xref: /linux/drivers/md/dm-crypt.c (revision b48543c451c30387b53ee6e202dda8d5303f6268)
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 used_tag_size;
218 	unsigned int tuple_size;
219 
220 	/*
221 	 * pool for per bio private data, crypto requests,
222 	 * encryption requeusts/buffer pages and integrity tags
223 	 */
224 	unsigned int tag_pool_max_sectors;
225 	mempool_t tag_pool;
226 	mempool_t req_pool;
227 	mempool_t page_pool;
228 
229 	struct bio_set bs;
230 	struct mutex bio_alloc_lock;
231 
232 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
233 	u8 key[] __counted_by(key_size);
234 };
235 
236 #define MIN_IOS		64
237 #define MAX_TAG_SIZE	480
238 #define POOL_ENTRY_SIZE	512
239 
240 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
241 static unsigned int dm_crypt_clients_n;
242 static volatile unsigned long dm_crypt_pages_per_client;
243 #define DM_CRYPT_MEMORY_PERCENT			2
244 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
245 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
246 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
247 
248 static unsigned int max_read_size = 0;
249 module_param(max_read_size, uint, 0644);
250 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
251 static unsigned int max_write_size = 0;
252 module_param(max_write_size, uint, 0644);
253 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
254 static unsigned get_max_request_size(struct crypt_config *cc, bool wrt)
255 {
256 	unsigned val, sector_align;
257 	val = !wrt ? READ_ONCE(max_read_size) : READ_ONCE(max_write_size);
258 	if (likely(!val))
259 		val = !wrt ? DM_CRYPT_DEFAULT_MAX_READ_SIZE : DM_CRYPT_DEFAULT_MAX_WRITE_SIZE;
260 	if (wrt || cc->used_tag_size) {
261 		if (unlikely(val > BIO_MAX_VECS << PAGE_SHIFT))
262 			val = BIO_MAX_VECS << PAGE_SHIFT;
263 	}
264 	sector_align = max(bdev_logical_block_size(cc->dev->bdev), (unsigned)cc->sector_size);
265 	val = round_down(val, sector_align);
266 	if (unlikely(!val))
267 		val = sector_align;
268 	return val >> SECTOR_SHIFT;
269 }
270 
271 static void crypt_endio(struct bio *clone);
272 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
273 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
274 					     struct scatterlist *sg);
275 
276 static bool crypt_integrity_aead(struct crypt_config *cc);
277 
278 /*
279  * Use this to access cipher attributes that are independent of the key.
280  */
281 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
282 {
283 	return cc->cipher_tfm.tfms[0];
284 }
285 
286 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
287 {
288 	return cc->cipher_tfm.tfms_aead[0];
289 }
290 
291 /*
292  * Different IV generation algorithms:
293  *
294  * plain: the initial vector is the 32-bit little-endian version of the sector
295  *        number, padded with zeros if necessary.
296  *
297  * plain64: the initial vector is the 64-bit little-endian version of the sector
298  *        number, padded with zeros if necessary.
299  *
300  * plain64be: the initial vector is the 64-bit big-endian version of the sector
301  *        number, padded with zeros if necessary.
302  *
303  * essiv: "encrypted sector|salt initial vector", the sector number is
304  *        encrypted with the bulk cipher using a salt as key. The salt
305  *        should be derived from the bulk cipher's key via hashing.
306  *
307  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
308  *        (needed for LRW-32-AES and possible other narrow block modes)
309  *
310  * null: the initial vector is always zero.  Provides compatibility with
311  *       obsolete loop_fish2 devices.  Do not use for new devices.
312  *
313  * lmk:  Compatible implementation of the block chaining mode used
314  *       by the Loop-AES block device encryption system
315  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
316  *       It operates on full 512 byte sectors and uses CBC
317  *       with an IV derived from the sector number, the data and
318  *       optionally extra IV seed.
319  *       This means that after decryption the first block
320  *       of sector must be tweaked according to decrypted data.
321  *       Loop-AES can use three encryption schemes:
322  *         version 1: is plain aes-cbc mode
323  *         version 2: uses 64 multikey scheme with lmk IV generator
324  *         version 3: the same as version 2 with additional IV seed
325  *                   (it uses 65 keys, last key is used as IV seed)
326  *
327  * tcw:  Compatible implementation of the block chaining mode used
328  *       by the TrueCrypt device encryption system (prior to version 4.1).
329  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
330  *       It operates on full 512 byte sectors and uses CBC
331  *       with an IV derived from initial key and the sector number.
332  *       In addition, whitening value is applied on every sector, whitening
333  *       is calculated from initial key, sector number and mixed using CRC32.
334  *       Note that this encryption scheme is vulnerable to watermarking attacks
335  *       and should be used for old compatible containers access only.
336  *
337  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
338  *        The IV is encrypted little-endian byte-offset (with the same key
339  *        and cipher as the volume).
340  *
341  * elephant: The extended version of eboiv with additional Elephant diffuser
342  *           used with Bitlocker CBC mode.
343  *           This mode was used in older Windows systems
344  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
345  */
346 
347 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
348 			      struct dm_crypt_request *dmreq)
349 {
350 	memset(iv, 0, cc->iv_size);
351 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
352 
353 	return 0;
354 }
355 
356 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
357 				struct dm_crypt_request *dmreq)
358 {
359 	memset(iv, 0, cc->iv_size);
360 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
361 
362 	return 0;
363 }
364 
365 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
366 				  struct dm_crypt_request *dmreq)
367 {
368 	memset(iv, 0, cc->iv_size);
369 	/* iv_size is at least of size u64; usually it is 16 bytes */
370 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
371 
372 	return 0;
373 }
374 
375 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
376 			      struct dm_crypt_request *dmreq)
377 {
378 	/*
379 	 * ESSIV encryption of the IV is now handled by the crypto API,
380 	 * so just pass the plain sector number here.
381 	 */
382 	memset(iv, 0, cc->iv_size);
383 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
384 
385 	return 0;
386 }
387 
388 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
389 			      const char *opts)
390 {
391 	unsigned int bs;
392 	int log;
393 
394 	if (crypt_integrity_aead(cc))
395 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
396 	else
397 		bs = crypto_skcipher_blocksize(any_tfm(cc));
398 	log = ilog2(bs);
399 
400 	/*
401 	 * We need to calculate how far we must shift the sector count
402 	 * to get the cipher block count, we use this shift in _gen.
403 	 */
404 	if (1 << log != bs) {
405 		ti->error = "cypher blocksize is not a power of 2";
406 		return -EINVAL;
407 	}
408 
409 	if (log > 9) {
410 		ti->error = "cypher blocksize is > 512";
411 		return -EINVAL;
412 	}
413 
414 	cc->iv_gen_private.benbi.shift = 9 - log;
415 
416 	return 0;
417 }
418 
419 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
420 {
421 }
422 
423 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
424 			      struct dm_crypt_request *dmreq)
425 {
426 	__be64 val;
427 
428 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
429 
430 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
431 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
432 
433 	return 0;
434 }
435 
436 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
437 			     struct dm_crypt_request *dmreq)
438 {
439 	memset(iv, 0, cc->iv_size);
440 
441 	return 0;
442 }
443 
444 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
445 {
446 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
447 
448 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
449 		crypto_free_shash(lmk->hash_tfm);
450 	lmk->hash_tfm = NULL;
451 
452 	kfree_sensitive(lmk->seed);
453 	lmk->seed = NULL;
454 }
455 
456 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
457 			    const char *opts)
458 {
459 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
460 
461 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
462 		ti->error = "Unsupported sector size for LMK";
463 		return -EINVAL;
464 	}
465 
466 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
467 					   CRYPTO_ALG_ALLOCATES_MEMORY);
468 	if (IS_ERR(lmk->hash_tfm)) {
469 		ti->error = "Error initializing LMK hash";
470 		return PTR_ERR(lmk->hash_tfm);
471 	}
472 
473 	/* No seed in LMK version 2 */
474 	if (cc->key_parts == cc->tfms_count) {
475 		lmk->seed = NULL;
476 		return 0;
477 	}
478 
479 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
480 	if (!lmk->seed) {
481 		crypt_iv_lmk_dtr(cc);
482 		ti->error = "Error kmallocing seed storage in LMK";
483 		return -ENOMEM;
484 	}
485 
486 	return 0;
487 }
488 
489 static int crypt_iv_lmk_init(struct crypt_config *cc)
490 {
491 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
492 	int subkey_size = cc->key_size / cc->key_parts;
493 
494 	/* LMK seed is on the position of LMK_KEYS + 1 key */
495 	if (lmk->seed)
496 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
497 		       crypto_shash_digestsize(lmk->hash_tfm));
498 
499 	return 0;
500 }
501 
502 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
503 {
504 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
505 
506 	if (lmk->seed)
507 		memset(lmk->seed, 0, LMK_SEED_SIZE);
508 
509 	return 0;
510 }
511 
512 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
513 			    struct dm_crypt_request *dmreq,
514 			    u8 *data)
515 {
516 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
517 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
518 	struct md5_state md5state;
519 	__le32 buf[4];
520 	int i, r;
521 
522 	desc->tfm = lmk->hash_tfm;
523 
524 	r = crypto_shash_init(desc);
525 	if (r)
526 		return r;
527 
528 	if (lmk->seed) {
529 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
530 		if (r)
531 			return r;
532 	}
533 
534 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
535 	r = crypto_shash_update(desc, data + 16, 16 * 31);
536 	if (r)
537 		return r;
538 
539 	/* Sector is cropped to 56 bits here */
540 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
541 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
542 	buf[2] = cpu_to_le32(4024);
543 	buf[3] = 0;
544 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
545 	if (r)
546 		return r;
547 
548 	/* No MD5 padding here */
549 	r = crypto_shash_export(desc, &md5state);
550 	if (r)
551 		return r;
552 
553 	for (i = 0; i < MD5_HASH_WORDS; i++)
554 		__cpu_to_le32s(&md5state.hash[i]);
555 	memcpy(iv, &md5state.hash, cc->iv_size);
556 
557 	return 0;
558 }
559 
560 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
561 			    struct dm_crypt_request *dmreq)
562 {
563 	struct scatterlist *sg;
564 	u8 *src;
565 	int r = 0;
566 
567 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
568 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
569 		src = kmap_local_page(sg_page(sg));
570 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
571 		kunmap_local(src);
572 	} else
573 		memset(iv, 0, cc->iv_size);
574 
575 	return r;
576 }
577 
578 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
579 			     struct dm_crypt_request *dmreq)
580 {
581 	struct scatterlist *sg;
582 	u8 *dst;
583 	int r;
584 
585 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
586 		return 0;
587 
588 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
589 	dst = kmap_local_page(sg_page(sg));
590 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
591 
592 	/* Tweak the first block of plaintext sector */
593 	if (!r)
594 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
595 
596 	kunmap_local(dst);
597 	return r;
598 }
599 
600 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
601 {
602 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
603 
604 	kfree_sensitive(tcw->iv_seed);
605 	tcw->iv_seed = NULL;
606 	kfree_sensitive(tcw->whitening);
607 	tcw->whitening = NULL;
608 
609 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
610 		crypto_free_shash(tcw->crc32_tfm);
611 	tcw->crc32_tfm = NULL;
612 }
613 
614 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
615 			    const char *opts)
616 {
617 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
618 
619 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
620 		ti->error = "Unsupported sector size for TCW";
621 		return -EINVAL;
622 	}
623 
624 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
625 		ti->error = "Wrong key size for TCW";
626 		return -EINVAL;
627 	}
628 
629 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
630 					    CRYPTO_ALG_ALLOCATES_MEMORY);
631 	if (IS_ERR(tcw->crc32_tfm)) {
632 		ti->error = "Error initializing CRC32 in TCW";
633 		return PTR_ERR(tcw->crc32_tfm);
634 	}
635 
636 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
637 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
638 	if (!tcw->iv_seed || !tcw->whitening) {
639 		crypt_iv_tcw_dtr(cc);
640 		ti->error = "Error allocating seed storage in TCW";
641 		return -ENOMEM;
642 	}
643 
644 	return 0;
645 }
646 
647 static int crypt_iv_tcw_init(struct crypt_config *cc)
648 {
649 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
650 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
651 
652 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
653 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
654 	       TCW_WHITENING_SIZE);
655 
656 	return 0;
657 }
658 
659 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
660 {
661 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
662 
663 	memset(tcw->iv_seed, 0, cc->iv_size);
664 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
665 
666 	return 0;
667 }
668 
669 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
670 				  struct dm_crypt_request *dmreq,
671 				  u8 *data)
672 {
673 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
674 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
675 	u8 buf[TCW_WHITENING_SIZE];
676 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
677 	int i, r;
678 
679 	/* xor whitening with sector number */
680 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
681 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
682 
683 	/* calculate crc32 for every 32bit part and xor it */
684 	desc->tfm = tcw->crc32_tfm;
685 	for (i = 0; i < 4; i++) {
686 		r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
687 		if (r)
688 			goto out;
689 	}
690 	crypto_xor(&buf[0], &buf[12], 4);
691 	crypto_xor(&buf[4], &buf[8], 4);
692 
693 	/* apply whitening (8 bytes) to whole sector */
694 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
695 		crypto_xor(data + i * 8, buf, 8);
696 out:
697 	memzero_explicit(buf, sizeof(buf));
698 	return r;
699 }
700 
701 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
702 			    struct dm_crypt_request *dmreq)
703 {
704 	struct scatterlist *sg;
705 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
706 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
707 	u8 *src;
708 	int r = 0;
709 
710 	/* Remove whitening from ciphertext */
711 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
712 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
713 		src = kmap_local_page(sg_page(sg));
714 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
715 		kunmap_local(src);
716 	}
717 
718 	/* Calculate IV */
719 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
720 	if (cc->iv_size > 8)
721 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
722 			       cc->iv_size - 8);
723 
724 	return r;
725 }
726 
727 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
728 			     struct dm_crypt_request *dmreq)
729 {
730 	struct scatterlist *sg;
731 	u8 *dst;
732 	int r;
733 
734 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
735 		return 0;
736 
737 	/* Apply whitening on ciphertext */
738 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
739 	dst = kmap_local_page(sg_page(sg));
740 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
741 	kunmap_local(dst);
742 
743 	return r;
744 }
745 
746 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
747 				struct dm_crypt_request *dmreq)
748 {
749 	/* Used only for writes, there must be an additional space to store IV */
750 	get_random_bytes(iv, cc->iv_size);
751 	return 0;
752 }
753 
754 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
755 			    const char *opts)
756 {
757 	if (crypt_integrity_aead(cc)) {
758 		ti->error = "AEAD transforms not supported for EBOIV";
759 		return -EINVAL;
760 	}
761 
762 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
763 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
764 		return -EINVAL;
765 	}
766 
767 	return 0;
768 }
769 
770 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
771 			    struct dm_crypt_request *dmreq)
772 {
773 	struct crypto_skcipher *tfm = any_tfm(cc);
774 	struct skcipher_request *req;
775 	struct scatterlist src, dst;
776 	DECLARE_CRYPTO_WAIT(wait);
777 	unsigned int reqsize;
778 	int err;
779 	u8 *buf;
780 
781 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
782 	reqsize = ALIGN(reqsize, __alignof__(__le64));
783 
784 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
785 	if (!req)
786 		return -ENOMEM;
787 
788 	skcipher_request_set_tfm(req, tfm);
789 
790 	buf = (u8 *)req + reqsize;
791 	memset(buf, 0, cc->iv_size);
792 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
793 
794 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
795 	sg_init_one(&dst, iv, cc->iv_size);
796 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
797 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
798 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
799 	kfree_sensitive(req);
800 
801 	return err;
802 }
803 
804 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
805 {
806 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
807 
808 	crypto_free_skcipher(elephant->tfm);
809 	elephant->tfm = NULL;
810 }
811 
812 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
813 			    const char *opts)
814 {
815 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
816 	int r;
817 
818 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
819 					      CRYPTO_ALG_ALLOCATES_MEMORY);
820 	if (IS_ERR(elephant->tfm)) {
821 		r = PTR_ERR(elephant->tfm);
822 		elephant->tfm = NULL;
823 		return r;
824 	}
825 
826 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
827 	if (r)
828 		crypt_iv_elephant_dtr(cc);
829 	return r;
830 }
831 
832 static void diffuser_disk_to_cpu(u32 *d, size_t n)
833 {
834 #ifndef __LITTLE_ENDIAN
835 	int i;
836 
837 	for (i = 0; i < n; i++)
838 		d[i] = le32_to_cpu((__le32)d[i]);
839 #endif
840 }
841 
842 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
843 {
844 #ifndef __LITTLE_ENDIAN
845 	int i;
846 
847 	for (i = 0; i < n; i++)
848 		d[i] = cpu_to_le32((u32)d[i]);
849 #endif
850 }
851 
852 static void diffuser_a_decrypt(u32 *d, size_t n)
853 {
854 	int i, i1, i2, i3;
855 
856 	for (i = 0; i < 5; i++) {
857 		i1 = 0;
858 		i2 = n - 2;
859 		i3 = n - 5;
860 
861 		while (i1 < (n - 1)) {
862 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
863 			i1++; i2++; i3++;
864 
865 			if (i3 >= n)
866 				i3 -= n;
867 
868 			d[i1] += d[i2] ^ d[i3];
869 			i1++; i2++; i3++;
870 
871 			if (i2 >= n)
872 				i2 -= n;
873 
874 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
875 			i1++; i2++; i3++;
876 
877 			d[i1] += d[i2] ^ d[i3];
878 			i1++; i2++; i3++;
879 		}
880 	}
881 }
882 
883 static void diffuser_a_encrypt(u32 *d, size_t n)
884 {
885 	int i, i1, i2, i3;
886 
887 	for (i = 0; i < 5; i++) {
888 		i1 = n - 1;
889 		i2 = n - 2 - 1;
890 		i3 = n - 5 - 1;
891 
892 		while (i1 > 0) {
893 			d[i1] -= d[i2] ^ d[i3];
894 			i1--; i2--; i3--;
895 
896 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
897 			i1--; i2--; i3--;
898 
899 			if (i2 < 0)
900 				i2 += n;
901 
902 			d[i1] -= d[i2] ^ d[i3];
903 			i1--; i2--; i3--;
904 
905 			if (i3 < 0)
906 				i3 += n;
907 
908 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
909 			i1--; i2--; i3--;
910 		}
911 	}
912 }
913 
914 static void diffuser_b_decrypt(u32 *d, size_t n)
915 {
916 	int i, i1, i2, i3;
917 
918 	for (i = 0; i < 3; i++) {
919 		i1 = 0;
920 		i2 = 2;
921 		i3 = 5;
922 
923 		while (i1 < (n - 1)) {
924 			d[i1] += d[i2] ^ d[i3];
925 			i1++; i2++; i3++;
926 
927 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
928 			i1++; i2++; i3++;
929 
930 			if (i2 >= n)
931 				i2 -= n;
932 
933 			d[i1] += d[i2] ^ d[i3];
934 			i1++; i2++; i3++;
935 
936 			if (i3 >= n)
937 				i3 -= n;
938 
939 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
940 			i1++; i2++; i3++;
941 		}
942 	}
943 }
944 
945 static void diffuser_b_encrypt(u32 *d, size_t n)
946 {
947 	int i, i1, i2, i3;
948 
949 	for (i = 0; i < 3; i++) {
950 		i1 = n - 1;
951 		i2 = 2 - 1;
952 		i3 = 5 - 1;
953 
954 		while (i1 > 0) {
955 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
956 			i1--; i2--; i3--;
957 
958 			if (i3 < 0)
959 				i3 += n;
960 
961 			d[i1] -= d[i2] ^ d[i3];
962 			i1--; i2--; i3--;
963 
964 			if (i2 < 0)
965 				i2 += n;
966 
967 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
968 			i1--; i2--; i3--;
969 
970 			d[i1] -= d[i2] ^ d[i3];
971 			i1--; i2--; i3--;
972 		}
973 	}
974 }
975 
976 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
977 {
978 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
979 	u8 *es, *ks, *data, *data2, *data_offset;
980 	struct skcipher_request *req;
981 	struct scatterlist *sg, *sg2, src, dst;
982 	DECLARE_CRYPTO_WAIT(wait);
983 	int i, r;
984 
985 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
986 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
987 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
988 
989 	if (!req || !es || !ks) {
990 		r = -ENOMEM;
991 		goto out;
992 	}
993 
994 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
995 
996 	/* E(Ks, e(s)) */
997 	sg_init_one(&src, es, 16);
998 	sg_init_one(&dst, ks, 16);
999 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
1000 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
1001 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1002 	if (r)
1003 		goto out;
1004 
1005 	/* E(Ks, e'(s)) */
1006 	es[15] = 0x80;
1007 	sg_init_one(&dst, &ks[16], 16);
1008 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1009 	if (r)
1010 		goto out;
1011 
1012 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
1013 	data = kmap_local_page(sg_page(sg));
1014 	data_offset = data + sg->offset;
1015 
1016 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
1017 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1018 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
1019 		data2 = kmap_local_page(sg_page(sg2));
1020 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
1021 		kunmap_local(data2);
1022 	}
1023 
1024 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
1025 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1026 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1027 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1028 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1029 	}
1030 
1031 	for (i = 0; i < (cc->sector_size / 32); i++)
1032 		crypto_xor(data_offset + i * 32, ks, 32);
1033 
1034 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1035 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1036 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1037 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1038 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1039 	}
1040 
1041 	kunmap_local(data);
1042 out:
1043 	kfree_sensitive(ks);
1044 	kfree_sensitive(es);
1045 	skcipher_request_free(req);
1046 	return r;
1047 }
1048 
1049 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1050 			    struct dm_crypt_request *dmreq)
1051 {
1052 	int r;
1053 
1054 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1055 		r = crypt_iv_elephant(cc, dmreq);
1056 		if (r)
1057 			return r;
1058 	}
1059 
1060 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1061 }
1062 
1063 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1064 				  struct dm_crypt_request *dmreq)
1065 {
1066 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1067 		return crypt_iv_elephant(cc, dmreq);
1068 
1069 	return 0;
1070 }
1071 
1072 static int crypt_iv_elephant_init(struct crypt_config *cc)
1073 {
1074 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1075 	int key_offset = cc->key_size - cc->key_extra_size;
1076 
1077 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1078 }
1079 
1080 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1081 {
1082 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1083 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1084 
1085 	memset(key, 0, cc->key_extra_size);
1086 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1087 }
1088 
1089 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1090 	.generator = crypt_iv_plain_gen
1091 };
1092 
1093 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1094 	.generator = crypt_iv_plain64_gen
1095 };
1096 
1097 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1098 	.generator = crypt_iv_plain64be_gen
1099 };
1100 
1101 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1102 	.generator = crypt_iv_essiv_gen
1103 };
1104 
1105 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1106 	.ctr	   = crypt_iv_benbi_ctr,
1107 	.dtr	   = crypt_iv_benbi_dtr,
1108 	.generator = crypt_iv_benbi_gen
1109 };
1110 
1111 static const struct crypt_iv_operations crypt_iv_null_ops = {
1112 	.generator = crypt_iv_null_gen
1113 };
1114 
1115 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1116 	.ctr	   = crypt_iv_lmk_ctr,
1117 	.dtr	   = crypt_iv_lmk_dtr,
1118 	.init	   = crypt_iv_lmk_init,
1119 	.wipe	   = crypt_iv_lmk_wipe,
1120 	.generator = crypt_iv_lmk_gen,
1121 	.post	   = crypt_iv_lmk_post
1122 };
1123 
1124 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1125 	.ctr	   = crypt_iv_tcw_ctr,
1126 	.dtr	   = crypt_iv_tcw_dtr,
1127 	.init	   = crypt_iv_tcw_init,
1128 	.wipe	   = crypt_iv_tcw_wipe,
1129 	.generator = crypt_iv_tcw_gen,
1130 	.post	   = crypt_iv_tcw_post
1131 };
1132 
1133 static const struct crypt_iv_operations crypt_iv_random_ops = {
1134 	.generator = crypt_iv_random_gen
1135 };
1136 
1137 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1138 	.ctr	   = crypt_iv_eboiv_ctr,
1139 	.generator = crypt_iv_eboiv_gen
1140 };
1141 
1142 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1143 	.ctr	   = crypt_iv_elephant_ctr,
1144 	.dtr	   = crypt_iv_elephant_dtr,
1145 	.init	   = crypt_iv_elephant_init,
1146 	.wipe	   = crypt_iv_elephant_wipe,
1147 	.generator = crypt_iv_elephant_gen,
1148 	.post	   = crypt_iv_elephant_post
1149 };
1150 
1151 /*
1152  * Integrity extensions
1153  */
1154 static bool crypt_integrity_aead(struct crypt_config *cc)
1155 {
1156 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1157 }
1158 
1159 static bool crypt_integrity_hmac(struct crypt_config *cc)
1160 {
1161 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1162 }
1163 
1164 /* Get sg containing data */
1165 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1166 					     struct scatterlist *sg)
1167 {
1168 	if (unlikely(crypt_integrity_aead(cc)))
1169 		return &sg[2];
1170 
1171 	return sg;
1172 }
1173 
1174 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1175 {
1176 	struct bio_integrity_payload *bip;
1177 	unsigned int tag_len;
1178 	int ret;
1179 
1180 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1181 		return 0;
1182 
1183 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1184 	if (IS_ERR(bip))
1185 		return PTR_ERR(bip);
1186 
1187 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1188 
1189 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1190 
1191 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1192 				     tag_len, offset_in_page(io->integrity_metadata));
1193 	if (unlikely(ret != tag_len))
1194 		return -ENOMEM;
1195 
1196 	return 0;
1197 }
1198 
1199 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1200 {
1201 #ifdef CONFIG_BLK_DEV_INTEGRITY
1202 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1203 	struct mapped_device *md = dm_table_get_md(ti->table);
1204 
1205 	/* We require an underlying device with non-PI metadata */
1206 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1207 		ti->error = "Integrity profile not supported.";
1208 		return -EINVAL;
1209 	}
1210 
1211 	if (bi->tuple_size < cc->used_tag_size) {
1212 		ti->error = "Integrity profile tag size mismatch.";
1213 		return -EINVAL;
1214 	}
1215 	cc->tuple_size = bi->tuple_size;
1216 	if (1 << bi->interval_exp != cc->sector_size) {
1217 		ti->error = "Integrity profile sector size mismatch.";
1218 		return -EINVAL;
1219 	}
1220 
1221 	if (crypt_integrity_aead(cc)) {
1222 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1223 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1224 		       cc->integrity_tag_size, cc->integrity_iv_size);
1225 
1226 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1227 			ti->error = "Integrity AEAD auth tag size is not supported.";
1228 			return -EINVAL;
1229 		}
1230 	} else if (cc->integrity_iv_size)
1231 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1232 		       cc->integrity_iv_size);
1233 
1234 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1235 		ti->error = "Not enough space for integrity tag in the profile.";
1236 		return -EINVAL;
1237 	}
1238 
1239 	return 0;
1240 #else
1241 	ti->error = "Integrity profile not supported.";
1242 	return -EINVAL;
1243 #endif
1244 }
1245 
1246 static void crypt_convert_init(struct crypt_config *cc,
1247 			       struct convert_context *ctx,
1248 			       struct bio *bio_out, struct bio *bio_in,
1249 			       sector_t sector)
1250 {
1251 	ctx->bio_in = bio_in;
1252 	ctx->bio_out = bio_out;
1253 	if (bio_in)
1254 		ctx->iter_in = bio_in->bi_iter;
1255 	if (bio_out)
1256 		ctx->iter_out = bio_out->bi_iter;
1257 	ctx->cc_sector = sector + cc->iv_offset;
1258 	init_completion(&ctx->restart);
1259 }
1260 
1261 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1262 					     void *req)
1263 {
1264 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1265 }
1266 
1267 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1268 {
1269 	return (void *)((char *)dmreq - cc->dmreq_start);
1270 }
1271 
1272 static u8 *iv_of_dmreq(struct crypt_config *cc,
1273 		       struct dm_crypt_request *dmreq)
1274 {
1275 	if (crypt_integrity_aead(cc))
1276 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1277 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1278 	else
1279 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1280 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1281 }
1282 
1283 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1284 		       struct dm_crypt_request *dmreq)
1285 {
1286 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1287 }
1288 
1289 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1290 		       struct dm_crypt_request *dmreq)
1291 {
1292 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1293 
1294 	return (__le64 *) ptr;
1295 }
1296 
1297 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1298 		       struct dm_crypt_request *dmreq)
1299 {
1300 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1301 		  cc->iv_size + sizeof(uint64_t);
1302 
1303 	return (unsigned int *)ptr;
1304 }
1305 
1306 static void *tag_from_dmreq(struct crypt_config *cc,
1307 				struct dm_crypt_request *dmreq)
1308 {
1309 	struct convert_context *ctx = dmreq->ctx;
1310 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1311 
1312 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1313 		cc->tuple_size];
1314 }
1315 
1316 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1317 			       struct dm_crypt_request *dmreq)
1318 {
1319 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1320 }
1321 
1322 static int crypt_convert_block_aead(struct crypt_config *cc,
1323 				     struct convert_context *ctx,
1324 				     struct aead_request *req,
1325 				     unsigned int tag_offset)
1326 {
1327 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1328 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1329 	struct dm_crypt_request *dmreq;
1330 	u8 *iv, *org_iv, *tag_iv, *tag;
1331 	__le64 *sector;
1332 	int r = 0;
1333 
1334 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1335 
1336 	/* Reject unexpected unaligned bio. */
1337 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1338 		return -EIO;
1339 
1340 	dmreq = dmreq_of_req(cc, req);
1341 	dmreq->iv_sector = ctx->cc_sector;
1342 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1343 		dmreq->iv_sector >>= cc->sector_shift;
1344 	dmreq->ctx = ctx;
1345 
1346 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1347 
1348 	sector = org_sector_of_dmreq(cc, dmreq);
1349 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1350 
1351 	iv = iv_of_dmreq(cc, dmreq);
1352 	org_iv = org_iv_of_dmreq(cc, dmreq);
1353 	tag = tag_from_dmreq(cc, dmreq);
1354 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1355 
1356 	/* AEAD request:
1357 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1358 	 *  | (authenticated) | (auth+encryption) |              |
1359 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1360 	 */
1361 	sg_init_table(dmreq->sg_in, 4);
1362 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1363 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1364 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1365 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1366 
1367 	sg_init_table(dmreq->sg_out, 4);
1368 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1369 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1370 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1371 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1372 
1373 	if (cc->iv_gen_ops) {
1374 		/* For READs use IV stored in integrity metadata */
1375 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1376 			memcpy(org_iv, tag_iv, cc->iv_size);
1377 		} else {
1378 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1379 			if (r < 0)
1380 				return r;
1381 			/* Store generated IV in integrity metadata */
1382 			if (cc->integrity_iv_size)
1383 				memcpy(tag_iv, org_iv, cc->iv_size);
1384 		}
1385 		/* Working copy of IV, to be modified in crypto API */
1386 		memcpy(iv, org_iv, cc->iv_size);
1387 	}
1388 
1389 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1390 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1391 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1392 				       cc->sector_size, iv);
1393 		r = crypto_aead_encrypt(req);
1394 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1395 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1396 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1397 	} else {
1398 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1399 				       cc->sector_size + cc->integrity_tag_size, iv);
1400 		r = crypto_aead_decrypt(req);
1401 	}
1402 
1403 	if (r == -EBADMSG) {
1404 		sector_t s = le64_to_cpu(*sector);
1405 
1406 		ctx->aead_failed = true;
1407 		if (ctx->aead_recheck) {
1408 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1409 				    ctx->bio_in->bi_bdev, s);
1410 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1411 					 ctx->bio_in, s, 0);
1412 		}
1413 	}
1414 
1415 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1416 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1417 
1418 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1419 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1420 
1421 	return r;
1422 }
1423 
1424 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1425 					struct convert_context *ctx,
1426 					struct skcipher_request *req,
1427 					unsigned int tag_offset)
1428 {
1429 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1430 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1431 	struct scatterlist *sg_in, *sg_out;
1432 	struct dm_crypt_request *dmreq;
1433 	u8 *iv, *org_iv, *tag_iv;
1434 	__le64 *sector;
1435 	int r = 0;
1436 
1437 	/* Reject unexpected unaligned bio. */
1438 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1439 		return -EIO;
1440 
1441 	dmreq = dmreq_of_req(cc, req);
1442 	dmreq->iv_sector = ctx->cc_sector;
1443 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1444 		dmreq->iv_sector >>= cc->sector_shift;
1445 	dmreq->ctx = ctx;
1446 
1447 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1448 
1449 	iv = iv_of_dmreq(cc, dmreq);
1450 	org_iv = org_iv_of_dmreq(cc, dmreq);
1451 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1452 
1453 	sector = org_sector_of_dmreq(cc, dmreq);
1454 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1455 
1456 	/* For skcipher we use only the first sg item */
1457 	sg_in  = &dmreq->sg_in[0];
1458 	sg_out = &dmreq->sg_out[0];
1459 
1460 	sg_init_table(sg_in, 1);
1461 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1462 
1463 	sg_init_table(sg_out, 1);
1464 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1465 
1466 	if (cc->iv_gen_ops) {
1467 		/* For READs use IV stored in integrity metadata */
1468 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1469 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1470 		} else {
1471 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1472 			if (r < 0)
1473 				return r;
1474 			/* Data can be already preprocessed in generator */
1475 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1476 				sg_in = sg_out;
1477 			/* Store generated IV in integrity metadata */
1478 			if (cc->integrity_iv_size)
1479 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1480 		}
1481 		/* Working copy of IV, to be modified in crypto API */
1482 		memcpy(iv, org_iv, cc->iv_size);
1483 	}
1484 
1485 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1486 
1487 	if (bio_data_dir(ctx->bio_in) == WRITE)
1488 		r = crypto_skcipher_encrypt(req);
1489 	else
1490 		r = crypto_skcipher_decrypt(req);
1491 
1492 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1493 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1494 
1495 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1496 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1497 
1498 	return r;
1499 }
1500 
1501 static void kcryptd_async_done(void *async_req, int error);
1502 
1503 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1504 				     struct convert_context *ctx)
1505 {
1506 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1507 
1508 	if (!ctx->r.req) {
1509 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1510 		if (!ctx->r.req)
1511 			return -ENOMEM;
1512 	}
1513 
1514 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1515 
1516 	/*
1517 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1518 	 * requests if driver request queue is full.
1519 	 */
1520 	skcipher_request_set_callback(ctx->r.req,
1521 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1522 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1523 
1524 	return 0;
1525 }
1526 
1527 static int crypt_alloc_req_aead(struct crypt_config *cc,
1528 				 struct convert_context *ctx)
1529 {
1530 	if (!ctx->r.req_aead) {
1531 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1532 		if (!ctx->r.req_aead)
1533 			return -ENOMEM;
1534 	}
1535 
1536 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1537 
1538 	/*
1539 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1540 	 * requests if driver request queue is full.
1541 	 */
1542 	aead_request_set_callback(ctx->r.req_aead,
1543 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1544 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1545 
1546 	return 0;
1547 }
1548 
1549 static int crypt_alloc_req(struct crypt_config *cc,
1550 			    struct convert_context *ctx)
1551 {
1552 	if (crypt_integrity_aead(cc))
1553 		return crypt_alloc_req_aead(cc, ctx);
1554 	else
1555 		return crypt_alloc_req_skcipher(cc, ctx);
1556 }
1557 
1558 static void crypt_free_req_skcipher(struct crypt_config *cc,
1559 				    struct skcipher_request *req, struct bio *base_bio)
1560 {
1561 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1562 
1563 	if ((struct skcipher_request *)(io + 1) != req)
1564 		mempool_free(req, &cc->req_pool);
1565 }
1566 
1567 static void crypt_free_req_aead(struct crypt_config *cc,
1568 				struct aead_request *req, struct bio *base_bio)
1569 {
1570 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1571 
1572 	if ((struct aead_request *)(io + 1) != req)
1573 		mempool_free(req, &cc->req_pool);
1574 }
1575 
1576 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1577 {
1578 	if (crypt_integrity_aead(cc))
1579 		crypt_free_req_aead(cc, req, base_bio);
1580 	else
1581 		crypt_free_req_skcipher(cc, req, base_bio);
1582 }
1583 
1584 /*
1585  * Encrypt / decrypt data from one bio to another one (can be the same one)
1586  */
1587 static blk_status_t crypt_convert(struct crypt_config *cc,
1588 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1589 {
1590 	unsigned int tag_offset = 0;
1591 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1592 	int r;
1593 
1594 	/*
1595 	 * if reset_pending is set we are dealing with the bio for the first time,
1596 	 * else we're continuing to work on the previous bio, so don't mess with
1597 	 * the cc_pending counter
1598 	 */
1599 	if (reset_pending)
1600 		atomic_set(&ctx->cc_pending, 1);
1601 
1602 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1603 
1604 		r = crypt_alloc_req(cc, ctx);
1605 		if (r) {
1606 			complete(&ctx->restart);
1607 			return BLK_STS_DEV_RESOURCE;
1608 		}
1609 
1610 		atomic_inc(&ctx->cc_pending);
1611 
1612 		if (crypt_integrity_aead(cc))
1613 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1614 		else
1615 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1616 
1617 		switch (r) {
1618 		/*
1619 		 * The request was queued by a crypto driver
1620 		 * but the driver request queue is full, let's wait.
1621 		 */
1622 		case -EBUSY:
1623 			if (in_interrupt()) {
1624 				if (try_wait_for_completion(&ctx->restart)) {
1625 					/*
1626 					 * we don't have to block to wait for completion,
1627 					 * so proceed
1628 					 */
1629 				} else {
1630 					/*
1631 					 * we can't wait for completion without blocking
1632 					 * exit and continue processing in a workqueue
1633 					 */
1634 					ctx->r.req = NULL;
1635 					ctx->cc_sector += sector_step;
1636 					tag_offset++;
1637 					return BLK_STS_DEV_RESOURCE;
1638 				}
1639 			} else {
1640 				wait_for_completion(&ctx->restart);
1641 			}
1642 			reinit_completion(&ctx->restart);
1643 			fallthrough;
1644 		/*
1645 		 * The request is queued and processed asynchronously,
1646 		 * completion function kcryptd_async_done() will be called.
1647 		 */
1648 		case -EINPROGRESS:
1649 			ctx->r.req = NULL;
1650 			ctx->cc_sector += sector_step;
1651 			tag_offset++;
1652 			continue;
1653 		/*
1654 		 * The request was already processed (synchronously).
1655 		 */
1656 		case 0:
1657 			atomic_dec(&ctx->cc_pending);
1658 			ctx->cc_sector += sector_step;
1659 			tag_offset++;
1660 			if (!atomic)
1661 				cond_resched();
1662 			continue;
1663 		/*
1664 		 * There was a data integrity error.
1665 		 */
1666 		case -EBADMSG:
1667 			atomic_dec(&ctx->cc_pending);
1668 			return BLK_STS_PROTECTION;
1669 		/*
1670 		 * There was an error while processing the request.
1671 		 */
1672 		default:
1673 			atomic_dec(&ctx->cc_pending);
1674 			return BLK_STS_IOERR;
1675 		}
1676 	}
1677 
1678 	return 0;
1679 }
1680 
1681 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1682 
1683 /*
1684  * Generate a new unfragmented bio with the given size
1685  * This should never violate the device limitations (but if it did then block
1686  * core should split the bio as needed).
1687  *
1688  * This function may be called concurrently. If we allocate from the mempool
1689  * concurrently, there is a possibility of deadlock. For example, if we have
1690  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1691  * the mempool concurrently, it may deadlock in a situation where both processes
1692  * have allocated 128 pages and the mempool is exhausted.
1693  *
1694  * In order to avoid this scenario we allocate the pages under a mutex.
1695  *
1696  * In order to not degrade performance with excessive locking, we try
1697  * non-blocking allocations without a mutex first but on failure we fallback
1698  * to blocking allocations with a mutex.
1699  *
1700  * In order to reduce allocation overhead, we try to allocate compound pages in
1701  * the first pass. If they are not available, we fall back to the mempool.
1702  */
1703 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1704 {
1705 	struct crypt_config *cc = io->cc;
1706 	struct bio *clone;
1707 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1708 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1709 	unsigned int remaining_size;
1710 	unsigned int order = MAX_PAGE_ORDER;
1711 
1712 retry:
1713 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1714 		mutex_lock(&cc->bio_alloc_lock);
1715 
1716 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1717 				 GFP_NOIO, &cc->bs);
1718 	clone->bi_private = io;
1719 	clone->bi_end_io = crypt_endio;
1720 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1721 
1722 	remaining_size = size;
1723 
1724 	while (remaining_size) {
1725 		struct page *pages;
1726 		unsigned size_to_add;
1727 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1728 		order = min(order, remaining_order);
1729 
1730 		while (order > 0) {
1731 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1732 					(1 << order) > dm_crypt_pages_per_client))
1733 				goto decrease_order;
1734 			pages = alloc_pages(gfp_mask
1735 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1736 				order);
1737 			if (likely(pages != NULL)) {
1738 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1739 				goto have_pages;
1740 			}
1741 decrease_order:
1742 			order--;
1743 		}
1744 
1745 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1746 		if (!pages) {
1747 			crypt_free_buffer_pages(cc, clone);
1748 			bio_put(clone);
1749 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1750 			order = 0;
1751 			goto retry;
1752 		}
1753 
1754 have_pages:
1755 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1756 		__bio_add_page(clone, pages, size_to_add, 0);
1757 		remaining_size -= size_to_add;
1758 	}
1759 
1760 	/* Allocate space for integrity tags */
1761 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1762 		crypt_free_buffer_pages(cc, clone);
1763 		bio_put(clone);
1764 		clone = NULL;
1765 	}
1766 
1767 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1768 		mutex_unlock(&cc->bio_alloc_lock);
1769 
1770 	return clone;
1771 }
1772 
1773 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1774 {
1775 	struct folio_iter fi;
1776 
1777 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1778 		bio_for_each_folio_all(fi, clone) {
1779 			if (folio_test_large(fi.folio)) {
1780 				percpu_counter_sub(&cc->n_allocated_pages,
1781 						1 << folio_order(fi.folio));
1782 				folio_put(fi.folio);
1783 			} else {
1784 				mempool_free(&fi.folio->page, &cc->page_pool);
1785 			}
1786 		}
1787 	}
1788 }
1789 
1790 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1791 			  struct bio *bio, sector_t sector)
1792 {
1793 	io->cc = cc;
1794 	io->base_bio = bio;
1795 	io->sector = sector;
1796 	io->error = 0;
1797 	io->ctx.aead_recheck = false;
1798 	io->ctx.aead_failed = false;
1799 	io->ctx.r.req = NULL;
1800 	io->integrity_metadata = NULL;
1801 	io->integrity_metadata_from_pool = false;
1802 	atomic_set(&io->io_pending, 0);
1803 }
1804 
1805 static void crypt_inc_pending(struct dm_crypt_io *io)
1806 {
1807 	atomic_inc(&io->io_pending);
1808 }
1809 
1810 static void kcryptd_queue_read(struct dm_crypt_io *io);
1811 
1812 /*
1813  * One of the bios was finished. Check for completion of
1814  * the whole request and correctly clean up the buffer.
1815  */
1816 static void crypt_dec_pending(struct dm_crypt_io *io)
1817 {
1818 	struct crypt_config *cc = io->cc;
1819 	struct bio *base_bio = io->base_bio;
1820 	blk_status_t error = io->error;
1821 
1822 	if (!atomic_dec_and_test(&io->io_pending))
1823 		return;
1824 
1825 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1826 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1827 		io->ctx.aead_recheck = true;
1828 		io->ctx.aead_failed = false;
1829 		io->error = 0;
1830 		kcryptd_queue_read(io);
1831 		return;
1832 	}
1833 
1834 	if (io->ctx.r.req)
1835 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1836 
1837 	if (unlikely(io->integrity_metadata_from_pool))
1838 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1839 	else
1840 		kfree(io->integrity_metadata);
1841 
1842 	base_bio->bi_status = error;
1843 
1844 	bio_endio(base_bio);
1845 }
1846 
1847 /*
1848  * kcryptd/kcryptd_io:
1849  *
1850  * Needed because it would be very unwise to do decryption in an
1851  * interrupt context.
1852  *
1853  * kcryptd performs the actual encryption or decryption.
1854  *
1855  * kcryptd_io performs the IO submission.
1856  *
1857  * They must be separated as otherwise the final stages could be
1858  * starved by new requests which can block in the first stages due
1859  * to memory allocation.
1860  *
1861  * The work is done per CPU global for all dm-crypt instances.
1862  * They should not depend on each other and do not block.
1863  */
1864 static void crypt_endio(struct bio *clone)
1865 {
1866 	struct dm_crypt_io *io = clone->bi_private;
1867 	struct crypt_config *cc = io->cc;
1868 	unsigned int rw = bio_data_dir(clone);
1869 	blk_status_t error = clone->bi_status;
1870 
1871 	if (io->ctx.aead_recheck && !error) {
1872 		kcryptd_queue_crypt(io);
1873 		return;
1874 	}
1875 
1876 	/*
1877 	 * free the processed pages
1878 	 */
1879 	if (rw == WRITE || io->ctx.aead_recheck)
1880 		crypt_free_buffer_pages(cc, clone);
1881 
1882 	bio_put(clone);
1883 
1884 	if (rw == READ && !error) {
1885 		kcryptd_queue_crypt(io);
1886 		return;
1887 	}
1888 
1889 	if (unlikely(error))
1890 		io->error = error;
1891 
1892 	crypt_dec_pending(io);
1893 }
1894 
1895 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1896 
1897 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1898 {
1899 	struct crypt_config *cc = io->cc;
1900 	struct bio *clone;
1901 
1902 	if (io->ctx.aead_recheck) {
1903 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1904 			return 1;
1905 		crypt_inc_pending(io);
1906 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1907 		if (unlikely(!clone)) {
1908 			crypt_dec_pending(io);
1909 			return 1;
1910 		}
1911 		clone->bi_iter.bi_sector = cc->start + io->sector;
1912 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1913 		io->saved_bi_iter = clone->bi_iter;
1914 		dm_submit_bio_remap(io->base_bio, clone);
1915 		return 0;
1916 	}
1917 
1918 	/*
1919 	 * We need the original biovec array in order to decrypt the whole bio
1920 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1921 	 * worry about the block layer modifying the biovec array; so leverage
1922 	 * bio_alloc_clone().
1923 	 */
1924 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1925 	if (!clone)
1926 		return 1;
1927 	clone->bi_private = io;
1928 	clone->bi_end_io = crypt_endio;
1929 
1930 	crypt_inc_pending(io);
1931 
1932 	clone->bi_iter.bi_sector = cc->start + io->sector;
1933 
1934 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1935 		crypt_dec_pending(io);
1936 		bio_put(clone);
1937 		return 1;
1938 	}
1939 
1940 	dm_submit_bio_remap(io->base_bio, clone);
1941 	return 0;
1942 }
1943 
1944 static void kcryptd_io_read_work(struct work_struct *work)
1945 {
1946 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1947 
1948 	crypt_inc_pending(io);
1949 	if (kcryptd_io_read(io, GFP_NOIO))
1950 		io->error = BLK_STS_RESOURCE;
1951 	crypt_dec_pending(io);
1952 }
1953 
1954 static void kcryptd_queue_read(struct dm_crypt_io *io)
1955 {
1956 	struct crypt_config *cc = io->cc;
1957 
1958 	INIT_WORK(&io->work, kcryptd_io_read_work);
1959 	queue_work(cc->io_queue, &io->work);
1960 }
1961 
1962 static void kcryptd_io_write(struct dm_crypt_io *io)
1963 {
1964 	struct bio *clone = io->ctx.bio_out;
1965 
1966 	dm_submit_bio_remap(io->base_bio, clone);
1967 }
1968 
1969 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1970 
1971 static int dmcrypt_write(void *data)
1972 {
1973 	struct crypt_config *cc = data;
1974 	struct dm_crypt_io *io;
1975 
1976 	while (1) {
1977 		struct rb_root write_tree;
1978 		struct blk_plug plug;
1979 
1980 		spin_lock_irq(&cc->write_thread_lock);
1981 continue_locked:
1982 
1983 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1984 			goto pop_from_list;
1985 
1986 		set_current_state(TASK_INTERRUPTIBLE);
1987 
1988 		spin_unlock_irq(&cc->write_thread_lock);
1989 
1990 		if (unlikely(kthread_should_stop())) {
1991 			set_current_state(TASK_RUNNING);
1992 			break;
1993 		}
1994 
1995 		schedule();
1996 
1997 		spin_lock_irq(&cc->write_thread_lock);
1998 		goto continue_locked;
1999 
2000 pop_from_list:
2001 		write_tree = cc->write_tree;
2002 		cc->write_tree = RB_ROOT;
2003 		spin_unlock_irq(&cc->write_thread_lock);
2004 
2005 		BUG_ON(rb_parent(write_tree.rb_node));
2006 
2007 		/*
2008 		 * Note: we cannot walk the tree here with rb_next because
2009 		 * the structures may be freed when kcryptd_io_write is called.
2010 		 */
2011 		blk_start_plug(&plug);
2012 		do {
2013 			io = crypt_io_from_node(rb_first(&write_tree));
2014 			rb_erase(&io->rb_node, &write_tree);
2015 			kcryptd_io_write(io);
2016 			cond_resched();
2017 		} while (!RB_EMPTY_ROOT(&write_tree));
2018 		blk_finish_plug(&plug);
2019 	}
2020 	return 0;
2021 }
2022 
2023 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
2024 {
2025 	struct bio *clone = io->ctx.bio_out;
2026 	struct crypt_config *cc = io->cc;
2027 	unsigned long flags;
2028 	sector_t sector;
2029 	struct rb_node **rbp, *parent;
2030 
2031 	if (unlikely(io->error)) {
2032 		crypt_free_buffer_pages(cc, clone);
2033 		bio_put(clone);
2034 		crypt_dec_pending(io);
2035 		return;
2036 	}
2037 
2038 	/* crypt_convert should have filled the clone bio */
2039 	BUG_ON(io->ctx.iter_out.bi_size);
2040 
2041 	clone->bi_iter.bi_sector = cc->start + io->sector;
2042 
2043 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2044 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2045 		dm_submit_bio_remap(io->base_bio, clone);
2046 		return;
2047 	}
2048 
2049 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2050 	if (RB_EMPTY_ROOT(&cc->write_tree))
2051 		wake_up_process(cc->write_thread);
2052 	rbp = &cc->write_tree.rb_node;
2053 	parent = NULL;
2054 	sector = io->sector;
2055 	while (*rbp) {
2056 		parent = *rbp;
2057 		if (sector < crypt_io_from_node(parent)->sector)
2058 			rbp = &(*rbp)->rb_left;
2059 		else
2060 			rbp = &(*rbp)->rb_right;
2061 	}
2062 	rb_link_node(&io->rb_node, parent, rbp);
2063 	rb_insert_color(&io->rb_node, &cc->write_tree);
2064 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2065 }
2066 
2067 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2068 				       struct convert_context *ctx)
2069 
2070 {
2071 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2072 		return false;
2073 
2074 	/*
2075 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2076 	 * constraints so they do not need to be issued inline by
2077 	 * kcryptd_crypt_write_convert().
2078 	 */
2079 	switch (bio_op(ctx->bio_in)) {
2080 	case REQ_OP_WRITE:
2081 	case REQ_OP_WRITE_ZEROES:
2082 		return true;
2083 	default:
2084 		return false;
2085 	}
2086 }
2087 
2088 static void kcryptd_crypt_write_continue(struct work_struct *work)
2089 {
2090 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2091 	struct crypt_config *cc = io->cc;
2092 	struct convert_context *ctx = &io->ctx;
2093 	int crypt_finished;
2094 	sector_t sector = io->sector;
2095 	blk_status_t r;
2096 
2097 	wait_for_completion(&ctx->restart);
2098 	reinit_completion(&ctx->restart);
2099 
2100 	r = crypt_convert(cc, &io->ctx, true, false);
2101 	if (r)
2102 		io->error = r;
2103 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2104 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2105 		/* Wait for completion signaled by kcryptd_async_done() */
2106 		wait_for_completion(&ctx->restart);
2107 		crypt_finished = 1;
2108 	}
2109 
2110 	/* Encryption was already finished, submit io now */
2111 	if (crypt_finished) {
2112 		kcryptd_crypt_write_io_submit(io, 0);
2113 		io->sector = sector;
2114 	}
2115 
2116 	crypt_dec_pending(io);
2117 }
2118 
2119 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2120 {
2121 	struct crypt_config *cc = io->cc;
2122 	struct convert_context *ctx = &io->ctx;
2123 	struct bio *clone;
2124 	int crypt_finished;
2125 	sector_t sector = io->sector;
2126 	blk_status_t r;
2127 
2128 	/*
2129 	 * Prevent io from disappearing until this function completes.
2130 	 */
2131 	crypt_inc_pending(io);
2132 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2133 
2134 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2135 	if (unlikely(!clone)) {
2136 		io->error = BLK_STS_IOERR;
2137 		goto dec;
2138 	}
2139 
2140 	io->ctx.bio_out = clone;
2141 	io->ctx.iter_out = clone->bi_iter;
2142 
2143 	if (crypt_integrity_aead(cc)) {
2144 		bio_copy_data(clone, io->base_bio);
2145 		io->ctx.bio_in = clone;
2146 		io->ctx.iter_in = clone->bi_iter;
2147 	}
2148 
2149 	sector += bio_sectors(clone);
2150 
2151 	crypt_inc_pending(io);
2152 	r = crypt_convert(cc, ctx,
2153 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2154 	/*
2155 	 * Crypto API backlogged the request, because its queue was full
2156 	 * and we're in softirq context, so continue from a workqueue
2157 	 * (TODO: is it actually possible to be in softirq in the write path?)
2158 	 */
2159 	if (r == BLK_STS_DEV_RESOURCE) {
2160 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2161 		queue_work(cc->crypt_queue, &io->work);
2162 		return;
2163 	}
2164 	if (r)
2165 		io->error = r;
2166 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2167 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2168 		/* Wait for completion signaled by kcryptd_async_done() */
2169 		wait_for_completion(&ctx->restart);
2170 		crypt_finished = 1;
2171 	}
2172 
2173 	/* Encryption was already finished, submit io now */
2174 	if (crypt_finished) {
2175 		kcryptd_crypt_write_io_submit(io, 0);
2176 		io->sector = sector;
2177 	}
2178 
2179 dec:
2180 	crypt_dec_pending(io);
2181 }
2182 
2183 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2184 {
2185 	if (io->ctx.aead_recheck) {
2186 		if (!io->error) {
2187 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2188 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2189 		}
2190 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2191 		bio_put(io->ctx.bio_in);
2192 	}
2193 	crypt_dec_pending(io);
2194 }
2195 
2196 static void kcryptd_crypt_read_continue(struct work_struct *work)
2197 {
2198 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2199 	struct crypt_config *cc = io->cc;
2200 	blk_status_t r;
2201 
2202 	wait_for_completion(&io->ctx.restart);
2203 	reinit_completion(&io->ctx.restart);
2204 
2205 	r = crypt_convert(cc, &io->ctx, true, false);
2206 	if (r)
2207 		io->error = r;
2208 
2209 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2210 		kcryptd_crypt_read_done(io);
2211 
2212 	crypt_dec_pending(io);
2213 }
2214 
2215 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2216 {
2217 	struct crypt_config *cc = io->cc;
2218 	blk_status_t r;
2219 
2220 	crypt_inc_pending(io);
2221 
2222 	if (io->ctx.aead_recheck) {
2223 		io->ctx.cc_sector = io->sector + cc->iv_offset;
2224 		r = crypt_convert(cc, &io->ctx,
2225 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2226 	} else {
2227 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2228 				   io->sector);
2229 
2230 		r = crypt_convert(cc, &io->ctx,
2231 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2232 	}
2233 	/*
2234 	 * Crypto API backlogged the request, because its queue was full
2235 	 * and we're in softirq context, so continue from a workqueue
2236 	 */
2237 	if (r == BLK_STS_DEV_RESOURCE) {
2238 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2239 		queue_work(cc->crypt_queue, &io->work);
2240 		return;
2241 	}
2242 	if (r)
2243 		io->error = r;
2244 
2245 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2246 		kcryptd_crypt_read_done(io);
2247 
2248 	crypt_dec_pending(io);
2249 }
2250 
2251 static void kcryptd_async_done(void *data, int error)
2252 {
2253 	struct dm_crypt_request *dmreq = data;
2254 	struct convert_context *ctx = dmreq->ctx;
2255 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2256 	struct crypt_config *cc = io->cc;
2257 
2258 	/*
2259 	 * A request from crypto driver backlog is going to be processed now,
2260 	 * finish the completion and continue in crypt_convert().
2261 	 * (Callback will be called for the second time for this request.)
2262 	 */
2263 	if (error == -EINPROGRESS) {
2264 		complete(&ctx->restart);
2265 		return;
2266 	}
2267 
2268 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2269 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2270 
2271 	if (error == -EBADMSG) {
2272 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2273 
2274 		ctx->aead_failed = true;
2275 		if (ctx->aead_recheck) {
2276 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2277 				    ctx->bio_in->bi_bdev, s);
2278 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2279 					 ctx->bio_in, s, 0);
2280 		}
2281 		io->error = BLK_STS_PROTECTION;
2282 	} else if (error < 0)
2283 		io->error = BLK_STS_IOERR;
2284 
2285 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2286 
2287 	if (!atomic_dec_and_test(&ctx->cc_pending))
2288 		return;
2289 
2290 	/*
2291 	 * The request is fully completed: for inline writes, let
2292 	 * kcryptd_crypt_write_convert() do the IO submission.
2293 	 */
2294 	if (bio_data_dir(io->base_bio) == READ) {
2295 		kcryptd_crypt_read_done(io);
2296 		return;
2297 	}
2298 
2299 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2300 		complete(&ctx->restart);
2301 		return;
2302 	}
2303 
2304 	kcryptd_crypt_write_io_submit(io, 1);
2305 }
2306 
2307 static void kcryptd_crypt(struct work_struct *work)
2308 {
2309 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2310 
2311 	if (bio_data_dir(io->base_bio) == READ)
2312 		kcryptd_crypt_read_convert(io);
2313 	else
2314 		kcryptd_crypt_write_convert(io);
2315 }
2316 
2317 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2318 {
2319 	struct crypt_config *cc = io->cc;
2320 
2321 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2322 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2323 		/*
2324 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2325 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2326 		 * it is being executed with irqs disabled.
2327 		 */
2328 		if (in_hardirq() || irqs_disabled()) {
2329 			INIT_WORK(&io->work, kcryptd_crypt);
2330 			queue_work(system_bh_wq, &io->work);
2331 			return;
2332 		} else {
2333 			kcryptd_crypt(&io->work);
2334 			return;
2335 		}
2336 	}
2337 
2338 	INIT_WORK(&io->work, kcryptd_crypt);
2339 	queue_work(cc->crypt_queue, &io->work);
2340 }
2341 
2342 static void crypt_free_tfms_aead(struct crypt_config *cc)
2343 {
2344 	if (!cc->cipher_tfm.tfms_aead)
2345 		return;
2346 
2347 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2348 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2349 		cc->cipher_tfm.tfms_aead[0] = NULL;
2350 	}
2351 
2352 	kfree(cc->cipher_tfm.tfms_aead);
2353 	cc->cipher_tfm.tfms_aead = NULL;
2354 }
2355 
2356 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2357 {
2358 	unsigned int i;
2359 
2360 	if (!cc->cipher_tfm.tfms)
2361 		return;
2362 
2363 	for (i = 0; i < cc->tfms_count; i++)
2364 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2365 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2366 			cc->cipher_tfm.tfms[i] = NULL;
2367 		}
2368 
2369 	kfree(cc->cipher_tfm.tfms);
2370 	cc->cipher_tfm.tfms = NULL;
2371 }
2372 
2373 static void crypt_free_tfms(struct crypt_config *cc)
2374 {
2375 	if (crypt_integrity_aead(cc))
2376 		crypt_free_tfms_aead(cc);
2377 	else
2378 		crypt_free_tfms_skcipher(cc);
2379 }
2380 
2381 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2382 {
2383 	unsigned int i;
2384 	int err;
2385 
2386 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2387 				      sizeof(struct crypto_skcipher *),
2388 				      GFP_KERNEL);
2389 	if (!cc->cipher_tfm.tfms)
2390 		return -ENOMEM;
2391 
2392 	for (i = 0; i < cc->tfms_count; i++) {
2393 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2394 						CRYPTO_ALG_ALLOCATES_MEMORY);
2395 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2396 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2397 			crypt_free_tfms(cc);
2398 			return err;
2399 		}
2400 	}
2401 
2402 	/*
2403 	 * dm-crypt performance can vary greatly depending on which crypto
2404 	 * algorithm implementation is used.  Help people debug performance
2405 	 * problems by logging the ->cra_driver_name.
2406 	 */
2407 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2408 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2409 	return 0;
2410 }
2411 
2412 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2413 {
2414 	int err;
2415 
2416 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2417 	if (!cc->cipher_tfm.tfms)
2418 		return -ENOMEM;
2419 
2420 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2421 						CRYPTO_ALG_ALLOCATES_MEMORY);
2422 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2423 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2424 		crypt_free_tfms(cc);
2425 		return err;
2426 	}
2427 
2428 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2429 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2430 	return 0;
2431 }
2432 
2433 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2434 {
2435 	if (crypt_integrity_aead(cc))
2436 		return crypt_alloc_tfms_aead(cc, ciphermode);
2437 	else
2438 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2439 }
2440 
2441 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2442 {
2443 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2444 }
2445 
2446 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2447 {
2448 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2449 }
2450 
2451 /*
2452  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2453  * the key must be for some reason in special format.
2454  * This funcion converts cc->key to this special format.
2455  */
2456 static void crypt_copy_authenckey(char *p, const void *key,
2457 				  unsigned int enckeylen, unsigned int authkeylen)
2458 {
2459 	struct crypto_authenc_key_param *param;
2460 	struct rtattr *rta;
2461 
2462 	rta = (struct rtattr *)p;
2463 	param = RTA_DATA(rta);
2464 	param->enckeylen = cpu_to_be32(enckeylen);
2465 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2466 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2467 	p += RTA_SPACE(sizeof(*param));
2468 	memcpy(p, key + enckeylen, authkeylen);
2469 	p += authkeylen;
2470 	memcpy(p, key, enckeylen);
2471 }
2472 
2473 static int crypt_setkey(struct crypt_config *cc)
2474 {
2475 	unsigned int subkey_size;
2476 	int err = 0, i, r;
2477 
2478 	/* Ignore extra keys (which are used for IV etc) */
2479 	subkey_size = crypt_subkey_size(cc);
2480 
2481 	if (crypt_integrity_hmac(cc)) {
2482 		if (subkey_size < cc->key_mac_size)
2483 			return -EINVAL;
2484 
2485 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2486 				      subkey_size - cc->key_mac_size,
2487 				      cc->key_mac_size);
2488 	}
2489 
2490 	for (i = 0; i < cc->tfms_count; i++) {
2491 		if (crypt_integrity_hmac(cc))
2492 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2493 				cc->authenc_key, crypt_authenckey_size(cc));
2494 		else if (crypt_integrity_aead(cc))
2495 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2496 					       cc->key + (i * subkey_size),
2497 					       subkey_size);
2498 		else
2499 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2500 						   cc->key + (i * subkey_size),
2501 						   subkey_size);
2502 		if (r)
2503 			err = r;
2504 	}
2505 
2506 	if (crypt_integrity_hmac(cc))
2507 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2508 
2509 	return err;
2510 }
2511 
2512 #ifdef CONFIG_KEYS
2513 
2514 static bool contains_whitespace(const char *str)
2515 {
2516 	while (*str)
2517 		if (isspace(*str++))
2518 			return true;
2519 	return false;
2520 }
2521 
2522 static int set_key_user(struct crypt_config *cc, struct key *key)
2523 {
2524 	const struct user_key_payload *ukp;
2525 
2526 	ukp = user_key_payload_locked(key);
2527 	if (!ukp)
2528 		return -EKEYREVOKED;
2529 
2530 	if (cc->key_size != ukp->datalen)
2531 		return -EINVAL;
2532 
2533 	memcpy(cc->key, ukp->data, cc->key_size);
2534 
2535 	return 0;
2536 }
2537 
2538 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2539 {
2540 	const struct encrypted_key_payload *ekp;
2541 
2542 	ekp = key->payload.data[0];
2543 	if (!ekp)
2544 		return -EKEYREVOKED;
2545 
2546 	if (cc->key_size != ekp->decrypted_datalen)
2547 		return -EINVAL;
2548 
2549 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2550 
2551 	return 0;
2552 }
2553 
2554 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2555 {
2556 	const struct trusted_key_payload *tkp;
2557 
2558 	tkp = key->payload.data[0];
2559 	if (!tkp)
2560 		return -EKEYREVOKED;
2561 
2562 	if (cc->key_size != tkp->key_len)
2563 		return -EINVAL;
2564 
2565 	memcpy(cc->key, tkp->key, cc->key_size);
2566 
2567 	return 0;
2568 }
2569 
2570 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2571 {
2572 	char *new_key_string, *key_desc;
2573 	int ret;
2574 	struct key_type *type;
2575 	struct key *key;
2576 	int (*set_key)(struct crypt_config *cc, struct key *key);
2577 
2578 	/*
2579 	 * Reject key_string with whitespace. dm core currently lacks code for
2580 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2581 	 */
2582 	if (contains_whitespace(key_string)) {
2583 		DMERR("whitespace chars not allowed in key string");
2584 		return -EINVAL;
2585 	}
2586 
2587 	/* look for next ':' separating key_type from key_description */
2588 	key_desc = strchr(key_string, ':');
2589 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2590 		return -EINVAL;
2591 
2592 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2593 		type = &key_type_logon;
2594 		set_key = set_key_user;
2595 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2596 		type = &key_type_user;
2597 		set_key = set_key_user;
2598 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2599 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2600 		type = &key_type_encrypted;
2601 		set_key = set_key_encrypted;
2602 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2603 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2604 		type = &key_type_trusted;
2605 		set_key = set_key_trusted;
2606 	} else {
2607 		return -EINVAL;
2608 	}
2609 
2610 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2611 	if (!new_key_string)
2612 		return -ENOMEM;
2613 
2614 	key = request_key(type, key_desc + 1, NULL);
2615 	if (IS_ERR(key)) {
2616 		kfree_sensitive(new_key_string);
2617 		return PTR_ERR(key);
2618 	}
2619 
2620 	down_read(&key->sem);
2621 
2622 	ret = set_key(cc, key);
2623 	if (ret < 0) {
2624 		up_read(&key->sem);
2625 		key_put(key);
2626 		kfree_sensitive(new_key_string);
2627 		return ret;
2628 	}
2629 
2630 	up_read(&key->sem);
2631 	key_put(key);
2632 
2633 	/* clear the flag since following operations may invalidate previously valid key */
2634 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2635 
2636 	ret = crypt_setkey(cc);
2637 
2638 	if (!ret) {
2639 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2640 		kfree_sensitive(cc->key_string);
2641 		cc->key_string = new_key_string;
2642 	} else
2643 		kfree_sensitive(new_key_string);
2644 
2645 	return ret;
2646 }
2647 
2648 static int get_key_size(char **key_string)
2649 {
2650 	char *colon, dummy;
2651 	int ret;
2652 
2653 	if (*key_string[0] != ':')
2654 		return strlen(*key_string) >> 1;
2655 
2656 	/* look for next ':' in key string */
2657 	colon = strpbrk(*key_string + 1, ":");
2658 	if (!colon)
2659 		return -EINVAL;
2660 
2661 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2662 		return -EINVAL;
2663 
2664 	*key_string = colon;
2665 
2666 	/* remaining key string should be :<logon|user>:<key_desc> */
2667 
2668 	return ret;
2669 }
2670 
2671 #else
2672 
2673 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2674 {
2675 	return -EINVAL;
2676 }
2677 
2678 static int get_key_size(char **key_string)
2679 {
2680 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2681 }
2682 
2683 #endif /* CONFIG_KEYS */
2684 
2685 static int crypt_set_key(struct crypt_config *cc, char *key)
2686 {
2687 	int r = -EINVAL;
2688 	int key_string_len = strlen(key);
2689 
2690 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2691 	if (!cc->key_size && strcmp(key, "-"))
2692 		goto out;
2693 
2694 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2695 	if (key[0] == ':') {
2696 		r = crypt_set_keyring_key(cc, key + 1);
2697 		goto out;
2698 	}
2699 
2700 	/* clear the flag since following operations may invalidate previously valid key */
2701 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2702 
2703 	/* wipe references to any kernel keyring key */
2704 	kfree_sensitive(cc->key_string);
2705 	cc->key_string = NULL;
2706 
2707 	/* Decode key from its hex representation. */
2708 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2709 		goto out;
2710 
2711 	r = crypt_setkey(cc);
2712 	if (!r)
2713 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2714 
2715 out:
2716 	/* Hex key string not needed after here, so wipe it. */
2717 	memset(key, '0', key_string_len);
2718 
2719 	return r;
2720 }
2721 
2722 static int crypt_wipe_key(struct crypt_config *cc)
2723 {
2724 	int r;
2725 
2726 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2727 	get_random_bytes(&cc->key, cc->key_size);
2728 
2729 	/* Wipe IV private keys */
2730 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2731 		r = cc->iv_gen_ops->wipe(cc);
2732 		if (r)
2733 			return r;
2734 	}
2735 
2736 	kfree_sensitive(cc->key_string);
2737 	cc->key_string = NULL;
2738 	r = crypt_setkey(cc);
2739 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2740 
2741 	return r;
2742 }
2743 
2744 static void crypt_calculate_pages_per_client(void)
2745 {
2746 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2747 
2748 	if (!dm_crypt_clients_n)
2749 		return;
2750 
2751 	pages /= dm_crypt_clients_n;
2752 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2753 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2754 	dm_crypt_pages_per_client = pages;
2755 }
2756 
2757 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2758 {
2759 	struct crypt_config *cc = pool_data;
2760 	struct page *page;
2761 
2762 	/*
2763 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2764 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2765 	 * but avoids potential spinlock contention of an exact result.
2766 	 */
2767 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2768 	    likely(gfp_mask & __GFP_NORETRY))
2769 		return NULL;
2770 
2771 	page = alloc_page(gfp_mask);
2772 	if (likely(page != NULL))
2773 		percpu_counter_add(&cc->n_allocated_pages, 1);
2774 
2775 	return page;
2776 }
2777 
2778 static void crypt_page_free(void *page, void *pool_data)
2779 {
2780 	struct crypt_config *cc = pool_data;
2781 
2782 	__free_page(page);
2783 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2784 }
2785 
2786 static void crypt_dtr(struct dm_target *ti)
2787 {
2788 	struct crypt_config *cc = ti->private;
2789 
2790 	ti->private = NULL;
2791 
2792 	if (!cc)
2793 		return;
2794 
2795 	if (cc->write_thread)
2796 		kthread_stop(cc->write_thread);
2797 
2798 	if (cc->io_queue)
2799 		destroy_workqueue(cc->io_queue);
2800 	if (cc->crypt_queue)
2801 		destroy_workqueue(cc->crypt_queue);
2802 
2803 	if (cc->workqueue_id)
2804 		ida_free(&workqueue_ida, cc->workqueue_id);
2805 
2806 	crypt_free_tfms(cc);
2807 
2808 	bioset_exit(&cc->bs);
2809 
2810 	mempool_exit(&cc->page_pool);
2811 	mempool_exit(&cc->req_pool);
2812 	mempool_exit(&cc->tag_pool);
2813 
2814 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2815 	percpu_counter_destroy(&cc->n_allocated_pages);
2816 
2817 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2818 		cc->iv_gen_ops->dtr(cc);
2819 
2820 	if (cc->dev)
2821 		dm_put_device(ti, cc->dev);
2822 
2823 	kfree_sensitive(cc->cipher_string);
2824 	kfree_sensitive(cc->key_string);
2825 	kfree_sensitive(cc->cipher_auth);
2826 	kfree_sensitive(cc->authenc_key);
2827 
2828 	mutex_destroy(&cc->bio_alloc_lock);
2829 
2830 	/* Must zero key material before freeing */
2831 	kfree_sensitive(cc);
2832 
2833 	spin_lock(&dm_crypt_clients_lock);
2834 	WARN_ON(!dm_crypt_clients_n);
2835 	dm_crypt_clients_n--;
2836 	crypt_calculate_pages_per_client();
2837 	spin_unlock(&dm_crypt_clients_lock);
2838 
2839 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2840 }
2841 
2842 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2843 {
2844 	struct crypt_config *cc = ti->private;
2845 
2846 	if (crypt_integrity_aead(cc))
2847 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2848 	else
2849 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2850 
2851 	if (cc->iv_size)
2852 		/* at least a 64 bit sector number should fit in our buffer */
2853 		cc->iv_size = max(cc->iv_size,
2854 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2855 	else if (ivmode) {
2856 		DMWARN("Selected cipher does not support IVs");
2857 		ivmode = NULL;
2858 	}
2859 
2860 	/* Choose ivmode, see comments at iv code. */
2861 	if (ivmode == NULL)
2862 		cc->iv_gen_ops = NULL;
2863 	else if (strcmp(ivmode, "plain") == 0)
2864 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2865 	else if (strcmp(ivmode, "plain64") == 0)
2866 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2867 	else if (strcmp(ivmode, "plain64be") == 0)
2868 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2869 	else if (strcmp(ivmode, "essiv") == 0)
2870 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2871 	else if (strcmp(ivmode, "benbi") == 0)
2872 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2873 	else if (strcmp(ivmode, "null") == 0)
2874 		cc->iv_gen_ops = &crypt_iv_null_ops;
2875 	else if (strcmp(ivmode, "eboiv") == 0)
2876 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2877 	else if (strcmp(ivmode, "elephant") == 0) {
2878 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2879 		cc->key_parts = 2;
2880 		cc->key_extra_size = cc->key_size / 2;
2881 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2882 			return -EINVAL;
2883 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2884 	} else if (strcmp(ivmode, "lmk") == 0) {
2885 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2886 		/*
2887 		 * Version 2 and 3 is recognised according
2888 		 * to length of provided multi-key string.
2889 		 * If present (version 3), last key is used as IV seed.
2890 		 * All keys (including IV seed) are always the same size.
2891 		 */
2892 		if (cc->key_size % cc->key_parts) {
2893 			cc->key_parts++;
2894 			cc->key_extra_size = cc->key_size / cc->key_parts;
2895 		}
2896 	} else if (strcmp(ivmode, "tcw") == 0) {
2897 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2898 		cc->key_parts += 2; /* IV + whitening */
2899 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2900 	} else if (strcmp(ivmode, "random") == 0) {
2901 		cc->iv_gen_ops = &crypt_iv_random_ops;
2902 		/* Need storage space in integrity fields. */
2903 		cc->integrity_iv_size = cc->iv_size;
2904 	} else {
2905 		ti->error = "Invalid IV mode";
2906 		return -EINVAL;
2907 	}
2908 
2909 	return 0;
2910 }
2911 
2912 /*
2913  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2914  * The HMAC is needed to calculate tag size (HMAC digest size).
2915  * This should be probably done by crypto-api calls (once available...)
2916  */
2917 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2918 {
2919 	char *start, *end, *mac_alg = NULL;
2920 	struct crypto_ahash *mac;
2921 
2922 	if (!strstarts(cipher_api, "authenc("))
2923 		return 0;
2924 
2925 	start = strchr(cipher_api, '(');
2926 	end = strchr(cipher_api, ',');
2927 	if (!start || !end || ++start > end)
2928 		return -EINVAL;
2929 
2930 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2931 	if (!mac_alg)
2932 		return -ENOMEM;
2933 
2934 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2935 	kfree(mac_alg);
2936 
2937 	if (IS_ERR(mac))
2938 		return PTR_ERR(mac);
2939 
2940 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2941 	crypto_free_ahash(mac);
2942 
2943 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2944 	if (!cc->authenc_key)
2945 		return -ENOMEM;
2946 
2947 	return 0;
2948 }
2949 
2950 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2951 				char **ivmode, char **ivopts)
2952 {
2953 	struct crypt_config *cc = ti->private;
2954 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2955 	int ret = -EINVAL;
2956 
2957 	cc->tfms_count = 1;
2958 
2959 	/*
2960 	 * New format (capi: prefix)
2961 	 * capi:cipher_api_spec-iv:ivopts
2962 	 */
2963 	tmp = &cipher_in[strlen("capi:")];
2964 
2965 	/* Separate IV options if present, it can contain another '-' in hash name */
2966 	*ivopts = strrchr(tmp, ':');
2967 	if (*ivopts) {
2968 		**ivopts = '\0';
2969 		(*ivopts)++;
2970 	}
2971 	/* Parse IV mode */
2972 	*ivmode = strrchr(tmp, '-');
2973 	if (*ivmode) {
2974 		**ivmode = '\0';
2975 		(*ivmode)++;
2976 	}
2977 	/* The rest is crypto API spec */
2978 	cipher_api = tmp;
2979 
2980 	/* Alloc AEAD, can be used only in new format. */
2981 	if (crypt_integrity_aead(cc)) {
2982 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2983 		if (ret < 0) {
2984 			ti->error = "Invalid AEAD cipher spec";
2985 			return ret;
2986 		}
2987 	}
2988 
2989 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2990 		cc->tfms_count = 64;
2991 
2992 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2993 		if (!*ivopts) {
2994 			ti->error = "Digest algorithm missing for ESSIV mode";
2995 			return -EINVAL;
2996 		}
2997 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2998 			       cipher_api, *ivopts);
2999 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3000 			ti->error = "Cannot allocate cipher string";
3001 			return -ENOMEM;
3002 		}
3003 		cipher_api = buf;
3004 	}
3005 
3006 	cc->key_parts = cc->tfms_count;
3007 
3008 	/* Allocate cipher */
3009 	ret = crypt_alloc_tfms(cc, cipher_api);
3010 	if (ret < 0) {
3011 		ti->error = "Error allocating crypto tfm";
3012 		return ret;
3013 	}
3014 
3015 	if (crypt_integrity_aead(cc))
3016 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
3017 	else
3018 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
3019 
3020 	return 0;
3021 }
3022 
3023 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
3024 				char **ivmode, char **ivopts)
3025 {
3026 	struct crypt_config *cc = ti->private;
3027 	char *tmp, *cipher, *chainmode, *keycount;
3028 	char *cipher_api = NULL;
3029 	int ret = -EINVAL;
3030 	char dummy;
3031 
3032 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
3033 		ti->error = "Bad cipher specification";
3034 		return -EINVAL;
3035 	}
3036 
3037 	/*
3038 	 * Legacy dm-crypt cipher specification
3039 	 * cipher[:keycount]-mode-iv:ivopts
3040 	 */
3041 	tmp = cipher_in;
3042 	keycount = strsep(&tmp, "-");
3043 	cipher = strsep(&keycount, ":");
3044 
3045 	if (!keycount)
3046 		cc->tfms_count = 1;
3047 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3048 		 !is_power_of_2(cc->tfms_count)) {
3049 		ti->error = "Bad cipher key count specification";
3050 		return -EINVAL;
3051 	}
3052 	cc->key_parts = cc->tfms_count;
3053 
3054 	chainmode = strsep(&tmp, "-");
3055 	*ivmode = strsep(&tmp, ":");
3056 	*ivopts = tmp;
3057 
3058 	/*
3059 	 * For compatibility with the original dm-crypt mapping format, if
3060 	 * only the cipher name is supplied, use cbc-plain.
3061 	 */
3062 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3063 		chainmode = "cbc";
3064 		*ivmode = "plain";
3065 	}
3066 
3067 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3068 		ti->error = "IV mechanism required";
3069 		return -EINVAL;
3070 	}
3071 
3072 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3073 	if (!cipher_api)
3074 		goto bad_mem;
3075 
3076 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3077 		if (!*ivopts) {
3078 			ti->error = "Digest algorithm missing for ESSIV mode";
3079 			kfree(cipher_api);
3080 			return -EINVAL;
3081 		}
3082 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3083 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3084 	} else {
3085 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3086 			       "%s(%s)", chainmode, cipher);
3087 	}
3088 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3089 		kfree(cipher_api);
3090 		goto bad_mem;
3091 	}
3092 
3093 	/* Allocate cipher */
3094 	ret = crypt_alloc_tfms(cc, cipher_api);
3095 	if (ret < 0) {
3096 		ti->error = "Error allocating crypto tfm";
3097 		kfree(cipher_api);
3098 		return ret;
3099 	}
3100 	kfree(cipher_api);
3101 
3102 	return 0;
3103 bad_mem:
3104 	ti->error = "Cannot allocate cipher strings";
3105 	return -ENOMEM;
3106 }
3107 
3108 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3109 {
3110 	struct crypt_config *cc = ti->private;
3111 	char *ivmode = NULL, *ivopts = NULL;
3112 	int ret;
3113 
3114 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3115 	if (!cc->cipher_string) {
3116 		ti->error = "Cannot allocate cipher strings";
3117 		return -ENOMEM;
3118 	}
3119 
3120 	if (strstarts(cipher_in, "capi:"))
3121 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3122 	else
3123 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3124 	if (ret)
3125 		return ret;
3126 
3127 	/* Initialize IV */
3128 	ret = crypt_ctr_ivmode(ti, ivmode);
3129 	if (ret < 0)
3130 		return ret;
3131 
3132 	/* Initialize and set key */
3133 	ret = crypt_set_key(cc, key);
3134 	if (ret < 0) {
3135 		ti->error = "Error decoding and setting key";
3136 		return ret;
3137 	}
3138 
3139 	/* Allocate IV */
3140 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3141 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3142 		if (ret < 0) {
3143 			ti->error = "Error creating IV";
3144 			return ret;
3145 		}
3146 	}
3147 
3148 	/* Initialize IV (set keys for ESSIV etc) */
3149 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3150 		ret = cc->iv_gen_ops->init(cc);
3151 		if (ret < 0) {
3152 			ti->error = "Error initialising IV";
3153 			return ret;
3154 		}
3155 	}
3156 
3157 	/* wipe the kernel key payload copy */
3158 	if (cc->key_string)
3159 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3160 
3161 	return ret;
3162 }
3163 
3164 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3165 {
3166 	struct crypt_config *cc = ti->private;
3167 	struct dm_arg_set as;
3168 	static const struct dm_arg _args[] = {
3169 		{0, 9, "Invalid number of feature args"},
3170 	};
3171 	unsigned int opt_params, val;
3172 	const char *opt_string, *sval;
3173 	char dummy;
3174 	int ret;
3175 
3176 	/* Optional parameters */
3177 	as.argc = argc;
3178 	as.argv = argv;
3179 
3180 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3181 	if (ret)
3182 		return ret;
3183 
3184 	while (opt_params--) {
3185 		opt_string = dm_shift_arg(&as);
3186 		if (!opt_string) {
3187 			ti->error = "Not enough feature arguments";
3188 			return -EINVAL;
3189 		}
3190 
3191 		if (!strcasecmp(opt_string, "allow_discards"))
3192 			ti->num_discard_bios = 1;
3193 
3194 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3195 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3196 		else if (!strcasecmp(opt_string, "high_priority"))
3197 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3198 
3199 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3200 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3201 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3202 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3203 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3204 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3205 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3206 			if (val == 0 || val > MAX_TAG_SIZE) {
3207 				ti->error = "Invalid integrity arguments";
3208 				return -EINVAL;
3209 			}
3210 			cc->used_tag_size = val;
3211 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3212 			if (!strcasecmp(sval, "aead")) {
3213 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3214 			} else if (strcasecmp(sval, "none")) {
3215 				ti->error = "Unknown integrity profile";
3216 				return -EINVAL;
3217 			}
3218 
3219 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3220 			if (!cc->cipher_auth)
3221 				return -ENOMEM;
3222 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3223 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3224 			    cc->sector_size > 4096 ||
3225 			    (cc->sector_size & (cc->sector_size - 1))) {
3226 				ti->error = "Invalid feature value for sector_size";
3227 				return -EINVAL;
3228 			}
3229 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3230 				ti->error = "Device size is not multiple of sector_size feature";
3231 				return -EINVAL;
3232 			}
3233 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3234 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3235 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3236 		else {
3237 			ti->error = "Invalid feature arguments";
3238 			return -EINVAL;
3239 		}
3240 	}
3241 
3242 	return 0;
3243 }
3244 
3245 #ifdef CONFIG_BLK_DEV_ZONED
3246 static int crypt_report_zones(struct dm_target *ti,
3247 		struct dm_report_zones_args *args, unsigned int nr_zones)
3248 {
3249 	struct crypt_config *cc = ti->private;
3250 
3251 	return dm_report_zones(cc->dev->bdev, cc->start,
3252 			cc->start + dm_target_offset(ti, args->next_sector),
3253 			args, nr_zones);
3254 }
3255 #else
3256 #define crypt_report_zones NULL
3257 #endif
3258 
3259 /*
3260  * Construct an encryption mapping:
3261  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3262  */
3263 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3264 {
3265 	struct crypt_config *cc;
3266 	const char *devname = dm_table_device_name(ti->table);
3267 	int key_size, wq_id;
3268 	unsigned int align_mask;
3269 	unsigned int common_wq_flags;
3270 	unsigned long long tmpll;
3271 	int ret;
3272 	size_t iv_size_padding, additional_req_size;
3273 	char dummy;
3274 
3275 	if (argc < 5) {
3276 		ti->error = "Not enough arguments";
3277 		return -EINVAL;
3278 	}
3279 
3280 	key_size = get_key_size(&argv[1]);
3281 	if (key_size < 0) {
3282 		ti->error = "Cannot parse key size";
3283 		return -EINVAL;
3284 	}
3285 
3286 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3287 	if (!cc) {
3288 		ti->error = "Cannot allocate encryption context";
3289 		return -ENOMEM;
3290 	}
3291 	cc->key_size = key_size;
3292 	cc->sector_size = (1 << SECTOR_SHIFT);
3293 	cc->sector_shift = 0;
3294 
3295 	ti->private = cc;
3296 
3297 	spin_lock(&dm_crypt_clients_lock);
3298 	dm_crypt_clients_n++;
3299 	crypt_calculate_pages_per_client();
3300 	spin_unlock(&dm_crypt_clients_lock);
3301 
3302 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3303 	if (ret < 0)
3304 		goto bad;
3305 
3306 	/* Optional parameters need to be read before cipher constructor */
3307 	if (argc > 5) {
3308 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3309 		if (ret)
3310 			goto bad;
3311 	}
3312 
3313 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3314 	if (ret < 0)
3315 		goto bad;
3316 
3317 	if (crypt_integrity_aead(cc)) {
3318 		cc->dmreq_start = sizeof(struct aead_request);
3319 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3320 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3321 	} else {
3322 		cc->dmreq_start = sizeof(struct skcipher_request);
3323 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3324 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3325 	}
3326 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3327 
3328 	if (align_mask < CRYPTO_MINALIGN) {
3329 		/* Allocate the padding exactly */
3330 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3331 				& align_mask;
3332 	} else {
3333 		/*
3334 		 * If the cipher requires greater alignment than kmalloc
3335 		 * alignment, we don't know the exact position of the
3336 		 * initialization vector. We must assume worst case.
3337 		 */
3338 		iv_size_padding = align_mask;
3339 	}
3340 
3341 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3342 	additional_req_size = sizeof(struct dm_crypt_request) +
3343 		iv_size_padding + cc->iv_size +
3344 		cc->iv_size +
3345 		sizeof(uint64_t) +
3346 		sizeof(unsigned int);
3347 
3348 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3349 	if (ret) {
3350 		ti->error = "Cannot allocate crypt request mempool";
3351 		goto bad;
3352 	}
3353 
3354 	cc->per_bio_data_size = ti->per_io_data_size =
3355 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3356 		      ARCH_DMA_MINALIGN);
3357 
3358 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3359 	if (ret) {
3360 		ti->error = "Cannot allocate page mempool";
3361 		goto bad;
3362 	}
3363 
3364 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3365 	if (ret) {
3366 		ti->error = "Cannot allocate crypt bioset";
3367 		goto bad;
3368 	}
3369 
3370 	mutex_init(&cc->bio_alloc_lock);
3371 
3372 	ret = -EINVAL;
3373 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3374 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3375 		ti->error = "Invalid iv_offset sector";
3376 		goto bad;
3377 	}
3378 	cc->iv_offset = tmpll;
3379 
3380 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3381 	if (ret) {
3382 		ti->error = "Device lookup failed";
3383 		goto bad;
3384 	}
3385 
3386 	ret = -EINVAL;
3387 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3388 		ti->error = "Invalid device sector";
3389 		goto bad;
3390 	}
3391 	cc->start = tmpll;
3392 
3393 	if (bdev_is_zoned(cc->dev->bdev)) {
3394 		/*
3395 		 * For zoned block devices, we need to preserve the issuer write
3396 		 * ordering. To do so, disable write workqueues and force inline
3397 		 * encryption completion.
3398 		 */
3399 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3400 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3401 
3402 		/*
3403 		 * All zone append writes to a zone of a zoned block device will
3404 		 * have the same BIO sector, the start of the zone. When the
3405 		 * cypher IV mode uses sector values, all data targeting a
3406 		 * zone will be encrypted using the first sector numbers of the
3407 		 * zone. This will not result in write errors but will
3408 		 * cause most reads to fail as reads will use the sector values
3409 		 * for the actual data locations, resulting in IV mismatch.
3410 		 * To avoid this problem, ask DM core to emulate zone append
3411 		 * operations with regular writes.
3412 		 */
3413 		DMDEBUG("Zone append operations will be emulated");
3414 		ti->emulate_zone_append = true;
3415 	}
3416 
3417 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3418 		ret = crypt_integrity_ctr(cc, ti);
3419 		if (ret)
3420 			goto bad;
3421 
3422 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3423 		if (!cc->tag_pool_max_sectors)
3424 			cc->tag_pool_max_sectors = 1;
3425 
3426 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3427 			cc->tag_pool_max_sectors * cc->tuple_size);
3428 		if (ret) {
3429 			ti->error = "Cannot allocate integrity tags mempool";
3430 			goto bad;
3431 		}
3432 
3433 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3434 	}
3435 
3436 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3437 	if (wq_id < 0) {
3438 		ti->error = "Couldn't get workqueue id";
3439 		ret = wq_id;
3440 		goto bad;
3441 	}
3442 	cc->workqueue_id = wq_id;
3443 
3444 	ret = -ENOMEM;
3445 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3446 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3447 		common_wq_flags |= WQ_HIGHPRI;
3448 
3449 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", common_wq_flags, 1, devname, wq_id);
3450 	if (!cc->io_queue) {
3451 		ti->error = "Couldn't create kcryptd io queue";
3452 		goto bad;
3453 	}
3454 
3455 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3456 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3457 						  common_wq_flags | WQ_CPU_INTENSIVE,
3458 						  1, devname, wq_id);
3459 	} else {
3460 		/*
3461 		 * While crypt_queue is certainly CPU intensive, the use of
3462 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3463 		 */
3464 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3465 						  common_wq_flags | WQ_UNBOUND,
3466 						  num_online_cpus(), devname, wq_id);
3467 	}
3468 	if (!cc->crypt_queue) {
3469 		ti->error = "Couldn't create kcryptd queue";
3470 		goto bad;
3471 	}
3472 
3473 	spin_lock_init(&cc->write_thread_lock);
3474 	cc->write_tree = RB_ROOT;
3475 
3476 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3477 	if (IS_ERR(cc->write_thread)) {
3478 		ret = PTR_ERR(cc->write_thread);
3479 		cc->write_thread = NULL;
3480 		ti->error = "Couldn't spawn write thread";
3481 		goto bad;
3482 	}
3483 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3484 		set_user_nice(cc->write_thread, MIN_NICE);
3485 
3486 	ti->num_flush_bios = 1;
3487 	ti->limit_swap_bios = true;
3488 	ti->accounts_remapped_io = true;
3489 
3490 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3491 	return 0;
3492 
3493 bad:
3494 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3495 	crypt_dtr(ti);
3496 	return ret;
3497 }
3498 
3499 static int crypt_map(struct dm_target *ti, struct bio *bio)
3500 {
3501 	struct dm_crypt_io *io;
3502 	struct crypt_config *cc = ti->private;
3503 	unsigned max_sectors;
3504 
3505 	/*
3506 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3507 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3508 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3509 	 */
3510 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3511 	    bio_op(bio) == REQ_OP_DISCARD)) {
3512 		bio_set_dev(bio, cc->dev->bdev);
3513 		if (bio_sectors(bio))
3514 			bio->bi_iter.bi_sector = cc->start +
3515 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3516 		return DM_MAPIO_REMAPPED;
3517 	}
3518 
3519 	/*
3520 	 * Check if bio is too large, split as needed.
3521 	 */
3522 	max_sectors = get_max_request_size(cc, bio_data_dir(bio) == WRITE);
3523 	if (unlikely(bio_sectors(bio) > max_sectors))
3524 		dm_accept_partial_bio(bio, max_sectors);
3525 
3526 	/*
3527 	 * Ensure that bio is a multiple of internal sector encryption size
3528 	 * and is aligned to this size as defined in IO hints.
3529 	 */
3530 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3531 		return DM_MAPIO_KILL;
3532 
3533 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3534 		return DM_MAPIO_KILL;
3535 
3536 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3537 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3538 
3539 	if (cc->tuple_size) {
3540 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3541 
3542 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3543 			io->integrity_metadata = NULL;
3544 		else
3545 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3546 
3547 		if (unlikely(!io->integrity_metadata)) {
3548 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3549 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3550 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3551 			io->integrity_metadata_from_pool = true;
3552 		}
3553 	}
3554 
3555 	if (crypt_integrity_aead(cc))
3556 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3557 	else
3558 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3559 
3560 	if (bio_data_dir(io->base_bio) == READ) {
3561 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3562 			kcryptd_queue_read(io);
3563 	} else
3564 		kcryptd_queue_crypt(io);
3565 
3566 	return DM_MAPIO_SUBMITTED;
3567 }
3568 
3569 static char hex2asc(unsigned char c)
3570 {
3571 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3572 }
3573 
3574 static void crypt_status(struct dm_target *ti, status_type_t type,
3575 			 unsigned int status_flags, char *result, unsigned int maxlen)
3576 {
3577 	struct crypt_config *cc = ti->private;
3578 	unsigned int i, sz = 0;
3579 	int num_feature_args = 0;
3580 
3581 	switch (type) {
3582 	case STATUSTYPE_INFO:
3583 		result[0] = '\0';
3584 		break;
3585 
3586 	case STATUSTYPE_TABLE:
3587 		DMEMIT("%s ", cc->cipher_string);
3588 
3589 		if (cc->key_size > 0) {
3590 			if (cc->key_string)
3591 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3592 			else {
3593 				for (i = 0; i < cc->key_size; i++) {
3594 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3595 					       hex2asc(cc->key[i] & 0xf));
3596 				}
3597 			}
3598 		} else
3599 			DMEMIT("-");
3600 
3601 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3602 				cc->dev->name, (unsigned long long)cc->start);
3603 
3604 		num_feature_args += !!ti->num_discard_bios;
3605 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3606 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3607 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3608 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3609 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3610 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3611 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3612 		if (cc->used_tag_size)
3613 			num_feature_args++;
3614 		if (num_feature_args) {
3615 			DMEMIT(" %d", num_feature_args);
3616 			if (ti->num_discard_bios)
3617 				DMEMIT(" allow_discards");
3618 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3619 				DMEMIT(" same_cpu_crypt");
3620 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3621 				DMEMIT(" high_priority");
3622 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3623 				DMEMIT(" submit_from_crypt_cpus");
3624 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3625 				DMEMIT(" no_read_workqueue");
3626 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3627 				DMEMIT(" no_write_workqueue");
3628 			if (cc->used_tag_size)
3629 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3630 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3631 				DMEMIT(" sector_size:%d", cc->sector_size);
3632 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3633 				DMEMIT(" iv_large_sectors");
3634 		}
3635 		break;
3636 
3637 	case STATUSTYPE_IMA:
3638 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3639 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3640 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3641 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3642 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3643 		       'y' : 'n');
3644 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3645 		       'y' : 'n');
3646 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3647 		       'y' : 'n');
3648 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3649 		       'y' : 'n');
3650 
3651 		if (cc->used_tag_size)
3652 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3653 			       cc->used_tag_size, cc->cipher_auth);
3654 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3655 			DMEMIT(",sector_size=%d", cc->sector_size);
3656 		if (cc->cipher_string)
3657 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3658 
3659 		DMEMIT(",key_size=%u", cc->key_size);
3660 		DMEMIT(",key_parts=%u", cc->key_parts);
3661 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3662 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3663 		DMEMIT(";");
3664 		break;
3665 	}
3666 }
3667 
3668 static void crypt_postsuspend(struct dm_target *ti)
3669 {
3670 	struct crypt_config *cc = ti->private;
3671 
3672 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3673 }
3674 
3675 static int crypt_preresume(struct dm_target *ti)
3676 {
3677 	struct crypt_config *cc = ti->private;
3678 
3679 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3680 		DMERR("aborting resume - crypt key is not set.");
3681 		return -EAGAIN;
3682 	}
3683 
3684 	return 0;
3685 }
3686 
3687 static void crypt_resume(struct dm_target *ti)
3688 {
3689 	struct crypt_config *cc = ti->private;
3690 
3691 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3692 }
3693 
3694 /* Message interface
3695  *	key set <key>
3696  *	key wipe
3697  */
3698 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3699 			 char *result, unsigned int maxlen)
3700 {
3701 	struct crypt_config *cc = ti->private;
3702 	int key_size, ret = -EINVAL;
3703 
3704 	if (argc < 2)
3705 		goto error;
3706 
3707 	if (!strcasecmp(argv[0], "key")) {
3708 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3709 			DMWARN("not suspended during key manipulation.");
3710 			return -EINVAL;
3711 		}
3712 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3713 			/* The key size may not be changed. */
3714 			key_size = get_key_size(&argv[2]);
3715 			if (key_size < 0 || cc->key_size != key_size) {
3716 				memset(argv[2], '0', strlen(argv[2]));
3717 				return -EINVAL;
3718 			}
3719 
3720 			ret = crypt_set_key(cc, argv[2]);
3721 			if (ret)
3722 				return ret;
3723 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3724 				ret = cc->iv_gen_ops->init(cc);
3725 			/* wipe the kernel key payload copy */
3726 			if (cc->key_string)
3727 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3728 			return ret;
3729 		}
3730 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3731 			return crypt_wipe_key(cc);
3732 	}
3733 
3734 error:
3735 	DMWARN("unrecognised message received.");
3736 	return -EINVAL;
3737 }
3738 
3739 static int crypt_iterate_devices(struct dm_target *ti,
3740 				 iterate_devices_callout_fn fn, void *data)
3741 {
3742 	struct crypt_config *cc = ti->private;
3743 
3744 	return fn(ti, cc->dev, cc->start, ti->len, data);
3745 }
3746 
3747 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3748 {
3749 	struct crypt_config *cc = ti->private;
3750 
3751 	limits->logical_block_size =
3752 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3753 	limits->physical_block_size =
3754 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3755 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3756 	limits->dma_alignment = limits->logical_block_size - 1;
3757 }
3758 
3759 static struct target_type crypt_target = {
3760 	.name   = "crypt",
3761 	.version = {1, 27, 0},
3762 	.module = THIS_MODULE,
3763 	.ctr    = crypt_ctr,
3764 	.dtr    = crypt_dtr,
3765 	.features = DM_TARGET_ZONED_HM,
3766 	.report_zones = crypt_report_zones,
3767 	.map    = crypt_map,
3768 	.status = crypt_status,
3769 	.postsuspend = crypt_postsuspend,
3770 	.preresume = crypt_preresume,
3771 	.resume = crypt_resume,
3772 	.message = crypt_message,
3773 	.iterate_devices = crypt_iterate_devices,
3774 	.io_hints = crypt_io_hints,
3775 };
3776 module_dm(crypt);
3777 
3778 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3779 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3780 MODULE_LICENSE("GPL");
3781