xref: /linux/drivers/md/dm-crypt.c (revision 9629d83f05bdc7fdc21363979e96696886bd129a)
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 <linux/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 	unsigned int tag_offset;
63 	u64 cc_sector;
64 	union {
65 		struct skcipher_request *req;
66 		struct aead_request *req_aead;
67 	} r;
68 	bool aead_recheck;
69 	bool aead_failed;
70 
71 };
72 
73 /*
74  * per bio private data
75  */
76 struct dm_crypt_io {
77 	struct crypt_config *cc;
78 	struct bio *base_bio;
79 	u8 *integrity_metadata;
80 	bool integrity_metadata_from_pool:1;
81 
82 	struct work_struct work;
83 
84 	struct convert_context ctx;
85 
86 	atomic_t io_pending;
87 	blk_status_t error;
88 	sector_t sector;
89 
90 	struct bvec_iter saved_bi_iter;
91 
92 	struct rb_node rb_node;
93 } CRYPTO_MINALIGN_ATTR;
94 
95 struct dm_crypt_request {
96 	struct convert_context *ctx;
97 	struct scatterlist sg_in[4];
98 	struct scatterlist sg_out[4];
99 	u64 iv_sector;
100 };
101 
102 struct crypt_config;
103 
104 struct crypt_iv_operations {
105 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
106 		   const char *opts);
107 	void (*dtr)(struct crypt_config *cc);
108 	int (*init)(struct crypt_config *cc);
109 	int (*wipe)(struct crypt_config *cc);
110 	int (*generator)(struct crypt_config *cc, u8 *iv,
111 			 struct dm_crypt_request *dmreq);
112 	int (*post)(struct crypt_config *cc, u8 *iv,
113 		    struct dm_crypt_request *dmreq);
114 };
115 
116 struct iv_benbi_private {
117 	int shift;
118 };
119 
120 #define LMK_SEED_SIZE 64 /* hash + 0 */
121 struct iv_lmk_private {
122 	struct crypto_shash *hash_tfm;
123 	u8 *seed;
124 };
125 
126 #define TCW_WHITENING_SIZE 16
127 struct iv_tcw_private {
128 	struct crypto_shash *crc32_tfm;
129 	u8 *iv_seed;
130 	u8 *whitening;
131 };
132 
133 #define ELEPHANT_MAX_KEY_SIZE 32
134 struct iv_elephant_private {
135 	struct crypto_skcipher *tfm;
136 };
137 
138 /*
139  * Crypt: maps a linear range of a block device
140  * and encrypts / decrypts at the same time.
141  */
142 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
143 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
144 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
145 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
146 
147 enum cipher_flags {
148 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
149 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
150 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
151 	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
152 };
153 
154 /*
155  * The fields in here must be read only after initialization.
156  */
157 struct crypt_config {
158 	struct dm_dev *dev;
159 	sector_t start;
160 
161 	struct percpu_counter n_allocated_pages;
162 
163 	struct workqueue_struct *io_queue;
164 	struct workqueue_struct *crypt_queue;
165 
166 	spinlock_t write_thread_lock;
167 	struct task_struct *write_thread;
168 	struct rb_root write_tree;
169 
170 	char *cipher_string;
171 	char *cipher_auth;
172 	char *key_string;
173 
174 	const struct crypt_iv_operations *iv_gen_ops;
175 	union {
176 		struct iv_benbi_private benbi;
177 		struct iv_lmk_private lmk;
178 		struct iv_tcw_private tcw;
179 		struct iv_elephant_private elephant;
180 	} iv_gen_private;
181 	u64 iv_offset;
182 	unsigned int iv_size;
183 	unsigned short sector_size;
184 	unsigned char sector_shift;
185 
186 	union {
187 		struct crypto_skcipher **tfms;
188 		struct crypto_aead **tfms_aead;
189 	} cipher_tfm;
190 	unsigned int tfms_count;
191 	int workqueue_id;
192 	unsigned long cipher_flags;
193 
194 	/*
195 	 * Layout of each crypto request:
196 	 *
197 	 *   struct skcipher_request
198 	 *      context
199 	 *      padding
200 	 *   struct dm_crypt_request
201 	 *      padding
202 	 *   IV
203 	 *
204 	 * The padding is added so that dm_crypt_request and the IV are
205 	 * correctly aligned.
206 	 */
207 	unsigned int dmreq_start;
208 
209 	unsigned int per_bio_data_size;
210 
211 	unsigned long flags;
212 	unsigned int key_size;
213 	unsigned int key_parts;      /* independent parts in key buffer */
214 	unsigned int key_extra_size; /* additional keys length */
215 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
216 
217 	unsigned int integrity_tag_size;
218 	unsigned int integrity_iv_size;
219 	unsigned int used_tag_size;
220 	unsigned int tuple_size;
221 
222 	/*
223 	 * pool for per bio private data, crypto requests,
224 	 * encryption requeusts/buffer pages and integrity tags
225 	 */
226 	unsigned int tag_pool_max_sectors;
227 	mempool_t tag_pool;
228 	mempool_t req_pool;
229 	mempool_t page_pool;
230 
231 	struct bio_set bs;
232 	struct mutex bio_alloc_lock;
233 
234 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
235 	u8 key[] __counted_by(key_size);
236 };
237 
238 #define MIN_IOS		64
239 #define MAX_TAG_SIZE	480
240 #define POOL_ENTRY_SIZE	512
241 
242 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
243 static unsigned int dm_crypt_clients_n;
244 static volatile unsigned long dm_crypt_pages_per_client;
245 #define DM_CRYPT_MEMORY_PERCENT			2
246 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
247 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
248 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
249 
250 static unsigned int max_read_size = 0;
251 module_param(max_read_size, uint, 0644);
252 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
253 static unsigned int max_write_size = 0;
254 module_param(max_write_size, uint, 0644);
255 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
256 static unsigned get_max_request_size(struct crypt_config *cc, bool wrt)
257 {
258 	unsigned val, sector_align;
259 	val = !wrt ? READ_ONCE(max_read_size) : READ_ONCE(max_write_size);
260 	if (likely(!val))
261 		val = !wrt ? DM_CRYPT_DEFAULT_MAX_READ_SIZE : DM_CRYPT_DEFAULT_MAX_WRITE_SIZE;
262 	if (wrt || cc->used_tag_size) {
263 		if (unlikely(val > BIO_MAX_VECS << PAGE_SHIFT))
264 			val = BIO_MAX_VECS << PAGE_SHIFT;
265 	}
266 	sector_align = max(bdev_logical_block_size(cc->dev->bdev), (unsigned)cc->sector_size);
267 	val = round_down(val, sector_align);
268 	if (unlikely(!val))
269 		val = sector_align;
270 	return val >> SECTOR_SHIFT;
271 }
272 
273 static void crypt_endio(struct bio *clone);
274 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
275 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
276 					     struct scatterlist *sg);
277 
278 static bool crypt_integrity_aead(struct crypt_config *cc);
279 
280 /*
281  * Use this to access cipher attributes that are independent of the key.
282  */
283 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
284 {
285 	return cc->cipher_tfm.tfms[0];
286 }
287 
288 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
289 {
290 	return cc->cipher_tfm.tfms_aead[0];
291 }
292 
293 /*
294  * Different IV generation algorithms:
295  *
296  * plain: the initial vector is the 32-bit little-endian version of the sector
297  *        number, padded with zeros if necessary.
298  *
299  * plain64: the initial vector is the 64-bit little-endian version of the sector
300  *        number, padded with zeros if necessary.
301  *
302  * plain64be: the initial vector is the 64-bit big-endian version of the sector
303  *        number, padded with zeros if necessary.
304  *
305  * essiv: "encrypted sector|salt initial vector", the sector number is
306  *        encrypted with the bulk cipher using a salt as key. The salt
307  *        should be derived from the bulk cipher's key via hashing.
308  *
309  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
310  *        (needed for LRW-32-AES and possible other narrow block modes)
311  *
312  * null: the initial vector is always zero.  Provides compatibility with
313  *       obsolete loop_fish2 devices.  Do not use for new devices.
314  *
315  * lmk:  Compatible implementation of the block chaining mode used
316  *       by the Loop-AES block device encryption system
317  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
318  *       It operates on full 512 byte sectors and uses CBC
319  *       with an IV derived from the sector number, the data and
320  *       optionally extra IV seed.
321  *       This means that after decryption the first block
322  *       of sector must be tweaked according to decrypted data.
323  *       Loop-AES can use three encryption schemes:
324  *         version 1: is plain aes-cbc mode
325  *         version 2: uses 64 multikey scheme with lmk IV generator
326  *         version 3: the same as version 2 with additional IV seed
327  *                   (it uses 65 keys, last key is used as IV seed)
328  *
329  * tcw:  Compatible implementation of the block chaining mode used
330  *       by the TrueCrypt device encryption system (prior to version 4.1).
331  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
332  *       It operates on full 512 byte sectors and uses CBC
333  *       with an IV derived from initial key and the sector number.
334  *       In addition, whitening value is applied on every sector, whitening
335  *       is calculated from initial key, sector number and mixed using CRC32.
336  *       Note that this encryption scheme is vulnerable to watermarking attacks
337  *       and should be used for old compatible containers access only.
338  *
339  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
340  *        The IV is encrypted little-endian byte-offset (with the same key
341  *        and cipher as the volume).
342  *
343  * elephant: The extended version of eboiv with additional Elephant diffuser
344  *           used with Bitlocker CBC mode.
345  *           This mode was used in older Windows systems
346  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
347  */
348 
349 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
350 			      struct dm_crypt_request *dmreq)
351 {
352 	memset(iv, 0, cc->iv_size);
353 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
354 
355 	return 0;
356 }
357 
358 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
359 				struct dm_crypt_request *dmreq)
360 {
361 	memset(iv, 0, cc->iv_size);
362 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
363 
364 	return 0;
365 }
366 
367 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
368 				  struct dm_crypt_request *dmreq)
369 {
370 	memset(iv, 0, cc->iv_size);
371 	/* iv_size is at least of size u64; usually it is 16 bytes */
372 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
373 
374 	return 0;
375 }
376 
377 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
378 			      struct dm_crypt_request *dmreq)
379 {
380 	/*
381 	 * ESSIV encryption of the IV is now handled by the crypto API,
382 	 * so just pass the plain sector number here.
383 	 */
384 	memset(iv, 0, cc->iv_size);
385 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
386 
387 	return 0;
388 }
389 
390 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
391 			      const char *opts)
392 {
393 	unsigned int bs;
394 	int log;
395 
396 	if (crypt_integrity_aead(cc))
397 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
398 	else
399 		bs = crypto_skcipher_blocksize(any_tfm(cc));
400 	log = ilog2(bs);
401 
402 	/*
403 	 * We need to calculate how far we must shift the sector count
404 	 * to get the cipher block count, we use this shift in _gen.
405 	 */
406 	if (1 << log != bs) {
407 		ti->error = "cypher blocksize is not a power of 2";
408 		return -EINVAL;
409 	}
410 
411 	if (log > 9) {
412 		ti->error = "cypher blocksize is > 512";
413 		return -EINVAL;
414 	}
415 
416 	cc->iv_gen_private.benbi.shift = 9 - log;
417 
418 	return 0;
419 }
420 
421 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
422 {
423 }
424 
425 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
426 			      struct dm_crypt_request *dmreq)
427 {
428 	__be64 val;
429 
430 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
431 
432 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
433 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
434 
435 	return 0;
436 }
437 
438 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
439 			     struct dm_crypt_request *dmreq)
440 {
441 	memset(iv, 0, cc->iv_size);
442 
443 	return 0;
444 }
445 
446 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
447 {
448 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
449 
450 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
451 		crypto_free_shash(lmk->hash_tfm);
452 	lmk->hash_tfm = NULL;
453 
454 	kfree_sensitive(lmk->seed);
455 	lmk->seed = NULL;
456 }
457 
458 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
459 			    const char *opts)
460 {
461 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
462 
463 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
464 		ti->error = "Unsupported sector size for LMK";
465 		return -EINVAL;
466 	}
467 
468 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
469 					   CRYPTO_ALG_ALLOCATES_MEMORY);
470 	if (IS_ERR(lmk->hash_tfm)) {
471 		ti->error = "Error initializing LMK hash";
472 		return PTR_ERR(lmk->hash_tfm);
473 	}
474 
475 	/* No seed in LMK version 2 */
476 	if (cc->key_parts == cc->tfms_count) {
477 		lmk->seed = NULL;
478 		return 0;
479 	}
480 
481 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
482 	if (!lmk->seed) {
483 		crypt_iv_lmk_dtr(cc);
484 		ti->error = "Error kmallocing seed storage in LMK";
485 		return -ENOMEM;
486 	}
487 
488 	return 0;
489 }
490 
491 static int crypt_iv_lmk_init(struct crypt_config *cc)
492 {
493 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
494 	int subkey_size = cc->key_size / cc->key_parts;
495 
496 	/* LMK seed is on the position of LMK_KEYS + 1 key */
497 	if (lmk->seed)
498 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
499 		       crypto_shash_digestsize(lmk->hash_tfm));
500 
501 	return 0;
502 }
503 
504 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
505 {
506 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
507 
508 	if (lmk->seed)
509 		memset(lmk->seed, 0, LMK_SEED_SIZE);
510 
511 	return 0;
512 }
513 
514 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
515 			    struct dm_crypt_request *dmreq,
516 			    u8 *data)
517 {
518 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
519 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
520 	struct md5_state md5state;
521 	__le32 buf[4];
522 	int i, r;
523 
524 	desc->tfm = lmk->hash_tfm;
525 
526 	r = crypto_shash_init(desc);
527 	if (r)
528 		return r;
529 
530 	if (lmk->seed) {
531 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
532 		if (r)
533 			return r;
534 	}
535 
536 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
537 	r = crypto_shash_update(desc, data + 16, 16 * 31);
538 	if (r)
539 		return r;
540 
541 	/* Sector is cropped to 56 bits here */
542 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
543 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
544 	buf[2] = cpu_to_le32(4024);
545 	buf[3] = 0;
546 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
547 	if (r)
548 		return r;
549 
550 	/* No MD5 padding here */
551 	r = crypto_shash_export(desc, &md5state);
552 	if (r)
553 		return r;
554 
555 	for (i = 0; i < MD5_HASH_WORDS; i++)
556 		__cpu_to_le32s(&md5state.hash[i]);
557 	memcpy(iv, &md5state.hash, cc->iv_size);
558 
559 	return 0;
560 }
561 
562 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
563 			    struct dm_crypt_request *dmreq)
564 {
565 	struct scatterlist *sg;
566 	u8 *src;
567 	int r = 0;
568 
569 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
570 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
571 		src = kmap_local_page(sg_page(sg));
572 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
573 		kunmap_local(src);
574 	} else
575 		memset(iv, 0, cc->iv_size);
576 
577 	return r;
578 }
579 
580 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
581 			     struct dm_crypt_request *dmreq)
582 {
583 	struct scatterlist *sg;
584 	u8 *dst;
585 	int r;
586 
587 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
588 		return 0;
589 
590 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
591 	dst = kmap_local_page(sg_page(sg));
592 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
593 
594 	/* Tweak the first block of plaintext sector */
595 	if (!r)
596 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
597 
598 	kunmap_local(dst);
599 	return r;
600 }
601 
602 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
603 {
604 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
605 
606 	kfree_sensitive(tcw->iv_seed);
607 	tcw->iv_seed = NULL;
608 	kfree_sensitive(tcw->whitening);
609 	tcw->whitening = NULL;
610 
611 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
612 		crypto_free_shash(tcw->crc32_tfm);
613 	tcw->crc32_tfm = NULL;
614 }
615 
616 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
617 			    const char *opts)
618 {
619 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
620 
621 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
622 		ti->error = "Unsupported sector size for TCW";
623 		return -EINVAL;
624 	}
625 
626 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
627 		ti->error = "Wrong key size for TCW";
628 		return -EINVAL;
629 	}
630 
631 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
632 					    CRYPTO_ALG_ALLOCATES_MEMORY);
633 	if (IS_ERR(tcw->crc32_tfm)) {
634 		ti->error = "Error initializing CRC32 in TCW";
635 		return PTR_ERR(tcw->crc32_tfm);
636 	}
637 
638 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
639 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
640 	if (!tcw->iv_seed || !tcw->whitening) {
641 		crypt_iv_tcw_dtr(cc);
642 		ti->error = "Error allocating seed storage in TCW";
643 		return -ENOMEM;
644 	}
645 
646 	return 0;
647 }
648 
649 static int crypt_iv_tcw_init(struct crypt_config *cc)
650 {
651 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
652 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
653 
654 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
655 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
656 	       TCW_WHITENING_SIZE);
657 
658 	return 0;
659 }
660 
661 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
662 {
663 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
664 
665 	memset(tcw->iv_seed, 0, cc->iv_size);
666 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
667 
668 	return 0;
669 }
670 
671 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
672 				  struct dm_crypt_request *dmreq,
673 				  u8 *data)
674 {
675 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
676 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
677 	u8 buf[TCW_WHITENING_SIZE];
678 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
679 	int i, r;
680 
681 	/* xor whitening with sector number */
682 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
683 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
684 
685 	/* calculate crc32 for every 32bit part and xor it */
686 	desc->tfm = tcw->crc32_tfm;
687 	for (i = 0; i < 4; i++) {
688 		r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
689 		if (r)
690 			goto out;
691 	}
692 	crypto_xor(&buf[0], &buf[12], 4);
693 	crypto_xor(&buf[4], &buf[8], 4);
694 
695 	/* apply whitening (8 bytes) to whole sector */
696 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
697 		crypto_xor(data + i * 8, buf, 8);
698 out:
699 	memzero_explicit(buf, sizeof(buf));
700 	return r;
701 }
702 
703 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
704 			    struct dm_crypt_request *dmreq)
705 {
706 	struct scatterlist *sg;
707 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
708 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
709 	u8 *src;
710 	int r = 0;
711 
712 	/* Remove whitening from ciphertext */
713 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
714 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
715 		src = kmap_local_page(sg_page(sg));
716 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
717 		kunmap_local(src);
718 	}
719 
720 	/* Calculate IV */
721 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
722 	if (cc->iv_size > 8)
723 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
724 			       cc->iv_size - 8);
725 
726 	return r;
727 }
728 
729 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
730 			     struct dm_crypt_request *dmreq)
731 {
732 	struct scatterlist *sg;
733 	u8 *dst;
734 	int r;
735 
736 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
737 		return 0;
738 
739 	/* Apply whitening on ciphertext */
740 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
741 	dst = kmap_local_page(sg_page(sg));
742 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
743 	kunmap_local(dst);
744 
745 	return r;
746 }
747 
748 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
749 				struct dm_crypt_request *dmreq)
750 {
751 	/* Used only for writes, there must be an additional space to store IV */
752 	get_random_bytes(iv, cc->iv_size);
753 	return 0;
754 }
755 
756 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
757 			    const char *opts)
758 {
759 	if (crypt_integrity_aead(cc)) {
760 		ti->error = "AEAD transforms not supported for EBOIV";
761 		return -EINVAL;
762 	}
763 
764 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
765 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
766 		return -EINVAL;
767 	}
768 
769 	return 0;
770 }
771 
772 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
773 			    struct dm_crypt_request *dmreq)
774 {
775 	struct crypto_skcipher *tfm = any_tfm(cc);
776 	struct skcipher_request *req;
777 	struct scatterlist src, dst;
778 	DECLARE_CRYPTO_WAIT(wait);
779 	unsigned int reqsize;
780 	int err;
781 	u8 *buf;
782 
783 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
784 	reqsize = ALIGN(reqsize, __alignof__(__le64));
785 
786 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
787 	if (!req)
788 		return -ENOMEM;
789 
790 	skcipher_request_set_tfm(req, tfm);
791 
792 	buf = (u8 *)req + reqsize;
793 	memset(buf, 0, cc->iv_size);
794 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
795 
796 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
797 	sg_init_one(&dst, iv, cc->iv_size);
798 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
799 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
800 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
801 	kfree_sensitive(req);
802 
803 	return err;
804 }
805 
806 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
807 {
808 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
809 
810 	crypto_free_skcipher(elephant->tfm);
811 	elephant->tfm = NULL;
812 }
813 
814 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
815 			    const char *opts)
816 {
817 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
818 	int r;
819 
820 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
821 					      CRYPTO_ALG_ALLOCATES_MEMORY);
822 	if (IS_ERR(elephant->tfm)) {
823 		r = PTR_ERR(elephant->tfm);
824 		elephant->tfm = NULL;
825 		return r;
826 	}
827 
828 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
829 	if (r)
830 		crypt_iv_elephant_dtr(cc);
831 	return r;
832 }
833 
834 static void diffuser_disk_to_cpu(u32 *d, size_t n)
835 {
836 #ifndef __LITTLE_ENDIAN
837 	int i;
838 
839 	for (i = 0; i < n; i++)
840 		d[i] = le32_to_cpu((__le32)d[i]);
841 #endif
842 }
843 
844 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
845 {
846 #ifndef __LITTLE_ENDIAN
847 	int i;
848 
849 	for (i = 0; i < n; i++)
850 		d[i] = cpu_to_le32((u32)d[i]);
851 #endif
852 }
853 
854 static void diffuser_a_decrypt(u32 *d, size_t n)
855 {
856 	int i, i1, i2, i3;
857 
858 	for (i = 0; i < 5; i++) {
859 		i1 = 0;
860 		i2 = n - 2;
861 		i3 = n - 5;
862 
863 		while (i1 < (n - 1)) {
864 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
865 			i1++; i2++; i3++;
866 
867 			if (i3 >= n)
868 				i3 -= n;
869 
870 			d[i1] += d[i2] ^ d[i3];
871 			i1++; i2++; i3++;
872 
873 			if (i2 >= n)
874 				i2 -= n;
875 
876 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
877 			i1++; i2++; i3++;
878 
879 			d[i1] += d[i2] ^ d[i3];
880 			i1++; i2++; i3++;
881 		}
882 	}
883 }
884 
885 static void diffuser_a_encrypt(u32 *d, size_t n)
886 {
887 	int i, i1, i2, i3;
888 
889 	for (i = 0; i < 5; i++) {
890 		i1 = n - 1;
891 		i2 = n - 2 - 1;
892 		i3 = n - 5 - 1;
893 
894 		while (i1 > 0) {
895 			d[i1] -= d[i2] ^ d[i3];
896 			i1--; i2--; i3--;
897 
898 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
899 			i1--; i2--; i3--;
900 
901 			if (i2 < 0)
902 				i2 += n;
903 
904 			d[i1] -= d[i2] ^ d[i3];
905 			i1--; i2--; i3--;
906 
907 			if (i3 < 0)
908 				i3 += n;
909 
910 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
911 			i1--; i2--; i3--;
912 		}
913 	}
914 }
915 
916 static void diffuser_b_decrypt(u32 *d, size_t n)
917 {
918 	int i, i1, i2, i3;
919 
920 	for (i = 0; i < 3; i++) {
921 		i1 = 0;
922 		i2 = 2;
923 		i3 = 5;
924 
925 		while (i1 < (n - 1)) {
926 			d[i1] += d[i2] ^ d[i3];
927 			i1++; i2++; i3++;
928 
929 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
930 			i1++; i2++; i3++;
931 
932 			if (i2 >= n)
933 				i2 -= n;
934 
935 			d[i1] += d[i2] ^ d[i3];
936 			i1++; i2++; i3++;
937 
938 			if (i3 >= n)
939 				i3 -= n;
940 
941 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
942 			i1++; i2++; i3++;
943 		}
944 	}
945 }
946 
947 static void diffuser_b_encrypt(u32 *d, size_t n)
948 {
949 	int i, i1, i2, i3;
950 
951 	for (i = 0; i < 3; i++) {
952 		i1 = n - 1;
953 		i2 = 2 - 1;
954 		i3 = 5 - 1;
955 
956 		while (i1 > 0) {
957 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
958 			i1--; i2--; i3--;
959 
960 			if (i3 < 0)
961 				i3 += n;
962 
963 			d[i1] -= d[i2] ^ d[i3];
964 			i1--; i2--; i3--;
965 
966 			if (i2 < 0)
967 				i2 += n;
968 
969 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
970 			i1--; i2--; i3--;
971 
972 			d[i1] -= d[i2] ^ d[i3];
973 			i1--; i2--; i3--;
974 		}
975 	}
976 }
977 
978 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
979 {
980 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
981 	u8 *es, *ks, *data, *data2, *data_offset;
982 	struct skcipher_request *req;
983 	struct scatterlist *sg, *sg2, src, dst;
984 	DECLARE_CRYPTO_WAIT(wait);
985 	int i, r;
986 
987 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
988 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
989 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
990 
991 	if (!req || !es || !ks) {
992 		r = -ENOMEM;
993 		goto out;
994 	}
995 
996 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
997 
998 	/* E(Ks, e(s)) */
999 	sg_init_one(&src, es, 16);
1000 	sg_init_one(&dst, ks, 16);
1001 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
1002 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
1003 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1004 	if (r)
1005 		goto out;
1006 
1007 	/* E(Ks, e'(s)) */
1008 	es[15] = 0x80;
1009 	sg_init_one(&dst, &ks[16], 16);
1010 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1011 	if (r)
1012 		goto out;
1013 
1014 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
1015 	data = kmap_local_page(sg_page(sg));
1016 	data_offset = data + sg->offset;
1017 
1018 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
1019 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1020 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
1021 		data2 = kmap_local_page(sg_page(sg2));
1022 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
1023 		kunmap_local(data2);
1024 	}
1025 
1026 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
1027 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1028 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1029 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1030 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1031 	}
1032 
1033 	for (i = 0; i < (cc->sector_size / 32); i++)
1034 		crypto_xor(data_offset + i * 32, ks, 32);
1035 
1036 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1037 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1038 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1039 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1040 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1041 	}
1042 
1043 	kunmap_local(data);
1044 out:
1045 	kfree_sensitive(ks);
1046 	kfree_sensitive(es);
1047 	skcipher_request_free(req);
1048 	return r;
1049 }
1050 
1051 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1052 			    struct dm_crypt_request *dmreq)
1053 {
1054 	int r;
1055 
1056 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1057 		r = crypt_iv_elephant(cc, dmreq);
1058 		if (r)
1059 			return r;
1060 	}
1061 
1062 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1063 }
1064 
1065 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1066 				  struct dm_crypt_request *dmreq)
1067 {
1068 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1069 		return crypt_iv_elephant(cc, dmreq);
1070 
1071 	return 0;
1072 }
1073 
1074 static int crypt_iv_elephant_init(struct crypt_config *cc)
1075 {
1076 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1077 	int key_offset = cc->key_size - cc->key_extra_size;
1078 
1079 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1080 }
1081 
1082 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1083 {
1084 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1085 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1086 
1087 	memset(key, 0, cc->key_extra_size);
1088 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1089 }
1090 
1091 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1092 	.generator = crypt_iv_plain_gen
1093 };
1094 
1095 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1096 	.generator = crypt_iv_plain64_gen
1097 };
1098 
1099 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1100 	.generator = crypt_iv_plain64be_gen
1101 };
1102 
1103 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1104 	.generator = crypt_iv_essiv_gen
1105 };
1106 
1107 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1108 	.ctr	   = crypt_iv_benbi_ctr,
1109 	.dtr	   = crypt_iv_benbi_dtr,
1110 	.generator = crypt_iv_benbi_gen
1111 };
1112 
1113 static const struct crypt_iv_operations crypt_iv_null_ops = {
1114 	.generator = crypt_iv_null_gen
1115 };
1116 
1117 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1118 	.ctr	   = crypt_iv_lmk_ctr,
1119 	.dtr	   = crypt_iv_lmk_dtr,
1120 	.init	   = crypt_iv_lmk_init,
1121 	.wipe	   = crypt_iv_lmk_wipe,
1122 	.generator = crypt_iv_lmk_gen,
1123 	.post	   = crypt_iv_lmk_post
1124 };
1125 
1126 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1127 	.ctr	   = crypt_iv_tcw_ctr,
1128 	.dtr	   = crypt_iv_tcw_dtr,
1129 	.init	   = crypt_iv_tcw_init,
1130 	.wipe	   = crypt_iv_tcw_wipe,
1131 	.generator = crypt_iv_tcw_gen,
1132 	.post	   = crypt_iv_tcw_post
1133 };
1134 
1135 static const struct crypt_iv_operations crypt_iv_random_ops = {
1136 	.generator = crypt_iv_random_gen
1137 };
1138 
1139 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1140 	.ctr	   = crypt_iv_eboiv_ctr,
1141 	.generator = crypt_iv_eboiv_gen
1142 };
1143 
1144 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1145 	.ctr	   = crypt_iv_elephant_ctr,
1146 	.dtr	   = crypt_iv_elephant_dtr,
1147 	.init	   = crypt_iv_elephant_init,
1148 	.wipe	   = crypt_iv_elephant_wipe,
1149 	.generator = crypt_iv_elephant_gen,
1150 	.post	   = crypt_iv_elephant_post
1151 };
1152 
1153 /*
1154  * Integrity extensions
1155  */
1156 static bool crypt_integrity_aead(struct crypt_config *cc)
1157 {
1158 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1159 }
1160 
1161 static bool crypt_integrity_hmac(struct crypt_config *cc)
1162 {
1163 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1164 }
1165 
1166 /* Get sg containing data */
1167 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1168 					     struct scatterlist *sg)
1169 {
1170 	if (unlikely(crypt_integrity_aead(cc)))
1171 		return &sg[2];
1172 
1173 	return sg;
1174 }
1175 
1176 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1177 {
1178 	struct bio_integrity_payload *bip;
1179 	unsigned int tag_len;
1180 	int ret;
1181 
1182 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1183 		return 0;
1184 
1185 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1186 	if (IS_ERR(bip))
1187 		return PTR_ERR(bip);
1188 
1189 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1190 
1191 	bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
1192 
1193 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1194 				     tag_len, offset_in_page(io->integrity_metadata));
1195 	if (unlikely(ret != tag_len))
1196 		return -ENOMEM;
1197 
1198 	return 0;
1199 }
1200 
1201 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1202 {
1203 #ifdef CONFIG_BLK_DEV_INTEGRITY
1204 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1205 	struct mapped_device *md = dm_table_get_md(ti->table);
1206 
1207 	/* We require an underlying device with non-PI metadata */
1208 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1209 		ti->error = "Integrity profile not supported.";
1210 		return -EINVAL;
1211 	}
1212 
1213 	if (bi->tuple_size < cc->used_tag_size) {
1214 		ti->error = "Integrity profile tag size mismatch.";
1215 		return -EINVAL;
1216 	}
1217 	cc->tuple_size = bi->tuple_size;
1218 	if (1 << bi->interval_exp != cc->sector_size) {
1219 		ti->error = "Integrity profile sector size mismatch.";
1220 		return -EINVAL;
1221 	}
1222 
1223 	if (crypt_integrity_aead(cc)) {
1224 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1225 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1226 		       cc->integrity_tag_size, cc->integrity_iv_size);
1227 
1228 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1229 			ti->error = "Integrity AEAD auth tag size is not supported.";
1230 			return -EINVAL;
1231 		}
1232 	} else if (cc->integrity_iv_size)
1233 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1234 		       cc->integrity_iv_size);
1235 
1236 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1237 		ti->error = "Not enough space for integrity tag in the profile.";
1238 		return -EINVAL;
1239 	}
1240 
1241 	return 0;
1242 #else
1243 	ti->error = "Integrity profile not supported.";
1244 	return -EINVAL;
1245 #endif
1246 }
1247 
1248 static void crypt_convert_init(struct crypt_config *cc,
1249 			       struct convert_context *ctx,
1250 			       struct bio *bio_out, struct bio *bio_in,
1251 			       sector_t sector)
1252 {
1253 	ctx->bio_in = bio_in;
1254 	ctx->bio_out = bio_out;
1255 	if (bio_in)
1256 		ctx->iter_in = bio_in->bi_iter;
1257 	if (bio_out)
1258 		ctx->iter_out = bio_out->bi_iter;
1259 	ctx->cc_sector = sector + cc->iv_offset;
1260 	ctx->tag_offset = 0;
1261 	init_completion(&ctx->restart);
1262 }
1263 
1264 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1265 					     void *req)
1266 {
1267 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1268 }
1269 
1270 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1271 {
1272 	return (void *)((char *)dmreq - cc->dmreq_start);
1273 }
1274 
1275 static u8 *iv_of_dmreq(struct crypt_config *cc,
1276 		       struct dm_crypt_request *dmreq)
1277 {
1278 	if (crypt_integrity_aead(cc))
1279 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1280 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1281 	else
1282 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1283 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1284 }
1285 
1286 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1287 		       struct dm_crypt_request *dmreq)
1288 {
1289 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1290 }
1291 
1292 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1293 		       struct dm_crypt_request *dmreq)
1294 {
1295 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1296 
1297 	return (__le64 *) ptr;
1298 }
1299 
1300 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1301 		       struct dm_crypt_request *dmreq)
1302 {
1303 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1304 		  cc->iv_size + sizeof(uint64_t);
1305 
1306 	return (unsigned int *)ptr;
1307 }
1308 
1309 static void *tag_from_dmreq(struct crypt_config *cc,
1310 				struct dm_crypt_request *dmreq)
1311 {
1312 	struct convert_context *ctx = dmreq->ctx;
1313 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1314 
1315 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1316 		cc->tuple_size];
1317 }
1318 
1319 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1320 			       struct dm_crypt_request *dmreq)
1321 {
1322 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1323 }
1324 
1325 static int crypt_convert_block_aead(struct crypt_config *cc,
1326 				     struct convert_context *ctx,
1327 				     struct aead_request *req,
1328 				     unsigned int tag_offset)
1329 {
1330 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1331 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1332 	struct dm_crypt_request *dmreq;
1333 	u8 *iv, *org_iv, *tag_iv, *tag;
1334 	__le64 *sector;
1335 	int r = 0;
1336 
1337 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1338 
1339 	/* Reject unexpected unaligned bio. */
1340 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1341 		return -EIO;
1342 
1343 	dmreq = dmreq_of_req(cc, req);
1344 	dmreq->iv_sector = ctx->cc_sector;
1345 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1346 		dmreq->iv_sector >>= cc->sector_shift;
1347 	dmreq->ctx = ctx;
1348 
1349 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1350 
1351 	sector = org_sector_of_dmreq(cc, dmreq);
1352 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1353 
1354 	iv = iv_of_dmreq(cc, dmreq);
1355 	org_iv = org_iv_of_dmreq(cc, dmreq);
1356 	tag = tag_from_dmreq(cc, dmreq);
1357 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1358 
1359 	/* AEAD request:
1360 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1361 	 *  | (authenticated) | (auth+encryption) |              |
1362 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1363 	 */
1364 	sg_init_table(dmreq->sg_in, 4);
1365 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1366 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1367 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1368 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1369 
1370 	sg_init_table(dmreq->sg_out, 4);
1371 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1372 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1373 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1374 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1375 
1376 	if (cc->iv_gen_ops) {
1377 		/* For READs use IV stored in integrity metadata */
1378 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1379 			memcpy(org_iv, tag_iv, cc->iv_size);
1380 		} else {
1381 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1382 			if (r < 0)
1383 				return r;
1384 			/* Store generated IV in integrity metadata */
1385 			if (cc->integrity_iv_size)
1386 				memcpy(tag_iv, org_iv, cc->iv_size);
1387 		}
1388 		/* Working copy of IV, to be modified in crypto API */
1389 		memcpy(iv, org_iv, cc->iv_size);
1390 	}
1391 
1392 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1393 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1394 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1395 				       cc->sector_size, iv);
1396 		r = crypto_aead_encrypt(req);
1397 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1398 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1399 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1400 	} else {
1401 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1402 				       cc->sector_size + cc->integrity_tag_size, iv);
1403 		r = crypto_aead_decrypt(req);
1404 	}
1405 
1406 	if (r == -EBADMSG) {
1407 		sector_t s = le64_to_cpu(*sector);
1408 
1409 		ctx->aead_failed = true;
1410 		if (ctx->aead_recheck) {
1411 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1412 				    ctx->bio_in->bi_bdev, s);
1413 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1414 					 ctx->bio_in, s, 0);
1415 		}
1416 	}
1417 
1418 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1419 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1420 
1421 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1422 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1423 
1424 	return r;
1425 }
1426 
1427 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1428 					struct convert_context *ctx,
1429 					struct skcipher_request *req,
1430 					unsigned int tag_offset)
1431 {
1432 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1433 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1434 	struct scatterlist *sg_in, *sg_out;
1435 	struct dm_crypt_request *dmreq;
1436 	u8 *iv, *org_iv, *tag_iv;
1437 	__le64 *sector;
1438 	int r = 0;
1439 
1440 	/* Reject unexpected unaligned bio. */
1441 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1442 		return -EIO;
1443 
1444 	dmreq = dmreq_of_req(cc, req);
1445 	dmreq->iv_sector = ctx->cc_sector;
1446 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1447 		dmreq->iv_sector >>= cc->sector_shift;
1448 	dmreq->ctx = ctx;
1449 
1450 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1451 
1452 	iv = iv_of_dmreq(cc, dmreq);
1453 	org_iv = org_iv_of_dmreq(cc, dmreq);
1454 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1455 
1456 	sector = org_sector_of_dmreq(cc, dmreq);
1457 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1458 
1459 	/* For skcipher we use only the first sg item */
1460 	sg_in  = &dmreq->sg_in[0];
1461 	sg_out = &dmreq->sg_out[0];
1462 
1463 	sg_init_table(sg_in, 1);
1464 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1465 
1466 	sg_init_table(sg_out, 1);
1467 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1468 
1469 	if (cc->iv_gen_ops) {
1470 		/* For READs use IV stored in integrity metadata */
1471 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1472 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1473 		} else {
1474 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1475 			if (r < 0)
1476 				return r;
1477 			/* Data can be already preprocessed in generator */
1478 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1479 				sg_in = sg_out;
1480 			/* Store generated IV in integrity metadata */
1481 			if (cc->integrity_iv_size)
1482 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1483 		}
1484 		/* Working copy of IV, to be modified in crypto API */
1485 		memcpy(iv, org_iv, cc->iv_size);
1486 	}
1487 
1488 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1489 
1490 	if (bio_data_dir(ctx->bio_in) == WRITE)
1491 		r = crypto_skcipher_encrypt(req);
1492 	else
1493 		r = crypto_skcipher_decrypt(req);
1494 
1495 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1496 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1497 
1498 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1499 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1500 
1501 	return r;
1502 }
1503 
1504 static void kcryptd_async_done(void *async_req, int error);
1505 
1506 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1507 				     struct convert_context *ctx)
1508 {
1509 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1510 
1511 	if (!ctx->r.req) {
1512 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1513 		if (!ctx->r.req)
1514 			return -ENOMEM;
1515 	}
1516 
1517 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1518 
1519 	/*
1520 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1521 	 * requests if driver request queue is full.
1522 	 */
1523 	skcipher_request_set_callback(ctx->r.req,
1524 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1525 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1526 
1527 	return 0;
1528 }
1529 
1530 static int crypt_alloc_req_aead(struct crypt_config *cc,
1531 				 struct convert_context *ctx)
1532 {
1533 	if (!ctx->r.req_aead) {
1534 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1535 		if (!ctx->r.req_aead)
1536 			return -ENOMEM;
1537 	}
1538 
1539 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1540 
1541 	/*
1542 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1543 	 * requests if driver request queue is full.
1544 	 */
1545 	aead_request_set_callback(ctx->r.req_aead,
1546 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1547 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1548 
1549 	return 0;
1550 }
1551 
1552 static int crypt_alloc_req(struct crypt_config *cc,
1553 			    struct convert_context *ctx)
1554 {
1555 	if (crypt_integrity_aead(cc))
1556 		return crypt_alloc_req_aead(cc, ctx);
1557 	else
1558 		return crypt_alloc_req_skcipher(cc, ctx);
1559 }
1560 
1561 static void crypt_free_req_skcipher(struct crypt_config *cc,
1562 				    struct skcipher_request *req, struct bio *base_bio)
1563 {
1564 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1565 
1566 	if ((struct skcipher_request *)(io + 1) != req)
1567 		mempool_free(req, &cc->req_pool);
1568 }
1569 
1570 static void crypt_free_req_aead(struct crypt_config *cc,
1571 				struct aead_request *req, struct bio *base_bio)
1572 {
1573 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1574 
1575 	if ((struct aead_request *)(io + 1) != req)
1576 		mempool_free(req, &cc->req_pool);
1577 }
1578 
1579 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1580 {
1581 	if (crypt_integrity_aead(cc))
1582 		crypt_free_req_aead(cc, req, base_bio);
1583 	else
1584 		crypt_free_req_skcipher(cc, req, base_bio);
1585 }
1586 
1587 /*
1588  * Encrypt / decrypt data from one bio to another one (can be the same one)
1589  */
1590 static blk_status_t crypt_convert(struct crypt_config *cc,
1591 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1592 {
1593 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1594 	int r;
1595 
1596 	/*
1597 	 * if reset_pending is set we are dealing with the bio for the first time,
1598 	 * else we're continuing to work on the previous bio, so don't mess with
1599 	 * the cc_pending counter
1600 	 */
1601 	if (reset_pending)
1602 		atomic_set(&ctx->cc_pending, 1);
1603 
1604 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1605 
1606 		r = crypt_alloc_req(cc, ctx);
1607 		if (r) {
1608 			complete(&ctx->restart);
1609 			return BLK_STS_DEV_RESOURCE;
1610 		}
1611 
1612 		atomic_inc(&ctx->cc_pending);
1613 
1614 		if (crypt_integrity_aead(cc))
1615 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, ctx->tag_offset);
1616 		else
1617 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->tag_offset);
1618 
1619 		switch (r) {
1620 		/*
1621 		 * The request was queued by a crypto driver
1622 		 * but the driver request queue is full, let's wait.
1623 		 */
1624 		case -EBUSY:
1625 			if (in_interrupt()) {
1626 				if (try_wait_for_completion(&ctx->restart)) {
1627 					/*
1628 					 * we don't have to block to wait for completion,
1629 					 * so proceed
1630 					 */
1631 				} else {
1632 					/*
1633 					 * we can't wait for completion without blocking
1634 					 * exit and continue processing in a workqueue
1635 					 */
1636 					ctx->r.req = NULL;
1637 					ctx->tag_offset++;
1638 					ctx->cc_sector += sector_step;
1639 					return BLK_STS_DEV_RESOURCE;
1640 				}
1641 			} else {
1642 				wait_for_completion(&ctx->restart);
1643 			}
1644 			reinit_completion(&ctx->restart);
1645 			fallthrough;
1646 		/*
1647 		 * The request is queued and processed asynchronously,
1648 		 * completion function kcryptd_async_done() will be called.
1649 		 */
1650 		case -EINPROGRESS:
1651 			ctx->r.req = NULL;
1652 			ctx->tag_offset++;
1653 			ctx->cc_sector += sector_step;
1654 			continue;
1655 		/*
1656 		 * The request was already processed (synchronously).
1657 		 */
1658 		case 0:
1659 			atomic_dec(&ctx->cc_pending);
1660 			ctx->cc_sector += sector_step;
1661 			ctx->tag_offset++;
1662 			if (!atomic)
1663 				cond_resched();
1664 			continue;
1665 		/*
1666 		 * There was a data integrity error.
1667 		 */
1668 		case -EBADMSG:
1669 			atomic_dec(&ctx->cc_pending);
1670 			return BLK_STS_PROTECTION;
1671 		/*
1672 		 * There was an error while processing the request.
1673 		 */
1674 		default:
1675 			atomic_dec(&ctx->cc_pending);
1676 			return BLK_STS_IOERR;
1677 		}
1678 	}
1679 
1680 	return 0;
1681 }
1682 
1683 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1684 
1685 /*
1686  * Generate a new unfragmented bio with the given size
1687  * This should never violate the device limitations (but if it did then block
1688  * core should split the bio as needed).
1689  *
1690  * This function may be called concurrently. If we allocate from the mempool
1691  * concurrently, there is a possibility of deadlock. For example, if we have
1692  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1693  * the mempool concurrently, it may deadlock in a situation where both processes
1694  * have allocated 128 pages and the mempool is exhausted.
1695  *
1696  * In order to avoid this scenario we allocate the pages under a mutex.
1697  *
1698  * In order to not degrade performance with excessive locking, we try
1699  * non-blocking allocations without a mutex first but on failure we fallback
1700  * to blocking allocations with a mutex.
1701  *
1702  * In order to reduce allocation overhead, we try to allocate compound pages in
1703  * the first pass. If they are not available, we fall back to the mempool.
1704  */
1705 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1706 {
1707 	struct crypt_config *cc = io->cc;
1708 	struct bio *clone;
1709 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1710 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1711 	unsigned int remaining_size;
1712 	unsigned int order = MAX_PAGE_ORDER;
1713 
1714 retry:
1715 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1716 		mutex_lock(&cc->bio_alloc_lock);
1717 
1718 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1719 				 GFP_NOIO, &cc->bs);
1720 	clone->bi_private = io;
1721 	clone->bi_end_io = crypt_endio;
1722 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1723 	clone->bi_iter.bi_sector = cc->start + io->sector;
1724 
1725 	remaining_size = size;
1726 
1727 	while (remaining_size) {
1728 		struct page *pages;
1729 		unsigned size_to_add;
1730 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1731 		order = min(order, remaining_order);
1732 
1733 		while (order > 0) {
1734 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1735 					(1 << order) > dm_crypt_pages_per_client))
1736 				goto decrease_order;
1737 			pages = alloc_pages(gfp_mask
1738 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1739 				order);
1740 			if (likely(pages != NULL)) {
1741 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1742 				goto have_pages;
1743 			}
1744 decrease_order:
1745 			order--;
1746 		}
1747 
1748 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1749 		if (!pages) {
1750 			crypt_free_buffer_pages(cc, clone);
1751 			bio_put(clone);
1752 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1753 			order = 0;
1754 			goto retry;
1755 		}
1756 
1757 have_pages:
1758 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1759 		__bio_add_page(clone, pages, size_to_add, 0);
1760 		remaining_size -= size_to_add;
1761 	}
1762 
1763 	/* Allocate space for integrity tags */
1764 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1765 		crypt_free_buffer_pages(cc, clone);
1766 		bio_put(clone);
1767 		clone = NULL;
1768 	}
1769 
1770 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1771 		mutex_unlock(&cc->bio_alloc_lock);
1772 
1773 	return clone;
1774 }
1775 
1776 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1777 {
1778 	struct folio_iter fi;
1779 
1780 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1781 		bio_for_each_folio_all(fi, clone) {
1782 			if (folio_test_large(fi.folio)) {
1783 				percpu_counter_sub(&cc->n_allocated_pages,
1784 						1 << folio_order(fi.folio));
1785 				folio_put(fi.folio);
1786 			} else {
1787 				mempool_free(&fi.folio->page, &cc->page_pool);
1788 			}
1789 		}
1790 	}
1791 }
1792 
1793 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1794 			  struct bio *bio, sector_t sector)
1795 {
1796 	io->cc = cc;
1797 	io->base_bio = bio;
1798 	io->sector = sector;
1799 	io->error = 0;
1800 	io->ctx.aead_recheck = false;
1801 	io->ctx.aead_failed = false;
1802 	io->ctx.r.req = NULL;
1803 	io->integrity_metadata = NULL;
1804 	io->integrity_metadata_from_pool = false;
1805 	atomic_set(&io->io_pending, 0);
1806 }
1807 
1808 static void crypt_inc_pending(struct dm_crypt_io *io)
1809 {
1810 	atomic_inc(&io->io_pending);
1811 }
1812 
1813 static void kcryptd_queue_read(struct dm_crypt_io *io);
1814 
1815 /*
1816  * One of the bios was finished. Check for completion of
1817  * the whole request and correctly clean up the buffer.
1818  */
1819 static void crypt_dec_pending(struct dm_crypt_io *io)
1820 {
1821 	struct crypt_config *cc = io->cc;
1822 	struct bio *base_bio = io->base_bio;
1823 	blk_status_t error = io->error;
1824 
1825 	if (!atomic_dec_and_test(&io->io_pending))
1826 		return;
1827 
1828 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1829 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1830 		io->ctx.aead_recheck = true;
1831 		io->ctx.aead_failed = false;
1832 		io->error = 0;
1833 		kcryptd_queue_read(io);
1834 		return;
1835 	}
1836 
1837 	if (io->ctx.r.req)
1838 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1839 
1840 	if (unlikely(io->integrity_metadata_from_pool))
1841 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1842 	else
1843 		kfree(io->integrity_metadata);
1844 
1845 	base_bio->bi_status = error;
1846 
1847 	bio_endio(base_bio);
1848 }
1849 
1850 /*
1851  * kcryptd/kcryptd_io:
1852  *
1853  * Needed because it would be very unwise to do decryption in an
1854  * interrupt context.
1855  *
1856  * kcryptd performs the actual encryption or decryption.
1857  *
1858  * kcryptd_io performs the IO submission.
1859  *
1860  * They must be separated as otherwise the final stages could be
1861  * starved by new requests which can block in the first stages due
1862  * to memory allocation.
1863  *
1864  * The work is done per CPU global for all dm-crypt instances.
1865  * They should not depend on each other and do not block.
1866  */
1867 static void crypt_endio(struct bio *clone)
1868 {
1869 	struct dm_crypt_io *io = clone->bi_private;
1870 	struct crypt_config *cc = io->cc;
1871 	unsigned int rw = bio_data_dir(clone);
1872 	blk_status_t error = clone->bi_status;
1873 
1874 	if (io->ctx.aead_recheck && !error) {
1875 		kcryptd_queue_crypt(io);
1876 		return;
1877 	}
1878 
1879 	/*
1880 	 * free the processed pages
1881 	 */
1882 	if (rw == WRITE || io->ctx.aead_recheck)
1883 		crypt_free_buffer_pages(cc, clone);
1884 
1885 	bio_put(clone);
1886 
1887 	if (rw == READ && !error) {
1888 		kcryptd_queue_crypt(io);
1889 		return;
1890 	}
1891 
1892 	if (unlikely(error))
1893 		io->error = error;
1894 
1895 	crypt_dec_pending(io);
1896 }
1897 
1898 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1899 
1900 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1901 {
1902 	struct crypt_config *cc = io->cc;
1903 	struct bio *clone;
1904 
1905 	if (io->ctx.aead_recheck) {
1906 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1907 			return 1;
1908 		crypt_inc_pending(io);
1909 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1910 		if (unlikely(!clone)) {
1911 			crypt_dec_pending(io);
1912 			return 1;
1913 		}
1914 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1915 		io->saved_bi_iter = clone->bi_iter;
1916 		dm_submit_bio_remap(io->base_bio, clone);
1917 		return 0;
1918 	}
1919 
1920 	/*
1921 	 * We need the original biovec array in order to decrypt the whole bio
1922 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1923 	 * worry about the block layer modifying the biovec array; so leverage
1924 	 * bio_alloc_clone().
1925 	 */
1926 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1927 	if (!clone)
1928 		return 1;
1929 
1930 	clone->bi_iter.bi_sector = cc->start + io->sector;
1931 	clone->bi_private = io;
1932 	clone->bi_end_io = crypt_endio;
1933 
1934 	crypt_inc_pending(io);
1935 
1936 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1937 		crypt_dec_pending(io);
1938 		bio_put(clone);
1939 		return 1;
1940 	}
1941 
1942 	dm_submit_bio_remap(io->base_bio, clone);
1943 	return 0;
1944 }
1945 
1946 static void kcryptd_io_read_work(struct work_struct *work)
1947 {
1948 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1949 
1950 	crypt_inc_pending(io);
1951 	if (kcryptd_io_read(io, GFP_NOIO))
1952 		io->error = BLK_STS_RESOURCE;
1953 	crypt_dec_pending(io);
1954 }
1955 
1956 static void kcryptd_queue_read(struct dm_crypt_io *io)
1957 {
1958 	struct crypt_config *cc = io->cc;
1959 
1960 	INIT_WORK(&io->work, kcryptd_io_read_work);
1961 	queue_work(cc->io_queue, &io->work);
1962 }
1963 
1964 static void kcryptd_io_write(struct dm_crypt_io *io)
1965 {
1966 	struct bio *clone = io->ctx.bio_out;
1967 
1968 	dm_submit_bio_remap(io->base_bio, clone);
1969 }
1970 
1971 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1972 
1973 static int dmcrypt_write(void *data)
1974 {
1975 	struct crypt_config *cc = data;
1976 	struct dm_crypt_io *io;
1977 
1978 	while (1) {
1979 		struct rb_root write_tree;
1980 		struct blk_plug plug;
1981 
1982 		spin_lock_irq(&cc->write_thread_lock);
1983 continue_locked:
1984 
1985 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1986 			goto pop_from_list;
1987 
1988 		set_current_state(TASK_INTERRUPTIBLE);
1989 
1990 		spin_unlock_irq(&cc->write_thread_lock);
1991 
1992 		if (unlikely(kthread_should_stop())) {
1993 			set_current_state(TASK_RUNNING);
1994 			break;
1995 		}
1996 
1997 		schedule();
1998 
1999 		spin_lock_irq(&cc->write_thread_lock);
2000 		goto continue_locked;
2001 
2002 pop_from_list:
2003 		write_tree = cc->write_tree;
2004 		cc->write_tree = RB_ROOT;
2005 		spin_unlock_irq(&cc->write_thread_lock);
2006 
2007 		BUG_ON(rb_parent(write_tree.rb_node));
2008 
2009 		/*
2010 		 * Note: we cannot walk the tree here with rb_next because
2011 		 * the structures may be freed when kcryptd_io_write is called.
2012 		 */
2013 		blk_start_plug(&plug);
2014 		do {
2015 			io = crypt_io_from_node(rb_first(&write_tree));
2016 			rb_erase(&io->rb_node, &write_tree);
2017 			kcryptd_io_write(io);
2018 			cond_resched();
2019 		} while (!RB_EMPTY_ROOT(&write_tree));
2020 		blk_finish_plug(&plug);
2021 	}
2022 	return 0;
2023 }
2024 
2025 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
2026 {
2027 	struct bio *clone = io->ctx.bio_out;
2028 	struct crypt_config *cc = io->cc;
2029 	unsigned long flags;
2030 	sector_t sector;
2031 	struct rb_node **rbp, *parent;
2032 
2033 	if (unlikely(io->error)) {
2034 		crypt_free_buffer_pages(cc, clone);
2035 		bio_put(clone);
2036 		crypt_dec_pending(io);
2037 		return;
2038 	}
2039 
2040 	/* crypt_convert should have filled the clone bio */
2041 	BUG_ON(io->ctx.iter_out.bi_size);
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 	blk_status_t r;
2095 
2096 	wait_for_completion(&ctx->restart);
2097 	reinit_completion(&ctx->restart);
2098 
2099 	r = crypt_convert(cc, &io->ctx, false, false);
2100 	if (r)
2101 		io->error = r;
2102 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2103 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2104 		/* Wait for completion signaled by kcryptd_async_done() */
2105 		wait_for_completion(&ctx->restart);
2106 		crypt_finished = 1;
2107 	}
2108 
2109 	/* Encryption was already finished, submit io now */
2110 	if (crypt_finished)
2111 		kcryptd_crypt_write_io_submit(io, 0);
2112 
2113 	crypt_dec_pending(io);
2114 }
2115 
2116 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2117 {
2118 	struct crypt_config *cc = io->cc;
2119 	struct convert_context *ctx = &io->ctx;
2120 	struct bio *clone;
2121 	int crypt_finished;
2122 	blk_status_t r;
2123 
2124 	/*
2125 	 * Prevent io from disappearing until this function completes.
2126 	 */
2127 	crypt_inc_pending(io);
2128 	crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector);
2129 
2130 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2131 	if (unlikely(!clone)) {
2132 		io->error = BLK_STS_IOERR;
2133 		goto dec;
2134 	}
2135 
2136 	io->ctx.bio_out = clone;
2137 	io->ctx.iter_out = clone->bi_iter;
2138 
2139 	if (crypt_integrity_aead(cc)) {
2140 		bio_copy_data(clone, io->base_bio);
2141 		io->ctx.bio_in = clone;
2142 		io->ctx.iter_in = clone->bi_iter;
2143 	}
2144 
2145 	crypt_inc_pending(io);
2146 	r = crypt_convert(cc, ctx,
2147 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2148 	/*
2149 	 * Crypto API backlogged the request, because its queue was full
2150 	 * and we're in softirq context, so continue from a workqueue
2151 	 * (TODO: is it actually possible to be in softirq in the write path?)
2152 	 */
2153 	if (r == BLK_STS_DEV_RESOURCE) {
2154 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2155 		queue_work(cc->crypt_queue, &io->work);
2156 		return;
2157 	}
2158 	if (r)
2159 		io->error = r;
2160 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2161 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2162 		/* Wait for completion signaled by kcryptd_async_done() */
2163 		wait_for_completion(&ctx->restart);
2164 		crypt_finished = 1;
2165 	}
2166 
2167 	/* Encryption was already finished, submit io now */
2168 	if (crypt_finished)
2169 		kcryptd_crypt_write_io_submit(io, 0);
2170 
2171 dec:
2172 	crypt_dec_pending(io);
2173 }
2174 
2175 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2176 {
2177 	if (io->ctx.aead_recheck) {
2178 		if (!io->error) {
2179 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2180 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2181 		}
2182 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2183 		bio_put(io->ctx.bio_in);
2184 	}
2185 	crypt_dec_pending(io);
2186 }
2187 
2188 static void kcryptd_crypt_read_continue(struct work_struct *work)
2189 {
2190 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2191 	struct crypt_config *cc = io->cc;
2192 	blk_status_t r;
2193 
2194 	wait_for_completion(&io->ctx.restart);
2195 	reinit_completion(&io->ctx.restart);
2196 
2197 	r = crypt_convert(cc, &io->ctx, false, false);
2198 	if (r)
2199 		io->error = r;
2200 
2201 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2202 		kcryptd_crypt_read_done(io);
2203 
2204 	crypt_dec_pending(io);
2205 }
2206 
2207 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2208 {
2209 	struct crypt_config *cc = io->cc;
2210 	blk_status_t r;
2211 
2212 	crypt_inc_pending(io);
2213 
2214 	if (io->ctx.aead_recheck) {
2215 		r = crypt_convert(cc, &io->ctx,
2216 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2217 	} else {
2218 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2219 				   io->sector);
2220 
2221 		r = crypt_convert(cc, &io->ctx,
2222 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2223 	}
2224 	/*
2225 	 * Crypto API backlogged the request, because its queue was full
2226 	 * and we're in softirq context, so continue from a workqueue
2227 	 */
2228 	if (r == BLK_STS_DEV_RESOURCE) {
2229 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2230 		queue_work(cc->crypt_queue, &io->work);
2231 		return;
2232 	}
2233 	if (r)
2234 		io->error = r;
2235 
2236 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2237 		kcryptd_crypt_read_done(io);
2238 
2239 	crypt_dec_pending(io);
2240 }
2241 
2242 static void kcryptd_async_done(void *data, int error)
2243 {
2244 	struct dm_crypt_request *dmreq = data;
2245 	struct convert_context *ctx = dmreq->ctx;
2246 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2247 	struct crypt_config *cc = io->cc;
2248 
2249 	/*
2250 	 * A request from crypto driver backlog is going to be processed now,
2251 	 * finish the completion and continue in crypt_convert().
2252 	 * (Callback will be called for the second time for this request.)
2253 	 */
2254 	if (error == -EINPROGRESS) {
2255 		complete(&ctx->restart);
2256 		return;
2257 	}
2258 
2259 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2260 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2261 
2262 	if (error == -EBADMSG) {
2263 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2264 
2265 		ctx->aead_failed = true;
2266 		if (ctx->aead_recheck) {
2267 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2268 				    ctx->bio_in->bi_bdev, s);
2269 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2270 					 ctx->bio_in, s, 0);
2271 		}
2272 		io->error = BLK_STS_PROTECTION;
2273 	} else if (error < 0)
2274 		io->error = BLK_STS_IOERR;
2275 
2276 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2277 
2278 	if (!atomic_dec_and_test(&ctx->cc_pending))
2279 		return;
2280 
2281 	/*
2282 	 * The request is fully completed: for inline writes, let
2283 	 * kcryptd_crypt_write_convert() do the IO submission.
2284 	 */
2285 	if (bio_data_dir(io->base_bio) == READ) {
2286 		kcryptd_crypt_read_done(io);
2287 		return;
2288 	}
2289 
2290 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2291 		complete(&ctx->restart);
2292 		return;
2293 	}
2294 
2295 	kcryptd_crypt_write_io_submit(io, 1);
2296 }
2297 
2298 static void kcryptd_crypt(struct work_struct *work)
2299 {
2300 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2301 
2302 	if (bio_data_dir(io->base_bio) == READ)
2303 		kcryptd_crypt_read_convert(io);
2304 	else
2305 		kcryptd_crypt_write_convert(io);
2306 }
2307 
2308 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2309 {
2310 	struct crypt_config *cc = io->cc;
2311 
2312 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2313 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2314 		/*
2315 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2316 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2317 		 * it is being executed with irqs disabled.
2318 		 */
2319 		if (in_hardirq() || irqs_disabled()) {
2320 			INIT_WORK(&io->work, kcryptd_crypt);
2321 			queue_work(system_bh_wq, &io->work);
2322 			return;
2323 		} else {
2324 			kcryptd_crypt(&io->work);
2325 			return;
2326 		}
2327 	}
2328 
2329 	INIT_WORK(&io->work, kcryptd_crypt);
2330 	queue_work(cc->crypt_queue, &io->work);
2331 }
2332 
2333 static void crypt_free_tfms_aead(struct crypt_config *cc)
2334 {
2335 	if (!cc->cipher_tfm.tfms_aead)
2336 		return;
2337 
2338 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2339 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2340 		cc->cipher_tfm.tfms_aead[0] = NULL;
2341 	}
2342 
2343 	kfree(cc->cipher_tfm.tfms_aead);
2344 	cc->cipher_tfm.tfms_aead = NULL;
2345 }
2346 
2347 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2348 {
2349 	unsigned int i;
2350 
2351 	if (!cc->cipher_tfm.tfms)
2352 		return;
2353 
2354 	for (i = 0; i < cc->tfms_count; i++)
2355 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2356 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2357 			cc->cipher_tfm.tfms[i] = NULL;
2358 		}
2359 
2360 	kfree(cc->cipher_tfm.tfms);
2361 	cc->cipher_tfm.tfms = NULL;
2362 }
2363 
2364 static void crypt_free_tfms(struct crypt_config *cc)
2365 {
2366 	if (crypt_integrity_aead(cc))
2367 		crypt_free_tfms_aead(cc);
2368 	else
2369 		crypt_free_tfms_skcipher(cc);
2370 }
2371 
2372 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2373 {
2374 	unsigned int i;
2375 	int err;
2376 
2377 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2378 				      sizeof(struct crypto_skcipher *),
2379 				      GFP_KERNEL);
2380 	if (!cc->cipher_tfm.tfms)
2381 		return -ENOMEM;
2382 
2383 	for (i = 0; i < cc->tfms_count; i++) {
2384 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2385 						CRYPTO_ALG_ALLOCATES_MEMORY);
2386 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2387 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2388 			crypt_free_tfms(cc);
2389 			return err;
2390 		}
2391 	}
2392 
2393 	/*
2394 	 * dm-crypt performance can vary greatly depending on which crypto
2395 	 * algorithm implementation is used.  Help people debug performance
2396 	 * problems by logging the ->cra_driver_name.
2397 	 */
2398 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2399 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2400 	return 0;
2401 }
2402 
2403 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2404 {
2405 	int err;
2406 
2407 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2408 	if (!cc->cipher_tfm.tfms)
2409 		return -ENOMEM;
2410 
2411 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2412 						CRYPTO_ALG_ALLOCATES_MEMORY);
2413 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2414 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2415 		crypt_free_tfms(cc);
2416 		return err;
2417 	}
2418 
2419 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2420 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2421 	return 0;
2422 }
2423 
2424 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2425 {
2426 	if (crypt_integrity_aead(cc))
2427 		return crypt_alloc_tfms_aead(cc, ciphermode);
2428 	else
2429 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2430 }
2431 
2432 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2433 {
2434 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2435 }
2436 
2437 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2438 {
2439 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2440 }
2441 
2442 /*
2443  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2444  * the key must be for some reason in special format.
2445  * This funcion converts cc->key to this special format.
2446  */
2447 static void crypt_copy_authenckey(char *p, const void *key,
2448 				  unsigned int enckeylen, unsigned int authkeylen)
2449 {
2450 	struct crypto_authenc_key_param *param;
2451 	struct rtattr *rta;
2452 
2453 	rta = (struct rtattr *)p;
2454 	param = RTA_DATA(rta);
2455 	param->enckeylen = cpu_to_be32(enckeylen);
2456 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2457 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2458 	p += RTA_SPACE(sizeof(*param));
2459 	memcpy(p, key + enckeylen, authkeylen);
2460 	p += authkeylen;
2461 	memcpy(p, key, enckeylen);
2462 }
2463 
2464 static int crypt_setkey(struct crypt_config *cc)
2465 {
2466 	unsigned int subkey_size;
2467 	int err = 0, i, r;
2468 
2469 	/* Ignore extra keys (which are used for IV etc) */
2470 	subkey_size = crypt_subkey_size(cc);
2471 
2472 	if (crypt_integrity_hmac(cc)) {
2473 		if (subkey_size < cc->key_mac_size)
2474 			return -EINVAL;
2475 
2476 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2477 				      subkey_size - cc->key_mac_size,
2478 				      cc->key_mac_size);
2479 	}
2480 
2481 	for (i = 0; i < cc->tfms_count; i++) {
2482 		if (crypt_integrity_hmac(cc))
2483 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2484 				cc->authenc_key, crypt_authenckey_size(cc));
2485 		else if (crypt_integrity_aead(cc))
2486 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2487 					       cc->key + (i * subkey_size),
2488 					       subkey_size);
2489 		else
2490 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2491 						   cc->key + (i * subkey_size),
2492 						   subkey_size);
2493 		if (r)
2494 			err = r;
2495 	}
2496 
2497 	if (crypt_integrity_hmac(cc))
2498 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2499 
2500 	return err;
2501 }
2502 
2503 #ifdef CONFIG_KEYS
2504 
2505 static bool contains_whitespace(const char *str)
2506 {
2507 	while (*str)
2508 		if (isspace(*str++))
2509 			return true;
2510 	return false;
2511 }
2512 
2513 static int set_key_user(struct crypt_config *cc, struct key *key)
2514 {
2515 	const struct user_key_payload *ukp;
2516 
2517 	ukp = user_key_payload_locked(key);
2518 	if (!ukp)
2519 		return -EKEYREVOKED;
2520 
2521 	if (cc->key_size != ukp->datalen)
2522 		return -EINVAL;
2523 
2524 	memcpy(cc->key, ukp->data, cc->key_size);
2525 
2526 	return 0;
2527 }
2528 
2529 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2530 {
2531 	const struct encrypted_key_payload *ekp;
2532 
2533 	ekp = key->payload.data[0];
2534 	if (!ekp)
2535 		return -EKEYREVOKED;
2536 
2537 	if (cc->key_size != ekp->decrypted_datalen)
2538 		return -EINVAL;
2539 
2540 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2541 
2542 	return 0;
2543 }
2544 
2545 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2546 {
2547 	const struct trusted_key_payload *tkp;
2548 
2549 	tkp = key->payload.data[0];
2550 	if (!tkp)
2551 		return -EKEYREVOKED;
2552 
2553 	if (cc->key_size != tkp->key_len)
2554 		return -EINVAL;
2555 
2556 	memcpy(cc->key, tkp->key, cc->key_size);
2557 
2558 	return 0;
2559 }
2560 
2561 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2562 {
2563 	char *new_key_string, *key_desc;
2564 	int ret;
2565 	struct key_type *type;
2566 	struct key *key;
2567 	int (*set_key)(struct crypt_config *cc, struct key *key);
2568 
2569 	/*
2570 	 * Reject key_string with whitespace. dm core currently lacks code for
2571 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2572 	 */
2573 	if (contains_whitespace(key_string)) {
2574 		DMERR("whitespace chars not allowed in key string");
2575 		return -EINVAL;
2576 	}
2577 
2578 	/* look for next ':' separating key_type from key_description */
2579 	key_desc = strchr(key_string, ':');
2580 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2581 		return -EINVAL;
2582 
2583 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2584 		type = &key_type_logon;
2585 		set_key = set_key_user;
2586 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2587 		type = &key_type_user;
2588 		set_key = set_key_user;
2589 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2590 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2591 		type = &key_type_encrypted;
2592 		set_key = set_key_encrypted;
2593 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2594 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2595 		type = &key_type_trusted;
2596 		set_key = set_key_trusted;
2597 	} else {
2598 		return -EINVAL;
2599 	}
2600 
2601 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2602 	if (!new_key_string)
2603 		return -ENOMEM;
2604 
2605 	key = request_key(type, key_desc + 1, NULL);
2606 	if (IS_ERR(key)) {
2607 		ret = PTR_ERR(key);
2608 		goto free_new_key_string;
2609 	}
2610 
2611 	down_read(&key->sem);
2612 	ret = set_key(cc, key);
2613 	up_read(&key->sem);
2614 	key_put(key);
2615 	if (ret < 0)
2616 		goto free_new_key_string;
2617 
2618 	/* clear the flag since following operations may invalidate previously valid key */
2619 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2620 
2621 	ret = crypt_setkey(cc);
2622 	if (ret)
2623 		goto free_new_key_string;
2624 
2625 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2626 	kfree_sensitive(cc->key_string);
2627 	cc->key_string = new_key_string;
2628 	return 0;
2629 
2630 free_new_key_string:
2631 	kfree_sensitive(new_key_string);
2632 	return ret;
2633 }
2634 
2635 static int get_key_size(char **key_string)
2636 {
2637 	char *colon, dummy;
2638 	int ret;
2639 
2640 	if (*key_string[0] != ':')
2641 		return strlen(*key_string) >> 1;
2642 
2643 	/* look for next ':' in key string */
2644 	colon = strpbrk(*key_string + 1, ":");
2645 	if (!colon)
2646 		return -EINVAL;
2647 
2648 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2649 		return -EINVAL;
2650 
2651 	*key_string = colon;
2652 
2653 	/* remaining key string should be :<logon|user>:<key_desc> */
2654 
2655 	return ret;
2656 }
2657 
2658 #else
2659 
2660 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2661 {
2662 	return -EINVAL;
2663 }
2664 
2665 static int get_key_size(char **key_string)
2666 {
2667 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2668 }
2669 
2670 #endif /* CONFIG_KEYS */
2671 
2672 static int crypt_set_key(struct crypt_config *cc, char *key)
2673 {
2674 	int r = -EINVAL;
2675 	int key_string_len = strlen(key);
2676 
2677 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2678 	if (!cc->key_size && strcmp(key, "-"))
2679 		goto out;
2680 
2681 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2682 	if (key[0] == ':') {
2683 		r = crypt_set_keyring_key(cc, key + 1);
2684 		goto out;
2685 	}
2686 
2687 	/* clear the flag since following operations may invalidate previously valid key */
2688 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2689 
2690 	/* wipe references to any kernel keyring key */
2691 	kfree_sensitive(cc->key_string);
2692 	cc->key_string = NULL;
2693 
2694 	/* Decode key from its hex representation. */
2695 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2696 		goto out;
2697 
2698 	r = crypt_setkey(cc);
2699 	if (!r)
2700 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2701 
2702 out:
2703 	/* Hex key string not needed after here, so wipe it. */
2704 	memset(key, '0', key_string_len);
2705 
2706 	return r;
2707 }
2708 
2709 static int crypt_wipe_key(struct crypt_config *cc)
2710 {
2711 	int r;
2712 
2713 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2714 	get_random_bytes(&cc->key, cc->key_size);
2715 
2716 	/* Wipe IV private keys */
2717 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2718 		r = cc->iv_gen_ops->wipe(cc);
2719 		if (r)
2720 			return r;
2721 	}
2722 
2723 	kfree_sensitive(cc->key_string);
2724 	cc->key_string = NULL;
2725 	r = crypt_setkey(cc);
2726 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2727 
2728 	return r;
2729 }
2730 
2731 static void crypt_calculate_pages_per_client(void)
2732 {
2733 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2734 
2735 	if (!dm_crypt_clients_n)
2736 		return;
2737 
2738 	pages /= dm_crypt_clients_n;
2739 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2740 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2741 	dm_crypt_pages_per_client = pages;
2742 }
2743 
2744 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2745 {
2746 	struct crypt_config *cc = pool_data;
2747 	struct page *page;
2748 
2749 	/*
2750 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2751 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2752 	 * but avoids potential spinlock contention of an exact result.
2753 	 */
2754 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2755 	    likely(gfp_mask & __GFP_NORETRY))
2756 		return NULL;
2757 
2758 	page = alloc_page(gfp_mask);
2759 	if (likely(page != NULL))
2760 		percpu_counter_add(&cc->n_allocated_pages, 1);
2761 
2762 	return page;
2763 }
2764 
2765 static void crypt_page_free(void *page, void *pool_data)
2766 {
2767 	struct crypt_config *cc = pool_data;
2768 
2769 	__free_page(page);
2770 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2771 }
2772 
2773 static void crypt_dtr(struct dm_target *ti)
2774 {
2775 	struct crypt_config *cc = ti->private;
2776 
2777 	ti->private = NULL;
2778 
2779 	if (!cc)
2780 		return;
2781 
2782 	if (cc->write_thread)
2783 		kthread_stop(cc->write_thread);
2784 
2785 	if (cc->io_queue)
2786 		destroy_workqueue(cc->io_queue);
2787 	if (cc->crypt_queue)
2788 		destroy_workqueue(cc->crypt_queue);
2789 
2790 	if (cc->workqueue_id)
2791 		ida_free(&workqueue_ida, cc->workqueue_id);
2792 
2793 	crypt_free_tfms(cc);
2794 
2795 	bioset_exit(&cc->bs);
2796 
2797 	mempool_exit(&cc->page_pool);
2798 	mempool_exit(&cc->req_pool);
2799 	mempool_exit(&cc->tag_pool);
2800 
2801 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2802 	percpu_counter_destroy(&cc->n_allocated_pages);
2803 
2804 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2805 		cc->iv_gen_ops->dtr(cc);
2806 
2807 	if (cc->dev)
2808 		dm_put_device(ti, cc->dev);
2809 
2810 	kfree_sensitive(cc->cipher_string);
2811 	kfree_sensitive(cc->key_string);
2812 	kfree_sensitive(cc->cipher_auth);
2813 	kfree_sensitive(cc->authenc_key);
2814 
2815 	mutex_destroy(&cc->bio_alloc_lock);
2816 
2817 	/* Must zero key material before freeing */
2818 	kfree_sensitive(cc);
2819 
2820 	spin_lock(&dm_crypt_clients_lock);
2821 	WARN_ON(!dm_crypt_clients_n);
2822 	dm_crypt_clients_n--;
2823 	crypt_calculate_pages_per_client();
2824 	spin_unlock(&dm_crypt_clients_lock);
2825 
2826 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2827 }
2828 
2829 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2830 {
2831 	struct crypt_config *cc = ti->private;
2832 
2833 	if (crypt_integrity_aead(cc))
2834 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2835 	else
2836 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2837 
2838 	if (cc->iv_size)
2839 		/* at least a 64 bit sector number should fit in our buffer */
2840 		cc->iv_size = max(cc->iv_size,
2841 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2842 	else if (ivmode) {
2843 		DMWARN("Selected cipher does not support IVs");
2844 		ivmode = NULL;
2845 	}
2846 
2847 	/* Choose ivmode, see comments at iv code. */
2848 	if (ivmode == NULL)
2849 		cc->iv_gen_ops = NULL;
2850 	else if (strcmp(ivmode, "plain") == 0)
2851 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2852 	else if (strcmp(ivmode, "plain64") == 0)
2853 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2854 	else if (strcmp(ivmode, "plain64be") == 0)
2855 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2856 	else if (strcmp(ivmode, "essiv") == 0)
2857 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2858 	else if (strcmp(ivmode, "benbi") == 0)
2859 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2860 	else if (strcmp(ivmode, "null") == 0)
2861 		cc->iv_gen_ops = &crypt_iv_null_ops;
2862 	else if (strcmp(ivmode, "eboiv") == 0)
2863 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2864 	else if (strcmp(ivmode, "elephant") == 0) {
2865 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2866 		cc->key_parts = 2;
2867 		cc->key_extra_size = cc->key_size / 2;
2868 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2869 			return -EINVAL;
2870 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2871 	} else if (strcmp(ivmode, "lmk") == 0) {
2872 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2873 		/*
2874 		 * Version 2 and 3 is recognised according
2875 		 * to length of provided multi-key string.
2876 		 * If present (version 3), last key is used as IV seed.
2877 		 * All keys (including IV seed) are always the same size.
2878 		 */
2879 		if (cc->key_size % cc->key_parts) {
2880 			cc->key_parts++;
2881 			cc->key_extra_size = cc->key_size / cc->key_parts;
2882 		}
2883 	} else if (strcmp(ivmode, "tcw") == 0) {
2884 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2885 		cc->key_parts += 2; /* IV + whitening */
2886 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2887 	} else if (strcmp(ivmode, "random") == 0) {
2888 		cc->iv_gen_ops = &crypt_iv_random_ops;
2889 		/* Need storage space in integrity fields. */
2890 		cc->integrity_iv_size = cc->iv_size;
2891 	} else {
2892 		ti->error = "Invalid IV mode";
2893 		return -EINVAL;
2894 	}
2895 
2896 	return 0;
2897 }
2898 
2899 /*
2900  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2901  * The HMAC is needed to calculate tag size (HMAC digest size).
2902  * This should be probably done by crypto-api calls (once available...)
2903  */
2904 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2905 {
2906 	char *start, *end, *mac_alg = NULL;
2907 	struct crypto_ahash *mac;
2908 
2909 	if (!strstarts(cipher_api, "authenc("))
2910 		return 0;
2911 
2912 	start = strchr(cipher_api, '(');
2913 	end = strchr(cipher_api, ',');
2914 	if (!start || !end || ++start > end)
2915 		return -EINVAL;
2916 
2917 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2918 	if (!mac_alg)
2919 		return -ENOMEM;
2920 
2921 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2922 	kfree(mac_alg);
2923 
2924 	if (IS_ERR(mac))
2925 		return PTR_ERR(mac);
2926 
2927 	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2928 		cc->key_mac_size = crypto_ahash_digestsize(mac);
2929 	crypto_free_ahash(mac);
2930 
2931 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2932 	if (!cc->authenc_key)
2933 		return -ENOMEM;
2934 
2935 	return 0;
2936 }
2937 
2938 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2939 				char **ivmode, char **ivopts)
2940 {
2941 	struct crypt_config *cc = ti->private;
2942 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2943 	int ret = -EINVAL;
2944 
2945 	cc->tfms_count = 1;
2946 
2947 	/*
2948 	 * New format (capi: prefix)
2949 	 * capi:cipher_api_spec-iv:ivopts
2950 	 */
2951 	tmp = &cipher_in[strlen("capi:")];
2952 
2953 	/* Separate IV options if present, it can contain another '-' in hash name */
2954 	*ivopts = strrchr(tmp, ':');
2955 	if (*ivopts) {
2956 		**ivopts = '\0';
2957 		(*ivopts)++;
2958 	}
2959 	/* Parse IV mode */
2960 	*ivmode = strrchr(tmp, '-');
2961 	if (*ivmode) {
2962 		**ivmode = '\0';
2963 		(*ivmode)++;
2964 	}
2965 	/* The rest is crypto API spec */
2966 	cipher_api = tmp;
2967 
2968 	/* Alloc AEAD, can be used only in new format. */
2969 	if (crypt_integrity_aead(cc)) {
2970 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2971 		if (ret < 0) {
2972 			ti->error = "Invalid AEAD cipher spec";
2973 			return ret;
2974 		}
2975 	}
2976 
2977 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2978 		cc->tfms_count = 64;
2979 
2980 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2981 		if (!*ivopts) {
2982 			ti->error = "Digest algorithm missing for ESSIV mode";
2983 			return -EINVAL;
2984 		}
2985 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2986 			       cipher_api, *ivopts);
2987 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2988 			ti->error = "Cannot allocate cipher string";
2989 			return -ENOMEM;
2990 		}
2991 		cipher_api = buf;
2992 	}
2993 
2994 	cc->key_parts = cc->tfms_count;
2995 
2996 	/* Allocate cipher */
2997 	ret = crypt_alloc_tfms(cc, cipher_api);
2998 	if (ret < 0) {
2999 		ti->error = "Error allocating crypto tfm";
3000 		return ret;
3001 	}
3002 
3003 	if (crypt_integrity_aead(cc))
3004 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
3005 	else
3006 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
3007 
3008 	return 0;
3009 }
3010 
3011 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
3012 				char **ivmode, char **ivopts)
3013 {
3014 	struct crypt_config *cc = ti->private;
3015 	char *tmp, *cipher, *chainmode, *keycount;
3016 	char *cipher_api = NULL;
3017 	int ret = -EINVAL;
3018 	char dummy;
3019 
3020 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
3021 		ti->error = "Bad cipher specification";
3022 		return -EINVAL;
3023 	}
3024 
3025 	/*
3026 	 * Legacy dm-crypt cipher specification
3027 	 * cipher[:keycount]-mode-iv:ivopts
3028 	 */
3029 	tmp = cipher_in;
3030 	keycount = strsep(&tmp, "-");
3031 	cipher = strsep(&keycount, ":");
3032 
3033 	if (!keycount)
3034 		cc->tfms_count = 1;
3035 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3036 		 !is_power_of_2(cc->tfms_count)) {
3037 		ti->error = "Bad cipher key count specification";
3038 		return -EINVAL;
3039 	}
3040 	cc->key_parts = cc->tfms_count;
3041 
3042 	chainmode = strsep(&tmp, "-");
3043 	*ivmode = strsep(&tmp, ":");
3044 	*ivopts = tmp;
3045 
3046 	/*
3047 	 * For compatibility with the original dm-crypt mapping format, if
3048 	 * only the cipher name is supplied, use cbc-plain.
3049 	 */
3050 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3051 		chainmode = "cbc";
3052 		*ivmode = "plain";
3053 	}
3054 
3055 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3056 		ti->error = "IV mechanism required";
3057 		return -EINVAL;
3058 	}
3059 
3060 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3061 	if (!cipher_api)
3062 		goto bad_mem;
3063 
3064 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3065 		if (!*ivopts) {
3066 			ti->error = "Digest algorithm missing for ESSIV mode";
3067 			kfree(cipher_api);
3068 			return -EINVAL;
3069 		}
3070 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3071 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3072 	} else {
3073 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3074 			       "%s(%s)", chainmode, cipher);
3075 	}
3076 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3077 		kfree(cipher_api);
3078 		goto bad_mem;
3079 	}
3080 
3081 	/* Allocate cipher */
3082 	ret = crypt_alloc_tfms(cc, cipher_api);
3083 	if (ret < 0) {
3084 		ti->error = "Error allocating crypto tfm";
3085 		kfree(cipher_api);
3086 		return ret;
3087 	}
3088 	kfree(cipher_api);
3089 
3090 	return 0;
3091 bad_mem:
3092 	ti->error = "Cannot allocate cipher strings";
3093 	return -ENOMEM;
3094 }
3095 
3096 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3097 {
3098 	struct crypt_config *cc = ti->private;
3099 	char *ivmode = NULL, *ivopts = NULL;
3100 	int ret;
3101 
3102 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3103 	if (!cc->cipher_string) {
3104 		ti->error = "Cannot allocate cipher strings";
3105 		return -ENOMEM;
3106 	}
3107 
3108 	if (strstarts(cipher_in, "capi:"))
3109 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3110 	else
3111 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3112 	if (ret)
3113 		return ret;
3114 
3115 	/* Initialize IV */
3116 	ret = crypt_ctr_ivmode(ti, ivmode);
3117 	if (ret < 0)
3118 		return ret;
3119 
3120 	/* Initialize and set key */
3121 	ret = crypt_set_key(cc, key);
3122 	if (ret < 0) {
3123 		ti->error = "Error decoding and setting key";
3124 		return ret;
3125 	}
3126 
3127 	/* Allocate IV */
3128 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3129 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3130 		if (ret < 0) {
3131 			ti->error = "Error creating IV";
3132 			return ret;
3133 		}
3134 	}
3135 
3136 	/* Initialize IV (set keys for ESSIV etc) */
3137 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3138 		ret = cc->iv_gen_ops->init(cc);
3139 		if (ret < 0) {
3140 			ti->error = "Error initialising IV";
3141 			return ret;
3142 		}
3143 	}
3144 
3145 	/* wipe the kernel key payload copy */
3146 	if (cc->key_string)
3147 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3148 
3149 	return ret;
3150 }
3151 
3152 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3153 {
3154 	struct crypt_config *cc = ti->private;
3155 	struct dm_arg_set as;
3156 	static const struct dm_arg _args[] = {
3157 		{0, 9, "Invalid number of feature args"},
3158 	};
3159 	unsigned int opt_params, val;
3160 	const char *opt_string, *sval;
3161 	char dummy;
3162 	int ret;
3163 
3164 	/* Optional parameters */
3165 	as.argc = argc;
3166 	as.argv = argv;
3167 
3168 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3169 	if (ret)
3170 		return ret;
3171 
3172 	while (opt_params--) {
3173 		opt_string = dm_shift_arg(&as);
3174 		if (!opt_string) {
3175 			ti->error = "Not enough feature arguments";
3176 			return -EINVAL;
3177 		}
3178 
3179 		if (!strcasecmp(opt_string, "allow_discards"))
3180 			ti->num_discard_bios = 1;
3181 
3182 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3183 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3184 		else if (!strcasecmp(opt_string, "high_priority"))
3185 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3186 
3187 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3188 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3189 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3190 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3191 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3192 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3193 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3194 			if (val == 0 || val > MAX_TAG_SIZE) {
3195 				ti->error = "Invalid integrity arguments";
3196 				return -EINVAL;
3197 			}
3198 			cc->used_tag_size = val;
3199 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3200 			if (!strcasecmp(sval, "aead")) {
3201 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3202 			} else if (strcasecmp(sval, "none")) {
3203 				ti->error = "Unknown integrity profile";
3204 				return -EINVAL;
3205 			}
3206 
3207 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3208 			if (!cc->cipher_auth)
3209 				return -ENOMEM;
3210 		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3211 			if (!val) {
3212 				ti->error = "Invalid integrity_key_size argument";
3213 				return -EINVAL;
3214 			}
3215 			cc->key_mac_size = val;
3216 			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3217 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3218 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3219 			    cc->sector_size > 4096 ||
3220 			    (cc->sector_size & (cc->sector_size - 1))) {
3221 				ti->error = "Invalid feature value for sector_size";
3222 				return -EINVAL;
3223 			}
3224 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3225 				ti->error = "Device size is not multiple of sector_size feature";
3226 				return -EINVAL;
3227 			}
3228 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3229 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3230 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3231 		else {
3232 			ti->error = "Invalid feature arguments";
3233 			return -EINVAL;
3234 		}
3235 	}
3236 
3237 	return 0;
3238 }
3239 
3240 #ifdef CONFIG_BLK_DEV_ZONED
3241 static int crypt_report_zones(struct dm_target *ti,
3242 		struct dm_report_zones_args *args, unsigned int nr_zones)
3243 {
3244 	struct crypt_config *cc = ti->private;
3245 
3246 	return dm_report_zones(cc->dev->bdev, cc->start,
3247 			cc->start + dm_target_offset(ti, args->next_sector),
3248 			args, nr_zones);
3249 }
3250 #else
3251 #define crypt_report_zones NULL
3252 #endif
3253 
3254 /*
3255  * Construct an encryption mapping:
3256  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3257  */
3258 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3259 {
3260 	struct crypt_config *cc;
3261 	const char *devname = dm_table_device_name(ti->table);
3262 	int key_size, wq_id;
3263 	unsigned int align_mask;
3264 	unsigned int common_wq_flags;
3265 	unsigned long long tmpll;
3266 	int ret;
3267 	size_t iv_size_padding, additional_req_size;
3268 	char dummy;
3269 
3270 	if (argc < 5) {
3271 		ti->error = "Not enough arguments";
3272 		return -EINVAL;
3273 	}
3274 
3275 	key_size = get_key_size(&argv[1]);
3276 	if (key_size < 0) {
3277 		ti->error = "Cannot parse key size";
3278 		return -EINVAL;
3279 	}
3280 
3281 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3282 	if (!cc) {
3283 		ti->error = "Cannot allocate encryption context";
3284 		return -ENOMEM;
3285 	}
3286 	cc->key_size = key_size;
3287 	cc->sector_size = (1 << SECTOR_SHIFT);
3288 	cc->sector_shift = 0;
3289 
3290 	ti->private = cc;
3291 
3292 	spin_lock(&dm_crypt_clients_lock);
3293 	dm_crypt_clients_n++;
3294 	crypt_calculate_pages_per_client();
3295 	spin_unlock(&dm_crypt_clients_lock);
3296 
3297 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3298 	if (ret < 0)
3299 		goto bad;
3300 
3301 	/* Optional parameters need to be read before cipher constructor */
3302 	if (argc > 5) {
3303 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3304 		if (ret)
3305 			goto bad;
3306 	}
3307 
3308 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3309 	if (ret < 0)
3310 		goto bad;
3311 
3312 	if (crypt_integrity_aead(cc)) {
3313 		cc->dmreq_start = sizeof(struct aead_request);
3314 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3315 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3316 	} else {
3317 		cc->dmreq_start = sizeof(struct skcipher_request);
3318 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3319 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3320 	}
3321 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3322 
3323 	if (align_mask < CRYPTO_MINALIGN) {
3324 		/* Allocate the padding exactly */
3325 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3326 				& align_mask;
3327 	} else {
3328 		/*
3329 		 * If the cipher requires greater alignment than kmalloc
3330 		 * alignment, we don't know the exact position of the
3331 		 * initialization vector. We must assume worst case.
3332 		 */
3333 		iv_size_padding = align_mask;
3334 	}
3335 
3336 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3337 	additional_req_size = sizeof(struct dm_crypt_request) +
3338 		iv_size_padding + cc->iv_size +
3339 		cc->iv_size +
3340 		sizeof(uint64_t) +
3341 		sizeof(unsigned int);
3342 
3343 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3344 	if (ret) {
3345 		ti->error = "Cannot allocate crypt request mempool";
3346 		goto bad;
3347 	}
3348 
3349 	cc->per_bio_data_size = ti->per_io_data_size =
3350 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3351 		      ARCH_DMA_MINALIGN);
3352 
3353 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3354 	if (ret) {
3355 		ti->error = "Cannot allocate page mempool";
3356 		goto bad;
3357 	}
3358 
3359 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3360 	if (ret) {
3361 		ti->error = "Cannot allocate crypt bioset";
3362 		goto bad;
3363 	}
3364 
3365 	mutex_init(&cc->bio_alloc_lock);
3366 
3367 	ret = -EINVAL;
3368 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3369 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3370 		ti->error = "Invalid iv_offset sector";
3371 		goto bad;
3372 	}
3373 	cc->iv_offset = tmpll;
3374 
3375 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3376 	if (ret) {
3377 		ti->error = "Device lookup failed";
3378 		goto bad;
3379 	}
3380 
3381 	ret = -EINVAL;
3382 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3383 		ti->error = "Invalid device sector";
3384 		goto bad;
3385 	}
3386 	cc->start = tmpll;
3387 
3388 	if (bdev_is_zoned(cc->dev->bdev)) {
3389 		/*
3390 		 * For zoned block devices, we need to preserve the issuer write
3391 		 * ordering. To do so, disable write workqueues and force inline
3392 		 * encryption completion.
3393 		 */
3394 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3395 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3396 
3397 		/*
3398 		 * All zone append writes to a zone of a zoned block device will
3399 		 * have the same BIO sector, the start of the zone. When the
3400 		 * cypher IV mode uses sector values, all data targeting a
3401 		 * zone will be encrypted using the first sector numbers of the
3402 		 * zone. This will not result in write errors but will
3403 		 * cause most reads to fail as reads will use the sector values
3404 		 * for the actual data locations, resulting in IV mismatch.
3405 		 * To avoid this problem, ask DM core to emulate zone append
3406 		 * operations with regular writes.
3407 		 */
3408 		DMDEBUG("Zone append operations will be emulated");
3409 		ti->emulate_zone_append = true;
3410 	}
3411 
3412 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3413 		ret = crypt_integrity_ctr(cc, ti);
3414 		if (ret)
3415 			goto bad;
3416 
3417 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3418 		if (!cc->tag_pool_max_sectors)
3419 			cc->tag_pool_max_sectors = 1;
3420 
3421 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3422 			cc->tag_pool_max_sectors * cc->tuple_size);
3423 		if (ret) {
3424 			ti->error = "Cannot allocate integrity tags mempool";
3425 			goto bad;
3426 		}
3427 
3428 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3429 	}
3430 
3431 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3432 	if (wq_id < 0) {
3433 		ti->error = "Couldn't get workqueue id";
3434 		ret = wq_id;
3435 		goto bad;
3436 	}
3437 	cc->workqueue_id = wq_id;
3438 
3439 	ret = -ENOMEM;
3440 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3441 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3442 		common_wq_flags |= WQ_HIGHPRI;
3443 
3444 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", common_wq_flags, 1, devname, wq_id);
3445 	if (!cc->io_queue) {
3446 		ti->error = "Couldn't create kcryptd io queue";
3447 		goto bad;
3448 	}
3449 
3450 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3451 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3452 						  common_wq_flags | WQ_CPU_INTENSIVE,
3453 						  1, devname, wq_id);
3454 	} else {
3455 		/*
3456 		 * While crypt_queue is certainly CPU intensive, the use of
3457 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3458 		 */
3459 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3460 						  common_wq_flags | WQ_UNBOUND,
3461 						  num_online_cpus(), devname, wq_id);
3462 	}
3463 	if (!cc->crypt_queue) {
3464 		ti->error = "Couldn't create kcryptd queue";
3465 		goto bad;
3466 	}
3467 
3468 	spin_lock_init(&cc->write_thread_lock);
3469 	cc->write_tree = RB_ROOT;
3470 
3471 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3472 	if (IS_ERR(cc->write_thread)) {
3473 		ret = PTR_ERR(cc->write_thread);
3474 		cc->write_thread = NULL;
3475 		ti->error = "Couldn't spawn write thread";
3476 		goto bad;
3477 	}
3478 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3479 		set_user_nice(cc->write_thread, MIN_NICE);
3480 
3481 	ti->num_flush_bios = 1;
3482 	ti->limit_swap_bios = true;
3483 	ti->accounts_remapped_io = true;
3484 
3485 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3486 	return 0;
3487 
3488 bad:
3489 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3490 	crypt_dtr(ti);
3491 	return ret;
3492 }
3493 
3494 static int crypt_map(struct dm_target *ti, struct bio *bio)
3495 {
3496 	struct dm_crypt_io *io;
3497 	struct crypt_config *cc = ti->private;
3498 	unsigned max_sectors;
3499 
3500 	/*
3501 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3502 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3503 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3504 	 */
3505 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3506 	    bio_op(bio) == REQ_OP_DISCARD)) {
3507 		bio_set_dev(bio, cc->dev->bdev);
3508 		if (bio_sectors(bio))
3509 			bio->bi_iter.bi_sector = cc->start +
3510 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3511 		return DM_MAPIO_REMAPPED;
3512 	}
3513 
3514 	/*
3515 	 * Check if bio is too large, split as needed.
3516 	 */
3517 	max_sectors = get_max_request_size(cc, bio_data_dir(bio) == WRITE);
3518 	if (unlikely(bio_sectors(bio) > max_sectors))
3519 		dm_accept_partial_bio(bio, max_sectors);
3520 
3521 	/*
3522 	 * Ensure that bio is a multiple of internal sector encryption size
3523 	 * and is aligned to this size as defined in IO hints.
3524 	 */
3525 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3526 		return DM_MAPIO_KILL;
3527 
3528 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3529 		return DM_MAPIO_KILL;
3530 
3531 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3532 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3533 
3534 	if (cc->tuple_size) {
3535 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3536 
3537 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3538 			io->integrity_metadata = NULL;
3539 		else
3540 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3541 
3542 		if (unlikely(!io->integrity_metadata)) {
3543 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3544 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3545 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3546 			io->integrity_metadata_from_pool = true;
3547 		}
3548 	}
3549 
3550 	if (crypt_integrity_aead(cc))
3551 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3552 	else
3553 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3554 
3555 	if (bio_data_dir(io->base_bio) == READ) {
3556 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3557 			kcryptd_queue_read(io);
3558 	} else
3559 		kcryptd_queue_crypt(io);
3560 
3561 	return DM_MAPIO_SUBMITTED;
3562 }
3563 
3564 static char hex2asc(unsigned char c)
3565 {
3566 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3567 }
3568 
3569 static void crypt_status(struct dm_target *ti, status_type_t type,
3570 			 unsigned int status_flags, char *result, unsigned int maxlen)
3571 {
3572 	struct crypt_config *cc = ti->private;
3573 	unsigned int i, sz = 0;
3574 	int num_feature_args = 0;
3575 
3576 	switch (type) {
3577 	case STATUSTYPE_INFO:
3578 		result[0] = '\0';
3579 		break;
3580 
3581 	case STATUSTYPE_TABLE:
3582 		DMEMIT("%s ", cc->cipher_string);
3583 
3584 		if (cc->key_size > 0) {
3585 			if (cc->key_string)
3586 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3587 			else {
3588 				for (i = 0; i < cc->key_size; i++) {
3589 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3590 					       hex2asc(cc->key[i] & 0xf));
3591 				}
3592 			}
3593 		} else
3594 			DMEMIT("-");
3595 
3596 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3597 				cc->dev->name, (unsigned long long)cc->start);
3598 
3599 		num_feature_args += !!ti->num_discard_bios;
3600 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3601 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3602 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3603 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3604 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3605 		num_feature_args += !!cc->used_tag_size;
3606 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3607 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3608 		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3609 		if (num_feature_args) {
3610 			DMEMIT(" %d", num_feature_args);
3611 			if (ti->num_discard_bios)
3612 				DMEMIT(" allow_discards");
3613 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3614 				DMEMIT(" same_cpu_crypt");
3615 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3616 				DMEMIT(" high_priority");
3617 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3618 				DMEMIT(" submit_from_crypt_cpus");
3619 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3620 				DMEMIT(" no_read_workqueue");
3621 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3622 				DMEMIT(" no_write_workqueue");
3623 			if (cc->used_tag_size)
3624 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3625 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3626 				DMEMIT(" sector_size:%d", cc->sector_size);
3627 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3628 				DMEMIT(" iv_large_sectors");
3629 			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3630 				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3631 		}
3632 		break;
3633 
3634 	case STATUSTYPE_IMA:
3635 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3636 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3637 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3638 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3639 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3640 		       'y' : 'n');
3641 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3642 		       'y' : 'n');
3643 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3644 		       'y' : 'n');
3645 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3646 		       'y' : 'n');
3647 
3648 		if (cc->used_tag_size)
3649 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3650 			       cc->used_tag_size, cc->cipher_auth);
3651 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3652 			DMEMIT(",sector_size=%d", cc->sector_size);
3653 		if (cc->cipher_string)
3654 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3655 
3656 		DMEMIT(",key_size=%u", cc->key_size);
3657 		DMEMIT(",key_parts=%u", cc->key_parts);
3658 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3659 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3660 		DMEMIT(";");
3661 		break;
3662 	}
3663 }
3664 
3665 static void crypt_postsuspend(struct dm_target *ti)
3666 {
3667 	struct crypt_config *cc = ti->private;
3668 
3669 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3670 }
3671 
3672 static int crypt_preresume(struct dm_target *ti)
3673 {
3674 	struct crypt_config *cc = ti->private;
3675 
3676 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3677 		DMERR("aborting resume - crypt key is not set.");
3678 		return -EAGAIN;
3679 	}
3680 
3681 	return 0;
3682 }
3683 
3684 static void crypt_resume(struct dm_target *ti)
3685 {
3686 	struct crypt_config *cc = ti->private;
3687 
3688 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3689 }
3690 
3691 /* Message interface
3692  *	key set <key>
3693  *	key wipe
3694  */
3695 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3696 			 char *result, unsigned int maxlen)
3697 {
3698 	struct crypt_config *cc = ti->private;
3699 	int key_size, ret = -EINVAL;
3700 
3701 	if (argc < 2)
3702 		goto error;
3703 
3704 	if (!strcasecmp(argv[0], "key")) {
3705 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3706 			DMWARN("not suspended during key manipulation.");
3707 			return -EINVAL;
3708 		}
3709 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3710 			/* The key size may not be changed. */
3711 			key_size = get_key_size(&argv[2]);
3712 			if (key_size < 0 || cc->key_size != key_size) {
3713 				memset(argv[2], '0', strlen(argv[2]));
3714 				return -EINVAL;
3715 			}
3716 
3717 			ret = crypt_set_key(cc, argv[2]);
3718 			if (ret)
3719 				return ret;
3720 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3721 				ret = cc->iv_gen_ops->init(cc);
3722 			/* wipe the kernel key payload copy */
3723 			if (cc->key_string)
3724 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3725 			return ret;
3726 		}
3727 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3728 			return crypt_wipe_key(cc);
3729 	}
3730 
3731 error:
3732 	DMWARN("unrecognised message received.");
3733 	return -EINVAL;
3734 }
3735 
3736 static int crypt_iterate_devices(struct dm_target *ti,
3737 				 iterate_devices_callout_fn fn, void *data)
3738 {
3739 	struct crypt_config *cc = ti->private;
3740 
3741 	return fn(ti, cc->dev, cc->start, ti->len, data);
3742 }
3743 
3744 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3745 {
3746 	struct crypt_config *cc = ti->private;
3747 
3748 	limits->logical_block_size =
3749 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3750 	limits->physical_block_size =
3751 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3752 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3753 	limits->dma_alignment = limits->logical_block_size - 1;
3754 }
3755 
3756 static struct target_type crypt_target = {
3757 	.name   = "crypt",
3758 	.version = {1, 28, 0},
3759 	.module = THIS_MODULE,
3760 	.ctr    = crypt_ctr,
3761 	.dtr    = crypt_dtr,
3762 	.features = DM_TARGET_ZONED_HM,
3763 	.report_zones = crypt_report_zones,
3764 	.map    = crypt_map,
3765 	.status = crypt_status,
3766 	.postsuspend = crypt_postsuspend,
3767 	.preresume = crypt_preresume,
3768 	.resume = crypt_resume,
3769 	.message = crypt_message,
3770 	.iterate_devices = crypt_iterate_devices,
3771 	.io_hints = crypt_io_hints,
3772 };
3773 module_dm(crypt);
3774 
3775 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3776 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3777 MODULE_LICENSE("GPL");
3778