xref: /linux/drivers/md/dm-crypt.c (revision 55d0969c451159cff86949b38c39171cab962069)
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 	u64 cc_sector;
63 	union {
64 		struct skcipher_request *req;
65 		struct aead_request *req_aead;
66 	} r;
67 	bool aead_recheck;
68 	bool aead_failed;
69 
70 };
71 
72 /*
73  * per bio private data
74  */
75 struct dm_crypt_io {
76 	struct crypt_config *cc;
77 	struct bio *base_bio;
78 	u8 *integrity_metadata;
79 	bool integrity_metadata_from_pool:1;
80 
81 	struct work_struct work;
82 
83 	struct convert_context ctx;
84 
85 	atomic_t io_pending;
86 	blk_status_t error;
87 	sector_t sector;
88 
89 	struct bvec_iter saved_bi_iter;
90 
91 	struct rb_node rb_node;
92 } CRYPTO_MINALIGN_ATTR;
93 
94 struct dm_crypt_request {
95 	struct convert_context *ctx;
96 	struct scatterlist sg_in[4];
97 	struct scatterlist sg_out[4];
98 	u64 iv_sector;
99 };
100 
101 struct crypt_config;
102 
103 struct crypt_iv_operations {
104 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
105 		   const char *opts);
106 	void (*dtr)(struct crypt_config *cc);
107 	int (*init)(struct crypt_config *cc);
108 	int (*wipe)(struct crypt_config *cc);
109 	int (*generator)(struct crypt_config *cc, u8 *iv,
110 			 struct dm_crypt_request *dmreq);
111 	int (*post)(struct crypt_config *cc, u8 *iv,
112 		    struct dm_crypt_request *dmreq);
113 };
114 
115 struct iv_benbi_private {
116 	int shift;
117 };
118 
119 #define LMK_SEED_SIZE 64 /* hash + 0 */
120 struct iv_lmk_private {
121 	struct crypto_shash *hash_tfm;
122 	u8 *seed;
123 };
124 
125 #define TCW_WHITENING_SIZE 16
126 struct iv_tcw_private {
127 	struct crypto_shash *crc32_tfm;
128 	u8 *iv_seed;
129 	u8 *whitening;
130 };
131 
132 #define ELEPHANT_MAX_KEY_SIZE 32
133 struct iv_elephant_private {
134 	struct crypto_skcipher *tfm;
135 };
136 
137 /*
138  * Crypt: maps a linear range of a block device
139  * and encrypts / decrypts at the same time.
140  */
141 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
142 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
143 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
144 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
145 
146 enum cipher_flags {
147 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
148 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
149 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
150 	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
151 };
152 
153 /*
154  * The fields in here must be read only after initialization.
155  */
156 struct crypt_config {
157 	struct dm_dev *dev;
158 	sector_t start;
159 
160 	struct percpu_counter n_allocated_pages;
161 
162 	struct workqueue_struct *io_queue;
163 	struct workqueue_struct *crypt_queue;
164 
165 	spinlock_t write_thread_lock;
166 	struct task_struct *write_thread;
167 	struct rb_root write_tree;
168 
169 	char *cipher_string;
170 	char *cipher_auth;
171 	char *key_string;
172 
173 	const struct crypt_iv_operations *iv_gen_ops;
174 	union {
175 		struct iv_benbi_private benbi;
176 		struct iv_lmk_private lmk;
177 		struct iv_tcw_private tcw;
178 		struct iv_elephant_private elephant;
179 	} iv_gen_private;
180 	u64 iv_offset;
181 	unsigned int iv_size;
182 	unsigned short sector_size;
183 	unsigned char sector_shift;
184 
185 	union {
186 		struct crypto_skcipher **tfms;
187 		struct crypto_aead **tfms_aead;
188 	} cipher_tfm;
189 	unsigned int tfms_count;
190 	int workqueue_id;
191 	unsigned long cipher_flags;
192 
193 	/*
194 	 * Layout of each crypto request:
195 	 *
196 	 *   struct skcipher_request
197 	 *      context
198 	 *      padding
199 	 *   struct dm_crypt_request
200 	 *      padding
201 	 *   IV
202 	 *
203 	 * The padding is added so that dm_crypt_request and the IV are
204 	 * correctly aligned.
205 	 */
206 	unsigned int dmreq_start;
207 
208 	unsigned int per_bio_data_size;
209 
210 	unsigned long flags;
211 	unsigned int key_size;
212 	unsigned int key_parts;      /* independent parts in key buffer */
213 	unsigned int key_extra_size; /* additional keys length */
214 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
215 
216 	unsigned int integrity_tag_size;
217 	unsigned int integrity_iv_size;
218 	unsigned int used_tag_size;
219 	unsigned int tuple_size;
220 
221 	/*
222 	 * pool for per bio private data, crypto requests,
223 	 * encryption requeusts/buffer pages and integrity tags
224 	 */
225 	unsigned int tag_pool_max_sectors;
226 	mempool_t tag_pool;
227 	mempool_t req_pool;
228 	mempool_t page_pool;
229 
230 	struct bio_set bs;
231 	struct mutex bio_alloc_lock;
232 
233 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
234 	u8 key[] __counted_by(key_size);
235 };
236 
237 #define MIN_IOS		64
238 #define MAX_TAG_SIZE	480
239 #define POOL_ENTRY_SIZE	512
240 
241 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
242 static unsigned int dm_crypt_clients_n;
243 static volatile unsigned long dm_crypt_pages_per_client;
244 #define DM_CRYPT_MEMORY_PERCENT			2
245 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
246 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
247 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
248 
249 static unsigned int max_read_size = 0;
250 module_param(max_read_size, uint, 0644);
251 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
252 static unsigned int max_write_size = 0;
253 module_param(max_write_size, uint, 0644);
254 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
255 static unsigned get_max_request_size(struct crypt_config *cc, bool wrt)
256 {
257 	unsigned val, sector_align;
258 	val = !wrt ? READ_ONCE(max_read_size) : READ_ONCE(max_write_size);
259 	if (likely(!val))
260 		val = !wrt ? DM_CRYPT_DEFAULT_MAX_READ_SIZE : DM_CRYPT_DEFAULT_MAX_WRITE_SIZE;
261 	if (wrt || cc->used_tag_size) {
262 		if (unlikely(val > BIO_MAX_VECS << PAGE_SHIFT))
263 			val = BIO_MAX_VECS << PAGE_SHIFT;
264 	}
265 	sector_align = max(bdev_logical_block_size(cc->dev->bdev), (unsigned)cc->sector_size);
266 	val = round_down(val, sector_align);
267 	if (unlikely(!val))
268 		val = sector_align;
269 	return val >> SECTOR_SHIFT;
270 }
271 
272 static void crypt_endio(struct bio *clone);
273 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
274 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
275 					     struct scatterlist *sg);
276 
277 static bool crypt_integrity_aead(struct crypt_config *cc);
278 
279 /*
280  * Use this to access cipher attributes that are independent of the key.
281  */
282 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
283 {
284 	return cc->cipher_tfm.tfms[0];
285 }
286 
287 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
288 {
289 	return cc->cipher_tfm.tfms_aead[0];
290 }
291 
292 /*
293  * Different IV generation algorithms:
294  *
295  * plain: the initial vector is the 32-bit little-endian version of the sector
296  *        number, padded with zeros if necessary.
297  *
298  * plain64: the initial vector is the 64-bit little-endian version of the sector
299  *        number, padded with zeros if necessary.
300  *
301  * plain64be: the initial vector is the 64-bit big-endian version of the sector
302  *        number, padded with zeros if necessary.
303  *
304  * essiv: "encrypted sector|salt initial vector", the sector number is
305  *        encrypted with the bulk cipher using a salt as key. The salt
306  *        should be derived from the bulk cipher's key via hashing.
307  *
308  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
309  *        (needed for LRW-32-AES and possible other narrow block modes)
310  *
311  * null: the initial vector is always zero.  Provides compatibility with
312  *       obsolete loop_fish2 devices.  Do not use for new devices.
313  *
314  * lmk:  Compatible implementation of the block chaining mode used
315  *       by the Loop-AES block device encryption system
316  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
317  *       It operates on full 512 byte sectors and uses CBC
318  *       with an IV derived from the sector number, the data and
319  *       optionally extra IV seed.
320  *       This means that after decryption the first block
321  *       of sector must be tweaked according to decrypted data.
322  *       Loop-AES can use three encryption schemes:
323  *         version 1: is plain aes-cbc mode
324  *         version 2: uses 64 multikey scheme with lmk IV generator
325  *         version 3: the same as version 2 with additional IV seed
326  *                   (it uses 65 keys, last key is used as IV seed)
327  *
328  * tcw:  Compatible implementation of the block chaining mode used
329  *       by the TrueCrypt device encryption system (prior to version 4.1).
330  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
331  *       It operates on full 512 byte sectors and uses CBC
332  *       with an IV derived from initial key and the sector number.
333  *       In addition, whitening value is applied on every sector, whitening
334  *       is calculated from initial key, sector number and mixed using CRC32.
335  *       Note that this encryption scheme is vulnerable to watermarking attacks
336  *       and should be used for old compatible containers access only.
337  *
338  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
339  *        The IV is encrypted little-endian byte-offset (with the same key
340  *        and cipher as the volume).
341  *
342  * elephant: The extended version of eboiv with additional Elephant diffuser
343  *           used with Bitlocker CBC mode.
344  *           This mode was used in older Windows systems
345  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
346  */
347 
348 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
349 			      struct dm_crypt_request *dmreq)
350 {
351 	memset(iv, 0, cc->iv_size);
352 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
353 
354 	return 0;
355 }
356 
357 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
358 				struct dm_crypt_request *dmreq)
359 {
360 	memset(iv, 0, cc->iv_size);
361 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
362 
363 	return 0;
364 }
365 
366 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
367 				  struct dm_crypt_request *dmreq)
368 {
369 	memset(iv, 0, cc->iv_size);
370 	/* iv_size is at least of size u64; usually it is 16 bytes */
371 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
372 
373 	return 0;
374 }
375 
376 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
377 			      struct dm_crypt_request *dmreq)
378 {
379 	/*
380 	 * ESSIV encryption of the IV is now handled by the crypto API,
381 	 * so just pass the plain sector number here.
382 	 */
383 	memset(iv, 0, cc->iv_size);
384 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
385 
386 	return 0;
387 }
388 
389 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
390 			      const char *opts)
391 {
392 	unsigned int bs;
393 	int log;
394 
395 	if (crypt_integrity_aead(cc))
396 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
397 	else
398 		bs = crypto_skcipher_blocksize(any_tfm(cc));
399 	log = ilog2(bs);
400 
401 	/*
402 	 * We need to calculate how far we must shift the sector count
403 	 * to get the cipher block count, we use this shift in _gen.
404 	 */
405 	if (1 << log != bs) {
406 		ti->error = "cypher blocksize is not a power of 2";
407 		return -EINVAL;
408 	}
409 
410 	if (log > 9) {
411 		ti->error = "cypher blocksize is > 512";
412 		return -EINVAL;
413 	}
414 
415 	cc->iv_gen_private.benbi.shift = 9 - log;
416 
417 	return 0;
418 }
419 
420 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
421 {
422 }
423 
424 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
425 			      struct dm_crypt_request *dmreq)
426 {
427 	__be64 val;
428 
429 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
430 
431 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
432 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
433 
434 	return 0;
435 }
436 
437 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
438 			     struct dm_crypt_request *dmreq)
439 {
440 	memset(iv, 0, cc->iv_size);
441 
442 	return 0;
443 }
444 
445 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
446 {
447 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
448 
449 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
450 		crypto_free_shash(lmk->hash_tfm);
451 	lmk->hash_tfm = NULL;
452 
453 	kfree_sensitive(lmk->seed);
454 	lmk->seed = NULL;
455 }
456 
457 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
458 			    const char *opts)
459 {
460 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
461 
462 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
463 		ti->error = "Unsupported sector size for LMK";
464 		return -EINVAL;
465 	}
466 
467 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
468 					   CRYPTO_ALG_ALLOCATES_MEMORY);
469 	if (IS_ERR(lmk->hash_tfm)) {
470 		ti->error = "Error initializing LMK hash";
471 		return PTR_ERR(lmk->hash_tfm);
472 	}
473 
474 	/* No seed in LMK version 2 */
475 	if (cc->key_parts == cc->tfms_count) {
476 		lmk->seed = NULL;
477 		return 0;
478 	}
479 
480 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
481 	if (!lmk->seed) {
482 		crypt_iv_lmk_dtr(cc);
483 		ti->error = "Error kmallocing seed storage in LMK";
484 		return -ENOMEM;
485 	}
486 
487 	return 0;
488 }
489 
490 static int crypt_iv_lmk_init(struct crypt_config *cc)
491 {
492 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
493 	int subkey_size = cc->key_size / cc->key_parts;
494 
495 	/* LMK seed is on the position of LMK_KEYS + 1 key */
496 	if (lmk->seed)
497 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
498 		       crypto_shash_digestsize(lmk->hash_tfm));
499 
500 	return 0;
501 }
502 
503 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
504 {
505 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
506 
507 	if (lmk->seed)
508 		memset(lmk->seed, 0, LMK_SEED_SIZE);
509 
510 	return 0;
511 }
512 
513 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
514 			    struct dm_crypt_request *dmreq,
515 			    u8 *data)
516 {
517 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
518 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
519 	struct md5_state md5state;
520 	__le32 buf[4];
521 	int i, r;
522 
523 	desc->tfm = lmk->hash_tfm;
524 
525 	r = crypto_shash_init(desc);
526 	if (r)
527 		return r;
528 
529 	if (lmk->seed) {
530 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
531 		if (r)
532 			return r;
533 	}
534 
535 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
536 	r = crypto_shash_update(desc, data + 16, 16 * 31);
537 	if (r)
538 		return r;
539 
540 	/* Sector is cropped to 56 bits here */
541 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
542 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
543 	buf[2] = cpu_to_le32(4024);
544 	buf[3] = 0;
545 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
546 	if (r)
547 		return r;
548 
549 	/* No MD5 padding here */
550 	r = crypto_shash_export(desc, &md5state);
551 	if (r)
552 		return r;
553 
554 	for (i = 0; i < MD5_HASH_WORDS; i++)
555 		__cpu_to_le32s(&md5state.hash[i]);
556 	memcpy(iv, &md5state.hash, cc->iv_size);
557 
558 	return 0;
559 }
560 
561 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
562 			    struct dm_crypt_request *dmreq)
563 {
564 	struct scatterlist *sg;
565 	u8 *src;
566 	int r = 0;
567 
568 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
569 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
570 		src = kmap_local_page(sg_page(sg));
571 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
572 		kunmap_local(src);
573 	} else
574 		memset(iv, 0, cc->iv_size);
575 
576 	return r;
577 }
578 
579 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
580 			     struct dm_crypt_request *dmreq)
581 {
582 	struct scatterlist *sg;
583 	u8 *dst;
584 	int r;
585 
586 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
587 		return 0;
588 
589 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
590 	dst = kmap_local_page(sg_page(sg));
591 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
592 
593 	/* Tweak the first block of plaintext sector */
594 	if (!r)
595 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
596 
597 	kunmap_local(dst);
598 	return r;
599 }
600 
601 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
602 {
603 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
604 
605 	kfree_sensitive(tcw->iv_seed);
606 	tcw->iv_seed = NULL;
607 	kfree_sensitive(tcw->whitening);
608 	tcw->whitening = NULL;
609 
610 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
611 		crypto_free_shash(tcw->crc32_tfm);
612 	tcw->crc32_tfm = NULL;
613 }
614 
615 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
616 			    const char *opts)
617 {
618 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
619 
620 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
621 		ti->error = "Unsupported sector size for TCW";
622 		return -EINVAL;
623 	}
624 
625 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
626 		ti->error = "Wrong key size for TCW";
627 		return -EINVAL;
628 	}
629 
630 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
631 					    CRYPTO_ALG_ALLOCATES_MEMORY);
632 	if (IS_ERR(tcw->crc32_tfm)) {
633 		ti->error = "Error initializing CRC32 in TCW";
634 		return PTR_ERR(tcw->crc32_tfm);
635 	}
636 
637 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
638 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
639 	if (!tcw->iv_seed || !tcw->whitening) {
640 		crypt_iv_tcw_dtr(cc);
641 		ti->error = "Error allocating seed storage in TCW";
642 		return -ENOMEM;
643 	}
644 
645 	return 0;
646 }
647 
648 static int crypt_iv_tcw_init(struct crypt_config *cc)
649 {
650 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
651 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
652 
653 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
654 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
655 	       TCW_WHITENING_SIZE);
656 
657 	return 0;
658 }
659 
660 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
661 {
662 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
663 
664 	memset(tcw->iv_seed, 0, cc->iv_size);
665 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
666 
667 	return 0;
668 }
669 
670 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
671 				  struct dm_crypt_request *dmreq,
672 				  u8 *data)
673 {
674 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
675 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
676 	u8 buf[TCW_WHITENING_SIZE];
677 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
678 	int i, r;
679 
680 	/* xor whitening with sector number */
681 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
682 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
683 
684 	/* calculate crc32 for every 32bit part and xor it */
685 	desc->tfm = tcw->crc32_tfm;
686 	for (i = 0; i < 4; i++) {
687 		r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
688 		if (r)
689 			goto out;
690 	}
691 	crypto_xor(&buf[0], &buf[12], 4);
692 	crypto_xor(&buf[4], &buf[8], 4);
693 
694 	/* apply whitening (8 bytes) to whole sector */
695 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
696 		crypto_xor(data + i * 8, buf, 8);
697 out:
698 	memzero_explicit(buf, sizeof(buf));
699 	return r;
700 }
701 
702 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
703 			    struct dm_crypt_request *dmreq)
704 {
705 	struct scatterlist *sg;
706 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
707 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
708 	u8 *src;
709 	int r = 0;
710 
711 	/* Remove whitening from ciphertext */
712 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
713 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
714 		src = kmap_local_page(sg_page(sg));
715 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
716 		kunmap_local(src);
717 	}
718 
719 	/* Calculate IV */
720 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
721 	if (cc->iv_size > 8)
722 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
723 			       cc->iv_size - 8);
724 
725 	return r;
726 }
727 
728 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
729 			     struct dm_crypt_request *dmreq)
730 {
731 	struct scatterlist *sg;
732 	u8 *dst;
733 	int r;
734 
735 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
736 		return 0;
737 
738 	/* Apply whitening on ciphertext */
739 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
740 	dst = kmap_local_page(sg_page(sg));
741 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
742 	kunmap_local(dst);
743 
744 	return r;
745 }
746 
747 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
748 				struct dm_crypt_request *dmreq)
749 {
750 	/* Used only for writes, there must be an additional space to store IV */
751 	get_random_bytes(iv, cc->iv_size);
752 	return 0;
753 }
754 
755 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
756 			    const char *opts)
757 {
758 	if (crypt_integrity_aead(cc)) {
759 		ti->error = "AEAD transforms not supported for EBOIV";
760 		return -EINVAL;
761 	}
762 
763 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
764 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
765 		return -EINVAL;
766 	}
767 
768 	return 0;
769 }
770 
771 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
772 			    struct dm_crypt_request *dmreq)
773 {
774 	struct crypto_skcipher *tfm = any_tfm(cc);
775 	struct skcipher_request *req;
776 	struct scatterlist src, dst;
777 	DECLARE_CRYPTO_WAIT(wait);
778 	unsigned int reqsize;
779 	int err;
780 	u8 *buf;
781 
782 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
783 	reqsize = ALIGN(reqsize, __alignof__(__le64));
784 
785 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
786 	if (!req)
787 		return -ENOMEM;
788 
789 	skcipher_request_set_tfm(req, tfm);
790 
791 	buf = (u8 *)req + reqsize;
792 	memset(buf, 0, cc->iv_size);
793 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
794 
795 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
796 	sg_init_one(&dst, iv, cc->iv_size);
797 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
798 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
799 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
800 	kfree_sensitive(req);
801 
802 	return err;
803 }
804 
805 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
806 {
807 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
808 
809 	crypto_free_skcipher(elephant->tfm);
810 	elephant->tfm = NULL;
811 }
812 
813 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
814 			    const char *opts)
815 {
816 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
817 	int r;
818 
819 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
820 					      CRYPTO_ALG_ALLOCATES_MEMORY);
821 	if (IS_ERR(elephant->tfm)) {
822 		r = PTR_ERR(elephant->tfm);
823 		elephant->tfm = NULL;
824 		return r;
825 	}
826 
827 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
828 	if (r)
829 		crypt_iv_elephant_dtr(cc);
830 	return r;
831 }
832 
833 static void diffuser_disk_to_cpu(u32 *d, size_t n)
834 {
835 #ifndef __LITTLE_ENDIAN
836 	int i;
837 
838 	for (i = 0; i < n; i++)
839 		d[i] = le32_to_cpu((__le32)d[i]);
840 #endif
841 }
842 
843 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
844 {
845 #ifndef __LITTLE_ENDIAN
846 	int i;
847 
848 	for (i = 0; i < n; i++)
849 		d[i] = cpu_to_le32((u32)d[i]);
850 #endif
851 }
852 
853 static void diffuser_a_decrypt(u32 *d, size_t n)
854 {
855 	int i, i1, i2, i3;
856 
857 	for (i = 0; i < 5; i++) {
858 		i1 = 0;
859 		i2 = n - 2;
860 		i3 = n - 5;
861 
862 		while (i1 < (n - 1)) {
863 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
864 			i1++; i2++; i3++;
865 
866 			if (i3 >= n)
867 				i3 -= n;
868 
869 			d[i1] += d[i2] ^ d[i3];
870 			i1++; i2++; i3++;
871 
872 			if (i2 >= n)
873 				i2 -= n;
874 
875 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
876 			i1++; i2++; i3++;
877 
878 			d[i1] += d[i2] ^ d[i3];
879 			i1++; i2++; i3++;
880 		}
881 	}
882 }
883 
884 static void diffuser_a_encrypt(u32 *d, size_t n)
885 {
886 	int i, i1, i2, i3;
887 
888 	for (i = 0; i < 5; i++) {
889 		i1 = n - 1;
890 		i2 = n - 2 - 1;
891 		i3 = n - 5 - 1;
892 
893 		while (i1 > 0) {
894 			d[i1] -= d[i2] ^ d[i3];
895 			i1--; i2--; i3--;
896 
897 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
898 			i1--; i2--; i3--;
899 
900 			if (i2 < 0)
901 				i2 += n;
902 
903 			d[i1] -= d[i2] ^ d[i3];
904 			i1--; i2--; i3--;
905 
906 			if (i3 < 0)
907 				i3 += n;
908 
909 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
910 			i1--; i2--; i3--;
911 		}
912 	}
913 }
914 
915 static void diffuser_b_decrypt(u32 *d, size_t n)
916 {
917 	int i, i1, i2, i3;
918 
919 	for (i = 0; i < 3; i++) {
920 		i1 = 0;
921 		i2 = 2;
922 		i3 = 5;
923 
924 		while (i1 < (n - 1)) {
925 			d[i1] += d[i2] ^ d[i3];
926 			i1++; i2++; i3++;
927 
928 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
929 			i1++; i2++; i3++;
930 
931 			if (i2 >= n)
932 				i2 -= n;
933 
934 			d[i1] += d[i2] ^ d[i3];
935 			i1++; i2++; i3++;
936 
937 			if (i3 >= n)
938 				i3 -= n;
939 
940 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
941 			i1++; i2++; i3++;
942 		}
943 	}
944 }
945 
946 static void diffuser_b_encrypt(u32 *d, size_t n)
947 {
948 	int i, i1, i2, i3;
949 
950 	for (i = 0; i < 3; i++) {
951 		i1 = n - 1;
952 		i2 = 2 - 1;
953 		i3 = 5 - 1;
954 
955 		while (i1 > 0) {
956 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
957 			i1--; i2--; i3--;
958 
959 			if (i3 < 0)
960 				i3 += n;
961 
962 			d[i1] -= d[i2] ^ d[i3];
963 			i1--; i2--; i3--;
964 
965 			if (i2 < 0)
966 				i2 += n;
967 
968 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
969 			i1--; i2--; i3--;
970 
971 			d[i1] -= d[i2] ^ d[i3];
972 			i1--; i2--; i3--;
973 		}
974 	}
975 }
976 
977 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
978 {
979 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
980 	u8 *es, *ks, *data, *data2, *data_offset;
981 	struct skcipher_request *req;
982 	struct scatterlist *sg, *sg2, src, dst;
983 	DECLARE_CRYPTO_WAIT(wait);
984 	int i, r;
985 
986 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
987 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
988 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
989 
990 	if (!req || !es || !ks) {
991 		r = -ENOMEM;
992 		goto out;
993 	}
994 
995 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
996 
997 	/* E(Ks, e(s)) */
998 	sg_init_one(&src, es, 16);
999 	sg_init_one(&dst, ks, 16);
1000 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
1001 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
1002 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1003 	if (r)
1004 		goto out;
1005 
1006 	/* E(Ks, e'(s)) */
1007 	es[15] = 0x80;
1008 	sg_init_one(&dst, &ks[16], 16);
1009 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
1010 	if (r)
1011 		goto out;
1012 
1013 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
1014 	data = kmap_local_page(sg_page(sg));
1015 	data_offset = data + sg->offset;
1016 
1017 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
1018 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1019 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
1020 		data2 = kmap_local_page(sg_page(sg2));
1021 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
1022 		kunmap_local(data2);
1023 	}
1024 
1025 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
1026 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1027 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1028 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1029 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1030 	}
1031 
1032 	for (i = 0; i < (cc->sector_size / 32); i++)
1033 		crypto_xor(data_offset + i * 32, ks, 32);
1034 
1035 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1036 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1037 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1038 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1039 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1040 	}
1041 
1042 	kunmap_local(data);
1043 out:
1044 	kfree_sensitive(ks);
1045 	kfree_sensitive(es);
1046 	skcipher_request_free(req);
1047 	return r;
1048 }
1049 
1050 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1051 			    struct dm_crypt_request *dmreq)
1052 {
1053 	int r;
1054 
1055 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1056 		r = crypt_iv_elephant(cc, dmreq);
1057 		if (r)
1058 			return r;
1059 	}
1060 
1061 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1062 }
1063 
1064 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1065 				  struct dm_crypt_request *dmreq)
1066 {
1067 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1068 		return crypt_iv_elephant(cc, dmreq);
1069 
1070 	return 0;
1071 }
1072 
1073 static int crypt_iv_elephant_init(struct crypt_config *cc)
1074 {
1075 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1076 	int key_offset = cc->key_size - cc->key_extra_size;
1077 
1078 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1079 }
1080 
1081 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1082 {
1083 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1084 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1085 
1086 	memset(key, 0, cc->key_extra_size);
1087 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1088 }
1089 
1090 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1091 	.generator = crypt_iv_plain_gen
1092 };
1093 
1094 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1095 	.generator = crypt_iv_plain64_gen
1096 };
1097 
1098 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1099 	.generator = crypt_iv_plain64be_gen
1100 };
1101 
1102 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1103 	.generator = crypt_iv_essiv_gen
1104 };
1105 
1106 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1107 	.ctr	   = crypt_iv_benbi_ctr,
1108 	.dtr	   = crypt_iv_benbi_dtr,
1109 	.generator = crypt_iv_benbi_gen
1110 };
1111 
1112 static const struct crypt_iv_operations crypt_iv_null_ops = {
1113 	.generator = crypt_iv_null_gen
1114 };
1115 
1116 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1117 	.ctr	   = crypt_iv_lmk_ctr,
1118 	.dtr	   = crypt_iv_lmk_dtr,
1119 	.init	   = crypt_iv_lmk_init,
1120 	.wipe	   = crypt_iv_lmk_wipe,
1121 	.generator = crypt_iv_lmk_gen,
1122 	.post	   = crypt_iv_lmk_post
1123 };
1124 
1125 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1126 	.ctr	   = crypt_iv_tcw_ctr,
1127 	.dtr	   = crypt_iv_tcw_dtr,
1128 	.init	   = crypt_iv_tcw_init,
1129 	.wipe	   = crypt_iv_tcw_wipe,
1130 	.generator = crypt_iv_tcw_gen,
1131 	.post	   = crypt_iv_tcw_post
1132 };
1133 
1134 static const struct crypt_iv_operations crypt_iv_random_ops = {
1135 	.generator = crypt_iv_random_gen
1136 };
1137 
1138 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1139 	.ctr	   = crypt_iv_eboiv_ctr,
1140 	.generator = crypt_iv_eboiv_gen
1141 };
1142 
1143 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1144 	.ctr	   = crypt_iv_elephant_ctr,
1145 	.dtr	   = crypt_iv_elephant_dtr,
1146 	.init	   = crypt_iv_elephant_init,
1147 	.wipe	   = crypt_iv_elephant_wipe,
1148 	.generator = crypt_iv_elephant_gen,
1149 	.post	   = crypt_iv_elephant_post
1150 };
1151 
1152 /*
1153  * Integrity extensions
1154  */
1155 static bool crypt_integrity_aead(struct crypt_config *cc)
1156 {
1157 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1158 }
1159 
1160 static bool crypt_integrity_hmac(struct crypt_config *cc)
1161 {
1162 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1163 }
1164 
1165 /* Get sg containing data */
1166 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1167 					     struct scatterlist *sg)
1168 {
1169 	if (unlikely(crypt_integrity_aead(cc)))
1170 		return &sg[2];
1171 
1172 	return sg;
1173 }
1174 
1175 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1176 {
1177 	struct bio_integrity_payload *bip;
1178 	unsigned int tag_len;
1179 	int ret;
1180 
1181 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1182 		return 0;
1183 
1184 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1185 	if (IS_ERR(bip))
1186 		return PTR_ERR(bip);
1187 
1188 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1189 
1190 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1191 
1192 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1193 				     tag_len, offset_in_page(io->integrity_metadata));
1194 	if (unlikely(ret != tag_len))
1195 		return -ENOMEM;
1196 
1197 	return 0;
1198 }
1199 
1200 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1201 {
1202 #ifdef CONFIG_BLK_DEV_INTEGRITY
1203 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1204 	struct mapped_device *md = dm_table_get_md(ti->table);
1205 
1206 	/* We require an underlying device with non-PI metadata */
1207 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1208 		ti->error = "Integrity profile not supported.";
1209 		return -EINVAL;
1210 	}
1211 
1212 	if (bi->tuple_size < cc->used_tag_size) {
1213 		ti->error = "Integrity profile tag size mismatch.";
1214 		return -EINVAL;
1215 	}
1216 	cc->tuple_size = bi->tuple_size;
1217 	if (1 << bi->interval_exp != cc->sector_size) {
1218 		ti->error = "Integrity profile sector size mismatch.";
1219 		return -EINVAL;
1220 	}
1221 
1222 	if (crypt_integrity_aead(cc)) {
1223 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1224 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1225 		       cc->integrity_tag_size, cc->integrity_iv_size);
1226 
1227 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1228 			ti->error = "Integrity AEAD auth tag size is not supported.";
1229 			return -EINVAL;
1230 		}
1231 	} else if (cc->integrity_iv_size)
1232 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1233 		       cc->integrity_iv_size);
1234 
1235 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1236 		ti->error = "Not enough space for integrity tag in the profile.";
1237 		return -EINVAL;
1238 	}
1239 
1240 	return 0;
1241 #else
1242 	ti->error = "Integrity profile not supported.";
1243 	return -EINVAL;
1244 #endif
1245 }
1246 
1247 static void crypt_convert_init(struct crypt_config *cc,
1248 			       struct convert_context *ctx,
1249 			       struct bio *bio_out, struct bio *bio_in,
1250 			       sector_t sector)
1251 {
1252 	ctx->bio_in = bio_in;
1253 	ctx->bio_out = bio_out;
1254 	if (bio_in)
1255 		ctx->iter_in = bio_in->bi_iter;
1256 	if (bio_out)
1257 		ctx->iter_out = bio_out->bi_iter;
1258 	ctx->cc_sector = sector + cc->iv_offset;
1259 	init_completion(&ctx->restart);
1260 }
1261 
1262 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1263 					     void *req)
1264 {
1265 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1266 }
1267 
1268 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1269 {
1270 	return (void *)((char *)dmreq - cc->dmreq_start);
1271 }
1272 
1273 static u8 *iv_of_dmreq(struct crypt_config *cc,
1274 		       struct dm_crypt_request *dmreq)
1275 {
1276 	if (crypt_integrity_aead(cc))
1277 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1278 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1279 	else
1280 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1281 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1282 }
1283 
1284 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1285 		       struct dm_crypt_request *dmreq)
1286 {
1287 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1288 }
1289 
1290 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1291 		       struct dm_crypt_request *dmreq)
1292 {
1293 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1294 
1295 	return (__le64 *) ptr;
1296 }
1297 
1298 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1299 		       struct dm_crypt_request *dmreq)
1300 {
1301 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1302 		  cc->iv_size + sizeof(uint64_t);
1303 
1304 	return (unsigned int *)ptr;
1305 }
1306 
1307 static void *tag_from_dmreq(struct crypt_config *cc,
1308 				struct dm_crypt_request *dmreq)
1309 {
1310 	struct convert_context *ctx = dmreq->ctx;
1311 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1312 
1313 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1314 		cc->tuple_size];
1315 }
1316 
1317 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1318 			       struct dm_crypt_request *dmreq)
1319 {
1320 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1321 }
1322 
1323 static int crypt_convert_block_aead(struct crypt_config *cc,
1324 				     struct convert_context *ctx,
1325 				     struct aead_request *req,
1326 				     unsigned int tag_offset)
1327 {
1328 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1329 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1330 	struct dm_crypt_request *dmreq;
1331 	u8 *iv, *org_iv, *tag_iv, *tag;
1332 	__le64 *sector;
1333 	int r = 0;
1334 
1335 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1336 
1337 	/* Reject unexpected unaligned bio. */
1338 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1339 		return -EIO;
1340 
1341 	dmreq = dmreq_of_req(cc, req);
1342 	dmreq->iv_sector = ctx->cc_sector;
1343 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1344 		dmreq->iv_sector >>= cc->sector_shift;
1345 	dmreq->ctx = ctx;
1346 
1347 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1348 
1349 	sector = org_sector_of_dmreq(cc, dmreq);
1350 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1351 
1352 	iv = iv_of_dmreq(cc, dmreq);
1353 	org_iv = org_iv_of_dmreq(cc, dmreq);
1354 	tag = tag_from_dmreq(cc, dmreq);
1355 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1356 
1357 	/* AEAD request:
1358 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1359 	 *  | (authenticated) | (auth+encryption) |              |
1360 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1361 	 */
1362 	sg_init_table(dmreq->sg_in, 4);
1363 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1364 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1365 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1366 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1367 
1368 	sg_init_table(dmreq->sg_out, 4);
1369 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1370 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1371 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1372 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1373 
1374 	if (cc->iv_gen_ops) {
1375 		/* For READs use IV stored in integrity metadata */
1376 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1377 			memcpy(org_iv, tag_iv, cc->iv_size);
1378 		} else {
1379 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1380 			if (r < 0)
1381 				return r;
1382 			/* Store generated IV in integrity metadata */
1383 			if (cc->integrity_iv_size)
1384 				memcpy(tag_iv, org_iv, cc->iv_size);
1385 		}
1386 		/* Working copy of IV, to be modified in crypto API */
1387 		memcpy(iv, org_iv, cc->iv_size);
1388 	}
1389 
1390 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1391 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1392 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1393 				       cc->sector_size, iv);
1394 		r = crypto_aead_encrypt(req);
1395 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1396 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1397 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1398 	} else {
1399 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1400 				       cc->sector_size + cc->integrity_tag_size, iv);
1401 		r = crypto_aead_decrypt(req);
1402 	}
1403 
1404 	if (r == -EBADMSG) {
1405 		sector_t s = le64_to_cpu(*sector);
1406 
1407 		ctx->aead_failed = true;
1408 		if (ctx->aead_recheck) {
1409 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1410 				    ctx->bio_in->bi_bdev, s);
1411 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1412 					 ctx->bio_in, s, 0);
1413 		}
1414 	}
1415 
1416 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1417 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1418 
1419 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1420 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1421 
1422 	return r;
1423 }
1424 
1425 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1426 					struct convert_context *ctx,
1427 					struct skcipher_request *req,
1428 					unsigned int tag_offset)
1429 {
1430 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1431 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1432 	struct scatterlist *sg_in, *sg_out;
1433 	struct dm_crypt_request *dmreq;
1434 	u8 *iv, *org_iv, *tag_iv;
1435 	__le64 *sector;
1436 	int r = 0;
1437 
1438 	/* Reject unexpected unaligned bio. */
1439 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1440 		return -EIO;
1441 
1442 	dmreq = dmreq_of_req(cc, req);
1443 	dmreq->iv_sector = ctx->cc_sector;
1444 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1445 		dmreq->iv_sector >>= cc->sector_shift;
1446 	dmreq->ctx = ctx;
1447 
1448 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1449 
1450 	iv = iv_of_dmreq(cc, dmreq);
1451 	org_iv = org_iv_of_dmreq(cc, dmreq);
1452 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1453 
1454 	sector = org_sector_of_dmreq(cc, dmreq);
1455 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1456 
1457 	/* For skcipher we use only the first sg item */
1458 	sg_in  = &dmreq->sg_in[0];
1459 	sg_out = &dmreq->sg_out[0];
1460 
1461 	sg_init_table(sg_in, 1);
1462 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1463 
1464 	sg_init_table(sg_out, 1);
1465 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1466 
1467 	if (cc->iv_gen_ops) {
1468 		/* For READs use IV stored in integrity metadata */
1469 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1470 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1471 		} else {
1472 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1473 			if (r < 0)
1474 				return r;
1475 			/* Data can be already preprocessed in generator */
1476 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1477 				sg_in = sg_out;
1478 			/* Store generated IV in integrity metadata */
1479 			if (cc->integrity_iv_size)
1480 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1481 		}
1482 		/* Working copy of IV, to be modified in crypto API */
1483 		memcpy(iv, org_iv, cc->iv_size);
1484 	}
1485 
1486 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1487 
1488 	if (bio_data_dir(ctx->bio_in) == WRITE)
1489 		r = crypto_skcipher_encrypt(req);
1490 	else
1491 		r = crypto_skcipher_decrypt(req);
1492 
1493 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1494 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1495 
1496 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1497 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1498 
1499 	return r;
1500 }
1501 
1502 static void kcryptd_async_done(void *async_req, int error);
1503 
1504 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1505 				     struct convert_context *ctx)
1506 {
1507 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1508 
1509 	if (!ctx->r.req) {
1510 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1511 		if (!ctx->r.req)
1512 			return -ENOMEM;
1513 	}
1514 
1515 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1516 
1517 	/*
1518 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1519 	 * requests if driver request queue is full.
1520 	 */
1521 	skcipher_request_set_callback(ctx->r.req,
1522 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1523 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1524 
1525 	return 0;
1526 }
1527 
1528 static int crypt_alloc_req_aead(struct crypt_config *cc,
1529 				 struct convert_context *ctx)
1530 {
1531 	if (!ctx->r.req_aead) {
1532 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1533 		if (!ctx->r.req_aead)
1534 			return -ENOMEM;
1535 	}
1536 
1537 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1538 
1539 	/*
1540 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1541 	 * requests if driver request queue is full.
1542 	 */
1543 	aead_request_set_callback(ctx->r.req_aead,
1544 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1545 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1546 
1547 	return 0;
1548 }
1549 
1550 static int crypt_alloc_req(struct crypt_config *cc,
1551 			    struct convert_context *ctx)
1552 {
1553 	if (crypt_integrity_aead(cc))
1554 		return crypt_alloc_req_aead(cc, ctx);
1555 	else
1556 		return crypt_alloc_req_skcipher(cc, ctx);
1557 }
1558 
1559 static void crypt_free_req_skcipher(struct crypt_config *cc,
1560 				    struct skcipher_request *req, struct bio *base_bio)
1561 {
1562 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1563 
1564 	if ((struct skcipher_request *)(io + 1) != req)
1565 		mempool_free(req, &cc->req_pool);
1566 }
1567 
1568 static void crypt_free_req_aead(struct crypt_config *cc,
1569 				struct aead_request *req, struct bio *base_bio)
1570 {
1571 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1572 
1573 	if ((struct aead_request *)(io + 1) != req)
1574 		mempool_free(req, &cc->req_pool);
1575 }
1576 
1577 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1578 {
1579 	if (crypt_integrity_aead(cc))
1580 		crypt_free_req_aead(cc, req, base_bio);
1581 	else
1582 		crypt_free_req_skcipher(cc, req, base_bio);
1583 }
1584 
1585 /*
1586  * Encrypt / decrypt data from one bio to another one (can be the same one)
1587  */
1588 static blk_status_t crypt_convert(struct crypt_config *cc,
1589 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1590 {
1591 	unsigned int tag_offset = 0;
1592 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1593 	int r;
1594 
1595 	/*
1596 	 * if reset_pending is set we are dealing with the bio for the first time,
1597 	 * else we're continuing to work on the previous bio, so don't mess with
1598 	 * the cc_pending counter
1599 	 */
1600 	if (reset_pending)
1601 		atomic_set(&ctx->cc_pending, 1);
1602 
1603 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1604 
1605 		r = crypt_alloc_req(cc, ctx);
1606 		if (r) {
1607 			complete(&ctx->restart);
1608 			return BLK_STS_DEV_RESOURCE;
1609 		}
1610 
1611 		atomic_inc(&ctx->cc_pending);
1612 
1613 		if (crypt_integrity_aead(cc))
1614 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1615 		else
1616 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1617 
1618 		switch (r) {
1619 		/*
1620 		 * The request was queued by a crypto driver
1621 		 * but the driver request queue is full, let's wait.
1622 		 */
1623 		case -EBUSY:
1624 			if (in_interrupt()) {
1625 				if (try_wait_for_completion(&ctx->restart)) {
1626 					/*
1627 					 * we don't have to block to wait for completion,
1628 					 * so proceed
1629 					 */
1630 				} else {
1631 					/*
1632 					 * we can't wait for completion without blocking
1633 					 * exit and continue processing in a workqueue
1634 					 */
1635 					ctx->r.req = NULL;
1636 					ctx->cc_sector += sector_step;
1637 					tag_offset++;
1638 					return BLK_STS_DEV_RESOURCE;
1639 				}
1640 			} else {
1641 				wait_for_completion(&ctx->restart);
1642 			}
1643 			reinit_completion(&ctx->restart);
1644 			fallthrough;
1645 		/*
1646 		 * The request is queued and processed asynchronously,
1647 		 * completion function kcryptd_async_done() will be called.
1648 		 */
1649 		case -EINPROGRESS:
1650 			ctx->r.req = NULL;
1651 			ctx->cc_sector += sector_step;
1652 			tag_offset++;
1653 			continue;
1654 		/*
1655 		 * The request was already processed (synchronously).
1656 		 */
1657 		case 0:
1658 			atomic_dec(&ctx->cc_pending);
1659 			ctx->cc_sector += sector_step;
1660 			tag_offset++;
1661 			if (!atomic)
1662 				cond_resched();
1663 			continue;
1664 		/*
1665 		 * There was a data integrity error.
1666 		 */
1667 		case -EBADMSG:
1668 			atomic_dec(&ctx->cc_pending);
1669 			return BLK_STS_PROTECTION;
1670 		/*
1671 		 * There was an error while processing the request.
1672 		 */
1673 		default:
1674 			atomic_dec(&ctx->cc_pending);
1675 			return BLK_STS_IOERR;
1676 		}
1677 	}
1678 
1679 	return 0;
1680 }
1681 
1682 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1683 
1684 /*
1685  * Generate a new unfragmented bio with the given size
1686  * This should never violate the device limitations (but if it did then block
1687  * core should split the bio as needed).
1688  *
1689  * This function may be called concurrently. If we allocate from the mempool
1690  * concurrently, there is a possibility of deadlock. For example, if we have
1691  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1692  * the mempool concurrently, it may deadlock in a situation where both processes
1693  * have allocated 128 pages and the mempool is exhausted.
1694  *
1695  * In order to avoid this scenario we allocate the pages under a mutex.
1696  *
1697  * In order to not degrade performance with excessive locking, we try
1698  * non-blocking allocations without a mutex first but on failure we fallback
1699  * to blocking allocations with a mutex.
1700  *
1701  * In order to reduce allocation overhead, we try to allocate compound pages in
1702  * the first pass. If they are not available, we fall back to the mempool.
1703  */
1704 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1705 {
1706 	struct crypt_config *cc = io->cc;
1707 	struct bio *clone;
1708 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1709 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1710 	unsigned int remaining_size;
1711 	unsigned int order = MAX_PAGE_ORDER;
1712 
1713 retry:
1714 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1715 		mutex_lock(&cc->bio_alloc_lock);
1716 
1717 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1718 				 GFP_NOIO, &cc->bs);
1719 	clone->bi_private = io;
1720 	clone->bi_end_io = crypt_endio;
1721 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1722 
1723 	remaining_size = size;
1724 
1725 	while (remaining_size) {
1726 		struct page *pages;
1727 		unsigned size_to_add;
1728 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1729 		order = min(order, remaining_order);
1730 
1731 		while (order > 0) {
1732 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1733 					(1 << order) > dm_crypt_pages_per_client))
1734 				goto decrease_order;
1735 			pages = alloc_pages(gfp_mask
1736 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1737 				order);
1738 			if (likely(pages != NULL)) {
1739 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1740 				goto have_pages;
1741 			}
1742 decrease_order:
1743 			order--;
1744 		}
1745 
1746 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1747 		if (!pages) {
1748 			crypt_free_buffer_pages(cc, clone);
1749 			bio_put(clone);
1750 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1751 			order = 0;
1752 			goto retry;
1753 		}
1754 
1755 have_pages:
1756 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1757 		__bio_add_page(clone, pages, size_to_add, 0);
1758 		remaining_size -= size_to_add;
1759 	}
1760 
1761 	/* Allocate space for integrity tags */
1762 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1763 		crypt_free_buffer_pages(cc, clone);
1764 		bio_put(clone);
1765 		clone = NULL;
1766 	}
1767 
1768 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1769 		mutex_unlock(&cc->bio_alloc_lock);
1770 
1771 	return clone;
1772 }
1773 
1774 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1775 {
1776 	struct folio_iter fi;
1777 
1778 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1779 		bio_for_each_folio_all(fi, clone) {
1780 			if (folio_test_large(fi.folio)) {
1781 				percpu_counter_sub(&cc->n_allocated_pages,
1782 						1 << folio_order(fi.folio));
1783 				folio_put(fi.folio);
1784 			} else {
1785 				mempool_free(&fi.folio->page, &cc->page_pool);
1786 			}
1787 		}
1788 	}
1789 }
1790 
1791 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1792 			  struct bio *bio, sector_t sector)
1793 {
1794 	io->cc = cc;
1795 	io->base_bio = bio;
1796 	io->sector = sector;
1797 	io->error = 0;
1798 	io->ctx.aead_recheck = false;
1799 	io->ctx.aead_failed = false;
1800 	io->ctx.r.req = NULL;
1801 	io->integrity_metadata = NULL;
1802 	io->integrity_metadata_from_pool = false;
1803 	atomic_set(&io->io_pending, 0);
1804 }
1805 
1806 static void crypt_inc_pending(struct dm_crypt_io *io)
1807 {
1808 	atomic_inc(&io->io_pending);
1809 }
1810 
1811 static void kcryptd_queue_read(struct dm_crypt_io *io);
1812 
1813 /*
1814  * One of the bios was finished. Check for completion of
1815  * the whole request and correctly clean up the buffer.
1816  */
1817 static void crypt_dec_pending(struct dm_crypt_io *io)
1818 {
1819 	struct crypt_config *cc = io->cc;
1820 	struct bio *base_bio = io->base_bio;
1821 	blk_status_t error = io->error;
1822 
1823 	if (!atomic_dec_and_test(&io->io_pending))
1824 		return;
1825 
1826 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1827 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1828 		io->ctx.aead_recheck = true;
1829 		io->ctx.aead_failed = false;
1830 		io->error = 0;
1831 		kcryptd_queue_read(io);
1832 		return;
1833 	}
1834 
1835 	if (io->ctx.r.req)
1836 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1837 
1838 	if (unlikely(io->integrity_metadata_from_pool))
1839 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1840 	else
1841 		kfree(io->integrity_metadata);
1842 
1843 	base_bio->bi_status = error;
1844 
1845 	bio_endio(base_bio);
1846 }
1847 
1848 /*
1849  * kcryptd/kcryptd_io:
1850  *
1851  * Needed because it would be very unwise to do decryption in an
1852  * interrupt context.
1853  *
1854  * kcryptd performs the actual encryption or decryption.
1855  *
1856  * kcryptd_io performs the IO submission.
1857  *
1858  * They must be separated as otherwise the final stages could be
1859  * starved by new requests which can block in the first stages due
1860  * to memory allocation.
1861  *
1862  * The work is done per CPU global for all dm-crypt instances.
1863  * They should not depend on each other and do not block.
1864  */
1865 static void crypt_endio(struct bio *clone)
1866 {
1867 	struct dm_crypt_io *io = clone->bi_private;
1868 	struct crypt_config *cc = io->cc;
1869 	unsigned int rw = bio_data_dir(clone);
1870 	blk_status_t error = clone->bi_status;
1871 
1872 	if (io->ctx.aead_recheck && !error) {
1873 		kcryptd_queue_crypt(io);
1874 		return;
1875 	}
1876 
1877 	/*
1878 	 * free the processed pages
1879 	 */
1880 	if (rw == WRITE || io->ctx.aead_recheck)
1881 		crypt_free_buffer_pages(cc, clone);
1882 
1883 	bio_put(clone);
1884 
1885 	if (rw == READ && !error) {
1886 		kcryptd_queue_crypt(io);
1887 		return;
1888 	}
1889 
1890 	if (unlikely(error))
1891 		io->error = error;
1892 
1893 	crypt_dec_pending(io);
1894 }
1895 
1896 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1897 
1898 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1899 {
1900 	struct crypt_config *cc = io->cc;
1901 	struct bio *clone;
1902 
1903 	if (io->ctx.aead_recheck) {
1904 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1905 			return 1;
1906 		crypt_inc_pending(io);
1907 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1908 		if (unlikely(!clone)) {
1909 			crypt_dec_pending(io);
1910 			return 1;
1911 		}
1912 		clone->bi_iter.bi_sector = cc->start + io->sector;
1913 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1914 		io->saved_bi_iter = clone->bi_iter;
1915 		dm_submit_bio_remap(io->base_bio, clone);
1916 		return 0;
1917 	}
1918 
1919 	/*
1920 	 * We need the original biovec array in order to decrypt the whole bio
1921 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1922 	 * worry about the block layer modifying the biovec array; so leverage
1923 	 * bio_alloc_clone().
1924 	 */
1925 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1926 	if (!clone)
1927 		return 1;
1928 	clone->bi_private = io;
1929 	clone->bi_end_io = crypt_endio;
1930 
1931 	crypt_inc_pending(io);
1932 
1933 	clone->bi_iter.bi_sector = cc->start + io->sector;
1934 
1935 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1936 		crypt_dec_pending(io);
1937 		bio_put(clone);
1938 		return 1;
1939 	}
1940 
1941 	dm_submit_bio_remap(io->base_bio, clone);
1942 	return 0;
1943 }
1944 
1945 static void kcryptd_io_read_work(struct work_struct *work)
1946 {
1947 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1948 
1949 	crypt_inc_pending(io);
1950 	if (kcryptd_io_read(io, GFP_NOIO))
1951 		io->error = BLK_STS_RESOURCE;
1952 	crypt_dec_pending(io);
1953 }
1954 
1955 static void kcryptd_queue_read(struct dm_crypt_io *io)
1956 {
1957 	struct crypt_config *cc = io->cc;
1958 
1959 	INIT_WORK(&io->work, kcryptd_io_read_work);
1960 	queue_work(cc->io_queue, &io->work);
1961 }
1962 
1963 static void kcryptd_io_write(struct dm_crypt_io *io)
1964 {
1965 	struct bio *clone = io->ctx.bio_out;
1966 
1967 	dm_submit_bio_remap(io->base_bio, clone);
1968 }
1969 
1970 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1971 
1972 static int dmcrypt_write(void *data)
1973 {
1974 	struct crypt_config *cc = data;
1975 	struct dm_crypt_io *io;
1976 
1977 	while (1) {
1978 		struct rb_root write_tree;
1979 		struct blk_plug plug;
1980 
1981 		spin_lock_irq(&cc->write_thread_lock);
1982 continue_locked:
1983 
1984 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1985 			goto pop_from_list;
1986 
1987 		set_current_state(TASK_INTERRUPTIBLE);
1988 
1989 		spin_unlock_irq(&cc->write_thread_lock);
1990 
1991 		if (unlikely(kthread_should_stop())) {
1992 			set_current_state(TASK_RUNNING);
1993 			break;
1994 		}
1995 
1996 		schedule();
1997 
1998 		spin_lock_irq(&cc->write_thread_lock);
1999 		goto continue_locked;
2000 
2001 pop_from_list:
2002 		write_tree = cc->write_tree;
2003 		cc->write_tree = RB_ROOT;
2004 		spin_unlock_irq(&cc->write_thread_lock);
2005 
2006 		BUG_ON(rb_parent(write_tree.rb_node));
2007 
2008 		/*
2009 		 * Note: we cannot walk the tree here with rb_next because
2010 		 * the structures may be freed when kcryptd_io_write is called.
2011 		 */
2012 		blk_start_plug(&plug);
2013 		do {
2014 			io = crypt_io_from_node(rb_first(&write_tree));
2015 			rb_erase(&io->rb_node, &write_tree);
2016 			kcryptd_io_write(io);
2017 			cond_resched();
2018 		} while (!RB_EMPTY_ROOT(&write_tree));
2019 		blk_finish_plug(&plug);
2020 	}
2021 	return 0;
2022 }
2023 
2024 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
2025 {
2026 	struct bio *clone = io->ctx.bio_out;
2027 	struct crypt_config *cc = io->cc;
2028 	unsigned long flags;
2029 	sector_t sector;
2030 	struct rb_node **rbp, *parent;
2031 
2032 	if (unlikely(io->error)) {
2033 		crypt_free_buffer_pages(cc, clone);
2034 		bio_put(clone);
2035 		crypt_dec_pending(io);
2036 		return;
2037 	}
2038 
2039 	/* crypt_convert should have filled the clone bio */
2040 	BUG_ON(io->ctx.iter_out.bi_size);
2041 
2042 	clone->bi_iter.bi_sector = cc->start + io->sector;
2043 
2044 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2045 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2046 		dm_submit_bio_remap(io->base_bio, clone);
2047 		return;
2048 	}
2049 
2050 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2051 	if (RB_EMPTY_ROOT(&cc->write_tree))
2052 		wake_up_process(cc->write_thread);
2053 	rbp = &cc->write_tree.rb_node;
2054 	parent = NULL;
2055 	sector = io->sector;
2056 	while (*rbp) {
2057 		parent = *rbp;
2058 		if (sector < crypt_io_from_node(parent)->sector)
2059 			rbp = &(*rbp)->rb_left;
2060 		else
2061 			rbp = &(*rbp)->rb_right;
2062 	}
2063 	rb_link_node(&io->rb_node, parent, rbp);
2064 	rb_insert_color(&io->rb_node, &cc->write_tree);
2065 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2066 }
2067 
2068 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2069 				       struct convert_context *ctx)
2070 
2071 {
2072 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2073 		return false;
2074 
2075 	/*
2076 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2077 	 * constraints so they do not need to be issued inline by
2078 	 * kcryptd_crypt_write_convert().
2079 	 */
2080 	switch (bio_op(ctx->bio_in)) {
2081 	case REQ_OP_WRITE:
2082 	case REQ_OP_WRITE_ZEROES:
2083 		return true;
2084 	default:
2085 		return false;
2086 	}
2087 }
2088 
2089 static void kcryptd_crypt_write_continue(struct work_struct *work)
2090 {
2091 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2092 	struct crypt_config *cc = io->cc;
2093 	struct convert_context *ctx = &io->ctx;
2094 	int crypt_finished;
2095 	sector_t sector = io->sector;
2096 	blk_status_t r;
2097 
2098 	wait_for_completion(&ctx->restart);
2099 	reinit_completion(&ctx->restart);
2100 
2101 	r = crypt_convert(cc, &io->ctx, true, false);
2102 	if (r)
2103 		io->error = r;
2104 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2105 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2106 		/* Wait for completion signaled by kcryptd_async_done() */
2107 		wait_for_completion(&ctx->restart);
2108 		crypt_finished = 1;
2109 	}
2110 
2111 	/* Encryption was already finished, submit io now */
2112 	if (crypt_finished) {
2113 		kcryptd_crypt_write_io_submit(io, 0);
2114 		io->sector = sector;
2115 	}
2116 
2117 	crypt_dec_pending(io);
2118 }
2119 
2120 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2121 {
2122 	struct crypt_config *cc = io->cc;
2123 	struct convert_context *ctx = &io->ctx;
2124 	struct bio *clone;
2125 	int crypt_finished;
2126 	sector_t sector = io->sector;
2127 	blk_status_t r;
2128 
2129 	/*
2130 	 * Prevent io from disappearing until this function completes.
2131 	 */
2132 	crypt_inc_pending(io);
2133 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2134 
2135 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2136 	if (unlikely(!clone)) {
2137 		io->error = BLK_STS_IOERR;
2138 		goto dec;
2139 	}
2140 
2141 	io->ctx.bio_out = clone;
2142 	io->ctx.iter_out = clone->bi_iter;
2143 
2144 	if (crypt_integrity_aead(cc)) {
2145 		bio_copy_data(clone, io->base_bio);
2146 		io->ctx.bio_in = clone;
2147 		io->ctx.iter_in = clone->bi_iter;
2148 	}
2149 
2150 	sector += bio_sectors(clone);
2151 
2152 	crypt_inc_pending(io);
2153 	r = crypt_convert(cc, ctx,
2154 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2155 	/*
2156 	 * Crypto API backlogged the request, because its queue was full
2157 	 * and we're in softirq context, so continue from a workqueue
2158 	 * (TODO: is it actually possible to be in softirq in the write path?)
2159 	 */
2160 	if (r == BLK_STS_DEV_RESOURCE) {
2161 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2162 		queue_work(cc->crypt_queue, &io->work);
2163 		return;
2164 	}
2165 	if (r)
2166 		io->error = r;
2167 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2168 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2169 		/* Wait for completion signaled by kcryptd_async_done() */
2170 		wait_for_completion(&ctx->restart);
2171 		crypt_finished = 1;
2172 	}
2173 
2174 	/* Encryption was already finished, submit io now */
2175 	if (crypt_finished) {
2176 		kcryptd_crypt_write_io_submit(io, 0);
2177 		io->sector = sector;
2178 	}
2179 
2180 dec:
2181 	crypt_dec_pending(io);
2182 }
2183 
2184 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2185 {
2186 	if (io->ctx.aead_recheck) {
2187 		if (!io->error) {
2188 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2189 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2190 		}
2191 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2192 		bio_put(io->ctx.bio_in);
2193 	}
2194 	crypt_dec_pending(io);
2195 }
2196 
2197 static void kcryptd_crypt_read_continue(struct work_struct *work)
2198 {
2199 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2200 	struct crypt_config *cc = io->cc;
2201 	blk_status_t r;
2202 
2203 	wait_for_completion(&io->ctx.restart);
2204 	reinit_completion(&io->ctx.restart);
2205 
2206 	r = crypt_convert(cc, &io->ctx, true, false);
2207 	if (r)
2208 		io->error = r;
2209 
2210 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2211 		kcryptd_crypt_read_done(io);
2212 
2213 	crypt_dec_pending(io);
2214 }
2215 
2216 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2217 {
2218 	struct crypt_config *cc = io->cc;
2219 	blk_status_t r;
2220 
2221 	crypt_inc_pending(io);
2222 
2223 	if (io->ctx.aead_recheck) {
2224 		io->ctx.cc_sector = io->sector + cc->iv_offset;
2225 		r = crypt_convert(cc, &io->ctx,
2226 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2227 	} else {
2228 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2229 				   io->sector);
2230 
2231 		r = crypt_convert(cc, &io->ctx,
2232 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2233 	}
2234 	/*
2235 	 * Crypto API backlogged the request, because its queue was full
2236 	 * and we're in softirq context, so continue from a workqueue
2237 	 */
2238 	if (r == BLK_STS_DEV_RESOURCE) {
2239 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2240 		queue_work(cc->crypt_queue, &io->work);
2241 		return;
2242 	}
2243 	if (r)
2244 		io->error = r;
2245 
2246 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2247 		kcryptd_crypt_read_done(io);
2248 
2249 	crypt_dec_pending(io);
2250 }
2251 
2252 static void kcryptd_async_done(void *data, int error)
2253 {
2254 	struct dm_crypt_request *dmreq = data;
2255 	struct convert_context *ctx = dmreq->ctx;
2256 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2257 	struct crypt_config *cc = io->cc;
2258 
2259 	/*
2260 	 * A request from crypto driver backlog is going to be processed now,
2261 	 * finish the completion and continue in crypt_convert().
2262 	 * (Callback will be called for the second time for this request.)
2263 	 */
2264 	if (error == -EINPROGRESS) {
2265 		complete(&ctx->restart);
2266 		return;
2267 	}
2268 
2269 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2270 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2271 
2272 	if (error == -EBADMSG) {
2273 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2274 
2275 		ctx->aead_failed = true;
2276 		if (ctx->aead_recheck) {
2277 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2278 				    ctx->bio_in->bi_bdev, s);
2279 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2280 					 ctx->bio_in, s, 0);
2281 		}
2282 		io->error = BLK_STS_PROTECTION;
2283 	} else if (error < 0)
2284 		io->error = BLK_STS_IOERR;
2285 
2286 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2287 
2288 	if (!atomic_dec_and_test(&ctx->cc_pending))
2289 		return;
2290 
2291 	/*
2292 	 * The request is fully completed: for inline writes, let
2293 	 * kcryptd_crypt_write_convert() do the IO submission.
2294 	 */
2295 	if (bio_data_dir(io->base_bio) == READ) {
2296 		kcryptd_crypt_read_done(io);
2297 		return;
2298 	}
2299 
2300 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2301 		complete(&ctx->restart);
2302 		return;
2303 	}
2304 
2305 	kcryptd_crypt_write_io_submit(io, 1);
2306 }
2307 
2308 static void kcryptd_crypt(struct work_struct *work)
2309 {
2310 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2311 
2312 	if (bio_data_dir(io->base_bio) == READ)
2313 		kcryptd_crypt_read_convert(io);
2314 	else
2315 		kcryptd_crypt_write_convert(io);
2316 }
2317 
2318 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2319 {
2320 	struct crypt_config *cc = io->cc;
2321 
2322 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2323 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2324 		/*
2325 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2326 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2327 		 * it is being executed with irqs disabled.
2328 		 */
2329 		if (in_hardirq() || irqs_disabled()) {
2330 			INIT_WORK(&io->work, kcryptd_crypt);
2331 			queue_work(system_bh_wq, &io->work);
2332 			return;
2333 		} else {
2334 			kcryptd_crypt(&io->work);
2335 			return;
2336 		}
2337 	}
2338 
2339 	INIT_WORK(&io->work, kcryptd_crypt);
2340 	queue_work(cc->crypt_queue, &io->work);
2341 }
2342 
2343 static void crypt_free_tfms_aead(struct crypt_config *cc)
2344 {
2345 	if (!cc->cipher_tfm.tfms_aead)
2346 		return;
2347 
2348 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2349 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2350 		cc->cipher_tfm.tfms_aead[0] = NULL;
2351 	}
2352 
2353 	kfree(cc->cipher_tfm.tfms_aead);
2354 	cc->cipher_tfm.tfms_aead = NULL;
2355 }
2356 
2357 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2358 {
2359 	unsigned int i;
2360 
2361 	if (!cc->cipher_tfm.tfms)
2362 		return;
2363 
2364 	for (i = 0; i < cc->tfms_count; i++)
2365 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2366 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2367 			cc->cipher_tfm.tfms[i] = NULL;
2368 		}
2369 
2370 	kfree(cc->cipher_tfm.tfms);
2371 	cc->cipher_tfm.tfms = NULL;
2372 }
2373 
2374 static void crypt_free_tfms(struct crypt_config *cc)
2375 {
2376 	if (crypt_integrity_aead(cc))
2377 		crypt_free_tfms_aead(cc);
2378 	else
2379 		crypt_free_tfms_skcipher(cc);
2380 }
2381 
2382 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2383 {
2384 	unsigned int i;
2385 	int err;
2386 
2387 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2388 				      sizeof(struct crypto_skcipher *),
2389 				      GFP_KERNEL);
2390 	if (!cc->cipher_tfm.tfms)
2391 		return -ENOMEM;
2392 
2393 	for (i = 0; i < cc->tfms_count; i++) {
2394 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2395 						CRYPTO_ALG_ALLOCATES_MEMORY);
2396 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2397 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2398 			crypt_free_tfms(cc);
2399 			return err;
2400 		}
2401 	}
2402 
2403 	/*
2404 	 * dm-crypt performance can vary greatly depending on which crypto
2405 	 * algorithm implementation is used.  Help people debug performance
2406 	 * problems by logging the ->cra_driver_name.
2407 	 */
2408 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2409 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2410 	return 0;
2411 }
2412 
2413 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2414 {
2415 	int err;
2416 
2417 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2418 	if (!cc->cipher_tfm.tfms)
2419 		return -ENOMEM;
2420 
2421 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2422 						CRYPTO_ALG_ALLOCATES_MEMORY);
2423 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2424 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2425 		crypt_free_tfms(cc);
2426 		return err;
2427 	}
2428 
2429 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2430 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2431 	return 0;
2432 }
2433 
2434 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2435 {
2436 	if (crypt_integrity_aead(cc))
2437 		return crypt_alloc_tfms_aead(cc, ciphermode);
2438 	else
2439 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2440 }
2441 
2442 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2443 {
2444 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2445 }
2446 
2447 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2448 {
2449 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2450 }
2451 
2452 /*
2453  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2454  * the key must be for some reason in special format.
2455  * This funcion converts cc->key to this special format.
2456  */
2457 static void crypt_copy_authenckey(char *p, const void *key,
2458 				  unsigned int enckeylen, unsigned int authkeylen)
2459 {
2460 	struct crypto_authenc_key_param *param;
2461 	struct rtattr *rta;
2462 
2463 	rta = (struct rtattr *)p;
2464 	param = RTA_DATA(rta);
2465 	param->enckeylen = cpu_to_be32(enckeylen);
2466 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2467 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2468 	p += RTA_SPACE(sizeof(*param));
2469 	memcpy(p, key + enckeylen, authkeylen);
2470 	p += authkeylen;
2471 	memcpy(p, key, enckeylen);
2472 }
2473 
2474 static int crypt_setkey(struct crypt_config *cc)
2475 {
2476 	unsigned int subkey_size;
2477 	int err = 0, i, r;
2478 
2479 	/* Ignore extra keys (which are used for IV etc) */
2480 	subkey_size = crypt_subkey_size(cc);
2481 
2482 	if (crypt_integrity_hmac(cc)) {
2483 		if (subkey_size < cc->key_mac_size)
2484 			return -EINVAL;
2485 
2486 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2487 				      subkey_size - cc->key_mac_size,
2488 				      cc->key_mac_size);
2489 	}
2490 
2491 	for (i = 0; i < cc->tfms_count; i++) {
2492 		if (crypt_integrity_hmac(cc))
2493 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2494 				cc->authenc_key, crypt_authenckey_size(cc));
2495 		else if (crypt_integrity_aead(cc))
2496 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2497 					       cc->key + (i * subkey_size),
2498 					       subkey_size);
2499 		else
2500 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2501 						   cc->key + (i * subkey_size),
2502 						   subkey_size);
2503 		if (r)
2504 			err = r;
2505 	}
2506 
2507 	if (crypt_integrity_hmac(cc))
2508 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2509 
2510 	return err;
2511 }
2512 
2513 #ifdef CONFIG_KEYS
2514 
2515 static bool contains_whitespace(const char *str)
2516 {
2517 	while (*str)
2518 		if (isspace(*str++))
2519 			return true;
2520 	return false;
2521 }
2522 
2523 static int set_key_user(struct crypt_config *cc, struct key *key)
2524 {
2525 	const struct user_key_payload *ukp;
2526 
2527 	ukp = user_key_payload_locked(key);
2528 	if (!ukp)
2529 		return -EKEYREVOKED;
2530 
2531 	if (cc->key_size != ukp->datalen)
2532 		return -EINVAL;
2533 
2534 	memcpy(cc->key, ukp->data, cc->key_size);
2535 
2536 	return 0;
2537 }
2538 
2539 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2540 {
2541 	const struct encrypted_key_payload *ekp;
2542 
2543 	ekp = key->payload.data[0];
2544 	if (!ekp)
2545 		return -EKEYREVOKED;
2546 
2547 	if (cc->key_size != ekp->decrypted_datalen)
2548 		return -EINVAL;
2549 
2550 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2551 
2552 	return 0;
2553 }
2554 
2555 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2556 {
2557 	const struct trusted_key_payload *tkp;
2558 
2559 	tkp = key->payload.data[0];
2560 	if (!tkp)
2561 		return -EKEYREVOKED;
2562 
2563 	if (cc->key_size != tkp->key_len)
2564 		return -EINVAL;
2565 
2566 	memcpy(cc->key, tkp->key, cc->key_size);
2567 
2568 	return 0;
2569 }
2570 
2571 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2572 {
2573 	char *new_key_string, *key_desc;
2574 	int ret;
2575 	struct key_type *type;
2576 	struct key *key;
2577 	int (*set_key)(struct crypt_config *cc, struct key *key);
2578 
2579 	/*
2580 	 * Reject key_string with whitespace. dm core currently lacks code for
2581 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2582 	 */
2583 	if (contains_whitespace(key_string)) {
2584 		DMERR("whitespace chars not allowed in key string");
2585 		return -EINVAL;
2586 	}
2587 
2588 	/* look for next ':' separating key_type from key_description */
2589 	key_desc = strchr(key_string, ':');
2590 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2591 		return -EINVAL;
2592 
2593 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2594 		type = &key_type_logon;
2595 		set_key = set_key_user;
2596 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2597 		type = &key_type_user;
2598 		set_key = set_key_user;
2599 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2600 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2601 		type = &key_type_encrypted;
2602 		set_key = set_key_encrypted;
2603 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2604 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2605 		type = &key_type_trusted;
2606 		set_key = set_key_trusted;
2607 	} else {
2608 		return -EINVAL;
2609 	}
2610 
2611 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2612 	if (!new_key_string)
2613 		return -ENOMEM;
2614 
2615 	key = request_key(type, key_desc + 1, NULL);
2616 	if (IS_ERR(key)) {
2617 		ret = PTR_ERR(key);
2618 		goto free_new_key_string;
2619 	}
2620 
2621 	down_read(&key->sem);
2622 	ret = set_key(cc, key);
2623 	up_read(&key->sem);
2624 	key_put(key);
2625 	if (ret < 0)
2626 		goto free_new_key_string;
2627 
2628 	/* clear the flag since following operations may invalidate previously valid key */
2629 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2630 
2631 	ret = crypt_setkey(cc);
2632 	if (ret)
2633 		goto free_new_key_string;
2634 
2635 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2636 	kfree_sensitive(cc->key_string);
2637 	cc->key_string = new_key_string;
2638 	return 0;
2639 
2640 free_new_key_string:
2641 	kfree_sensitive(new_key_string);
2642 	return ret;
2643 }
2644 
2645 static int get_key_size(char **key_string)
2646 {
2647 	char *colon, dummy;
2648 	int ret;
2649 
2650 	if (*key_string[0] != ':')
2651 		return strlen(*key_string) >> 1;
2652 
2653 	/* look for next ':' in key string */
2654 	colon = strpbrk(*key_string + 1, ":");
2655 	if (!colon)
2656 		return -EINVAL;
2657 
2658 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2659 		return -EINVAL;
2660 
2661 	*key_string = colon;
2662 
2663 	/* remaining key string should be :<logon|user>:<key_desc> */
2664 
2665 	return ret;
2666 }
2667 
2668 #else
2669 
2670 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2671 {
2672 	return -EINVAL;
2673 }
2674 
2675 static int get_key_size(char **key_string)
2676 {
2677 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2678 }
2679 
2680 #endif /* CONFIG_KEYS */
2681 
2682 static int crypt_set_key(struct crypt_config *cc, char *key)
2683 {
2684 	int r = -EINVAL;
2685 	int key_string_len = strlen(key);
2686 
2687 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2688 	if (!cc->key_size && strcmp(key, "-"))
2689 		goto out;
2690 
2691 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2692 	if (key[0] == ':') {
2693 		r = crypt_set_keyring_key(cc, key + 1);
2694 		goto out;
2695 	}
2696 
2697 	/* clear the flag since following operations may invalidate previously valid key */
2698 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2699 
2700 	/* wipe references to any kernel keyring key */
2701 	kfree_sensitive(cc->key_string);
2702 	cc->key_string = NULL;
2703 
2704 	/* Decode key from its hex representation. */
2705 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2706 		goto out;
2707 
2708 	r = crypt_setkey(cc);
2709 	if (!r)
2710 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2711 
2712 out:
2713 	/* Hex key string not needed after here, so wipe it. */
2714 	memset(key, '0', key_string_len);
2715 
2716 	return r;
2717 }
2718 
2719 static int crypt_wipe_key(struct crypt_config *cc)
2720 {
2721 	int r;
2722 
2723 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2724 	get_random_bytes(&cc->key, cc->key_size);
2725 
2726 	/* Wipe IV private keys */
2727 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2728 		r = cc->iv_gen_ops->wipe(cc);
2729 		if (r)
2730 			return r;
2731 	}
2732 
2733 	kfree_sensitive(cc->key_string);
2734 	cc->key_string = NULL;
2735 	r = crypt_setkey(cc);
2736 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2737 
2738 	return r;
2739 }
2740 
2741 static void crypt_calculate_pages_per_client(void)
2742 {
2743 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2744 
2745 	if (!dm_crypt_clients_n)
2746 		return;
2747 
2748 	pages /= dm_crypt_clients_n;
2749 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2750 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2751 	dm_crypt_pages_per_client = pages;
2752 }
2753 
2754 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2755 {
2756 	struct crypt_config *cc = pool_data;
2757 	struct page *page;
2758 
2759 	/*
2760 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2761 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2762 	 * but avoids potential spinlock contention of an exact result.
2763 	 */
2764 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2765 	    likely(gfp_mask & __GFP_NORETRY))
2766 		return NULL;
2767 
2768 	page = alloc_page(gfp_mask);
2769 	if (likely(page != NULL))
2770 		percpu_counter_add(&cc->n_allocated_pages, 1);
2771 
2772 	return page;
2773 }
2774 
2775 static void crypt_page_free(void *page, void *pool_data)
2776 {
2777 	struct crypt_config *cc = pool_data;
2778 
2779 	__free_page(page);
2780 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2781 }
2782 
2783 static void crypt_dtr(struct dm_target *ti)
2784 {
2785 	struct crypt_config *cc = ti->private;
2786 
2787 	ti->private = NULL;
2788 
2789 	if (!cc)
2790 		return;
2791 
2792 	if (cc->write_thread)
2793 		kthread_stop(cc->write_thread);
2794 
2795 	if (cc->io_queue)
2796 		destroy_workqueue(cc->io_queue);
2797 	if (cc->crypt_queue)
2798 		destroy_workqueue(cc->crypt_queue);
2799 
2800 	if (cc->workqueue_id)
2801 		ida_free(&workqueue_ida, cc->workqueue_id);
2802 
2803 	crypt_free_tfms(cc);
2804 
2805 	bioset_exit(&cc->bs);
2806 
2807 	mempool_exit(&cc->page_pool);
2808 	mempool_exit(&cc->req_pool);
2809 	mempool_exit(&cc->tag_pool);
2810 
2811 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2812 	percpu_counter_destroy(&cc->n_allocated_pages);
2813 
2814 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2815 		cc->iv_gen_ops->dtr(cc);
2816 
2817 	if (cc->dev)
2818 		dm_put_device(ti, cc->dev);
2819 
2820 	kfree_sensitive(cc->cipher_string);
2821 	kfree_sensitive(cc->key_string);
2822 	kfree_sensitive(cc->cipher_auth);
2823 	kfree_sensitive(cc->authenc_key);
2824 
2825 	mutex_destroy(&cc->bio_alloc_lock);
2826 
2827 	/* Must zero key material before freeing */
2828 	kfree_sensitive(cc);
2829 
2830 	spin_lock(&dm_crypt_clients_lock);
2831 	WARN_ON(!dm_crypt_clients_n);
2832 	dm_crypt_clients_n--;
2833 	crypt_calculate_pages_per_client();
2834 	spin_unlock(&dm_crypt_clients_lock);
2835 
2836 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2837 }
2838 
2839 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2840 {
2841 	struct crypt_config *cc = ti->private;
2842 
2843 	if (crypt_integrity_aead(cc))
2844 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2845 	else
2846 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2847 
2848 	if (cc->iv_size)
2849 		/* at least a 64 bit sector number should fit in our buffer */
2850 		cc->iv_size = max(cc->iv_size,
2851 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2852 	else if (ivmode) {
2853 		DMWARN("Selected cipher does not support IVs");
2854 		ivmode = NULL;
2855 	}
2856 
2857 	/* Choose ivmode, see comments at iv code. */
2858 	if (ivmode == NULL)
2859 		cc->iv_gen_ops = NULL;
2860 	else if (strcmp(ivmode, "plain") == 0)
2861 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2862 	else if (strcmp(ivmode, "plain64") == 0)
2863 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2864 	else if (strcmp(ivmode, "plain64be") == 0)
2865 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2866 	else if (strcmp(ivmode, "essiv") == 0)
2867 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2868 	else if (strcmp(ivmode, "benbi") == 0)
2869 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2870 	else if (strcmp(ivmode, "null") == 0)
2871 		cc->iv_gen_ops = &crypt_iv_null_ops;
2872 	else if (strcmp(ivmode, "eboiv") == 0)
2873 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2874 	else if (strcmp(ivmode, "elephant") == 0) {
2875 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2876 		cc->key_parts = 2;
2877 		cc->key_extra_size = cc->key_size / 2;
2878 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2879 			return -EINVAL;
2880 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2881 	} else if (strcmp(ivmode, "lmk") == 0) {
2882 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2883 		/*
2884 		 * Version 2 and 3 is recognised according
2885 		 * to length of provided multi-key string.
2886 		 * If present (version 3), last key is used as IV seed.
2887 		 * All keys (including IV seed) are always the same size.
2888 		 */
2889 		if (cc->key_size % cc->key_parts) {
2890 			cc->key_parts++;
2891 			cc->key_extra_size = cc->key_size / cc->key_parts;
2892 		}
2893 	} else if (strcmp(ivmode, "tcw") == 0) {
2894 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2895 		cc->key_parts += 2; /* IV + whitening */
2896 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2897 	} else if (strcmp(ivmode, "random") == 0) {
2898 		cc->iv_gen_ops = &crypt_iv_random_ops;
2899 		/* Need storage space in integrity fields. */
2900 		cc->integrity_iv_size = cc->iv_size;
2901 	} else {
2902 		ti->error = "Invalid IV mode";
2903 		return -EINVAL;
2904 	}
2905 
2906 	return 0;
2907 }
2908 
2909 /*
2910  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2911  * The HMAC is needed to calculate tag size (HMAC digest size).
2912  * This should be probably done by crypto-api calls (once available...)
2913  */
2914 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2915 {
2916 	char *start, *end, *mac_alg = NULL;
2917 	struct crypto_ahash *mac;
2918 
2919 	if (!strstarts(cipher_api, "authenc("))
2920 		return 0;
2921 
2922 	start = strchr(cipher_api, '(');
2923 	end = strchr(cipher_api, ',');
2924 	if (!start || !end || ++start > end)
2925 		return -EINVAL;
2926 
2927 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2928 	if (!mac_alg)
2929 		return -ENOMEM;
2930 
2931 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2932 	kfree(mac_alg);
2933 
2934 	if (IS_ERR(mac))
2935 		return PTR_ERR(mac);
2936 
2937 	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2938 		cc->key_mac_size = crypto_ahash_digestsize(mac);
2939 	crypto_free_ahash(mac);
2940 
2941 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2942 	if (!cc->authenc_key)
2943 		return -ENOMEM;
2944 
2945 	return 0;
2946 }
2947 
2948 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2949 				char **ivmode, char **ivopts)
2950 {
2951 	struct crypt_config *cc = ti->private;
2952 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2953 	int ret = -EINVAL;
2954 
2955 	cc->tfms_count = 1;
2956 
2957 	/*
2958 	 * New format (capi: prefix)
2959 	 * capi:cipher_api_spec-iv:ivopts
2960 	 */
2961 	tmp = &cipher_in[strlen("capi:")];
2962 
2963 	/* Separate IV options if present, it can contain another '-' in hash name */
2964 	*ivopts = strrchr(tmp, ':');
2965 	if (*ivopts) {
2966 		**ivopts = '\0';
2967 		(*ivopts)++;
2968 	}
2969 	/* Parse IV mode */
2970 	*ivmode = strrchr(tmp, '-');
2971 	if (*ivmode) {
2972 		**ivmode = '\0';
2973 		(*ivmode)++;
2974 	}
2975 	/* The rest is crypto API spec */
2976 	cipher_api = tmp;
2977 
2978 	/* Alloc AEAD, can be used only in new format. */
2979 	if (crypt_integrity_aead(cc)) {
2980 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2981 		if (ret < 0) {
2982 			ti->error = "Invalid AEAD cipher spec";
2983 			return ret;
2984 		}
2985 	}
2986 
2987 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2988 		cc->tfms_count = 64;
2989 
2990 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2991 		if (!*ivopts) {
2992 			ti->error = "Digest algorithm missing for ESSIV mode";
2993 			return -EINVAL;
2994 		}
2995 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2996 			       cipher_api, *ivopts);
2997 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2998 			ti->error = "Cannot allocate cipher string";
2999 			return -ENOMEM;
3000 		}
3001 		cipher_api = buf;
3002 	}
3003 
3004 	cc->key_parts = cc->tfms_count;
3005 
3006 	/* Allocate cipher */
3007 	ret = crypt_alloc_tfms(cc, cipher_api);
3008 	if (ret < 0) {
3009 		ti->error = "Error allocating crypto tfm";
3010 		return ret;
3011 	}
3012 
3013 	if (crypt_integrity_aead(cc))
3014 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
3015 	else
3016 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
3017 
3018 	return 0;
3019 }
3020 
3021 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
3022 				char **ivmode, char **ivopts)
3023 {
3024 	struct crypt_config *cc = ti->private;
3025 	char *tmp, *cipher, *chainmode, *keycount;
3026 	char *cipher_api = NULL;
3027 	int ret = -EINVAL;
3028 	char dummy;
3029 
3030 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
3031 		ti->error = "Bad cipher specification";
3032 		return -EINVAL;
3033 	}
3034 
3035 	/*
3036 	 * Legacy dm-crypt cipher specification
3037 	 * cipher[:keycount]-mode-iv:ivopts
3038 	 */
3039 	tmp = cipher_in;
3040 	keycount = strsep(&tmp, "-");
3041 	cipher = strsep(&keycount, ":");
3042 
3043 	if (!keycount)
3044 		cc->tfms_count = 1;
3045 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3046 		 !is_power_of_2(cc->tfms_count)) {
3047 		ti->error = "Bad cipher key count specification";
3048 		return -EINVAL;
3049 	}
3050 	cc->key_parts = cc->tfms_count;
3051 
3052 	chainmode = strsep(&tmp, "-");
3053 	*ivmode = strsep(&tmp, ":");
3054 	*ivopts = tmp;
3055 
3056 	/*
3057 	 * For compatibility with the original dm-crypt mapping format, if
3058 	 * only the cipher name is supplied, use cbc-plain.
3059 	 */
3060 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3061 		chainmode = "cbc";
3062 		*ivmode = "plain";
3063 	}
3064 
3065 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3066 		ti->error = "IV mechanism required";
3067 		return -EINVAL;
3068 	}
3069 
3070 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3071 	if (!cipher_api)
3072 		goto bad_mem;
3073 
3074 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3075 		if (!*ivopts) {
3076 			ti->error = "Digest algorithm missing for ESSIV mode";
3077 			kfree(cipher_api);
3078 			return -EINVAL;
3079 		}
3080 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3081 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3082 	} else {
3083 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3084 			       "%s(%s)", chainmode, cipher);
3085 	}
3086 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3087 		kfree(cipher_api);
3088 		goto bad_mem;
3089 	}
3090 
3091 	/* Allocate cipher */
3092 	ret = crypt_alloc_tfms(cc, cipher_api);
3093 	if (ret < 0) {
3094 		ti->error = "Error allocating crypto tfm";
3095 		kfree(cipher_api);
3096 		return ret;
3097 	}
3098 	kfree(cipher_api);
3099 
3100 	return 0;
3101 bad_mem:
3102 	ti->error = "Cannot allocate cipher strings";
3103 	return -ENOMEM;
3104 }
3105 
3106 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3107 {
3108 	struct crypt_config *cc = ti->private;
3109 	char *ivmode = NULL, *ivopts = NULL;
3110 	int ret;
3111 
3112 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3113 	if (!cc->cipher_string) {
3114 		ti->error = "Cannot allocate cipher strings";
3115 		return -ENOMEM;
3116 	}
3117 
3118 	if (strstarts(cipher_in, "capi:"))
3119 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3120 	else
3121 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3122 	if (ret)
3123 		return ret;
3124 
3125 	/* Initialize IV */
3126 	ret = crypt_ctr_ivmode(ti, ivmode);
3127 	if (ret < 0)
3128 		return ret;
3129 
3130 	/* Initialize and set key */
3131 	ret = crypt_set_key(cc, key);
3132 	if (ret < 0) {
3133 		ti->error = "Error decoding and setting key";
3134 		return ret;
3135 	}
3136 
3137 	/* Allocate IV */
3138 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3139 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3140 		if (ret < 0) {
3141 			ti->error = "Error creating IV";
3142 			return ret;
3143 		}
3144 	}
3145 
3146 	/* Initialize IV (set keys for ESSIV etc) */
3147 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3148 		ret = cc->iv_gen_ops->init(cc);
3149 		if (ret < 0) {
3150 			ti->error = "Error initialising IV";
3151 			return ret;
3152 		}
3153 	}
3154 
3155 	/* wipe the kernel key payload copy */
3156 	if (cc->key_string)
3157 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3158 
3159 	return ret;
3160 }
3161 
3162 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3163 {
3164 	struct crypt_config *cc = ti->private;
3165 	struct dm_arg_set as;
3166 	static const struct dm_arg _args[] = {
3167 		{0, 9, "Invalid number of feature args"},
3168 	};
3169 	unsigned int opt_params, val;
3170 	const char *opt_string, *sval;
3171 	char dummy;
3172 	int ret;
3173 
3174 	/* Optional parameters */
3175 	as.argc = argc;
3176 	as.argv = argv;
3177 
3178 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3179 	if (ret)
3180 		return ret;
3181 
3182 	while (opt_params--) {
3183 		opt_string = dm_shift_arg(&as);
3184 		if (!opt_string) {
3185 			ti->error = "Not enough feature arguments";
3186 			return -EINVAL;
3187 		}
3188 
3189 		if (!strcasecmp(opt_string, "allow_discards"))
3190 			ti->num_discard_bios = 1;
3191 
3192 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3193 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3194 		else if (!strcasecmp(opt_string, "high_priority"))
3195 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3196 
3197 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3198 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3199 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3200 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3201 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3202 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3203 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3204 			if (val == 0 || val > MAX_TAG_SIZE) {
3205 				ti->error = "Invalid integrity arguments";
3206 				return -EINVAL;
3207 			}
3208 			cc->used_tag_size = val;
3209 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3210 			if (!strcasecmp(sval, "aead")) {
3211 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3212 			} else if (strcasecmp(sval, "none")) {
3213 				ti->error = "Unknown integrity profile";
3214 				return -EINVAL;
3215 			}
3216 
3217 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3218 			if (!cc->cipher_auth)
3219 				return -ENOMEM;
3220 		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3221 			if (!val) {
3222 				ti->error = "Invalid integrity_key_size argument";
3223 				return -EINVAL;
3224 			}
3225 			cc->key_mac_size = val;
3226 			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3227 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3228 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3229 			    cc->sector_size > 4096 ||
3230 			    (cc->sector_size & (cc->sector_size - 1))) {
3231 				ti->error = "Invalid feature value for sector_size";
3232 				return -EINVAL;
3233 			}
3234 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3235 				ti->error = "Device size is not multiple of sector_size feature";
3236 				return -EINVAL;
3237 			}
3238 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3239 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3240 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3241 		else {
3242 			ti->error = "Invalid feature arguments";
3243 			return -EINVAL;
3244 		}
3245 	}
3246 
3247 	return 0;
3248 }
3249 
3250 #ifdef CONFIG_BLK_DEV_ZONED
3251 static int crypt_report_zones(struct dm_target *ti,
3252 		struct dm_report_zones_args *args, unsigned int nr_zones)
3253 {
3254 	struct crypt_config *cc = ti->private;
3255 
3256 	return dm_report_zones(cc->dev->bdev, cc->start,
3257 			cc->start + dm_target_offset(ti, args->next_sector),
3258 			args, nr_zones);
3259 }
3260 #else
3261 #define crypt_report_zones NULL
3262 #endif
3263 
3264 /*
3265  * Construct an encryption mapping:
3266  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3267  */
3268 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3269 {
3270 	struct crypt_config *cc;
3271 	const char *devname = dm_table_device_name(ti->table);
3272 	int key_size, wq_id;
3273 	unsigned int align_mask;
3274 	unsigned int common_wq_flags;
3275 	unsigned long long tmpll;
3276 	int ret;
3277 	size_t iv_size_padding, additional_req_size;
3278 	char dummy;
3279 
3280 	if (argc < 5) {
3281 		ti->error = "Not enough arguments";
3282 		return -EINVAL;
3283 	}
3284 
3285 	key_size = get_key_size(&argv[1]);
3286 	if (key_size < 0) {
3287 		ti->error = "Cannot parse key size";
3288 		return -EINVAL;
3289 	}
3290 
3291 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3292 	if (!cc) {
3293 		ti->error = "Cannot allocate encryption context";
3294 		return -ENOMEM;
3295 	}
3296 	cc->key_size = key_size;
3297 	cc->sector_size = (1 << SECTOR_SHIFT);
3298 	cc->sector_shift = 0;
3299 
3300 	ti->private = cc;
3301 
3302 	spin_lock(&dm_crypt_clients_lock);
3303 	dm_crypt_clients_n++;
3304 	crypt_calculate_pages_per_client();
3305 	spin_unlock(&dm_crypt_clients_lock);
3306 
3307 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3308 	if (ret < 0)
3309 		goto bad;
3310 
3311 	/* Optional parameters need to be read before cipher constructor */
3312 	if (argc > 5) {
3313 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3314 		if (ret)
3315 			goto bad;
3316 	}
3317 
3318 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3319 	if (ret < 0)
3320 		goto bad;
3321 
3322 	if (crypt_integrity_aead(cc)) {
3323 		cc->dmreq_start = sizeof(struct aead_request);
3324 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3325 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3326 	} else {
3327 		cc->dmreq_start = sizeof(struct skcipher_request);
3328 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3329 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3330 	}
3331 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3332 
3333 	if (align_mask < CRYPTO_MINALIGN) {
3334 		/* Allocate the padding exactly */
3335 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3336 				& align_mask;
3337 	} else {
3338 		/*
3339 		 * If the cipher requires greater alignment than kmalloc
3340 		 * alignment, we don't know the exact position of the
3341 		 * initialization vector. We must assume worst case.
3342 		 */
3343 		iv_size_padding = align_mask;
3344 	}
3345 
3346 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3347 	additional_req_size = sizeof(struct dm_crypt_request) +
3348 		iv_size_padding + cc->iv_size +
3349 		cc->iv_size +
3350 		sizeof(uint64_t) +
3351 		sizeof(unsigned int);
3352 
3353 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3354 	if (ret) {
3355 		ti->error = "Cannot allocate crypt request mempool";
3356 		goto bad;
3357 	}
3358 
3359 	cc->per_bio_data_size = ti->per_io_data_size =
3360 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3361 		      ARCH_DMA_MINALIGN);
3362 
3363 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3364 	if (ret) {
3365 		ti->error = "Cannot allocate page mempool";
3366 		goto bad;
3367 	}
3368 
3369 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3370 	if (ret) {
3371 		ti->error = "Cannot allocate crypt bioset";
3372 		goto bad;
3373 	}
3374 
3375 	mutex_init(&cc->bio_alloc_lock);
3376 
3377 	ret = -EINVAL;
3378 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3379 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3380 		ti->error = "Invalid iv_offset sector";
3381 		goto bad;
3382 	}
3383 	cc->iv_offset = tmpll;
3384 
3385 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3386 	if (ret) {
3387 		ti->error = "Device lookup failed";
3388 		goto bad;
3389 	}
3390 
3391 	ret = -EINVAL;
3392 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3393 		ti->error = "Invalid device sector";
3394 		goto bad;
3395 	}
3396 	cc->start = tmpll;
3397 
3398 	if (bdev_is_zoned(cc->dev->bdev)) {
3399 		/*
3400 		 * For zoned block devices, we need to preserve the issuer write
3401 		 * ordering. To do so, disable write workqueues and force inline
3402 		 * encryption completion.
3403 		 */
3404 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3405 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3406 
3407 		/*
3408 		 * All zone append writes to a zone of a zoned block device will
3409 		 * have the same BIO sector, the start of the zone. When the
3410 		 * cypher IV mode uses sector values, all data targeting a
3411 		 * zone will be encrypted using the first sector numbers of the
3412 		 * zone. This will not result in write errors but will
3413 		 * cause most reads to fail as reads will use the sector values
3414 		 * for the actual data locations, resulting in IV mismatch.
3415 		 * To avoid this problem, ask DM core to emulate zone append
3416 		 * operations with regular writes.
3417 		 */
3418 		DMDEBUG("Zone append operations will be emulated");
3419 		ti->emulate_zone_append = true;
3420 	}
3421 
3422 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3423 		ret = crypt_integrity_ctr(cc, ti);
3424 		if (ret)
3425 			goto bad;
3426 
3427 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3428 		if (!cc->tag_pool_max_sectors)
3429 			cc->tag_pool_max_sectors = 1;
3430 
3431 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3432 			cc->tag_pool_max_sectors * cc->tuple_size);
3433 		if (ret) {
3434 			ti->error = "Cannot allocate integrity tags mempool";
3435 			goto bad;
3436 		}
3437 
3438 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3439 	}
3440 
3441 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3442 	if (wq_id < 0) {
3443 		ti->error = "Couldn't get workqueue id";
3444 		ret = wq_id;
3445 		goto bad;
3446 	}
3447 	cc->workqueue_id = wq_id;
3448 
3449 	ret = -ENOMEM;
3450 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3451 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3452 		common_wq_flags |= WQ_HIGHPRI;
3453 
3454 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", common_wq_flags, 1, devname, wq_id);
3455 	if (!cc->io_queue) {
3456 		ti->error = "Couldn't create kcryptd io queue";
3457 		goto bad;
3458 	}
3459 
3460 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3461 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3462 						  common_wq_flags | WQ_CPU_INTENSIVE,
3463 						  1, devname, wq_id);
3464 	} else {
3465 		/*
3466 		 * While crypt_queue is certainly CPU intensive, the use of
3467 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3468 		 */
3469 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3470 						  common_wq_flags | WQ_UNBOUND,
3471 						  num_online_cpus(), devname, wq_id);
3472 	}
3473 	if (!cc->crypt_queue) {
3474 		ti->error = "Couldn't create kcryptd queue";
3475 		goto bad;
3476 	}
3477 
3478 	spin_lock_init(&cc->write_thread_lock);
3479 	cc->write_tree = RB_ROOT;
3480 
3481 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3482 	if (IS_ERR(cc->write_thread)) {
3483 		ret = PTR_ERR(cc->write_thread);
3484 		cc->write_thread = NULL;
3485 		ti->error = "Couldn't spawn write thread";
3486 		goto bad;
3487 	}
3488 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3489 		set_user_nice(cc->write_thread, MIN_NICE);
3490 
3491 	ti->num_flush_bios = 1;
3492 	ti->limit_swap_bios = true;
3493 	ti->accounts_remapped_io = true;
3494 
3495 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3496 	return 0;
3497 
3498 bad:
3499 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3500 	crypt_dtr(ti);
3501 	return ret;
3502 }
3503 
3504 static int crypt_map(struct dm_target *ti, struct bio *bio)
3505 {
3506 	struct dm_crypt_io *io;
3507 	struct crypt_config *cc = ti->private;
3508 	unsigned max_sectors;
3509 
3510 	/*
3511 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3512 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3513 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3514 	 */
3515 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3516 	    bio_op(bio) == REQ_OP_DISCARD)) {
3517 		bio_set_dev(bio, cc->dev->bdev);
3518 		if (bio_sectors(bio))
3519 			bio->bi_iter.bi_sector = cc->start +
3520 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3521 		return DM_MAPIO_REMAPPED;
3522 	}
3523 
3524 	/*
3525 	 * Check if bio is too large, split as needed.
3526 	 */
3527 	max_sectors = get_max_request_size(cc, bio_data_dir(bio) == WRITE);
3528 	if (unlikely(bio_sectors(bio) > max_sectors))
3529 		dm_accept_partial_bio(bio, max_sectors);
3530 
3531 	/*
3532 	 * Ensure that bio is a multiple of internal sector encryption size
3533 	 * and is aligned to this size as defined in IO hints.
3534 	 */
3535 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3536 		return DM_MAPIO_KILL;
3537 
3538 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3539 		return DM_MAPIO_KILL;
3540 
3541 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3542 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3543 
3544 	if (cc->tuple_size) {
3545 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3546 
3547 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3548 			io->integrity_metadata = NULL;
3549 		else
3550 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3551 
3552 		if (unlikely(!io->integrity_metadata)) {
3553 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3554 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3555 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3556 			io->integrity_metadata_from_pool = true;
3557 		}
3558 	}
3559 
3560 	if (crypt_integrity_aead(cc))
3561 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3562 	else
3563 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3564 
3565 	if (bio_data_dir(io->base_bio) == READ) {
3566 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3567 			kcryptd_queue_read(io);
3568 	} else
3569 		kcryptd_queue_crypt(io);
3570 
3571 	return DM_MAPIO_SUBMITTED;
3572 }
3573 
3574 static char hex2asc(unsigned char c)
3575 {
3576 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3577 }
3578 
3579 static void crypt_status(struct dm_target *ti, status_type_t type,
3580 			 unsigned int status_flags, char *result, unsigned int maxlen)
3581 {
3582 	struct crypt_config *cc = ti->private;
3583 	unsigned int i, sz = 0;
3584 	int num_feature_args = 0;
3585 
3586 	switch (type) {
3587 	case STATUSTYPE_INFO:
3588 		result[0] = '\0';
3589 		break;
3590 
3591 	case STATUSTYPE_TABLE:
3592 		DMEMIT("%s ", cc->cipher_string);
3593 
3594 		if (cc->key_size > 0) {
3595 			if (cc->key_string)
3596 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3597 			else {
3598 				for (i = 0; i < cc->key_size; i++) {
3599 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3600 					       hex2asc(cc->key[i] & 0xf));
3601 				}
3602 			}
3603 		} else
3604 			DMEMIT("-");
3605 
3606 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3607 				cc->dev->name, (unsigned long long)cc->start);
3608 
3609 		num_feature_args += !!ti->num_discard_bios;
3610 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3611 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3612 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3613 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3614 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3615 		num_feature_args += !!cc->used_tag_size;
3616 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3617 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3618 		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3619 		if (num_feature_args) {
3620 			DMEMIT(" %d", num_feature_args);
3621 			if (ti->num_discard_bios)
3622 				DMEMIT(" allow_discards");
3623 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3624 				DMEMIT(" same_cpu_crypt");
3625 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3626 				DMEMIT(" high_priority");
3627 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3628 				DMEMIT(" submit_from_crypt_cpus");
3629 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3630 				DMEMIT(" no_read_workqueue");
3631 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3632 				DMEMIT(" no_write_workqueue");
3633 			if (cc->used_tag_size)
3634 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3635 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3636 				DMEMIT(" sector_size:%d", cc->sector_size);
3637 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3638 				DMEMIT(" iv_large_sectors");
3639 			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3640 				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3641 		}
3642 		break;
3643 
3644 	case STATUSTYPE_IMA:
3645 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3646 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3647 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3648 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3649 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3650 		       'y' : 'n');
3651 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3652 		       'y' : 'n');
3653 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3654 		       'y' : 'n');
3655 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3656 		       'y' : 'n');
3657 
3658 		if (cc->used_tag_size)
3659 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3660 			       cc->used_tag_size, cc->cipher_auth);
3661 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3662 			DMEMIT(",sector_size=%d", cc->sector_size);
3663 		if (cc->cipher_string)
3664 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3665 
3666 		DMEMIT(",key_size=%u", cc->key_size);
3667 		DMEMIT(",key_parts=%u", cc->key_parts);
3668 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3669 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3670 		DMEMIT(";");
3671 		break;
3672 	}
3673 }
3674 
3675 static void crypt_postsuspend(struct dm_target *ti)
3676 {
3677 	struct crypt_config *cc = ti->private;
3678 
3679 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3680 }
3681 
3682 static int crypt_preresume(struct dm_target *ti)
3683 {
3684 	struct crypt_config *cc = ti->private;
3685 
3686 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3687 		DMERR("aborting resume - crypt key is not set.");
3688 		return -EAGAIN;
3689 	}
3690 
3691 	return 0;
3692 }
3693 
3694 static void crypt_resume(struct dm_target *ti)
3695 {
3696 	struct crypt_config *cc = ti->private;
3697 
3698 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3699 }
3700 
3701 /* Message interface
3702  *	key set <key>
3703  *	key wipe
3704  */
3705 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3706 			 char *result, unsigned int maxlen)
3707 {
3708 	struct crypt_config *cc = ti->private;
3709 	int key_size, ret = -EINVAL;
3710 
3711 	if (argc < 2)
3712 		goto error;
3713 
3714 	if (!strcasecmp(argv[0], "key")) {
3715 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3716 			DMWARN("not suspended during key manipulation.");
3717 			return -EINVAL;
3718 		}
3719 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3720 			/* The key size may not be changed. */
3721 			key_size = get_key_size(&argv[2]);
3722 			if (key_size < 0 || cc->key_size != key_size) {
3723 				memset(argv[2], '0', strlen(argv[2]));
3724 				return -EINVAL;
3725 			}
3726 
3727 			ret = crypt_set_key(cc, argv[2]);
3728 			if (ret)
3729 				return ret;
3730 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3731 				ret = cc->iv_gen_ops->init(cc);
3732 			/* wipe the kernel key payload copy */
3733 			if (cc->key_string)
3734 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3735 			return ret;
3736 		}
3737 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3738 			return crypt_wipe_key(cc);
3739 	}
3740 
3741 error:
3742 	DMWARN("unrecognised message received.");
3743 	return -EINVAL;
3744 }
3745 
3746 static int crypt_iterate_devices(struct dm_target *ti,
3747 				 iterate_devices_callout_fn fn, void *data)
3748 {
3749 	struct crypt_config *cc = ti->private;
3750 
3751 	return fn(ti, cc->dev, cc->start, ti->len, data);
3752 }
3753 
3754 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3755 {
3756 	struct crypt_config *cc = ti->private;
3757 
3758 	limits->logical_block_size =
3759 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3760 	limits->physical_block_size =
3761 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3762 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3763 	limits->dma_alignment = limits->logical_block_size - 1;
3764 }
3765 
3766 static struct target_type crypt_target = {
3767 	.name   = "crypt",
3768 	.version = {1, 28, 0},
3769 	.module = THIS_MODULE,
3770 	.ctr    = crypt_ctr,
3771 	.dtr    = crypt_dtr,
3772 	.features = DM_TARGET_ZONED_HM,
3773 	.report_zones = crypt_report_zones,
3774 	.map    = crypt_map,
3775 	.status = crypt_status,
3776 	.postsuspend = crypt_postsuspend,
3777 	.preresume = crypt_preresume,
3778 	.resume = crypt_resume,
3779 	.message = crypt_message,
3780 	.iterate_devices = crypt_iterate_devices,
3781 	.io_hints = crypt_io_hints,
3782 };
3783 module_dm(crypt);
3784 
3785 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3786 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3787 MODULE_LICENSE("GPL");
3788