xref: /linux/drivers/md/dm-crypt.c (revision 8ba1ade3e6b595b9091040feaa3341d7e45ba95d)
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/hex.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/key.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-integrity.h>
21 #include <linux/crc32.h>
22 #include <linux/mempool.h>
23 #include <linux/slab.h>
24 #include <linux/crypto.h>
25 #include <linux/fips.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/backing-dev.h>
29 #include <linux/atomic.h>
30 #include <linux/scatterlist.h>
31 #include <linux/rbtree.h>
32 #include <linux/ctype.h>
33 #include <asm/page.h>
34 #include <linux/unaligned.h>
35 #include <crypto/aes.h>
36 #include <crypto/hash.h>
37 #include <crypto/md5.h>
38 #include <crypto/skcipher.h>
39 #include <crypto/aead.h>
40 #include <crypto/authenc.h>
41 #include <crypto/utils.h>
42 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
43 #include <linux/key-type.h>
44 #include <keys/user-type.h>
45 #include <keys/encrypted-type.h>
46 #include <keys/trusted-type.h>
47 
48 #include <linux/device-mapper.h>
49 
50 #include "dm-audit.h"
51 
52 #define DM_MSG_PREFIX "crypt"
53 
54 static DEFINE_IDA(workqueue_ida);
55 
56 /*
57  * context holding the current state of a multi-part conversion
58  */
59 struct convert_context {
60 	struct completion restart;
61 	struct bio *bio_in;
62 	struct bvec_iter iter_in;
63 	struct bio *bio_out;
64 	struct bvec_iter iter_out;
65 	atomic_t cc_pending;
66 	unsigned int tag_offset;
67 	u64 cc_sector;
68 	union {
69 		struct skcipher_request *req;
70 		struct aead_request *req_aead;
71 	} r;
72 	bool aead_recheck;
73 	bool aead_failed;
74 
75 };
76 
77 /*
78  * per bio private data
79  */
80 struct dm_crypt_io {
81 	struct crypt_config *cc;
82 	struct bio *base_bio;
83 	u8 *integrity_metadata;
84 	bool integrity_metadata_from_pool:1;
85 
86 	struct work_struct work;
87 
88 	struct convert_context ctx;
89 
90 	atomic_t io_pending;
91 	blk_status_t error;
92 	sector_t sector;
93 
94 	struct bvec_iter saved_bi_iter;
95 
96 	struct rb_node rb_node;
97 } CRYPTO_MINALIGN_ATTR;
98 
99 struct dm_crypt_request {
100 	struct convert_context *ctx;
101 	struct scatterlist sg_in[4];
102 	struct scatterlist sg_out[4];
103 	u64 iv_sector;
104 };
105 
106 struct crypt_config;
107 
108 struct crypt_iv_operations {
109 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
110 		   const char *opts);
111 	void (*dtr)(struct crypt_config *cc);
112 	int (*init)(struct crypt_config *cc);
113 	void (*wipe)(struct crypt_config *cc);
114 	int (*generator)(struct crypt_config *cc, u8 *iv,
115 			 struct dm_crypt_request *dmreq);
116 	void (*post)(struct crypt_config *cc, u8 *iv,
117 		     struct dm_crypt_request *dmreq);
118 };
119 
120 struct iv_benbi_private {
121 	int shift;
122 };
123 
124 #define LMK_SEED_SIZE 64 /* hash + 0 */
125 struct iv_lmk_private {
126 	u8 *seed;
127 };
128 
129 #define TCW_WHITENING_SIZE 16
130 struct iv_tcw_private {
131 	u8 *iv_seed;
132 	u8 *whitening;
133 };
134 
135 #define ELEPHANT_MAX_KEY_SIZE 32
136 struct iv_elephant_private {
137 	struct aes_enckey *key;
138 };
139 
140 /*
141  * Crypt: maps a linear range of a block device
142  * and encrypts / decrypts at the same time.
143  */
144 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
145 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
146 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
147 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
148 
149 enum cipher_flags {
150 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
151 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
152 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
153 	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
154 };
155 
156 /*
157  * The fields in here must be read only after initialization.
158  */
159 struct crypt_config {
160 	struct dm_dev *dev;
161 	sector_t start;
162 
163 	struct percpu_counter n_allocated_pages;
164 
165 	struct workqueue_struct *io_queue;
166 	struct workqueue_struct *crypt_queue;
167 
168 	spinlock_t write_thread_lock;
169 	struct task_struct *write_thread;
170 	struct rb_root write_tree;
171 
172 	char *cipher_string;
173 	char *cipher_auth;
174 	char *key_string;
175 
176 	const struct crypt_iv_operations *iv_gen_ops;
177 	union {
178 		struct iv_benbi_private benbi;
179 		struct iv_lmk_private lmk;
180 		struct iv_tcw_private tcw;
181 		struct iv_elephant_private elephant;
182 	} iv_gen_private;
183 	u64 iv_offset;
184 	unsigned int iv_size;
185 	unsigned int sector_size;
186 	unsigned char sector_shift;
187 
188 	union {
189 		struct crypto_skcipher **tfms;
190 		struct crypto_aead **tfms_aead;
191 	} cipher_tfm;
192 	unsigned int tfms_count;
193 	int workqueue_id;
194 	unsigned long cipher_flags;
195 
196 	/*
197 	 * Layout of each crypto request:
198 	 *
199 	 *   struct skcipher_request
200 	 *      context
201 	 *      padding
202 	 *   struct dm_crypt_request
203 	 *      padding
204 	 *   IV
205 	 *
206 	 * The padding is added so that dm_crypt_request and the IV are
207 	 * correctly aligned.
208 	 */
209 	unsigned int dmreq_start;
210 
211 	unsigned int per_bio_data_size;
212 
213 	unsigned long flags;
214 	unsigned int key_size;
215 	unsigned int key_parts;      /* independent parts in key buffer */
216 	unsigned int key_extra_size; /* additional keys length */
217 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
218 
219 	unsigned int integrity_tag_size;
220 	unsigned int integrity_iv_size;
221 	unsigned int used_tag_size;
222 	unsigned int tuple_size;
223 
224 	/*
225 	 * pool for per bio private data, crypto requests,
226 	 * encryption requeusts/buffer pages and integrity tags
227 	 */
228 	unsigned int tag_pool_max_sectors;
229 	mempool_t tag_pool;
230 	mempool_t req_pool;
231 	mempool_t page_pool;
232 
233 	struct bio_set bs;
234 	struct mutex bio_alloc_lock;
235 
236 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
237 	u8 key[] __counted_by(key_size);
238 };
239 
240 #define MIN_IOS		64
241 #define MAX_TAG_SIZE	480
242 #define POOL_ENTRY_SIZE	512
243 
244 /*
245  * Largest encryption sector size that can be requested with the
246  * "sector_size:<bytes>" option.
247  *
248  * A sector is handed to the crypto API as a single scatterlist entry, so it
249  * has to be covered by one bio_vec.  bio_iter_iovec() never returns more than
250  * PAGE_SIZE bytes, and crypt_alloc_buffer() may fall back to order-0 pages
251  * for the write bounce buffer, so PAGE_SIZE is the ceiling.
252  *
253  * crypt_io_hints() announces the sector size as the logical block size, which
254  * the block layer caps at BLK_MAX_BLOCK_SIZE.  That cap is never below
255  * PAGE_SIZE in any configuration today, so it does not lower the limit; take
256  * the minimum anyway so that this target cannot announce a block size
257  * blk_validate_limits() would reject.
258  */
259 #define DM_CRYPT_MAX_SECTOR_SIZE	MIN_T(unsigned int, PAGE_SIZE, \
260 					      BLK_MAX_BLOCK_SIZE)
261 
262 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
263 static unsigned int dm_crypt_clients_n;
264 static volatile unsigned long dm_crypt_pages_per_client;
265 #define DM_CRYPT_MEMORY_PERCENT			2
266 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
267 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
268 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
269 
270 static unsigned int max_read_size = 0;
271 module_param(max_read_size, uint, 0644);
272 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
273 static unsigned int max_write_size = 0;
274 module_param(max_write_size, uint, 0644);
275 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
276 
277 static unsigned get_max_request_sectors(struct dm_target *ti, struct bio *bio, bool no_split)
278 {
279 	struct crypt_config *cc = ti->private;
280 	unsigned val, sector_align;
281 	bool wrt = op_is_write(bio_op(bio));
282 
283 	if (no_split) {
284 		val = -1;
285 	} else if (wrt) {
286 		val = min_not_zero(READ_ONCE(max_write_size),
287 				   DM_CRYPT_DEFAULT_MAX_WRITE_SIZE);
288 	} else {
289 		val = min_not_zero(READ_ONCE(max_read_size),
290 				   DM_CRYPT_DEFAULT_MAX_READ_SIZE);
291 	}
292 
293 	if (wrt || cc->used_tag_size)
294 		val = min(val, BIO_MAX_VECS << PAGE_SHIFT);
295 
296 	sector_align = max(bdev_logical_block_size(cc->dev->bdev),
297 			   (unsigned)cc->sector_size);
298 	val = round_down(val, sector_align);
299 	if (unlikely(!val))
300 		val = sector_align;
301 	return val >> SECTOR_SHIFT;
302 }
303 
304 static void crypt_endio(struct bio *clone);
305 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
306 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
307 					     struct scatterlist *sg);
308 
309 static bool crypt_integrity_aead(struct crypt_config *cc);
310 
311 /*
312  * Use this to access cipher attributes that are independent of the key.
313  */
314 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
315 {
316 	return cc->cipher_tfm.tfms[0];
317 }
318 
319 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
320 {
321 	return cc->cipher_tfm.tfms_aead[0];
322 }
323 
324 /*
325  * Different IV generation algorithms:
326  *
327  * plain: the initial vector is the 32-bit little-endian version of the sector
328  *        number, padded with zeros if necessary.
329  *
330  * plain64: the initial vector is the 64-bit little-endian version of the sector
331  *        number, padded with zeros if necessary.
332  *
333  * plain64be: the initial vector is the 64-bit big-endian version of the sector
334  *        number, padded with zeros if necessary.
335  *
336  * essiv: "encrypted sector|salt initial vector", the sector number is
337  *        encrypted with the bulk cipher using a salt as key. The salt
338  *        should be derived from the bulk cipher's key via hashing.
339  *
340  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
341  *        (needed for LRW-32-AES and possible other narrow block modes)
342  *
343  * null: the initial vector is always zero.  Provides compatibility with
344  *       obsolete loop_fish2 devices.  Do not use for new devices.
345  *
346  * lmk:  Compatible implementation of the block chaining mode used
347  *       by the Loop-AES block device encryption system
348  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
349  *       It operates on full 512 byte sectors and uses CBC
350  *       with an IV derived from the sector number, the data and
351  *       optionally extra IV seed.
352  *       This means that after decryption the first block
353  *       of sector must be tweaked according to decrypted data.
354  *       Loop-AES can use three encryption schemes:
355  *         version 1: is plain aes-cbc mode
356  *         version 2: uses 64 multikey scheme with lmk IV generator
357  *         version 3: the same as version 2 with additional IV seed
358  *                   (it uses 65 keys, last key is used as IV seed)
359  *
360  * tcw:  Compatible implementation of the block chaining mode used
361  *       by the TrueCrypt device encryption system (prior to version 4.1).
362  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
363  *       It operates on full 512 byte sectors and uses CBC
364  *       with an IV derived from initial key and the sector number.
365  *       In addition, whitening value is applied on every sector, whitening
366  *       is calculated from initial key, sector number and mixed using CRC32.
367  *       Note that this encryption scheme is vulnerable to watermarking attacks
368  *       and should be used for old compatible containers access only.
369  *
370  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
371  *        The IV is encrypted little-endian byte-offset (with the same key
372  *        and cipher as the volume).
373  *
374  * elephant: The extended version of eboiv with additional Elephant diffuser
375  *           used with Bitlocker CBC mode.
376  *           This mode was used in older Windows systems
377  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
378  */
379 
380 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
381 			      struct dm_crypt_request *dmreq)
382 {
383 	memset(iv, 0, cc->iv_size);
384 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
385 
386 	return 0;
387 }
388 
389 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
390 				struct dm_crypt_request *dmreq)
391 {
392 	memset(iv, 0, cc->iv_size);
393 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
394 
395 	return 0;
396 }
397 
398 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
399 				  struct dm_crypt_request *dmreq)
400 {
401 	memset(iv, 0, cc->iv_size);
402 	/* iv_size is at least of size u64; usually it is 16 bytes */
403 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
404 
405 	return 0;
406 }
407 
408 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
409 			      struct dm_crypt_request *dmreq)
410 {
411 	/*
412 	 * ESSIV encryption of the IV is now handled by the crypto API,
413 	 * so just pass the plain sector number here.
414 	 */
415 	memset(iv, 0, cc->iv_size);
416 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
417 
418 	return 0;
419 }
420 
421 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
422 			      const char *opts)
423 {
424 	unsigned int bs;
425 	int log;
426 
427 	if (crypt_integrity_aead(cc))
428 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
429 	else
430 		bs = crypto_skcipher_blocksize(any_tfm(cc));
431 	log = ilog2(bs);
432 
433 	/*
434 	 * We need to calculate how far we must shift the sector count
435 	 * to get the cipher block count, we use this shift in _gen.
436 	 */
437 	if (1 << log != bs) {
438 		ti->error = "cypher blocksize is not a power of 2";
439 		return -EINVAL;
440 	}
441 
442 	if (log > 9) {
443 		ti->error = "cypher blocksize is > 512";
444 		return -EINVAL;
445 	}
446 
447 	cc->iv_gen_private.benbi.shift = 9 - log;
448 
449 	return 0;
450 }
451 
452 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
453 {
454 }
455 
456 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
457 			      struct dm_crypt_request *dmreq)
458 {
459 	__be64 val;
460 
461 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
462 
463 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
464 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
465 
466 	return 0;
467 }
468 
469 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
470 			     struct dm_crypt_request *dmreq)
471 {
472 	memset(iv, 0, cc->iv_size);
473 
474 	return 0;
475 }
476 
477 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
478 {
479 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
480 
481 	kfree_sensitive(lmk->seed);
482 	lmk->seed = NULL;
483 }
484 
485 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
486 			    const char *opts)
487 {
488 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
489 
490 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
491 		ti->error = "Unsupported sector size for LMK";
492 		return -EINVAL;
493 	}
494 
495 	if (fips_enabled) {
496 		ti->error = "LMK support is disabled due to FIPS";
497 		/* ... because it uses MD5. */
498 		return -EINVAL;
499 	}
500 
501 	/* No seed in LMK version 2 */
502 	if (cc->key_parts == cc->tfms_count) {
503 		lmk->seed = NULL;
504 		return 0;
505 	}
506 
507 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
508 	if (!lmk->seed) {
509 		ti->error = "Error kmallocing seed storage in LMK";
510 		return -ENOMEM;
511 	}
512 
513 	return 0;
514 }
515 
516 static int crypt_iv_lmk_init(struct crypt_config *cc)
517 {
518 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
519 	int subkey_size = cc->key_size / cc->key_parts;
520 
521 	/* LMK seed is on the position of LMK_KEYS + 1 key */
522 	if (lmk->seed)
523 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
524 		       MD5_DIGEST_SIZE);
525 
526 	return 0;
527 }
528 
529 static void crypt_iv_lmk_wipe(struct crypt_config *cc)
530 {
531 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
532 
533 	if (lmk->seed)
534 		memzero_explicit(lmk->seed, LMK_SEED_SIZE);
535 }
536 
537 static void crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
538 			     struct dm_crypt_request *dmreq, u8 *data)
539 {
540 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
541 	struct md5_ctx ctx;
542 	__le32 buf[4];
543 
544 	md5_init(&ctx);
545 
546 	if (lmk->seed)
547 		md5_update(&ctx, lmk->seed, LMK_SEED_SIZE);
548 
549 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
550 	md5_update(&ctx, data + 16, 16 * 31);
551 
552 	/* Sector is cropped to 56 bits here */
553 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
554 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
555 	buf[2] = cpu_to_le32(4024);
556 	buf[3] = 0;
557 	md5_update(&ctx, (u8 *)buf, sizeof(buf));
558 
559 	/* No MD5 padding here */
560 	cpu_to_le32_array(ctx.state.h, ARRAY_SIZE(ctx.state.h));
561 	memcpy(iv, ctx.state.h, cc->iv_size);
562 }
563 
564 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
565 			    struct dm_crypt_request *dmreq)
566 {
567 	struct scatterlist *sg;
568 	u8 *src;
569 
570 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
571 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
572 		src = kmap_local_page(sg_page(sg));
573 		crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
574 		kunmap_local(src);
575 	} else
576 		memset(iv, 0, cc->iv_size);
577 	return 0;
578 }
579 
580 static void crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
581 			      struct dm_crypt_request *dmreq)
582 {
583 	struct scatterlist *sg;
584 	u8 *dst;
585 
586 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
587 		return;
588 
589 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
590 	dst = kmap_local_page(sg_page(sg));
591 	crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
592 
593 	/* Tweak the first block of plaintext sector */
594 	crypto_xor(dst + sg->offset, iv, cc->iv_size);
595 
596 	kunmap_local(dst);
597 }
598 
599 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
600 {
601 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
602 
603 	kfree_sensitive(tcw->iv_seed);
604 	tcw->iv_seed = NULL;
605 	kfree_sensitive(tcw->whitening);
606 	tcw->whitening = NULL;
607 }
608 
609 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
610 			    const char *opts)
611 {
612 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
613 
614 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
615 		ti->error = "Unsupported sector size for TCW";
616 		return -EINVAL;
617 	}
618 
619 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
620 		ti->error = "Wrong key size for TCW";
621 		return -EINVAL;
622 	}
623 
624 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
625 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
626 	if (!tcw->iv_seed || !tcw->whitening) {
627 		crypt_iv_tcw_dtr(cc);
628 		ti->error = "Error allocating seed storage in TCW";
629 		return -ENOMEM;
630 	}
631 
632 	return 0;
633 }
634 
635 static int crypt_iv_tcw_init(struct crypt_config *cc)
636 {
637 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
638 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
639 
640 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
641 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
642 	       TCW_WHITENING_SIZE);
643 
644 	return 0;
645 }
646 
647 static void crypt_iv_tcw_wipe(struct crypt_config *cc)
648 {
649 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
650 
651 	memzero_explicit(tcw->iv_seed, cc->iv_size);
652 	memzero_explicit(tcw->whitening, TCW_WHITENING_SIZE);
653 }
654 
655 static void crypt_iv_tcw_whitening(struct crypt_config *cc,
656 				   struct dm_crypt_request *dmreq, u8 *data)
657 {
658 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
659 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
660 	u8 buf[TCW_WHITENING_SIZE];
661 	int i;
662 
663 	/* xor whitening with sector number */
664 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
665 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
666 
667 	/* calculate crc32 for every 32bit part and xor it */
668 	for (i = 0; i < 4; i++)
669 		put_unaligned_le32(crc32(0, &buf[i * 4], 4), &buf[i * 4]);
670 	crypto_xor(&buf[0], &buf[12], 4);
671 	crypto_xor(&buf[4], &buf[8], 4);
672 
673 	/* apply whitening (8 bytes) to whole sector */
674 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
675 		crypto_xor(data + i * 8, buf, 8);
676 	memzero_explicit(buf, sizeof(buf));
677 }
678 
679 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
680 			    struct dm_crypt_request *dmreq)
681 {
682 	struct scatterlist *sg;
683 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
684 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
685 	u8 *src;
686 
687 	/* Remove whitening from ciphertext */
688 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
689 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
690 		src = kmap_local_page(sg_page(sg));
691 		crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
692 		kunmap_local(src);
693 	}
694 
695 	/* Calculate IV */
696 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
697 	if (cc->iv_size > 8)
698 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
699 			       cc->iv_size - 8);
700 
701 	return 0;
702 }
703 
704 static void crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
705 			      struct dm_crypt_request *dmreq)
706 {
707 	struct scatterlist *sg;
708 	u8 *dst;
709 
710 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
711 		return;
712 
713 	/* Apply whitening on ciphertext */
714 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
715 	dst = kmap_local_page(sg_page(sg));
716 	crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
717 	kunmap_local(dst);
718 }
719 
720 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
721 				struct dm_crypt_request *dmreq)
722 {
723 	/* Used only for writes, there must be an additional space to store IV */
724 	get_random_bytes(iv, cc->iv_size);
725 	return 0;
726 }
727 
728 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
729 			    const char *opts)
730 {
731 	if (crypt_integrity_aead(cc)) {
732 		ti->error = "AEAD transforms not supported for EBOIV";
733 		return -EINVAL;
734 	}
735 
736 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
737 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
738 		return -EINVAL;
739 	}
740 
741 	return 0;
742 }
743 
744 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
745 			    struct dm_crypt_request *dmreq)
746 {
747 	struct crypto_skcipher *tfm = any_tfm(cc);
748 	struct skcipher_request *req;
749 	struct scatterlist src, dst;
750 	DECLARE_CRYPTO_WAIT(wait);
751 	unsigned int reqsize;
752 	int err;
753 	u8 *buf;
754 
755 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
756 	reqsize = ALIGN(reqsize, __alignof__(__le64));
757 
758 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
759 	if (!req)
760 		return -ENOMEM;
761 
762 	skcipher_request_set_tfm(req, tfm);
763 
764 	buf = (u8 *)req + reqsize;
765 	memset(buf, 0, cc->iv_size);
766 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
767 
768 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
769 	sg_init_one(&dst, iv, cc->iv_size);
770 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
771 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
772 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
773 	kfree_sensitive(req);
774 
775 	return err;
776 }
777 
778 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
779 {
780 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
781 
782 	kfree_sensitive(elephant->key);
783 	elephant->key = NULL;
784 }
785 
786 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
787 			    const char *opts)
788 {
789 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
790 	int r;
791 
792 	elephant->key = kmalloc_obj(*elephant->key);
793 	if (!elephant->key)
794 		return -ENOMEM;
795 
796 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
797 	if (r)
798 		crypt_iv_elephant_dtr(cc);
799 	return r;
800 }
801 
802 static void diffuser_disk_to_cpu(u32 *d, size_t n)
803 {
804 #ifndef __LITTLE_ENDIAN
805 	int i;
806 
807 	for (i = 0; i < n; i++)
808 		d[i] = le32_to_cpu((__le32)d[i]);
809 #endif
810 }
811 
812 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
813 {
814 #ifndef __LITTLE_ENDIAN
815 	int i;
816 
817 	for (i = 0; i < n; i++)
818 		d[i] = cpu_to_le32((u32)d[i]);
819 #endif
820 }
821 
822 static void diffuser_a_decrypt(u32 *d, size_t n)
823 {
824 	int i, i1, i2, i3;
825 
826 	for (i = 0; i < 5; i++) {
827 		i1 = 0;
828 		i2 = n - 2;
829 		i3 = n - 5;
830 
831 		while (i1 < (n - 1)) {
832 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
833 			i1++; i2++; i3++;
834 
835 			if (i3 >= n)
836 				i3 -= n;
837 
838 			d[i1] += d[i2] ^ d[i3];
839 			i1++; i2++; i3++;
840 
841 			if (i2 >= n)
842 				i2 -= n;
843 
844 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
845 			i1++; i2++; i3++;
846 
847 			d[i1] += d[i2] ^ d[i3];
848 			i1++; i2++; i3++;
849 		}
850 	}
851 }
852 
853 static void diffuser_a_encrypt(u32 *d, size_t n)
854 {
855 	int i, i1, i2, i3;
856 
857 	for (i = 0; i < 5; i++) {
858 		i1 = n - 1;
859 		i2 = n - 2 - 1;
860 		i3 = n - 5 - 1;
861 
862 		while (i1 > 0) {
863 			d[i1] -= d[i2] ^ d[i3];
864 			i1--; i2--; i3--;
865 
866 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
867 			i1--; i2--; i3--;
868 
869 			if (i2 < 0)
870 				i2 += n;
871 
872 			d[i1] -= d[i2] ^ d[i3];
873 			i1--; i2--; i3--;
874 
875 			if (i3 < 0)
876 				i3 += n;
877 
878 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
879 			i1--; i2--; i3--;
880 		}
881 	}
882 }
883 
884 static void diffuser_b_decrypt(u32 *d, size_t n)
885 {
886 	int i, i1, i2, i3;
887 
888 	for (i = 0; i < 3; i++) {
889 		i1 = 0;
890 		i2 = 2;
891 		i3 = 5;
892 
893 		while (i1 < (n - 1)) {
894 			d[i1] += d[i2] ^ d[i3];
895 			i1++; i2++; i3++;
896 
897 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
898 			i1++; i2++; i3++;
899 
900 			if (i2 >= n)
901 				i2 -= n;
902 
903 			d[i1] += d[i2] ^ d[i3];
904 			i1++; i2++; i3++;
905 
906 			if (i3 >= n)
907 				i3 -= n;
908 
909 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
910 			i1++; i2++; i3++;
911 		}
912 	}
913 }
914 
915 static void diffuser_b_encrypt(u32 *d, size_t n)
916 {
917 	int i, i1, i2, i3;
918 
919 	for (i = 0; i < 3; i++) {
920 		i1 = n - 1;
921 		i2 = 2 - 1;
922 		i3 = 5 - 1;
923 
924 		while (i1 > 0) {
925 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
926 			i1--; i2--; i3--;
927 
928 			if (i3 < 0)
929 				i3 += n;
930 
931 			d[i1] -= d[i2] ^ d[i3];
932 			i1--; i2--; i3--;
933 
934 			if (i2 < 0)
935 				i2 += n;
936 
937 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
938 			i1--; i2--; i3--;
939 
940 			d[i1] -= d[i2] ^ d[i3];
941 			i1--; i2--; i3--;
942 		}
943 	}
944 }
945 
946 static void crypt_iv_elephant(struct crypt_config *cc,
947 			      struct dm_crypt_request *dmreq)
948 {
949 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
950 	u8 *data, *data2, *data_offset;
951 	struct scatterlist *sg, *sg2;
952 	union {
953 		__le64 w[2];
954 		u8 b[16];
955 	} es;
956 	u8 ks[32] __aligned(__alignof(long)); /* Elephant sector key */
957 	int i;
958 
959 	es.w[0] = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
960 	es.w[1] = 0;
961 
962 	/* E(Ks, e(s)) */
963 	aes_encrypt(elephant->key, &ks[0], es.b);
964 
965 	/* E(Ks, e'(s)) */
966 	es.b[15] = 0x80;
967 	aes_encrypt(elephant->key, &ks[16], es.b);
968 
969 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
970 	data = kmap_local_page(sg_page(sg));
971 	data_offset = data + sg->offset;
972 
973 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
974 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
975 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
976 		data2 = kmap_local_page(sg_page(sg2));
977 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
978 		kunmap_local(data2);
979 	}
980 
981 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
982 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
983 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
984 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
985 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
986 	}
987 
988 	for (i = 0; i < (cc->sector_size / 32); i++)
989 		crypto_xor(data_offset + i * 32, ks, 32);
990 
991 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
992 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
993 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
994 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
995 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
996 	}
997 
998 	kunmap_local(data);
999 	memzero_explicit(ks, sizeof(ks));
1000 	memzero_explicit(&es, sizeof(es));
1001 }
1002 
1003 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1004 			    struct dm_crypt_request *dmreq)
1005 {
1006 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
1007 		crypt_iv_elephant(cc, dmreq);
1008 
1009 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1010 }
1011 
1012 static void crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1013 				   struct dm_crypt_request *dmreq)
1014 {
1015 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1016 		crypt_iv_elephant(cc, dmreq);
1017 }
1018 
1019 static int crypt_iv_elephant_init(struct crypt_config *cc)
1020 {
1021 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1022 	int key_offset = cc->key_size - cc->key_extra_size;
1023 
1024 	return aes_prepareenckey(elephant->key, &cc->key[key_offset], cc->key_extra_size);
1025 }
1026 
1027 static void crypt_iv_elephant_wipe(struct crypt_config *cc)
1028 {
1029 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1030 
1031 	memzero_explicit(elephant->key, sizeof(*elephant->key));
1032 }
1033 
1034 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1035 	.generator = crypt_iv_plain_gen
1036 };
1037 
1038 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1039 	.generator = crypt_iv_plain64_gen
1040 };
1041 
1042 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1043 	.generator = crypt_iv_plain64be_gen
1044 };
1045 
1046 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1047 	.generator = crypt_iv_essiv_gen
1048 };
1049 
1050 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1051 	.ctr	   = crypt_iv_benbi_ctr,
1052 	.dtr	   = crypt_iv_benbi_dtr,
1053 	.generator = crypt_iv_benbi_gen
1054 };
1055 
1056 static const struct crypt_iv_operations crypt_iv_null_ops = {
1057 	.generator = crypt_iv_null_gen
1058 };
1059 
1060 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1061 	.ctr	   = crypt_iv_lmk_ctr,
1062 	.dtr	   = crypt_iv_lmk_dtr,
1063 	.init	   = crypt_iv_lmk_init,
1064 	.wipe	   = crypt_iv_lmk_wipe,
1065 	.generator = crypt_iv_lmk_gen,
1066 	.post	   = crypt_iv_lmk_post
1067 };
1068 
1069 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1070 	.ctr	   = crypt_iv_tcw_ctr,
1071 	.dtr	   = crypt_iv_tcw_dtr,
1072 	.init	   = crypt_iv_tcw_init,
1073 	.wipe	   = crypt_iv_tcw_wipe,
1074 	.generator = crypt_iv_tcw_gen,
1075 	.post	   = crypt_iv_tcw_post
1076 };
1077 
1078 static const struct crypt_iv_operations crypt_iv_random_ops = {
1079 	.generator = crypt_iv_random_gen
1080 };
1081 
1082 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1083 	.ctr	   = crypt_iv_eboiv_ctr,
1084 	.generator = crypt_iv_eboiv_gen
1085 };
1086 
1087 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1088 	.ctr	   = crypt_iv_elephant_ctr,
1089 	.dtr	   = crypt_iv_elephant_dtr,
1090 	.init	   = crypt_iv_elephant_init,
1091 	.wipe	   = crypt_iv_elephant_wipe,
1092 	.generator = crypt_iv_elephant_gen,
1093 	.post	   = crypt_iv_elephant_post
1094 };
1095 
1096 /*
1097  * Integrity extensions
1098  */
1099 static bool crypt_integrity_aead(struct crypt_config *cc)
1100 {
1101 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1102 }
1103 
1104 static bool crypt_integrity_hmac(struct crypt_config *cc)
1105 {
1106 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1107 }
1108 
1109 /* Get sg containing data */
1110 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1111 					     struct scatterlist *sg)
1112 {
1113 	if (unlikely(crypt_integrity_aead(cc)))
1114 		return &sg[2];
1115 
1116 	return sg;
1117 }
1118 
1119 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1120 {
1121 	struct bio_integrity_payload *bip;
1122 	unsigned int tag_len;
1123 	int ret;
1124 
1125 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1126 		return 0;
1127 
1128 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1129 	if (IS_ERR(bip))
1130 		return PTR_ERR(bip);
1131 
1132 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1133 
1134 	bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
1135 
1136 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1137 				     tag_len, offset_in_page(io->integrity_metadata));
1138 	if (unlikely(ret != tag_len))
1139 		return -ENOMEM;
1140 
1141 	return 0;
1142 }
1143 
1144 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1145 {
1146 #ifdef CONFIG_BLK_DEV_INTEGRITY
1147 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1148 	struct mapped_device *md = dm_table_get_md(ti->table);
1149 
1150 	/* We require an underlying device with non-PI metadata */
1151 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1152 		ti->error = "Integrity profile not supported.";
1153 		return -EINVAL;
1154 	}
1155 
1156 	if (bi->metadata_size < cc->used_tag_size) {
1157 		ti->error = "Integrity profile tag size mismatch.";
1158 		return -EINVAL;
1159 	}
1160 	cc->tuple_size = bi->metadata_size;
1161 	if (1 << bi->interval_exp != cc->sector_size) {
1162 		ti->error = "Integrity profile sector size mismatch.";
1163 		return -EINVAL;
1164 	}
1165 
1166 	if (crypt_integrity_aead(cc)) {
1167 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1168 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1169 		       cc->integrity_tag_size, cc->integrity_iv_size);
1170 
1171 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1172 			ti->error = "Integrity AEAD auth tag size is not supported.";
1173 			return -EINVAL;
1174 		}
1175 	} else if (cc->integrity_iv_size)
1176 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1177 		       cc->integrity_iv_size);
1178 
1179 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1180 		ti->error = "Not enough space for integrity tag in the profile.";
1181 		return -EINVAL;
1182 	}
1183 
1184 	return 0;
1185 #else
1186 	ti->error = "Integrity profile not supported.";
1187 	return -EINVAL;
1188 #endif
1189 }
1190 
1191 static void crypt_convert_init(struct crypt_config *cc,
1192 			       struct convert_context *ctx,
1193 			       struct bio *bio_out, struct bio *bio_in,
1194 			       sector_t sector)
1195 {
1196 	ctx->bio_in = bio_in;
1197 	ctx->bio_out = bio_out;
1198 	if (bio_in)
1199 		ctx->iter_in = bio_in->bi_iter;
1200 	if (bio_out)
1201 		ctx->iter_out = bio_out->bi_iter;
1202 	ctx->cc_sector = sector + cc->iv_offset;
1203 	ctx->tag_offset = 0;
1204 	init_completion(&ctx->restart);
1205 }
1206 
1207 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1208 					     void *req)
1209 {
1210 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1211 }
1212 
1213 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1214 {
1215 	return (void *)((char *)dmreq - cc->dmreq_start);
1216 }
1217 
1218 static u8 *iv_of_dmreq(struct crypt_config *cc,
1219 		       struct dm_crypt_request *dmreq)
1220 {
1221 	if (crypt_integrity_aead(cc))
1222 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1223 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1224 	else
1225 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1226 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1227 }
1228 
1229 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1230 		       struct dm_crypt_request *dmreq)
1231 {
1232 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1233 }
1234 
1235 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1236 		       struct dm_crypt_request *dmreq)
1237 {
1238 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1239 
1240 	return (__le64 *) ptr;
1241 }
1242 
1243 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1244 		       struct dm_crypt_request *dmreq)
1245 {
1246 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1247 		  cc->iv_size + sizeof(uint64_t);
1248 
1249 	return (unsigned int *)ptr;
1250 }
1251 
1252 static void *tag_from_dmreq(struct crypt_config *cc,
1253 				struct dm_crypt_request *dmreq)
1254 {
1255 	struct convert_context *ctx = dmreq->ctx;
1256 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1257 
1258 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1259 		cc->tuple_size];
1260 }
1261 
1262 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1263 			       struct dm_crypt_request *dmreq)
1264 {
1265 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1266 }
1267 
1268 static int crypt_convert_block_aead(struct crypt_config *cc,
1269 				     struct convert_context *ctx,
1270 				     struct aead_request *req,
1271 				     unsigned int tag_offset)
1272 {
1273 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1274 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1275 	struct dm_crypt_request *dmreq;
1276 	u8 *iv, *org_iv, *tag_iv, *tag;
1277 	__le64 *sector;
1278 	int r = 0;
1279 
1280 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1281 
1282 	/* Reject unexpected unaligned bio. */
1283 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1284 		return -EIO;
1285 
1286 	dmreq = dmreq_of_req(cc, req);
1287 	dmreq->iv_sector = ctx->cc_sector;
1288 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1289 		dmreq->iv_sector >>= cc->sector_shift;
1290 	dmreq->ctx = ctx;
1291 
1292 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1293 
1294 	sector = org_sector_of_dmreq(cc, dmreq);
1295 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1296 
1297 	iv = iv_of_dmreq(cc, dmreq);
1298 	org_iv = org_iv_of_dmreq(cc, dmreq);
1299 	tag = tag_from_dmreq(cc, dmreq);
1300 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1301 
1302 	/* AEAD request:
1303 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1304 	 *  | (authenticated) | (auth+encryption) |              |
1305 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1306 	 */
1307 	sg_init_table(dmreq->sg_in, 4);
1308 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1309 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1310 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1311 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1312 
1313 	sg_init_table(dmreq->sg_out, 4);
1314 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1315 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1316 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1317 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1318 
1319 	if (cc->iv_gen_ops) {
1320 		/* For READs use IV stored in integrity metadata */
1321 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1322 			memcpy(org_iv, tag_iv, cc->iv_size);
1323 		} else {
1324 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1325 			if (r < 0)
1326 				return r;
1327 			/* Store generated IV in integrity metadata */
1328 			if (cc->integrity_iv_size)
1329 				memcpy(tag_iv, org_iv, cc->iv_size);
1330 		}
1331 		/* Working copy of IV, to be modified in crypto API */
1332 		memcpy(iv, org_iv, cc->iv_size);
1333 	}
1334 
1335 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1336 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1337 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1338 				       cc->sector_size, iv);
1339 		r = crypto_aead_encrypt(req);
1340 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1341 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1342 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1343 	} else {
1344 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1345 				       cc->sector_size + cc->integrity_tag_size, iv);
1346 		r = crypto_aead_decrypt(req);
1347 	}
1348 
1349 	if (r == -EBADMSG) {
1350 		sector_t s = le64_to_cpu(*sector);
1351 
1352 		ctx->aead_failed = true;
1353 		if (ctx->aead_recheck) {
1354 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1355 				    ctx->bio_in->bi_bdev, s);
1356 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1357 					 ctx->bio_in, s, 0);
1358 		}
1359 	}
1360 
1361 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1362 		cc->iv_gen_ops->post(cc, org_iv, dmreq);
1363 
1364 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1365 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1366 
1367 	return r;
1368 }
1369 
1370 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1371 					struct convert_context *ctx,
1372 					struct skcipher_request *req,
1373 					unsigned int tag_offset)
1374 {
1375 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1376 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1377 	struct scatterlist *sg_in, *sg_out;
1378 	struct dm_crypt_request *dmreq;
1379 	u8 *iv, *org_iv, *tag_iv;
1380 	__le64 *sector;
1381 	int r = 0;
1382 
1383 	/* Reject unexpected unaligned bio. */
1384 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1385 		return -EIO;
1386 
1387 	dmreq = dmreq_of_req(cc, req);
1388 	dmreq->iv_sector = ctx->cc_sector;
1389 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1390 		dmreq->iv_sector >>= cc->sector_shift;
1391 	dmreq->ctx = ctx;
1392 
1393 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1394 
1395 	iv = iv_of_dmreq(cc, dmreq);
1396 	org_iv = org_iv_of_dmreq(cc, dmreq);
1397 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1398 
1399 	sector = org_sector_of_dmreq(cc, dmreq);
1400 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1401 
1402 	/* For skcipher we use only the first sg item */
1403 	sg_in  = &dmreq->sg_in[0];
1404 	sg_out = &dmreq->sg_out[0];
1405 
1406 	sg_init_table(sg_in, 1);
1407 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1408 
1409 	sg_init_table(sg_out, 1);
1410 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1411 
1412 	if (cc->iv_gen_ops) {
1413 		/* For READs use IV stored in integrity metadata */
1414 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1415 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1416 		} else {
1417 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1418 			if (r < 0)
1419 				return r;
1420 			/* Data can be already preprocessed in generator */
1421 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1422 				sg_in = sg_out;
1423 			/* Store generated IV in integrity metadata */
1424 			if (cc->integrity_iv_size)
1425 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1426 		}
1427 		/* Working copy of IV, to be modified in crypto API */
1428 		memcpy(iv, org_iv, cc->iv_size);
1429 	}
1430 
1431 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1432 
1433 	if (bio_data_dir(ctx->bio_in) == WRITE)
1434 		r = crypto_skcipher_encrypt(req);
1435 	else
1436 		r = crypto_skcipher_decrypt(req);
1437 
1438 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1439 		cc->iv_gen_ops->post(cc, org_iv, dmreq);
1440 
1441 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1442 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1443 
1444 	return r;
1445 }
1446 
1447 static void kcryptd_async_done(void *async_req, int error);
1448 
1449 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1450 				     struct convert_context *ctx)
1451 {
1452 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1453 
1454 	if (!ctx->r.req) {
1455 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1456 		if (!ctx->r.req)
1457 			return -ENOMEM;
1458 	}
1459 
1460 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1461 
1462 	/*
1463 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1464 	 * requests if driver request queue is full.
1465 	 */
1466 	skcipher_request_set_callback(ctx->r.req,
1467 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1468 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1469 
1470 	return 0;
1471 }
1472 
1473 static int crypt_alloc_req_aead(struct crypt_config *cc,
1474 				 struct convert_context *ctx)
1475 {
1476 	if (!ctx->r.req_aead) {
1477 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1478 		if (!ctx->r.req_aead)
1479 			return -ENOMEM;
1480 	}
1481 
1482 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1483 
1484 	/*
1485 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1486 	 * requests if driver request queue is full.
1487 	 */
1488 	aead_request_set_callback(ctx->r.req_aead,
1489 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1490 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1491 
1492 	return 0;
1493 }
1494 
1495 static int crypt_alloc_req(struct crypt_config *cc,
1496 			    struct convert_context *ctx)
1497 {
1498 	if (crypt_integrity_aead(cc))
1499 		return crypt_alloc_req_aead(cc, ctx);
1500 	else
1501 		return crypt_alloc_req_skcipher(cc, ctx);
1502 }
1503 
1504 static void crypt_free_req_skcipher(struct crypt_config *cc,
1505 				    struct skcipher_request *req, struct bio *base_bio)
1506 {
1507 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1508 
1509 	if ((struct skcipher_request *)(io + 1) != req)
1510 		mempool_free(req, &cc->req_pool);
1511 }
1512 
1513 static void crypt_free_req_aead(struct crypt_config *cc,
1514 				struct aead_request *req, struct bio *base_bio)
1515 {
1516 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1517 
1518 	if ((struct aead_request *)(io + 1) != req)
1519 		mempool_free(req, &cc->req_pool);
1520 }
1521 
1522 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1523 {
1524 	if (crypt_integrity_aead(cc))
1525 		crypt_free_req_aead(cc, req, base_bio);
1526 	else
1527 		crypt_free_req_skcipher(cc, req, base_bio);
1528 }
1529 
1530 /*
1531  * Encrypt / decrypt data from one bio to another one (can be the same one)
1532  */
1533 static blk_status_t crypt_convert(struct crypt_config *cc,
1534 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1535 {
1536 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1537 	int r;
1538 
1539 	/*
1540 	 * if reset_pending is set we are dealing with the bio for the first time,
1541 	 * else we're continuing to work on the previous bio, so don't mess with
1542 	 * the cc_pending counter
1543 	 */
1544 	if (reset_pending)
1545 		atomic_set(&ctx->cc_pending, 1);
1546 
1547 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1548 
1549 		r = crypt_alloc_req(cc, ctx);
1550 		if (r) {
1551 			complete(&ctx->restart);
1552 			return BLK_STS_DEV_RESOURCE;
1553 		}
1554 
1555 		atomic_inc(&ctx->cc_pending);
1556 
1557 		if (crypt_integrity_aead(cc))
1558 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, ctx->tag_offset);
1559 		else
1560 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->tag_offset);
1561 
1562 		switch (r) {
1563 		/*
1564 		 * The request was queued by a crypto driver
1565 		 * but the driver request queue is full, let's wait.
1566 		 */
1567 		case -EBUSY:
1568 			if (in_interrupt()) {
1569 				if (try_wait_for_completion(&ctx->restart)) {
1570 					/*
1571 					 * we don't have to block to wait for completion,
1572 					 * so proceed
1573 					 */
1574 				} else {
1575 					/*
1576 					 * we can't wait for completion without blocking
1577 					 * exit and continue processing in a workqueue
1578 					 */
1579 					ctx->r.req = NULL;
1580 					ctx->tag_offset++;
1581 					ctx->cc_sector += sector_step;
1582 					return BLK_STS_DEV_RESOURCE;
1583 				}
1584 			} else {
1585 				wait_for_completion(&ctx->restart);
1586 			}
1587 			reinit_completion(&ctx->restart);
1588 			fallthrough;
1589 		/*
1590 		 * The request is queued and processed asynchronously,
1591 		 * completion function kcryptd_async_done() will be called.
1592 		 */
1593 		case -EINPROGRESS:
1594 			ctx->r.req = NULL;
1595 			ctx->tag_offset++;
1596 			ctx->cc_sector += sector_step;
1597 			continue;
1598 		/*
1599 		 * The request was already processed (synchronously).
1600 		 */
1601 		case 0:
1602 			atomic_dec(&ctx->cc_pending);
1603 			ctx->cc_sector += sector_step;
1604 			ctx->tag_offset++;
1605 			if (!atomic)
1606 				cond_resched();
1607 			continue;
1608 		/*
1609 		 * There was a data integrity error.
1610 		 */
1611 		case -EBADMSG:
1612 			atomic_dec(&ctx->cc_pending);
1613 			return BLK_STS_PROTECTION;
1614 		/*
1615 		 * There was an error while processing the request.
1616 		 */
1617 		default:
1618 			atomic_dec(&ctx->cc_pending);
1619 			return BLK_STS_IOERR;
1620 		}
1621 	}
1622 
1623 	return 0;
1624 }
1625 
1626 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1627 
1628 /*
1629  * Generate a new unfragmented bio with the given size
1630  * This should never violate the device limitations (but if it did then block
1631  * core should split the bio as needed).
1632  *
1633  * This function may be called concurrently. If we allocate from the mempool
1634  * concurrently, there is a possibility of deadlock. For example, if we have
1635  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1636  * the mempool concurrently, it may deadlock in a situation where both processes
1637  * have allocated 128 pages and the mempool is exhausted.
1638  *
1639  * In order to avoid this scenario we allocate the pages under a mutex.
1640  *
1641  * In order to not degrade performance with excessive locking, we try
1642  * non-blocking allocations without a mutex first but on failure we fallback
1643  * to blocking allocations with a mutex.
1644  *
1645  * In order to reduce allocation overhead, we try to allocate compound pages in
1646  * the first pass. If they are not available, we fall back to the mempool.
1647  */
1648 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1649 {
1650 	struct crypt_config *cc = io->cc;
1651 	struct bio *clone;
1652 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1653 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1654 	unsigned int remaining_size;
1655 	unsigned int order = MAX_PAGE_ORDER;
1656 
1657 retry:
1658 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1659 		mutex_lock(&cc->bio_alloc_lock);
1660 
1661 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1662 				 GFP_NOIO, &cc->bs);
1663 	clone->bi_private = io;
1664 	clone->bi_end_io = crypt_endio;
1665 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1666 	clone->bi_iter.bi_sector = cc->start + io->sector;
1667 
1668 	remaining_size = size;
1669 
1670 	while (remaining_size) {
1671 		struct page *pages;
1672 		unsigned size_to_add;
1673 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1674 		order = min(order, remaining_order);
1675 
1676 		while (order > 0) {
1677 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1678 					(1 << order) > dm_crypt_pages_per_client))
1679 				goto decrease_order;
1680 			pages = alloc_pages(gfp_mask
1681 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1682 				order);
1683 			if (likely(pages != NULL)) {
1684 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1685 				goto have_pages;
1686 			}
1687 decrease_order:
1688 			order--;
1689 		}
1690 
1691 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1692 		if (!pages) {
1693 			crypt_free_buffer_pages(cc, clone);
1694 			bio_put(clone);
1695 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1696 			order = 0;
1697 			goto retry;
1698 		}
1699 
1700 have_pages:
1701 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1702 		__bio_add_page(clone, pages, size_to_add, 0);
1703 		remaining_size -= size_to_add;
1704 	}
1705 
1706 	/* Allocate space for integrity tags */
1707 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1708 		crypt_free_buffer_pages(cc, clone);
1709 		bio_put(clone);
1710 		clone = NULL;
1711 	}
1712 
1713 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1714 		mutex_unlock(&cc->bio_alloc_lock);
1715 
1716 	return clone;
1717 }
1718 
1719 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1720 {
1721 	struct folio_iter fi;
1722 
1723 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1724 		bio_for_each_folio_all(fi, clone) {
1725 			if (folio_test_large(fi.folio)) {
1726 				percpu_counter_sub(&cc->n_allocated_pages,
1727 						folio_nr_pages(fi.folio));
1728 				folio_put(fi.folio);
1729 			} else {
1730 				mempool_free(&fi.folio->page, &cc->page_pool);
1731 			}
1732 		}
1733 	}
1734 }
1735 
1736 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1737 			  struct bio *bio, sector_t sector)
1738 {
1739 	io->cc = cc;
1740 	io->base_bio = bio;
1741 	io->sector = sector;
1742 	io->error = 0;
1743 	io->ctx.aead_recheck = false;
1744 	io->ctx.aead_failed = false;
1745 	io->ctx.r.req = NULL;
1746 	io->integrity_metadata = NULL;
1747 	io->integrity_metadata_from_pool = false;
1748 	atomic_set(&io->io_pending, 0);
1749 }
1750 
1751 static void crypt_inc_pending(struct dm_crypt_io *io)
1752 {
1753 	atomic_inc(&io->io_pending);
1754 }
1755 
1756 static void kcryptd_queue_read(struct dm_crypt_io *io);
1757 
1758 /*
1759  * One of the bios was finished. Check for completion of
1760  * the whole request and correctly clean up the buffer.
1761  */
1762 static void crypt_dec_pending(struct dm_crypt_io *io)
1763 {
1764 	struct crypt_config *cc = io->cc;
1765 	struct bio *base_bio = io->base_bio;
1766 
1767 	if (!atomic_dec_and_test(&io->io_pending))
1768 		return;
1769 
1770 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1771 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1772 		io->ctx.aead_recheck = true;
1773 		io->ctx.aead_failed = false;
1774 		io->error = 0;
1775 		kcryptd_queue_read(io);
1776 		return;
1777 	}
1778 
1779 	if (io->ctx.r.req)
1780 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1781 
1782 	if (unlikely(io->integrity_metadata_from_pool))
1783 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1784 	else
1785 		kfree(io->integrity_metadata);
1786 
1787 	base_bio->bi_status = io->error;
1788 
1789 	bio_endio(base_bio);
1790 }
1791 
1792 /*
1793  * kcryptd/kcryptd_io:
1794  *
1795  * Needed because it would be very unwise to do decryption in an
1796  * interrupt context.
1797  *
1798  * kcryptd performs the actual encryption or decryption.
1799  *
1800  * kcryptd_io performs the IO submission.
1801  *
1802  * They must be separated as otherwise the final stages could be
1803  * starved by new requests which can block in the first stages due
1804  * to memory allocation.
1805  *
1806  * The work is done per CPU global for all dm-crypt instances.
1807  * They should not depend on each other and do not block.
1808  */
1809 static void crypt_endio(struct bio *clone)
1810 {
1811 	struct dm_crypt_io *io = clone->bi_private;
1812 	struct crypt_config *cc = io->cc;
1813 	unsigned int rw = bio_data_dir(clone);
1814 	blk_status_t error = clone->bi_status;
1815 
1816 	if (io->ctx.aead_recheck && !error) {
1817 		kcryptd_queue_crypt(io);
1818 		return;
1819 	}
1820 
1821 	/*
1822 	 * free the processed pages
1823 	 */
1824 	if (rw == WRITE || io->ctx.aead_recheck)
1825 		crypt_free_buffer_pages(cc, clone);
1826 
1827 	bio_put(clone);
1828 
1829 	if (rw == READ && !error) {
1830 		kcryptd_queue_crypt(io);
1831 		return;
1832 	}
1833 
1834 	if (unlikely(error))
1835 		io->error = error;
1836 
1837 	crypt_dec_pending(io);
1838 }
1839 
1840 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1841 
1842 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1843 {
1844 	struct crypt_config *cc = io->cc;
1845 	struct bio *clone;
1846 
1847 	if (io->ctx.aead_recheck) {
1848 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1849 			return 1;
1850 		crypt_inc_pending(io);
1851 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1852 		if (unlikely(!clone)) {
1853 			crypt_dec_pending(io);
1854 			return 1;
1855 		}
1856 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1857 		io->saved_bi_iter = clone->bi_iter;
1858 		dm_submit_bio_remap(io->base_bio, clone);
1859 		return 0;
1860 	}
1861 
1862 	/*
1863 	 * We need the original biovec array in order to decrypt the whole bio
1864 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1865 	 * worry about the block layer modifying the biovec array; so leverage
1866 	 * bio_alloc_clone().
1867 	 */
1868 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1869 	if (!clone)
1870 		return 1;
1871 
1872 	clone->bi_iter.bi_sector = cc->start + io->sector;
1873 	clone->bi_private = io;
1874 	clone->bi_end_io = crypt_endio;
1875 
1876 	crypt_inc_pending(io);
1877 
1878 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1879 		crypt_dec_pending(io);
1880 		bio_put(clone);
1881 		return 1;
1882 	}
1883 
1884 	dm_submit_bio_remap(io->base_bio, clone);
1885 	return 0;
1886 }
1887 
1888 static void kcryptd_io_read_work(struct work_struct *work)
1889 {
1890 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1891 
1892 	crypt_inc_pending(io);
1893 	if (kcryptd_io_read(io, GFP_NOIO))
1894 		io->error = BLK_STS_RESOURCE;
1895 	crypt_dec_pending(io);
1896 }
1897 
1898 static void kcryptd_queue_read(struct dm_crypt_io *io)
1899 {
1900 	struct crypt_config *cc = io->cc;
1901 
1902 	INIT_WORK(&io->work, kcryptd_io_read_work);
1903 	queue_work(cc->io_queue, &io->work);
1904 }
1905 
1906 static void kcryptd_io_write(struct dm_crypt_io *io)
1907 {
1908 	struct bio *clone = io->ctx.bio_out;
1909 
1910 	dm_submit_bio_remap(io->base_bio, clone);
1911 }
1912 
1913 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1914 
1915 static int dmcrypt_write(void *data)
1916 {
1917 	struct crypt_config *cc = data;
1918 	struct dm_crypt_io *io;
1919 
1920 	while (1) {
1921 		struct rb_root write_tree;
1922 		struct blk_plug plug;
1923 
1924 		spin_lock_irq(&cc->write_thread_lock);
1925 continue_locked:
1926 
1927 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1928 			goto pop_from_list;
1929 
1930 		set_current_state(TASK_INTERRUPTIBLE);
1931 
1932 		spin_unlock_irq(&cc->write_thread_lock);
1933 
1934 		if (unlikely(kthread_should_stop())) {
1935 			set_current_state(TASK_RUNNING);
1936 			break;
1937 		}
1938 
1939 		schedule();
1940 
1941 		spin_lock_irq(&cc->write_thread_lock);
1942 		goto continue_locked;
1943 
1944 pop_from_list:
1945 		write_tree = cc->write_tree;
1946 		cc->write_tree = RB_ROOT;
1947 		spin_unlock_irq(&cc->write_thread_lock);
1948 
1949 		BUG_ON(rb_parent(write_tree.rb_node));
1950 
1951 		/*
1952 		 * Note: we cannot walk the tree here with rb_next because
1953 		 * the structures may be freed when kcryptd_io_write is called.
1954 		 */
1955 		blk_start_plug(&plug);
1956 		do {
1957 			io = crypt_io_from_node(rb_first(&write_tree));
1958 			rb_erase(&io->rb_node, &write_tree);
1959 			kcryptd_io_write(io);
1960 			cond_resched();
1961 		} while (!RB_EMPTY_ROOT(&write_tree));
1962 		blk_finish_plug(&plug);
1963 	}
1964 	return 0;
1965 }
1966 
1967 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1968 {
1969 	struct bio *clone = io->ctx.bio_out;
1970 	struct crypt_config *cc = io->cc;
1971 	unsigned long flags;
1972 	sector_t sector;
1973 	struct rb_node **rbp, *parent;
1974 
1975 	if (unlikely(io->error)) {
1976 		crypt_free_buffer_pages(cc, clone);
1977 		bio_put(clone);
1978 		crypt_dec_pending(io);
1979 		return;
1980 	}
1981 
1982 	/* crypt_convert should have filled the clone bio */
1983 	BUG_ON(io->ctx.iter_out.bi_size);
1984 
1985 	if (likely(!async) &&
1986 	    (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ||
1987 	     test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
1988 		dm_submit_bio_remap(io->base_bio, clone);
1989 		return;
1990 	}
1991 
1992 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1993 	if (RB_EMPTY_ROOT(&cc->write_tree))
1994 		wake_up_process(cc->write_thread);
1995 	rbp = &cc->write_tree.rb_node;
1996 	parent = NULL;
1997 	sector = io->sector;
1998 	while (*rbp) {
1999 		parent = *rbp;
2000 		if (sector < crypt_io_from_node(parent)->sector)
2001 			rbp = &(*rbp)->rb_left;
2002 		else
2003 			rbp = &(*rbp)->rb_right;
2004 	}
2005 	rb_link_node(&io->rb_node, parent, rbp);
2006 	rb_insert_color(&io->rb_node, &cc->write_tree);
2007 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2008 }
2009 
2010 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2011 				       struct convert_context *ctx)
2012 
2013 {
2014 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2015 		return false;
2016 
2017 	/*
2018 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2019 	 * constraints so they do not need to be issued inline by
2020 	 * kcryptd_crypt_write_convert().
2021 	 */
2022 	switch (bio_op(ctx->bio_in)) {
2023 	case REQ_OP_WRITE:
2024 	case REQ_OP_WRITE_ZEROES:
2025 		return true;
2026 	default:
2027 		return false;
2028 	}
2029 }
2030 
2031 static void kcryptd_crypt_write_continue(struct work_struct *work)
2032 {
2033 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2034 	struct crypt_config *cc = io->cc;
2035 	struct convert_context *ctx = &io->ctx;
2036 	int crypt_finished;
2037 	blk_status_t r;
2038 
2039 	wait_for_completion(&ctx->restart);
2040 	reinit_completion(&ctx->restart);
2041 
2042 	r = crypt_convert(cc, &io->ctx, false, false);
2043 	if (r)
2044 		io->error = r;
2045 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2046 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2047 		/* Wait for completion signaled by kcryptd_async_done() */
2048 		wait_for_completion(&ctx->restart);
2049 		crypt_finished = 1;
2050 	}
2051 
2052 	/* Encryption was already finished, submit io now */
2053 	if (crypt_finished)
2054 		kcryptd_crypt_write_io_submit(io, 0);
2055 
2056 	crypt_dec_pending(io);
2057 }
2058 
2059 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2060 {
2061 	struct crypt_config *cc = io->cc;
2062 	struct convert_context *ctx = &io->ctx;
2063 	struct bio *clone;
2064 	int crypt_finished;
2065 	blk_status_t r;
2066 
2067 	/*
2068 	 * Prevent io from disappearing until this function completes.
2069 	 */
2070 	crypt_inc_pending(io);
2071 	crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector);
2072 
2073 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2074 	if (unlikely(!clone)) {
2075 		io->error = BLK_STS_IOERR;
2076 		goto dec;
2077 	}
2078 
2079 	io->ctx.bio_out = clone;
2080 	io->ctx.iter_out = clone->bi_iter;
2081 
2082 	if (crypt_integrity_aead(cc)) {
2083 		bio_copy_data(clone, io->base_bio);
2084 		io->ctx.bio_in = clone;
2085 		io->ctx.iter_in = clone->bi_iter;
2086 	}
2087 
2088 	crypt_inc_pending(io);
2089 	r = crypt_convert(cc, ctx,
2090 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2091 	/*
2092 	 * Crypto API backlogged the request, because its queue was full
2093 	 * and we're in softirq context, so continue from a workqueue
2094 	 * (TODO: is it actually possible to be in softirq in the write path?)
2095 	 */
2096 	if (r == BLK_STS_DEV_RESOURCE) {
2097 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2098 		queue_work(cc->crypt_queue, &io->work);
2099 		return;
2100 	}
2101 	if (r)
2102 		io->error = r;
2103 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2104 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2105 		/* Wait for completion signaled by kcryptd_async_done() */
2106 		wait_for_completion(&ctx->restart);
2107 		crypt_finished = 1;
2108 	}
2109 
2110 	/* Encryption was already finished, submit io now */
2111 	if (crypt_finished)
2112 		kcryptd_crypt_write_io_submit(io, 0);
2113 
2114 dec:
2115 	crypt_dec_pending(io);
2116 }
2117 
2118 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2119 {
2120 	if (io->ctx.aead_recheck) {
2121 		if (!io->error) {
2122 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2123 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2124 		}
2125 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2126 		bio_put(io->ctx.bio_in);
2127 	}
2128 	crypt_dec_pending(io);
2129 }
2130 
2131 static void kcryptd_crypt_read_continue(struct work_struct *work)
2132 {
2133 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2134 	struct crypt_config *cc = io->cc;
2135 	blk_status_t r;
2136 
2137 	wait_for_completion(&io->ctx.restart);
2138 	reinit_completion(&io->ctx.restart);
2139 
2140 	r = crypt_convert(cc, &io->ctx, false, false);
2141 	if (r)
2142 		io->error = r;
2143 
2144 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2145 		kcryptd_crypt_read_done(io);
2146 
2147 	crypt_dec_pending(io);
2148 }
2149 
2150 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2151 {
2152 	struct crypt_config *cc = io->cc;
2153 	blk_status_t r;
2154 
2155 	crypt_inc_pending(io);
2156 
2157 	if (io->ctx.aead_recheck) {
2158 		r = crypt_convert(cc, &io->ctx,
2159 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2160 	} else {
2161 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2162 				   io->sector);
2163 
2164 		r = crypt_convert(cc, &io->ctx,
2165 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2166 	}
2167 	/*
2168 	 * Crypto API backlogged the request, because its queue was full
2169 	 * and we're in softirq context, so continue from a workqueue
2170 	 */
2171 	if (r == BLK_STS_DEV_RESOURCE) {
2172 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2173 		queue_work(cc->crypt_queue, &io->work);
2174 		return;
2175 	}
2176 	if (r)
2177 		io->error = r;
2178 
2179 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2180 		kcryptd_crypt_read_done(io);
2181 
2182 	crypt_dec_pending(io);
2183 }
2184 
2185 static void kcryptd_async_done(void *data, int error)
2186 {
2187 	struct dm_crypt_request *dmreq = data;
2188 	struct convert_context *ctx = dmreq->ctx;
2189 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2190 	struct crypt_config *cc = io->cc;
2191 
2192 	/*
2193 	 * A request from crypto driver backlog is going to be processed now,
2194 	 * finish the completion and continue in crypt_convert().
2195 	 * (Callback will be called for the second time for this request.)
2196 	 */
2197 	if (error == -EINPROGRESS) {
2198 		complete(&ctx->restart);
2199 		return;
2200 	}
2201 
2202 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2203 		cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2204 
2205 	if (error == -EBADMSG) {
2206 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2207 
2208 		ctx->aead_failed = true;
2209 		if (ctx->aead_recheck) {
2210 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2211 				    ctx->bio_in->bi_bdev, s);
2212 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2213 					 ctx->bio_in, s, 0);
2214 		}
2215 		io->error = BLK_STS_PROTECTION;
2216 	} else if (error < 0)
2217 		io->error = BLK_STS_IOERR;
2218 
2219 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2220 
2221 	if (!atomic_dec_and_test(&ctx->cc_pending))
2222 		return;
2223 
2224 	/*
2225 	 * The request is fully completed: for inline writes, let
2226 	 * kcryptd_crypt_write_convert() do the IO submission.
2227 	 */
2228 	if (bio_data_dir(io->base_bio) == READ) {
2229 		kcryptd_crypt_read_done(io);
2230 		return;
2231 	}
2232 
2233 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2234 		complete(&ctx->restart);
2235 		return;
2236 	}
2237 
2238 	kcryptd_crypt_write_io_submit(io, 1);
2239 }
2240 
2241 static void kcryptd_crypt(struct work_struct *work)
2242 {
2243 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2244 
2245 	if (bio_data_dir(io->base_bio) == READ)
2246 		kcryptd_crypt_read_convert(io);
2247 	else
2248 		kcryptd_crypt_write_convert(io);
2249 }
2250 
2251 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2252 {
2253 	struct crypt_config *cc = io->cc;
2254 
2255 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2256 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2257 		/*
2258 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2259 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2260 		 * it is being executed with irqs disabled.
2261 		 */
2262 		if (in_hardirq() || irqs_disabled()) {
2263 			INIT_WORK(&io->work, kcryptd_crypt);
2264 			queue_work(system_bh_wq, &io->work);
2265 			return;
2266 		} else {
2267 			kcryptd_crypt(&io->work);
2268 			return;
2269 		}
2270 	}
2271 
2272 	INIT_WORK(&io->work, kcryptd_crypt);
2273 	queue_work(cc->crypt_queue, &io->work);
2274 }
2275 
2276 static void crypt_free_tfms_aead(struct crypt_config *cc)
2277 {
2278 	if (!cc->cipher_tfm.tfms_aead)
2279 		return;
2280 
2281 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2282 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2283 		cc->cipher_tfm.tfms_aead[0] = NULL;
2284 	}
2285 
2286 	kfree(cc->cipher_tfm.tfms_aead);
2287 	cc->cipher_tfm.tfms_aead = NULL;
2288 }
2289 
2290 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2291 {
2292 	unsigned int i;
2293 
2294 	if (!cc->cipher_tfm.tfms)
2295 		return;
2296 
2297 	for (i = 0; i < cc->tfms_count; i++)
2298 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2299 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2300 			cc->cipher_tfm.tfms[i] = NULL;
2301 		}
2302 
2303 	kfree(cc->cipher_tfm.tfms);
2304 	cc->cipher_tfm.tfms = NULL;
2305 }
2306 
2307 static void crypt_free_tfms(struct crypt_config *cc)
2308 {
2309 	if (crypt_integrity_aead(cc))
2310 		crypt_free_tfms_aead(cc);
2311 	else
2312 		crypt_free_tfms_skcipher(cc);
2313 }
2314 
2315 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2316 {
2317 	unsigned int i;
2318 	int err;
2319 
2320 	cc->cipher_tfm.tfms = kzalloc_objs(struct crypto_skcipher *,
2321 					   cc->tfms_count);
2322 	if (!cc->cipher_tfm.tfms)
2323 		return -ENOMEM;
2324 
2325 	for (i = 0; i < cc->tfms_count; i++) {
2326 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2327 						CRYPTO_ALG_ALLOCATES_MEMORY);
2328 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2329 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2330 			crypt_free_tfms(cc);
2331 			return err;
2332 		}
2333 	}
2334 
2335 	/*
2336 	 * dm-crypt performance can vary greatly depending on which crypto
2337 	 * algorithm implementation is used.  Help people debug performance
2338 	 * problems by logging the ->cra_driver_name.
2339 	 */
2340 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2341 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2342 	return 0;
2343 }
2344 
2345 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2346 {
2347 	int err;
2348 
2349 	cc->cipher_tfm.tfms = kmalloc_obj(struct crypto_skcipher *);
2350 	if (!cc->cipher_tfm.tfms)
2351 		return -ENOMEM;
2352 
2353 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2354 						CRYPTO_ALG_ALLOCATES_MEMORY);
2355 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2356 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2357 		crypt_free_tfms(cc);
2358 		return err;
2359 	}
2360 
2361 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2362 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2363 	return 0;
2364 }
2365 
2366 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2367 {
2368 	if (crypt_integrity_aead(cc))
2369 		return crypt_alloc_tfms_aead(cc, ciphermode);
2370 	else
2371 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2372 }
2373 
2374 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2375 {
2376 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2377 }
2378 
2379 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2380 {
2381 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2382 }
2383 
2384 /*
2385  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2386  * the key must be for some reason in special format.
2387  * This function converts cc->key to this special format.
2388  */
2389 static void crypt_copy_authenckey(char *p, const void *key,
2390 				  unsigned int enckeylen, unsigned int authkeylen)
2391 {
2392 	struct crypto_authenc_key_param *param;
2393 	struct rtattr *rta;
2394 
2395 	rta = (struct rtattr *)p;
2396 	param = RTA_DATA(rta);
2397 	param->enckeylen = cpu_to_be32(enckeylen);
2398 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2399 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2400 	p += RTA_SPACE(sizeof(*param));
2401 	memcpy(p, key + enckeylen, authkeylen);
2402 	p += authkeylen;
2403 	memcpy(p, key, enckeylen);
2404 }
2405 
2406 static int crypt_setkey(struct crypt_config *cc)
2407 {
2408 	unsigned int subkey_size;
2409 	int err = 0, i, r;
2410 
2411 	/* Ignore extra keys (which are used for IV etc) */
2412 	subkey_size = crypt_subkey_size(cc);
2413 
2414 	if (crypt_integrity_hmac(cc)) {
2415 		if (subkey_size < cc->key_mac_size)
2416 			return -EINVAL;
2417 
2418 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2419 				      subkey_size - cc->key_mac_size,
2420 				      cc->key_mac_size);
2421 	}
2422 
2423 	for (i = 0; i < cc->tfms_count; i++) {
2424 		if (crypt_integrity_hmac(cc))
2425 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2426 				cc->authenc_key, crypt_authenckey_size(cc));
2427 		else if (crypt_integrity_aead(cc))
2428 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2429 					       cc->key + (i * subkey_size),
2430 					       subkey_size);
2431 		else
2432 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2433 						   cc->key + (i * subkey_size),
2434 						   subkey_size);
2435 		if (r)
2436 			err = r;
2437 	}
2438 
2439 	if (crypt_integrity_hmac(cc))
2440 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2441 
2442 	return err;
2443 }
2444 
2445 #ifdef CONFIG_KEYS
2446 
2447 static bool contains_whitespace(const char *str)
2448 {
2449 	while (*str)
2450 		if (isspace(*str++))
2451 			return true;
2452 	return false;
2453 }
2454 
2455 static int set_key_user(struct crypt_config *cc, struct key *key)
2456 {
2457 	const struct user_key_payload *ukp;
2458 
2459 	ukp = user_key_payload_locked(key);
2460 	if (!ukp)
2461 		return -EKEYREVOKED;
2462 
2463 	if (cc->key_size != ukp->datalen)
2464 		return -EINVAL;
2465 
2466 	memcpy(cc->key, ukp->data, cc->key_size);
2467 
2468 	return 0;
2469 }
2470 
2471 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2472 {
2473 	const struct encrypted_key_payload *ekp;
2474 
2475 	ekp = key->payload.data[0];
2476 	if (!ekp)
2477 		return -EKEYREVOKED;
2478 
2479 	if (cc->key_size != ekp->decrypted_datalen)
2480 		return -EINVAL;
2481 
2482 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2483 
2484 	return 0;
2485 }
2486 
2487 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2488 {
2489 	const struct trusted_key_payload *tkp;
2490 
2491 	tkp = key->payload.data[0];
2492 	if (!tkp)
2493 		return -EKEYREVOKED;
2494 
2495 	if (cc->key_size != tkp->key_len)
2496 		return -EINVAL;
2497 
2498 	memcpy(cc->key, tkp->key, cc->key_size);
2499 
2500 	return 0;
2501 }
2502 
2503 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2504 {
2505 	char *new_key_string, *key_desc;
2506 	int ret;
2507 	struct key_type *type;
2508 	struct key *key;
2509 	int (*set_key)(struct crypt_config *cc, struct key *key);
2510 
2511 	/*
2512 	 * Reject key_string with whitespace. dm core currently lacks code for
2513 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2514 	 */
2515 	if (contains_whitespace(key_string)) {
2516 		DMERR("whitespace chars not allowed in key string");
2517 		return -EINVAL;
2518 	}
2519 
2520 	/* look for next ':' separating key_type from key_description */
2521 	key_desc = strchr(key_string, ':');
2522 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2523 		return -EINVAL;
2524 
2525 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2526 		type = &key_type_logon;
2527 		set_key = set_key_user;
2528 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2529 		type = &key_type_user;
2530 		set_key = set_key_user;
2531 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2532 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2533 		type = &key_type_encrypted;
2534 		set_key = set_key_encrypted;
2535 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2536 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2537 		type = &key_type_trusted;
2538 		set_key = set_key_trusted;
2539 	} else {
2540 		return -EINVAL;
2541 	}
2542 
2543 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2544 	if (!new_key_string)
2545 		return -ENOMEM;
2546 
2547 	key = request_key(type, key_desc + 1, NULL);
2548 	if (IS_ERR(key)) {
2549 		ret = PTR_ERR(key);
2550 		goto free_new_key_string;
2551 	}
2552 
2553 	down_read(&key->sem);
2554 	ret = set_key(cc, key);
2555 	up_read(&key->sem);
2556 	key_put(key);
2557 	if (ret < 0)
2558 		goto free_new_key_string;
2559 
2560 	/* clear the flag since following operations may invalidate previously valid key */
2561 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2562 
2563 	ret = crypt_setkey(cc);
2564 	if (ret)
2565 		goto free_new_key_string;
2566 
2567 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2568 	kfree_sensitive(cc->key_string);
2569 	cc->key_string = new_key_string;
2570 	return 0;
2571 
2572 free_new_key_string:
2573 	kfree_sensitive(new_key_string);
2574 	return ret;
2575 }
2576 
2577 static int get_key_size(char **key_string)
2578 {
2579 	char *colon, dummy;
2580 	int ret;
2581 
2582 	if (*key_string[0] != ':')
2583 		return strlen(*key_string) >> 1;
2584 
2585 	/* look for next ':' in key string */
2586 	colon = strpbrk(*key_string + 1, ":");
2587 	if (!colon)
2588 		return -EINVAL;
2589 
2590 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2591 		return -EINVAL;
2592 
2593 	*key_string = colon;
2594 
2595 	/* remaining key string should be :<logon|user>:<key_desc> */
2596 
2597 	return ret;
2598 }
2599 
2600 #else
2601 
2602 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2603 {
2604 	return -EINVAL;
2605 }
2606 
2607 static int get_key_size(char **key_string)
2608 {
2609 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2610 }
2611 
2612 #endif /* CONFIG_KEYS */
2613 
2614 static int crypt_set_key(struct crypt_config *cc, char *key)
2615 {
2616 	int r = -EINVAL;
2617 	size_t key_string_len = strlen(key);
2618 
2619 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2620 	if (!cc->key_size && strcmp(key, "-"))
2621 		goto out;
2622 
2623 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2624 	if (key[0] == ':') {
2625 		r = crypt_set_keyring_key(cc, key + 1);
2626 		goto out;
2627 	}
2628 
2629 	/* clear the flag since following operations may invalidate previously valid key */
2630 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2631 
2632 	/* wipe references to any kernel keyring key */
2633 	kfree_sensitive(cc->key_string);
2634 	cc->key_string = NULL;
2635 
2636 	if (cc->key_size && key_string_len != cc->key_size * 2)
2637 		goto out;
2638 
2639 	/* Decode key from its hex representation. */
2640 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2641 		goto out;
2642 
2643 	r = crypt_setkey(cc);
2644 	if (!r)
2645 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2646 
2647 out:
2648 	/* Hex key string not needed after here, so wipe it. */
2649 	memzero_explicit(key, key_string_len);
2650 
2651 	return r;
2652 }
2653 
2654 static int crypt_wipe_key(struct crypt_config *cc)
2655 {
2656 	int r;
2657 
2658 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2659 	get_random_bytes(&cc->key, cc->key_size);
2660 
2661 	/* Wipe IV private keys */
2662 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe)
2663 		cc->iv_gen_ops->wipe(cc);
2664 
2665 	kfree_sensitive(cc->key_string);
2666 	cc->key_string = NULL;
2667 	r = crypt_setkey(cc);
2668 	memzero_explicit(cc->key, cc->key_size);
2669 
2670 	return r;
2671 }
2672 
2673 static void crypt_calculate_pages_per_client(void)
2674 {
2675 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2676 
2677 	if (!dm_crypt_clients_n)
2678 		return;
2679 
2680 	pages /= dm_crypt_clients_n;
2681 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2682 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2683 	dm_crypt_pages_per_client = pages;
2684 }
2685 
2686 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2687 {
2688 	struct crypt_config *cc = pool_data;
2689 	struct page *page;
2690 
2691 	/*
2692 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2693 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2694 	 * but avoids potential spinlock contention of an exact result.
2695 	 */
2696 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2697 	    likely(gfp_mask & __GFP_NORETRY))
2698 		return NULL;
2699 
2700 	page = alloc_page(gfp_mask);
2701 	if (likely(page != NULL))
2702 		percpu_counter_add(&cc->n_allocated_pages, 1);
2703 
2704 	return page;
2705 }
2706 
2707 static void crypt_page_free(void *page, void *pool_data)
2708 {
2709 	struct crypt_config *cc = pool_data;
2710 
2711 	__free_page(page);
2712 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2713 }
2714 
2715 static void crypt_dtr(struct dm_target *ti)
2716 {
2717 	struct crypt_config *cc = ti->private;
2718 
2719 	ti->private = NULL;
2720 
2721 	if (!cc)
2722 		return;
2723 
2724 	if (cc->write_thread)
2725 		kthread_stop(cc->write_thread);
2726 
2727 	if (cc->io_queue)
2728 		destroy_workqueue(cc->io_queue);
2729 	if (cc->crypt_queue)
2730 		destroy_workqueue(cc->crypt_queue);
2731 
2732 	if (cc->workqueue_id)
2733 		ida_free(&workqueue_ida, cc->workqueue_id);
2734 
2735 	crypt_free_tfms(cc);
2736 
2737 	bioset_exit(&cc->bs);
2738 
2739 	mempool_exit(&cc->page_pool);
2740 	mempool_exit(&cc->req_pool);
2741 	mempool_exit(&cc->tag_pool);
2742 
2743 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2744 	percpu_counter_destroy(&cc->n_allocated_pages);
2745 
2746 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2747 		cc->iv_gen_ops->dtr(cc);
2748 
2749 	if (cc->dev)
2750 		dm_put_device(ti, cc->dev);
2751 
2752 	kfree_sensitive(cc->cipher_string);
2753 	kfree_sensitive(cc->key_string);
2754 	kfree_sensitive(cc->cipher_auth);
2755 	kfree_sensitive(cc->authenc_key);
2756 
2757 	mutex_destroy(&cc->bio_alloc_lock);
2758 
2759 	/* Must zero key material before freeing */
2760 	kfree_sensitive(cc);
2761 
2762 	spin_lock(&dm_crypt_clients_lock);
2763 	WARN_ON(!dm_crypt_clients_n);
2764 	dm_crypt_clients_n--;
2765 	crypt_calculate_pages_per_client();
2766 	spin_unlock(&dm_crypt_clients_lock);
2767 
2768 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2769 }
2770 
2771 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2772 {
2773 	struct crypt_config *cc = ti->private;
2774 
2775 	if (crypt_integrity_aead(cc))
2776 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2777 	else
2778 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2779 
2780 	if (cc->iv_size)
2781 		/* at least a 64 bit sector number should fit in our buffer */
2782 		cc->iv_size = max(cc->iv_size,
2783 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2784 	else if (ivmode) {
2785 		DMWARN("Selected cipher does not support IVs");
2786 		ivmode = NULL;
2787 	}
2788 
2789 	/* Choose ivmode, see comments at iv code. */
2790 	if (ivmode == NULL)
2791 		cc->iv_gen_ops = NULL;
2792 	else if (strcmp(ivmode, "plain") == 0)
2793 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2794 	else if (strcmp(ivmode, "plain64") == 0)
2795 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2796 	else if (strcmp(ivmode, "plain64be") == 0)
2797 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2798 	else if (strcmp(ivmode, "essiv") == 0)
2799 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2800 	else if (strcmp(ivmode, "benbi") == 0)
2801 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2802 	else if (strcmp(ivmode, "null") == 0)
2803 		cc->iv_gen_ops = &crypt_iv_null_ops;
2804 	else if (strcmp(ivmode, "eboiv") == 0)
2805 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2806 	else if (strcmp(ivmode, "elephant") == 0) {
2807 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2808 		cc->key_parts = 2;
2809 		cc->key_extra_size = cc->key_size / 2;
2810 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2811 			return -EINVAL;
2812 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2813 	} else if (strcmp(ivmode, "lmk") == 0) {
2814 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2815 		/*
2816 		 * Version 2 and 3 is recognised according
2817 		 * to length of provided multi-key string.
2818 		 * If present (version 3), last key is used as IV seed.
2819 		 * All keys (including IV seed) are always the same size.
2820 		 */
2821 		if (cc->key_size % cc->key_parts) {
2822 			cc->key_parts++;
2823 			cc->key_extra_size = cc->key_size / cc->key_parts;
2824 		}
2825 	} else if (strcmp(ivmode, "tcw") == 0) {
2826 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2827 		cc->key_parts += 2; /* IV + whitening */
2828 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2829 	} else if (strcmp(ivmode, "random") == 0) {
2830 		cc->iv_gen_ops = &crypt_iv_random_ops;
2831 		/* Need storage space in integrity fields. */
2832 		cc->integrity_iv_size = cc->iv_size;
2833 	} else {
2834 		ti->error = "Invalid IV mode";
2835 		return -EINVAL;
2836 	}
2837 
2838 	return 0;
2839 }
2840 
2841 /*
2842  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2843  * The HMAC is needed to calculate tag size (HMAC digest size).
2844  * This should be probably done by crypto-api calls (once available...)
2845  */
2846 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2847 {
2848 	char *start, *end, *mac_alg = NULL;
2849 	struct crypto_ahash *mac;
2850 
2851 	if (!strstarts(cipher_api, "authenc(")) {
2852 		if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2853 			return -EINVAL;
2854 		return 0;
2855 	}
2856 
2857 	start = strchr(cipher_api, '(');
2858 	end = strchr(cipher_api, ',');
2859 	if (!start || !end || ++start > end)
2860 		return -EINVAL;
2861 
2862 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2863 	if (!mac_alg)
2864 		return -ENOMEM;
2865 
2866 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2867 	kfree(mac_alg);
2868 
2869 	if (IS_ERR(mac))
2870 		return PTR_ERR(mac);
2871 
2872 	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2873 		cc->key_mac_size = crypto_ahash_digestsize(mac);
2874 	crypto_free_ahash(mac);
2875 
2876 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2877 	if (!cc->authenc_key)
2878 		return -ENOMEM;
2879 
2880 	return 0;
2881 }
2882 
2883 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2884 				char **ivmode, char **ivopts)
2885 {
2886 	struct crypt_config *cc = ti->private;
2887 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2888 	int ret = -EINVAL;
2889 
2890 	cc->tfms_count = 1;
2891 
2892 	/*
2893 	 * New format (capi: prefix)
2894 	 * capi:cipher_api_spec-iv:ivopts
2895 	 */
2896 	tmp = &cipher_in[strlen("capi:")];
2897 
2898 	/* Separate IV options if present, it can contain another '-' in hash name */
2899 	*ivopts = strrchr(tmp, ':');
2900 	if (*ivopts) {
2901 		**ivopts = '\0';
2902 		(*ivopts)++;
2903 	}
2904 	/* Parse IV mode */
2905 	*ivmode = strrchr(tmp, '-');
2906 	if (*ivmode) {
2907 		**ivmode = '\0';
2908 		(*ivmode)++;
2909 	}
2910 	/* The rest is crypto API spec */
2911 	cipher_api = tmp;
2912 
2913 	/* Alloc AEAD, can be used only in new format. */
2914 	if (crypt_integrity_aead(cc)) {
2915 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2916 		if (ret < 0) {
2917 			ti->error = "Invalid AEAD cipher spec";
2918 			return ret;
2919 		}
2920 	}
2921 
2922 	if (*ivmode && !strcmp(*ivmode, "lmk")) {
2923 		if (crypt_integrity_aead(cc)) {
2924 			ti->error = "AEAD transforms not supported for LMK";
2925 			return -EINVAL;
2926 		}
2927 		cc->tfms_count = 64;
2928 	}
2929 
2930 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2931 		if (!*ivopts) {
2932 			ti->error = "Digest algorithm missing for ESSIV mode";
2933 			return -EINVAL;
2934 		}
2935 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2936 			       cipher_api, *ivopts);
2937 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2938 			ti->error = "Cannot allocate cipher string";
2939 			return -ENOMEM;
2940 		}
2941 		cipher_api = buf;
2942 	}
2943 
2944 	cc->key_parts = cc->tfms_count;
2945 
2946 	/* Allocate cipher */
2947 	ret = crypt_alloc_tfms(cc, cipher_api);
2948 	if (ret < 0) {
2949 		ti->error = "Error allocating crypto tfm";
2950 		return ret;
2951 	}
2952 
2953 	if (crypt_integrity_aead(cc))
2954 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2955 	else
2956 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2957 
2958 	return 0;
2959 }
2960 
2961 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2962 				char **ivmode, char **ivopts)
2963 {
2964 	struct crypt_config *cc = ti->private;
2965 	char *tmp, *cipher, *chainmode, *keycount;
2966 	char *cipher_api = NULL;
2967 	int ret = -EINVAL;
2968 	char dummy;
2969 
2970 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2971 		ti->error = "Bad cipher specification";
2972 		return -EINVAL;
2973 	}
2974 
2975 	/*
2976 	 * Legacy dm-crypt cipher specification
2977 	 * cipher[:keycount]-mode-iv:ivopts
2978 	 */
2979 	tmp = cipher_in;
2980 	keycount = strsep(&tmp, "-");
2981 	cipher = strsep(&keycount, ":");
2982 
2983 	if (!keycount)
2984 		cc->tfms_count = 1;
2985 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2986 		 !is_power_of_2(cc->tfms_count)) {
2987 		ti->error = "Bad cipher key count specification";
2988 		return -EINVAL;
2989 	}
2990 	cc->key_parts = cc->tfms_count;
2991 
2992 	chainmode = strsep(&tmp, "-");
2993 	*ivmode = strsep(&tmp, ":");
2994 	*ivopts = tmp;
2995 
2996 	/*
2997 	 * For compatibility with the original dm-crypt mapping format, if
2998 	 * only the cipher name is supplied, use cbc-plain.
2999 	 */
3000 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3001 		chainmode = "cbc";
3002 		*ivmode = "plain";
3003 	}
3004 
3005 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3006 		ti->error = "IV mechanism required";
3007 		return -EINVAL;
3008 	}
3009 
3010 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3011 	if (!cipher_api)
3012 		goto bad_mem;
3013 
3014 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3015 		if (!*ivopts) {
3016 			ti->error = "Digest algorithm missing for ESSIV mode";
3017 			kfree(cipher_api);
3018 			return -EINVAL;
3019 		}
3020 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3021 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3022 	} else {
3023 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3024 			       "%s(%s)", chainmode, cipher);
3025 	}
3026 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3027 		kfree(cipher_api);
3028 		goto bad_mem;
3029 	}
3030 
3031 	/* Allocate cipher */
3032 	ret = crypt_alloc_tfms(cc, cipher_api);
3033 	if (ret < 0) {
3034 		ti->error = "Error allocating crypto tfm";
3035 		kfree(cipher_api);
3036 		return ret;
3037 	}
3038 	kfree(cipher_api);
3039 
3040 	return 0;
3041 bad_mem:
3042 	ti->error = "Cannot allocate cipher strings";
3043 	return -ENOMEM;
3044 }
3045 
3046 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3047 {
3048 	struct crypt_config *cc = ti->private;
3049 	char *ivmode = NULL, *ivopts = NULL;
3050 	int ret;
3051 
3052 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3053 	if (!cc->cipher_string) {
3054 		ti->error = "Cannot allocate cipher strings";
3055 		return -ENOMEM;
3056 	}
3057 
3058 	if (strstarts(cipher_in, "capi:"))
3059 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3060 	else
3061 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3062 	if (ret)
3063 		return ret;
3064 
3065 	/* Initialize IV */
3066 	ret = crypt_ctr_ivmode(ti, ivmode);
3067 	if (ret < 0)
3068 		return ret;
3069 
3070 	/* Initialize and set key */
3071 	ret = crypt_set_key(cc, key);
3072 	if (ret < 0) {
3073 		ti->error = "Error decoding and setting key";
3074 		return ret;
3075 	}
3076 
3077 	/* Allocate IV */
3078 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3079 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3080 		if (ret < 0) {
3081 			ti->error = "Error creating IV";
3082 			return ret;
3083 		}
3084 	}
3085 
3086 	/* Initialize IV (set keys for ESSIV etc) */
3087 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3088 		ret = cc->iv_gen_ops->init(cc);
3089 		if (ret < 0) {
3090 			ti->error = "Error initialising IV";
3091 			return ret;
3092 		}
3093 	}
3094 
3095 	/* wipe the kernel key payload copy */
3096 	if (cc->key_string)
3097 		memzero_explicit(cc->key, cc->key_size);
3098 
3099 	return ret;
3100 }
3101 
3102 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3103 {
3104 	struct crypt_config *cc = ti->private;
3105 	struct dm_arg_set as;
3106 	static const struct dm_arg _args[] = {
3107 		{0, 9, "Invalid number of feature args"},
3108 	};
3109 	unsigned int opt_params, val;
3110 	const char *opt_string, *sval;
3111 	char dummy;
3112 	int ret;
3113 
3114 	/* Optional parameters */
3115 	as.argc = argc;
3116 	as.argv = argv;
3117 
3118 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3119 	if (ret)
3120 		return ret;
3121 
3122 	while (opt_params--) {
3123 		opt_string = dm_shift_arg(&as);
3124 		if (!opt_string) {
3125 			ti->error = "Not enough feature arguments";
3126 			return -EINVAL;
3127 		}
3128 
3129 		if (!strcasecmp(opt_string, "allow_discards"))
3130 			ti->num_discard_bios = 1;
3131 
3132 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3133 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3134 		else if (!strcasecmp(opt_string, "high_priority"))
3135 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3136 
3137 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3138 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3139 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3140 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3141 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3142 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3143 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3144 			if (val == 0 || val > MAX_TAG_SIZE) {
3145 				ti->error = "Invalid integrity arguments";
3146 				return -EINVAL;
3147 			}
3148 			cc->used_tag_size = val;
3149 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3150 			if (!strcasecmp(sval, "aead")) {
3151 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3152 			} else if (strcasecmp(sval, "none")) {
3153 				ti->error = "Unknown integrity profile";
3154 				return -EINVAL;
3155 			}
3156 
3157 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3158 			if (!cc->cipher_auth)
3159 				return -ENOMEM;
3160 		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3161 			if (!val) {
3162 				ti->error = "Invalid integrity_key_size argument";
3163 				return -EINVAL;
3164 			}
3165 			cc->key_mac_size = val;
3166 			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3167 		} else if (sscanf(opt_string, "sector_size:%u%c", &cc->sector_size, &dummy) == 1) {
3168 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3169 			    cc->sector_size > DM_CRYPT_MAX_SECTOR_SIZE ||
3170 			    (cc->sector_size & (cc->sector_size - 1))) {
3171 				ti->error = "Invalid feature value for sector_size";
3172 				return -EINVAL;
3173 			}
3174 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3175 				ti->error = "Device size is not multiple of sector_size feature";
3176 				return -EINVAL;
3177 			}
3178 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3179 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3180 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3181 		else {
3182 			ti->error = "Invalid feature arguments";
3183 			return -EINVAL;
3184 		}
3185 	}
3186 
3187 	return 0;
3188 }
3189 
3190 #ifdef CONFIG_BLK_DEV_ZONED
3191 static int crypt_report_zones(struct dm_target *ti,
3192 		struct dm_report_zones_args *args, unsigned int nr_zones)
3193 {
3194 	struct crypt_config *cc = ti->private;
3195 
3196 	return dm_report_zones(cc->dev->bdev, cc->start,
3197 			cc->start + dm_target_offset(ti, args->next_sector),
3198 			args, nr_zones);
3199 }
3200 #else
3201 #define crypt_report_zones NULL
3202 #endif
3203 
3204 /*
3205  * Construct an encryption mapping:
3206  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3207  */
3208 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3209 {
3210 	struct crypt_config *cc;
3211 	const char *devname = dm_table_device_name(ti->table);
3212 	int key_size, wq_id;
3213 	unsigned int align_mask;
3214 	unsigned int common_wq_flags;
3215 	unsigned long long tmpll;
3216 	int ret;
3217 	size_t iv_size_padding, additional_req_size;
3218 	char dummy;
3219 
3220 	if (argc < 5) {
3221 		ti->error = "Not enough arguments";
3222 		return -EINVAL;
3223 	}
3224 
3225 	key_size = get_key_size(&argv[1]);
3226 	if (key_size < 0) {
3227 		ti->error = "Cannot parse key size";
3228 		return -EINVAL;
3229 	}
3230 
3231 	cc = kzalloc_flex(*cc, key, key_size);
3232 	if (!cc) {
3233 		ti->error = "Cannot allocate encryption context";
3234 		return -ENOMEM;
3235 	}
3236 
3237 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3238 	if (ret < 0) {
3239 		ti->error = "Cannot initialize page counter";
3240 		kfree_sensitive(cc);
3241 		return ret;
3242 	}
3243 
3244 	cc->key_size = key_size;
3245 	cc->sector_size = (1 << SECTOR_SHIFT);
3246 	cc->sector_shift = 0;
3247 
3248 	ti->private = cc;
3249 
3250 	spin_lock(&dm_crypt_clients_lock);
3251 	dm_crypt_clients_n++;
3252 	crypt_calculate_pages_per_client();
3253 	spin_unlock(&dm_crypt_clients_lock);
3254 
3255 	/* Optional parameters need to be read before cipher constructor */
3256 	if (argc > 5) {
3257 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3258 		if (ret)
3259 			goto bad;
3260 	}
3261 
3262 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3263 	if (ret < 0)
3264 		goto bad;
3265 
3266 	if (crypt_integrity_aead(cc)) {
3267 		cc->dmreq_start = sizeof(struct aead_request);
3268 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3269 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3270 	} else {
3271 		cc->dmreq_start = sizeof(struct skcipher_request);
3272 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3273 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3274 	}
3275 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3276 
3277 	if (align_mask < CRYPTO_MINALIGN) {
3278 		/* Allocate the padding exactly */
3279 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3280 				& align_mask;
3281 	} else {
3282 		/*
3283 		 * If the cipher requires greater alignment than kmalloc
3284 		 * alignment, we don't know the exact position of the
3285 		 * initialization vector. We must assume worst case.
3286 		 */
3287 		iv_size_padding = align_mask;
3288 	}
3289 
3290 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3291 	additional_req_size = sizeof(struct dm_crypt_request) +
3292 		iv_size_padding + cc->iv_size +
3293 		cc->iv_size +
3294 		sizeof(uint64_t) +
3295 		sizeof(unsigned int);
3296 
3297 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3298 	if (ret) {
3299 		ti->error = "Cannot allocate crypt request mempool";
3300 		goto bad;
3301 	}
3302 
3303 	cc->per_bio_data_size = ti->per_io_data_size =
3304 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3305 		      ARCH_DMA_MINALIGN);
3306 
3307 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3308 	if (ret) {
3309 		ti->error = "Cannot allocate page mempool";
3310 		goto bad;
3311 	}
3312 
3313 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3314 	if (ret) {
3315 		ti->error = "Cannot allocate crypt bioset";
3316 		goto bad;
3317 	}
3318 
3319 	mutex_init(&cc->bio_alloc_lock);
3320 
3321 	ret = -EINVAL;
3322 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3323 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3324 		ti->error = "Invalid iv_offset sector";
3325 		goto bad;
3326 	}
3327 	cc->iv_offset = tmpll;
3328 
3329 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3330 	if (ret) {
3331 		ti->error = "Device lookup failed";
3332 		goto bad;
3333 	}
3334 
3335 	ret = -EINVAL;
3336 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3337 		ti->error = "Invalid device sector";
3338 		goto bad;
3339 	}
3340 	cc->start = tmpll;
3341 
3342 	if (bdev_is_zoned(cc->dev->bdev)) {
3343 		/*
3344 		 * For zoned block devices, we need to preserve the issuer write
3345 		 * ordering. To do so, disable write workqueues and force inline
3346 		 * encryption completion.
3347 		 */
3348 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3349 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3350 
3351 		/*
3352 		 * All zone append writes to a zone of a zoned block device will
3353 		 * have the same BIO sector, the start of the zone. When the
3354 		 * cypher IV mode uses sector values, all data targeting a
3355 		 * zone will be encrypted using the first sector numbers of the
3356 		 * zone. This will not result in write errors but will
3357 		 * cause most reads to fail as reads will use the sector values
3358 		 * for the actual data locations, resulting in IV mismatch.
3359 		 * To avoid this problem, ask DM core to emulate zone append
3360 		 * operations with regular writes.
3361 		 */
3362 		DMDEBUG("Zone append operations will be emulated");
3363 		ti->emulate_zone_append = true;
3364 	}
3365 
3366 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3367 		ret = crypt_integrity_ctr(cc, ti);
3368 		if (ret)
3369 			goto bad;
3370 
3371 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3372 		if (!cc->tag_pool_max_sectors)
3373 			cc->tag_pool_max_sectors = 1;
3374 
3375 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3376 			cc->tag_pool_max_sectors * cc->tuple_size);
3377 		if (ret) {
3378 			ti->error = "Cannot allocate integrity tags mempool";
3379 			goto bad;
3380 		}
3381 
3382 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3383 	}
3384 
3385 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3386 	if (wq_id < 0) {
3387 		ti->error = "Couldn't get workqueue id";
3388 		ret = wq_id;
3389 		goto bad;
3390 	}
3391 	cc->workqueue_id = wq_id;
3392 
3393 	ret = -ENOMEM;
3394 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3395 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3396 		common_wq_flags |= WQ_HIGHPRI;
3397 
3398 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d",
3399 				       common_wq_flags | WQ_PERCPU, 1,
3400 				       devname, wq_id);
3401 	if (!cc->io_queue) {
3402 		ti->error = "Couldn't create kcryptd io queue";
3403 		goto bad;
3404 	}
3405 
3406 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3407 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3408 						  common_wq_flags | WQ_CPU_INTENSIVE | WQ_PERCPU,
3409 						  1, devname, wq_id);
3410 	} else {
3411 		/*
3412 		 * While crypt_queue is certainly CPU intensive, the use of
3413 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3414 		 */
3415 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3416 						  common_wq_flags | WQ_UNBOUND,
3417 						  num_online_cpus(), devname, wq_id);
3418 	}
3419 	if (!cc->crypt_queue) {
3420 		ti->error = "Couldn't create kcryptd queue";
3421 		goto bad;
3422 	}
3423 
3424 	spin_lock_init(&cc->write_thread_lock);
3425 	cc->write_tree = RB_ROOT;
3426 
3427 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3428 	if (IS_ERR(cc->write_thread)) {
3429 		ret = PTR_ERR(cc->write_thread);
3430 		cc->write_thread = NULL;
3431 		ti->error = "Couldn't spawn write thread";
3432 		goto bad;
3433 	}
3434 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3435 		set_user_nice(cc->write_thread, MIN_NICE);
3436 
3437 	ti->num_flush_bios = 1;
3438 	ti->limit_swap_bios = true;
3439 	ti->accounts_remapped_io = true;
3440 
3441 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3442 	return 0;
3443 
3444 bad:
3445 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3446 	crypt_dtr(ti);
3447 	return ret;
3448 }
3449 
3450 static int crypt_map(struct dm_target *ti, struct bio *bio)
3451 {
3452 	struct dm_crypt_io *io;
3453 	struct crypt_config *cc = ti->private;
3454 	unsigned max_sectors;
3455 	bool no_split;
3456 
3457 	/*
3458 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3459 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3460 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3461 	 */
3462 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3463 	    bio_op(bio) == REQ_OP_DISCARD)) {
3464 		bio_set_dev(bio, cc->dev->bdev);
3465 		if (bio_sectors(bio))
3466 			bio->bi_iter.bi_sector = cc->start +
3467 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3468 		return DM_MAPIO_REMAPPED;
3469 	}
3470 
3471 	/*
3472 	 * Check if bio is too large, split as needed.
3473 	 *
3474 	 * For zoned devices, splitting write operations creates the
3475 	 * risk of deadlocking queue freeze operations with zone write
3476 	 * plugging BIO work when the reminder of a split BIO is
3477 	 * issued. So always allow the entire BIO to proceed.
3478 	 */
3479 	no_split = (ti->emulate_zone_append && op_is_write(bio_op(bio))) ||
3480 		   (bio->bi_opf & REQ_ATOMIC);
3481 	max_sectors = get_max_request_sectors(ti, bio, no_split);
3482 	if (unlikely(bio_sectors(bio) > max_sectors)) {
3483 		if (unlikely(no_split))
3484 			return DM_MAPIO_KILL;
3485 		dm_accept_partial_bio(bio, max_sectors);
3486 	}
3487 
3488 	/*
3489 	 * Ensure that bio is a multiple of internal sector encryption size
3490 	 * and is aligned to this size as defined in IO hints.
3491 	 */
3492 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3493 		return DM_MAPIO_KILL;
3494 
3495 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3496 		return DM_MAPIO_KILL;
3497 
3498 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3499 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3500 
3501 	if (cc->tuple_size) {
3502 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3503 
3504 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3505 			io->integrity_metadata = NULL;
3506 		else
3507 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3508 
3509 		if (unlikely(!io->integrity_metadata)) {
3510 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3511 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3512 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3513 			io->integrity_metadata_from_pool = true;
3514 		}
3515 	}
3516 
3517 	if (crypt_integrity_aead(cc))
3518 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3519 	else
3520 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3521 
3522 	if (bio_data_dir(io->base_bio) == READ) {
3523 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3524 			kcryptd_queue_read(io);
3525 	} else
3526 		kcryptd_queue_crypt(io);
3527 
3528 	return DM_MAPIO_SUBMITTED;
3529 }
3530 
3531 static char hex2asc(unsigned char c)
3532 {
3533 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3534 }
3535 
3536 static void crypt_status(struct dm_target *ti, status_type_t type,
3537 			 unsigned int status_flags, char *result, unsigned int maxlen)
3538 {
3539 	struct crypt_config *cc = ti->private;
3540 	unsigned int i, sz = 0;
3541 	int num_feature_args = 0;
3542 
3543 	switch (type) {
3544 	case STATUSTYPE_INFO:
3545 		result[0] = '\0';
3546 		break;
3547 
3548 	case STATUSTYPE_TABLE:
3549 		DMEMIT("%s ", cc->cipher_string);
3550 
3551 		if (cc->key_size > 0) {
3552 			if (cc->key_string)
3553 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3554 			else {
3555 				for (i = 0; i < cc->key_size; i++) {
3556 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3557 					       hex2asc(cc->key[i] & 0xf));
3558 				}
3559 			}
3560 		} else
3561 			DMEMIT("-");
3562 
3563 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3564 				cc->dev->name, (unsigned long long)cc->start);
3565 
3566 		num_feature_args += !!ti->num_discard_bios;
3567 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3568 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3569 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3570 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3571 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3572 		num_feature_args += !!cc->used_tag_size;
3573 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3574 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3575 		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3576 		if (num_feature_args) {
3577 			DMEMIT(" %d", num_feature_args);
3578 			if (ti->num_discard_bios)
3579 				DMEMIT(" allow_discards");
3580 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3581 				DMEMIT(" same_cpu_crypt");
3582 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3583 				DMEMIT(" high_priority");
3584 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3585 				DMEMIT(" submit_from_crypt_cpus");
3586 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3587 				DMEMIT(" no_read_workqueue");
3588 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3589 				DMEMIT(" no_write_workqueue");
3590 			if (cc->used_tag_size)
3591 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3592 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3593 				DMEMIT(" sector_size:%u", cc->sector_size);
3594 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3595 				DMEMIT(" iv_large_sectors");
3596 			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3597 				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3598 		}
3599 		break;
3600 
3601 	case STATUSTYPE_IMA:
3602 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3603 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3604 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3605 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3606 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3607 		       'y' : 'n');
3608 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3609 		       'y' : 'n');
3610 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3611 		       'y' : 'n');
3612 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3613 		       'y' : 'n');
3614 
3615 		if (cc->used_tag_size)
3616 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3617 			       cc->used_tag_size, cc->cipher_auth);
3618 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3619 			DMEMIT(",sector_size=%u", cc->sector_size);
3620 		if (cc->cipher_string)
3621 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3622 
3623 		DMEMIT(",key_size=%u", cc->key_size);
3624 		DMEMIT(",key_parts=%u", cc->key_parts);
3625 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3626 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3627 		DMEMIT(";");
3628 		break;
3629 	}
3630 }
3631 
3632 static void crypt_postsuspend(struct dm_target *ti)
3633 {
3634 	struct crypt_config *cc = ti->private;
3635 
3636 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3637 }
3638 
3639 static int crypt_preresume(struct dm_target *ti)
3640 {
3641 	struct crypt_config *cc = ti->private;
3642 
3643 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3644 		DMERR("aborting resume - crypt key is not set.");
3645 		return -EAGAIN;
3646 	}
3647 
3648 	return 0;
3649 }
3650 
3651 static void crypt_resume(struct dm_target *ti)
3652 {
3653 	struct crypt_config *cc = ti->private;
3654 
3655 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3656 }
3657 
3658 /* Message interface
3659  *	key set <key>
3660  *	key wipe
3661  */
3662 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3663 			 char *result, unsigned int maxlen)
3664 {
3665 	struct crypt_config *cc = ti->private;
3666 	int key_size, ret = -EINVAL;
3667 
3668 	if (argc < 2)
3669 		goto error;
3670 
3671 	if (!strcasecmp(argv[0], "key")) {
3672 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3673 			DMWARN("not suspended during key manipulation.");
3674 			return -EINVAL;
3675 		}
3676 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3677 			/* The key size may not be changed. */
3678 			key_size = get_key_size(&argv[2]);
3679 			if (key_size < 0 || cc->key_size != key_size) {
3680 				memzero_explicit(argv[2], strlen(argv[2]));
3681 				return -EINVAL;
3682 			}
3683 
3684 			ret = crypt_set_key(cc, argv[2]);
3685 			if (ret)
3686 				return ret;
3687 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3688 				ret = cc->iv_gen_ops->init(cc);
3689 			/* wipe the kernel key payload copy */
3690 			if (cc->key_string)
3691 				memzero_explicit(cc->key, cc->key_size);
3692 			return ret;
3693 		}
3694 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3695 			return crypt_wipe_key(cc);
3696 	}
3697 
3698 error:
3699 	DMWARN("unrecognised message received.");
3700 	return -EINVAL;
3701 }
3702 
3703 static int crypt_iterate_devices(struct dm_target *ti,
3704 				 iterate_devices_callout_fn fn, void *data)
3705 {
3706 	struct crypt_config *cc = ti->private;
3707 
3708 	return fn(ti, cc->dev, cc->start, ti->len, data);
3709 }
3710 
3711 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3712 {
3713 	struct crypt_config *cc = ti->private;
3714 
3715 	dm_stack_bs_limits(limits, cc->sector_size);
3716 	limits->dma_alignment = limits->logical_block_size - 1;
3717 
3718 	/*
3719 	 * For zoned dm-crypt targets, there will be no internal splitting of
3720 	 * write BIOs to avoid exceeding BIO_MAX_VECS vectors per BIO. But
3721 	 * without respecting this limit, crypt_alloc_buffer() will trigger a
3722 	 * BUG(). Avoid this by forcing DM core to split write BIOs to this
3723 	 * limit.
3724 	 */
3725 	if (ti->emulate_zone_append)
3726 		limits->max_hw_sectors = min(limits->max_hw_sectors,
3727 					     BIO_MAX_VECS << PAGE_SECTORS_SHIFT);
3728 
3729 	limits->atomic_write_hw_unit_max = min(limits->atomic_write_hw_unit_max,
3730 					       BIO_MAX_VECS << PAGE_SHIFT);
3731 	limits->atomic_write_hw_max = min(limits->atomic_write_hw_max,
3732 					  BIO_MAX_VECS << PAGE_SHIFT);
3733 }
3734 
3735 static struct target_type crypt_target = {
3736 	.name   = "crypt",
3737 	.version = {1, 30, 0},
3738 	.module = THIS_MODULE,
3739 	.ctr    = crypt_ctr,
3740 	.dtr    = crypt_dtr,
3741 	.features = DM_TARGET_ZONED_HM | DM_TARGET_ATOMIC_WRITES,
3742 	.report_zones = crypt_report_zones,
3743 	.map    = crypt_map,
3744 	.status = crypt_status,
3745 	.postsuspend = crypt_postsuspend,
3746 	.preresume = crypt_preresume,
3747 	.resume = crypt_resume,
3748 	.message = crypt_message,
3749 	.iterate_devices = crypt_iterate_devices,
3750 	.io_hints = crypt_io_hints,
3751 };
3752 module_dm(crypt);
3753 
3754 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3755 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3756 MODULE_LICENSE("GPL");
3757