xref: /openbmc/linux/drivers/md/dm-crypt.c (revision c4c3c32d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2003 Jana Saout <jana@saout.de>
4  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
5  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
6  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
7  *
8  * This file is released under the GPL.
9  */
10 
11 #include <linux/completion.h>
12 #include <linux/err.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/key.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-integrity.h>
20 #include <linux/mempool.h>
21 #include <linux/slab.h>
22 #include <linux/crypto.h>
23 #include <linux/workqueue.h>
24 #include <linux/kthread.h>
25 #include <linux/backing-dev.h>
26 #include <linux/atomic.h>
27 #include <linux/scatterlist.h>
28 #include <linux/rbtree.h>
29 #include <linux/ctype.h>
30 #include <asm/page.h>
31 #include <asm/unaligned.h>
32 #include <crypto/hash.h>
33 #include <crypto/md5.h>
34 #include <crypto/skcipher.h>
35 #include <crypto/aead.h>
36 #include <crypto/authenc.h>
37 #include <crypto/utils.h>
38 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
39 #include <linux/key-type.h>
40 #include <keys/user-type.h>
41 #include <keys/encrypted-type.h>
42 #include <keys/trusted-type.h>
43 
44 #include <linux/device-mapper.h>
45 
46 #include "dm-audit.h"
47 
48 #define DM_MSG_PREFIX "crypt"
49 
50 /*
51  * context holding the current state of a multi-part conversion
52  */
53 struct convert_context {
54 	struct completion restart;
55 	struct bio *bio_in;
56 	struct bio *bio_out;
57 	struct bvec_iter iter_in;
58 	struct bvec_iter iter_out;
59 	u64 cc_sector;
60 	atomic_t cc_pending;
61 	union {
62 		struct skcipher_request *req;
63 		struct aead_request *req_aead;
64 	} r;
65 
66 };
67 
68 /*
69  * per bio private data
70  */
71 struct dm_crypt_io {
72 	struct crypt_config *cc;
73 	struct bio *base_bio;
74 	u8 *integrity_metadata;
75 	bool integrity_metadata_from_pool:1;
76 	bool in_tasklet:1;
77 
78 	struct work_struct work;
79 	struct tasklet_struct tasklet;
80 
81 	struct convert_context ctx;
82 
83 	atomic_t io_pending;
84 	blk_status_t error;
85 	sector_t sector;
86 
87 	struct rb_node rb_node;
88 } CRYPTO_MINALIGN_ATTR;
89 
90 struct dm_crypt_request {
91 	struct convert_context *ctx;
92 	struct scatterlist sg_in[4];
93 	struct scatterlist sg_out[4];
94 	u64 iv_sector;
95 };
96 
97 struct crypt_config;
98 
99 struct crypt_iv_operations {
100 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
101 		   const char *opts);
102 	void (*dtr)(struct crypt_config *cc);
103 	int (*init)(struct crypt_config *cc);
104 	int (*wipe)(struct crypt_config *cc);
105 	int (*generator)(struct crypt_config *cc, u8 *iv,
106 			 struct dm_crypt_request *dmreq);
107 	int (*post)(struct crypt_config *cc, u8 *iv,
108 		    struct dm_crypt_request *dmreq);
109 };
110 
111 struct iv_benbi_private {
112 	int shift;
113 };
114 
115 #define LMK_SEED_SIZE 64 /* hash + 0 */
116 struct iv_lmk_private {
117 	struct crypto_shash *hash_tfm;
118 	u8 *seed;
119 };
120 
121 #define TCW_WHITENING_SIZE 16
122 struct iv_tcw_private {
123 	struct crypto_shash *crc32_tfm;
124 	u8 *iv_seed;
125 	u8 *whitening;
126 };
127 
128 #define ELEPHANT_MAX_KEY_SIZE 32
129 struct iv_elephant_private {
130 	struct crypto_skcipher *tfm;
131 };
132 
133 /*
134  * Crypt: maps a linear range of a block device
135  * and encrypts / decrypts at the same time.
136  */
137 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
138 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
139 	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
140 	     DM_CRYPT_WRITE_INLINE };
141 
142 enum cipher_flags {
143 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
144 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
145 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
146 };
147 
148 /*
149  * The fields in here must be read only after initialization.
150  */
151 struct crypt_config {
152 	struct dm_dev *dev;
153 	sector_t start;
154 
155 	struct percpu_counter n_allocated_pages;
156 
157 	struct workqueue_struct *io_queue;
158 	struct workqueue_struct *crypt_queue;
159 
160 	spinlock_t write_thread_lock;
161 	struct task_struct *write_thread;
162 	struct rb_root write_tree;
163 
164 	char *cipher_string;
165 	char *cipher_auth;
166 	char *key_string;
167 
168 	const struct crypt_iv_operations *iv_gen_ops;
169 	union {
170 		struct iv_benbi_private benbi;
171 		struct iv_lmk_private lmk;
172 		struct iv_tcw_private tcw;
173 		struct iv_elephant_private elephant;
174 	} iv_gen_private;
175 	u64 iv_offset;
176 	unsigned int iv_size;
177 	unsigned short sector_size;
178 	unsigned char sector_shift;
179 
180 	union {
181 		struct crypto_skcipher **tfms;
182 		struct crypto_aead **tfms_aead;
183 	} cipher_tfm;
184 	unsigned int tfms_count;
185 	unsigned long cipher_flags;
186 
187 	/*
188 	 * Layout of each crypto request:
189 	 *
190 	 *   struct skcipher_request
191 	 *      context
192 	 *      padding
193 	 *   struct dm_crypt_request
194 	 *      padding
195 	 *   IV
196 	 *
197 	 * The padding is added so that dm_crypt_request and the IV are
198 	 * correctly aligned.
199 	 */
200 	unsigned int dmreq_start;
201 
202 	unsigned int per_bio_data_size;
203 
204 	unsigned long flags;
205 	unsigned int key_size;
206 	unsigned int key_parts;      /* independent parts in key buffer */
207 	unsigned int key_extra_size; /* additional keys length */
208 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
209 
210 	unsigned int integrity_tag_size;
211 	unsigned int integrity_iv_size;
212 	unsigned int on_disk_tag_size;
213 
214 	/*
215 	 * pool for per bio private data, crypto requests,
216 	 * encryption requeusts/buffer pages and integrity tags
217 	 */
218 	unsigned int tag_pool_max_sectors;
219 	mempool_t tag_pool;
220 	mempool_t req_pool;
221 	mempool_t page_pool;
222 
223 	struct bio_set bs;
224 	struct mutex bio_alloc_lock;
225 
226 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
227 	u8 key[];
228 };
229 
230 #define MIN_IOS		64
231 #define MAX_TAG_SIZE	480
232 #define POOL_ENTRY_SIZE	512
233 
234 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
235 static unsigned int dm_crypt_clients_n;
236 static volatile unsigned long dm_crypt_pages_per_client;
237 #define DM_CRYPT_MEMORY_PERCENT			2
238 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
239 
240 static void crypt_endio(struct bio *clone);
241 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
242 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
243 					     struct scatterlist *sg);
244 
245 static bool crypt_integrity_aead(struct crypt_config *cc);
246 
247 /*
248  * Use this to access cipher attributes that are independent of the key.
249  */
250 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
251 {
252 	return cc->cipher_tfm.tfms[0];
253 }
254 
255 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
256 {
257 	return cc->cipher_tfm.tfms_aead[0];
258 }
259 
260 /*
261  * Different IV generation algorithms:
262  *
263  * plain: the initial vector is the 32-bit little-endian version of the sector
264  *        number, padded with zeros if necessary.
265  *
266  * plain64: the initial vector is the 64-bit little-endian version of the sector
267  *        number, padded with zeros if necessary.
268  *
269  * plain64be: the initial vector is the 64-bit big-endian version of the sector
270  *        number, padded with zeros if necessary.
271  *
272  * essiv: "encrypted sector|salt initial vector", the sector number is
273  *        encrypted with the bulk cipher using a salt as key. The salt
274  *        should be derived from the bulk cipher's key via hashing.
275  *
276  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
277  *        (needed for LRW-32-AES and possible other narrow block modes)
278  *
279  * null: the initial vector is always zero.  Provides compatibility with
280  *       obsolete loop_fish2 devices.  Do not use for new devices.
281  *
282  * lmk:  Compatible implementation of the block chaining mode used
283  *       by the Loop-AES block device encryption system
284  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
285  *       It operates on full 512 byte sectors and uses CBC
286  *       with an IV derived from the sector number, the data and
287  *       optionally extra IV seed.
288  *       This means that after decryption the first block
289  *       of sector must be tweaked according to decrypted data.
290  *       Loop-AES can use three encryption schemes:
291  *         version 1: is plain aes-cbc mode
292  *         version 2: uses 64 multikey scheme with lmk IV generator
293  *         version 3: the same as version 2 with additional IV seed
294  *                   (it uses 65 keys, last key is used as IV seed)
295  *
296  * tcw:  Compatible implementation of the block chaining mode used
297  *       by the TrueCrypt device encryption system (prior to version 4.1).
298  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
299  *       It operates on full 512 byte sectors and uses CBC
300  *       with an IV derived from initial key and the sector number.
301  *       In addition, whitening value is applied on every sector, whitening
302  *       is calculated from initial key, sector number and mixed using CRC32.
303  *       Note that this encryption scheme is vulnerable to watermarking attacks
304  *       and should be used for old compatible containers access only.
305  *
306  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
307  *        The IV is encrypted little-endian byte-offset (with the same key
308  *        and cipher as the volume).
309  *
310  * elephant: The extended version of eboiv with additional Elephant diffuser
311  *           used with Bitlocker CBC mode.
312  *           This mode was used in older Windows systems
313  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
314  */
315 
316 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
317 			      struct dm_crypt_request *dmreq)
318 {
319 	memset(iv, 0, cc->iv_size);
320 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
321 
322 	return 0;
323 }
324 
325 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
326 				struct dm_crypt_request *dmreq)
327 {
328 	memset(iv, 0, cc->iv_size);
329 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
330 
331 	return 0;
332 }
333 
334 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
335 				  struct dm_crypt_request *dmreq)
336 {
337 	memset(iv, 0, cc->iv_size);
338 	/* iv_size is at least of size u64; usually it is 16 bytes */
339 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
340 
341 	return 0;
342 }
343 
344 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
345 			      struct dm_crypt_request *dmreq)
346 {
347 	/*
348 	 * ESSIV encryption of the IV is now handled by the crypto API,
349 	 * so just pass the plain sector number here.
350 	 */
351 	memset(iv, 0, cc->iv_size);
352 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
353 
354 	return 0;
355 }
356 
357 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
358 			      const char *opts)
359 {
360 	unsigned int bs;
361 	int log;
362 
363 	if (crypt_integrity_aead(cc))
364 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
365 	else
366 		bs = crypto_skcipher_blocksize(any_tfm(cc));
367 	log = ilog2(bs);
368 
369 	/*
370 	 * We need to calculate how far we must shift the sector count
371 	 * to get the cipher block count, we use this shift in _gen.
372 	 */
373 	if (1 << log != bs) {
374 		ti->error = "cypher blocksize is not a power of 2";
375 		return -EINVAL;
376 	}
377 
378 	if (log > 9) {
379 		ti->error = "cypher blocksize is > 512";
380 		return -EINVAL;
381 	}
382 
383 	cc->iv_gen_private.benbi.shift = 9 - log;
384 
385 	return 0;
386 }
387 
388 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
389 {
390 }
391 
392 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
393 			      struct dm_crypt_request *dmreq)
394 {
395 	__be64 val;
396 
397 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
398 
399 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
400 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
401 
402 	return 0;
403 }
404 
405 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
406 			     struct dm_crypt_request *dmreq)
407 {
408 	memset(iv, 0, cc->iv_size);
409 
410 	return 0;
411 }
412 
413 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
414 {
415 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
416 
417 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
418 		crypto_free_shash(lmk->hash_tfm);
419 	lmk->hash_tfm = NULL;
420 
421 	kfree_sensitive(lmk->seed);
422 	lmk->seed = NULL;
423 }
424 
425 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
426 			    const char *opts)
427 {
428 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
429 
430 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
431 		ti->error = "Unsupported sector size for LMK";
432 		return -EINVAL;
433 	}
434 
435 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
436 					   CRYPTO_ALG_ALLOCATES_MEMORY);
437 	if (IS_ERR(lmk->hash_tfm)) {
438 		ti->error = "Error initializing LMK hash";
439 		return PTR_ERR(lmk->hash_tfm);
440 	}
441 
442 	/* No seed in LMK version 2 */
443 	if (cc->key_parts == cc->tfms_count) {
444 		lmk->seed = NULL;
445 		return 0;
446 	}
447 
448 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
449 	if (!lmk->seed) {
450 		crypt_iv_lmk_dtr(cc);
451 		ti->error = "Error kmallocing seed storage in LMK";
452 		return -ENOMEM;
453 	}
454 
455 	return 0;
456 }
457 
458 static int crypt_iv_lmk_init(struct crypt_config *cc)
459 {
460 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
461 	int subkey_size = cc->key_size / cc->key_parts;
462 
463 	/* LMK seed is on the position of LMK_KEYS + 1 key */
464 	if (lmk->seed)
465 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
466 		       crypto_shash_digestsize(lmk->hash_tfm));
467 
468 	return 0;
469 }
470 
471 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
472 {
473 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
474 
475 	if (lmk->seed)
476 		memset(lmk->seed, 0, LMK_SEED_SIZE);
477 
478 	return 0;
479 }
480 
481 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
482 			    struct dm_crypt_request *dmreq,
483 			    u8 *data)
484 {
485 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
486 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
487 	struct md5_state md5state;
488 	__le32 buf[4];
489 	int i, r;
490 
491 	desc->tfm = lmk->hash_tfm;
492 
493 	r = crypto_shash_init(desc);
494 	if (r)
495 		return r;
496 
497 	if (lmk->seed) {
498 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
499 		if (r)
500 			return r;
501 	}
502 
503 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
504 	r = crypto_shash_update(desc, data + 16, 16 * 31);
505 	if (r)
506 		return r;
507 
508 	/* Sector is cropped to 56 bits here */
509 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
510 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
511 	buf[2] = cpu_to_le32(4024);
512 	buf[3] = 0;
513 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
514 	if (r)
515 		return r;
516 
517 	/* No MD5 padding here */
518 	r = crypto_shash_export(desc, &md5state);
519 	if (r)
520 		return r;
521 
522 	for (i = 0; i < MD5_HASH_WORDS; i++)
523 		__cpu_to_le32s(&md5state.hash[i]);
524 	memcpy(iv, &md5state.hash, cc->iv_size);
525 
526 	return 0;
527 }
528 
529 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
530 			    struct dm_crypt_request *dmreq)
531 {
532 	struct scatterlist *sg;
533 	u8 *src;
534 	int r = 0;
535 
536 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
537 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
538 		src = kmap_local_page(sg_page(sg));
539 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
540 		kunmap_local(src);
541 	} else
542 		memset(iv, 0, cc->iv_size);
543 
544 	return r;
545 }
546 
547 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
548 			     struct dm_crypt_request *dmreq)
549 {
550 	struct scatterlist *sg;
551 	u8 *dst;
552 	int r;
553 
554 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
555 		return 0;
556 
557 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
558 	dst = kmap_local_page(sg_page(sg));
559 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
560 
561 	/* Tweak the first block of plaintext sector */
562 	if (!r)
563 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
564 
565 	kunmap_local(dst);
566 	return r;
567 }
568 
569 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
570 {
571 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
572 
573 	kfree_sensitive(tcw->iv_seed);
574 	tcw->iv_seed = NULL;
575 	kfree_sensitive(tcw->whitening);
576 	tcw->whitening = NULL;
577 
578 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
579 		crypto_free_shash(tcw->crc32_tfm);
580 	tcw->crc32_tfm = NULL;
581 }
582 
583 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
584 			    const char *opts)
585 {
586 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
587 
588 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
589 		ti->error = "Unsupported sector size for TCW";
590 		return -EINVAL;
591 	}
592 
593 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
594 		ti->error = "Wrong key size for TCW";
595 		return -EINVAL;
596 	}
597 
598 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
599 					    CRYPTO_ALG_ALLOCATES_MEMORY);
600 	if (IS_ERR(tcw->crc32_tfm)) {
601 		ti->error = "Error initializing CRC32 in TCW";
602 		return PTR_ERR(tcw->crc32_tfm);
603 	}
604 
605 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
606 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
607 	if (!tcw->iv_seed || !tcw->whitening) {
608 		crypt_iv_tcw_dtr(cc);
609 		ti->error = "Error allocating seed storage in TCW";
610 		return -ENOMEM;
611 	}
612 
613 	return 0;
614 }
615 
616 static int crypt_iv_tcw_init(struct crypt_config *cc)
617 {
618 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
619 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
620 
621 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
622 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
623 	       TCW_WHITENING_SIZE);
624 
625 	return 0;
626 }
627 
628 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
629 {
630 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
631 
632 	memset(tcw->iv_seed, 0, cc->iv_size);
633 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
634 
635 	return 0;
636 }
637 
638 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
639 				  struct dm_crypt_request *dmreq,
640 				  u8 *data)
641 {
642 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
643 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
644 	u8 buf[TCW_WHITENING_SIZE];
645 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
646 	int i, r;
647 
648 	/* xor whitening with sector number */
649 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
650 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
651 
652 	/* calculate crc32 for every 32bit part and xor it */
653 	desc->tfm = tcw->crc32_tfm;
654 	for (i = 0; i < 4; i++) {
655 		r = crypto_shash_init(desc);
656 		if (r)
657 			goto out;
658 		r = crypto_shash_update(desc, &buf[i * 4], 4);
659 		if (r)
660 			goto out;
661 		r = crypto_shash_final(desc, &buf[i * 4]);
662 		if (r)
663 			goto out;
664 	}
665 	crypto_xor(&buf[0], &buf[12], 4);
666 	crypto_xor(&buf[4], &buf[8], 4);
667 
668 	/* apply whitening (8 bytes) to whole sector */
669 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
670 		crypto_xor(data + i * 8, buf, 8);
671 out:
672 	memzero_explicit(buf, sizeof(buf));
673 	return r;
674 }
675 
676 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
677 			    struct dm_crypt_request *dmreq)
678 {
679 	struct scatterlist *sg;
680 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
681 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
682 	u8 *src;
683 	int r = 0;
684 
685 	/* Remove whitening from ciphertext */
686 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
687 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
688 		src = kmap_local_page(sg_page(sg));
689 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
690 		kunmap_local(src);
691 	}
692 
693 	/* Calculate IV */
694 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
695 	if (cc->iv_size > 8)
696 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
697 			       cc->iv_size - 8);
698 
699 	return r;
700 }
701 
702 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
703 			     struct dm_crypt_request *dmreq)
704 {
705 	struct scatterlist *sg;
706 	u8 *dst;
707 	int r;
708 
709 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
710 		return 0;
711 
712 	/* Apply whitening on ciphertext */
713 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
714 	dst = kmap_local_page(sg_page(sg));
715 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
716 	kunmap_local(dst);
717 
718 	return r;
719 }
720 
721 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
722 				struct dm_crypt_request *dmreq)
723 {
724 	/* Used only for writes, there must be an additional space to store IV */
725 	get_random_bytes(iv, cc->iv_size);
726 	return 0;
727 }
728 
729 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
730 			    const char *opts)
731 {
732 	if (crypt_integrity_aead(cc)) {
733 		ti->error = "AEAD transforms not supported for EBOIV";
734 		return -EINVAL;
735 	}
736 
737 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
738 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
739 		return -EINVAL;
740 	}
741 
742 	return 0;
743 }
744 
745 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
746 			    struct dm_crypt_request *dmreq)
747 {
748 	struct crypto_skcipher *tfm = any_tfm(cc);
749 	struct skcipher_request *req;
750 	struct scatterlist src, dst;
751 	DECLARE_CRYPTO_WAIT(wait);
752 	unsigned int reqsize;
753 	int err;
754 	u8 *buf;
755 
756 	reqsize = ALIGN(crypto_skcipher_reqsize(tfm), __alignof__(__le64));
757 
758 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
759 	if (!req)
760 		return -ENOMEM;
761 
762 	skcipher_request_set_tfm(req, tfm);
763 
764 	buf = (u8 *)req + reqsize;
765 	memset(buf, 0, cc->iv_size);
766 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
767 
768 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
769 	sg_init_one(&dst, iv, cc->iv_size);
770 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
771 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
772 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
773 	kfree_sensitive(req);
774 
775 	return err;
776 }
777 
778 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
779 {
780 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
781 
782 	crypto_free_skcipher(elephant->tfm);
783 	elephant->tfm = NULL;
784 }
785 
786 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
787 			    const char *opts)
788 {
789 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
790 	int r;
791 
792 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
793 					      CRYPTO_ALG_ALLOCATES_MEMORY);
794 	if (IS_ERR(elephant->tfm)) {
795 		r = PTR_ERR(elephant->tfm);
796 		elephant->tfm = NULL;
797 		return r;
798 	}
799 
800 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
801 	if (r)
802 		crypt_iv_elephant_dtr(cc);
803 	return r;
804 }
805 
806 static void diffuser_disk_to_cpu(u32 *d, size_t n)
807 {
808 #ifndef __LITTLE_ENDIAN
809 	int i;
810 
811 	for (i = 0; i < n; i++)
812 		d[i] = le32_to_cpu((__le32)d[i]);
813 #endif
814 }
815 
816 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
817 {
818 #ifndef __LITTLE_ENDIAN
819 	int i;
820 
821 	for (i = 0; i < n; i++)
822 		d[i] = cpu_to_le32((u32)d[i]);
823 #endif
824 }
825 
826 static void diffuser_a_decrypt(u32 *d, size_t n)
827 {
828 	int i, i1, i2, i3;
829 
830 	for (i = 0; i < 5; i++) {
831 		i1 = 0;
832 		i2 = n - 2;
833 		i3 = n - 5;
834 
835 		while (i1 < (n - 1)) {
836 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
837 			i1++; i2++; i3++;
838 
839 			if (i3 >= n)
840 				i3 -= n;
841 
842 			d[i1] += d[i2] ^ d[i3];
843 			i1++; i2++; i3++;
844 
845 			if (i2 >= n)
846 				i2 -= n;
847 
848 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
849 			i1++; i2++; i3++;
850 
851 			d[i1] += d[i2] ^ d[i3];
852 			i1++; i2++; i3++;
853 		}
854 	}
855 }
856 
857 static void diffuser_a_encrypt(u32 *d, size_t n)
858 {
859 	int i, i1, i2, i3;
860 
861 	for (i = 0; i < 5; i++) {
862 		i1 = n - 1;
863 		i2 = n - 2 - 1;
864 		i3 = n - 5 - 1;
865 
866 		while (i1 > 0) {
867 			d[i1] -= d[i2] ^ d[i3];
868 			i1--; i2--; i3--;
869 
870 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
871 			i1--; i2--; i3--;
872 
873 			if (i2 < 0)
874 				i2 += n;
875 
876 			d[i1] -= d[i2] ^ d[i3];
877 			i1--; i2--; i3--;
878 
879 			if (i3 < 0)
880 				i3 += n;
881 
882 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
883 			i1--; i2--; i3--;
884 		}
885 	}
886 }
887 
888 static void diffuser_b_decrypt(u32 *d, size_t n)
889 {
890 	int i, i1, i2, i3;
891 
892 	for (i = 0; i < 3; i++) {
893 		i1 = 0;
894 		i2 = 2;
895 		i3 = 5;
896 
897 		while (i1 < (n - 1)) {
898 			d[i1] += d[i2] ^ d[i3];
899 			i1++; i2++; i3++;
900 
901 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
902 			i1++; i2++; i3++;
903 
904 			if (i2 >= n)
905 				i2 -= n;
906 
907 			d[i1] += d[i2] ^ d[i3];
908 			i1++; i2++; i3++;
909 
910 			if (i3 >= n)
911 				i3 -= n;
912 
913 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
914 			i1++; i2++; i3++;
915 		}
916 	}
917 }
918 
919 static void diffuser_b_encrypt(u32 *d, size_t n)
920 {
921 	int i, i1, i2, i3;
922 
923 	for (i = 0; i < 3; i++) {
924 		i1 = n - 1;
925 		i2 = 2 - 1;
926 		i3 = 5 - 1;
927 
928 		while (i1 > 0) {
929 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
930 			i1--; i2--; i3--;
931 
932 			if (i3 < 0)
933 				i3 += n;
934 
935 			d[i1] -= d[i2] ^ d[i3];
936 			i1--; i2--; i3--;
937 
938 			if (i2 < 0)
939 				i2 += n;
940 
941 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
942 			i1--; i2--; i3--;
943 
944 			d[i1] -= d[i2] ^ d[i3];
945 			i1--; i2--; i3--;
946 		}
947 	}
948 }
949 
950 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
951 {
952 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
953 	u8 *es, *ks, *data, *data2, *data_offset;
954 	struct skcipher_request *req;
955 	struct scatterlist *sg, *sg2, src, dst;
956 	DECLARE_CRYPTO_WAIT(wait);
957 	int i, r;
958 
959 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
960 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
961 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
962 
963 	if (!req || !es || !ks) {
964 		r = -ENOMEM;
965 		goto out;
966 	}
967 
968 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
969 
970 	/* E(Ks, e(s)) */
971 	sg_init_one(&src, es, 16);
972 	sg_init_one(&dst, ks, 16);
973 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
974 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
975 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
976 	if (r)
977 		goto out;
978 
979 	/* E(Ks, e'(s)) */
980 	es[15] = 0x80;
981 	sg_init_one(&dst, &ks[16], 16);
982 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
983 	if (r)
984 		goto out;
985 
986 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
987 	data = kmap_local_page(sg_page(sg));
988 	data_offset = data + sg->offset;
989 
990 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
991 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
992 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
993 		data2 = kmap_local_page(sg_page(sg2));
994 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
995 		kunmap_local(data2);
996 	}
997 
998 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
999 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1000 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1001 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1002 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1003 	}
1004 
1005 	for (i = 0; i < (cc->sector_size / 32); i++)
1006 		crypto_xor(data_offset + i * 32, ks, 32);
1007 
1008 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1009 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1010 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1011 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1012 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1013 	}
1014 
1015 	kunmap_local(data);
1016 out:
1017 	kfree_sensitive(ks);
1018 	kfree_sensitive(es);
1019 	skcipher_request_free(req);
1020 	return r;
1021 }
1022 
1023 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1024 			    struct dm_crypt_request *dmreq)
1025 {
1026 	int r;
1027 
1028 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1029 		r = crypt_iv_elephant(cc, dmreq);
1030 		if (r)
1031 			return r;
1032 	}
1033 
1034 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1035 }
1036 
1037 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1038 				  struct dm_crypt_request *dmreq)
1039 {
1040 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1041 		return crypt_iv_elephant(cc, dmreq);
1042 
1043 	return 0;
1044 }
1045 
1046 static int crypt_iv_elephant_init(struct crypt_config *cc)
1047 {
1048 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1049 	int key_offset = cc->key_size - cc->key_extra_size;
1050 
1051 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1052 }
1053 
1054 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1055 {
1056 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1057 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1058 
1059 	memset(key, 0, cc->key_extra_size);
1060 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1061 }
1062 
1063 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1064 	.generator = crypt_iv_plain_gen
1065 };
1066 
1067 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1068 	.generator = crypt_iv_plain64_gen
1069 };
1070 
1071 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1072 	.generator = crypt_iv_plain64be_gen
1073 };
1074 
1075 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1076 	.generator = crypt_iv_essiv_gen
1077 };
1078 
1079 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1080 	.ctr	   = crypt_iv_benbi_ctr,
1081 	.dtr	   = crypt_iv_benbi_dtr,
1082 	.generator = crypt_iv_benbi_gen
1083 };
1084 
1085 static const struct crypt_iv_operations crypt_iv_null_ops = {
1086 	.generator = crypt_iv_null_gen
1087 };
1088 
1089 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1090 	.ctr	   = crypt_iv_lmk_ctr,
1091 	.dtr	   = crypt_iv_lmk_dtr,
1092 	.init	   = crypt_iv_lmk_init,
1093 	.wipe	   = crypt_iv_lmk_wipe,
1094 	.generator = crypt_iv_lmk_gen,
1095 	.post	   = crypt_iv_lmk_post
1096 };
1097 
1098 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1099 	.ctr	   = crypt_iv_tcw_ctr,
1100 	.dtr	   = crypt_iv_tcw_dtr,
1101 	.init	   = crypt_iv_tcw_init,
1102 	.wipe	   = crypt_iv_tcw_wipe,
1103 	.generator = crypt_iv_tcw_gen,
1104 	.post	   = crypt_iv_tcw_post
1105 };
1106 
1107 static const struct crypt_iv_operations crypt_iv_random_ops = {
1108 	.generator = crypt_iv_random_gen
1109 };
1110 
1111 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1112 	.ctr	   = crypt_iv_eboiv_ctr,
1113 	.generator = crypt_iv_eboiv_gen
1114 };
1115 
1116 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1117 	.ctr	   = crypt_iv_elephant_ctr,
1118 	.dtr	   = crypt_iv_elephant_dtr,
1119 	.init	   = crypt_iv_elephant_init,
1120 	.wipe	   = crypt_iv_elephant_wipe,
1121 	.generator = crypt_iv_elephant_gen,
1122 	.post	   = crypt_iv_elephant_post
1123 };
1124 
1125 /*
1126  * Integrity extensions
1127  */
1128 static bool crypt_integrity_aead(struct crypt_config *cc)
1129 {
1130 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1131 }
1132 
1133 static bool crypt_integrity_hmac(struct crypt_config *cc)
1134 {
1135 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1136 }
1137 
1138 /* Get sg containing data */
1139 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1140 					     struct scatterlist *sg)
1141 {
1142 	if (unlikely(crypt_integrity_aead(cc)))
1143 		return &sg[2];
1144 
1145 	return sg;
1146 }
1147 
1148 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1149 {
1150 	struct bio_integrity_payload *bip;
1151 	unsigned int tag_len;
1152 	int ret;
1153 
1154 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1155 		return 0;
1156 
1157 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1158 	if (IS_ERR(bip))
1159 		return PTR_ERR(bip);
1160 
1161 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1162 
1163 	bip->bip_iter.bi_size = tag_len;
1164 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1165 
1166 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1167 				     tag_len, offset_in_page(io->integrity_metadata));
1168 	if (unlikely(ret != tag_len))
1169 		return -ENOMEM;
1170 
1171 	return 0;
1172 }
1173 
1174 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1175 {
1176 #ifdef CONFIG_BLK_DEV_INTEGRITY
1177 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1178 	struct mapped_device *md = dm_table_get_md(ti->table);
1179 
1180 	/* From now we require underlying device with our integrity profile */
1181 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1182 		ti->error = "Integrity profile not supported.";
1183 		return -EINVAL;
1184 	}
1185 
1186 	if (bi->tag_size != cc->on_disk_tag_size ||
1187 	    bi->tuple_size != cc->on_disk_tag_size) {
1188 		ti->error = "Integrity profile tag size mismatch.";
1189 		return -EINVAL;
1190 	}
1191 	if (1 << bi->interval_exp != cc->sector_size) {
1192 		ti->error = "Integrity profile sector size mismatch.";
1193 		return -EINVAL;
1194 	}
1195 
1196 	if (crypt_integrity_aead(cc)) {
1197 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1198 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1199 		       cc->integrity_tag_size, cc->integrity_iv_size);
1200 
1201 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1202 			ti->error = "Integrity AEAD auth tag size is not supported.";
1203 			return -EINVAL;
1204 		}
1205 	} else if (cc->integrity_iv_size)
1206 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1207 		       cc->integrity_iv_size);
1208 
1209 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1210 		ti->error = "Not enough space for integrity tag in the profile.";
1211 		return -EINVAL;
1212 	}
1213 
1214 	return 0;
1215 #else
1216 	ti->error = "Integrity profile not supported.";
1217 	return -EINVAL;
1218 #endif
1219 }
1220 
1221 static void crypt_convert_init(struct crypt_config *cc,
1222 			       struct convert_context *ctx,
1223 			       struct bio *bio_out, struct bio *bio_in,
1224 			       sector_t sector)
1225 {
1226 	ctx->bio_in = bio_in;
1227 	ctx->bio_out = bio_out;
1228 	if (bio_in)
1229 		ctx->iter_in = bio_in->bi_iter;
1230 	if (bio_out)
1231 		ctx->iter_out = bio_out->bi_iter;
1232 	ctx->cc_sector = sector + cc->iv_offset;
1233 	init_completion(&ctx->restart);
1234 }
1235 
1236 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1237 					     void *req)
1238 {
1239 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1240 }
1241 
1242 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1243 {
1244 	return (void *)((char *)dmreq - cc->dmreq_start);
1245 }
1246 
1247 static u8 *iv_of_dmreq(struct crypt_config *cc,
1248 		       struct dm_crypt_request *dmreq)
1249 {
1250 	if (crypt_integrity_aead(cc))
1251 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1252 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1253 	else
1254 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1255 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1256 }
1257 
1258 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1259 		       struct dm_crypt_request *dmreq)
1260 {
1261 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1262 }
1263 
1264 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1265 		       struct dm_crypt_request *dmreq)
1266 {
1267 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1268 
1269 	return (__le64 *) ptr;
1270 }
1271 
1272 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1273 		       struct dm_crypt_request *dmreq)
1274 {
1275 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1276 		  cc->iv_size + sizeof(uint64_t);
1277 
1278 	return (unsigned int *)ptr;
1279 }
1280 
1281 static void *tag_from_dmreq(struct crypt_config *cc,
1282 				struct dm_crypt_request *dmreq)
1283 {
1284 	struct convert_context *ctx = dmreq->ctx;
1285 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1286 
1287 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1288 		cc->on_disk_tag_size];
1289 }
1290 
1291 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1292 			       struct dm_crypt_request *dmreq)
1293 {
1294 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1295 }
1296 
1297 static int crypt_convert_block_aead(struct crypt_config *cc,
1298 				     struct convert_context *ctx,
1299 				     struct aead_request *req,
1300 				     unsigned int tag_offset)
1301 {
1302 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1303 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1304 	struct dm_crypt_request *dmreq;
1305 	u8 *iv, *org_iv, *tag_iv, *tag;
1306 	__le64 *sector;
1307 	int r = 0;
1308 
1309 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1310 
1311 	/* Reject unexpected unaligned bio. */
1312 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1313 		return -EIO;
1314 
1315 	dmreq = dmreq_of_req(cc, req);
1316 	dmreq->iv_sector = ctx->cc_sector;
1317 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1318 		dmreq->iv_sector >>= cc->sector_shift;
1319 	dmreq->ctx = ctx;
1320 
1321 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1322 
1323 	sector = org_sector_of_dmreq(cc, dmreq);
1324 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1325 
1326 	iv = iv_of_dmreq(cc, dmreq);
1327 	org_iv = org_iv_of_dmreq(cc, dmreq);
1328 	tag = tag_from_dmreq(cc, dmreq);
1329 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1330 
1331 	/* AEAD request:
1332 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1333 	 *  | (authenticated) | (auth+encryption) |              |
1334 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1335 	 */
1336 	sg_init_table(dmreq->sg_in, 4);
1337 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1338 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1339 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1340 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1341 
1342 	sg_init_table(dmreq->sg_out, 4);
1343 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1344 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1345 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1346 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1347 
1348 	if (cc->iv_gen_ops) {
1349 		/* For READs use IV stored in integrity metadata */
1350 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1351 			memcpy(org_iv, tag_iv, cc->iv_size);
1352 		} else {
1353 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1354 			if (r < 0)
1355 				return r;
1356 			/* Store generated IV in integrity metadata */
1357 			if (cc->integrity_iv_size)
1358 				memcpy(tag_iv, org_iv, cc->iv_size);
1359 		}
1360 		/* Working copy of IV, to be modified in crypto API */
1361 		memcpy(iv, org_iv, cc->iv_size);
1362 	}
1363 
1364 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1365 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1366 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1367 				       cc->sector_size, iv);
1368 		r = crypto_aead_encrypt(req);
1369 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1370 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1371 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1372 	} else {
1373 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1374 				       cc->sector_size + cc->integrity_tag_size, iv);
1375 		r = crypto_aead_decrypt(req);
1376 	}
1377 
1378 	if (r == -EBADMSG) {
1379 		sector_t s = le64_to_cpu(*sector);
1380 
1381 		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1382 			    ctx->bio_in->bi_bdev, s);
1383 		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1384 				 ctx->bio_in, s, 0);
1385 	}
1386 
1387 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1388 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1389 
1390 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1391 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1392 
1393 	return r;
1394 }
1395 
1396 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1397 					struct convert_context *ctx,
1398 					struct skcipher_request *req,
1399 					unsigned int tag_offset)
1400 {
1401 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1402 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1403 	struct scatterlist *sg_in, *sg_out;
1404 	struct dm_crypt_request *dmreq;
1405 	u8 *iv, *org_iv, *tag_iv;
1406 	__le64 *sector;
1407 	int r = 0;
1408 
1409 	/* Reject unexpected unaligned bio. */
1410 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1411 		return -EIO;
1412 
1413 	dmreq = dmreq_of_req(cc, req);
1414 	dmreq->iv_sector = ctx->cc_sector;
1415 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1416 		dmreq->iv_sector >>= cc->sector_shift;
1417 	dmreq->ctx = ctx;
1418 
1419 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1420 
1421 	iv = iv_of_dmreq(cc, dmreq);
1422 	org_iv = org_iv_of_dmreq(cc, dmreq);
1423 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1424 
1425 	sector = org_sector_of_dmreq(cc, dmreq);
1426 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1427 
1428 	/* For skcipher we use only the first sg item */
1429 	sg_in  = &dmreq->sg_in[0];
1430 	sg_out = &dmreq->sg_out[0];
1431 
1432 	sg_init_table(sg_in, 1);
1433 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1434 
1435 	sg_init_table(sg_out, 1);
1436 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1437 
1438 	if (cc->iv_gen_ops) {
1439 		/* For READs use IV stored in integrity metadata */
1440 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1441 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1442 		} else {
1443 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1444 			if (r < 0)
1445 				return r;
1446 			/* Data can be already preprocessed in generator */
1447 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1448 				sg_in = sg_out;
1449 			/* Store generated IV in integrity metadata */
1450 			if (cc->integrity_iv_size)
1451 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1452 		}
1453 		/* Working copy of IV, to be modified in crypto API */
1454 		memcpy(iv, org_iv, cc->iv_size);
1455 	}
1456 
1457 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1458 
1459 	if (bio_data_dir(ctx->bio_in) == WRITE)
1460 		r = crypto_skcipher_encrypt(req);
1461 	else
1462 		r = crypto_skcipher_decrypt(req);
1463 
1464 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1465 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1466 
1467 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1468 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1469 
1470 	return r;
1471 }
1472 
1473 static void kcryptd_async_done(void *async_req, int error);
1474 
1475 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1476 				     struct convert_context *ctx)
1477 {
1478 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1479 
1480 	if (!ctx->r.req) {
1481 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1482 		if (!ctx->r.req)
1483 			return -ENOMEM;
1484 	}
1485 
1486 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1487 
1488 	/*
1489 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1490 	 * requests if driver request queue is full.
1491 	 */
1492 	skcipher_request_set_callback(ctx->r.req,
1493 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1494 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1495 
1496 	return 0;
1497 }
1498 
1499 static int crypt_alloc_req_aead(struct crypt_config *cc,
1500 				 struct convert_context *ctx)
1501 {
1502 	if (!ctx->r.req_aead) {
1503 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1504 		if (!ctx->r.req_aead)
1505 			return -ENOMEM;
1506 	}
1507 
1508 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1509 
1510 	/*
1511 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1512 	 * requests if driver request queue is full.
1513 	 */
1514 	aead_request_set_callback(ctx->r.req_aead,
1515 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1516 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1517 
1518 	return 0;
1519 }
1520 
1521 static int crypt_alloc_req(struct crypt_config *cc,
1522 			    struct convert_context *ctx)
1523 {
1524 	if (crypt_integrity_aead(cc))
1525 		return crypt_alloc_req_aead(cc, ctx);
1526 	else
1527 		return crypt_alloc_req_skcipher(cc, ctx);
1528 }
1529 
1530 static void crypt_free_req_skcipher(struct crypt_config *cc,
1531 				    struct skcipher_request *req, struct bio *base_bio)
1532 {
1533 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1534 
1535 	if ((struct skcipher_request *)(io + 1) != req)
1536 		mempool_free(req, &cc->req_pool);
1537 }
1538 
1539 static void crypt_free_req_aead(struct crypt_config *cc,
1540 				struct aead_request *req, struct bio *base_bio)
1541 {
1542 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1543 
1544 	if ((struct aead_request *)(io + 1) != req)
1545 		mempool_free(req, &cc->req_pool);
1546 }
1547 
1548 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1549 {
1550 	if (crypt_integrity_aead(cc))
1551 		crypt_free_req_aead(cc, req, base_bio);
1552 	else
1553 		crypt_free_req_skcipher(cc, req, base_bio);
1554 }
1555 
1556 /*
1557  * Encrypt / decrypt data from one bio to another one (can be the same one)
1558  */
1559 static blk_status_t crypt_convert(struct crypt_config *cc,
1560 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1561 {
1562 	unsigned int tag_offset = 0;
1563 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1564 	int r;
1565 
1566 	/*
1567 	 * if reset_pending is set we are dealing with the bio for the first time,
1568 	 * else we're continuing to work on the previous bio, so don't mess with
1569 	 * the cc_pending counter
1570 	 */
1571 	if (reset_pending)
1572 		atomic_set(&ctx->cc_pending, 1);
1573 
1574 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1575 
1576 		r = crypt_alloc_req(cc, ctx);
1577 		if (r) {
1578 			complete(&ctx->restart);
1579 			return BLK_STS_DEV_RESOURCE;
1580 		}
1581 
1582 		atomic_inc(&ctx->cc_pending);
1583 
1584 		if (crypt_integrity_aead(cc))
1585 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1586 		else
1587 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1588 
1589 		switch (r) {
1590 		/*
1591 		 * The request was queued by a crypto driver
1592 		 * but the driver request queue is full, let's wait.
1593 		 */
1594 		case -EBUSY:
1595 			if (in_interrupt()) {
1596 				if (try_wait_for_completion(&ctx->restart)) {
1597 					/*
1598 					 * we don't have to block to wait for completion,
1599 					 * so proceed
1600 					 */
1601 				} else {
1602 					/*
1603 					 * we can't wait for completion without blocking
1604 					 * exit and continue processing in a workqueue
1605 					 */
1606 					ctx->r.req = NULL;
1607 					ctx->cc_sector += sector_step;
1608 					tag_offset++;
1609 					return BLK_STS_DEV_RESOURCE;
1610 				}
1611 			} else {
1612 				wait_for_completion(&ctx->restart);
1613 			}
1614 			reinit_completion(&ctx->restart);
1615 			fallthrough;
1616 		/*
1617 		 * The request is queued and processed asynchronously,
1618 		 * completion function kcryptd_async_done() will be called.
1619 		 */
1620 		case -EINPROGRESS:
1621 			ctx->r.req = NULL;
1622 			ctx->cc_sector += sector_step;
1623 			tag_offset++;
1624 			continue;
1625 		/*
1626 		 * The request was already processed (synchronously).
1627 		 */
1628 		case 0:
1629 			atomic_dec(&ctx->cc_pending);
1630 			ctx->cc_sector += sector_step;
1631 			tag_offset++;
1632 			if (!atomic)
1633 				cond_resched();
1634 			continue;
1635 		/*
1636 		 * There was a data integrity error.
1637 		 */
1638 		case -EBADMSG:
1639 			atomic_dec(&ctx->cc_pending);
1640 			return BLK_STS_PROTECTION;
1641 		/*
1642 		 * There was an error while processing the request.
1643 		 */
1644 		default:
1645 			atomic_dec(&ctx->cc_pending);
1646 			return BLK_STS_IOERR;
1647 		}
1648 	}
1649 
1650 	return 0;
1651 }
1652 
1653 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1654 
1655 /*
1656  * Generate a new unfragmented bio with the given size
1657  * This should never violate the device limitations (but only because
1658  * max_segment_size is being constrained to PAGE_SIZE).
1659  *
1660  * This function may be called concurrently. If we allocate from the mempool
1661  * concurrently, there is a possibility of deadlock. For example, if we have
1662  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1663  * the mempool concurrently, it may deadlock in a situation where both processes
1664  * have allocated 128 pages and the mempool is exhausted.
1665  *
1666  * In order to avoid this scenario we allocate the pages under a mutex.
1667  *
1668  * In order to not degrade performance with excessive locking, we try
1669  * non-blocking allocations without a mutex first but on failure we fallback
1670  * to blocking allocations with a mutex.
1671  *
1672  * In order to reduce allocation overhead, we try to allocate compound pages in
1673  * the first pass. If they are not available, we fall back to the mempool.
1674  */
1675 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1676 {
1677 	struct crypt_config *cc = io->cc;
1678 	struct bio *clone;
1679 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1680 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1681 	unsigned int remaining_size;
1682 	unsigned int order = MAX_ORDER - 1;
1683 
1684 retry:
1685 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1686 		mutex_lock(&cc->bio_alloc_lock);
1687 
1688 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1689 				 GFP_NOIO, &cc->bs);
1690 	clone->bi_private = io;
1691 	clone->bi_end_io = crypt_endio;
1692 
1693 	remaining_size = size;
1694 
1695 	while (remaining_size) {
1696 		struct page *pages;
1697 		unsigned size_to_add;
1698 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1699 		order = min(order, remaining_order);
1700 
1701 		while (order > 0) {
1702 			pages = alloc_pages(gfp_mask
1703 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1704 				order);
1705 			if (likely(pages != NULL))
1706 				goto have_pages;
1707 			order--;
1708 		}
1709 
1710 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1711 		if (!pages) {
1712 			crypt_free_buffer_pages(cc, clone);
1713 			bio_put(clone);
1714 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1715 			order = 0;
1716 			goto retry;
1717 		}
1718 
1719 have_pages:
1720 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1721 		__bio_add_page(clone, pages, size_to_add, 0);
1722 		remaining_size -= size_to_add;
1723 	}
1724 
1725 	/* Allocate space for integrity tags */
1726 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1727 		crypt_free_buffer_pages(cc, clone);
1728 		bio_put(clone);
1729 		clone = NULL;
1730 	}
1731 
1732 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1733 		mutex_unlock(&cc->bio_alloc_lock);
1734 
1735 	return clone;
1736 }
1737 
1738 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1739 {
1740 	struct folio_iter fi;
1741 
1742 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1743 		bio_for_each_folio_all(fi, clone) {
1744 			if (folio_test_large(fi.folio))
1745 				folio_put(fi.folio);
1746 			else
1747 				mempool_free(&fi.folio->page, &cc->page_pool);
1748 		}
1749 	}
1750 }
1751 
1752 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1753 			  struct bio *bio, sector_t sector)
1754 {
1755 	io->cc = cc;
1756 	io->base_bio = bio;
1757 	io->sector = sector;
1758 	io->error = 0;
1759 	io->ctx.r.req = NULL;
1760 	io->integrity_metadata = NULL;
1761 	io->integrity_metadata_from_pool = false;
1762 	io->in_tasklet = false;
1763 	atomic_set(&io->io_pending, 0);
1764 }
1765 
1766 static void crypt_inc_pending(struct dm_crypt_io *io)
1767 {
1768 	atomic_inc(&io->io_pending);
1769 }
1770 
1771 static void kcryptd_io_bio_endio(struct work_struct *work)
1772 {
1773 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1774 
1775 	bio_endio(io->base_bio);
1776 }
1777 
1778 /*
1779  * One of the bios was finished. Check for completion of
1780  * the whole request and correctly clean up the buffer.
1781  */
1782 static void crypt_dec_pending(struct dm_crypt_io *io)
1783 {
1784 	struct crypt_config *cc = io->cc;
1785 	struct bio *base_bio = io->base_bio;
1786 	blk_status_t error = io->error;
1787 
1788 	if (!atomic_dec_and_test(&io->io_pending))
1789 		return;
1790 
1791 	if (io->ctx.r.req)
1792 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1793 
1794 	if (unlikely(io->integrity_metadata_from_pool))
1795 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1796 	else
1797 		kfree(io->integrity_metadata);
1798 
1799 	base_bio->bi_status = error;
1800 
1801 	/*
1802 	 * If we are running this function from our tasklet,
1803 	 * we can't call bio_endio() here, because it will call
1804 	 * clone_endio() from dm.c, which in turn will
1805 	 * free the current struct dm_crypt_io structure with
1806 	 * our tasklet. In this case we need to delay bio_endio()
1807 	 * execution to after the tasklet is done and dequeued.
1808 	 */
1809 	if (io->in_tasklet) {
1810 		INIT_WORK(&io->work, kcryptd_io_bio_endio);
1811 		queue_work(cc->io_queue, &io->work);
1812 		return;
1813 	}
1814 
1815 	bio_endio(base_bio);
1816 }
1817 
1818 /*
1819  * kcryptd/kcryptd_io:
1820  *
1821  * Needed because it would be very unwise to do decryption in an
1822  * interrupt context.
1823  *
1824  * kcryptd performs the actual encryption or decryption.
1825  *
1826  * kcryptd_io performs the IO submission.
1827  *
1828  * They must be separated as otherwise the final stages could be
1829  * starved by new requests which can block in the first stages due
1830  * to memory allocation.
1831  *
1832  * The work is done per CPU global for all dm-crypt instances.
1833  * They should not depend on each other and do not block.
1834  */
1835 static void crypt_endio(struct bio *clone)
1836 {
1837 	struct dm_crypt_io *io = clone->bi_private;
1838 	struct crypt_config *cc = io->cc;
1839 	unsigned int rw = bio_data_dir(clone);
1840 	blk_status_t error;
1841 
1842 	/*
1843 	 * free the processed pages
1844 	 */
1845 	if (rw == WRITE)
1846 		crypt_free_buffer_pages(cc, clone);
1847 
1848 	error = clone->bi_status;
1849 	bio_put(clone);
1850 
1851 	if (rw == READ && !error) {
1852 		kcryptd_queue_crypt(io);
1853 		return;
1854 	}
1855 
1856 	if (unlikely(error))
1857 		io->error = error;
1858 
1859 	crypt_dec_pending(io);
1860 }
1861 
1862 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1863 
1864 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1865 {
1866 	struct crypt_config *cc = io->cc;
1867 	struct bio *clone;
1868 
1869 	/*
1870 	 * We need the original biovec array in order to decrypt the whole bio
1871 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1872 	 * worry about the block layer modifying the biovec array; so leverage
1873 	 * bio_alloc_clone().
1874 	 */
1875 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1876 	if (!clone)
1877 		return 1;
1878 	clone->bi_private = io;
1879 	clone->bi_end_io = crypt_endio;
1880 
1881 	crypt_inc_pending(io);
1882 
1883 	clone->bi_iter.bi_sector = cc->start + io->sector;
1884 
1885 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1886 		crypt_dec_pending(io);
1887 		bio_put(clone);
1888 		return 1;
1889 	}
1890 
1891 	dm_submit_bio_remap(io->base_bio, clone);
1892 	return 0;
1893 }
1894 
1895 static void kcryptd_io_read_work(struct work_struct *work)
1896 {
1897 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1898 
1899 	crypt_inc_pending(io);
1900 	if (kcryptd_io_read(io, GFP_NOIO))
1901 		io->error = BLK_STS_RESOURCE;
1902 	crypt_dec_pending(io);
1903 }
1904 
1905 static void kcryptd_queue_read(struct dm_crypt_io *io)
1906 {
1907 	struct crypt_config *cc = io->cc;
1908 
1909 	INIT_WORK(&io->work, kcryptd_io_read_work);
1910 	queue_work(cc->io_queue, &io->work);
1911 }
1912 
1913 static void kcryptd_io_write(struct dm_crypt_io *io)
1914 {
1915 	struct bio *clone = io->ctx.bio_out;
1916 
1917 	dm_submit_bio_remap(io->base_bio, clone);
1918 }
1919 
1920 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1921 
1922 static int dmcrypt_write(void *data)
1923 {
1924 	struct crypt_config *cc = data;
1925 	struct dm_crypt_io *io;
1926 
1927 	while (1) {
1928 		struct rb_root write_tree;
1929 		struct blk_plug plug;
1930 
1931 		spin_lock_irq(&cc->write_thread_lock);
1932 continue_locked:
1933 
1934 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1935 			goto pop_from_list;
1936 
1937 		set_current_state(TASK_INTERRUPTIBLE);
1938 
1939 		spin_unlock_irq(&cc->write_thread_lock);
1940 
1941 		if (unlikely(kthread_should_stop())) {
1942 			set_current_state(TASK_RUNNING);
1943 			break;
1944 		}
1945 
1946 		schedule();
1947 
1948 		set_current_state(TASK_RUNNING);
1949 		spin_lock_irq(&cc->write_thread_lock);
1950 		goto continue_locked;
1951 
1952 pop_from_list:
1953 		write_tree = cc->write_tree;
1954 		cc->write_tree = RB_ROOT;
1955 		spin_unlock_irq(&cc->write_thread_lock);
1956 
1957 		BUG_ON(rb_parent(write_tree.rb_node));
1958 
1959 		/*
1960 		 * Note: we cannot walk the tree here with rb_next because
1961 		 * the structures may be freed when kcryptd_io_write is called.
1962 		 */
1963 		blk_start_plug(&plug);
1964 		do {
1965 			io = crypt_io_from_node(rb_first(&write_tree));
1966 			rb_erase(&io->rb_node, &write_tree);
1967 			kcryptd_io_write(io);
1968 			cond_resched();
1969 		} while (!RB_EMPTY_ROOT(&write_tree));
1970 		blk_finish_plug(&plug);
1971 	}
1972 	return 0;
1973 }
1974 
1975 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1976 {
1977 	struct bio *clone = io->ctx.bio_out;
1978 	struct crypt_config *cc = io->cc;
1979 	unsigned long flags;
1980 	sector_t sector;
1981 	struct rb_node **rbp, *parent;
1982 
1983 	if (unlikely(io->error)) {
1984 		crypt_free_buffer_pages(cc, clone);
1985 		bio_put(clone);
1986 		crypt_dec_pending(io);
1987 		return;
1988 	}
1989 
1990 	/* crypt_convert should have filled the clone bio */
1991 	BUG_ON(io->ctx.iter_out.bi_size);
1992 
1993 	clone->bi_iter.bi_sector = cc->start + io->sector;
1994 
1995 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1996 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1997 		dm_submit_bio_remap(io->base_bio, clone);
1998 		return;
1999 	}
2000 
2001 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2002 	if (RB_EMPTY_ROOT(&cc->write_tree))
2003 		wake_up_process(cc->write_thread);
2004 	rbp = &cc->write_tree.rb_node;
2005 	parent = NULL;
2006 	sector = io->sector;
2007 	while (*rbp) {
2008 		parent = *rbp;
2009 		if (sector < crypt_io_from_node(parent)->sector)
2010 			rbp = &(*rbp)->rb_left;
2011 		else
2012 			rbp = &(*rbp)->rb_right;
2013 	}
2014 	rb_link_node(&io->rb_node, parent, rbp);
2015 	rb_insert_color(&io->rb_node, &cc->write_tree);
2016 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2017 }
2018 
2019 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2020 				       struct convert_context *ctx)
2021 
2022 {
2023 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2024 		return false;
2025 
2026 	/*
2027 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2028 	 * constraints so they do not need to be issued inline by
2029 	 * kcryptd_crypt_write_convert().
2030 	 */
2031 	switch (bio_op(ctx->bio_in)) {
2032 	case REQ_OP_WRITE:
2033 	case REQ_OP_WRITE_ZEROES:
2034 		return true;
2035 	default:
2036 		return false;
2037 	}
2038 }
2039 
2040 static void kcryptd_crypt_write_continue(struct work_struct *work)
2041 {
2042 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2043 	struct crypt_config *cc = io->cc;
2044 	struct convert_context *ctx = &io->ctx;
2045 	int crypt_finished;
2046 	sector_t sector = io->sector;
2047 	blk_status_t r;
2048 
2049 	wait_for_completion(&ctx->restart);
2050 	reinit_completion(&ctx->restart);
2051 
2052 	r = crypt_convert(cc, &io->ctx, true, false);
2053 	if (r)
2054 		io->error = r;
2055 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2056 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2057 		/* Wait for completion signaled by kcryptd_async_done() */
2058 		wait_for_completion(&ctx->restart);
2059 		crypt_finished = 1;
2060 	}
2061 
2062 	/* Encryption was already finished, submit io now */
2063 	if (crypt_finished) {
2064 		kcryptd_crypt_write_io_submit(io, 0);
2065 		io->sector = sector;
2066 	}
2067 
2068 	crypt_dec_pending(io);
2069 }
2070 
2071 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2072 {
2073 	struct crypt_config *cc = io->cc;
2074 	struct convert_context *ctx = &io->ctx;
2075 	struct bio *clone;
2076 	int crypt_finished;
2077 	sector_t sector = io->sector;
2078 	blk_status_t r;
2079 
2080 	/*
2081 	 * Prevent io from disappearing until this function completes.
2082 	 */
2083 	crypt_inc_pending(io);
2084 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2085 
2086 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2087 	if (unlikely(!clone)) {
2088 		io->error = BLK_STS_IOERR;
2089 		goto dec;
2090 	}
2091 
2092 	io->ctx.bio_out = clone;
2093 	io->ctx.iter_out = clone->bi_iter;
2094 
2095 	sector += bio_sectors(clone);
2096 
2097 	crypt_inc_pending(io);
2098 	r = crypt_convert(cc, ctx,
2099 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2100 	/*
2101 	 * Crypto API backlogged the request, because its queue was full
2102 	 * and we're in softirq context, so continue from a workqueue
2103 	 * (TODO: is it actually possible to be in softirq in the write path?)
2104 	 */
2105 	if (r == BLK_STS_DEV_RESOURCE) {
2106 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2107 		queue_work(cc->crypt_queue, &io->work);
2108 		return;
2109 	}
2110 	if (r)
2111 		io->error = r;
2112 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2113 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2114 		/* Wait for completion signaled by kcryptd_async_done() */
2115 		wait_for_completion(&ctx->restart);
2116 		crypt_finished = 1;
2117 	}
2118 
2119 	/* Encryption was already finished, submit io now */
2120 	if (crypt_finished) {
2121 		kcryptd_crypt_write_io_submit(io, 0);
2122 		io->sector = sector;
2123 	}
2124 
2125 dec:
2126 	crypt_dec_pending(io);
2127 }
2128 
2129 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2130 {
2131 	crypt_dec_pending(io);
2132 }
2133 
2134 static void kcryptd_crypt_read_continue(struct work_struct *work)
2135 {
2136 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2137 	struct crypt_config *cc = io->cc;
2138 	blk_status_t r;
2139 
2140 	wait_for_completion(&io->ctx.restart);
2141 	reinit_completion(&io->ctx.restart);
2142 
2143 	r = crypt_convert(cc, &io->ctx, true, false);
2144 	if (r)
2145 		io->error = r;
2146 
2147 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2148 		kcryptd_crypt_read_done(io);
2149 
2150 	crypt_dec_pending(io);
2151 }
2152 
2153 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2154 {
2155 	struct crypt_config *cc = io->cc;
2156 	blk_status_t r;
2157 
2158 	crypt_inc_pending(io);
2159 
2160 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2161 			   io->sector);
2162 
2163 	r = crypt_convert(cc, &io->ctx,
2164 			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2165 	/*
2166 	 * Crypto API backlogged the request, because its queue was full
2167 	 * and we're in softirq context, so continue from a workqueue
2168 	 */
2169 	if (r == BLK_STS_DEV_RESOURCE) {
2170 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2171 		queue_work(cc->crypt_queue, &io->work);
2172 		return;
2173 	}
2174 	if (r)
2175 		io->error = r;
2176 
2177 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2178 		kcryptd_crypt_read_done(io);
2179 
2180 	crypt_dec_pending(io);
2181 }
2182 
2183 static void kcryptd_async_done(void *data, int error)
2184 {
2185 	struct dm_crypt_request *dmreq = data;
2186 	struct convert_context *ctx = dmreq->ctx;
2187 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2188 	struct crypt_config *cc = io->cc;
2189 
2190 	/*
2191 	 * A request from crypto driver backlog is going to be processed now,
2192 	 * finish the completion and continue in crypt_convert().
2193 	 * (Callback will be called for the second time for this request.)
2194 	 */
2195 	if (error == -EINPROGRESS) {
2196 		complete(&ctx->restart);
2197 		return;
2198 	}
2199 
2200 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2201 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2202 
2203 	if (error == -EBADMSG) {
2204 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2205 
2206 		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2207 			    ctx->bio_in->bi_bdev, s);
2208 		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2209 				 ctx->bio_in, s, 0);
2210 		io->error = BLK_STS_PROTECTION;
2211 	} else if (error < 0)
2212 		io->error = BLK_STS_IOERR;
2213 
2214 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2215 
2216 	if (!atomic_dec_and_test(&ctx->cc_pending))
2217 		return;
2218 
2219 	/*
2220 	 * The request is fully completed: for inline writes, let
2221 	 * kcryptd_crypt_write_convert() do the IO submission.
2222 	 */
2223 	if (bio_data_dir(io->base_bio) == READ) {
2224 		kcryptd_crypt_read_done(io);
2225 		return;
2226 	}
2227 
2228 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2229 		complete(&ctx->restart);
2230 		return;
2231 	}
2232 
2233 	kcryptd_crypt_write_io_submit(io, 1);
2234 }
2235 
2236 static void kcryptd_crypt(struct work_struct *work)
2237 {
2238 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2239 
2240 	if (bio_data_dir(io->base_bio) == READ)
2241 		kcryptd_crypt_read_convert(io);
2242 	else
2243 		kcryptd_crypt_write_convert(io);
2244 }
2245 
2246 static void kcryptd_crypt_tasklet(unsigned long work)
2247 {
2248 	kcryptd_crypt((struct work_struct *)work);
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 			io->in_tasklet = true;
2264 			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2265 			tasklet_schedule(&io->tasklet);
2266 			return;
2267 		}
2268 
2269 		kcryptd_crypt(&io->work);
2270 		return;
2271 	}
2272 
2273 	INIT_WORK(&io->work, kcryptd_crypt);
2274 	queue_work(cc->crypt_queue, &io->work);
2275 }
2276 
2277 static void crypt_free_tfms_aead(struct crypt_config *cc)
2278 {
2279 	if (!cc->cipher_tfm.tfms_aead)
2280 		return;
2281 
2282 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2283 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2284 		cc->cipher_tfm.tfms_aead[0] = NULL;
2285 	}
2286 
2287 	kfree(cc->cipher_tfm.tfms_aead);
2288 	cc->cipher_tfm.tfms_aead = NULL;
2289 }
2290 
2291 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2292 {
2293 	unsigned int i;
2294 
2295 	if (!cc->cipher_tfm.tfms)
2296 		return;
2297 
2298 	for (i = 0; i < cc->tfms_count; i++)
2299 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2300 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2301 			cc->cipher_tfm.tfms[i] = NULL;
2302 		}
2303 
2304 	kfree(cc->cipher_tfm.tfms);
2305 	cc->cipher_tfm.tfms = NULL;
2306 }
2307 
2308 static void crypt_free_tfms(struct crypt_config *cc)
2309 {
2310 	if (crypt_integrity_aead(cc))
2311 		crypt_free_tfms_aead(cc);
2312 	else
2313 		crypt_free_tfms_skcipher(cc);
2314 }
2315 
2316 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2317 {
2318 	unsigned int i;
2319 	int err;
2320 
2321 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2322 				      sizeof(struct crypto_skcipher *),
2323 				      GFP_KERNEL);
2324 	if (!cc->cipher_tfm.tfms)
2325 		return -ENOMEM;
2326 
2327 	for (i = 0; i < cc->tfms_count; i++) {
2328 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2329 						CRYPTO_ALG_ALLOCATES_MEMORY);
2330 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2331 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2332 			crypt_free_tfms(cc);
2333 			return err;
2334 		}
2335 	}
2336 
2337 	/*
2338 	 * dm-crypt performance can vary greatly depending on which crypto
2339 	 * algorithm implementation is used.  Help people debug performance
2340 	 * problems by logging the ->cra_driver_name.
2341 	 */
2342 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2343 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2344 	return 0;
2345 }
2346 
2347 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2348 {
2349 	int err;
2350 
2351 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2352 	if (!cc->cipher_tfm.tfms)
2353 		return -ENOMEM;
2354 
2355 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2356 						CRYPTO_ALG_ALLOCATES_MEMORY);
2357 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2358 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2359 		crypt_free_tfms(cc);
2360 		return err;
2361 	}
2362 
2363 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2364 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2365 	return 0;
2366 }
2367 
2368 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2369 {
2370 	if (crypt_integrity_aead(cc))
2371 		return crypt_alloc_tfms_aead(cc, ciphermode);
2372 	else
2373 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2374 }
2375 
2376 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2377 {
2378 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2379 }
2380 
2381 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2382 {
2383 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2384 }
2385 
2386 /*
2387  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2388  * the key must be for some reason in special format.
2389  * This funcion converts cc->key to this special format.
2390  */
2391 static void crypt_copy_authenckey(char *p, const void *key,
2392 				  unsigned int enckeylen, unsigned int authkeylen)
2393 {
2394 	struct crypto_authenc_key_param *param;
2395 	struct rtattr *rta;
2396 
2397 	rta = (struct rtattr *)p;
2398 	param = RTA_DATA(rta);
2399 	param->enckeylen = cpu_to_be32(enckeylen);
2400 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2401 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2402 	p += RTA_SPACE(sizeof(*param));
2403 	memcpy(p, key + enckeylen, authkeylen);
2404 	p += authkeylen;
2405 	memcpy(p, key, enckeylen);
2406 }
2407 
2408 static int crypt_setkey(struct crypt_config *cc)
2409 {
2410 	unsigned int subkey_size;
2411 	int err = 0, i, r;
2412 
2413 	/* Ignore extra keys (which are used for IV etc) */
2414 	subkey_size = crypt_subkey_size(cc);
2415 
2416 	if (crypt_integrity_hmac(cc)) {
2417 		if (subkey_size < cc->key_mac_size)
2418 			return -EINVAL;
2419 
2420 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2421 				      subkey_size - cc->key_mac_size,
2422 				      cc->key_mac_size);
2423 	}
2424 
2425 	for (i = 0; i < cc->tfms_count; i++) {
2426 		if (crypt_integrity_hmac(cc))
2427 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2428 				cc->authenc_key, crypt_authenckey_size(cc));
2429 		else if (crypt_integrity_aead(cc))
2430 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2431 					       cc->key + (i * subkey_size),
2432 					       subkey_size);
2433 		else
2434 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2435 						   cc->key + (i * subkey_size),
2436 						   subkey_size);
2437 		if (r)
2438 			err = r;
2439 	}
2440 
2441 	if (crypt_integrity_hmac(cc))
2442 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2443 
2444 	return err;
2445 }
2446 
2447 #ifdef CONFIG_KEYS
2448 
2449 static bool contains_whitespace(const char *str)
2450 {
2451 	while (*str)
2452 		if (isspace(*str++))
2453 			return true;
2454 	return false;
2455 }
2456 
2457 static int set_key_user(struct crypt_config *cc, struct key *key)
2458 {
2459 	const struct user_key_payload *ukp;
2460 
2461 	ukp = user_key_payload_locked(key);
2462 	if (!ukp)
2463 		return -EKEYREVOKED;
2464 
2465 	if (cc->key_size != ukp->datalen)
2466 		return -EINVAL;
2467 
2468 	memcpy(cc->key, ukp->data, cc->key_size);
2469 
2470 	return 0;
2471 }
2472 
2473 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2474 {
2475 	const struct encrypted_key_payload *ekp;
2476 
2477 	ekp = key->payload.data[0];
2478 	if (!ekp)
2479 		return -EKEYREVOKED;
2480 
2481 	if (cc->key_size != ekp->decrypted_datalen)
2482 		return -EINVAL;
2483 
2484 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2485 
2486 	return 0;
2487 }
2488 
2489 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2490 {
2491 	const struct trusted_key_payload *tkp;
2492 
2493 	tkp = key->payload.data[0];
2494 	if (!tkp)
2495 		return -EKEYREVOKED;
2496 
2497 	if (cc->key_size != tkp->key_len)
2498 		return -EINVAL;
2499 
2500 	memcpy(cc->key, tkp->key, cc->key_size);
2501 
2502 	return 0;
2503 }
2504 
2505 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2506 {
2507 	char *new_key_string, *key_desc;
2508 	int ret;
2509 	struct key_type *type;
2510 	struct key *key;
2511 	int (*set_key)(struct crypt_config *cc, struct key *key);
2512 
2513 	/*
2514 	 * Reject key_string with whitespace. dm core currently lacks code for
2515 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2516 	 */
2517 	if (contains_whitespace(key_string)) {
2518 		DMERR("whitespace chars not allowed in key string");
2519 		return -EINVAL;
2520 	}
2521 
2522 	/* look for next ':' separating key_type from key_description */
2523 	key_desc = strchr(key_string, ':');
2524 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2525 		return -EINVAL;
2526 
2527 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2528 		type = &key_type_logon;
2529 		set_key = set_key_user;
2530 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2531 		type = &key_type_user;
2532 		set_key = set_key_user;
2533 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2534 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2535 		type = &key_type_encrypted;
2536 		set_key = set_key_encrypted;
2537 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2538 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2539 		type = &key_type_trusted;
2540 		set_key = set_key_trusted;
2541 	} else {
2542 		return -EINVAL;
2543 	}
2544 
2545 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2546 	if (!new_key_string)
2547 		return -ENOMEM;
2548 
2549 	key = request_key(type, key_desc + 1, NULL);
2550 	if (IS_ERR(key)) {
2551 		kfree_sensitive(new_key_string);
2552 		return PTR_ERR(key);
2553 	}
2554 
2555 	down_read(&key->sem);
2556 
2557 	ret = set_key(cc, key);
2558 	if (ret < 0) {
2559 		up_read(&key->sem);
2560 		key_put(key);
2561 		kfree_sensitive(new_key_string);
2562 		return ret;
2563 	}
2564 
2565 	up_read(&key->sem);
2566 	key_put(key);
2567 
2568 	/* clear the flag since following operations may invalidate previously valid key */
2569 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2570 
2571 	ret = crypt_setkey(cc);
2572 
2573 	if (!ret) {
2574 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2575 		kfree_sensitive(cc->key_string);
2576 		cc->key_string = new_key_string;
2577 	} else
2578 		kfree_sensitive(new_key_string);
2579 
2580 	return ret;
2581 }
2582 
2583 static int get_key_size(char **key_string)
2584 {
2585 	char *colon, dummy;
2586 	int ret;
2587 
2588 	if (*key_string[0] != ':')
2589 		return strlen(*key_string) >> 1;
2590 
2591 	/* look for next ':' in key string */
2592 	colon = strpbrk(*key_string + 1, ":");
2593 	if (!colon)
2594 		return -EINVAL;
2595 
2596 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2597 		return -EINVAL;
2598 
2599 	*key_string = colon;
2600 
2601 	/* remaining key string should be :<logon|user>:<key_desc> */
2602 
2603 	return ret;
2604 }
2605 
2606 #else
2607 
2608 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2609 {
2610 	return -EINVAL;
2611 }
2612 
2613 static int get_key_size(char **key_string)
2614 {
2615 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2616 }
2617 
2618 #endif /* CONFIG_KEYS */
2619 
2620 static int crypt_set_key(struct crypt_config *cc, char *key)
2621 {
2622 	int r = -EINVAL;
2623 	int key_string_len = strlen(key);
2624 
2625 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2626 	if (!cc->key_size && strcmp(key, "-"))
2627 		goto out;
2628 
2629 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2630 	if (key[0] == ':') {
2631 		r = crypt_set_keyring_key(cc, key + 1);
2632 		goto out;
2633 	}
2634 
2635 	/* clear the flag since following operations may invalidate previously valid key */
2636 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2637 
2638 	/* wipe references to any kernel keyring key */
2639 	kfree_sensitive(cc->key_string);
2640 	cc->key_string = NULL;
2641 
2642 	/* Decode key from its hex representation. */
2643 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2644 		goto out;
2645 
2646 	r = crypt_setkey(cc);
2647 	if (!r)
2648 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2649 
2650 out:
2651 	/* Hex key string not needed after here, so wipe it. */
2652 	memset(key, '0', key_string_len);
2653 
2654 	return r;
2655 }
2656 
2657 static int crypt_wipe_key(struct crypt_config *cc)
2658 {
2659 	int r;
2660 
2661 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2662 	get_random_bytes(&cc->key, cc->key_size);
2663 
2664 	/* Wipe IV private keys */
2665 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2666 		r = cc->iv_gen_ops->wipe(cc);
2667 		if (r)
2668 			return r;
2669 	}
2670 
2671 	kfree_sensitive(cc->key_string);
2672 	cc->key_string = NULL;
2673 	r = crypt_setkey(cc);
2674 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2675 
2676 	return r;
2677 }
2678 
2679 static void crypt_calculate_pages_per_client(void)
2680 {
2681 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2682 
2683 	if (!dm_crypt_clients_n)
2684 		return;
2685 
2686 	pages /= dm_crypt_clients_n;
2687 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2688 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2689 	dm_crypt_pages_per_client = pages;
2690 }
2691 
2692 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2693 {
2694 	struct crypt_config *cc = pool_data;
2695 	struct page *page;
2696 
2697 	/*
2698 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2699 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2700 	 * but avoids potential spinlock contention of an exact result.
2701 	 */
2702 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2703 	    likely(gfp_mask & __GFP_NORETRY))
2704 		return NULL;
2705 
2706 	page = alloc_page(gfp_mask);
2707 	if (likely(page != NULL))
2708 		percpu_counter_add(&cc->n_allocated_pages, 1);
2709 
2710 	return page;
2711 }
2712 
2713 static void crypt_page_free(void *page, void *pool_data)
2714 {
2715 	struct crypt_config *cc = pool_data;
2716 
2717 	__free_page(page);
2718 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2719 }
2720 
2721 static void crypt_dtr(struct dm_target *ti)
2722 {
2723 	struct crypt_config *cc = ti->private;
2724 
2725 	ti->private = NULL;
2726 
2727 	if (!cc)
2728 		return;
2729 
2730 	if (cc->write_thread)
2731 		kthread_stop(cc->write_thread);
2732 
2733 	if (cc->io_queue)
2734 		destroy_workqueue(cc->io_queue);
2735 	if (cc->crypt_queue)
2736 		destroy_workqueue(cc->crypt_queue);
2737 
2738 	crypt_free_tfms(cc);
2739 
2740 	bioset_exit(&cc->bs);
2741 
2742 	mempool_exit(&cc->page_pool);
2743 	mempool_exit(&cc->req_pool);
2744 	mempool_exit(&cc->tag_pool);
2745 
2746 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2747 	percpu_counter_destroy(&cc->n_allocated_pages);
2748 
2749 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2750 		cc->iv_gen_ops->dtr(cc);
2751 
2752 	if (cc->dev)
2753 		dm_put_device(ti, cc->dev);
2754 
2755 	kfree_sensitive(cc->cipher_string);
2756 	kfree_sensitive(cc->key_string);
2757 	kfree_sensitive(cc->cipher_auth);
2758 	kfree_sensitive(cc->authenc_key);
2759 
2760 	mutex_destroy(&cc->bio_alloc_lock);
2761 
2762 	/* Must zero key material before freeing */
2763 	kfree_sensitive(cc);
2764 
2765 	spin_lock(&dm_crypt_clients_lock);
2766 	WARN_ON(!dm_crypt_clients_n);
2767 	dm_crypt_clients_n--;
2768 	crypt_calculate_pages_per_client();
2769 	spin_unlock(&dm_crypt_clients_lock);
2770 
2771 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2772 }
2773 
2774 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2775 {
2776 	struct crypt_config *cc = ti->private;
2777 
2778 	if (crypt_integrity_aead(cc))
2779 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2780 	else
2781 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2782 
2783 	if (cc->iv_size)
2784 		/* at least a 64 bit sector number should fit in our buffer */
2785 		cc->iv_size = max(cc->iv_size,
2786 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2787 	else if (ivmode) {
2788 		DMWARN("Selected cipher does not support IVs");
2789 		ivmode = NULL;
2790 	}
2791 
2792 	/* Choose ivmode, see comments at iv code. */
2793 	if (ivmode == NULL)
2794 		cc->iv_gen_ops = NULL;
2795 	else if (strcmp(ivmode, "plain") == 0)
2796 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2797 	else if (strcmp(ivmode, "plain64") == 0)
2798 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2799 	else if (strcmp(ivmode, "plain64be") == 0)
2800 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2801 	else if (strcmp(ivmode, "essiv") == 0)
2802 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2803 	else if (strcmp(ivmode, "benbi") == 0)
2804 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2805 	else if (strcmp(ivmode, "null") == 0)
2806 		cc->iv_gen_ops = &crypt_iv_null_ops;
2807 	else if (strcmp(ivmode, "eboiv") == 0)
2808 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2809 	else if (strcmp(ivmode, "elephant") == 0) {
2810 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2811 		cc->key_parts = 2;
2812 		cc->key_extra_size = cc->key_size / 2;
2813 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2814 			return -EINVAL;
2815 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2816 	} else if (strcmp(ivmode, "lmk") == 0) {
2817 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2818 		/*
2819 		 * Version 2 and 3 is recognised according
2820 		 * to length of provided multi-key string.
2821 		 * If present (version 3), last key is used as IV seed.
2822 		 * All keys (including IV seed) are always the same size.
2823 		 */
2824 		if (cc->key_size % cc->key_parts) {
2825 			cc->key_parts++;
2826 			cc->key_extra_size = cc->key_size / cc->key_parts;
2827 		}
2828 	} else if (strcmp(ivmode, "tcw") == 0) {
2829 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2830 		cc->key_parts += 2; /* IV + whitening */
2831 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2832 	} else if (strcmp(ivmode, "random") == 0) {
2833 		cc->iv_gen_ops = &crypt_iv_random_ops;
2834 		/* Need storage space in integrity fields. */
2835 		cc->integrity_iv_size = cc->iv_size;
2836 	} else {
2837 		ti->error = "Invalid IV mode";
2838 		return -EINVAL;
2839 	}
2840 
2841 	return 0;
2842 }
2843 
2844 /*
2845  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2846  * The HMAC is needed to calculate tag size (HMAC digest size).
2847  * This should be probably done by crypto-api calls (once available...)
2848  */
2849 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2850 {
2851 	char *start, *end, *mac_alg = NULL;
2852 	struct crypto_ahash *mac;
2853 
2854 	if (!strstarts(cipher_api, "authenc("))
2855 		return 0;
2856 
2857 	start = strchr(cipher_api, '(');
2858 	end = strchr(cipher_api, ',');
2859 	if (!start || !end || ++start > end)
2860 		return -EINVAL;
2861 
2862 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2863 	if (!mac_alg)
2864 		return -ENOMEM;
2865 	strncpy(mac_alg, start, end - start);
2866 
2867 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2868 	kfree(mac_alg);
2869 
2870 	if (IS_ERR(mac))
2871 		return PTR_ERR(mac);
2872 
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 		cc->tfms_count = 64;
2924 
2925 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2926 		if (!*ivopts) {
2927 			ti->error = "Digest algorithm missing for ESSIV mode";
2928 			return -EINVAL;
2929 		}
2930 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2931 			       cipher_api, *ivopts);
2932 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2933 			ti->error = "Cannot allocate cipher string";
2934 			return -ENOMEM;
2935 		}
2936 		cipher_api = buf;
2937 	}
2938 
2939 	cc->key_parts = cc->tfms_count;
2940 
2941 	/* Allocate cipher */
2942 	ret = crypt_alloc_tfms(cc, cipher_api);
2943 	if (ret < 0) {
2944 		ti->error = "Error allocating crypto tfm";
2945 		return ret;
2946 	}
2947 
2948 	if (crypt_integrity_aead(cc))
2949 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2950 	else
2951 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2952 
2953 	return 0;
2954 }
2955 
2956 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2957 				char **ivmode, char **ivopts)
2958 {
2959 	struct crypt_config *cc = ti->private;
2960 	char *tmp, *cipher, *chainmode, *keycount;
2961 	char *cipher_api = NULL;
2962 	int ret = -EINVAL;
2963 	char dummy;
2964 
2965 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2966 		ti->error = "Bad cipher specification";
2967 		return -EINVAL;
2968 	}
2969 
2970 	/*
2971 	 * Legacy dm-crypt cipher specification
2972 	 * cipher[:keycount]-mode-iv:ivopts
2973 	 */
2974 	tmp = cipher_in;
2975 	keycount = strsep(&tmp, "-");
2976 	cipher = strsep(&keycount, ":");
2977 
2978 	if (!keycount)
2979 		cc->tfms_count = 1;
2980 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2981 		 !is_power_of_2(cc->tfms_count)) {
2982 		ti->error = "Bad cipher key count specification";
2983 		return -EINVAL;
2984 	}
2985 	cc->key_parts = cc->tfms_count;
2986 
2987 	chainmode = strsep(&tmp, "-");
2988 	*ivmode = strsep(&tmp, ":");
2989 	*ivopts = tmp;
2990 
2991 	/*
2992 	 * For compatibility with the original dm-crypt mapping format, if
2993 	 * only the cipher name is supplied, use cbc-plain.
2994 	 */
2995 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2996 		chainmode = "cbc";
2997 		*ivmode = "plain";
2998 	}
2999 
3000 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3001 		ti->error = "IV mechanism required";
3002 		return -EINVAL;
3003 	}
3004 
3005 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3006 	if (!cipher_api)
3007 		goto bad_mem;
3008 
3009 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3010 		if (!*ivopts) {
3011 			ti->error = "Digest algorithm missing for ESSIV mode";
3012 			kfree(cipher_api);
3013 			return -EINVAL;
3014 		}
3015 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3016 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3017 	} else {
3018 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3019 			       "%s(%s)", chainmode, cipher);
3020 	}
3021 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3022 		kfree(cipher_api);
3023 		goto bad_mem;
3024 	}
3025 
3026 	/* Allocate cipher */
3027 	ret = crypt_alloc_tfms(cc, cipher_api);
3028 	if (ret < 0) {
3029 		ti->error = "Error allocating crypto tfm";
3030 		kfree(cipher_api);
3031 		return ret;
3032 	}
3033 	kfree(cipher_api);
3034 
3035 	return 0;
3036 bad_mem:
3037 	ti->error = "Cannot allocate cipher strings";
3038 	return -ENOMEM;
3039 }
3040 
3041 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3042 {
3043 	struct crypt_config *cc = ti->private;
3044 	char *ivmode = NULL, *ivopts = NULL;
3045 	int ret;
3046 
3047 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3048 	if (!cc->cipher_string) {
3049 		ti->error = "Cannot allocate cipher strings";
3050 		return -ENOMEM;
3051 	}
3052 
3053 	if (strstarts(cipher_in, "capi:"))
3054 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3055 	else
3056 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3057 	if (ret)
3058 		return ret;
3059 
3060 	/* Initialize IV */
3061 	ret = crypt_ctr_ivmode(ti, ivmode);
3062 	if (ret < 0)
3063 		return ret;
3064 
3065 	/* Initialize and set key */
3066 	ret = crypt_set_key(cc, key);
3067 	if (ret < 0) {
3068 		ti->error = "Error decoding and setting key";
3069 		return ret;
3070 	}
3071 
3072 	/* Allocate IV */
3073 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3074 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3075 		if (ret < 0) {
3076 			ti->error = "Error creating IV";
3077 			return ret;
3078 		}
3079 	}
3080 
3081 	/* Initialize IV (set keys for ESSIV etc) */
3082 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3083 		ret = cc->iv_gen_ops->init(cc);
3084 		if (ret < 0) {
3085 			ti->error = "Error initialising IV";
3086 			return ret;
3087 		}
3088 	}
3089 
3090 	/* wipe the kernel key payload copy */
3091 	if (cc->key_string)
3092 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3093 
3094 	return ret;
3095 }
3096 
3097 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3098 {
3099 	struct crypt_config *cc = ti->private;
3100 	struct dm_arg_set as;
3101 	static const struct dm_arg _args[] = {
3102 		{0, 8, "Invalid number of feature args"},
3103 	};
3104 	unsigned int opt_params, val;
3105 	const char *opt_string, *sval;
3106 	char dummy;
3107 	int ret;
3108 
3109 	/* Optional parameters */
3110 	as.argc = argc;
3111 	as.argv = argv;
3112 
3113 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3114 	if (ret)
3115 		return ret;
3116 
3117 	while (opt_params--) {
3118 		opt_string = dm_shift_arg(&as);
3119 		if (!opt_string) {
3120 			ti->error = "Not enough feature arguments";
3121 			return -EINVAL;
3122 		}
3123 
3124 		if (!strcasecmp(opt_string, "allow_discards"))
3125 			ti->num_discard_bios = 1;
3126 
3127 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3128 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3129 
3130 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3131 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3132 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3133 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3134 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3135 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3136 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3137 			if (val == 0 || val > MAX_TAG_SIZE) {
3138 				ti->error = "Invalid integrity arguments";
3139 				return -EINVAL;
3140 			}
3141 			cc->on_disk_tag_size = val;
3142 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3143 			if (!strcasecmp(sval, "aead")) {
3144 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3145 			} else  if (strcasecmp(sval, "none")) {
3146 				ti->error = "Unknown integrity profile";
3147 				return -EINVAL;
3148 			}
3149 
3150 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3151 			if (!cc->cipher_auth)
3152 				return -ENOMEM;
3153 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3154 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3155 			    cc->sector_size > 4096 ||
3156 			    (cc->sector_size & (cc->sector_size - 1))) {
3157 				ti->error = "Invalid feature value for sector_size";
3158 				return -EINVAL;
3159 			}
3160 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3161 				ti->error = "Device size is not multiple of sector_size feature";
3162 				return -EINVAL;
3163 			}
3164 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3165 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3166 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3167 		else {
3168 			ti->error = "Invalid feature arguments";
3169 			return -EINVAL;
3170 		}
3171 	}
3172 
3173 	return 0;
3174 }
3175 
3176 #ifdef CONFIG_BLK_DEV_ZONED
3177 static int crypt_report_zones(struct dm_target *ti,
3178 		struct dm_report_zones_args *args, unsigned int nr_zones)
3179 {
3180 	struct crypt_config *cc = ti->private;
3181 
3182 	return dm_report_zones(cc->dev->bdev, cc->start,
3183 			cc->start + dm_target_offset(ti, args->next_sector),
3184 			args, nr_zones);
3185 }
3186 #else
3187 #define crypt_report_zones NULL
3188 #endif
3189 
3190 /*
3191  * Construct an encryption mapping:
3192  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3193  */
3194 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3195 {
3196 	struct crypt_config *cc;
3197 	const char *devname = dm_table_device_name(ti->table);
3198 	int key_size;
3199 	unsigned int align_mask;
3200 	unsigned long long tmpll;
3201 	int ret;
3202 	size_t iv_size_padding, additional_req_size;
3203 	char dummy;
3204 
3205 	if (argc < 5) {
3206 		ti->error = "Not enough arguments";
3207 		return -EINVAL;
3208 	}
3209 
3210 	key_size = get_key_size(&argv[1]);
3211 	if (key_size < 0) {
3212 		ti->error = "Cannot parse key size";
3213 		return -EINVAL;
3214 	}
3215 
3216 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3217 	if (!cc) {
3218 		ti->error = "Cannot allocate encryption context";
3219 		return -ENOMEM;
3220 	}
3221 	cc->key_size = key_size;
3222 	cc->sector_size = (1 << SECTOR_SHIFT);
3223 	cc->sector_shift = 0;
3224 
3225 	ti->private = cc;
3226 
3227 	spin_lock(&dm_crypt_clients_lock);
3228 	dm_crypt_clients_n++;
3229 	crypt_calculate_pages_per_client();
3230 	spin_unlock(&dm_crypt_clients_lock);
3231 
3232 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3233 	if (ret < 0)
3234 		goto bad;
3235 
3236 	/* Optional parameters need to be read before cipher constructor */
3237 	if (argc > 5) {
3238 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3239 		if (ret)
3240 			goto bad;
3241 	}
3242 
3243 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3244 	if (ret < 0)
3245 		goto bad;
3246 
3247 	if (crypt_integrity_aead(cc)) {
3248 		cc->dmreq_start = sizeof(struct aead_request);
3249 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3250 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3251 	} else {
3252 		cc->dmreq_start = sizeof(struct skcipher_request);
3253 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3254 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3255 	}
3256 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3257 
3258 	if (align_mask < CRYPTO_MINALIGN) {
3259 		/* Allocate the padding exactly */
3260 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3261 				& align_mask;
3262 	} else {
3263 		/*
3264 		 * If the cipher requires greater alignment than kmalloc
3265 		 * alignment, we don't know the exact position of the
3266 		 * initialization vector. We must assume worst case.
3267 		 */
3268 		iv_size_padding = align_mask;
3269 	}
3270 
3271 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3272 	additional_req_size = sizeof(struct dm_crypt_request) +
3273 		iv_size_padding + cc->iv_size +
3274 		cc->iv_size +
3275 		sizeof(uint64_t) +
3276 		sizeof(unsigned int);
3277 
3278 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3279 	if (ret) {
3280 		ti->error = "Cannot allocate crypt request mempool";
3281 		goto bad;
3282 	}
3283 
3284 	cc->per_bio_data_size = ti->per_io_data_size =
3285 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3286 		      ARCH_DMA_MINALIGN);
3287 
3288 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3289 	if (ret) {
3290 		ti->error = "Cannot allocate page mempool";
3291 		goto bad;
3292 	}
3293 
3294 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3295 	if (ret) {
3296 		ti->error = "Cannot allocate crypt bioset";
3297 		goto bad;
3298 	}
3299 
3300 	mutex_init(&cc->bio_alloc_lock);
3301 
3302 	ret = -EINVAL;
3303 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3304 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3305 		ti->error = "Invalid iv_offset sector";
3306 		goto bad;
3307 	}
3308 	cc->iv_offset = tmpll;
3309 
3310 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3311 	if (ret) {
3312 		ti->error = "Device lookup failed";
3313 		goto bad;
3314 	}
3315 
3316 	ret = -EINVAL;
3317 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3318 		ti->error = "Invalid device sector";
3319 		goto bad;
3320 	}
3321 	cc->start = tmpll;
3322 
3323 	if (bdev_is_zoned(cc->dev->bdev)) {
3324 		/*
3325 		 * For zoned block devices, we need to preserve the issuer write
3326 		 * ordering. To do so, disable write workqueues and force inline
3327 		 * encryption completion.
3328 		 */
3329 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3330 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3331 
3332 		/*
3333 		 * All zone append writes to a zone of a zoned block device will
3334 		 * have the same BIO sector, the start of the zone. When the
3335 		 * cypher IV mode uses sector values, all data targeting a
3336 		 * zone will be encrypted using the first sector numbers of the
3337 		 * zone. This will not result in write errors but will
3338 		 * cause most reads to fail as reads will use the sector values
3339 		 * for the actual data locations, resulting in IV mismatch.
3340 		 * To avoid this problem, ask DM core to emulate zone append
3341 		 * operations with regular writes.
3342 		 */
3343 		DMDEBUG("Zone append operations will be emulated");
3344 		ti->emulate_zone_append = true;
3345 	}
3346 
3347 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3348 		ret = crypt_integrity_ctr(cc, ti);
3349 		if (ret)
3350 			goto bad;
3351 
3352 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3353 		if (!cc->tag_pool_max_sectors)
3354 			cc->tag_pool_max_sectors = 1;
3355 
3356 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3357 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3358 		if (ret) {
3359 			ti->error = "Cannot allocate integrity tags mempool";
3360 			goto bad;
3361 		}
3362 
3363 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3364 	}
3365 
3366 	ret = -ENOMEM;
3367 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3368 	if (!cc->io_queue) {
3369 		ti->error = "Couldn't create kcryptd io queue";
3370 		goto bad;
3371 	}
3372 
3373 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3374 		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3375 						  1, devname);
3376 	else
3377 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3378 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3379 						  num_online_cpus(), devname);
3380 	if (!cc->crypt_queue) {
3381 		ti->error = "Couldn't create kcryptd queue";
3382 		goto bad;
3383 	}
3384 
3385 	spin_lock_init(&cc->write_thread_lock);
3386 	cc->write_tree = RB_ROOT;
3387 
3388 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3389 	if (IS_ERR(cc->write_thread)) {
3390 		ret = PTR_ERR(cc->write_thread);
3391 		cc->write_thread = NULL;
3392 		ti->error = "Couldn't spawn write thread";
3393 		goto bad;
3394 	}
3395 
3396 	ti->num_flush_bios = 1;
3397 	ti->limit_swap_bios = true;
3398 	ti->accounts_remapped_io = true;
3399 
3400 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3401 	return 0;
3402 
3403 bad:
3404 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3405 	crypt_dtr(ti);
3406 	return ret;
3407 }
3408 
3409 static int crypt_map(struct dm_target *ti, struct bio *bio)
3410 {
3411 	struct dm_crypt_io *io;
3412 	struct crypt_config *cc = ti->private;
3413 
3414 	/*
3415 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3416 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3417 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3418 	 */
3419 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3420 	    bio_op(bio) == REQ_OP_DISCARD)) {
3421 		bio_set_dev(bio, cc->dev->bdev);
3422 		if (bio_sectors(bio))
3423 			bio->bi_iter.bi_sector = cc->start +
3424 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3425 		return DM_MAPIO_REMAPPED;
3426 	}
3427 
3428 	/*
3429 	 * Check if bio is too large, split as needed.
3430 	 */
3431 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3432 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3433 		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3434 
3435 	/*
3436 	 * Ensure that bio is a multiple of internal sector encryption size
3437 	 * and is aligned to this size as defined in IO hints.
3438 	 */
3439 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3440 		return DM_MAPIO_KILL;
3441 
3442 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3443 		return DM_MAPIO_KILL;
3444 
3445 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3446 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3447 
3448 	if (cc->on_disk_tag_size) {
3449 		unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3450 
3451 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3452 			io->integrity_metadata = NULL;
3453 		else
3454 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3455 
3456 		if (unlikely(!io->integrity_metadata)) {
3457 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3458 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3459 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3460 			io->integrity_metadata_from_pool = true;
3461 		}
3462 	}
3463 
3464 	if (crypt_integrity_aead(cc))
3465 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3466 	else
3467 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3468 
3469 	if (bio_data_dir(io->base_bio) == READ) {
3470 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3471 			kcryptd_queue_read(io);
3472 	} else
3473 		kcryptd_queue_crypt(io);
3474 
3475 	return DM_MAPIO_SUBMITTED;
3476 }
3477 
3478 static char hex2asc(unsigned char c)
3479 {
3480 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3481 }
3482 
3483 static void crypt_status(struct dm_target *ti, status_type_t type,
3484 			 unsigned int status_flags, char *result, unsigned int maxlen)
3485 {
3486 	struct crypt_config *cc = ti->private;
3487 	unsigned int i, sz = 0;
3488 	int num_feature_args = 0;
3489 
3490 	switch (type) {
3491 	case STATUSTYPE_INFO:
3492 		result[0] = '\0';
3493 		break;
3494 
3495 	case STATUSTYPE_TABLE:
3496 		DMEMIT("%s ", cc->cipher_string);
3497 
3498 		if (cc->key_size > 0) {
3499 			if (cc->key_string)
3500 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3501 			else {
3502 				for (i = 0; i < cc->key_size; i++) {
3503 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3504 					       hex2asc(cc->key[i] & 0xf));
3505 				}
3506 			}
3507 		} else
3508 			DMEMIT("-");
3509 
3510 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3511 				cc->dev->name, (unsigned long long)cc->start);
3512 
3513 		num_feature_args += !!ti->num_discard_bios;
3514 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3515 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3516 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3517 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3518 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3519 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3520 		if (cc->on_disk_tag_size)
3521 			num_feature_args++;
3522 		if (num_feature_args) {
3523 			DMEMIT(" %d", num_feature_args);
3524 			if (ti->num_discard_bios)
3525 				DMEMIT(" allow_discards");
3526 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3527 				DMEMIT(" same_cpu_crypt");
3528 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3529 				DMEMIT(" submit_from_crypt_cpus");
3530 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3531 				DMEMIT(" no_read_workqueue");
3532 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3533 				DMEMIT(" no_write_workqueue");
3534 			if (cc->on_disk_tag_size)
3535 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3536 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3537 				DMEMIT(" sector_size:%d", cc->sector_size);
3538 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3539 				DMEMIT(" iv_large_sectors");
3540 		}
3541 		break;
3542 
3543 	case STATUSTYPE_IMA:
3544 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3545 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3546 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3547 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3548 		       'y' : 'n');
3549 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3550 		       'y' : 'n');
3551 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3552 		       'y' : 'n');
3553 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3554 		       'y' : 'n');
3555 
3556 		if (cc->on_disk_tag_size)
3557 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3558 			       cc->on_disk_tag_size, cc->cipher_auth);
3559 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3560 			DMEMIT(",sector_size=%d", cc->sector_size);
3561 		if (cc->cipher_string)
3562 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3563 
3564 		DMEMIT(",key_size=%u", cc->key_size);
3565 		DMEMIT(",key_parts=%u", cc->key_parts);
3566 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3567 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3568 		DMEMIT(";");
3569 		break;
3570 	}
3571 }
3572 
3573 static void crypt_postsuspend(struct dm_target *ti)
3574 {
3575 	struct crypt_config *cc = ti->private;
3576 
3577 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3578 }
3579 
3580 static int crypt_preresume(struct dm_target *ti)
3581 {
3582 	struct crypt_config *cc = ti->private;
3583 
3584 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3585 		DMERR("aborting resume - crypt key is not set.");
3586 		return -EAGAIN;
3587 	}
3588 
3589 	return 0;
3590 }
3591 
3592 static void crypt_resume(struct dm_target *ti)
3593 {
3594 	struct crypt_config *cc = ti->private;
3595 
3596 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3597 }
3598 
3599 /* Message interface
3600  *	key set <key>
3601  *	key wipe
3602  */
3603 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3604 			 char *result, unsigned int maxlen)
3605 {
3606 	struct crypt_config *cc = ti->private;
3607 	int key_size, ret = -EINVAL;
3608 
3609 	if (argc < 2)
3610 		goto error;
3611 
3612 	if (!strcasecmp(argv[0], "key")) {
3613 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3614 			DMWARN("not suspended during key manipulation.");
3615 			return -EINVAL;
3616 		}
3617 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3618 			/* The key size may not be changed. */
3619 			key_size = get_key_size(&argv[2]);
3620 			if (key_size < 0 || cc->key_size != key_size) {
3621 				memset(argv[2], '0', strlen(argv[2]));
3622 				return -EINVAL;
3623 			}
3624 
3625 			ret = crypt_set_key(cc, argv[2]);
3626 			if (ret)
3627 				return ret;
3628 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3629 				ret = cc->iv_gen_ops->init(cc);
3630 			/* wipe the kernel key payload copy */
3631 			if (cc->key_string)
3632 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3633 			return ret;
3634 		}
3635 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3636 			return crypt_wipe_key(cc);
3637 	}
3638 
3639 error:
3640 	DMWARN("unrecognised message received.");
3641 	return -EINVAL;
3642 }
3643 
3644 static int crypt_iterate_devices(struct dm_target *ti,
3645 				 iterate_devices_callout_fn fn, void *data)
3646 {
3647 	struct crypt_config *cc = ti->private;
3648 
3649 	return fn(ti, cc->dev, cc->start, ti->len, data);
3650 }
3651 
3652 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3653 {
3654 	struct crypt_config *cc = ti->private;
3655 
3656 	/*
3657 	 * Unfortunate constraint that is required to avoid the potential
3658 	 * for exceeding underlying device's max_segments limits -- due to
3659 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3660 	 * bio that are not as physically contiguous as the original bio.
3661 	 */
3662 	limits->max_segment_size = PAGE_SIZE;
3663 
3664 	limits->logical_block_size =
3665 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3666 	limits->physical_block_size =
3667 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3668 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3669 	limits->dma_alignment = limits->logical_block_size - 1;
3670 }
3671 
3672 static struct target_type crypt_target = {
3673 	.name   = "crypt",
3674 	.version = {1, 24, 0},
3675 	.module = THIS_MODULE,
3676 	.ctr    = crypt_ctr,
3677 	.dtr    = crypt_dtr,
3678 	.features = DM_TARGET_ZONED_HM,
3679 	.report_zones = crypt_report_zones,
3680 	.map    = crypt_map,
3681 	.status = crypt_status,
3682 	.postsuspend = crypt_postsuspend,
3683 	.preresume = crypt_preresume,
3684 	.resume = crypt_resume,
3685 	.message = crypt_message,
3686 	.iterate_devices = crypt_iterate_devices,
3687 	.io_hints = crypt_io_hints,
3688 };
3689 module_dm(crypt);
3690 
3691 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3692 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3693 MODULE_LICENSE("GPL");
3694