xref: /linux/drivers/md/dm-crypt.c (revision 02091cbe9cc4f18167208eec1d6de636cc731817)
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_sector = io->cc->start + io->sector;
1164 
1165 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1166 				     tag_len, offset_in_page(io->integrity_metadata));
1167 	if (unlikely(ret != tag_len))
1168 		return -ENOMEM;
1169 
1170 	return 0;
1171 }
1172 
1173 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1174 {
1175 #ifdef CONFIG_BLK_DEV_INTEGRITY
1176 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1177 	struct mapped_device *md = dm_table_get_md(ti->table);
1178 
1179 	/* From now we require underlying device with our integrity profile */
1180 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1181 		ti->error = "Integrity profile not supported.";
1182 		return -EINVAL;
1183 	}
1184 
1185 	if (bi->tag_size != cc->on_disk_tag_size ||
1186 	    bi->tuple_size != cc->on_disk_tag_size) {
1187 		ti->error = "Integrity profile tag size mismatch.";
1188 		return -EINVAL;
1189 	}
1190 	if (1 << bi->interval_exp != cc->sector_size) {
1191 		ti->error = "Integrity profile sector size mismatch.";
1192 		return -EINVAL;
1193 	}
1194 
1195 	if (crypt_integrity_aead(cc)) {
1196 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1197 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1198 		       cc->integrity_tag_size, cc->integrity_iv_size);
1199 
1200 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1201 			ti->error = "Integrity AEAD auth tag size is not supported.";
1202 			return -EINVAL;
1203 		}
1204 	} else if (cc->integrity_iv_size)
1205 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1206 		       cc->integrity_iv_size);
1207 
1208 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1209 		ti->error = "Not enough space for integrity tag in the profile.";
1210 		return -EINVAL;
1211 	}
1212 
1213 	return 0;
1214 #else
1215 	ti->error = "Integrity profile not supported.";
1216 	return -EINVAL;
1217 #endif
1218 }
1219 
1220 static void crypt_convert_init(struct crypt_config *cc,
1221 			       struct convert_context *ctx,
1222 			       struct bio *bio_out, struct bio *bio_in,
1223 			       sector_t sector)
1224 {
1225 	ctx->bio_in = bio_in;
1226 	ctx->bio_out = bio_out;
1227 	if (bio_in)
1228 		ctx->iter_in = bio_in->bi_iter;
1229 	if (bio_out)
1230 		ctx->iter_out = bio_out->bi_iter;
1231 	ctx->cc_sector = sector + cc->iv_offset;
1232 	init_completion(&ctx->restart);
1233 }
1234 
1235 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1236 					     void *req)
1237 {
1238 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1239 }
1240 
1241 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1242 {
1243 	return (void *)((char *)dmreq - cc->dmreq_start);
1244 }
1245 
1246 static u8 *iv_of_dmreq(struct crypt_config *cc,
1247 		       struct dm_crypt_request *dmreq)
1248 {
1249 	if (crypt_integrity_aead(cc))
1250 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1251 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1252 	else
1253 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1254 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1255 }
1256 
1257 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1258 		       struct dm_crypt_request *dmreq)
1259 {
1260 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1261 }
1262 
1263 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1264 		       struct dm_crypt_request *dmreq)
1265 {
1266 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1267 
1268 	return (__le64 *) ptr;
1269 }
1270 
1271 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1272 		       struct dm_crypt_request *dmreq)
1273 {
1274 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1275 		  cc->iv_size + sizeof(uint64_t);
1276 
1277 	return (unsigned int *)ptr;
1278 }
1279 
1280 static void *tag_from_dmreq(struct crypt_config *cc,
1281 				struct dm_crypt_request *dmreq)
1282 {
1283 	struct convert_context *ctx = dmreq->ctx;
1284 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1285 
1286 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1287 		cc->on_disk_tag_size];
1288 }
1289 
1290 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1291 			       struct dm_crypt_request *dmreq)
1292 {
1293 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1294 }
1295 
1296 static int crypt_convert_block_aead(struct crypt_config *cc,
1297 				     struct convert_context *ctx,
1298 				     struct aead_request *req,
1299 				     unsigned int tag_offset)
1300 {
1301 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1302 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1303 	struct dm_crypt_request *dmreq;
1304 	u8 *iv, *org_iv, *tag_iv, *tag;
1305 	__le64 *sector;
1306 	int r = 0;
1307 
1308 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1309 
1310 	/* Reject unexpected unaligned bio. */
1311 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1312 		return -EIO;
1313 
1314 	dmreq = dmreq_of_req(cc, req);
1315 	dmreq->iv_sector = ctx->cc_sector;
1316 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1317 		dmreq->iv_sector >>= cc->sector_shift;
1318 	dmreq->ctx = ctx;
1319 
1320 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1321 
1322 	sector = org_sector_of_dmreq(cc, dmreq);
1323 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1324 
1325 	iv = iv_of_dmreq(cc, dmreq);
1326 	org_iv = org_iv_of_dmreq(cc, dmreq);
1327 	tag = tag_from_dmreq(cc, dmreq);
1328 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1329 
1330 	/* AEAD request:
1331 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1332 	 *  | (authenticated) | (auth+encryption) |              |
1333 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1334 	 */
1335 	sg_init_table(dmreq->sg_in, 4);
1336 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1337 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1338 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1339 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1340 
1341 	sg_init_table(dmreq->sg_out, 4);
1342 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1343 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1344 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1345 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1346 
1347 	if (cc->iv_gen_ops) {
1348 		/* For READs use IV stored in integrity metadata */
1349 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1350 			memcpy(org_iv, tag_iv, cc->iv_size);
1351 		} else {
1352 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1353 			if (r < 0)
1354 				return r;
1355 			/* Store generated IV in integrity metadata */
1356 			if (cc->integrity_iv_size)
1357 				memcpy(tag_iv, org_iv, cc->iv_size);
1358 		}
1359 		/* Working copy of IV, to be modified in crypto API */
1360 		memcpy(iv, org_iv, cc->iv_size);
1361 	}
1362 
1363 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1364 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1365 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1366 				       cc->sector_size, iv);
1367 		r = crypto_aead_encrypt(req);
1368 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1369 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1370 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1371 	} else {
1372 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1373 				       cc->sector_size + cc->integrity_tag_size, iv);
1374 		r = crypto_aead_decrypt(req);
1375 	}
1376 
1377 	if (r == -EBADMSG) {
1378 		sector_t s = le64_to_cpu(*sector);
1379 
1380 		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1381 			    ctx->bio_in->bi_bdev, s);
1382 		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1383 				 ctx->bio_in, s, 0);
1384 	}
1385 
1386 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1387 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1388 
1389 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1390 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1391 
1392 	return r;
1393 }
1394 
1395 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1396 					struct convert_context *ctx,
1397 					struct skcipher_request *req,
1398 					unsigned int tag_offset)
1399 {
1400 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1401 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1402 	struct scatterlist *sg_in, *sg_out;
1403 	struct dm_crypt_request *dmreq;
1404 	u8 *iv, *org_iv, *tag_iv;
1405 	__le64 *sector;
1406 	int r = 0;
1407 
1408 	/* Reject unexpected unaligned bio. */
1409 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1410 		return -EIO;
1411 
1412 	dmreq = dmreq_of_req(cc, req);
1413 	dmreq->iv_sector = ctx->cc_sector;
1414 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1415 		dmreq->iv_sector >>= cc->sector_shift;
1416 	dmreq->ctx = ctx;
1417 
1418 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1419 
1420 	iv = iv_of_dmreq(cc, dmreq);
1421 	org_iv = org_iv_of_dmreq(cc, dmreq);
1422 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1423 
1424 	sector = org_sector_of_dmreq(cc, dmreq);
1425 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1426 
1427 	/* For skcipher we use only the first sg item */
1428 	sg_in  = &dmreq->sg_in[0];
1429 	sg_out = &dmreq->sg_out[0];
1430 
1431 	sg_init_table(sg_in, 1);
1432 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1433 
1434 	sg_init_table(sg_out, 1);
1435 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1436 
1437 	if (cc->iv_gen_ops) {
1438 		/* For READs use IV stored in integrity metadata */
1439 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1440 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1441 		} else {
1442 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1443 			if (r < 0)
1444 				return r;
1445 			/* Data can be already preprocessed in generator */
1446 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1447 				sg_in = sg_out;
1448 			/* Store generated IV in integrity metadata */
1449 			if (cc->integrity_iv_size)
1450 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1451 		}
1452 		/* Working copy of IV, to be modified in crypto API */
1453 		memcpy(iv, org_iv, cc->iv_size);
1454 	}
1455 
1456 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1457 
1458 	if (bio_data_dir(ctx->bio_in) == WRITE)
1459 		r = crypto_skcipher_encrypt(req);
1460 	else
1461 		r = crypto_skcipher_decrypt(req);
1462 
1463 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1464 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1465 
1466 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1467 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1468 
1469 	return r;
1470 }
1471 
1472 static void kcryptd_async_done(void *async_req, int error);
1473 
1474 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1475 				     struct convert_context *ctx)
1476 {
1477 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1478 
1479 	if (!ctx->r.req) {
1480 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1481 		if (!ctx->r.req)
1482 			return -ENOMEM;
1483 	}
1484 
1485 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1486 
1487 	/*
1488 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1489 	 * requests if driver request queue is full.
1490 	 */
1491 	skcipher_request_set_callback(ctx->r.req,
1492 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1493 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1494 
1495 	return 0;
1496 }
1497 
1498 static int crypt_alloc_req_aead(struct crypt_config *cc,
1499 				 struct convert_context *ctx)
1500 {
1501 	if (!ctx->r.req_aead) {
1502 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1503 		if (!ctx->r.req_aead)
1504 			return -ENOMEM;
1505 	}
1506 
1507 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1508 
1509 	/*
1510 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1511 	 * requests if driver request queue is full.
1512 	 */
1513 	aead_request_set_callback(ctx->r.req_aead,
1514 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1515 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1516 
1517 	return 0;
1518 }
1519 
1520 static int crypt_alloc_req(struct crypt_config *cc,
1521 			    struct convert_context *ctx)
1522 {
1523 	if (crypt_integrity_aead(cc))
1524 		return crypt_alloc_req_aead(cc, ctx);
1525 	else
1526 		return crypt_alloc_req_skcipher(cc, ctx);
1527 }
1528 
1529 static void crypt_free_req_skcipher(struct crypt_config *cc,
1530 				    struct skcipher_request *req, struct bio *base_bio)
1531 {
1532 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1533 
1534 	if ((struct skcipher_request *)(io + 1) != req)
1535 		mempool_free(req, &cc->req_pool);
1536 }
1537 
1538 static void crypt_free_req_aead(struct crypt_config *cc,
1539 				struct aead_request *req, struct bio *base_bio)
1540 {
1541 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1542 
1543 	if ((struct aead_request *)(io + 1) != req)
1544 		mempool_free(req, &cc->req_pool);
1545 }
1546 
1547 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1548 {
1549 	if (crypt_integrity_aead(cc))
1550 		crypt_free_req_aead(cc, req, base_bio);
1551 	else
1552 		crypt_free_req_skcipher(cc, req, base_bio);
1553 }
1554 
1555 /*
1556  * Encrypt / decrypt data from one bio to another one (can be the same one)
1557  */
1558 static blk_status_t crypt_convert(struct crypt_config *cc,
1559 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1560 {
1561 	unsigned int tag_offset = 0;
1562 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1563 	int r;
1564 
1565 	/*
1566 	 * if reset_pending is set we are dealing with the bio for the first time,
1567 	 * else we're continuing to work on the previous bio, so don't mess with
1568 	 * the cc_pending counter
1569 	 */
1570 	if (reset_pending)
1571 		atomic_set(&ctx->cc_pending, 1);
1572 
1573 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1574 
1575 		r = crypt_alloc_req(cc, ctx);
1576 		if (r) {
1577 			complete(&ctx->restart);
1578 			return BLK_STS_DEV_RESOURCE;
1579 		}
1580 
1581 		atomic_inc(&ctx->cc_pending);
1582 
1583 		if (crypt_integrity_aead(cc))
1584 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1585 		else
1586 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1587 
1588 		switch (r) {
1589 		/*
1590 		 * The request was queued by a crypto driver
1591 		 * but the driver request queue is full, let's wait.
1592 		 */
1593 		case -EBUSY:
1594 			if (in_interrupt()) {
1595 				if (try_wait_for_completion(&ctx->restart)) {
1596 					/*
1597 					 * we don't have to block to wait for completion,
1598 					 * so proceed
1599 					 */
1600 				} else {
1601 					/*
1602 					 * we can't wait for completion without blocking
1603 					 * exit and continue processing in a workqueue
1604 					 */
1605 					ctx->r.req = NULL;
1606 					ctx->cc_sector += sector_step;
1607 					tag_offset++;
1608 					return BLK_STS_DEV_RESOURCE;
1609 				}
1610 			} else {
1611 				wait_for_completion(&ctx->restart);
1612 			}
1613 			reinit_completion(&ctx->restart);
1614 			fallthrough;
1615 		/*
1616 		 * The request is queued and processed asynchronously,
1617 		 * completion function kcryptd_async_done() will be called.
1618 		 */
1619 		case -EINPROGRESS:
1620 			ctx->r.req = NULL;
1621 			ctx->cc_sector += sector_step;
1622 			tag_offset++;
1623 			continue;
1624 		/*
1625 		 * The request was already processed (synchronously).
1626 		 */
1627 		case 0:
1628 			atomic_dec(&ctx->cc_pending);
1629 			ctx->cc_sector += sector_step;
1630 			tag_offset++;
1631 			if (!atomic)
1632 				cond_resched();
1633 			continue;
1634 		/*
1635 		 * There was a data integrity error.
1636 		 */
1637 		case -EBADMSG:
1638 			atomic_dec(&ctx->cc_pending);
1639 			return BLK_STS_PROTECTION;
1640 		/*
1641 		 * There was an error while processing the request.
1642 		 */
1643 		default:
1644 			atomic_dec(&ctx->cc_pending);
1645 			return BLK_STS_IOERR;
1646 		}
1647 	}
1648 
1649 	return 0;
1650 }
1651 
1652 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1653 
1654 /*
1655  * Generate a new unfragmented bio with the given size
1656  * This should never violate the device limitations (but only because
1657  * max_segment_size is being constrained to PAGE_SIZE).
1658  *
1659  * This function may be called concurrently. If we allocate from the mempool
1660  * concurrently, there is a possibility of deadlock. For example, if we have
1661  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1662  * the mempool concurrently, it may deadlock in a situation where both processes
1663  * have allocated 128 pages and the mempool is exhausted.
1664  *
1665  * In order to avoid this scenario we allocate the pages under a mutex.
1666  *
1667  * In order to not degrade performance with excessive locking, we try
1668  * non-blocking allocations without a mutex first but on failure we fallback
1669  * to blocking allocations with a mutex.
1670  *
1671  * In order to reduce allocation overhead, we try to allocate compound pages in
1672  * the first pass. If they are not available, we fall back to the mempool.
1673  */
1674 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1675 {
1676 	struct crypt_config *cc = io->cc;
1677 	struct bio *clone;
1678 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1679 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1680 	unsigned int remaining_size;
1681 	unsigned int order = MAX_ORDER - 1;
1682 
1683 retry:
1684 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1685 		mutex_lock(&cc->bio_alloc_lock);
1686 
1687 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1688 				 GFP_NOIO, &cc->bs);
1689 	clone->bi_private = io;
1690 	clone->bi_end_io = crypt_endio;
1691 
1692 	remaining_size = size;
1693 
1694 	while (remaining_size) {
1695 		struct page *pages;
1696 		unsigned size_to_add;
1697 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1698 		order = min(order, remaining_order);
1699 
1700 		while (order > 0) {
1701 			pages = alloc_pages(gfp_mask
1702 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1703 				order);
1704 			if (likely(pages != NULL))
1705 				goto have_pages;
1706 			order--;
1707 		}
1708 
1709 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1710 		if (!pages) {
1711 			crypt_free_buffer_pages(cc, clone);
1712 			bio_put(clone);
1713 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1714 			order = 0;
1715 			goto retry;
1716 		}
1717 
1718 have_pages:
1719 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1720 		__bio_add_page(clone, pages, size_to_add, 0);
1721 		remaining_size -= size_to_add;
1722 	}
1723 
1724 	/* Allocate space for integrity tags */
1725 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1726 		crypt_free_buffer_pages(cc, clone);
1727 		bio_put(clone);
1728 		clone = NULL;
1729 	}
1730 
1731 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1732 		mutex_unlock(&cc->bio_alloc_lock);
1733 
1734 	return clone;
1735 }
1736 
1737 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1738 {
1739 	struct folio_iter fi;
1740 
1741 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1742 		bio_for_each_folio_all(fi, clone) {
1743 			if (folio_test_large(fi.folio))
1744 				folio_put(fi.folio);
1745 			else
1746 				mempool_free(&fi.folio->page, &cc->page_pool);
1747 		}
1748 	}
1749 }
1750 
1751 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1752 			  struct bio *bio, sector_t sector)
1753 {
1754 	io->cc = cc;
1755 	io->base_bio = bio;
1756 	io->sector = sector;
1757 	io->error = 0;
1758 	io->ctx.r.req = NULL;
1759 	io->integrity_metadata = NULL;
1760 	io->integrity_metadata_from_pool = false;
1761 	io->in_tasklet = false;
1762 	atomic_set(&io->io_pending, 0);
1763 }
1764 
1765 static void crypt_inc_pending(struct dm_crypt_io *io)
1766 {
1767 	atomic_inc(&io->io_pending);
1768 }
1769 
1770 static void kcryptd_io_bio_endio(struct work_struct *work)
1771 {
1772 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1773 
1774 	bio_endio(io->base_bio);
1775 }
1776 
1777 /*
1778  * One of the bios was finished. Check for completion of
1779  * the whole request and correctly clean up the buffer.
1780  */
1781 static void crypt_dec_pending(struct dm_crypt_io *io)
1782 {
1783 	struct crypt_config *cc = io->cc;
1784 	struct bio *base_bio = io->base_bio;
1785 	blk_status_t error = io->error;
1786 
1787 	if (!atomic_dec_and_test(&io->io_pending))
1788 		return;
1789 
1790 	if (io->ctx.r.req)
1791 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1792 
1793 	if (unlikely(io->integrity_metadata_from_pool))
1794 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1795 	else
1796 		kfree(io->integrity_metadata);
1797 
1798 	base_bio->bi_status = error;
1799 
1800 	/*
1801 	 * If we are running this function from our tasklet,
1802 	 * we can't call bio_endio() here, because it will call
1803 	 * clone_endio() from dm.c, which in turn will
1804 	 * free the current struct dm_crypt_io structure with
1805 	 * our tasklet. In this case we need to delay bio_endio()
1806 	 * execution to after the tasklet is done and dequeued.
1807 	 */
1808 	if (io->in_tasklet) {
1809 		INIT_WORK(&io->work, kcryptd_io_bio_endio);
1810 		queue_work(cc->io_queue, &io->work);
1811 		return;
1812 	}
1813 
1814 	bio_endio(base_bio);
1815 }
1816 
1817 /*
1818  * kcryptd/kcryptd_io:
1819  *
1820  * Needed because it would be very unwise to do decryption in an
1821  * interrupt context.
1822  *
1823  * kcryptd performs the actual encryption or decryption.
1824  *
1825  * kcryptd_io performs the IO submission.
1826  *
1827  * They must be separated as otherwise the final stages could be
1828  * starved by new requests which can block in the first stages due
1829  * to memory allocation.
1830  *
1831  * The work is done per CPU global for all dm-crypt instances.
1832  * They should not depend on each other and do not block.
1833  */
1834 static void crypt_endio(struct bio *clone)
1835 {
1836 	struct dm_crypt_io *io = clone->bi_private;
1837 	struct crypt_config *cc = io->cc;
1838 	unsigned int rw = bio_data_dir(clone);
1839 	blk_status_t error;
1840 
1841 	/*
1842 	 * free the processed pages
1843 	 */
1844 	if (rw == WRITE)
1845 		crypt_free_buffer_pages(cc, clone);
1846 
1847 	error = clone->bi_status;
1848 	bio_put(clone);
1849 
1850 	if (rw == READ && !error) {
1851 		kcryptd_queue_crypt(io);
1852 		return;
1853 	}
1854 
1855 	if (unlikely(error))
1856 		io->error = error;
1857 
1858 	crypt_dec_pending(io);
1859 }
1860 
1861 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1862 
1863 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1864 {
1865 	struct crypt_config *cc = io->cc;
1866 	struct bio *clone;
1867 
1868 	/*
1869 	 * We need the original biovec array in order to decrypt the whole bio
1870 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1871 	 * worry about the block layer modifying the biovec array; so leverage
1872 	 * bio_alloc_clone().
1873 	 */
1874 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1875 	if (!clone)
1876 		return 1;
1877 	clone->bi_private = io;
1878 	clone->bi_end_io = crypt_endio;
1879 
1880 	crypt_inc_pending(io);
1881 
1882 	clone->bi_iter.bi_sector = cc->start + io->sector;
1883 
1884 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1885 		crypt_dec_pending(io);
1886 		bio_put(clone);
1887 		return 1;
1888 	}
1889 
1890 	dm_submit_bio_remap(io->base_bio, clone);
1891 	return 0;
1892 }
1893 
1894 static void kcryptd_io_read_work(struct work_struct *work)
1895 {
1896 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1897 
1898 	crypt_inc_pending(io);
1899 	if (kcryptd_io_read(io, GFP_NOIO))
1900 		io->error = BLK_STS_RESOURCE;
1901 	crypt_dec_pending(io);
1902 }
1903 
1904 static void kcryptd_queue_read(struct dm_crypt_io *io)
1905 {
1906 	struct crypt_config *cc = io->cc;
1907 
1908 	INIT_WORK(&io->work, kcryptd_io_read_work);
1909 	queue_work(cc->io_queue, &io->work);
1910 }
1911 
1912 static void kcryptd_io_write(struct dm_crypt_io *io)
1913 {
1914 	struct bio *clone = io->ctx.bio_out;
1915 
1916 	dm_submit_bio_remap(io->base_bio, clone);
1917 }
1918 
1919 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1920 
1921 static int dmcrypt_write(void *data)
1922 {
1923 	struct crypt_config *cc = data;
1924 	struct dm_crypt_io *io;
1925 
1926 	while (1) {
1927 		struct rb_root write_tree;
1928 		struct blk_plug plug;
1929 
1930 		spin_lock_irq(&cc->write_thread_lock);
1931 continue_locked:
1932 
1933 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1934 			goto pop_from_list;
1935 
1936 		set_current_state(TASK_INTERRUPTIBLE);
1937 
1938 		spin_unlock_irq(&cc->write_thread_lock);
1939 
1940 		if (unlikely(kthread_should_stop())) {
1941 			set_current_state(TASK_RUNNING);
1942 			break;
1943 		}
1944 
1945 		schedule();
1946 
1947 		set_current_state(TASK_RUNNING);
1948 		spin_lock_irq(&cc->write_thread_lock);
1949 		goto continue_locked;
1950 
1951 pop_from_list:
1952 		write_tree = cc->write_tree;
1953 		cc->write_tree = RB_ROOT;
1954 		spin_unlock_irq(&cc->write_thread_lock);
1955 
1956 		BUG_ON(rb_parent(write_tree.rb_node));
1957 
1958 		/*
1959 		 * Note: we cannot walk the tree here with rb_next because
1960 		 * the structures may be freed when kcryptd_io_write is called.
1961 		 */
1962 		blk_start_plug(&plug);
1963 		do {
1964 			io = crypt_io_from_node(rb_first(&write_tree));
1965 			rb_erase(&io->rb_node, &write_tree);
1966 			kcryptd_io_write(io);
1967 			cond_resched();
1968 		} while (!RB_EMPTY_ROOT(&write_tree));
1969 		blk_finish_plug(&plug);
1970 	}
1971 	return 0;
1972 }
1973 
1974 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1975 {
1976 	struct bio *clone = io->ctx.bio_out;
1977 	struct crypt_config *cc = io->cc;
1978 	unsigned long flags;
1979 	sector_t sector;
1980 	struct rb_node **rbp, *parent;
1981 
1982 	if (unlikely(io->error)) {
1983 		crypt_free_buffer_pages(cc, clone);
1984 		bio_put(clone);
1985 		crypt_dec_pending(io);
1986 		return;
1987 	}
1988 
1989 	/* crypt_convert should have filled the clone bio */
1990 	BUG_ON(io->ctx.iter_out.bi_size);
1991 
1992 	clone->bi_iter.bi_sector = cc->start + io->sector;
1993 
1994 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1995 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1996 		dm_submit_bio_remap(io->base_bio, clone);
1997 		return;
1998 	}
1999 
2000 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2001 	if (RB_EMPTY_ROOT(&cc->write_tree))
2002 		wake_up_process(cc->write_thread);
2003 	rbp = &cc->write_tree.rb_node;
2004 	parent = NULL;
2005 	sector = io->sector;
2006 	while (*rbp) {
2007 		parent = *rbp;
2008 		if (sector < crypt_io_from_node(parent)->sector)
2009 			rbp = &(*rbp)->rb_left;
2010 		else
2011 			rbp = &(*rbp)->rb_right;
2012 	}
2013 	rb_link_node(&io->rb_node, parent, rbp);
2014 	rb_insert_color(&io->rb_node, &cc->write_tree);
2015 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2016 }
2017 
2018 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2019 				       struct convert_context *ctx)
2020 
2021 {
2022 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2023 		return false;
2024 
2025 	/*
2026 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2027 	 * constraints so they do not need to be issued inline by
2028 	 * kcryptd_crypt_write_convert().
2029 	 */
2030 	switch (bio_op(ctx->bio_in)) {
2031 	case REQ_OP_WRITE:
2032 	case REQ_OP_WRITE_ZEROES:
2033 		return true;
2034 	default:
2035 		return false;
2036 	}
2037 }
2038 
2039 static void kcryptd_crypt_write_continue(struct work_struct *work)
2040 {
2041 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2042 	struct crypt_config *cc = io->cc;
2043 	struct convert_context *ctx = &io->ctx;
2044 	int crypt_finished;
2045 	sector_t sector = io->sector;
2046 	blk_status_t r;
2047 
2048 	wait_for_completion(&ctx->restart);
2049 	reinit_completion(&ctx->restart);
2050 
2051 	r = crypt_convert(cc, &io->ctx, true, false);
2052 	if (r)
2053 		io->error = r;
2054 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2055 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2056 		/* Wait for completion signaled by kcryptd_async_done() */
2057 		wait_for_completion(&ctx->restart);
2058 		crypt_finished = 1;
2059 	}
2060 
2061 	/* Encryption was already finished, submit io now */
2062 	if (crypt_finished) {
2063 		kcryptd_crypt_write_io_submit(io, 0);
2064 		io->sector = sector;
2065 	}
2066 
2067 	crypt_dec_pending(io);
2068 }
2069 
2070 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2071 {
2072 	struct crypt_config *cc = io->cc;
2073 	struct convert_context *ctx = &io->ctx;
2074 	struct bio *clone;
2075 	int crypt_finished;
2076 	sector_t sector = io->sector;
2077 	blk_status_t r;
2078 
2079 	/*
2080 	 * Prevent io from disappearing until this function completes.
2081 	 */
2082 	crypt_inc_pending(io);
2083 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2084 
2085 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2086 	if (unlikely(!clone)) {
2087 		io->error = BLK_STS_IOERR;
2088 		goto dec;
2089 	}
2090 
2091 	io->ctx.bio_out = clone;
2092 	io->ctx.iter_out = clone->bi_iter;
2093 
2094 	sector += bio_sectors(clone);
2095 
2096 	crypt_inc_pending(io);
2097 	r = crypt_convert(cc, ctx,
2098 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2099 	/*
2100 	 * Crypto API backlogged the request, because its queue was full
2101 	 * and we're in softirq context, so continue from a workqueue
2102 	 * (TODO: is it actually possible to be in softirq in the write path?)
2103 	 */
2104 	if (r == BLK_STS_DEV_RESOURCE) {
2105 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2106 		queue_work(cc->crypt_queue, &io->work);
2107 		return;
2108 	}
2109 	if (r)
2110 		io->error = r;
2111 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2112 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2113 		/* Wait for completion signaled by kcryptd_async_done() */
2114 		wait_for_completion(&ctx->restart);
2115 		crypt_finished = 1;
2116 	}
2117 
2118 	/* Encryption was already finished, submit io now */
2119 	if (crypt_finished) {
2120 		kcryptd_crypt_write_io_submit(io, 0);
2121 		io->sector = sector;
2122 	}
2123 
2124 dec:
2125 	crypt_dec_pending(io);
2126 }
2127 
2128 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2129 {
2130 	crypt_dec_pending(io);
2131 }
2132 
2133 static void kcryptd_crypt_read_continue(struct work_struct *work)
2134 {
2135 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2136 	struct crypt_config *cc = io->cc;
2137 	blk_status_t r;
2138 
2139 	wait_for_completion(&io->ctx.restart);
2140 	reinit_completion(&io->ctx.restart);
2141 
2142 	r = crypt_convert(cc, &io->ctx, true, false);
2143 	if (r)
2144 		io->error = r;
2145 
2146 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2147 		kcryptd_crypt_read_done(io);
2148 
2149 	crypt_dec_pending(io);
2150 }
2151 
2152 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2153 {
2154 	struct crypt_config *cc = io->cc;
2155 	blk_status_t r;
2156 
2157 	crypt_inc_pending(io);
2158 
2159 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2160 			   io->sector);
2161 
2162 	r = crypt_convert(cc, &io->ctx,
2163 			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2164 	/*
2165 	 * Crypto API backlogged the request, because its queue was full
2166 	 * and we're in softirq context, so continue from a workqueue
2167 	 */
2168 	if (r == BLK_STS_DEV_RESOURCE) {
2169 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2170 		queue_work(cc->crypt_queue, &io->work);
2171 		return;
2172 	}
2173 	if (r)
2174 		io->error = r;
2175 
2176 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2177 		kcryptd_crypt_read_done(io);
2178 
2179 	crypt_dec_pending(io);
2180 }
2181 
2182 static void kcryptd_async_done(void *data, int error)
2183 {
2184 	struct dm_crypt_request *dmreq = data;
2185 	struct convert_context *ctx = dmreq->ctx;
2186 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2187 	struct crypt_config *cc = io->cc;
2188 
2189 	/*
2190 	 * A request from crypto driver backlog is going to be processed now,
2191 	 * finish the completion and continue in crypt_convert().
2192 	 * (Callback will be called for the second time for this request.)
2193 	 */
2194 	if (error == -EINPROGRESS) {
2195 		complete(&ctx->restart);
2196 		return;
2197 	}
2198 
2199 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2200 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2201 
2202 	if (error == -EBADMSG) {
2203 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2204 
2205 		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2206 			    ctx->bio_in->bi_bdev, s);
2207 		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2208 				 ctx->bio_in, s, 0);
2209 		io->error = BLK_STS_PROTECTION;
2210 	} else if (error < 0)
2211 		io->error = BLK_STS_IOERR;
2212 
2213 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2214 
2215 	if (!atomic_dec_and_test(&ctx->cc_pending))
2216 		return;
2217 
2218 	/*
2219 	 * The request is fully completed: for inline writes, let
2220 	 * kcryptd_crypt_write_convert() do the IO submission.
2221 	 */
2222 	if (bio_data_dir(io->base_bio) == READ) {
2223 		kcryptd_crypt_read_done(io);
2224 		return;
2225 	}
2226 
2227 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2228 		complete(&ctx->restart);
2229 		return;
2230 	}
2231 
2232 	kcryptd_crypt_write_io_submit(io, 1);
2233 }
2234 
2235 static void kcryptd_crypt(struct work_struct *work)
2236 {
2237 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2238 
2239 	if (bio_data_dir(io->base_bio) == READ)
2240 		kcryptd_crypt_read_convert(io);
2241 	else
2242 		kcryptd_crypt_write_convert(io);
2243 }
2244 
2245 static void kcryptd_crypt_tasklet(unsigned long work)
2246 {
2247 	kcryptd_crypt((struct work_struct *)work);
2248 }
2249 
2250 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2251 {
2252 	struct crypt_config *cc = io->cc;
2253 
2254 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2255 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2256 		/*
2257 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2258 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2259 		 * it is being executed with irqs disabled.
2260 		 */
2261 		if (in_hardirq() || irqs_disabled()) {
2262 			io->in_tasklet = true;
2263 			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2264 			tasklet_schedule(&io->tasklet);
2265 			return;
2266 		}
2267 
2268 		kcryptd_crypt(&io->work);
2269 		return;
2270 	}
2271 
2272 	INIT_WORK(&io->work, kcryptd_crypt);
2273 	queue_work(cc->crypt_queue, &io->work);
2274 }
2275 
2276 static void crypt_free_tfms_aead(struct crypt_config *cc)
2277 {
2278 	if (!cc->cipher_tfm.tfms_aead)
2279 		return;
2280 
2281 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2282 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2283 		cc->cipher_tfm.tfms_aead[0] = NULL;
2284 	}
2285 
2286 	kfree(cc->cipher_tfm.tfms_aead);
2287 	cc->cipher_tfm.tfms_aead = NULL;
2288 }
2289 
2290 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2291 {
2292 	unsigned int i;
2293 
2294 	if (!cc->cipher_tfm.tfms)
2295 		return;
2296 
2297 	for (i = 0; i < cc->tfms_count; i++)
2298 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2299 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2300 			cc->cipher_tfm.tfms[i] = NULL;
2301 		}
2302 
2303 	kfree(cc->cipher_tfm.tfms);
2304 	cc->cipher_tfm.tfms = NULL;
2305 }
2306 
2307 static void crypt_free_tfms(struct crypt_config *cc)
2308 {
2309 	if (crypt_integrity_aead(cc))
2310 		crypt_free_tfms_aead(cc);
2311 	else
2312 		crypt_free_tfms_skcipher(cc);
2313 }
2314 
2315 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2316 {
2317 	unsigned int i;
2318 	int err;
2319 
2320 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2321 				      sizeof(struct crypto_skcipher *),
2322 				      GFP_KERNEL);
2323 	if (!cc->cipher_tfm.tfms)
2324 		return -ENOMEM;
2325 
2326 	for (i = 0; i < cc->tfms_count; i++) {
2327 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2328 						CRYPTO_ALG_ALLOCATES_MEMORY);
2329 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2330 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2331 			crypt_free_tfms(cc);
2332 			return err;
2333 		}
2334 	}
2335 
2336 	/*
2337 	 * dm-crypt performance can vary greatly depending on which crypto
2338 	 * algorithm implementation is used.  Help people debug performance
2339 	 * problems by logging the ->cra_driver_name.
2340 	 */
2341 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2342 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2343 	return 0;
2344 }
2345 
2346 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2347 {
2348 	int err;
2349 
2350 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2351 	if (!cc->cipher_tfm.tfms)
2352 		return -ENOMEM;
2353 
2354 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2355 						CRYPTO_ALG_ALLOCATES_MEMORY);
2356 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2357 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2358 		crypt_free_tfms(cc);
2359 		return err;
2360 	}
2361 
2362 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2363 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2364 	return 0;
2365 }
2366 
2367 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2368 {
2369 	if (crypt_integrity_aead(cc))
2370 		return crypt_alloc_tfms_aead(cc, ciphermode);
2371 	else
2372 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2373 }
2374 
2375 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2376 {
2377 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2378 }
2379 
2380 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2381 {
2382 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2383 }
2384 
2385 /*
2386  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2387  * the key must be for some reason in special format.
2388  * This funcion converts cc->key to this special format.
2389  */
2390 static void crypt_copy_authenckey(char *p, const void *key,
2391 				  unsigned int enckeylen, unsigned int authkeylen)
2392 {
2393 	struct crypto_authenc_key_param *param;
2394 	struct rtattr *rta;
2395 
2396 	rta = (struct rtattr *)p;
2397 	param = RTA_DATA(rta);
2398 	param->enckeylen = cpu_to_be32(enckeylen);
2399 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2400 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2401 	p += RTA_SPACE(sizeof(*param));
2402 	memcpy(p, key + enckeylen, authkeylen);
2403 	p += authkeylen;
2404 	memcpy(p, key, enckeylen);
2405 }
2406 
2407 static int crypt_setkey(struct crypt_config *cc)
2408 {
2409 	unsigned int subkey_size;
2410 	int err = 0, i, r;
2411 
2412 	/* Ignore extra keys (which are used for IV etc) */
2413 	subkey_size = crypt_subkey_size(cc);
2414 
2415 	if (crypt_integrity_hmac(cc)) {
2416 		if (subkey_size < cc->key_mac_size)
2417 			return -EINVAL;
2418 
2419 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2420 				      subkey_size - cc->key_mac_size,
2421 				      cc->key_mac_size);
2422 	}
2423 
2424 	for (i = 0; i < cc->tfms_count; i++) {
2425 		if (crypt_integrity_hmac(cc))
2426 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2427 				cc->authenc_key, crypt_authenckey_size(cc));
2428 		else if (crypt_integrity_aead(cc))
2429 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2430 					       cc->key + (i * subkey_size),
2431 					       subkey_size);
2432 		else
2433 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2434 						   cc->key + (i * subkey_size),
2435 						   subkey_size);
2436 		if (r)
2437 			err = r;
2438 	}
2439 
2440 	if (crypt_integrity_hmac(cc))
2441 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2442 
2443 	return err;
2444 }
2445 
2446 #ifdef CONFIG_KEYS
2447 
2448 static bool contains_whitespace(const char *str)
2449 {
2450 	while (*str)
2451 		if (isspace(*str++))
2452 			return true;
2453 	return false;
2454 }
2455 
2456 static int set_key_user(struct crypt_config *cc, struct key *key)
2457 {
2458 	const struct user_key_payload *ukp;
2459 
2460 	ukp = user_key_payload_locked(key);
2461 	if (!ukp)
2462 		return -EKEYREVOKED;
2463 
2464 	if (cc->key_size != ukp->datalen)
2465 		return -EINVAL;
2466 
2467 	memcpy(cc->key, ukp->data, cc->key_size);
2468 
2469 	return 0;
2470 }
2471 
2472 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2473 {
2474 	const struct encrypted_key_payload *ekp;
2475 
2476 	ekp = key->payload.data[0];
2477 	if (!ekp)
2478 		return -EKEYREVOKED;
2479 
2480 	if (cc->key_size != ekp->decrypted_datalen)
2481 		return -EINVAL;
2482 
2483 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2484 
2485 	return 0;
2486 }
2487 
2488 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2489 {
2490 	const struct trusted_key_payload *tkp;
2491 
2492 	tkp = key->payload.data[0];
2493 	if (!tkp)
2494 		return -EKEYREVOKED;
2495 
2496 	if (cc->key_size != tkp->key_len)
2497 		return -EINVAL;
2498 
2499 	memcpy(cc->key, tkp->key, cc->key_size);
2500 
2501 	return 0;
2502 }
2503 
2504 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2505 {
2506 	char *new_key_string, *key_desc;
2507 	int ret;
2508 	struct key_type *type;
2509 	struct key *key;
2510 	int (*set_key)(struct crypt_config *cc, struct key *key);
2511 
2512 	/*
2513 	 * Reject key_string with whitespace. dm core currently lacks code for
2514 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2515 	 */
2516 	if (contains_whitespace(key_string)) {
2517 		DMERR("whitespace chars not allowed in key string");
2518 		return -EINVAL;
2519 	}
2520 
2521 	/* look for next ':' separating key_type from key_description */
2522 	key_desc = strchr(key_string, ':');
2523 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2524 		return -EINVAL;
2525 
2526 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2527 		type = &key_type_logon;
2528 		set_key = set_key_user;
2529 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2530 		type = &key_type_user;
2531 		set_key = set_key_user;
2532 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2533 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2534 		type = &key_type_encrypted;
2535 		set_key = set_key_encrypted;
2536 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2537 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2538 		type = &key_type_trusted;
2539 		set_key = set_key_trusted;
2540 	} else {
2541 		return -EINVAL;
2542 	}
2543 
2544 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2545 	if (!new_key_string)
2546 		return -ENOMEM;
2547 
2548 	key = request_key(type, key_desc + 1, NULL);
2549 	if (IS_ERR(key)) {
2550 		kfree_sensitive(new_key_string);
2551 		return PTR_ERR(key);
2552 	}
2553 
2554 	down_read(&key->sem);
2555 
2556 	ret = set_key(cc, key);
2557 	if (ret < 0) {
2558 		up_read(&key->sem);
2559 		key_put(key);
2560 		kfree_sensitive(new_key_string);
2561 		return ret;
2562 	}
2563 
2564 	up_read(&key->sem);
2565 	key_put(key);
2566 
2567 	/* clear the flag since following operations may invalidate previously valid key */
2568 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2569 
2570 	ret = crypt_setkey(cc);
2571 
2572 	if (!ret) {
2573 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2574 		kfree_sensitive(cc->key_string);
2575 		cc->key_string = new_key_string;
2576 	} else
2577 		kfree_sensitive(new_key_string);
2578 
2579 	return ret;
2580 }
2581 
2582 static int get_key_size(char **key_string)
2583 {
2584 	char *colon, dummy;
2585 	int ret;
2586 
2587 	if (*key_string[0] != ':')
2588 		return strlen(*key_string) >> 1;
2589 
2590 	/* look for next ':' in key string */
2591 	colon = strpbrk(*key_string + 1, ":");
2592 	if (!colon)
2593 		return -EINVAL;
2594 
2595 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2596 		return -EINVAL;
2597 
2598 	*key_string = colon;
2599 
2600 	/* remaining key string should be :<logon|user>:<key_desc> */
2601 
2602 	return ret;
2603 }
2604 
2605 #else
2606 
2607 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2608 {
2609 	return -EINVAL;
2610 }
2611 
2612 static int get_key_size(char **key_string)
2613 {
2614 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2615 }
2616 
2617 #endif /* CONFIG_KEYS */
2618 
2619 static int crypt_set_key(struct crypt_config *cc, char *key)
2620 {
2621 	int r = -EINVAL;
2622 	int key_string_len = strlen(key);
2623 
2624 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2625 	if (!cc->key_size && strcmp(key, "-"))
2626 		goto out;
2627 
2628 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2629 	if (key[0] == ':') {
2630 		r = crypt_set_keyring_key(cc, key + 1);
2631 		goto out;
2632 	}
2633 
2634 	/* clear the flag since following operations may invalidate previously valid key */
2635 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2636 
2637 	/* wipe references to any kernel keyring key */
2638 	kfree_sensitive(cc->key_string);
2639 	cc->key_string = NULL;
2640 
2641 	/* Decode key from its hex representation. */
2642 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2643 		goto out;
2644 
2645 	r = crypt_setkey(cc);
2646 	if (!r)
2647 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2648 
2649 out:
2650 	/* Hex key string not needed after here, so wipe it. */
2651 	memset(key, '0', key_string_len);
2652 
2653 	return r;
2654 }
2655 
2656 static int crypt_wipe_key(struct crypt_config *cc)
2657 {
2658 	int r;
2659 
2660 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2661 	get_random_bytes(&cc->key, cc->key_size);
2662 
2663 	/* Wipe IV private keys */
2664 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2665 		r = cc->iv_gen_ops->wipe(cc);
2666 		if (r)
2667 			return r;
2668 	}
2669 
2670 	kfree_sensitive(cc->key_string);
2671 	cc->key_string = NULL;
2672 	r = crypt_setkey(cc);
2673 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2674 
2675 	return r;
2676 }
2677 
2678 static void crypt_calculate_pages_per_client(void)
2679 {
2680 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2681 
2682 	if (!dm_crypt_clients_n)
2683 		return;
2684 
2685 	pages /= dm_crypt_clients_n;
2686 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2687 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2688 	dm_crypt_pages_per_client = pages;
2689 }
2690 
2691 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2692 {
2693 	struct crypt_config *cc = pool_data;
2694 	struct page *page;
2695 
2696 	/*
2697 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2698 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2699 	 * but avoids potential spinlock contention of an exact result.
2700 	 */
2701 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2702 	    likely(gfp_mask & __GFP_NORETRY))
2703 		return NULL;
2704 
2705 	page = alloc_page(gfp_mask);
2706 	if (likely(page != NULL))
2707 		percpu_counter_add(&cc->n_allocated_pages, 1);
2708 
2709 	return page;
2710 }
2711 
2712 static void crypt_page_free(void *page, void *pool_data)
2713 {
2714 	struct crypt_config *cc = pool_data;
2715 
2716 	__free_page(page);
2717 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2718 }
2719 
2720 static void crypt_dtr(struct dm_target *ti)
2721 {
2722 	struct crypt_config *cc = ti->private;
2723 
2724 	ti->private = NULL;
2725 
2726 	if (!cc)
2727 		return;
2728 
2729 	if (cc->write_thread)
2730 		kthread_stop(cc->write_thread);
2731 
2732 	if (cc->io_queue)
2733 		destroy_workqueue(cc->io_queue);
2734 	if (cc->crypt_queue)
2735 		destroy_workqueue(cc->crypt_queue);
2736 
2737 	crypt_free_tfms(cc);
2738 
2739 	bioset_exit(&cc->bs);
2740 
2741 	mempool_exit(&cc->page_pool);
2742 	mempool_exit(&cc->req_pool);
2743 	mempool_exit(&cc->tag_pool);
2744 
2745 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2746 	percpu_counter_destroy(&cc->n_allocated_pages);
2747 
2748 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2749 		cc->iv_gen_ops->dtr(cc);
2750 
2751 	if (cc->dev)
2752 		dm_put_device(ti, cc->dev);
2753 
2754 	kfree_sensitive(cc->cipher_string);
2755 	kfree_sensitive(cc->key_string);
2756 	kfree_sensitive(cc->cipher_auth);
2757 	kfree_sensitive(cc->authenc_key);
2758 
2759 	mutex_destroy(&cc->bio_alloc_lock);
2760 
2761 	/* Must zero key material before freeing */
2762 	kfree_sensitive(cc);
2763 
2764 	spin_lock(&dm_crypt_clients_lock);
2765 	WARN_ON(!dm_crypt_clients_n);
2766 	dm_crypt_clients_n--;
2767 	crypt_calculate_pages_per_client();
2768 	spin_unlock(&dm_crypt_clients_lock);
2769 
2770 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2771 }
2772 
2773 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2774 {
2775 	struct crypt_config *cc = ti->private;
2776 
2777 	if (crypt_integrity_aead(cc))
2778 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2779 	else
2780 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2781 
2782 	if (cc->iv_size)
2783 		/* at least a 64 bit sector number should fit in our buffer */
2784 		cc->iv_size = max(cc->iv_size,
2785 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2786 	else if (ivmode) {
2787 		DMWARN("Selected cipher does not support IVs");
2788 		ivmode = NULL;
2789 	}
2790 
2791 	/* Choose ivmode, see comments at iv code. */
2792 	if (ivmode == NULL)
2793 		cc->iv_gen_ops = NULL;
2794 	else if (strcmp(ivmode, "plain") == 0)
2795 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2796 	else if (strcmp(ivmode, "plain64") == 0)
2797 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2798 	else if (strcmp(ivmode, "plain64be") == 0)
2799 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2800 	else if (strcmp(ivmode, "essiv") == 0)
2801 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2802 	else if (strcmp(ivmode, "benbi") == 0)
2803 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2804 	else if (strcmp(ivmode, "null") == 0)
2805 		cc->iv_gen_ops = &crypt_iv_null_ops;
2806 	else if (strcmp(ivmode, "eboiv") == 0)
2807 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2808 	else if (strcmp(ivmode, "elephant") == 0) {
2809 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2810 		cc->key_parts = 2;
2811 		cc->key_extra_size = cc->key_size / 2;
2812 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2813 			return -EINVAL;
2814 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2815 	} else if (strcmp(ivmode, "lmk") == 0) {
2816 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2817 		/*
2818 		 * Version 2 and 3 is recognised according
2819 		 * to length of provided multi-key string.
2820 		 * If present (version 3), last key is used as IV seed.
2821 		 * All keys (including IV seed) are always the same size.
2822 		 */
2823 		if (cc->key_size % cc->key_parts) {
2824 			cc->key_parts++;
2825 			cc->key_extra_size = cc->key_size / cc->key_parts;
2826 		}
2827 	} else if (strcmp(ivmode, "tcw") == 0) {
2828 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2829 		cc->key_parts += 2; /* IV + whitening */
2830 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2831 	} else if (strcmp(ivmode, "random") == 0) {
2832 		cc->iv_gen_ops = &crypt_iv_random_ops;
2833 		/* Need storage space in integrity fields. */
2834 		cc->integrity_iv_size = cc->iv_size;
2835 	} else {
2836 		ti->error = "Invalid IV mode";
2837 		return -EINVAL;
2838 	}
2839 
2840 	return 0;
2841 }
2842 
2843 /*
2844  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2845  * The HMAC is needed to calculate tag size (HMAC digest size).
2846  * This should be probably done by crypto-api calls (once available...)
2847  */
2848 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2849 {
2850 	char *start, *end, *mac_alg = NULL;
2851 	struct crypto_ahash *mac;
2852 
2853 	if (!strstarts(cipher_api, "authenc("))
2854 		return 0;
2855 
2856 	start = strchr(cipher_api, '(');
2857 	end = strchr(cipher_api, ',');
2858 	if (!start || !end || ++start > end)
2859 		return -EINVAL;
2860 
2861 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2862 	if (!mac_alg)
2863 		return -ENOMEM;
2864 	strncpy(mac_alg, start, end - start);
2865 
2866 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2867 	kfree(mac_alg);
2868 
2869 	if (IS_ERR(mac))
2870 		return PTR_ERR(mac);
2871 
2872 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2873 	crypto_free_ahash(mac);
2874 
2875 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2876 	if (!cc->authenc_key)
2877 		return -ENOMEM;
2878 
2879 	return 0;
2880 }
2881 
2882 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2883 				char **ivmode, char **ivopts)
2884 {
2885 	struct crypt_config *cc = ti->private;
2886 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2887 	int ret = -EINVAL;
2888 
2889 	cc->tfms_count = 1;
2890 
2891 	/*
2892 	 * New format (capi: prefix)
2893 	 * capi:cipher_api_spec-iv:ivopts
2894 	 */
2895 	tmp = &cipher_in[strlen("capi:")];
2896 
2897 	/* Separate IV options if present, it can contain another '-' in hash name */
2898 	*ivopts = strrchr(tmp, ':');
2899 	if (*ivopts) {
2900 		**ivopts = '\0';
2901 		(*ivopts)++;
2902 	}
2903 	/* Parse IV mode */
2904 	*ivmode = strrchr(tmp, '-');
2905 	if (*ivmode) {
2906 		**ivmode = '\0';
2907 		(*ivmode)++;
2908 	}
2909 	/* The rest is crypto API spec */
2910 	cipher_api = tmp;
2911 
2912 	/* Alloc AEAD, can be used only in new format. */
2913 	if (crypt_integrity_aead(cc)) {
2914 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2915 		if (ret < 0) {
2916 			ti->error = "Invalid AEAD cipher spec";
2917 			return ret;
2918 		}
2919 	}
2920 
2921 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2922 		cc->tfms_count = 64;
2923 
2924 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2925 		if (!*ivopts) {
2926 			ti->error = "Digest algorithm missing for ESSIV mode";
2927 			return -EINVAL;
2928 		}
2929 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2930 			       cipher_api, *ivopts);
2931 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2932 			ti->error = "Cannot allocate cipher string";
2933 			return -ENOMEM;
2934 		}
2935 		cipher_api = buf;
2936 	}
2937 
2938 	cc->key_parts = cc->tfms_count;
2939 
2940 	/* Allocate cipher */
2941 	ret = crypt_alloc_tfms(cc, cipher_api);
2942 	if (ret < 0) {
2943 		ti->error = "Error allocating crypto tfm";
2944 		return ret;
2945 	}
2946 
2947 	if (crypt_integrity_aead(cc))
2948 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2949 	else
2950 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2951 
2952 	return 0;
2953 }
2954 
2955 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2956 				char **ivmode, char **ivopts)
2957 {
2958 	struct crypt_config *cc = ti->private;
2959 	char *tmp, *cipher, *chainmode, *keycount;
2960 	char *cipher_api = NULL;
2961 	int ret = -EINVAL;
2962 	char dummy;
2963 
2964 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2965 		ti->error = "Bad cipher specification";
2966 		return -EINVAL;
2967 	}
2968 
2969 	/*
2970 	 * Legacy dm-crypt cipher specification
2971 	 * cipher[:keycount]-mode-iv:ivopts
2972 	 */
2973 	tmp = cipher_in;
2974 	keycount = strsep(&tmp, "-");
2975 	cipher = strsep(&keycount, ":");
2976 
2977 	if (!keycount)
2978 		cc->tfms_count = 1;
2979 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2980 		 !is_power_of_2(cc->tfms_count)) {
2981 		ti->error = "Bad cipher key count specification";
2982 		return -EINVAL;
2983 	}
2984 	cc->key_parts = cc->tfms_count;
2985 
2986 	chainmode = strsep(&tmp, "-");
2987 	*ivmode = strsep(&tmp, ":");
2988 	*ivopts = tmp;
2989 
2990 	/*
2991 	 * For compatibility with the original dm-crypt mapping format, if
2992 	 * only the cipher name is supplied, use cbc-plain.
2993 	 */
2994 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2995 		chainmode = "cbc";
2996 		*ivmode = "plain";
2997 	}
2998 
2999 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3000 		ti->error = "IV mechanism required";
3001 		return -EINVAL;
3002 	}
3003 
3004 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3005 	if (!cipher_api)
3006 		goto bad_mem;
3007 
3008 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3009 		if (!*ivopts) {
3010 			ti->error = "Digest algorithm missing for ESSIV mode";
3011 			kfree(cipher_api);
3012 			return -EINVAL;
3013 		}
3014 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3015 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3016 	} else {
3017 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3018 			       "%s(%s)", chainmode, cipher);
3019 	}
3020 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3021 		kfree(cipher_api);
3022 		goto bad_mem;
3023 	}
3024 
3025 	/* Allocate cipher */
3026 	ret = crypt_alloc_tfms(cc, cipher_api);
3027 	if (ret < 0) {
3028 		ti->error = "Error allocating crypto tfm";
3029 		kfree(cipher_api);
3030 		return ret;
3031 	}
3032 	kfree(cipher_api);
3033 
3034 	return 0;
3035 bad_mem:
3036 	ti->error = "Cannot allocate cipher strings";
3037 	return -ENOMEM;
3038 }
3039 
3040 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3041 {
3042 	struct crypt_config *cc = ti->private;
3043 	char *ivmode = NULL, *ivopts = NULL;
3044 	int ret;
3045 
3046 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3047 	if (!cc->cipher_string) {
3048 		ti->error = "Cannot allocate cipher strings";
3049 		return -ENOMEM;
3050 	}
3051 
3052 	if (strstarts(cipher_in, "capi:"))
3053 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3054 	else
3055 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3056 	if (ret)
3057 		return ret;
3058 
3059 	/* Initialize IV */
3060 	ret = crypt_ctr_ivmode(ti, ivmode);
3061 	if (ret < 0)
3062 		return ret;
3063 
3064 	/* Initialize and set key */
3065 	ret = crypt_set_key(cc, key);
3066 	if (ret < 0) {
3067 		ti->error = "Error decoding and setting key";
3068 		return ret;
3069 	}
3070 
3071 	/* Allocate IV */
3072 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3073 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3074 		if (ret < 0) {
3075 			ti->error = "Error creating IV";
3076 			return ret;
3077 		}
3078 	}
3079 
3080 	/* Initialize IV (set keys for ESSIV etc) */
3081 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3082 		ret = cc->iv_gen_ops->init(cc);
3083 		if (ret < 0) {
3084 			ti->error = "Error initialising IV";
3085 			return ret;
3086 		}
3087 	}
3088 
3089 	/* wipe the kernel key payload copy */
3090 	if (cc->key_string)
3091 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3092 
3093 	return ret;
3094 }
3095 
3096 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3097 {
3098 	struct crypt_config *cc = ti->private;
3099 	struct dm_arg_set as;
3100 	static const struct dm_arg _args[] = {
3101 		{0, 8, "Invalid number of feature args"},
3102 	};
3103 	unsigned int opt_params, val;
3104 	const char *opt_string, *sval;
3105 	char dummy;
3106 	int ret;
3107 
3108 	/* Optional parameters */
3109 	as.argc = argc;
3110 	as.argv = argv;
3111 
3112 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3113 	if (ret)
3114 		return ret;
3115 
3116 	while (opt_params--) {
3117 		opt_string = dm_shift_arg(&as);
3118 		if (!opt_string) {
3119 			ti->error = "Not enough feature arguments";
3120 			return -EINVAL;
3121 		}
3122 
3123 		if (!strcasecmp(opt_string, "allow_discards"))
3124 			ti->num_discard_bios = 1;
3125 
3126 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3127 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3128 
3129 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3130 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3131 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3132 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3133 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3134 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3135 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3136 			if (val == 0 || val > MAX_TAG_SIZE) {
3137 				ti->error = "Invalid integrity arguments";
3138 				return -EINVAL;
3139 			}
3140 			cc->on_disk_tag_size = val;
3141 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3142 			if (!strcasecmp(sval, "aead")) {
3143 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3144 			} else  if (strcasecmp(sval, "none")) {
3145 				ti->error = "Unknown integrity profile";
3146 				return -EINVAL;
3147 			}
3148 
3149 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3150 			if (!cc->cipher_auth)
3151 				return -ENOMEM;
3152 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3153 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3154 			    cc->sector_size > 4096 ||
3155 			    (cc->sector_size & (cc->sector_size - 1))) {
3156 				ti->error = "Invalid feature value for sector_size";
3157 				return -EINVAL;
3158 			}
3159 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3160 				ti->error = "Device size is not multiple of sector_size feature";
3161 				return -EINVAL;
3162 			}
3163 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3164 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3165 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3166 		else {
3167 			ti->error = "Invalid feature arguments";
3168 			return -EINVAL;
3169 		}
3170 	}
3171 
3172 	return 0;
3173 }
3174 
3175 #ifdef CONFIG_BLK_DEV_ZONED
3176 static int crypt_report_zones(struct dm_target *ti,
3177 		struct dm_report_zones_args *args, unsigned int nr_zones)
3178 {
3179 	struct crypt_config *cc = ti->private;
3180 
3181 	return dm_report_zones(cc->dev->bdev, cc->start,
3182 			cc->start + dm_target_offset(ti, args->next_sector),
3183 			args, nr_zones);
3184 }
3185 #else
3186 #define crypt_report_zones NULL
3187 #endif
3188 
3189 /*
3190  * Construct an encryption mapping:
3191  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3192  */
3193 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3194 {
3195 	struct crypt_config *cc;
3196 	const char *devname = dm_table_device_name(ti->table);
3197 	int key_size;
3198 	unsigned int align_mask;
3199 	unsigned long long tmpll;
3200 	int ret;
3201 	size_t iv_size_padding, additional_req_size;
3202 	char dummy;
3203 
3204 	if (argc < 5) {
3205 		ti->error = "Not enough arguments";
3206 		return -EINVAL;
3207 	}
3208 
3209 	key_size = get_key_size(&argv[1]);
3210 	if (key_size < 0) {
3211 		ti->error = "Cannot parse key size";
3212 		return -EINVAL;
3213 	}
3214 
3215 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3216 	if (!cc) {
3217 		ti->error = "Cannot allocate encryption context";
3218 		return -ENOMEM;
3219 	}
3220 	cc->key_size = key_size;
3221 	cc->sector_size = (1 << SECTOR_SHIFT);
3222 	cc->sector_shift = 0;
3223 
3224 	ti->private = cc;
3225 
3226 	spin_lock(&dm_crypt_clients_lock);
3227 	dm_crypt_clients_n++;
3228 	crypt_calculate_pages_per_client();
3229 	spin_unlock(&dm_crypt_clients_lock);
3230 
3231 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3232 	if (ret < 0)
3233 		goto bad;
3234 
3235 	/* Optional parameters need to be read before cipher constructor */
3236 	if (argc > 5) {
3237 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3238 		if (ret)
3239 			goto bad;
3240 	}
3241 
3242 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3243 	if (ret < 0)
3244 		goto bad;
3245 
3246 	if (crypt_integrity_aead(cc)) {
3247 		cc->dmreq_start = sizeof(struct aead_request);
3248 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3249 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3250 	} else {
3251 		cc->dmreq_start = sizeof(struct skcipher_request);
3252 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3253 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3254 	}
3255 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3256 
3257 	if (align_mask < CRYPTO_MINALIGN) {
3258 		/* Allocate the padding exactly */
3259 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3260 				& align_mask;
3261 	} else {
3262 		/*
3263 		 * If the cipher requires greater alignment than kmalloc
3264 		 * alignment, we don't know the exact position of the
3265 		 * initialization vector. We must assume worst case.
3266 		 */
3267 		iv_size_padding = align_mask;
3268 	}
3269 
3270 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3271 	additional_req_size = sizeof(struct dm_crypt_request) +
3272 		iv_size_padding + cc->iv_size +
3273 		cc->iv_size +
3274 		sizeof(uint64_t) +
3275 		sizeof(unsigned int);
3276 
3277 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3278 	if (ret) {
3279 		ti->error = "Cannot allocate crypt request mempool";
3280 		goto bad;
3281 	}
3282 
3283 	cc->per_bio_data_size = ti->per_io_data_size =
3284 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3285 		      ARCH_DMA_MINALIGN);
3286 
3287 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3288 	if (ret) {
3289 		ti->error = "Cannot allocate page mempool";
3290 		goto bad;
3291 	}
3292 
3293 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3294 	if (ret) {
3295 		ti->error = "Cannot allocate crypt bioset";
3296 		goto bad;
3297 	}
3298 
3299 	mutex_init(&cc->bio_alloc_lock);
3300 
3301 	ret = -EINVAL;
3302 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3303 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3304 		ti->error = "Invalid iv_offset sector";
3305 		goto bad;
3306 	}
3307 	cc->iv_offset = tmpll;
3308 
3309 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3310 	if (ret) {
3311 		ti->error = "Device lookup failed";
3312 		goto bad;
3313 	}
3314 
3315 	ret = -EINVAL;
3316 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3317 		ti->error = "Invalid device sector";
3318 		goto bad;
3319 	}
3320 	cc->start = tmpll;
3321 
3322 	if (bdev_is_zoned(cc->dev->bdev)) {
3323 		/*
3324 		 * For zoned block devices, we need to preserve the issuer write
3325 		 * ordering. To do so, disable write workqueues and force inline
3326 		 * encryption completion.
3327 		 */
3328 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3329 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3330 
3331 		/*
3332 		 * All zone append writes to a zone of a zoned block device will
3333 		 * have the same BIO sector, the start of the zone. When the
3334 		 * cypher IV mode uses sector values, all data targeting a
3335 		 * zone will be encrypted using the first sector numbers of the
3336 		 * zone. This will not result in write errors but will
3337 		 * cause most reads to fail as reads will use the sector values
3338 		 * for the actual data locations, resulting in IV mismatch.
3339 		 * To avoid this problem, ask DM core to emulate zone append
3340 		 * operations with regular writes.
3341 		 */
3342 		DMDEBUG("Zone append operations will be emulated");
3343 		ti->emulate_zone_append = true;
3344 	}
3345 
3346 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3347 		ret = crypt_integrity_ctr(cc, ti);
3348 		if (ret)
3349 			goto bad;
3350 
3351 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3352 		if (!cc->tag_pool_max_sectors)
3353 			cc->tag_pool_max_sectors = 1;
3354 
3355 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3356 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3357 		if (ret) {
3358 			ti->error = "Cannot allocate integrity tags mempool";
3359 			goto bad;
3360 		}
3361 
3362 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3363 	}
3364 
3365 	ret = -ENOMEM;
3366 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3367 	if (!cc->io_queue) {
3368 		ti->error = "Couldn't create kcryptd io queue";
3369 		goto bad;
3370 	}
3371 
3372 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3373 		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3374 						  1, devname);
3375 	else
3376 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3377 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3378 						  num_online_cpus(), devname);
3379 	if (!cc->crypt_queue) {
3380 		ti->error = "Couldn't create kcryptd queue";
3381 		goto bad;
3382 	}
3383 
3384 	spin_lock_init(&cc->write_thread_lock);
3385 	cc->write_tree = RB_ROOT;
3386 
3387 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3388 	if (IS_ERR(cc->write_thread)) {
3389 		ret = PTR_ERR(cc->write_thread);
3390 		cc->write_thread = NULL;
3391 		ti->error = "Couldn't spawn write thread";
3392 		goto bad;
3393 	}
3394 
3395 	ti->num_flush_bios = 1;
3396 	ti->limit_swap_bios = true;
3397 	ti->accounts_remapped_io = true;
3398 
3399 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3400 	return 0;
3401 
3402 bad:
3403 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3404 	crypt_dtr(ti);
3405 	return ret;
3406 }
3407 
3408 static int crypt_map(struct dm_target *ti, struct bio *bio)
3409 {
3410 	struct dm_crypt_io *io;
3411 	struct crypt_config *cc = ti->private;
3412 
3413 	/*
3414 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3415 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3416 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3417 	 */
3418 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3419 	    bio_op(bio) == REQ_OP_DISCARD)) {
3420 		bio_set_dev(bio, cc->dev->bdev);
3421 		if (bio_sectors(bio))
3422 			bio->bi_iter.bi_sector = cc->start +
3423 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3424 		return DM_MAPIO_REMAPPED;
3425 	}
3426 
3427 	/*
3428 	 * Check if bio is too large, split as needed.
3429 	 */
3430 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3431 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3432 		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3433 
3434 	/*
3435 	 * Ensure that bio is a multiple of internal sector encryption size
3436 	 * and is aligned to this size as defined in IO hints.
3437 	 */
3438 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3439 		return DM_MAPIO_KILL;
3440 
3441 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3442 		return DM_MAPIO_KILL;
3443 
3444 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3445 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3446 
3447 	if (cc->on_disk_tag_size) {
3448 		unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3449 
3450 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3451 			io->integrity_metadata = NULL;
3452 		else
3453 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3454 
3455 		if (unlikely(!io->integrity_metadata)) {
3456 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3457 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3458 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3459 			io->integrity_metadata_from_pool = true;
3460 		}
3461 	}
3462 
3463 	if (crypt_integrity_aead(cc))
3464 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3465 	else
3466 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3467 
3468 	if (bio_data_dir(io->base_bio) == READ) {
3469 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3470 			kcryptd_queue_read(io);
3471 	} else
3472 		kcryptd_queue_crypt(io);
3473 
3474 	return DM_MAPIO_SUBMITTED;
3475 }
3476 
3477 static char hex2asc(unsigned char c)
3478 {
3479 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3480 }
3481 
3482 static void crypt_status(struct dm_target *ti, status_type_t type,
3483 			 unsigned int status_flags, char *result, unsigned int maxlen)
3484 {
3485 	struct crypt_config *cc = ti->private;
3486 	unsigned int i, sz = 0;
3487 	int num_feature_args = 0;
3488 
3489 	switch (type) {
3490 	case STATUSTYPE_INFO:
3491 		result[0] = '\0';
3492 		break;
3493 
3494 	case STATUSTYPE_TABLE:
3495 		DMEMIT("%s ", cc->cipher_string);
3496 
3497 		if (cc->key_size > 0) {
3498 			if (cc->key_string)
3499 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3500 			else {
3501 				for (i = 0; i < cc->key_size; i++) {
3502 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3503 					       hex2asc(cc->key[i] & 0xf));
3504 				}
3505 			}
3506 		} else
3507 			DMEMIT("-");
3508 
3509 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3510 				cc->dev->name, (unsigned long long)cc->start);
3511 
3512 		num_feature_args += !!ti->num_discard_bios;
3513 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3514 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3515 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3516 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3517 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3518 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3519 		if (cc->on_disk_tag_size)
3520 			num_feature_args++;
3521 		if (num_feature_args) {
3522 			DMEMIT(" %d", num_feature_args);
3523 			if (ti->num_discard_bios)
3524 				DMEMIT(" allow_discards");
3525 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3526 				DMEMIT(" same_cpu_crypt");
3527 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3528 				DMEMIT(" submit_from_crypt_cpus");
3529 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3530 				DMEMIT(" no_read_workqueue");
3531 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3532 				DMEMIT(" no_write_workqueue");
3533 			if (cc->on_disk_tag_size)
3534 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3535 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3536 				DMEMIT(" sector_size:%d", cc->sector_size);
3537 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3538 				DMEMIT(" iv_large_sectors");
3539 		}
3540 		break;
3541 
3542 	case STATUSTYPE_IMA:
3543 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3544 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3545 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3546 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3547 		       'y' : 'n');
3548 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3549 		       'y' : 'n');
3550 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3551 		       'y' : 'n');
3552 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3553 		       'y' : 'n');
3554 
3555 		if (cc->on_disk_tag_size)
3556 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3557 			       cc->on_disk_tag_size, cc->cipher_auth);
3558 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3559 			DMEMIT(",sector_size=%d", cc->sector_size);
3560 		if (cc->cipher_string)
3561 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3562 
3563 		DMEMIT(",key_size=%u", cc->key_size);
3564 		DMEMIT(",key_parts=%u", cc->key_parts);
3565 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3566 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3567 		DMEMIT(";");
3568 		break;
3569 	}
3570 }
3571 
3572 static void crypt_postsuspend(struct dm_target *ti)
3573 {
3574 	struct crypt_config *cc = ti->private;
3575 
3576 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3577 }
3578 
3579 static int crypt_preresume(struct dm_target *ti)
3580 {
3581 	struct crypt_config *cc = ti->private;
3582 
3583 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3584 		DMERR("aborting resume - crypt key is not set.");
3585 		return -EAGAIN;
3586 	}
3587 
3588 	return 0;
3589 }
3590 
3591 static void crypt_resume(struct dm_target *ti)
3592 {
3593 	struct crypt_config *cc = ti->private;
3594 
3595 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3596 }
3597 
3598 /* Message interface
3599  *	key set <key>
3600  *	key wipe
3601  */
3602 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3603 			 char *result, unsigned int maxlen)
3604 {
3605 	struct crypt_config *cc = ti->private;
3606 	int key_size, ret = -EINVAL;
3607 
3608 	if (argc < 2)
3609 		goto error;
3610 
3611 	if (!strcasecmp(argv[0], "key")) {
3612 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3613 			DMWARN("not suspended during key manipulation.");
3614 			return -EINVAL;
3615 		}
3616 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3617 			/* The key size may not be changed. */
3618 			key_size = get_key_size(&argv[2]);
3619 			if (key_size < 0 || cc->key_size != key_size) {
3620 				memset(argv[2], '0', strlen(argv[2]));
3621 				return -EINVAL;
3622 			}
3623 
3624 			ret = crypt_set_key(cc, argv[2]);
3625 			if (ret)
3626 				return ret;
3627 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3628 				ret = cc->iv_gen_ops->init(cc);
3629 			/* wipe the kernel key payload copy */
3630 			if (cc->key_string)
3631 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3632 			return ret;
3633 		}
3634 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3635 			return crypt_wipe_key(cc);
3636 	}
3637 
3638 error:
3639 	DMWARN("unrecognised message received.");
3640 	return -EINVAL;
3641 }
3642 
3643 static int crypt_iterate_devices(struct dm_target *ti,
3644 				 iterate_devices_callout_fn fn, void *data)
3645 {
3646 	struct crypt_config *cc = ti->private;
3647 
3648 	return fn(ti, cc->dev, cc->start, ti->len, data);
3649 }
3650 
3651 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3652 {
3653 	struct crypt_config *cc = ti->private;
3654 
3655 	/*
3656 	 * Unfortunate constraint that is required to avoid the potential
3657 	 * for exceeding underlying device's max_segments limits -- due to
3658 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3659 	 * bio that are not as physically contiguous as the original bio.
3660 	 */
3661 	limits->max_segment_size = PAGE_SIZE;
3662 
3663 	limits->logical_block_size =
3664 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3665 	limits->physical_block_size =
3666 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3667 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3668 	limits->dma_alignment = limits->logical_block_size - 1;
3669 }
3670 
3671 static struct target_type crypt_target = {
3672 	.name   = "crypt",
3673 	.version = {1, 24, 0},
3674 	.module = THIS_MODULE,
3675 	.ctr    = crypt_ctr,
3676 	.dtr    = crypt_dtr,
3677 	.features = DM_TARGET_ZONED_HM,
3678 	.report_zones = crypt_report_zones,
3679 	.map    = crypt_map,
3680 	.status = crypt_status,
3681 	.postsuspend = crypt_postsuspend,
3682 	.preresume = crypt_preresume,
3683 	.resume = crypt_resume,
3684 	.message = crypt_message,
3685 	.iterate_devices = crypt_iterate_devices,
3686 	.io_hints = crypt_io_hints,
3687 };
3688 module_dm(crypt);
3689 
3690 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3691 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3692 MODULE_LICENSE("GPL");
3693