xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 0bc92e7f)
1 /*
2  * Copyright (C) 2003 Jana Saout <jana@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
5  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
6  *
7  * This file is released under the GPL.
8  */
9 
10 #include <linux/completion.h>
11 #include <linux/err.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/key.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/crypto.h>
21 #include <linux/workqueue.h>
22 #include <linux/kthread.h>
23 #include <linux/backing-dev.h>
24 #include <linux/atomic.h>
25 #include <linux/scatterlist.h>
26 #include <linux/rbtree.h>
27 #include <linux/ctype.h>
28 #include <asm/page.h>
29 #include <asm/unaligned.h>
30 #include <crypto/hash.h>
31 #include <crypto/md5.h>
32 #include <crypto/algapi.h>
33 #include <crypto/skcipher.h>
34 #include <crypto/aead.h>
35 #include <crypto/authenc.h>
36 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
37 #include <linux/key-type.h>
38 #include <keys/user-type.h>
39 #include <keys/encrypted-type.h>
40 
41 #include <linux/device-mapper.h>
42 
43 #define DM_MSG_PREFIX "crypt"
44 
45 /*
46  * context holding the current state of a multi-part conversion
47  */
48 struct convert_context {
49 	struct completion restart;
50 	struct bio *bio_in;
51 	struct bio *bio_out;
52 	struct bvec_iter iter_in;
53 	struct bvec_iter iter_out;
54 	u64 cc_sector;
55 	atomic_t cc_pending;
56 	union {
57 		struct skcipher_request *req;
58 		struct aead_request *req_aead;
59 	} r;
60 
61 };
62 
63 /*
64  * per bio private data
65  */
66 struct dm_crypt_io {
67 	struct crypt_config *cc;
68 	struct bio *base_bio;
69 	u8 *integrity_metadata;
70 	bool integrity_metadata_from_pool;
71 	struct work_struct work;
72 	struct tasklet_struct tasklet;
73 
74 	struct convert_context ctx;
75 
76 	atomic_t io_pending;
77 	blk_status_t error;
78 	sector_t sector;
79 
80 	struct rb_node rb_node;
81 } CRYPTO_MINALIGN_ATTR;
82 
83 struct dm_crypt_request {
84 	struct convert_context *ctx;
85 	struct scatterlist sg_in[4];
86 	struct scatterlist sg_out[4];
87 	u64 iv_sector;
88 };
89 
90 struct crypt_config;
91 
92 struct crypt_iv_operations {
93 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
94 		   const char *opts);
95 	void (*dtr)(struct crypt_config *cc);
96 	int (*init)(struct crypt_config *cc);
97 	int (*wipe)(struct crypt_config *cc);
98 	int (*generator)(struct crypt_config *cc, u8 *iv,
99 			 struct dm_crypt_request *dmreq);
100 	int (*post)(struct crypt_config *cc, u8 *iv,
101 		    struct dm_crypt_request *dmreq);
102 };
103 
104 struct iv_benbi_private {
105 	int shift;
106 };
107 
108 #define LMK_SEED_SIZE 64 /* hash + 0 */
109 struct iv_lmk_private {
110 	struct crypto_shash *hash_tfm;
111 	u8 *seed;
112 };
113 
114 #define TCW_WHITENING_SIZE 16
115 struct iv_tcw_private {
116 	struct crypto_shash *crc32_tfm;
117 	u8 *iv_seed;
118 	u8 *whitening;
119 };
120 
121 #define ELEPHANT_MAX_KEY_SIZE 32
122 struct iv_elephant_private {
123 	struct crypto_skcipher *tfm;
124 };
125 
126 /*
127  * Crypt: maps a linear range of a block device
128  * and encrypts / decrypts at the same time.
129  */
130 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
131 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
132 	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
133 	     DM_CRYPT_WRITE_INLINE };
134 
135 enum cipher_flags {
136 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cihper */
137 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
138 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
139 };
140 
141 /*
142  * The fields in here must be read only after initialization.
143  */
144 struct crypt_config {
145 	struct dm_dev *dev;
146 	sector_t start;
147 
148 	struct percpu_counter n_allocated_pages;
149 
150 	struct workqueue_struct *io_queue;
151 	struct workqueue_struct *crypt_queue;
152 
153 	spinlock_t write_thread_lock;
154 	struct task_struct *write_thread;
155 	struct rb_root write_tree;
156 
157 	char *cipher_string;
158 	char *cipher_auth;
159 	char *key_string;
160 
161 	const struct crypt_iv_operations *iv_gen_ops;
162 	union {
163 		struct iv_benbi_private benbi;
164 		struct iv_lmk_private lmk;
165 		struct iv_tcw_private tcw;
166 		struct iv_elephant_private elephant;
167 	} iv_gen_private;
168 	u64 iv_offset;
169 	unsigned int iv_size;
170 	unsigned short int sector_size;
171 	unsigned char sector_shift;
172 
173 	union {
174 		struct crypto_skcipher **tfms;
175 		struct crypto_aead **tfms_aead;
176 	} cipher_tfm;
177 	unsigned tfms_count;
178 	unsigned long cipher_flags;
179 
180 	/*
181 	 * Layout of each crypto request:
182 	 *
183 	 *   struct skcipher_request
184 	 *      context
185 	 *      padding
186 	 *   struct dm_crypt_request
187 	 *      padding
188 	 *   IV
189 	 *
190 	 * The padding is added so that dm_crypt_request and the IV are
191 	 * correctly aligned.
192 	 */
193 	unsigned int dmreq_start;
194 
195 	unsigned int per_bio_data_size;
196 
197 	unsigned long flags;
198 	unsigned int key_size;
199 	unsigned int key_parts;      /* independent parts in key buffer */
200 	unsigned int key_extra_size; /* additional keys length */
201 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
202 
203 	unsigned int integrity_tag_size;
204 	unsigned int integrity_iv_size;
205 	unsigned int on_disk_tag_size;
206 
207 	/*
208 	 * pool for per bio private data, crypto requests,
209 	 * encryption requeusts/buffer pages and integrity tags
210 	 */
211 	unsigned tag_pool_max_sectors;
212 	mempool_t tag_pool;
213 	mempool_t req_pool;
214 	mempool_t page_pool;
215 
216 	struct bio_set bs;
217 	struct mutex bio_alloc_lock;
218 
219 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
220 	u8 key[];
221 };
222 
223 #define MIN_IOS		64
224 #define MAX_TAG_SIZE	480
225 #define POOL_ENTRY_SIZE	512
226 
227 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
228 static unsigned dm_crypt_clients_n = 0;
229 static volatile unsigned long dm_crypt_pages_per_client;
230 #define DM_CRYPT_MEMORY_PERCENT			2
231 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_PAGES * 16)
232 
233 static void clone_init(struct dm_crypt_io *, struct bio *);
234 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
235 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
236 					     struct scatterlist *sg);
237 
238 static bool crypt_integrity_aead(struct crypt_config *cc);
239 
240 /*
241  * Use this to access cipher attributes that are independent of the key.
242  */
243 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
244 {
245 	return cc->cipher_tfm.tfms[0];
246 }
247 
248 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
249 {
250 	return cc->cipher_tfm.tfms_aead[0];
251 }
252 
253 /*
254  * Different IV generation algorithms:
255  *
256  * plain: the initial vector is the 32-bit little-endian version of the sector
257  *        number, padded with zeros if necessary.
258  *
259  * plain64: the initial vector is the 64-bit little-endian version of the sector
260  *        number, padded with zeros if necessary.
261  *
262  * plain64be: the initial vector is the 64-bit big-endian version of the sector
263  *        number, padded with zeros if necessary.
264  *
265  * essiv: "encrypted sector|salt initial vector", the sector number is
266  *        encrypted with the bulk cipher using a salt as key. The salt
267  *        should be derived from the bulk cipher's key via hashing.
268  *
269  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
270  *        (needed for LRW-32-AES and possible other narrow block modes)
271  *
272  * null: the initial vector is always zero.  Provides compatibility with
273  *       obsolete loop_fish2 devices.  Do not use for new devices.
274  *
275  * lmk:  Compatible implementation of the block chaining mode used
276  *       by the Loop-AES block device encryption system
277  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
278  *       It operates on full 512 byte sectors and uses CBC
279  *       with an IV derived from the sector number, the data and
280  *       optionally extra IV seed.
281  *       This means that after decryption the first block
282  *       of sector must be tweaked according to decrypted data.
283  *       Loop-AES can use three encryption schemes:
284  *         version 1: is plain aes-cbc mode
285  *         version 2: uses 64 multikey scheme with lmk IV generator
286  *         version 3: the same as version 2 with additional IV seed
287  *                   (it uses 65 keys, last key is used as IV seed)
288  *
289  * tcw:  Compatible implementation of the block chaining mode used
290  *       by the TrueCrypt device encryption system (prior to version 4.1).
291  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
292  *       It operates on full 512 byte sectors and uses CBC
293  *       with an IV derived from initial key and the sector number.
294  *       In addition, whitening value is applied on every sector, whitening
295  *       is calculated from initial key, sector number and mixed using CRC32.
296  *       Note that this encryption scheme is vulnerable to watermarking attacks
297  *       and should be used for old compatible containers access only.
298  *
299  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
300  *        The IV is encrypted little-endian byte-offset (with the same key
301  *        and cipher as the volume).
302  *
303  * elephant: The extended version of eboiv with additional Elephant diffuser
304  *           used with Bitlocker CBC mode.
305  *           This mode was used in older Windows systems
306  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
307  */
308 
309 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
310 			      struct dm_crypt_request *dmreq)
311 {
312 	memset(iv, 0, cc->iv_size);
313 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
314 
315 	return 0;
316 }
317 
318 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
319 				struct dm_crypt_request *dmreq)
320 {
321 	memset(iv, 0, cc->iv_size);
322 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
323 
324 	return 0;
325 }
326 
327 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
328 				  struct dm_crypt_request *dmreq)
329 {
330 	memset(iv, 0, cc->iv_size);
331 	/* iv_size is at least of size u64; usually it is 16 bytes */
332 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
333 
334 	return 0;
335 }
336 
337 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
338 			      struct dm_crypt_request *dmreq)
339 {
340 	/*
341 	 * ESSIV encryption of the IV is now handled by the crypto API,
342 	 * so just pass the plain sector number here.
343 	 */
344 	memset(iv, 0, cc->iv_size);
345 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
346 
347 	return 0;
348 }
349 
350 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
351 			      const char *opts)
352 {
353 	unsigned bs;
354 	int log;
355 
356 	if (crypt_integrity_aead(cc))
357 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
358 	else
359 		bs = crypto_skcipher_blocksize(any_tfm(cc));
360 	log = ilog2(bs);
361 
362 	/* we need to calculate how far we must shift the sector count
363 	 * to get the cipher block count, we use this shift in _gen */
364 
365 	if (1 << log != bs) {
366 		ti->error = "cypher blocksize is not a power of 2";
367 		return -EINVAL;
368 	}
369 
370 	if (log > 9) {
371 		ti->error = "cypher blocksize is > 512";
372 		return -EINVAL;
373 	}
374 
375 	cc->iv_gen_private.benbi.shift = 9 - log;
376 
377 	return 0;
378 }
379 
380 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
381 {
382 }
383 
384 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
385 			      struct dm_crypt_request *dmreq)
386 {
387 	__be64 val;
388 
389 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
390 
391 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
392 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
393 
394 	return 0;
395 }
396 
397 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
398 			     struct dm_crypt_request *dmreq)
399 {
400 	memset(iv, 0, cc->iv_size);
401 
402 	return 0;
403 }
404 
405 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
406 {
407 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
408 
409 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
410 		crypto_free_shash(lmk->hash_tfm);
411 	lmk->hash_tfm = NULL;
412 
413 	kfree_sensitive(lmk->seed);
414 	lmk->seed = NULL;
415 }
416 
417 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
418 			    const char *opts)
419 {
420 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
421 
422 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
423 		ti->error = "Unsupported sector size for LMK";
424 		return -EINVAL;
425 	}
426 
427 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
428 					   CRYPTO_ALG_ALLOCATES_MEMORY);
429 	if (IS_ERR(lmk->hash_tfm)) {
430 		ti->error = "Error initializing LMK hash";
431 		return PTR_ERR(lmk->hash_tfm);
432 	}
433 
434 	/* No seed in LMK version 2 */
435 	if (cc->key_parts == cc->tfms_count) {
436 		lmk->seed = NULL;
437 		return 0;
438 	}
439 
440 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
441 	if (!lmk->seed) {
442 		crypt_iv_lmk_dtr(cc);
443 		ti->error = "Error kmallocing seed storage in LMK";
444 		return -ENOMEM;
445 	}
446 
447 	return 0;
448 }
449 
450 static int crypt_iv_lmk_init(struct crypt_config *cc)
451 {
452 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
453 	int subkey_size = cc->key_size / cc->key_parts;
454 
455 	/* LMK seed is on the position of LMK_KEYS + 1 key */
456 	if (lmk->seed)
457 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
458 		       crypto_shash_digestsize(lmk->hash_tfm));
459 
460 	return 0;
461 }
462 
463 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
464 {
465 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
466 
467 	if (lmk->seed)
468 		memset(lmk->seed, 0, LMK_SEED_SIZE);
469 
470 	return 0;
471 }
472 
473 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
474 			    struct dm_crypt_request *dmreq,
475 			    u8 *data)
476 {
477 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
478 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
479 	struct md5_state md5state;
480 	__le32 buf[4];
481 	int i, r;
482 
483 	desc->tfm = lmk->hash_tfm;
484 
485 	r = crypto_shash_init(desc);
486 	if (r)
487 		return r;
488 
489 	if (lmk->seed) {
490 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
491 		if (r)
492 			return r;
493 	}
494 
495 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
496 	r = crypto_shash_update(desc, data + 16, 16 * 31);
497 	if (r)
498 		return r;
499 
500 	/* Sector is cropped to 56 bits here */
501 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
502 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
503 	buf[2] = cpu_to_le32(4024);
504 	buf[3] = 0;
505 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
506 	if (r)
507 		return r;
508 
509 	/* No MD5 padding here */
510 	r = crypto_shash_export(desc, &md5state);
511 	if (r)
512 		return r;
513 
514 	for (i = 0; i < MD5_HASH_WORDS; i++)
515 		__cpu_to_le32s(&md5state.hash[i]);
516 	memcpy(iv, &md5state.hash, cc->iv_size);
517 
518 	return 0;
519 }
520 
521 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
522 			    struct dm_crypt_request *dmreq)
523 {
524 	struct scatterlist *sg;
525 	u8 *src;
526 	int r = 0;
527 
528 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
529 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
530 		src = kmap_atomic(sg_page(sg));
531 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
532 		kunmap_atomic(src);
533 	} else
534 		memset(iv, 0, cc->iv_size);
535 
536 	return r;
537 }
538 
539 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
540 			     struct dm_crypt_request *dmreq)
541 {
542 	struct scatterlist *sg;
543 	u8 *dst;
544 	int r;
545 
546 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
547 		return 0;
548 
549 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
550 	dst = kmap_atomic(sg_page(sg));
551 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
552 
553 	/* Tweak the first block of plaintext sector */
554 	if (!r)
555 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
556 
557 	kunmap_atomic(dst);
558 	return r;
559 }
560 
561 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
562 {
563 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
564 
565 	kfree_sensitive(tcw->iv_seed);
566 	tcw->iv_seed = NULL;
567 	kfree_sensitive(tcw->whitening);
568 	tcw->whitening = NULL;
569 
570 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
571 		crypto_free_shash(tcw->crc32_tfm);
572 	tcw->crc32_tfm = NULL;
573 }
574 
575 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
576 			    const char *opts)
577 {
578 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
579 
580 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
581 		ti->error = "Unsupported sector size for TCW";
582 		return -EINVAL;
583 	}
584 
585 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
586 		ti->error = "Wrong key size for TCW";
587 		return -EINVAL;
588 	}
589 
590 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
591 					    CRYPTO_ALG_ALLOCATES_MEMORY);
592 	if (IS_ERR(tcw->crc32_tfm)) {
593 		ti->error = "Error initializing CRC32 in TCW";
594 		return PTR_ERR(tcw->crc32_tfm);
595 	}
596 
597 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
598 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
599 	if (!tcw->iv_seed || !tcw->whitening) {
600 		crypt_iv_tcw_dtr(cc);
601 		ti->error = "Error allocating seed storage in TCW";
602 		return -ENOMEM;
603 	}
604 
605 	return 0;
606 }
607 
608 static int crypt_iv_tcw_init(struct crypt_config *cc)
609 {
610 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
611 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
612 
613 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
614 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
615 	       TCW_WHITENING_SIZE);
616 
617 	return 0;
618 }
619 
620 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
621 {
622 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
623 
624 	memset(tcw->iv_seed, 0, cc->iv_size);
625 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
626 
627 	return 0;
628 }
629 
630 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
631 				  struct dm_crypt_request *dmreq,
632 				  u8 *data)
633 {
634 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
635 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
636 	u8 buf[TCW_WHITENING_SIZE];
637 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
638 	int i, r;
639 
640 	/* xor whitening with sector number */
641 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
642 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
643 
644 	/* calculate crc32 for every 32bit part and xor it */
645 	desc->tfm = tcw->crc32_tfm;
646 	for (i = 0; i < 4; i++) {
647 		r = crypto_shash_init(desc);
648 		if (r)
649 			goto out;
650 		r = crypto_shash_update(desc, &buf[i * 4], 4);
651 		if (r)
652 			goto out;
653 		r = crypto_shash_final(desc, &buf[i * 4]);
654 		if (r)
655 			goto out;
656 	}
657 	crypto_xor(&buf[0], &buf[12], 4);
658 	crypto_xor(&buf[4], &buf[8], 4);
659 
660 	/* apply whitening (8 bytes) to whole sector */
661 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
662 		crypto_xor(data + i * 8, buf, 8);
663 out:
664 	memzero_explicit(buf, sizeof(buf));
665 	return r;
666 }
667 
668 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
669 			    struct dm_crypt_request *dmreq)
670 {
671 	struct scatterlist *sg;
672 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
673 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
674 	u8 *src;
675 	int r = 0;
676 
677 	/* Remove whitening from ciphertext */
678 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
679 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
680 		src = kmap_atomic(sg_page(sg));
681 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
682 		kunmap_atomic(src);
683 	}
684 
685 	/* Calculate IV */
686 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
687 	if (cc->iv_size > 8)
688 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
689 			       cc->iv_size - 8);
690 
691 	return r;
692 }
693 
694 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
695 			     struct dm_crypt_request *dmreq)
696 {
697 	struct scatterlist *sg;
698 	u8 *dst;
699 	int r;
700 
701 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
702 		return 0;
703 
704 	/* Apply whitening on ciphertext */
705 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
706 	dst = kmap_atomic(sg_page(sg));
707 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
708 	kunmap_atomic(dst);
709 
710 	return r;
711 }
712 
713 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
714 				struct dm_crypt_request *dmreq)
715 {
716 	/* Used only for writes, there must be an additional space to store IV */
717 	get_random_bytes(iv, cc->iv_size);
718 	return 0;
719 }
720 
721 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
722 			    const char *opts)
723 {
724 	if (crypt_integrity_aead(cc)) {
725 		ti->error = "AEAD transforms not supported for EBOIV";
726 		return -EINVAL;
727 	}
728 
729 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
730 		ti->error = "Block size of EBOIV cipher does "
731 			    "not match IV size of block cipher";
732 		return -EINVAL;
733 	}
734 
735 	return 0;
736 }
737 
738 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
739 			    struct dm_crypt_request *dmreq)
740 {
741 	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
742 	struct skcipher_request *req;
743 	struct scatterlist src, dst;
744 	DECLARE_CRYPTO_WAIT(wait);
745 	int err;
746 
747 	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
748 	if (!req)
749 		return -ENOMEM;
750 
751 	memset(buf, 0, cc->iv_size);
752 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
753 
754 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
755 	sg_init_one(&dst, iv, cc->iv_size);
756 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
757 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
758 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
759 	skcipher_request_free(req);
760 
761 	return err;
762 }
763 
764 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
765 {
766 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
767 
768 	crypto_free_skcipher(elephant->tfm);
769 	elephant->tfm = NULL;
770 }
771 
772 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
773 			    const char *opts)
774 {
775 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
776 	int r;
777 
778 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
779 					      CRYPTO_ALG_ALLOCATES_MEMORY);
780 	if (IS_ERR(elephant->tfm)) {
781 		r = PTR_ERR(elephant->tfm);
782 		elephant->tfm = NULL;
783 		return r;
784 	}
785 
786 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
787 	if (r)
788 		crypt_iv_elephant_dtr(cc);
789 	return r;
790 }
791 
792 static void diffuser_disk_to_cpu(u32 *d, size_t n)
793 {
794 #ifndef __LITTLE_ENDIAN
795 	int i;
796 
797 	for (i = 0; i < n; i++)
798 		d[i] = le32_to_cpu((__le32)d[i]);
799 #endif
800 }
801 
802 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
803 {
804 #ifndef __LITTLE_ENDIAN
805 	int i;
806 
807 	for (i = 0; i < n; i++)
808 		d[i] = cpu_to_le32((u32)d[i]);
809 #endif
810 }
811 
812 static void diffuser_a_decrypt(u32 *d, size_t n)
813 {
814 	int i, i1, i2, i3;
815 
816 	for (i = 0; i < 5; i++) {
817 		i1 = 0;
818 		i2 = n - 2;
819 		i3 = n - 5;
820 
821 		while (i1 < (n - 1)) {
822 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
823 			i1++; i2++; i3++;
824 
825 			if (i3 >= n)
826 				i3 -= n;
827 
828 			d[i1] += d[i2] ^ d[i3];
829 			i1++; i2++; i3++;
830 
831 			if (i2 >= n)
832 				i2 -= n;
833 
834 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
835 			i1++; i2++; i3++;
836 
837 			d[i1] += d[i2] ^ d[i3];
838 			i1++; i2++; i3++;
839 		}
840 	}
841 }
842 
843 static void diffuser_a_encrypt(u32 *d, size_t n)
844 {
845 	int i, i1, i2, i3;
846 
847 	for (i = 0; i < 5; i++) {
848 		i1 = n - 1;
849 		i2 = n - 2 - 1;
850 		i3 = n - 5 - 1;
851 
852 		while (i1 > 0) {
853 			d[i1] -= d[i2] ^ d[i3];
854 			i1--; i2--; i3--;
855 
856 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
857 			i1--; i2--; i3--;
858 
859 			if (i2 < 0)
860 				i2 += n;
861 
862 			d[i1] -= d[i2] ^ d[i3];
863 			i1--; i2--; i3--;
864 
865 			if (i3 < 0)
866 				i3 += n;
867 
868 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
869 			i1--; i2--; i3--;
870 		}
871 	}
872 }
873 
874 static void diffuser_b_decrypt(u32 *d, size_t n)
875 {
876 	int i, i1, i2, i3;
877 
878 	for (i = 0; i < 3; i++) {
879 		i1 = 0;
880 		i2 = 2;
881 		i3 = 5;
882 
883 		while (i1 < (n - 1)) {
884 			d[i1] += d[i2] ^ d[i3];
885 			i1++; i2++; i3++;
886 
887 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
888 			i1++; i2++; i3++;
889 
890 			if (i2 >= n)
891 				i2 -= n;
892 
893 			d[i1] += d[i2] ^ d[i3];
894 			i1++; i2++; i3++;
895 
896 			if (i3 >= n)
897 				i3 -= n;
898 
899 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
900 			i1++; i2++; i3++;
901 		}
902 	}
903 }
904 
905 static void diffuser_b_encrypt(u32 *d, size_t n)
906 {
907 	int i, i1, i2, i3;
908 
909 	for (i = 0; i < 3; i++) {
910 		i1 = n - 1;
911 		i2 = 2 - 1;
912 		i3 = 5 - 1;
913 
914 		while (i1 > 0) {
915 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
916 			i1--; i2--; i3--;
917 
918 			if (i3 < 0)
919 				i3 += n;
920 
921 			d[i1] -= d[i2] ^ d[i3];
922 			i1--; i2--; i3--;
923 
924 			if (i2 < 0)
925 				i2 += n;
926 
927 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
928 			i1--; i2--; i3--;
929 
930 			d[i1] -= d[i2] ^ d[i3];
931 			i1--; i2--; i3--;
932 		}
933 	}
934 }
935 
936 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
937 {
938 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
939 	u8 *es, *ks, *data, *data2, *data_offset;
940 	struct skcipher_request *req;
941 	struct scatterlist *sg, *sg2, src, dst;
942 	DECLARE_CRYPTO_WAIT(wait);
943 	int i, r;
944 
945 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
946 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
947 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
948 
949 	if (!req || !es || !ks) {
950 		r = -ENOMEM;
951 		goto out;
952 	}
953 
954 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
955 
956 	/* E(Ks, e(s)) */
957 	sg_init_one(&src, es, 16);
958 	sg_init_one(&dst, ks, 16);
959 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
960 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
961 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
962 	if (r)
963 		goto out;
964 
965 	/* E(Ks, e'(s)) */
966 	es[15] = 0x80;
967 	sg_init_one(&dst, &ks[16], 16);
968 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
969 	if (r)
970 		goto out;
971 
972 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
973 	data = kmap_atomic(sg_page(sg));
974 	data_offset = data + sg->offset;
975 
976 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
977 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
978 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
979 		data2 = kmap_atomic(sg_page(sg2));
980 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
981 		kunmap_atomic(data2);
982 	}
983 
984 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
985 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
986 		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
987 		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
988 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
989 	}
990 
991 	for (i = 0; i < (cc->sector_size / 32); i++)
992 		crypto_xor(data_offset + i * 32, ks, 32);
993 
994 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
995 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
996 		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
997 		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
998 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
999 	}
1000 
1001 	kunmap_atomic(data);
1002 out:
1003 	kfree_sensitive(ks);
1004 	kfree_sensitive(es);
1005 	skcipher_request_free(req);
1006 	return r;
1007 }
1008 
1009 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1010 			    struct dm_crypt_request *dmreq)
1011 {
1012 	int r;
1013 
1014 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1015 		r = crypt_iv_elephant(cc, dmreq);
1016 		if (r)
1017 			return r;
1018 	}
1019 
1020 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1021 }
1022 
1023 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1024 				  struct dm_crypt_request *dmreq)
1025 {
1026 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1027 		return crypt_iv_elephant(cc, dmreq);
1028 
1029 	return 0;
1030 }
1031 
1032 static int crypt_iv_elephant_init(struct crypt_config *cc)
1033 {
1034 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1035 	int key_offset = cc->key_size - cc->key_extra_size;
1036 
1037 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1038 }
1039 
1040 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1041 {
1042 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1043 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1044 
1045 	memset(key, 0, cc->key_extra_size);
1046 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1047 }
1048 
1049 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1050 	.generator = crypt_iv_plain_gen
1051 };
1052 
1053 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1054 	.generator = crypt_iv_plain64_gen
1055 };
1056 
1057 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1058 	.generator = crypt_iv_plain64be_gen
1059 };
1060 
1061 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1062 	.generator = crypt_iv_essiv_gen
1063 };
1064 
1065 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1066 	.ctr	   = crypt_iv_benbi_ctr,
1067 	.dtr	   = crypt_iv_benbi_dtr,
1068 	.generator = crypt_iv_benbi_gen
1069 };
1070 
1071 static const struct crypt_iv_operations crypt_iv_null_ops = {
1072 	.generator = crypt_iv_null_gen
1073 };
1074 
1075 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1076 	.ctr	   = crypt_iv_lmk_ctr,
1077 	.dtr	   = crypt_iv_lmk_dtr,
1078 	.init	   = crypt_iv_lmk_init,
1079 	.wipe	   = crypt_iv_lmk_wipe,
1080 	.generator = crypt_iv_lmk_gen,
1081 	.post	   = crypt_iv_lmk_post
1082 };
1083 
1084 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1085 	.ctr	   = crypt_iv_tcw_ctr,
1086 	.dtr	   = crypt_iv_tcw_dtr,
1087 	.init	   = crypt_iv_tcw_init,
1088 	.wipe	   = crypt_iv_tcw_wipe,
1089 	.generator = crypt_iv_tcw_gen,
1090 	.post	   = crypt_iv_tcw_post
1091 };
1092 
1093 static const struct crypt_iv_operations crypt_iv_random_ops = {
1094 	.generator = crypt_iv_random_gen
1095 };
1096 
1097 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1098 	.ctr	   = crypt_iv_eboiv_ctr,
1099 	.generator = crypt_iv_eboiv_gen
1100 };
1101 
1102 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1103 	.ctr	   = crypt_iv_elephant_ctr,
1104 	.dtr	   = crypt_iv_elephant_dtr,
1105 	.init	   = crypt_iv_elephant_init,
1106 	.wipe	   = crypt_iv_elephant_wipe,
1107 	.generator = crypt_iv_elephant_gen,
1108 	.post	   = crypt_iv_elephant_post
1109 };
1110 
1111 /*
1112  * Integrity extensions
1113  */
1114 static bool crypt_integrity_aead(struct crypt_config *cc)
1115 {
1116 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1117 }
1118 
1119 static bool crypt_integrity_hmac(struct crypt_config *cc)
1120 {
1121 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1122 }
1123 
1124 /* Get sg containing data */
1125 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1126 					     struct scatterlist *sg)
1127 {
1128 	if (unlikely(crypt_integrity_aead(cc)))
1129 		return &sg[2];
1130 
1131 	return sg;
1132 }
1133 
1134 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1135 {
1136 	struct bio_integrity_payload *bip;
1137 	unsigned int tag_len;
1138 	int ret;
1139 
1140 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1141 		return 0;
1142 
1143 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1144 	if (IS_ERR(bip))
1145 		return PTR_ERR(bip);
1146 
1147 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1148 
1149 	bip->bip_iter.bi_size = tag_len;
1150 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1151 
1152 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1153 				     tag_len, offset_in_page(io->integrity_metadata));
1154 	if (unlikely(ret != tag_len))
1155 		return -ENOMEM;
1156 
1157 	return 0;
1158 }
1159 
1160 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1161 {
1162 #ifdef CONFIG_BLK_DEV_INTEGRITY
1163 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1164 	struct mapped_device *md = dm_table_get_md(ti->table);
1165 
1166 	/* From now we require underlying device with our integrity profile */
1167 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1168 		ti->error = "Integrity profile not supported.";
1169 		return -EINVAL;
1170 	}
1171 
1172 	if (bi->tag_size != cc->on_disk_tag_size ||
1173 	    bi->tuple_size != cc->on_disk_tag_size) {
1174 		ti->error = "Integrity profile tag size mismatch.";
1175 		return -EINVAL;
1176 	}
1177 	if (1 << bi->interval_exp != cc->sector_size) {
1178 		ti->error = "Integrity profile sector size mismatch.";
1179 		return -EINVAL;
1180 	}
1181 
1182 	if (crypt_integrity_aead(cc)) {
1183 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1184 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1185 		       cc->integrity_tag_size, cc->integrity_iv_size);
1186 
1187 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1188 			ti->error = "Integrity AEAD auth tag size is not supported.";
1189 			return -EINVAL;
1190 		}
1191 	} else if (cc->integrity_iv_size)
1192 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1193 		       cc->integrity_iv_size);
1194 
1195 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1196 		ti->error = "Not enough space for integrity tag in the profile.";
1197 		return -EINVAL;
1198 	}
1199 
1200 	return 0;
1201 #else
1202 	ti->error = "Integrity profile not supported.";
1203 	return -EINVAL;
1204 #endif
1205 }
1206 
1207 static void crypt_convert_init(struct crypt_config *cc,
1208 			       struct convert_context *ctx,
1209 			       struct bio *bio_out, struct bio *bio_in,
1210 			       sector_t sector)
1211 {
1212 	ctx->bio_in = bio_in;
1213 	ctx->bio_out = bio_out;
1214 	if (bio_in)
1215 		ctx->iter_in = bio_in->bi_iter;
1216 	if (bio_out)
1217 		ctx->iter_out = bio_out->bi_iter;
1218 	ctx->cc_sector = sector + cc->iv_offset;
1219 	init_completion(&ctx->restart);
1220 }
1221 
1222 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1223 					     void *req)
1224 {
1225 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1226 }
1227 
1228 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1229 {
1230 	return (void *)((char *)dmreq - cc->dmreq_start);
1231 }
1232 
1233 static u8 *iv_of_dmreq(struct crypt_config *cc,
1234 		       struct dm_crypt_request *dmreq)
1235 {
1236 	if (crypt_integrity_aead(cc))
1237 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1238 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1239 	else
1240 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1241 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1242 }
1243 
1244 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1245 		       struct dm_crypt_request *dmreq)
1246 {
1247 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1248 }
1249 
1250 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1251 		       struct dm_crypt_request *dmreq)
1252 {
1253 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1254 	return (__le64 *) ptr;
1255 }
1256 
1257 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1258 		       struct dm_crypt_request *dmreq)
1259 {
1260 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1261 		  cc->iv_size + sizeof(uint64_t);
1262 	return (unsigned int*)ptr;
1263 }
1264 
1265 static void *tag_from_dmreq(struct crypt_config *cc,
1266 				struct dm_crypt_request *dmreq)
1267 {
1268 	struct convert_context *ctx = dmreq->ctx;
1269 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1270 
1271 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1272 		cc->on_disk_tag_size];
1273 }
1274 
1275 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1276 			       struct dm_crypt_request *dmreq)
1277 {
1278 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1279 }
1280 
1281 static int crypt_convert_block_aead(struct crypt_config *cc,
1282 				     struct convert_context *ctx,
1283 				     struct aead_request *req,
1284 				     unsigned int tag_offset)
1285 {
1286 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1287 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1288 	struct dm_crypt_request *dmreq;
1289 	u8 *iv, *org_iv, *tag_iv, *tag;
1290 	__le64 *sector;
1291 	int r = 0;
1292 
1293 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1294 
1295 	/* Reject unexpected unaligned bio. */
1296 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1297 		return -EIO;
1298 
1299 	dmreq = dmreq_of_req(cc, req);
1300 	dmreq->iv_sector = ctx->cc_sector;
1301 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1302 		dmreq->iv_sector >>= cc->sector_shift;
1303 	dmreq->ctx = ctx;
1304 
1305 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1306 
1307 	sector = org_sector_of_dmreq(cc, dmreq);
1308 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1309 
1310 	iv = iv_of_dmreq(cc, dmreq);
1311 	org_iv = org_iv_of_dmreq(cc, dmreq);
1312 	tag = tag_from_dmreq(cc, dmreq);
1313 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1314 
1315 	/* AEAD request:
1316 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1317 	 *  | (authenticated) | (auth+encryption) |              |
1318 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1319 	 */
1320 	sg_init_table(dmreq->sg_in, 4);
1321 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1322 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1323 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1324 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1325 
1326 	sg_init_table(dmreq->sg_out, 4);
1327 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1328 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1329 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1330 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1331 
1332 	if (cc->iv_gen_ops) {
1333 		/* For READs use IV stored in integrity metadata */
1334 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1335 			memcpy(org_iv, tag_iv, cc->iv_size);
1336 		} else {
1337 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1338 			if (r < 0)
1339 				return r;
1340 			/* Store generated IV in integrity metadata */
1341 			if (cc->integrity_iv_size)
1342 				memcpy(tag_iv, org_iv, cc->iv_size);
1343 		}
1344 		/* Working copy of IV, to be modified in crypto API */
1345 		memcpy(iv, org_iv, cc->iv_size);
1346 	}
1347 
1348 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1349 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1350 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1351 				       cc->sector_size, iv);
1352 		r = crypto_aead_encrypt(req);
1353 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1354 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1355 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1356 	} else {
1357 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1358 				       cc->sector_size + cc->integrity_tag_size, iv);
1359 		r = crypto_aead_decrypt(req);
1360 	}
1361 
1362 	if (r == -EBADMSG) {
1363 		char b[BDEVNAME_SIZE];
1364 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1365 			    (unsigned long long)le64_to_cpu(*sector));
1366 	}
1367 
1368 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1369 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1370 
1371 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1372 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1373 
1374 	return r;
1375 }
1376 
1377 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1378 					struct convert_context *ctx,
1379 					struct skcipher_request *req,
1380 					unsigned int tag_offset)
1381 {
1382 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1383 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1384 	struct scatterlist *sg_in, *sg_out;
1385 	struct dm_crypt_request *dmreq;
1386 	u8 *iv, *org_iv, *tag_iv;
1387 	__le64 *sector;
1388 	int r = 0;
1389 
1390 	/* Reject unexpected unaligned bio. */
1391 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1392 		return -EIO;
1393 
1394 	dmreq = dmreq_of_req(cc, req);
1395 	dmreq->iv_sector = ctx->cc_sector;
1396 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1397 		dmreq->iv_sector >>= cc->sector_shift;
1398 	dmreq->ctx = ctx;
1399 
1400 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1401 
1402 	iv = iv_of_dmreq(cc, dmreq);
1403 	org_iv = org_iv_of_dmreq(cc, dmreq);
1404 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1405 
1406 	sector = org_sector_of_dmreq(cc, dmreq);
1407 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1408 
1409 	/* For skcipher we use only the first sg item */
1410 	sg_in  = &dmreq->sg_in[0];
1411 	sg_out = &dmreq->sg_out[0];
1412 
1413 	sg_init_table(sg_in, 1);
1414 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1415 
1416 	sg_init_table(sg_out, 1);
1417 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1418 
1419 	if (cc->iv_gen_ops) {
1420 		/* For READs use IV stored in integrity metadata */
1421 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1422 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1423 		} else {
1424 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1425 			if (r < 0)
1426 				return r;
1427 			/* Data can be already preprocessed in generator */
1428 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1429 				sg_in = sg_out;
1430 			/* Store generated IV in integrity metadata */
1431 			if (cc->integrity_iv_size)
1432 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1433 		}
1434 		/* Working copy of IV, to be modified in crypto API */
1435 		memcpy(iv, org_iv, cc->iv_size);
1436 	}
1437 
1438 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1439 
1440 	if (bio_data_dir(ctx->bio_in) == WRITE)
1441 		r = crypto_skcipher_encrypt(req);
1442 	else
1443 		r = crypto_skcipher_decrypt(req);
1444 
1445 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1446 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1447 
1448 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1449 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1450 
1451 	return r;
1452 }
1453 
1454 static void kcryptd_async_done(struct crypto_async_request *async_req,
1455 			       int error);
1456 
1457 static void crypt_alloc_req_skcipher(struct crypt_config *cc,
1458 				     struct convert_context *ctx)
1459 {
1460 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1461 
1462 	if (!ctx->r.req)
1463 		ctx->r.req = mempool_alloc(&cc->req_pool, GFP_NOIO);
1464 
1465 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1466 
1467 	/*
1468 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1469 	 * requests if driver request queue is full.
1470 	 */
1471 	skcipher_request_set_callback(ctx->r.req,
1472 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1473 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1474 }
1475 
1476 static void crypt_alloc_req_aead(struct crypt_config *cc,
1477 				 struct convert_context *ctx)
1478 {
1479 	if (!ctx->r.req_aead)
1480 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, GFP_NOIO);
1481 
1482 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1483 
1484 	/*
1485 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1486 	 * requests if driver request queue is full.
1487 	 */
1488 	aead_request_set_callback(ctx->r.req_aead,
1489 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1490 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1491 }
1492 
1493 static void crypt_alloc_req(struct crypt_config *cc,
1494 			    struct convert_context *ctx)
1495 {
1496 	if (crypt_integrity_aead(cc))
1497 		crypt_alloc_req_aead(cc, ctx);
1498 	else
1499 		crypt_alloc_req_skcipher(cc, ctx);
1500 }
1501 
1502 static void crypt_free_req_skcipher(struct crypt_config *cc,
1503 				    struct skcipher_request *req, struct bio *base_bio)
1504 {
1505 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1506 
1507 	if ((struct skcipher_request *)(io + 1) != req)
1508 		mempool_free(req, &cc->req_pool);
1509 }
1510 
1511 static void crypt_free_req_aead(struct crypt_config *cc,
1512 				struct aead_request *req, struct bio *base_bio)
1513 {
1514 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1515 
1516 	if ((struct aead_request *)(io + 1) != req)
1517 		mempool_free(req, &cc->req_pool);
1518 }
1519 
1520 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1521 {
1522 	if (crypt_integrity_aead(cc))
1523 		crypt_free_req_aead(cc, req, base_bio);
1524 	else
1525 		crypt_free_req_skcipher(cc, req, base_bio);
1526 }
1527 
1528 /*
1529  * Encrypt / decrypt data from one bio to another one (can be the same one)
1530  */
1531 static blk_status_t crypt_convert(struct crypt_config *cc,
1532 			 struct convert_context *ctx, bool atomic)
1533 {
1534 	unsigned int tag_offset = 0;
1535 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1536 	int r;
1537 
1538 	atomic_set(&ctx->cc_pending, 1);
1539 
1540 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1541 
1542 		crypt_alloc_req(cc, ctx);
1543 		atomic_inc(&ctx->cc_pending);
1544 
1545 		if (crypt_integrity_aead(cc))
1546 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1547 		else
1548 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1549 
1550 		switch (r) {
1551 		/*
1552 		 * The request was queued by a crypto driver
1553 		 * but the driver request queue is full, let's wait.
1554 		 */
1555 		case -EBUSY:
1556 			wait_for_completion(&ctx->restart);
1557 			reinit_completion(&ctx->restart);
1558 			fallthrough;
1559 		/*
1560 		 * The request is queued and processed asynchronously,
1561 		 * completion function kcryptd_async_done() will be called.
1562 		 */
1563 		case -EINPROGRESS:
1564 			ctx->r.req = NULL;
1565 			ctx->cc_sector += sector_step;
1566 			tag_offset++;
1567 			continue;
1568 		/*
1569 		 * The request was already processed (synchronously).
1570 		 */
1571 		case 0:
1572 			atomic_dec(&ctx->cc_pending);
1573 			ctx->cc_sector += sector_step;
1574 			tag_offset++;
1575 			if (!atomic)
1576 				cond_resched();
1577 			continue;
1578 		/*
1579 		 * There was a data integrity error.
1580 		 */
1581 		case -EBADMSG:
1582 			atomic_dec(&ctx->cc_pending);
1583 			return BLK_STS_PROTECTION;
1584 		/*
1585 		 * There was an error while processing the request.
1586 		 */
1587 		default:
1588 			atomic_dec(&ctx->cc_pending);
1589 			return BLK_STS_IOERR;
1590 		}
1591 	}
1592 
1593 	return 0;
1594 }
1595 
1596 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1597 
1598 /*
1599  * Generate a new unfragmented bio with the given size
1600  * This should never violate the device limitations (but only because
1601  * max_segment_size is being constrained to PAGE_SIZE).
1602  *
1603  * This function may be called concurrently. If we allocate from the mempool
1604  * concurrently, there is a possibility of deadlock. For example, if we have
1605  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1606  * the mempool concurrently, it may deadlock in a situation where both processes
1607  * have allocated 128 pages and the mempool is exhausted.
1608  *
1609  * In order to avoid this scenario we allocate the pages under a mutex.
1610  *
1611  * In order to not degrade performance with excessive locking, we try
1612  * non-blocking allocations without a mutex first but on failure we fallback
1613  * to blocking allocations with a mutex.
1614  */
1615 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1616 {
1617 	struct crypt_config *cc = io->cc;
1618 	struct bio *clone;
1619 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1620 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1621 	unsigned i, len, remaining_size;
1622 	struct page *page;
1623 
1624 retry:
1625 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1626 		mutex_lock(&cc->bio_alloc_lock);
1627 
1628 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1629 	if (!clone)
1630 		goto out;
1631 
1632 	clone_init(io, clone);
1633 
1634 	remaining_size = size;
1635 
1636 	for (i = 0; i < nr_iovecs; i++) {
1637 		page = mempool_alloc(&cc->page_pool, gfp_mask);
1638 		if (!page) {
1639 			crypt_free_buffer_pages(cc, clone);
1640 			bio_put(clone);
1641 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1642 			goto retry;
1643 		}
1644 
1645 		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1646 
1647 		bio_add_page(clone, page, len, 0);
1648 
1649 		remaining_size -= len;
1650 	}
1651 
1652 	/* Allocate space for integrity tags */
1653 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1654 		crypt_free_buffer_pages(cc, clone);
1655 		bio_put(clone);
1656 		clone = NULL;
1657 	}
1658 out:
1659 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1660 		mutex_unlock(&cc->bio_alloc_lock);
1661 
1662 	return clone;
1663 }
1664 
1665 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1666 {
1667 	struct bio_vec *bv;
1668 	struct bvec_iter_all iter_all;
1669 
1670 	bio_for_each_segment_all(bv, clone, iter_all) {
1671 		BUG_ON(!bv->bv_page);
1672 		mempool_free(bv->bv_page, &cc->page_pool);
1673 	}
1674 }
1675 
1676 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1677 			  struct bio *bio, sector_t sector)
1678 {
1679 	io->cc = cc;
1680 	io->base_bio = bio;
1681 	io->sector = sector;
1682 	io->error = 0;
1683 	io->ctx.r.req = NULL;
1684 	io->integrity_metadata = NULL;
1685 	io->integrity_metadata_from_pool = false;
1686 	atomic_set(&io->io_pending, 0);
1687 }
1688 
1689 static void crypt_inc_pending(struct dm_crypt_io *io)
1690 {
1691 	atomic_inc(&io->io_pending);
1692 }
1693 
1694 /*
1695  * One of the bios was finished. Check for completion of
1696  * the whole request and correctly clean up the buffer.
1697  */
1698 static void crypt_dec_pending(struct dm_crypt_io *io)
1699 {
1700 	struct crypt_config *cc = io->cc;
1701 	struct bio *base_bio = io->base_bio;
1702 	blk_status_t error = io->error;
1703 
1704 	if (!atomic_dec_and_test(&io->io_pending))
1705 		return;
1706 
1707 	if (io->ctx.r.req)
1708 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1709 
1710 	if (unlikely(io->integrity_metadata_from_pool))
1711 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1712 	else
1713 		kfree(io->integrity_metadata);
1714 
1715 	base_bio->bi_status = error;
1716 	bio_endio(base_bio);
1717 }
1718 
1719 /*
1720  * kcryptd/kcryptd_io:
1721  *
1722  * Needed because it would be very unwise to do decryption in an
1723  * interrupt context.
1724  *
1725  * kcryptd performs the actual encryption or decryption.
1726  *
1727  * kcryptd_io performs the IO submission.
1728  *
1729  * They must be separated as otherwise the final stages could be
1730  * starved by new requests which can block in the first stages due
1731  * to memory allocation.
1732  *
1733  * The work is done per CPU global for all dm-crypt instances.
1734  * They should not depend on each other and do not block.
1735  */
1736 static void crypt_endio(struct bio *clone)
1737 {
1738 	struct dm_crypt_io *io = clone->bi_private;
1739 	struct crypt_config *cc = io->cc;
1740 	unsigned rw = bio_data_dir(clone);
1741 	blk_status_t error;
1742 
1743 	/*
1744 	 * free the processed pages
1745 	 */
1746 	if (rw == WRITE)
1747 		crypt_free_buffer_pages(cc, clone);
1748 
1749 	error = clone->bi_status;
1750 	bio_put(clone);
1751 
1752 	if (rw == READ && !error) {
1753 		kcryptd_queue_crypt(io);
1754 		return;
1755 	}
1756 
1757 	if (unlikely(error))
1758 		io->error = error;
1759 
1760 	crypt_dec_pending(io);
1761 }
1762 
1763 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1764 {
1765 	struct crypt_config *cc = io->cc;
1766 
1767 	clone->bi_private = io;
1768 	clone->bi_end_io  = crypt_endio;
1769 	bio_set_dev(clone, cc->dev->bdev);
1770 	clone->bi_opf	  = io->base_bio->bi_opf;
1771 }
1772 
1773 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1774 {
1775 	struct crypt_config *cc = io->cc;
1776 	struct bio *clone;
1777 
1778 	/*
1779 	 * We need the original biovec array in order to decrypt
1780 	 * the whole bio data *afterwards* -- thanks to immutable
1781 	 * biovecs we don't need to worry about the block layer
1782 	 * modifying the biovec array; so leverage bio_clone_fast().
1783 	 */
1784 	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1785 	if (!clone)
1786 		return 1;
1787 
1788 	crypt_inc_pending(io);
1789 
1790 	clone_init(io, clone);
1791 	clone->bi_iter.bi_sector = cc->start + io->sector;
1792 
1793 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1794 		crypt_dec_pending(io);
1795 		bio_put(clone);
1796 		return 1;
1797 	}
1798 
1799 	submit_bio_noacct(clone);
1800 	return 0;
1801 }
1802 
1803 static void kcryptd_io_read_work(struct work_struct *work)
1804 {
1805 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1806 
1807 	crypt_inc_pending(io);
1808 	if (kcryptd_io_read(io, GFP_NOIO))
1809 		io->error = BLK_STS_RESOURCE;
1810 	crypt_dec_pending(io);
1811 }
1812 
1813 static void kcryptd_queue_read(struct dm_crypt_io *io)
1814 {
1815 	struct crypt_config *cc = io->cc;
1816 
1817 	INIT_WORK(&io->work, kcryptd_io_read_work);
1818 	queue_work(cc->io_queue, &io->work);
1819 }
1820 
1821 static void kcryptd_io_write(struct dm_crypt_io *io)
1822 {
1823 	struct bio *clone = io->ctx.bio_out;
1824 
1825 	submit_bio_noacct(clone);
1826 }
1827 
1828 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1829 
1830 static int dmcrypt_write(void *data)
1831 {
1832 	struct crypt_config *cc = data;
1833 	struct dm_crypt_io *io;
1834 
1835 	while (1) {
1836 		struct rb_root write_tree;
1837 		struct blk_plug plug;
1838 
1839 		spin_lock_irq(&cc->write_thread_lock);
1840 continue_locked:
1841 
1842 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1843 			goto pop_from_list;
1844 
1845 		set_current_state(TASK_INTERRUPTIBLE);
1846 
1847 		spin_unlock_irq(&cc->write_thread_lock);
1848 
1849 		if (unlikely(kthread_should_stop())) {
1850 			set_current_state(TASK_RUNNING);
1851 			break;
1852 		}
1853 
1854 		schedule();
1855 
1856 		set_current_state(TASK_RUNNING);
1857 		spin_lock_irq(&cc->write_thread_lock);
1858 		goto continue_locked;
1859 
1860 pop_from_list:
1861 		write_tree = cc->write_tree;
1862 		cc->write_tree = RB_ROOT;
1863 		spin_unlock_irq(&cc->write_thread_lock);
1864 
1865 		BUG_ON(rb_parent(write_tree.rb_node));
1866 
1867 		/*
1868 		 * Note: we cannot walk the tree here with rb_next because
1869 		 * the structures may be freed when kcryptd_io_write is called.
1870 		 */
1871 		blk_start_plug(&plug);
1872 		do {
1873 			io = crypt_io_from_node(rb_first(&write_tree));
1874 			rb_erase(&io->rb_node, &write_tree);
1875 			kcryptd_io_write(io);
1876 		} while (!RB_EMPTY_ROOT(&write_tree));
1877 		blk_finish_plug(&plug);
1878 	}
1879 	return 0;
1880 }
1881 
1882 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1883 {
1884 	struct bio *clone = io->ctx.bio_out;
1885 	struct crypt_config *cc = io->cc;
1886 	unsigned long flags;
1887 	sector_t sector;
1888 	struct rb_node **rbp, *parent;
1889 
1890 	if (unlikely(io->error)) {
1891 		crypt_free_buffer_pages(cc, clone);
1892 		bio_put(clone);
1893 		crypt_dec_pending(io);
1894 		return;
1895 	}
1896 
1897 	/* crypt_convert should have filled the clone bio */
1898 	BUG_ON(io->ctx.iter_out.bi_size);
1899 
1900 	clone->bi_iter.bi_sector = cc->start + io->sector;
1901 
1902 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1903 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1904 		submit_bio_noacct(clone);
1905 		return;
1906 	}
1907 
1908 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1909 	if (RB_EMPTY_ROOT(&cc->write_tree))
1910 		wake_up_process(cc->write_thread);
1911 	rbp = &cc->write_tree.rb_node;
1912 	parent = NULL;
1913 	sector = io->sector;
1914 	while (*rbp) {
1915 		parent = *rbp;
1916 		if (sector < crypt_io_from_node(parent)->sector)
1917 			rbp = &(*rbp)->rb_left;
1918 		else
1919 			rbp = &(*rbp)->rb_right;
1920 	}
1921 	rb_link_node(&io->rb_node, parent, rbp);
1922 	rb_insert_color(&io->rb_node, &cc->write_tree);
1923 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1924 }
1925 
1926 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1927 				       struct convert_context *ctx)
1928 
1929 {
1930 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1931 		return false;
1932 
1933 	/*
1934 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1935 	 * constraints so they do not need to be issued inline by
1936 	 * kcryptd_crypt_write_convert().
1937 	 */
1938 	switch (bio_op(ctx->bio_in)) {
1939 	case REQ_OP_WRITE:
1940 	case REQ_OP_WRITE_SAME:
1941 	case REQ_OP_WRITE_ZEROES:
1942 		return true;
1943 	default:
1944 		return false;
1945 	}
1946 }
1947 
1948 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1949 {
1950 	struct crypt_config *cc = io->cc;
1951 	struct convert_context *ctx = &io->ctx;
1952 	struct bio *clone;
1953 	int crypt_finished;
1954 	sector_t sector = io->sector;
1955 	blk_status_t r;
1956 
1957 	/*
1958 	 * Prevent io from disappearing until this function completes.
1959 	 */
1960 	crypt_inc_pending(io);
1961 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
1962 
1963 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1964 	if (unlikely(!clone)) {
1965 		io->error = BLK_STS_IOERR;
1966 		goto dec;
1967 	}
1968 
1969 	io->ctx.bio_out = clone;
1970 	io->ctx.iter_out = clone->bi_iter;
1971 
1972 	sector += bio_sectors(clone);
1973 
1974 	crypt_inc_pending(io);
1975 	r = crypt_convert(cc, ctx,
1976 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags));
1977 	if (r)
1978 		io->error = r;
1979 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
1980 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
1981 		/* Wait for completion signaled by kcryptd_async_done() */
1982 		wait_for_completion(&ctx->restart);
1983 		crypt_finished = 1;
1984 	}
1985 
1986 	/* Encryption was already finished, submit io now */
1987 	if (crypt_finished) {
1988 		kcryptd_crypt_write_io_submit(io, 0);
1989 		io->sector = sector;
1990 	}
1991 
1992 dec:
1993 	crypt_dec_pending(io);
1994 }
1995 
1996 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1997 {
1998 	crypt_dec_pending(io);
1999 }
2000 
2001 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2002 {
2003 	struct crypt_config *cc = io->cc;
2004 	blk_status_t r;
2005 
2006 	crypt_inc_pending(io);
2007 
2008 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2009 			   io->sector);
2010 
2011 	r = crypt_convert(cc, &io->ctx,
2012 			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags));
2013 	if (r)
2014 		io->error = r;
2015 
2016 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2017 		kcryptd_crypt_read_done(io);
2018 
2019 	crypt_dec_pending(io);
2020 }
2021 
2022 static void kcryptd_async_done(struct crypto_async_request *async_req,
2023 			       int error)
2024 {
2025 	struct dm_crypt_request *dmreq = async_req->data;
2026 	struct convert_context *ctx = dmreq->ctx;
2027 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2028 	struct crypt_config *cc = io->cc;
2029 
2030 	/*
2031 	 * A request from crypto driver backlog is going to be processed now,
2032 	 * finish the completion and continue in crypt_convert().
2033 	 * (Callback will be called for the second time for this request.)
2034 	 */
2035 	if (error == -EINPROGRESS) {
2036 		complete(&ctx->restart);
2037 		return;
2038 	}
2039 
2040 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2041 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2042 
2043 	if (error == -EBADMSG) {
2044 		char b[BDEVNAME_SIZE];
2045 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2046 			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
2047 		io->error = BLK_STS_PROTECTION;
2048 	} else if (error < 0)
2049 		io->error = BLK_STS_IOERR;
2050 
2051 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2052 
2053 	if (!atomic_dec_and_test(&ctx->cc_pending))
2054 		return;
2055 
2056 	/*
2057 	 * The request is fully completed: for inline writes, let
2058 	 * kcryptd_crypt_write_convert() do the IO submission.
2059 	 */
2060 	if (bio_data_dir(io->base_bio) == READ) {
2061 		kcryptd_crypt_read_done(io);
2062 		return;
2063 	}
2064 
2065 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2066 		complete(&ctx->restart);
2067 		return;
2068 	}
2069 
2070 	kcryptd_crypt_write_io_submit(io, 1);
2071 }
2072 
2073 static void kcryptd_crypt(struct work_struct *work)
2074 {
2075 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2076 
2077 	if (bio_data_dir(io->base_bio) == READ)
2078 		kcryptd_crypt_read_convert(io);
2079 	else
2080 		kcryptd_crypt_write_convert(io);
2081 }
2082 
2083 static void kcryptd_crypt_tasklet(unsigned long work)
2084 {
2085 	kcryptd_crypt((struct work_struct *)work);
2086 }
2087 
2088 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2089 {
2090 	struct crypt_config *cc = io->cc;
2091 
2092 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2093 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2094 		if (in_irq()) {
2095 			/* Crypto API's "skcipher_walk_first() refuses to work in hard IRQ context */
2096 			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2097 			tasklet_schedule(&io->tasklet);
2098 			return;
2099 		}
2100 
2101 		kcryptd_crypt(&io->work);
2102 		return;
2103 	}
2104 
2105 	INIT_WORK(&io->work, kcryptd_crypt);
2106 	queue_work(cc->crypt_queue, &io->work);
2107 }
2108 
2109 static void crypt_free_tfms_aead(struct crypt_config *cc)
2110 {
2111 	if (!cc->cipher_tfm.tfms_aead)
2112 		return;
2113 
2114 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2115 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2116 		cc->cipher_tfm.tfms_aead[0] = NULL;
2117 	}
2118 
2119 	kfree(cc->cipher_tfm.tfms_aead);
2120 	cc->cipher_tfm.tfms_aead = NULL;
2121 }
2122 
2123 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2124 {
2125 	unsigned i;
2126 
2127 	if (!cc->cipher_tfm.tfms)
2128 		return;
2129 
2130 	for (i = 0; i < cc->tfms_count; i++)
2131 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2132 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2133 			cc->cipher_tfm.tfms[i] = NULL;
2134 		}
2135 
2136 	kfree(cc->cipher_tfm.tfms);
2137 	cc->cipher_tfm.tfms = NULL;
2138 }
2139 
2140 static void crypt_free_tfms(struct crypt_config *cc)
2141 {
2142 	if (crypt_integrity_aead(cc))
2143 		crypt_free_tfms_aead(cc);
2144 	else
2145 		crypt_free_tfms_skcipher(cc);
2146 }
2147 
2148 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2149 {
2150 	unsigned i;
2151 	int err;
2152 
2153 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2154 				      sizeof(struct crypto_skcipher *),
2155 				      GFP_KERNEL);
2156 	if (!cc->cipher_tfm.tfms)
2157 		return -ENOMEM;
2158 
2159 	for (i = 0; i < cc->tfms_count; i++) {
2160 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2161 						CRYPTO_ALG_ALLOCATES_MEMORY);
2162 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2163 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2164 			crypt_free_tfms(cc);
2165 			return err;
2166 		}
2167 	}
2168 
2169 	/*
2170 	 * dm-crypt performance can vary greatly depending on which crypto
2171 	 * algorithm implementation is used.  Help people debug performance
2172 	 * problems by logging the ->cra_driver_name.
2173 	 */
2174 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2175 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2176 	return 0;
2177 }
2178 
2179 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2180 {
2181 	int err;
2182 
2183 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2184 	if (!cc->cipher_tfm.tfms)
2185 		return -ENOMEM;
2186 
2187 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2188 						CRYPTO_ALG_ALLOCATES_MEMORY);
2189 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2190 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2191 		crypt_free_tfms(cc);
2192 		return err;
2193 	}
2194 
2195 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2196 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2197 	return 0;
2198 }
2199 
2200 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2201 {
2202 	if (crypt_integrity_aead(cc))
2203 		return crypt_alloc_tfms_aead(cc, ciphermode);
2204 	else
2205 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2206 }
2207 
2208 static unsigned crypt_subkey_size(struct crypt_config *cc)
2209 {
2210 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2211 }
2212 
2213 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2214 {
2215 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2216 }
2217 
2218 /*
2219  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2220  * the key must be for some reason in special format.
2221  * This funcion converts cc->key to this special format.
2222  */
2223 static void crypt_copy_authenckey(char *p, const void *key,
2224 				  unsigned enckeylen, unsigned authkeylen)
2225 {
2226 	struct crypto_authenc_key_param *param;
2227 	struct rtattr *rta;
2228 
2229 	rta = (struct rtattr *)p;
2230 	param = RTA_DATA(rta);
2231 	param->enckeylen = cpu_to_be32(enckeylen);
2232 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2233 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2234 	p += RTA_SPACE(sizeof(*param));
2235 	memcpy(p, key + enckeylen, authkeylen);
2236 	p += authkeylen;
2237 	memcpy(p, key, enckeylen);
2238 }
2239 
2240 static int crypt_setkey(struct crypt_config *cc)
2241 {
2242 	unsigned subkey_size;
2243 	int err = 0, i, r;
2244 
2245 	/* Ignore extra keys (which are used for IV etc) */
2246 	subkey_size = crypt_subkey_size(cc);
2247 
2248 	if (crypt_integrity_hmac(cc)) {
2249 		if (subkey_size < cc->key_mac_size)
2250 			return -EINVAL;
2251 
2252 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2253 				      subkey_size - cc->key_mac_size,
2254 				      cc->key_mac_size);
2255 	}
2256 
2257 	for (i = 0; i < cc->tfms_count; i++) {
2258 		if (crypt_integrity_hmac(cc))
2259 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2260 				cc->authenc_key, crypt_authenckey_size(cc));
2261 		else if (crypt_integrity_aead(cc))
2262 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2263 					       cc->key + (i * subkey_size),
2264 					       subkey_size);
2265 		else
2266 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2267 						   cc->key + (i * subkey_size),
2268 						   subkey_size);
2269 		if (r)
2270 			err = r;
2271 	}
2272 
2273 	if (crypt_integrity_hmac(cc))
2274 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2275 
2276 	return err;
2277 }
2278 
2279 #ifdef CONFIG_KEYS
2280 
2281 static bool contains_whitespace(const char *str)
2282 {
2283 	while (*str)
2284 		if (isspace(*str++))
2285 			return true;
2286 	return false;
2287 }
2288 
2289 static int set_key_user(struct crypt_config *cc, struct key *key)
2290 {
2291 	const struct user_key_payload *ukp;
2292 
2293 	ukp = user_key_payload_locked(key);
2294 	if (!ukp)
2295 		return -EKEYREVOKED;
2296 
2297 	if (cc->key_size != ukp->datalen)
2298 		return -EINVAL;
2299 
2300 	memcpy(cc->key, ukp->data, cc->key_size);
2301 
2302 	return 0;
2303 }
2304 
2305 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
2306 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2307 {
2308 	const struct encrypted_key_payload *ekp;
2309 
2310 	ekp = key->payload.data[0];
2311 	if (!ekp)
2312 		return -EKEYREVOKED;
2313 
2314 	if (cc->key_size != ekp->decrypted_datalen)
2315 		return -EINVAL;
2316 
2317 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2318 
2319 	return 0;
2320 }
2321 #endif /* CONFIG_ENCRYPTED_KEYS */
2322 
2323 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2324 {
2325 	char *new_key_string, *key_desc;
2326 	int ret;
2327 	struct key_type *type;
2328 	struct key *key;
2329 	int (*set_key)(struct crypt_config *cc, struct key *key);
2330 
2331 	/*
2332 	 * Reject key_string with whitespace. dm core currently lacks code for
2333 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2334 	 */
2335 	if (contains_whitespace(key_string)) {
2336 		DMERR("whitespace chars not allowed in key string");
2337 		return -EINVAL;
2338 	}
2339 
2340 	/* look for next ':' separating key_type from key_description */
2341 	key_desc = strpbrk(key_string, ":");
2342 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2343 		return -EINVAL;
2344 
2345 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2346 		type = &key_type_logon;
2347 		set_key = set_key_user;
2348 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2349 		type = &key_type_user;
2350 		set_key = set_key_user;
2351 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
2352 	} else if (!strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2353 		type = &key_type_encrypted;
2354 		set_key = set_key_encrypted;
2355 #endif
2356 	} else {
2357 		return -EINVAL;
2358 	}
2359 
2360 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2361 	if (!new_key_string)
2362 		return -ENOMEM;
2363 
2364 	key = request_key(type, key_desc + 1, NULL);
2365 	if (IS_ERR(key)) {
2366 		kfree_sensitive(new_key_string);
2367 		return PTR_ERR(key);
2368 	}
2369 
2370 	down_read(&key->sem);
2371 
2372 	ret = set_key(cc, key);
2373 	if (ret < 0) {
2374 		up_read(&key->sem);
2375 		key_put(key);
2376 		kfree_sensitive(new_key_string);
2377 		return ret;
2378 	}
2379 
2380 	up_read(&key->sem);
2381 	key_put(key);
2382 
2383 	/* clear the flag since following operations may invalidate previously valid key */
2384 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2385 
2386 	ret = crypt_setkey(cc);
2387 
2388 	if (!ret) {
2389 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2390 		kfree_sensitive(cc->key_string);
2391 		cc->key_string = new_key_string;
2392 	} else
2393 		kfree_sensitive(new_key_string);
2394 
2395 	return ret;
2396 }
2397 
2398 static int get_key_size(char **key_string)
2399 {
2400 	char *colon, dummy;
2401 	int ret;
2402 
2403 	if (*key_string[0] != ':')
2404 		return strlen(*key_string) >> 1;
2405 
2406 	/* look for next ':' in key string */
2407 	colon = strpbrk(*key_string + 1, ":");
2408 	if (!colon)
2409 		return -EINVAL;
2410 
2411 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2412 		return -EINVAL;
2413 
2414 	*key_string = colon;
2415 
2416 	/* remaining key string should be :<logon|user>:<key_desc> */
2417 
2418 	return ret;
2419 }
2420 
2421 #else
2422 
2423 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2424 {
2425 	return -EINVAL;
2426 }
2427 
2428 static int get_key_size(char **key_string)
2429 {
2430 	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2431 }
2432 
2433 #endif /* CONFIG_KEYS */
2434 
2435 static int crypt_set_key(struct crypt_config *cc, char *key)
2436 {
2437 	int r = -EINVAL;
2438 	int key_string_len = strlen(key);
2439 
2440 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2441 	if (!cc->key_size && strcmp(key, "-"))
2442 		goto out;
2443 
2444 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2445 	if (key[0] == ':') {
2446 		r = crypt_set_keyring_key(cc, key + 1);
2447 		goto out;
2448 	}
2449 
2450 	/* clear the flag since following operations may invalidate previously valid key */
2451 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2452 
2453 	/* wipe references to any kernel keyring key */
2454 	kfree_sensitive(cc->key_string);
2455 	cc->key_string = NULL;
2456 
2457 	/* Decode key from its hex representation. */
2458 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2459 		goto out;
2460 
2461 	r = crypt_setkey(cc);
2462 	if (!r)
2463 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2464 
2465 out:
2466 	/* Hex key string not needed after here, so wipe it. */
2467 	memset(key, '0', key_string_len);
2468 
2469 	return r;
2470 }
2471 
2472 static int crypt_wipe_key(struct crypt_config *cc)
2473 {
2474 	int r;
2475 
2476 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2477 	get_random_bytes(&cc->key, cc->key_size);
2478 
2479 	/* Wipe IV private keys */
2480 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2481 		r = cc->iv_gen_ops->wipe(cc);
2482 		if (r)
2483 			return r;
2484 	}
2485 
2486 	kfree_sensitive(cc->key_string);
2487 	cc->key_string = NULL;
2488 	r = crypt_setkey(cc);
2489 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2490 
2491 	return r;
2492 }
2493 
2494 static void crypt_calculate_pages_per_client(void)
2495 {
2496 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2497 
2498 	if (!dm_crypt_clients_n)
2499 		return;
2500 
2501 	pages /= dm_crypt_clients_n;
2502 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2503 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2504 	dm_crypt_pages_per_client = pages;
2505 }
2506 
2507 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2508 {
2509 	struct crypt_config *cc = pool_data;
2510 	struct page *page;
2511 
2512 	if (unlikely(percpu_counter_compare(&cc->n_allocated_pages, dm_crypt_pages_per_client) >= 0) &&
2513 	    likely(gfp_mask & __GFP_NORETRY))
2514 		return NULL;
2515 
2516 	page = alloc_page(gfp_mask);
2517 	if (likely(page != NULL))
2518 		percpu_counter_add(&cc->n_allocated_pages, 1);
2519 
2520 	return page;
2521 }
2522 
2523 static void crypt_page_free(void *page, void *pool_data)
2524 {
2525 	struct crypt_config *cc = pool_data;
2526 
2527 	__free_page(page);
2528 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2529 }
2530 
2531 static void crypt_dtr(struct dm_target *ti)
2532 {
2533 	struct crypt_config *cc = ti->private;
2534 
2535 	ti->private = NULL;
2536 
2537 	if (!cc)
2538 		return;
2539 
2540 	if (cc->write_thread)
2541 		kthread_stop(cc->write_thread);
2542 
2543 	if (cc->io_queue)
2544 		destroy_workqueue(cc->io_queue);
2545 	if (cc->crypt_queue)
2546 		destroy_workqueue(cc->crypt_queue);
2547 
2548 	crypt_free_tfms(cc);
2549 
2550 	bioset_exit(&cc->bs);
2551 
2552 	mempool_exit(&cc->page_pool);
2553 	mempool_exit(&cc->req_pool);
2554 	mempool_exit(&cc->tag_pool);
2555 
2556 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2557 	percpu_counter_destroy(&cc->n_allocated_pages);
2558 
2559 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2560 		cc->iv_gen_ops->dtr(cc);
2561 
2562 	if (cc->dev)
2563 		dm_put_device(ti, cc->dev);
2564 
2565 	kfree_sensitive(cc->cipher_string);
2566 	kfree_sensitive(cc->key_string);
2567 	kfree_sensitive(cc->cipher_auth);
2568 	kfree_sensitive(cc->authenc_key);
2569 
2570 	mutex_destroy(&cc->bio_alloc_lock);
2571 
2572 	/* Must zero key material before freeing */
2573 	kfree_sensitive(cc);
2574 
2575 	spin_lock(&dm_crypt_clients_lock);
2576 	WARN_ON(!dm_crypt_clients_n);
2577 	dm_crypt_clients_n--;
2578 	crypt_calculate_pages_per_client();
2579 	spin_unlock(&dm_crypt_clients_lock);
2580 }
2581 
2582 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2583 {
2584 	struct crypt_config *cc = ti->private;
2585 
2586 	if (crypt_integrity_aead(cc))
2587 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2588 	else
2589 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2590 
2591 	if (cc->iv_size)
2592 		/* at least a 64 bit sector number should fit in our buffer */
2593 		cc->iv_size = max(cc->iv_size,
2594 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2595 	else if (ivmode) {
2596 		DMWARN("Selected cipher does not support IVs");
2597 		ivmode = NULL;
2598 	}
2599 
2600 	/* Choose ivmode, see comments at iv code. */
2601 	if (ivmode == NULL)
2602 		cc->iv_gen_ops = NULL;
2603 	else if (strcmp(ivmode, "plain") == 0)
2604 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2605 	else if (strcmp(ivmode, "plain64") == 0)
2606 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2607 	else if (strcmp(ivmode, "plain64be") == 0)
2608 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2609 	else if (strcmp(ivmode, "essiv") == 0)
2610 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2611 	else if (strcmp(ivmode, "benbi") == 0)
2612 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2613 	else if (strcmp(ivmode, "null") == 0)
2614 		cc->iv_gen_ops = &crypt_iv_null_ops;
2615 	else if (strcmp(ivmode, "eboiv") == 0)
2616 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2617 	else if (strcmp(ivmode, "elephant") == 0) {
2618 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2619 		cc->key_parts = 2;
2620 		cc->key_extra_size = cc->key_size / 2;
2621 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2622 			return -EINVAL;
2623 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2624 	} else if (strcmp(ivmode, "lmk") == 0) {
2625 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2626 		/*
2627 		 * Version 2 and 3 is recognised according
2628 		 * to length of provided multi-key string.
2629 		 * If present (version 3), last key is used as IV seed.
2630 		 * All keys (including IV seed) are always the same size.
2631 		 */
2632 		if (cc->key_size % cc->key_parts) {
2633 			cc->key_parts++;
2634 			cc->key_extra_size = cc->key_size / cc->key_parts;
2635 		}
2636 	} else if (strcmp(ivmode, "tcw") == 0) {
2637 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2638 		cc->key_parts += 2; /* IV + whitening */
2639 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2640 	} else if (strcmp(ivmode, "random") == 0) {
2641 		cc->iv_gen_ops = &crypt_iv_random_ops;
2642 		/* Need storage space in integrity fields. */
2643 		cc->integrity_iv_size = cc->iv_size;
2644 	} else {
2645 		ti->error = "Invalid IV mode";
2646 		return -EINVAL;
2647 	}
2648 
2649 	return 0;
2650 }
2651 
2652 /*
2653  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2654  * The HMAC is needed to calculate tag size (HMAC digest size).
2655  * This should be probably done by crypto-api calls (once available...)
2656  */
2657 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2658 {
2659 	char *start, *end, *mac_alg = NULL;
2660 	struct crypto_ahash *mac;
2661 
2662 	if (!strstarts(cipher_api, "authenc("))
2663 		return 0;
2664 
2665 	start = strchr(cipher_api, '(');
2666 	end = strchr(cipher_api, ',');
2667 	if (!start || !end || ++start > end)
2668 		return -EINVAL;
2669 
2670 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2671 	if (!mac_alg)
2672 		return -ENOMEM;
2673 	strncpy(mac_alg, start, end - start);
2674 
2675 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2676 	kfree(mac_alg);
2677 
2678 	if (IS_ERR(mac))
2679 		return PTR_ERR(mac);
2680 
2681 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2682 	crypto_free_ahash(mac);
2683 
2684 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2685 	if (!cc->authenc_key)
2686 		return -ENOMEM;
2687 
2688 	return 0;
2689 }
2690 
2691 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2692 				char **ivmode, char **ivopts)
2693 {
2694 	struct crypt_config *cc = ti->private;
2695 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2696 	int ret = -EINVAL;
2697 
2698 	cc->tfms_count = 1;
2699 
2700 	/*
2701 	 * New format (capi: prefix)
2702 	 * capi:cipher_api_spec-iv:ivopts
2703 	 */
2704 	tmp = &cipher_in[strlen("capi:")];
2705 
2706 	/* Separate IV options if present, it can contain another '-' in hash name */
2707 	*ivopts = strrchr(tmp, ':');
2708 	if (*ivopts) {
2709 		**ivopts = '\0';
2710 		(*ivopts)++;
2711 	}
2712 	/* Parse IV mode */
2713 	*ivmode = strrchr(tmp, '-');
2714 	if (*ivmode) {
2715 		**ivmode = '\0';
2716 		(*ivmode)++;
2717 	}
2718 	/* The rest is crypto API spec */
2719 	cipher_api = tmp;
2720 
2721 	/* Alloc AEAD, can be used only in new format. */
2722 	if (crypt_integrity_aead(cc)) {
2723 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2724 		if (ret < 0) {
2725 			ti->error = "Invalid AEAD cipher spec";
2726 			return -ENOMEM;
2727 		}
2728 	}
2729 
2730 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2731 		cc->tfms_count = 64;
2732 
2733 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2734 		if (!*ivopts) {
2735 			ti->error = "Digest algorithm missing for ESSIV mode";
2736 			return -EINVAL;
2737 		}
2738 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2739 			       cipher_api, *ivopts);
2740 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2741 			ti->error = "Cannot allocate cipher string";
2742 			return -ENOMEM;
2743 		}
2744 		cipher_api = buf;
2745 	}
2746 
2747 	cc->key_parts = cc->tfms_count;
2748 
2749 	/* Allocate cipher */
2750 	ret = crypt_alloc_tfms(cc, cipher_api);
2751 	if (ret < 0) {
2752 		ti->error = "Error allocating crypto tfm";
2753 		return ret;
2754 	}
2755 
2756 	if (crypt_integrity_aead(cc))
2757 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2758 	else
2759 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2760 
2761 	return 0;
2762 }
2763 
2764 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2765 				char **ivmode, char **ivopts)
2766 {
2767 	struct crypt_config *cc = ti->private;
2768 	char *tmp, *cipher, *chainmode, *keycount;
2769 	char *cipher_api = NULL;
2770 	int ret = -EINVAL;
2771 	char dummy;
2772 
2773 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2774 		ti->error = "Bad cipher specification";
2775 		return -EINVAL;
2776 	}
2777 
2778 	/*
2779 	 * Legacy dm-crypt cipher specification
2780 	 * cipher[:keycount]-mode-iv:ivopts
2781 	 */
2782 	tmp = cipher_in;
2783 	keycount = strsep(&tmp, "-");
2784 	cipher = strsep(&keycount, ":");
2785 
2786 	if (!keycount)
2787 		cc->tfms_count = 1;
2788 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2789 		 !is_power_of_2(cc->tfms_count)) {
2790 		ti->error = "Bad cipher key count specification";
2791 		return -EINVAL;
2792 	}
2793 	cc->key_parts = cc->tfms_count;
2794 
2795 	chainmode = strsep(&tmp, "-");
2796 	*ivmode = strsep(&tmp, ":");
2797 	*ivopts = tmp;
2798 
2799 	/*
2800 	 * For compatibility with the original dm-crypt mapping format, if
2801 	 * only the cipher name is supplied, use cbc-plain.
2802 	 */
2803 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2804 		chainmode = "cbc";
2805 		*ivmode = "plain";
2806 	}
2807 
2808 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2809 		ti->error = "IV mechanism required";
2810 		return -EINVAL;
2811 	}
2812 
2813 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2814 	if (!cipher_api)
2815 		goto bad_mem;
2816 
2817 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2818 		if (!*ivopts) {
2819 			ti->error = "Digest algorithm missing for ESSIV mode";
2820 			kfree(cipher_api);
2821 			return -EINVAL;
2822 		}
2823 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2824 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2825 	} else {
2826 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2827 			       "%s(%s)", chainmode, cipher);
2828 	}
2829 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2830 		kfree(cipher_api);
2831 		goto bad_mem;
2832 	}
2833 
2834 	/* Allocate cipher */
2835 	ret = crypt_alloc_tfms(cc, cipher_api);
2836 	if (ret < 0) {
2837 		ti->error = "Error allocating crypto tfm";
2838 		kfree(cipher_api);
2839 		return ret;
2840 	}
2841 	kfree(cipher_api);
2842 
2843 	return 0;
2844 bad_mem:
2845 	ti->error = "Cannot allocate cipher strings";
2846 	return -ENOMEM;
2847 }
2848 
2849 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
2850 {
2851 	struct crypt_config *cc = ti->private;
2852 	char *ivmode = NULL, *ivopts = NULL;
2853 	int ret;
2854 
2855 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
2856 	if (!cc->cipher_string) {
2857 		ti->error = "Cannot allocate cipher strings";
2858 		return -ENOMEM;
2859 	}
2860 
2861 	if (strstarts(cipher_in, "capi:"))
2862 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
2863 	else
2864 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
2865 	if (ret)
2866 		return ret;
2867 
2868 	/* Initialize IV */
2869 	ret = crypt_ctr_ivmode(ti, ivmode);
2870 	if (ret < 0)
2871 		return ret;
2872 
2873 	/* Initialize and set key */
2874 	ret = crypt_set_key(cc, key);
2875 	if (ret < 0) {
2876 		ti->error = "Error decoding and setting key";
2877 		return ret;
2878 	}
2879 
2880 	/* Allocate IV */
2881 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
2882 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
2883 		if (ret < 0) {
2884 			ti->error = "Error creating IV";
2885 			return ret;
2886 		}
2887 	}
2888 
2889 	/* Initialize IV (set keys for ESSIV etc) */
2890 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
2891 		ret = cc->iv_gen_ops->init(cc);
2892 		if (ret < 0) {
2893 			ti->error = "Error initialising IV";
2894 			return ret;
2895 		}
2896 	}
2897 
2898 	/* wipe the kernel key payload copy */
2899 	if (cc->key_string)
2900 		memset(cc->key, 0, cc->key_size * sizeof(u8));
2901 
2902 	return ret;
2903 }
2904 
2905 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
2906 {
2907 	struct crypt_config *cc = ti->private;
2908 	struct dm_arg_set as;
2909 	static const struct dm_arg _args[] = {
2910 		{0, 8, "Invalid number of feature args"},
2911 	};
2912 	unsigned int opt_params, val;
2913 	const char *opt_string, *sval;
2914 	char dummy;
2915 	int ret;
2916 
2917 	/* Optional parameters */
2918 	as.argc = argc;
2919 	as.argv = argv;
2920 
2921 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
2922 	if (ret)
2923 		return ret;
2924 
2925 	while (opt_params--) {
2926 		opt_string = dm_shift_arg(&as);
2927 		if (!opt_string) {
2928 			ti->error = "Not enough feature arguments";
2929 			return -EINVAL;
2930 		}
2931 
2932 		if (!strcasecmp(opt_string, "allow_discards"))
2933 			ti->num_discard_bios = 1;
2934 
2935 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
2936 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
2937 
2938 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
2939 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
2940 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
2941 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
2942 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
2943 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
2944 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
2945 			if (val == 0 || val > MAX_TAG_SIZE) {
2946 				ti->error = "Invalid integrity arguments";
2947 				return -EINVAL;
2948 			}
2949 			cc->on_disk_tag_size = val;
2950 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
2951 			if (!strcasecmp(sval, "aead")) {
2952 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
2953 			} else  if (strcasecmp(sval, "none")) {
2954 				ti->error = "Unknown integrity profile";
2955 				return -EINVAL;
2956 			}
2957 
2958 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
2959 			if (!cc->cipher_auth)
2960 				return -ENOMEM;
2961 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
2962 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
2963 			    cc->sector_size > 4096 ||
2964 			    (cc->sector_size & (cc->sector_size - 1))) {
2965 				ti->error = "Invalid feature value for sector_size";
2966 				return -EINVAL;
2967 			}
2968 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
2969 				ti->error = "Device size is not multiple of sector_size feature";
2970 				return -EINVAL;
2971 			}
2972 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
2973 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
2974 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
2975 		else {
2976 			ti->error = "Invalid feature arguments";
2977 			return -EINVAL;
2978 		}
2979 	}
2980 
2981 	return 0;
2982 }
2983 
2984 #ifdef CONFIG_BLK_DEV_ZONED
2985 
2986 static int crypt_report_zones(struct dm_target *ti,
2987 		struct dm_report_zones_args *args, unsigned int nr_zones)
2988 {
2989 	struct crypt_config *cc = ti->private;
2990 	sector_t sector = cc->start + dm_target_offset(ti, args->next_sector);
2991 
2992 	args->start = cc->start;
2993 	return blkdev_report_zones(cc->dev->bdev, sector, nr_zones,
2994 				   dm_report_zones_cb, args);
2995 }
2996 
2997 #endif
2998 
2999 /*
3000  * Construct an encryption mapping:
3001  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3002  */
3003 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3004 {
3005 	struct crypt_config *cc;
3006 	const char *devname = dm_table_device_name(ti->table);
3007 	int key_size;
3008 	unsigned int align_mask;
3009 	unsigned long long tmpll;
3010 	int ret;
3011 	size_t iv_size_padding, additional_req_size;
3012 	char dummy;
3013 
3014 	if (argc < 5) {
3015 		ti->error = "Not enough arguments";
3016 		return -EINVAL;
3017 	}
3018 
3019 	key_size = get_key_size(&argv[1]);
3020 	if (key_size < 0) {
3021 		ti->error = "Cannot parse key size";
3022 		return -EINVAL;
3023 	}
3024 
3025 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3026 	if (!cc) {
3027 		ti->error = "Cannot allocate encryption context";
3028 		return -ENOMEM;
3029 	}
3030 	cc->key_size = key_size;
3031 	cc->sector_size = (1 << SECTOR_SHIFT);
3032 	cc->sector_shift = 0;
3033 
3034 	ti->private = cc;
3035 
3036 	spin_lock(&dm_crypt_clients_lock);
3037 	dm_crypt_clients_n++;
3038 	crypt_calculate_pages_per_client();
3039 	spin_unlock(&dm_crypt_clients_lock);
3040 
3041 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3042 	if (ret < 0)
3043 		goto bad;
3044 
3045 	/* Optional parameters need to be read before cipher constructor */
3046 	if (argc > 5) {
3047 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3048 		if (ret)
3049 			goto bad;
3050 	}
3051 
3052 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3053 	if (ret < 0)
3054 		goto bad;
3055 
3056 	if (crypt_integrity_aead(cc)) {
3057 		cc->dmreq_start = sizeof(struct aead_request);
3058 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3059 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3060 	} else {
3061 		cc->dmreq_start = sizeof(struct skcipher_request);
3062 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3063 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3064 	}
3065 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3066 
3067 	if (align_mask < CRYPTO_MINALIGN) {
3068 		/* Allocate the padding exactly */
3069 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3070 				& align_mask;
3071 	} else {
3072 		/*
3073 		 * If the cipher requires greater alignment than kmalloc
3074 		 * alignment, we don't know the exact position of the
3075 		 * initialization vector. We must assume worst case.
3076 		 */
3077 		iv_size_padding = align_mask;
3078 	}
3079 
3080 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3081 	additional_req_size = sizeof(struct dm_crypt_request) +
3082 		iv_size_padding + cc->iv_size +
3083 		cc->iv_size +
3084 		sizeof(uint64_t) +
3085 		sizeof(unsigned int);
3086 
3087 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3088 	if (ret) {
3089 		ti->error = "Cannot allocate crypt request mempool";
3090 		goto bad;
3091 	}
3092 
3093 	cc->per_bio_data_size = ti->per_io_data_size =
3094 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3095 		      ARCH_KMALLOC_MINALIGN);
3096 
3097 	ret = mempool_init(&cc->page_pool, BIO_MAX_PAGES, crypt_page_alloc, crypt_page_free, cc);
3098 	if (ret) {
3099 		ti->error = "Cannot allocate page mempool";
3100 		goto bad;
3101 	}
3102 
3103 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3104 	if (ret) {
3105 		ti->error = "Cannot allocate crypt bioset";
3106 		goto bad;
3107 	}
3108 
3109 	mutex_init(&cc->bio_alloc_lock);
3110 
3111 	ret = -EINVAL;
3112 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3113 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3114 		ti->error = "Invalid iv_offset sector";
3115 		goto bad;
3116 	}
3117 	cc->iv_offset = tmpll;
3118 
3119 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3120 	if (ret) {
3121 		ti->error = "Device lookup failed";
3122 		goto bad;
3123 	}
3124 
3125 	ret = -EINVAL;
3126 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3127 		ti->error = "Invalid device sector";
3128 		goto bad;
3129 	}
3130 	cc->start = tmpll;
3131 
3132 	/*
3133 	 * For zoned block devices, we need to preserve the issuer write
3134 	 * ordering. To do so, disable write workqueues and force inline
3135 	 * encryption completion.
3136 	 */
3137 	if (bdev_is_zoned(cc->dev->bdev)) {
3138 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3139 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3140 	}
3141 
3142 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3143 		ret = crypt_integrity_ctr(cc, ti);
3144 		if (ret)
3145 			goto bad;
3146 
3147 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3148 		if (!cc->tag_pool_max_sectors)
3149 			cc->tag_pool_max_sectors = 1;
3150 
3151 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3152 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3153 		if (ret) {
3154 			ti->error = "Cannot allocate integrity tags mempool";
3155 			goto bad;
3156 		}
3157 
3158 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3159 	}
3160 
3161 	ret = -ENOMEM;
3162 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3163 	if (!cc->io_queue) {
3164 		ti->error = "Couldn't create kcryptd io queue";
3165 		goto bad;
3166 	}
3167 
3168 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3169 		cc->crypt_queue = alloc_workqueue("kcryptd-%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3170 						  1, devname);
3171 	else
3172 		cc->crypt_queue = alloc_workqueue("kcryptd-%s",
3173 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM |
3174 						  WQ_UNBOUND | WQ_SYSFS,
3175 						  num_online_cpus(), devname);
3176 	if (!cc->crypt_queue) {
3177 		ti->error = "Couldn't create kcryptd queue";
3178 		goto bad;
3179 	}
3180 
3181 	spin_lock_init(&cc->write_thread_lock);
3182 	cc->write_tree = RB_ROOT;
3183 
3184 	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3185 	if (IS_ERR(cc->write_thread)) {
3186 		ret = PTR_ERR(cc->write_thread);
3187 		cc->write_thread = NULL;
3188 		ti->error = "Couldn't spawn write thread";
3189 		goto bad;
3190 	}
3191 	wake_up_process(cc->write_thread);
3192 
3193 	ti->num_flush_bios = 1;
3194 
3195 	return 0;
3196 
3197 bad:
3198 	crypt_dtr(ti);
3199 	return ret;
3200 }
3201 
3202 static int crypt_map(struct dm_target *ti, struct bio *bio)
3203 {
3204 	struct dm_crypt_io *io;
3205 	struct crypt_config *cc = ti->private;
3206 
3207 	/*
3208 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3209 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3210 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3211 	 */
3212 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3213 	    bio_op(bio) == REQ_OP_DISCARD)) {
3214 		bio_set_dev(bio, cc->dev->bdev);
3215 		if (bio_sectors(bio))
3216 			bio->bi_iter.bi_sector = cc->start +
3217 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3218 		return DM_MAPIO_REMAPPED;
3219 	}
3220 
3221 	/*
3222 	 * Check if bio is too large, split as needed.
3223 	 */
3224 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_PAGES << PAGE_SHIFT)) &&
3225 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3226 		dm_accept_partial_bio(bio, ((BIO_MAX_PAGES << PAGE_SHIFT) >> SECTOR_SHIFT));
3227 
3228 	/*
3229 	 * Ensure that bio is a multiple of internal sector encryption size
3230 	 * and is aligned to this size as defined in IO hints.
3231 	 */
3232 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3233 		return DM_MAPIO_KILL;
3234 
3235 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3236 		return DM_MAPIO_KILL;
3237 
3238 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3239 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3240 
3241 	if (cc->on_disk_tag_size) {
3242 		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3243 
3244 		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3245 		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3246 				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3247 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3248 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3249 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3250 			io->integrity_metadata_from_pool = true;
3251 		}
3252 	}
3253 
3254 	if (crypt_integrity_aead(cc))
3255 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3256 	else
3257 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3258 
3259 	if (bio_data_dir(io->base_bio) == READ) {
3260 		if (kcryptd_io_read(io, GFP_NOWAIT))
3261 			kcryptd_queue_read(io);
3262 	} else
3263 		kcryptd_queue_crypt(io);
3264 
3265 	return DM_MAPIO_SUBMITTED;
3266 }
3267 
3268 static void crypt_status(struct dm_target *ti, status_type_t type,
3269 			 unsigned status_flags, char *result, unsigned maxlen)
3270 {
3271 	struct crypt_config *cc = ti->private;
3272 	unsigned i, sz = 0;
3273 	int num_feature_args = 0;
3274 
3275 	switch (type) {
3276 	case STATUSTYPE_INFO:
3277 		result[0] = '\0';
3278 		break;
3279 
3280 	case STATUSTYPE_TABLE:
3281 		DMEMIT("%s ", cc->cipher_string);
3282 
3283 		if (cc->key_size > 0) {
3284 			if (cc->key_string)
3285 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3286 			else
3287 				for (i = 0; i < cc->key_size; i++)
3288 					DMEMIT("%02x", cc->key[i]);
3289 		} else
3290 			DMEMIT("-");
3291 
3292 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3293 				cc->dev->name, (unsigned long long)cc->start);
3294 
3295 		num_feature_args += !!ti->num_discard_bios;
3296 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3297 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3298 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3299 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3300 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3301 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3302 		if (cc->on_disk_tag_size)
3303 			num_feature_args++;
3304 		if (num_feature_args) {
3305 			DMEMIT(" %d", num_feature_args);
3306 			if (ti->num_discard_bios)
3307 				DMEMIT(" allow_discards");
3308 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3309 				DMEMIT(" same_cpu_crypt");
3310 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3311 				DMEMIT(" submit_from_crypt_cpus");
3312 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3313 				DMEMIT(" no_read_workqueue");
3314 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3315 				DMEMIT(" no_write_workqueue");
3316 			if (cc->on_disk_tag_size)
3317 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3318 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3319 				DMEMIT(" sector_size:%d", cc->sector_size);
3320 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3321 				DMEMIT(" iv_large_sectors");
3322 		}
3323 
3324 		break;
3325 	}
3326 }
3327 
3328 static void crypt_postsuspend(struct dm_target *ti)
3329 {
3330 	struct crypt_config *cc = ti->private;
3331 
3332 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3333 }
3334 
3335 static int crypt_preresume(struct dm_target *ti)
3336 {
3337 	struct crypt_config *cc = ti->private;
3338 
3339 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3340 		DMERR("aborting resume - crypt key is not set.");
3341 		return -EAGAIN;
3342 	}
3343 
3344 	return 0;
3345 }
3346 
3347 static void crypt_resume(struct dm_target *ti)
3348 {
3349 	struct crypt_config *cc = ti->private;
3350 
3351 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3352 }
3353 
3354 /* Message interface
3355  *	key set <key>
3356  *	key wipe
3357  */
3358 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3359 			 char *result, unsigned maxlen)
3360 {
3361 	struct crypt_config *cc = ti->private;
3362 	int key_size, ret = -EINVAL;
3363 
3364 	if (argc < 2)
3365 		goto error;
3366 
3367 	if (!strcasecmp(argv[0], "key")) {
3368 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3369 			DMWARN("not suspended during key manipulation.");
3370 			return -EINVAL;
3371 		}
3372 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3373 			/* The key size may not be changed. */
3374 			key_size = get_key_size(&argv[2]);
3375 			if (key_size < 0 || cc->key_size != key_size) {
3376 				memset(argv[2], '0', strlen(argv[2]));
3377 				return -EINVAL;
3378 			}
3379 
3380 			ret = crypt_set_key(cc, argv[2]);
3381 			if (ret)
3382 				return ret;
3383 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3384 				ret = cc->iv_gen_ops->init(cc);
3385 			/* wipe the kernel key payload copy */
3386 			if (cc->key_string)
3387 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3388 			return ret;
3389 		}
3390 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3391 			return crypt_wipe_key(cc);
3392 	}
3393 
3394 error:
3395 	DMWARN("unrecognised message received.");
3396 	return -EINVAL;
3397 }
3398 
3399 static int crypt_iterate_devices(struct dm_target *ti,
3400 				 iterate_devices_callout_fn fn, void *data)
3401 {
3402 	struct crypt_config *cc = ti->private;
3403 
3404 	return fn(ti, cc->dev, cc->start, ti->len, data);
3405 }
3406 
3407 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3408 {
3409 	struct crypt_config *cc = ti->private;
3410 
3411 	/*
3412 	 * Unfortunate constraint that is required to avoid the potential
3413 	 * for exceeding underlying device's max_segments limits -- due to
3414 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3415 	 * bio that are not as physically contiguous as the original bio.
3416 	 */
3417 	limits->max_segment_size = PAGE_SIZE;
3418 
3419 	limits->logical_block_size =
3420 		max_t(unsigned, limits->logical_block_size, cc->sector_size);
3421 	limits->physical_block_size =
3422 		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3423 	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3424 }
3425 
3426 static struct target_type crypt_target = {
3427 	.name   = "crypt",
3428 	.version = {1, 22, 0},
3429 	.module = THIS_MODULE,
3430 	.ctr    = crypt_ctr,
3431 	.dtr    = crypt_dtr,
3432 #ifdef CONFIG_BLK_DEV_ZONED
3433 	.features = DM_TARGET_ZONED_HM,
3434 	.report_zones = crypt_report_zones,
3435 #endif
3436 	.map    = crypt_map,
3437 	.status = crypt_status,
3438 	.postsuspend = crypt_postsuspend,
3439 	.preresume = crypt_preresume,
3440 	.resume = crypt_resume,
3441 	.message = crypt_message,
3442 	.iterate_devices = crypt_iterate_devices,
3443 	.io_hints = crypt_io_hints,
3444 };
3445 
3446 static int __init dm_crypt_init(void)
3447 {
3448 	int r;
3449 
3450 	r = dm_register_target(&crypt_target);
3451 	if (r < 0)
3452 		DMERR("register failed %d", r);
3453 
3454 	return r;
3455 }
3456 
3457 static void __exit dm_crypt_exit(void)
3458 {
3459 	dm_unregister_target(&crypt_target);
3460 }
3461 
3462 module_init(dm_crypt_init);
3463 module_exit(dm_crypt_exit);
3464 
3465 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3466 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3467 MODULE_LICENSE("GPL");
3468