xref: /openbmc/linux/drivers/md/dm-crypt.c (revision e1181b5b)
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 #include <keys/trusted-type.h>
41 
42 #include <linux/device-mapper.h>
43 
44 #define DM_MSG_PREFIX "crypt"
45 
46 /*
47  * context holding the current state of a multi-part conversion
48  */
49 struct convert_context {
50 	struct completion restart;
51 	struct bio *bio_in;
52 	struct bio *bio_out;
53 	struct bvec_iter iter_in;
54 	struct bvec_iter iter_out;
55 	u64 cc_sector;
56 	atomic_t cc_pending;
57 	union {
58 		struct skcipher_request *req;
59 		struct aead_request *req_aead;
60 	} r;
61 
62 };
63 
64 /*
65  * per bio private data
66  */
67 struct dm_crypt_io {
68 	struct crypt_config *cc;
69 	struct bio *base_bio;
70 	u8 *integrity_metadata;
71 	bool integrity_metadata_from_pool;
72 	struct work_struct work;
73 	struct tasklet_struct tasklet;
74 
75 	struct convert_context ctx;
76 
77 	atomic_t io_pending;
78 	blk_status_t error;
79 	sector_t sector;
80 
81 	struct rb_node rb_node;
82 } CRYPTO_MINALIGN_ATTR;
83 
84 struct dm_crypt_request {
85 	struct convert_context *ctx;
86 	struct scatterlist sg_in[4];
87 	struct scatterlist sg_out[4];
88 	u64 iv_sector;
89 };
90 
91 struct crypt_config;
92 
93 struct crypt_iv_operations {
94 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
95 		   const char *opts);
96 	void (*dtr)(struct crypt_config *cc);
97 	int (*init)(struct crypt_config *cc);
98 	int (*wipe)(struct crypt_config *cc);
99 	int (*generator)(struct crypt_config *cc, u8 *iv,
100 			 struct dm_crypt_request *dmreq);
101 	int (*post)(struct crypt_config *cc, u8 *iv,
102 		    struct dm_crypt_request *dmreq);
103 };
104 
105 struct iv_benbi_private {
106 	int shift;
107 };
108 
109 #define LMK_SEED_SIZE 64 /* hash + 0 */
110 struct iv_lmk_private {
111 	struct crypto_shash *hash_tfm;
112 	u8 *seed;
113 };
114 
115 #define TCW_WHITENING_SIZE 16
116 struct iv_tcw_private {
117 	struct crypto_shash *crc32_tfm;
118 	u8 *iv_seed;
119 	u8 *whitening;
120 };
121 
122 #define ELEPHANT_MAX_KEY_SIZE 32
123 struct iv_elephant_private {
124 	struct crypto_skcipher *tfm;
125 };
126 
127 /*
128  * Crypt: maps a linear range of a block device
129  * and encrypts / decrypts at the same time.
130  */
131 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
132 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
133 	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
134 	     DM_CRYPT_WRITE_INLINE };
135 
136 enum cipher_flags {
137 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
138 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
139 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
140 };
141 
142 /*
143  * The fields in here must be read only after initialization.
144  */
145 struct crypt_config {
146 	struct dm_dev *dev;
147 	sector_t start;
148 
149 	struct percpu_counter n_allocated_pages;
150 
151 	struct workqueue_struct *io_queue;
152 	struct workqueue_struct *crypt_queue;
153 
154 	spinlock_t write_thread_lock;
155 	struct task_struct *write_thread;
156 	struct rb_root write_tree;
157 
158 	char *cipher_string;
159 	char *cipher_auth;
160 	char *key_string;
161 
162 	const struct crypt_iv_operations *iv_gen_ops;
163 	union {
164 		struct iv_benbi_private benbi;
165 		struct iv_lmk_private lmk;
166 		struct iv_tcw_private tcw;
167 		struct iv_elephant_private elephant;
168 	} iv_gen_private;
169 	u64 iv_offset;
170 	unsigned int iv_size;
171 	unsigned short int sector_size;
172 	unsigned char sector_shift;
173 
174 	union {
175 		struct crypto_skcipher **tfms;
176 		struct crypto_aead **tfms_aead;
177 	} cipher_tfm;
178 	unsigned tfms_count;
179 	unsigned long cipher_flags;
180 
181 	/*
182 	 * Layout of each crypto request:
183 	 *
184 	 *   struct skcipher_request
185 	 *      context
186 	 *      padding
187 	 *   struct dm_crypt_request
188 	 *      padding
189 	 *   IV
190 	 *
191 	 * The padding is added so that dm_crypt_request and the IV are
192 	 * correctly aligned.
193 	 */
194 	unsigned int dmreq_start;
195 
196 	unsigned int per_bio_data_size;
197 
198 	unsigned long flags;
199 	unsigned int key_size;
200 	unsigned int key_parts;      /* independent parts in key buffer */
201 	unsigned int key_extra_size; /* additional keys length */
202 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
203 
204 	unsigned int integrity_tag_size;
205 	unsigned int integrity_iv_size;
206 	unsigned int on_disk_tag_size;
207 
208 	/*
209 	 * pool for per bio private data, crypto requests,
210 	 * encryption requeusts/buffer pages and integrity tags
211 	 */
212 	unsigned tag_pool_max_sectors;
213 	mempool_t tag_pool;
214 	mempool_t req_pool;
215 	mempool_t page_pool;
216 
217 	struct bio_set bs;
218 	struct mutex bio_alloc_lock;
219 
220 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
221 	u8 key[];
222 };
223 
224 #define MIN_IOS		64
225 #define MAX_TAG_SIZE	480
226 #define POOL_ENTRY_SIZE	512
227 
228 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
229 static unsigned dm_crypt_clients_n = 0;
230 static volatile unsigned long dm_crypt_pages_per_client;
231 #define DM_CRYPT_MEMORY_PERCENT			2
232 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
233 
234 static void clone_init(struct dm_crypt_io *, struct bio *);
235 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
236 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
237 					     struct scatterlist *sg);
238 
239 static bool crypt_integrity_aead(struct crypt_config *cc);
240 
241 /*
242  * Use this to access cipher attributes that are independent of the key.
243  */
244 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
245 {
246 	return cc->cipher_tfm.tfms[0];
247 }
248 
249 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
250 {
251 	return cc->cipher_tfm.tfms_aead[0];
252 }
253 
254 /*
255  * Different IV generation algorithms:
256  *
257  * plain: the initial vector is the 32-bit little-endian version of the sector
258  *        number, padded with zeros if necessary.
259  *
260  * plain64: the initial vector is the 64-bit little-endian version of the sector
261  *        number, padded with zeros if necessary.
262  *
263  * plain64be: the initial vector is the 64-bit big-endian version of the sector
264  *        number, padded with zeros if necessary.
265  *
266  * essiv: "encrypted sector|salt initial vector", the sector number is
267  *        encrypted with the bulk cipher using a salt as key. The salt
268  *        should be derived from the bulk cipher's key via hashing.
269  *
270  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
271  *        (needed for LRW-32-AES and possible other narrow block modes)
272  *
273  * null: the initial vector is always zero.  Provides compatibility with
274  *       obsolete loop_fish2 devices.  Do not use for new devices.
275  *
276  * lmk:  Compatible implementation of the block chaining mode used
277  *       by the Loop-AES block device encryption system
278  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
279  *       It operates on full 512 byte sectors and uses CBC
280  *       with an IV derived from the sector number, the data and
281  *       optionally extra IV seed.
282  *       This means that after decryption the first block
283  *       of sector must be tweaked according to decrypted data.
284  *       Loop-AES can use three encryption schemes:
285  *         version 1: is plain aes-cbc mode
286  *         version 2: uses 64 multikey scheme with lmk IV generator
287  *         version 3: the same as version 2 with additional IV seed
288  *                   (it uses 65 keys, last key is used as IV seed)
289  *
290  * tcw:  Compatible implementation of the block chaining mode used
291  *       by the TrueCrypt device encryption system (prior to version 4.1).
292  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
293  *       It operates on full 512 byte sectors and uses CBC
294  *       with an IV derived from initial key and the sector number.
295  *       In addition, whitening value is applied on every sector, whitening
296  *       is calculated from initial key, sector number and mixed using CRC32.
297  *       Note that this encryption scheme is vulnerable to watermarking attacks
298  *       and should be used for old compatible containers access only.
299  *
300  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
301  *        The IV is encrypted little-endian byte-offset (with the same key
302  *        and cipher as the volume).
303  *
304  * elephant: The extended version of eboiv with additional Elephant diffuser
305  *           used with Bitlocker CBC mode.
306  *           This mode was used in older Windows systems
307  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
308  */
309 
310 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
311 			      struct dm_crypt_request *dmreq)
312 {
313 	memset(iv, 0, cc->iv_size);
314 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
315 
316 	return 0;
317 }
318 
319 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
320 				struct dm_crypt_request *dmreq)
321 {
322 	memset(iv, 0, cc->iv_size);
323 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
324 
325 	return 0;
326 }
327 
328 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
329 				  struct dm_crypt_request *dmreq)
330 {
331 	memset(iv, 0, cc->iv_size);
332 	/* iv_size is at least of size u64; usually it is 16 bytes */
333 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
334 
335 	return 0;
336 }
337 
338 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
339 			      struct dm_crypt_request *dmreq)
340 {
341 	/*
342 	 * ESSIV encryption of the IV is now handled by the crypto API,
343 	 * so just pass the plain sector number here.
344 	 */
345 	memset(iv, 0, cc->iv_size);
346 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
347 
348 	return 0;
349 }
350 
351 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
352 			      const char *opts)
353 {
354 	unsigned bs;
355 	int log;
356 
357 	if (crypt_integrity_aead(cc))
358 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
359 	else
360 		bs = crypto_skcipher_blocksize(any_tfm(cc));
361 	log = ilog2(bs);
362 
363 	/* we need to calculate how far we must shift the sector count
364 	 * to get the cipher block count, we use this shift in _gen */
365 
366 	if (1 << log != bs) {
367 		ti->error = "cypher blocksize is not a power of 2";
368 		return -EINVAL;
369 	}
370 
371 	if (log > 9) {
372 		ti->error = "cypher blocksize is > 512";
373 		return -EINVAL;
374 	}
375 
376 	cc->iv_gen_private.benbi.shift = 9 - log;
377 
378 	return 0;
379 }
380 
381 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
382 {
383 }
384 
385 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
386 			      struct dm_crypt_request *dmreq)
387 {
388 	__be64 val;
389 
390 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
391 
392 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
393 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
394 
395 	return 0;
396 }
397 
398 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
399 			     struct dm_crypt_request *dmreq)
400 {
401 	memset(iv, 0, cc->iv_size);
402 
403 	return 0;
404 }
405 
406 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
407 {
408 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
409 
410 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
411 		crypto_free_shash(lmk->hash_tfm);
412 	lmk->hash_tfm = NULL;
413 
414 	kfree_sensitive(lmk->seed);
415 	lmk->seed = NULL;
416 }
417 
418 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
419 			    const char *opts)
420 {
421 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
422 
423 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
424 		ti->error = "Unsupported sector size for LMK";
425 		return -EINVAL;
426 	}
427 
428 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
429 					   CRYPTO_ALG_ALLOCATES_MEMORY);
430 	if (IS_ERR(lmk->hash_tfm)) {
431 		ti->error = "Error initializing LMK hash";
432 		return PTR_ERR(lmk->hash_tfm);
433 	}
434 
435 	/* No seed in LMK version 2 */
436 	if (cc->key_parts == cc->tfms_count) {
437 		lmk->seed = NULL;
438 		return 0;
439 	}
440 
441 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
442 	if (!lmk->seed) {
443 		crypt_iv_lmk_dtr(cc);
444 		ti->error = "Error kmallocing seed storage in LMK";
445 		return -ENOMEM;
446 	}
447 
448 	return 0;
449 }
450 
451 static int crypt_iv_lmk_init(struct crypt_config *cc)
452 {
453 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
454 	int subkey_size = cc->key_size / cc->key_parts;
455 
456 	/* LMK seed is on the position of LMK_KEYS + 1 key */
457 	if (lmk->seed)
458 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
459 		       crypto_shash_digestsize(lmk->hash_tfm));
460 
461 	return 0;
462 }
463 
464 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
465 {
466 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
467 
468 	if (lmk->seed)
469 		memset(lmk->seed, 0, LMK_SEED_SIZE);
470 
471 	return 0;
472 }
473 
474 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
475 			    struct dm_crypt_request *dmreq,
476 			    u8 *data)
477 {
478 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
479 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
480 	struct md5_state md5state;
481 	__le32 buf[4];
482 	int i, r;
483 
484 	desc->tfm = lmk->hash_tfm;
485 
486 	r = crypto_shash_init(desc);
487 	if (r)
488 		return r;
489 
490 	if (lmk->seed) {
491 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
492 		if (r)
493 			return r;
494 	}
495 
496 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
497 	r = crypto_shash_update(desc, data + 16, 16 * 31);
498 	if (r)
499 		return r;
500 
501 	/* Sector is cropped to 56 bits here */
502 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
503 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
504 	buf[2] = cpu_to_le32(4024);
505 	buf[3] = 0;
506 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
507 	if (r)
508 		return r;
509 
510 	/* No MD5 padding here */
511 	r = crypto_shash_export(desc, &md5state);
512 	if (r)
513 		return r;
514 
515 	for (i = 0; i < MD5_HASH_WORDS; i++)
516 		__cpu_to_le32s(&md5state.hash[i]);
517 	memcpy(iv, &md5state.hash, cc->iv_size);
518 
519 	return 0;
520 }
521 
522 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
523 			    struct dm_crypt_request *dmreq)
524 {
525 	struct scatterlist *sg;
526 	u8 *src;
527 	int r = 0;
528 
529 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
530 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
531 		src = kmap_atomic(sg_page(sg));
532 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
533 		kunmap_atomic(src);
534 	} else
535 		memset(iv, 0, cc->iv_size);
536 
537 	return r;
538 }
539 
540 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
541 			     struct dm_crypt_request *dmreq)
542 {
543 	struct scatterlist *sg;
544 	u8 *dst;
545 	int r;
546 
547 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
548 		return 0;
549 
550 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
551 	dst = kmap_atomic(sg_page(sg));
552 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
553 
554 	/* Tweak the first block of plaintext sector */
555 	if (!r)
556 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
557 
558 	kunmap_atomic(dst);
559 	return r;
560 }
561 
562 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
563 {
564 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
565 
566 	kfree_sensitive(tcw->iv_seed);
567 	tcw->iv_seed = NULL;
568 	kfree_sensitive(tcw->whitening);
569 	tcw->whitening = NULL;
570 
571 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
572 		crypto_free_shash(tcw->crc32_tfm);
573 	tcw->crc32_tfm = NULL;
574 }
575 
576 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
577 			    const char *opts)
578 {
579 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
580 
581 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
582 		ti->error = "Unsupported sector size for TCW";
583 		return -EINVAL;
584 	}
585 
586 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
587 		ti->error = "Wrong key size for TCW";
588 		return -EINVAL;
589 	}
590 
591 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
592 					    CRYPTO_ALG_ALLOCATES_MEMORY);
593 	if (IS_ERR(tcw->crc32_tfm)) {
594 		ti->error = "Error initializing CRC32 in TCW";
595 		return PTR_ERR(tcw->crc32_tfm);
596 	}
597 
598 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
599 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
600 	if (!tcw->iv_seed || !tcw->whitening) {
601 		crypt_iv_tcw_dtr(cc);
602 		ti->error = "Error allocating seed storage in TCW";
603 		return -ENOMEM;
604 	}
605 
606 	return 0;
607 }
608 
609 static int crypt_iv_tcw_init(struct crypt_config *cc)
610 {
611 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
612 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
613 
614 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
615 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
616 	       TCW_WHITENING_SIZE);
617 
618 	return 0;
619 }
620 
621 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
622 {
623 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
624 
625 	memset(tcw->iv_seed, 0, cc->iv_size);
626 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
627 
628 	return 0;
629 }
630 
631 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
632 				  struct dm_crypt_request *dmreq,
633 				  u8 *data)
634 {
635 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
636 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
637 	u8 buf[TCW_WHITENING_SIZE];
638 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
639 	int i, r;
640 
641 	/* xor whitening with sector number */
642 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
643 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
644 
645 	/* calculate crc32 for every 32bit part and xor it */
646 	desc->tfm = tcw->crc32_tfm;
647 	for (i = 0; i < 4; i++) {
648 		r = crypto_shash_init(desc);
649 		if (r)
650 			goto out;
651 		r = crypto_shash_update(desc, &buf[i * 4], 4);
652 		if (r)
653 			goto out;
654 		r = crypto_shash_final(desc, &buf[i * 4]);
655 		if (r)
656 			goto out;
657 	}
658 	crypto_xor(&buf[0], &buf[12], 4);
659 	crypto_xor(&buf[4], &buf[8], 4);
660 
661 	/* apply whitening (8 bytes) to whole sector */
662 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
663 		crypto_xor(data + i * 8, buf, 8);
664 out:
665 	memzero_explicit(buf, sizeof(buf));
666 	return r;
667 }
668 
669 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
670 			    struct dm_crypt_request *dmreq)
671 {
672 	struct scatterlist *sg;
673 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
674 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
675 	u8 *src;
676 	int r = 0;
677 
678 	/* Remove whitening from ciphertext */
679 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
680 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
681 		src = kmap_atomic(sg_page(sg));
682 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
683 		kunmap_atomic(src);
684 	}
685 
686 	/* Calculate IV */
687 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
688 	if (cc->iv_size > 8)
689 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
690 			       cc->iv_size - 8);
691 
692 	return r;
693 }
694 
695 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
696 			     struct dm_crypt_request *dmreq)
697 {
698 	struct scatterlist *sg;
699 	u8 *dst;
700 	int r;
701 
702 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
703 		return 0;
704 
705 	/* Apply whitening on ciphertext */
706 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
707 	dst = kmap_atomic(sg_page(sg));
708 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
709 	kunmap_atomic(dst);
710 
711 	return r;
712 }
713 
714 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
715 				struct dm_crypt_request *dmreq)
716 {
717 	/* Used only for writes, there must be an additional space to store IV */
718 	get_random_bytes(iv, cc->iv_size);
719 	return 0;
720 }
721 
722 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
723 			    const char *opts)
724 {
725 	if (crypt_integrity_aead(cc)) {
726 		ti->error = "AEAD transforms not supported for EBOIV";
727 		return -EINVAL;
728 	}
729 
730 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
731 		ti->error = "Block size of EBOIV cipher does "
732 			    "not match IV size of block cipher";
733 		return -EINVAL;
734 	}
735 
736 	return 0;
737 }
738 
739 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
740 			    struct dm_crypt_request *dmreq)
741 {
742 	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
743 	struct skcipher_request *req;
744 	struct scatterlist src, dst;
745 	DECLARE_CRYPTO_WAIT(wait);
746 	int err;
747 
748 	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
749 	if (!req)
750 		return -ENOMEM;
751 
752 	memset(buf, 0, cc->iv_size);
753 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
754 
755 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
756 	sg_init_one(&dst, iv, cc->iv_size);
757 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
758 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
759 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
760 	skcipher_request_free(req);
761 
762 	return err;
763 }
764 
765 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
766 {
767 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
768 
769 	crypto_free_skcipher(elephant->tfm);
770 	elephant->tfm = NULL;
771 }
772 
773 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
774 			    const char *opts)
775 {
776 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
777 	int r;
778 
779 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
780 					      CRYPTO_ALG_ALLOCATES_MEMORY);
781 	if (IS_ERR(elephant->tfm)) {
782 		r = PTR_ERR(elephant->tfm);
783 		elephant->tfm = NULL;
784 		return r;
785 	}
786 
787 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
788 	if (r)
789 		crypt_iv_elephant_dtr(cc);
790 	return r;
791 }
792 
793 static void diffuser_disk_to_cpu(u32 *d, size_t n)
794 {
795 #ifndef __LITTLE_ENDIAN
796 	int i;
797 
798 	for (i = 0; i < n; i++)
799 		d[i] = le32_to_cpu((__le32)d[i]);
800 #endif
801 }
802 
803 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
804 {
805 #ifndef __LITTLE_ENDIAN
806 	int i;
807 
808 	for (i = 0; i < n; i++)
809 		d[i] = cpu_to_le32((u32)d[i]);
810 #endif
811 }
812 
813 static void diffuser_a_decrypt(u32 *d, size_t n)
814 {
815 	int i, i1, i2, i3;
816 
817 	for (i = 0; i < 5; i++) {
818 		i1 = 0;
819 		i2 = n - 2;
820 		i3 = n - 5;
821 
822 		while (i1 < (n - 1)) {
823 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
824 			i1++; i2++; i3++;
825 
826 			if (i3 >= n)
827 				i3 -= n;
828 
829 			d[i1] += d[i2] ^ d[i3];
830 			i1++; i2++; i3++;
831 
832 			if (i2 >= n)
833 				i2 -= n;
834 
835 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
836 			i1++; i2++; i3++;
837 
838 			d[i1] += d[i2] ^ d[i3];
839 			i1++; i2++; i3++;
840 		}
841 	}
842 }
843 
844 static void diffuser_a_encrypt(u32 *d, size_t n)
845 {
846 	int i, i1, i2, i3;
847 
848 	for (i = 0; i < 5; i++) {
849 		i1 = n - 1;
850 		i2 = n - 2 - 1;
851 		i3 = n - 5 - 1;
852 
853 		while (i1 > 0) {
854 			d[i1] -= d[i2] ^ d[i3];
855 			i1--; i2--; i3--;
856 
857 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
858 			i1--; i2--; i3--;
859 
860 			if (i2 < 0)
861 				i2 += n;
862 
863 			d[i1] -= d[i2] ^ d[i3];
864 			i1--; i2--; i3--;
865 
866 			if (i3 < 0)
867 				i3 += n;
868 
869 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
870 			i1--; i2--; i3--;
871 		}
872 	}
873 }
874 
875 static void diffuser_b_decrypt(u32 *d, size_t n)
876 {
877 	int i, i1, i2, i3;
878 
879 	for (i = 0; i < 3; i++) {
880 		i1 = 0;
881 		i2 = 2;
882 		i3 = 5;
883 
884 		while (i1 < (n - 1)) {
885 			d[i1] += d[i2] ^ d[i3];
886 			i1++; i2++; i3++;
887 
888 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
889 			i1++; i2++; i3++;
890 
891 			if (i2 >= n)
892 				i2 -= n;
893 
894 			d[i1] += d[i2] ^ d[i3];
895 			i1++; i2++; i3++;
896 
897 			if (i3 >= n)
898 				i3 -= n;
899 
900 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
901 			i1++; i2++; i3++;
902 		}
903 	}
904 }
905 
906 static void diffuser_b_encrypt(u32 *d, size_t n)
907 {
908 	int i, i1, i2, i3;
909 
910 	for (i = 0; i < 3; i++) {
911 		i1 = n - 1;
912 		i2 = 2 - 1;
913 		i3 = 5 - 1;
914 
915 		while (i1 > 0) {
916 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
917 			i1--; i2--; i3--;
918 
919 			if (i3 < 0)
920 				i3 += n;
921 
922 			d[i1] -= d[i2] ^ d[i3];
923 			i1--; i2--; i3--;
924 
925 			if (i2 < 0)
926 				i2 += n;
927 
928 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
929 			i1--; i2--; i3--;
930 
931 			d[i1] -= d[i2] ^ d[i3];
932 			i1--; i2--; i3--;
933 		}
934 	}
935 }
936 
937 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
938 {
939 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
940 	u8 *es, *ks, *data, *data2, *data_offset;
941 	struct skcipher_request *req;
942 	struct scatterlist *sg, *sg2, src, dst;
943 	DECLARE_CRYPTO_WAIT(wait);
944 	int i, r;
945 
946 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
947 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
948 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
949 
950 	if (!req || !es || !ks) {
951 		r = -ENOMEM;
952 		goto out;
953 	}
954 
955 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
956 
957 	/* E(Ks, e(s)) */
958 	sg_init_one(&src, es, 16);
959 	sg_init_one(&dst, ks, 16);
960 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
961 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
962 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
963 	if (r)
964 		goto out;
965 
966 	/* E(Ks, e'(s)) */
967 	es[15] = 0x80;
968 	sg_init_one(&dst, &ks[16], 16);
969 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
970 	if (r)
971 		goto out;
972 
973 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
974 	data = kmap_atomic(sg_page(sg));
975 	data_offset = data + sg->offset;
976 
977 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
978 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
979 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
980 		data2 = kmap_atomic(sg_page(sg2));
981 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
982 		kunmap_atomic(data2);
983 	}
984 
985 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
986 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
987 		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
988 		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
989 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
990 	}
991 
992 	for (i = 0; i < (cc->sector_size / 32); i++)
993 		crypto_xor(data_offset + i * 32, ks, 32);
994 
995 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
996 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
997 		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
998 		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
999 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1000 	}
1001 
1002 	kunmap_atomic(data);
1003 out:
1004 	kfree_sensitive(ks);
1005 	kfree_sensitive(es);
1006 	skcipher_request_free(req);
1007 	return r;
1008 }
1009 
1010 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1011 			    struct dm_crypt_request *dmreq)
1012 {
1013 	int r;
1014 
1015 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1016 		r = crypt_iv_elephant(cc, dmreq);
1017 		if (r)
1018 			return r;
1019 	}
1020 
1021 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1022 }
1023 
1024 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1025 				  struct dm_crypt_request *dmreq)
1026 {
1027 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1028 		return crypt_iv_elephant(cc, dmreq);
1029 
1030 	return 0;
1031 }
1032 
1033 static int crypt_iv_elephant_init(struct crypt_config *cc)
1034 {
1035 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1036 	int key_offset = cc->key_size - cc->key_extra_size;
1037 
1038 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1039 }
1040 
1041 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1042 {
1043 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1044 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1045 
1046 	memset(key, 0, cc->key_extra_size);
1047 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1048 }
1049 
1050 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1051 	.generator = crypt_iv_plain_gen
1052 };
1053 
1054 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1055 	.generator = crypt_iv_plain64_gen
1056 };
1057 
1058 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1059 	.generator = crypt_iv_plain64be_gen
1060 };
1061 
1062 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1063 	.generator = crypt_iv_essiv_gen
1064 };
1065 
1066 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1067 	.ctr	   = crypt_iv_benbi_ctr,
1068 	.dtr	   = crypt_iv_benbi_dtr,
1069 	.generator = crypt_iv_benbi_gen
1070 };
1071 
1072 static const struct crypt_iv_operations crypt_iv_null_ops = {
1073 	.generator = crypt_iv_null_gen
1074 };
1075 
1076 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1077 	.ctr	   = crypt_iv_lmk_ctr,
1078 	.dtr	   = crypt_iv_lmk_dtr,
1079 	.init	   = crypt_iv_lmk_init,
1080 	.wipe	   = crypt_iv_lmk_wipe,
1081 	.generator = crypt_iv_lmk_gen,
1082 	.post	   = crypt_iv_lmk_post
1083 };
1084 
1085 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1086 	.ctr	   = crypt_iv_tcw_ctr,
1087 	.dtr	   = crypt_iv_tcw_dtr,
1088 	.init	   = crypt_iv_tcw_init,
1089 	.wipe	   = crypt_iv_tcw_wipe,
1090 	.generator = crypt_iv_tcw_gen,
1091 	.post	   = crypt_iv_tcw_post
1092 };
1093 
1094 static const struct crypt_iv_operations crypt_iv_random_ops = {
1095 	.generator = crypt_iv_random_gen
1096 };
1097 
1098 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1099 	.ctr	   = crypt_iv_eboiv_ctr,
1100 	.generator = crypt_iv_eboiv_gen
1101 };
1102 
1103 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1104 	.ctr	   = crypt_iv_elephant_ctr,
1105 	.dtr	   = crypt_iv_elephant_dtr,
1106 	.init	   = crypt_iv_elephant_init,
1107 	.wipe	   = crypt_iv_elephant_wipe,
1108 	.generator = crypt_iv_elephant_gen,
1109 	.post	   = crypt_iv_elephant_post
1110 };
1111 
1112 /*
1113  * Integrity extensions
1114  */
1115 static bool crypt_integrity_aead(struct crypt_config *cc)
1116 {
1117 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1118 }
1119 
1120 static bool crypt_integrity_hmac(struct crypt_config *cc)
1121 {
1122 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1123 }
1124 
1125 /* Get sg containing data */
1126 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1127 					     struct scatterlist *sg)
1128 {
1129 	if (unlikely(crypt_integrity_aead(cc)))
1130 		return &sg[2];
1131 
1132 	return sg;
1133 }
1134 
1135 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1136 {
1137 	struct bio_integrity_payload *bip;
1138 	unsigned int tag_len;
1139 	int ret;
1140 
1141 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1142 		return 0;
1143 
1144 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1145 	if (IS_ERR(bip))
1146 		return PTR_ERR(bip);
1147 
1148 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1149 
1150 	bip->bip_iter.bi_size = tag_len;
1151 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1152 
1153 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1154 				     tag_len, offset_in_page(io->integrity_metadata));
1155 	if (unlikely(ret != tag_len))
1156 		return -ENOMEM;
1157 
1158 	return 0;
1159 }
1160 
1161 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1162 {
1163 #ifdef CONFIG_BLK_DEV_INTEGRITY
1164 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1165 	struct mapped_device *md = dm_table_get_md(ti->table);
1166 
1167 	/* From now we require underlying device with our integrity profile */
1168 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1169 		ti->error = "Integrity profile not supported.";
1170 		return -EINVAL;
1171 	}
1172 
1173 	if (bi->tag_size != cc->on_disk_tag_size ||
1174 	    bi->tuple_size != cc->on_disk_tag_size) {
1175 		ti->error = "Integrity profile tag size mismatch.";
1176 		return -EINVAL;
1177 	}
1178 	if (1 << bi->interval_exp != cc->sector_size) {
1179 		ti->error = "Integrity profile sector size mismatch.";
1180 		return -EINVAL;
1181 	}
1182 
1183 	if (crypt_integrity_aead(cc)) {
1184 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1185 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1186 		       cc->integrity_tag_size, cc->integrity_iv_size);
1187 
1188 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1189 			ti->error = "Integrity AEAD auth tag size is not supported.";
1190 			return -EINVAL;
1191 		}
1192 	} else if (cc->integrity_iv_size)
1193 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1194 		       cc->integrity_iv_size);
1195 
1196 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1197 		ti->error = "Not enough space for integrity tag in the profile.";
1198 		return -EINVAL;
1199 	}
1200 
1201 	return 0;
1202 #else
1203 	ti->error = "Integrity profile not supported.";
1204 	return -EINVAL;
1205 #endif
1206 }
1207 
1208 static void crypt_convert_init(struct crypt_config *cc,
1209 			       struct convert_context *ctx,
1210 			       struct bio *bio_out, struct bio *bio_in,
1211 			       sector_t sector)
1212 {
1213 	ctx->bio_in = bio_in;
1214 	ctx->bio_out = bio_out;
1215 	if (bio_in)
1216 		ctx->iter_in = bio_in->bi_iter;
1217 	if (bio_out)
1218 		ctx->iter_out = bio_out->bi_iter;
1219 	ctx->cc_sector = sector + cc->iv_offset;
1220 	init_completion(&ctx->restart);
1221 }
1222 
1223 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1224 					     void *req)
1225 {
1226 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1227 }
1228 
1229 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1230 {
1231 	return (void *)((char *)dmreq - cc->dmreq_start);
1232 }
1233 
1234 static u8 *iv_of_dmreq(struct crypt_config *cc,
1235 		       struct dm_crypt_request *dmreq)
1236 {
1237 	if (crypt_integrity_aead(cc))
1238 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1239 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1240 	else
1241 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1242 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1243 }
1244 
1245 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1246 		       struct dm_crypt_request *dmreq)
1247 {
1248 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1249 }
1250 
1251 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1252 		       struct dm_crypt_request *dmreq)
1253 {
1254 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1255 	return (__le64 *) ptr;
1256 }
1257 
1258 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1259 		       struct dm_crypt_request *dmreq)
1260 {
1261 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1262 		  cc->iv_size + sizeof(uint64_t);
1263 	return (unsigned int*)ptr;
1264 }
1265 
1266 static void *tag_from_dmreq(struct crypt_config *cc,
1267 				struct dm_crypt_request *dmreq)
1268 {
1269 	struct convert_context *ctx = dmreq->ctx;
1270 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1271 
1272 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1273 		cc->on_disk_tag_size];
1274 }
1275 
1276 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1277 			       struct dm_crypt_request *dmreq)
1278 {
1279 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1280 }
1281 
1282 static int crypt_convert_block_aead(struct crypt_config *cc,
1283 				     struct convert_context *ctx,
1284 				     struct aead_request *req,
1285 				     unsigned int tag_offset)
1286 {
1287 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1288 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1289 	struct dm_crypt_request *dmreq;
1290 	u8 *iv, *org_iv, *tag_iv, *tag;
1291 	__le64 *sector;
1292 	int r = 0;
1293 
1294 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1295 
1296 	/* Reject unexpected unaligned bio. */
1297 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1298 		return -EIO;
1299 
1300 	dmreq = dmreq_of_req(cc, req);
1301 	dmreq->iv_sector = ctx->cc_sector;
1302 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1303 		dmreq->iv_sector >>= cc->sector_shift;
1304 	dmreq->ctx = ctx;
1305 
1306 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1307 
1308 	sector = org_sector_of_dmreq(cc, dmreq);
1309 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1310 
1311 	iv = iv_of_dmreq(cc, dmreq);
1312 	org_iv = org_iv_of_dmreq(cc, dmreq);
1313 	tag = tag_from_dmreq(cc, dmreq);
1314 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1315 
1316 	/* AEAD request:
1317 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1318 	 *  | (authenticated) | (auth+encryption) |              |
1319 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1320 	 */
1321 	sg_init_table(dmreq->sg_in, 4);
1322 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1323 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1324 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1325 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1326 
1327 	sg_init_table(dmreq->sg_out, 4);
1328 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1329 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1330 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1331 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1332 
1333 	if (cc->iv_gen_ops) {
1334 		/* For READs use IV stored in integrity metadata */
1335 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1336 			memcpy(org_iv, tag_iv, cc->iv_size);
1337 		} else {
1338 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1339 			if (r < 0)
1340 				return r;
1341 			/* Store generated IV in integrity metadata */
1342 			if (cc->integrity_iv_size)
1343 				memcpy(tag_iv, org_iv, cc->iv_size);
1344 		}
1345 		/* Working copy of IV, to be modified in crypto API */
1346 		memcpy(iv, org_iv, cc->iv_size);
1347 	}
1348 
1349 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1350 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1351 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1352 				       cc->sector_size, iv);
1353 		r = crypto_aead_encrypt(req);
1354 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1355 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1356 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1357 	} else {
1358 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1359 				       cc->sector_size + cc->integrity_tag_size, iv);
1360 		r = crypto_aead_decrypt(req);
1361 	}
1362 
1363 	if (r == -EBADMSG) {
1364 		char b[BDEVNAME_SIZE];
1365 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1366 			    (unsigned long long)le64_to_cpu(*sector));
1367 	}
1368 
1369 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1370 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1371 
1372 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1373 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1374 
1375 	return r;
1376 }
1377 
1378 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1379 					struct convert_context *ctx,
1380 					struct skcipher_request *req,
1381 					unsigned int tag_offset)
1382 {
1383 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1384 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1385 	struct scatterlist *sg_in, *sg_out;
1386 	struct dm_crypt_request *dmreq;
1387 	u8 *iv, *org_iv, *tag_iv;
1388 	__le64 *sector;
1389 	int r = 0;
1390 
1391 	/* Reject unexpected unaligned bio. */
1392 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1393 		return -EIO;
1394 
1395 	dmreq = dmreq_of_req(cc, req);
1396 	dmreq->iv_sector = ctx->cc_sector;
1397 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1398 		dmreq->iv_sector >>= cc->sector_shift;
1399 	dmreq->ctx = ctx;
1400 
1401 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1402 
1403 	iv = iv_of_dmreq(cc, dmreq);
1404 	org_iv = org_iv_of_dmreq(cc, dmreq);
1405 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1406 
1407 	sector = org_sector_of_dmreq(cc, dmreq);
1408 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1409 
1410 	/* For skcipher we use only the first sg item */
1411 	sg_in  = &dmreq->sg_in[0];
1412 	sg_out = &dmreq->sg_out[0];
1413 
1414 	sg_init_table(sg_in, 1);
1415 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1416 
1417 	sg_init_table(sg_out, 1);
1418 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1419 
1420 	if (cc->iv_gen_ops) {
1421 		/* For READs use IV stored in integrity metadata */
1422 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1423 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1424 		} else {
1425 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1426 			if (r < 0)
1427 				return r;
1428 			/* Data can be already preprocessed in generator */
1429 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1430 				sg_in = sg_out;
1431 			/* Store generated IV in integrity metadata */
1432 			if (cc->integrity_iv_size)
1433 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1434 		}
1435 		/* Working copy of IV, to be modified in crypto API */
1436 		memcpy(iv, org_iv, cc->iv_size);
1437 	}
1438 
1439 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1440 
1441 	if (bio_data_dir(ctx->bio_in) == WRITE)
1442 		r = crypto_skcipher_encrypt(req);
1443 	else
1444 		r = crypto_skcipher_decrypt(req);
1445 
1446 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1447 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1448 
1449 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1450 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1451 
1452 	return r;
1453 }
1454 
1455 static void kcryptd_async_done(struct crypto_async_request *async_req,
1456 			       int error);
1457 
1458 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1459 				     struct convert_context *ctx)
1460 {
1461 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1462 
1463 	if (!ctx->r.req) {
1464 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1465 		if (!ctx->r.req)
1466 			return -ENOMEM;
1467 	}
1468 
1469 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1470 
1471 	/*
1472 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1473 	 * requests if driver request queue is full.
1474 	 */
1475 	skcipher_request_set_callback(ctx->r.req,
1476 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1477 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1478 
1479 	return 0;
1480 }
1481 
1482 static int crypt_alloc_req_aead(struct crypt_config *cc,
1483 				 struct convert_context *ctx)
1484 {
1485 	if (!ctx->r.req_aead) {
1486 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1487 		if (!ctx->r.req_aead)
1488 			return -ENOMEM;
1489 	}
1490 
1491 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1492 
1493 	/*
1494 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1495 	 * requests if driver request queue is full.
1496 	 */
1497 	aead_request_set_callback(ctx->r.req_aead,
1498 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1499 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1500 
1501 	return 0;
1502 }
1503 
1504 static int crypt_alloc_req(struct crypt_config *cc,
1505 			    struct convert_context *ctx)
1506 {
1507 	if (crypt_integrity_aead(cc))
1508 		return crypt_alloc_req_aead(cc, ctx);
1509 	else
1510 		return crypt_alloc_req_skcipher(cc, ctx);
1511 }
1512 
1513 static void crypt_free_req_skcipher(struct crypt_config *cc,
1514 				    struct skcipher_request *req, struct bio *base_bio)
1515 {
1516 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1517 
1518 	if ((struct skcipher_request *)(io + 1) != req)
1519 		mempool_free(req, &cc->req_pool);
1520 }
1521 
1522 static void crypt_free_req_aead(struct crypt_config *cc,
1523 				struct aead_request *req, struct bio *base_bio)
1524 {
1525 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1526 
1527 	if ((struct aead_request *)(io + 1) != req)
1528 		mempool_free(req, &cc->req_pool);
1529 }
1530 
1531 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1532 {
1533 	if (crypt_integrity_aead(cc))
1534 		crypt_free_req_aead(cc, req, base_bio);
1535 	else
1536 		crypt_free_req_skcipher(cc, req, base_bio);
1537 }
1538 
1539 /*
1540  * Encrypt / decrypt data from one bio to another one (can be the same one)
1541  */
1542 static blk_status_t crypt_convert(struct crypt_config *cc,
1543 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1544 {
1545 	unsigned int tag_offset = 0;
1546 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1547 	int r;
1548 
1549 	/*
1550 	 * if reset_pending is set we are dealing with the bio for the first time,
1551 	 * else we're continuing to work on the previous bio, so don't mess with
1552 	 * the cc_pending counter
1553 	 */
1554 	if (reset_pending)
1555 		atomic_set(&ctx->cc_pending, 1);
1556 
1557 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1558 
1559 		r = crypt_alloc_req(cc, ctx);
1560 		if (r) {
1561 			complete(&ctx->restart);
1562 			return BLK_STS_DEV_RESOURCE;
1563 		}
1564 
1565 		atomic_inc(&ctx->cc_pending);
1566 
1567 		if (crypt_integrity_aead(cc))
1568 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1569 		else
1570 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1571 
1572 		switch (r) {
1573 		/*
1574 		 * The request was queued by a crypto driver
1575 		 * but the driver request queue is full, let's wait.
1576 		 */
1577 		case -EBUSY:
1578 			if (in_interrupt()) {
1579 				if (try_wait_for_completion(&ctx->restart)) {
1580 					/*
1581 					 * we don't have to block to wait for completion,
1582 					 * so proceed
1583 					 */
1584 				} else {
1585 					/*
1586 					 * we can't wait for completion without blocking
1587 					 * exit and continue processing in a workqueue
1588 					 */
1589 					ctx->r.req = NULL;
1590 					ctx->cc_sector += sector_step;
1591 					tag_offset++;
1592 					return BLK_STS_DEV_RESOURCE;
1593 				}
1594 			} else {
1595 				wait_for_completion(&ctx->restart);
1596 			}
1597 			reinit_completion(&ctx->restart);
1598 			fallthrough;
1599 		/*
1600 		 * The request is queued and processed asynchronously,
1601 		 * completion function kcryptd_async_done() will be called.
1602 		 */
1603 		case -EINPROGRESS:
1604 			ctx->r.req = NULL;
1605 			ctx->cc_sector += sector_step;
1606 			tag_offset++;
1607 			continue;
1608 		/*
1609 		 * The request was already processed (synchronously).
1610 		 */
1611 		case 0:
1612 			atomic_dec(&ctx->cc_pending);
1613 			ctx->cc_sector += sector_step;
1614 			tag_offset++;
1615 			if (!atomic)
1616 				cond_resched();
1617 			continue;
1618 		/*
1619 		 * There was a data integrity error.
1620 		 */
1621 		case -EBADMSG:
1622 			atomic_dec(&ctx->cc_pending);
1623 			return BLK_STS_PROTECTION;
1624 		/*
1625 		 * There was an error while processing the request.
1626 		 */
1627 		default:
1628 			atomic_dec(&ctx->cc_pending);
1629 			return BLK_STS_IOERR;
1630 		}
1631 	}
1632 
1633 	return 0;
1634 }
1635 
1636 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1637 
1638 /*
1639  * Generate a new unfragmented bio with the given size
1640  * This should never violate the device limitations (but only because
1641  * max_segment_size is being constrained to PAGE_SIZE).
1642  *
1643  * This function may be called concurrently. If we allocate from the mempool
1644  * concurrently, there is a possibility of deadlock. For example, if we have
1645  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1646  * the mempool concurrently, it may deadlock in a situation where both processes
1647  * have allocated 128 pages and the mempool is exhausted.
1648  *
1649  * In order to avoid this scenario we allocate the pages under a mutex.
1650  *
1651  * In order to not degrade performance with excessive locking, we try
1652  * non-blocking allocations without a mutex first but on failure we fallback
1653  * to blocking allocations with a mutex.
1654  */
1655 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1656 {
1657 	struct crypt_config *cc = io->cc;
1658 	struct bio *clone;
1659 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1660 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1661 	unsigned i, len, remaining_size;
1662 	struct page *page;
1663 
1664 retry:
1665 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1666 		mutex_lock(&cc->bio_alloc_lock);
1667 
1668 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1669 	if (!clone)
1670 		goto out;
1671 
1672 	clone_init(io, clone);
1673 
1674 	remaining_size = size;
1675 
1676 	for (i = 0; i < nr_iovecs; i++) {
1677 		page = mempool_alloc(&cc->page_pool, gfp_mask);
1678 		if (!page) {
1679 			crypt_free_buffer_pages(cc, clone);
1680 			bio_put(clone);
1681 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1682 			goto retry;
1683 		}
1684 
1685 		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1686 
1687 		bio_add_page(clone, page, len, 0);
1688 
1689 		remaining_size -= len;
1690 	}
1691 
1692 	/* Allocate space for integrity tags */
1693 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1694 		crypt_free_buffer_pages(cc, clone);
1695 		bio_put(clone);
1696 		clone = NULL;
1697 	}
1698 out:
1699 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1700 		mutex_unlock(&cc->bio_alloc_lock);
1701 
1702 	return clone;
1703 }
1704 
1705 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1706 {
1707 	struct bio_vec *bv;
1708 	struct bvec_iter_all iter_all;
1709 
1710 	bio_for_each_segment_all(bv, clone, iter_all) {
1711 		BUG_ON(!bv->bv_page);
1712 		mempool_free(bv->bv_page, &cc->page_pool);
1713 	}
1714 }
1715 
1716 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1717 			  struct bio *bio, sector_t sector)
1718 {
1719 	io->cc = cc;
1720 	io->base_bio = bio;
1721 	io->sector = sector;
1722 	io->error = 0;
1723 	io->ctx.r.req = NULL;
1724 	io->integrity_metadata = NULL;
1725 	io->integrity_metadata_from_pool = false;
1726 	atomic_set(&io->io_pending, 0);
1727 }
1728 
1729 static void crypt_inc_pending(struct dm_crypt_io *io)
1730 {
1731 	atomic_inc(&io->io_pending);
1732 }
1733 
1734 static void kcryptd_io_bio_endio(struct work_struct *work)
1735 {
1736 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1737 	bio_endio(io->base_bio);
1738 }
1739 
1740 /*
1741  * One of the bios was finished. Check for completion of
1742  * the whole request and correctly clean up the buffer.
1743  */
1744 static void crypt_dec_pending(struct dm_crypt_io *io)
1745 {
1746 	struct crypt_config *cc = io->cc;
1747 	struct bio *base_bio = io->base_bio;
1748 	blk_status_t error = io->error;
1749 
1750 	if (!atomic_dec_and_test(&io->io_pending))
1751 		return;
1752 
1753 	if (io->ctx.r.req)
1754 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1755 
1756 	if (unlikely(io->integrity_metadata_from_pool))
1757 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1758 	else
1759 		kfree(io->integrity_metadata);
1760 
1761 	base_bio->bi_status = error;
1762 
1763 	/*
1764 	 * If we are running this function from our tasklet,
1765 	 * we can't call bio_endio() here, because it will call
1766 	 * clone_endio() from dm.c, which in turn will
1767 	 * free the current struct dm_crypt_io structure with
1768 	 * our tasklet. In this case we need to delay bio_endio()
1769 	 * execution to after the tasklet is done and dequeued.
1770 	 */
1771 	if (tasklet_trylock(&io->tasklet)) {
1772 		tasklet_unlock(&io->tasklet);
1773 		bio_endio(base_bio);
1774 		return;
1775 	}
1776 
1777 	INIT_WORK(&io->work, kcryptd_io_bio_endio);
1778 	queue_work(cc->io_queue, &io->work);
1779 }
1780 
1781 /*
1782  * kcryptd/kcryptd_io:
1783  *
1784  * Needed because it would be very unwise to do decryption in an
1785  * interrupt context.
1786  *
1787  * kcryptd performs the actual encryption or decryption.
1788  *
1789  * kcryptd_io performs the IO submission.
1790  *
1791  * They must be separated as otherwise the final stages could be
1792  * starved by new requests which can block in the first stages due
1793  * to memory allocation.
1794  *
1795  * The work is done per CPU global for all dm-crypt instances.
1796  * They should not depend on each other and do not block.
1797  */
1798 static void crypt_endio(struct bio *clone)
1799 {
1800 	struct dm_crypt_io *io = clone->bi_private;
1801 	struct crypt_config *cc = io->cc;
1802 	unsigned rw = bio_data_dir(clone);
1803 	blk_status_t error;
1804 
1805 	/*
1806 	 * free the processed pages
1807 	 */
1808 	if (rw == WRITE)
1809 		crypt_free_buffer_pages(cc, clone);
1810 
1811 	error = clone->bi_status;
1812 	bio_put(clone);
1813 
1814 	if (rw == READ && !error) {
1815 		kcryptd_queue_crypt(io);
1816 		return;
1817 	}
1818 
1819 	if (unlikely(error))
1820 		io->error = error;
1821 
1822 	crypt_dec_pending(io);
1823 }
1824 
1825 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1826 {
1827 	struct crypt_config *cc = io->cc;
1828 
1829 	clone->bi_private = io;
1830 	clone->bi_end_io  = crypt_endio;
1831 	bio_set_dev(clone, cc->dev->bdev);
1832 	clone->bi_opf	  = io->base_bio->bi_opf;
1833 }
1834 
1835 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1836 {
1837 	struct crypt_config *cc = io->cc;
1838 	struct bio *clone;
1839 
1840 	/*
1841 	 * We need the original biovec array in order to decrypt
1842 	 * the whole bio data *afterwards* -- thanks to immutable
1843 	 * biovecs we don't need to worry about the block layer
1844 	 * modifying the biovec array; so leverage bio_clone_fast().
1845 	 */
1846 	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1847 	if (!clone)
1848 		return 1;
1849 
1850 	crypt_inc_pending(io);
1851 
1852 	clone_init(io, clone);
1853 	clone->bi_iter.bi_sector = cc->start + io->sector;
1854 
1855 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1856 		crypt_dec_pending(io);
1857 		bio_put(clone);
1858 		return 1;
1859 	}
1860 
1861 	submit_bio_noacct(clone);
1862 	return 0;
1863 }
1864 
1865 static void kcryptd_io_read_work(struct work_struct *work)
1866 {
1867 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1868 
1869 	crypt_inc_pending(io);
1870 	if (kcryptd_io_read(io, GFP_NOIO))
1871 		io->error = BLK_STS_RESOURCE;
1872 	crypt_dec_pending(io);
1873 }
1874 
1875 static void kcryptd_queue_read(struct dm_crypt_io *io)
1876 {
1877 	struct crypt_config *cc = io->cc;
1878 
1879 	INIT_WORK(&io->work, kcryptd_io_read_work);
1880 	queue_work(cc->io_queue, &io->work);
1881 }
1882 
1883 static void kcryptd_io_write(struct dm_crypt_io *io)
1884 {
1885 	struct bio *clone = io->ctx.bio_out;
1886 
1887 	submit_bio_noacct(clone);
1888 }
1889 
1890 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1891 
1892 static int dmcrypt_write(void *data)
1893 {
1894 	struct crypt_config *cc = data;
1895 	struct dm_crypt_io *io;
1896 
1897 	while (1) {
1898 		struct rb_root write_tree;
1899 		struct blk_plug plug;
1900 
1901 		spin_lock_irq(&cc->write_thread_lock);
1902 continue_locked:
1903 
1904 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1905 			goto pop_from_list;
1906 
1907 		set_current_state(TASK_INTERRUPTIBLE);
1908 
1909 		spin_unlock_irq(&cc->write_thread_lock);
1910 
1911 		if (unlikely(kthread_should_stop())) {
1912 			set_current_state(TASK_RUNNING);
1913 			break;
1914 		}
1915 
1916 		schedule();
1917 
1918 		set_current_state(TASK_RUNNING);
1919 		spin_lock_irq(&cc->write_thread_lock);
1920 		goto continue_locked;
1921 
1922 pop_from_list:
1923 		write_tree = cc->write_tree;
1924 		cc->write_tree = RB_ROOT;
1925 		spin_unlock_irq(&cc->write_thread_lock);
1926 
1927 		BUG_ON(rb_parent(write_tree.rb_node));
1928 
1929 		/*
1930 		 * Note: we cannot walk the tree here with rb_next because
1931 		 * the structures may be freed when kcryptd_io_write is called.
1932 		 */
1933 		blk_start_plug(&plug);
1934 		do {
1935 			io = crypt_io_from_node(rb_first(&write_tree));
1936 			rb_erase(&io->rb_node, &write_tree);
1937 			kcryptd_io_write(io);
1938 		} while (!RB_EMPTY_ROOT(&write_tree));
1939 		blk_finish_plug(&plug);
1940 	}
1941 	return 0;
1942 }
1943 
1944 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1945 {
1946 	struct bio *clone = io->ctx.bio_out;
1947 	struct crypt_config *cc = io->cc;
1948 	unsigned long flags;
1949 	sector_t sector;
1950 	struct rb_node **rbp, *parent;
1951 
1952 	if (unlikely(io->error)) {
1953 		crypt_free_buffer_pages(cc, clone);
1954 		bio_put(clone);
1955 		crypt_dec_pending(io);
1956 		return;
1957 	}
1958 
1959 	/* crypt_convert should have filled the clone bio */
1960 	BUG_ON(io->ctx.iter_out.bi_size);
1961 
1962 	clone->bi_iter.bi_sector = cc->start + io->sector;
1963 
1964 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1965 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1966 		submit_bio_noacct(clone);
1967 		return;
1968 	}
1969 
1970 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1971 	if (RB_EMPTY_ROOT(&cc->write_tree))
1972 		wake_up_process(cc->write_thread);
1973 	rbp = &cc->write_tree.rb_node;
1974 	parent = NULL;
1975 	sector = io->sector;
1976 	while (*rbp) {
1977 		parent = *rbp;
1978 		if (sector < crypt_io_from_node(parent)->sector)
1979 			rbp = &(*rbp)->rb_left;
1980 		else
1981 			rbp = &(*rbp)->rb_right;
1982 	}
1983 	rb_link_node(&io->rb_node, parent, rbp);
1984 	rb_insert_color(&io->rb_node, &cc->write_tree);
1985 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1986 }
1987 
1988 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1989 				       struct convert_context *ctx)
1990 
1991 {
1992 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1993 		return false;
1994 
1995 	/*
1996 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1997 	 * constraints so they do not need to be issued inline by
1998 	 * kcryptd_crypt_write_convert().
1999 	 */
2000 	switch (bio_op(ctx->bio_in)) {
2001 	case REQ_OP_WRITE:
2002 	case REQ_OP_WRITE_SAME:
2003 	case REQ_OP_WRITE_ZEROES:
2004 		return true;
2005 	default:
2006 		return false;
2007 	}
2008 }
2009 
2010 static void kcryptd_crypt_write_continue(struct work_struct *work)
2011 {
2012 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2013 	struct crypt_config *cc = io->cc;
2014 	struct convert_context *ctx = &io->ctx;
2015 	int crypt_finished;
2016 	sector_t sector = io->sector;
2017 	blk_status_t r;
2018 
2019 	wait_for_completion(&ctx->restart);
2020 	reinit_completion(&ctx->restart);
2021 
2022 	r = crypt_convert(cc, &io->ctx, true, false);
2023 	if (r)
2024 		io->error = r;
2025 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2026 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2027 		/* Wait for completion signaled by kcryptd_async_done() */
2028 		wait_for_completion(&ctx->restart);
2029 		crypt_finished = 1;
2030 	}
2031 
2032 	/* Encryption was already finished, submit io now */
2033 	if (crypt_finished) {
2034 		kcryptd_crypt_write_io_submit(io, 0);
2035 		io->sector = sector;
2036 	}
2037 
2038 	crypt_dec_pending(io);
2039 }
2040 
2041 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2042 {
2043 	struct crypt_config *cc = io->cc;
2044 	struct convert_context *ctx = &io->ctx;
2045 	struct bio *clone;
2046 	int crypt_finished;
2047 	sector_t sector = io->sector;
2048 	blk_status_t r;
2049 
2050 	/*
2051 	 * Prevent io from disappearing until this function completes.
2052 	 */
2053 	crypt_inc_pending(io);
2054 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2055 
2056 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2057 	if (unlikely(!clone)) {
2058 		io->error = BLK_STS_IOERR;
2059 		goto dec;
2060 	}
2061 
2062 	io->ctx.bio_out = clone;
2063 	io->ctx.iter_out = clone->bi_iter;
2064 
2065 	sector += bio_sectors(clone);
2066 
2067 	crypt_inc_pending(io);
2068 	r = crypt_convert(cc, ctx,
2069 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2070 	/*
2071 	 * Crypto API backlogged the request, because its queue was full
2072 	 * and we're in softirq context, so continue from a workqueue
2073 	 * (TODO: is it actually possible to be in softirq in the write path?)
2074 	 */
2075 	if (r == BLK_STS_DEV_RESOURCE) {
2076 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2077 		queue_work(cc->crypt_queue, &io->work);
2078 		return;
2079 	}
2080 	if (r)
2081 		io->error = r;
2082 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2083 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2084 		/* Wait for completion signaled by kcryptd_async_done() */
2085 		wait_for_completion(&ctx->restart);
2086 		crypt_finished = 1;
2087 	}
2088 
2089 	/* Encryption was already finished, submit io now */
2090 	if (crypt_finished) {
2091 		kcryptd_crypt_write_io_submit(io, 0);
2092 		io->sector = sector;
2093 	}
2094 
2095 dec:
2096 	crypt_dec_pending(io);
2097 }
2098 
2099 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2100 {
2101 	crypt_dec_pending(io);
2102 }
2103 
2104 static void kcryptd_crypt_read_continue(struct work_struct *work)
2105 {
2106 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2107 	struct crypt_config *cc = io->cc;
2108 	blk_status_t r;
2109 
2110 	wait_for_completion(&io->ctx.restart);
2111 	reinit_completion(&io->ctx.restart);
2112 
2113 	r = crypt_convert(cc, &io->ctx, true, false);
2114 	if (r)
2115 		io->error = r;
2116 
2117 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2118 		kcryptd_crypt_read_done(io);
2119 
2120 	crypt_dec_pending(io);
2121 }
2122 
2123 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2124 {
2125 	struct crypt_config *cc = io->cc;
2126 	blk_status_t r;
2127 
2128 	crypt_inc_pending(io);
2129 
2130 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2131 			   io->sector);
2132 
2133 	r = crypt_convert(cc, &io->ctx,
2134 			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2135 	/*
2136 	 * Crypto API backlogged the request, because its queue was full
2137 	 * and we're in softirq context, so continue from a workqueue
2138 	 */
2139 	if (r == BLK_STS_DEV_RESOURCE) {
2140 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2141 		queue_work(cc->crypt_queue, &io->work);
2142 		return;
2143 	}
2144 	if (r)
2145 		io->error = r;
2146 
2147 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2148 		kcryptd_crypt_read_done(io);
2149 
2150 	crypt_dec_pending(io);
2151 }
2152 
2153 static void kcryptd_async_done(struct crypto_async_request *async_req,
2154 			       int error)
2155 {
2156 	struct dm_crypt_request *dmreq = async_req->data;
2157 	struct convert_context *ctx = dmreq->ctx;
2158 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2159 	struct crypt_config *cc = io->cc;
2160 
2161 	/*
2162 	 * A request from crypto driver backlog is going to be processed now,
2163 	 * finish the completion and continue in crypt_convert().
2164 	 * (Callback will be called for the second time for this request.)
2165 	 */
2166 	if (error == -EINPROGRESS) {
2167 		complete(&ctx->restart);
2168 		return;
2169 	}
2170 
2171 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2172 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2173 
2174 	if (error == -EBADMSG) {
2175 		char b[BDEVNAME_SIZE];
2176 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2177 			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
2178 		io->error = BLK_STS_PROTECTION;
2179 	} else if (error < 0)
2180 		io->error = BLK_STS_IOERR;
2181 
2182 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2183 
2184 	if (!atomic_dec_and_test(&ctx->cc_pending))
2185 		return;
2186 
2187 	/*
2188 	 * The request is fully completed: for inline writes, let
2189 	 * kcryptd_crypt_write_convert() do the IO submission.
2190 	 */
2191 	if (bio_data_dir(io->base_bio) == READ) {
2192 		kcryptd_crypt_read_done(io);
2193 		return;
2194 	}
2195 
2196 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2197 		complete(&ctx->restart);
2198 		return;
2199 	}
2200 
2201 	kcryptd_crypt_write_io_submit(io, 1);
2202 }
2203 
2204 static void kcryptd_crypt(struct work_struct *work)
2205 {
2206 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2207 
2208 	if (bio_data_dir(io->base_bio) == READ)
2209 		kcryptd_crypt_read_convert(io);
2210 	else
2211 		kcryptd_crypt_write_convert(io);
2212 }
2213 
2214 static void kcryptd_crypt_tasklet(unsigned long work)
2215 {
2216 	kcryptd_crypt((struct work_struct *)work);
2217 }
2218 
2219 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2220 {
2221 	struct crypt_config *cc = io->cc;
2222 
2223 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2224 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2225 		/*
2226 		 * in_irq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2227 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2228 		 * it is being executed with irqs disabled.
2229 		 */
2230 		if (in_irq() || irqs_disabled()) {
2231 			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2232 			tasklet_schedule(&io->tasklet);
2233 			return;
2234 		}
2235 
2236 		kcryptd_crypt(&io->work);
2237 		return;
2238 	}
2239 
2240 	INIT_WORK(&io->work, kcryptd_crypt);
2241 	queue_work(cc->crypt_queue, &io->work);
2242 }
2243 
2244 static void crypt_free_tfms_aead(struct crypt_config *cc)
2245 {
2246 	if (!cc->cipher_tfm.tfms_aead)
2247 		return;
2248 
2249 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2250 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2251 		cc->cipher_tfm.tfms_aead[0] = NULL;
2252 	}
2253 
2254 	kfree(cc->cipher_tfm.tfms_aead);
2255 	cc->cipher_tfm.tfms_aead = NULL;
2256 }
2257 
2258 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2259 {
2260 	unsigned i;
2261 
2262 	if (!cc->cipher_tfm.tfms)
2263 		return;
2264 
2265 	for (i = 0; i < cc->tfms_count; i++)
2266 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2267 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2268 			cc->cipher_tfm.tfms[i] = NULL;
2269 		}
2270 
2271 	kfree(cc->cipher_tfm.tfms);
2272 	cc->cipher_tfm.tfms = NULL;
2273 }
2274 
2275 static void crypt_free_tfms(struct crypt_config *cc)
2276 {
2277 	if (crypt_integrity_aead(cc))
2278 		crypt_free_tfms_aead(cc);
2279 	else
2280 		crypt_free_tfms_skcipher(cc);
2281 }
2282 
2283 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2284 {
2285 	unsigned i;
2286 	int err;
2287 
2288 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2289 				      sizeof(struct crypto_skcipher *),
2290 				      GFP_KERNEL);
2291 	if (!cc->cipher_tfm.tfms)
2292 		return -ENOMEM;
2293 
2294 	for (i = 0; i < cc->tfms_count; i++) {
2295 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2296 						CRYPTO_ALG_ALLOCATES_MEMORY);
2297 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2298 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2299 			crypt_free_tfms(cc);
2300 			return err;
2301 		}
2302 	}
2303 
2304 	/*
2305 	 * dm-crypt performance can vary greatly depending on which crypto
2306 	 * algorithm implementation is used.  Help people debug performance
2307 	 * problems by logging the ->cra_driver_name.
2308 	 */
2309 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2310 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2311 	return 0;
2312 }
2313 
2314 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2315 {
2316 	int err;
2317 
2318 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2319 	if (!cc->cipher_tfm.tfms)
2320 		return -ENOMEM;
2321 
2322 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2323 						CRYPTO_ALG_ALLOCATES_MEMORY);
2324 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2325 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2326 		crypt_free_tfms(cc);
2327 		return err;
2328 	}
2329 
2330 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2331 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2332 	return 0;
2333 }
2334 
2335 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2336 {
2337 	if (crypt_integrity_aead(cc))
2338 		return crypt_alloc_tfms_aead(cc, ciphermode);
2339 	else
2340 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2341 }
2342 
2343 static unsigned crypt_subkey_size(struct crypt_config *cc)
2344 {
2345 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2346 }
2347 
2348 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2349 {
2350 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2351 }
2352 
2353 /*
2354  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2355  * the key must be for some reason in special format.
2356  * This funcion converts cc->key to this special format.
2357  */
2358 static void crypt_copy_authenckey(char *p, const void *key,
2359 				  unsigned enckeylen, unsigned authkeylen)
2360 {
2361 	struct crypto_authenc_key_param *param;
2362 	struct rtattr *rta;
2363 
2364 	rta = (struct rtattr *)p;
2365 	param = RTA_DATA(rta);
2366 	param->enckeylen = cpu_to_be32(enckeylen);
2367 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2368 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2369 	p += RTA_SPACE(sizeof(*param));
2370 	memcpy(p, key + enckeylen, authkeylen);
2371 	p += authkeylen;
2372 	memcpy(p, key, enckeylen);
2373 }
2374 
2375 static int crypt_setkey(struct crypt_config *cc)
2376 {
2377 	unsigned subkey_size;
2378 	int err = 0, i, r;
2379 
2380 	/* Ignore extra keys (which are used for IV etc) */
2381 	subkey_size = crypt_subkey_size(cc);
2382 
2383 	if (crypt_integrity_hmac(cc)) {
2384 		if (subkey_size < cc->key_mac_size)
2385 			return -EINVAL;
2386 
2387 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2388 				      subkey_size - cc->key_mac_size,
2389 				      cc->key_mac_size);
2390 	}
2391 
2392 	for (i = 0; i < cc->tfms_count; i++) {
2393 		if (crypt_integrity_hmac(cc))
2394 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2395 				cc->authenc_key, crypt_authenckey_size(cc));
2396 		else if (crypt_integrity_aead(cc))
2397 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2398 					       cc->key + (i * subkey_size),
2399 					       subkey_size);
2400 		else
2401 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2402 						   cc->key + (i * subkey_size),
2403 						   subkey_size);
2404 		if (r)
2405 			err = r;
2406 	}
2407 
2408 	if (crypt_integrity_hmac(cc))
2409 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2410 
2411 	return err;
2412 }
2413 
2414 #ifdef CONFIG_KEYS
2415 
2416 static bool contains_whitespace(const char *str)
2417 {
2418 	while (*str)
2419 		if (isspace(*str++))
2420 			return true;
2421 	return false;
2422 }
2423 
2424 static int set_key_user(struct crypt_config *cc, struct key *key)
2425 {
2426 	const struct user_key_payload *ukp;
2427 
2428 	ukp = user_key_payload_locked(key);
2429 	if (!ukp)
2430 		return -EKEYREVOKED;
2431 
2432 	if (cc->key_size != ukp->datalen)
2433 		return -EINVAL;
2434 
2435 	memcpy(cc->key, ukp->data, cc->key_size);
2436 
2437 	return 0;
2438 }
2439 
2440 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2441 {
2442 	const struct encrypted_key_payload *ekp;
2443 
2444 	ekp = key->payload.data[0];
2445 	if (!ekp)
2446 		return -EKEYREVOKED;
2447 
2448 	if (cc->key_size != ekp->decrypted_datalen)
2449 		return -EINVAL;
2450 
2451 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2452 
2453 	return 0;
2454 }
2455 
2456 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2457 {
2458 	const struct trusted_key_payload *tkp;
2459 
2460 	tkp = key->payload.data[0];
2461 	if (!tkp)
2462 		return -EKEYREVOKED;
2463 
2464 	if (cc->key_size != tkp->key_len)
2465 		return -EINVAL;
2466 
2467 	memcpy(cc->key, tkp->key, cc->key_size);
2468 
2469 	return 0;
2470 }
2471 
2472 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2473 {
2474 	char *new_key_string, *key_desc;
2475 	int ret;
2476 	struct key_type *type;
2477 	struct key *key;
2478 	int (*set_key)(struct crypt_config *cc, struct key *key);
2479 
2480 	/*
2481 	 * Reject key_string with whitespace. dm core currently lacks code for
2482 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2483 	 */
2484 	if (contains_whitespace(key_string)) {
2485 		DMERR("whitespace chars not allowed in key string");
2486 		return -EINVAL;
2487 	}
2488 
2489 	/* look for next ':' separating key_type from key_description */
2490 	key_desc = strpbrk(key_string, ":");
2491 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2492 		return -EINVAL;
2493 
2494 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2495 		type = &key_type_logon;
2496 		set_key = set_key_user;
2497 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2498 		type = &key_type_user;
2499 		set_key = set_key_user;
2500 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2501 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2502 		type = &key_type_encrypted;
2503 		set_key = set_key_encrypted;
2504 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2505 	           !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2506 		type = &key_type_trusted;
2507 		set_key = set_key_trusted;
2508 	} else {
2509 		return -EINVAL;
2510 	}
2511 
2512 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2513 	if (!new_key_string)
2514 		return -ENOMEM;
2515 
2516 	key = request_key(type, key_desc + 1, NULL);
2517 	if (IS_ERR(key)) {
2518 		kfree_sensitive(new_key_string);
2519 		return PTR_ERR(key);
2520 	}
2521 
2522 	down_read(&key->sem);
2523 
2524 	ret = set_key(cc, key);
2525 	if (ret < 0) {
2526 		up_read(&key->sem);
2527 		key_put(key);
2528 		kfree_sensitive(new_key_string);
2529 		return ret;
2530 	}
2531 
2532 	up_read(&key->sem);
2533 	key_put(key);
2534 
2535 	/* clear the flag since following operations may invalidate previously valid key */
2536 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2537 
2538 	ret = crypt_setkey(cc);
2539 
2540 	if (!ret) {
2541 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2542 		kfree_sensitive(cc->key_string);
2543 		cc->key_string = new_key_string;
2544 	} else
2545 		kfree_sensitive(new_key_string);
2546 
2547 	return ret;
2548 }
2549 
2550 static int get_key_size(char **key_string)
2551 {
2552 	char *colon, dummy;
2553 	int ret;
2554 
2555 	if (*key_string[0] != ':')
2556 		return strlen(*key_string) >> 1;
2557 
2558 	/* look for next ':' in key string */
2559 	colon = strpbrk(*key_string + 1, ":");
2560 	if (!colon)
2561 		return -EINVAL;
2562 
2563 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2564 		return -EINVAL;
2565 
2566 	*key_string = colon;
2567 
2568 	/* remaining key string should be :<logon|user>:<key_desc> */
2569 
2570 	return ret;
2571 }
2572 
2573 #else
2574 
2575 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2576 {
2577 	return -EINVAL;
2578 }
2579 
2580 static int get_key_size(char **key_string)
2581 {
2582 	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2583 }
2584 
2585 #endif /* CONFIG_KEYS */
2586 
2587 static int crypt_set_key(struct crypt_config *cc, char *key)
2588 {
2589 	int r = -EINVAL;
2590 	int key_string_len = strlen(key);
2591 
2592 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2593 	if (!cc->key_size && strcmp(key, "-"))
2594 		goto out;
2595 
2596 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2597 	if (key[0] == ':') {
2598 		r = crypt_set_keyring_key(cc, key + 1);
2599 		goto out;
2600 	}
2601 
2602 	/* clear the flag since following operations may invalidate previously valid key */
2603 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2604 
2605 	/* wipe references to any kernel keyring key */
2606 	kfree_sensitive(cc->key_string);
2607 	cc->key_string = NULL;
2608 
2609 	/* Decode key from its hex representation. */
2610 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2611 		goto out;
2612 
2613 	r = crypt_setkey(cc);
2614 	if (!r)
2615 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2616 
2617 out:
2618 	/* Hex key string not needed after here, so wipe it. */
2619 	memset(key, '0', key_string_len);
2620 
2621 	return r;
2622 }
2623 
2624 static int crypt_wipe_key(struct crypt_config *cc)
2625 {
2626 	int r;
2627 
2628 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2629 	get_random_bytes(&cc->key, cc->key_size);
2630 
2631 	/* Wipe IV private keys */
2632 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2633 		r = cc->iv_gen_ops->wipe(cc);
2634 		if (r)
2635 			return r;
2636 	}
2637 
2638 	kfree_sensitive(cc->key_string);
2639 	cc->key_string = NULL;
2640 	r = crypt_setkey(cc);
2641 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2642 
2643 	return r;
2644 }
2645 
2646 static void crypt_calculate_pages_per_client(void)
2647 {
2648 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2649 
2650 	if (!dm_crypt_clients_n)
2651 		return;
2652 
2653 	pages /= dm_crypt_clients_n;
2654 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2655 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2656 	dm_crypt_pages_per_client = pages;
2657 }
2658 
2659 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2660 {
2661 	struct crypt_config *cc = pool_data;
2662 	struct page *page;
2663 
2664 	if (unlikely(percpu_counter_compare(&cc->n_allocated_pages, dm_crypt_pages_per_client) >= 0) &&
2665 	    likely(gfp_mask & __GFP_NORETRY))
2666 		return NULL;
2667 
2668 	page = alloc_page(gfp_mask);
2669 	if (likely(page != NULL))
2670 		percpu_counter_add(&cc->n_allocated_pages, 1);
2671 
2672 	return page;
2673 }
2674 
2675 static void crypt_page_free(void *page, void *pool_data)
2676 {
2677 	struct crypt_config *cc = pool_data;
2678 
2679 	__free_page(page);
2680 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2681 }
2682 
2683 static void crypt_dtr(struct dm_target *ti)
2684 {
2685 	struct crypt_config *cc = ti->private;
2686 
2687 	ti->private = NULL;
2688 
2689 	if (!cc)
2690 		return;
2691 
2692 	if (cc->write_thread)
2693 		kthread_stop(cc->write_thread);
2694 
2695 	if (cc->io_queue)
2696 		destroy_workqueue(cc->io_queue);
2697 	if (cc->crypt_queue)
2698 		destroy_workqueue(cc->crypt_queue);
2699 
2700 	crypt_free_tfms(cc);
2701 
2702 	bioset_exit(&cc->bs);
2703 
2704 	mempool_exit(&cc->page_pool);
2705 	mempool_exit(&cc->req_pool);
2706 	mempool_exit(&cc->tag_pool);
2707 
2708 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2709 	percpu_counter_destroy(&cc->n_allocated_pages);
2710 
2711 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2712 		cc->iv_gen_ops->dtr(cc);
2713 
2714 	if (cc->dev)
2715 		dm_put_device(ti, cc->dev);
2716 
2717 	kfree_sensitive(cc->cipher_string);
2718 	kfree_sensitive(cc->key_string);
2719 	kfree_sensitive(cc->cipher_auth);
2720 	kfree_sensitive(cc->authenc_key);
2721 
2722 	mutex_destroy(&cc->bio_alloc_lock);
2723 
2724 	/* Must zero key material before freeing */
2725 	kfree_sensitive(cc);
2726 
2727 	spin_lock(&dm_crypt_clients_lock);
2728 	WARN_ON(!dm_crypt_clients_n);
2729 	dm_crypt_clients_n--;
2730 	crypt_calculate_pages_per_client();
2731 	spin_unlock(&dm_crypt_clients_lock);
2732 }
2733 
2734 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2735 {
2736 	struct crypt_config *cc = ti->private;
2737 
2738 	if (crypt_integrity_aead(cc))
2739 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2740 	else
2741 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2742 
2743 	if (cc->iv_size)
2744 		/* at least a 64 bit sector number should fit in our buffer */
2745 		cc->iv_size = max(cc->iv_size,
2746 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2747 	else if (ivmode) {
2748 		DMWARN("Selected cipher does not support IVs");
2749 		ivmode = NULL;
2750 	}
2751 
2752 	/* Choose ivmode, see comments at iv code. */
2753 	if (ivmode == NULL)
2754 		cc->iv_gen_ops = NULL;
2755 	else if (strcmp(ivmode, "plain") == 0)
2756 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2757 	else if (strcmp(ivmode, "plain64") == 0)
2758 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2759 	else if (strcmp(ivmode, "plain64be") == 0)
2760 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2761 	else if (strcmp(ivmode, "essiv") == 0)
2762 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2763 	else if (strcmp(ivmode, "benbi") == 0)
2764 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2765 	else if (strcmp(ivmode, "null") == 0)
2766 		cc->iv_gen_ops = &crypt_iv_null_ops;
2767 	else if (strcmp(ivmode, "eboiv") == 0)
2768 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2769 	else if (strcmp(ivmode, "elephant") == 0) {
2770 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2771 		cc->key_parts = 2;
2772 		cc->key_extra_size = cc->key_size / 2;
2773 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2774 			return -EINVAL;
2775 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2776 	} else if (strcmp(ivmode, "lmk") == 0) {
2777 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2778 		/*
2779 		 * Version 2 and 3 is recognised according
2780 		 * to length of provided multi-key string.
2781 		 * If present (version 3), last key is used as IV seed.
2782 		 * All keys (including IV seed) are always the same size.
2783 		 */
2784 		if (cc->key_size % cc->key_parts) {
2785 			cc->key_parts++;
2786 			cc->key_extra_size = cc->key_size / cc->key_parts;
2787 		}
2788 	} else if (strcmp(ivmode, "tcw") == 0) {
2789 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2790 		cc->key_parts += 2; /* IV + whitening */
2791 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2792 	} else if (strcmp(ivmode, "random") == 0) {
2793 		cc->iv_gen_ops = &crypt_iv_random_ops;
2794 		/* Need storage space in integrity fields. */
2795 		cc->integrity_iv_size = cc->iv_size;
2796 	} else {
2797 		ti->error = "Invalid IV mode";
2798 		return -EINVAL;
2799 	}
2800 
2801 	return 0;
2802 }
2803 
2804 /*
2805  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2806  * The HMAC is needed to calculate tag size (HMAC digest size).
2807  * This should be probably done by crypto-api calls (once available...)
2808  */
2809 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2810 {
2811 	char *start, *end, *mac_alg = NULL;
2812 	struct crypto_ahash *mac;
2813 
2814 	if (!strstarts(cipher_api, "authenc("))
2815 		return 0;
2816 
2817 	start = strchr(cipher_api, '(');
2818 	end = strchr(cipher_api, ',');
2819 	if (!start || !end || ++start > end)
2820 		return -EINVAL;
2821 
2822 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2823 	if (!mac_alg)
2824 		return -ENOMEM;
2825 	strncpy(mac_alg, start, end - start);
2826 
2827 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2828 	kfree(mac_alg);
2829 
2830 	if (IS_ERR(mac))
2831 		return PTR_ERR(mac);
2832 
2833 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2834 	crypto_free_ahash(mac);
2835 
2836 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2837 	if (!cc->authenc_key)
2838 		return -ENOMEM;
2839 
2840 	return 0;
2841 }
2842 
2843 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2844 				char **ivmode, char **ivopts)
2845 {
2846 	struct crypt_config *cc = ti->private;
2847 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2848 	int ret = -EINVAL;
2849 
2850 	cc->tfms_count = 1;
2851 
2852 	/*
2853 	 * New format (capi: prefix)
2854 	 * capi:cipher_api_spec-iv:ivopts
2855 	 */
2856 	tmp = &cipher_in[strlen("capi:")];
2857 
2858 	/* Separate IV options if present, it can contain another '-' in hash name */
2859 	*ivopts = strrchr(tmp, ':');
2860 	if (*ivopts) {
2861 		**ivopts = '\0';
2862 		(*ivopts)++;
2863 	}
2864 	/* Parse IV mode */
2865 	*ivmode = strrchr(tmp, '-');
2866 	if (*ivmode) {
2867 		**ivmode = '\0';
2868 		(*ivmode)++;
2869 	}
2870 	/* The rest is crypto API spec */
2871 	cipher_api = tmp;
2872 
2873 	/* Alloc AEAD, can be used only in new format. */
2874 	if (crypt_integrity_aead(cc)) {
2875 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2876 		if (ret < 0) {
2877 			ti->error = "Invalid AEAD cipher spec";
2878 			return -ENOMEM;
2879 		}
2880 	}
2881 
2882 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2883 		cc->tfms_count = 64;
2884 
2885 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2886 		if (!*ivopts) {
2887 			ti->error = "Digest algorithm missing for ESSIV mode";
2888 			return -EINVAL;
2889 		}
2890 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2891 			       cipher_api, *ivopts);
2892 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2893 			ti->error = "Cannot allocate cipher string";
2894 			return -ENOMEM;
2895 		}
2896 		cipher_api = buf;
2897 	}
2898 
2899 	cc->key_parts = cc->tfms_count;
2900 
2901 	/* Allocate cipher */
2902 	ret = crypt_alloc_tfms(cc, cipher_api);
2903 	if (ret < 0) {
2904 		ti->error = "Error allocating crypto tfm";
2905 		return ret;
2906 	}
2907 
2908 	if (crypt_integrity_aead(cc))
2909 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2910 	else
2911 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2912 
2913 	return 0;
2914 }
2915 
2916 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2917 				char **ivmode, char **ivopts)
2918 {
2919 	struct crypt_config *cc = ti->private;
2920 	char *tmp, *cipher, *chainmode, *keycount;
2921 	char *cipher_api = NULL;
2922 	int ret = -EINVAL;
2923 	char dummy;
2924 
2925 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2926 		ti->error = "Bad cipher specification";
2927 		return -EINVAL;
2928 	}
2929 
2930 	/*
2931 	 * Legacy dm-crypt cipher specification
2932 	 * cipher[:keycount]-mode-iv:ivopts
2933 	 */
2934 	tmp = cipher_in;
2935 	keycount = strsep(&tmp, "-");
2936 	cipher = strsep(&keycount, ":");
2937 
2938 	if (!keycount)
2939 		cc->tfms_count = 1;
2940 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2941 		 !is_power_of_2(cc->tfms_count)) {
2942 		ti->error = "Bad cipher key count specification";
2943 		return -EINVAL;
2944 	}
2945 	cc->key_parts = cc->tfms_count;
2946 
2947 	chainmode = strsep(&tmp, "-");
2948 	*ivmode = strsep(&tmp, ":");
2949 	*ivopts = tmp;
2950 
2951 	/*
2952 	 * For compatibility with the original dm-crypt mapping format, if
2953 	 * only the cipher name is supplied, use cbc-plain.
2954 	 */
2955 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2956 		chainmode = "cbc";
2957 		*ivmode = "plain";
2958 	}
2959 
2960 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2961 		ti->error = "IV mechanism required";
2962 		return -EINVAL;
2963 	}
2964 
2965 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2966 	if (!cipher_api)
2967 		goto bad_mem;
2968 
2969 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2970 		if (!*ivopts) {
2971 			ti->error = "Digest algorithm missing for ESSIV mode";
2972 			kfree(cipher_api);
2973 			return -EINVAL;
2974 		}
2975 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2976 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2977 	} else {
2978 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2979 			       "%s(%s)", chainmode, cipher);
2980 	}
2981 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2982 		kfree(cipher_api);
2983 		goto bad_mem;
2984 	}
2985 
2986 	/* Allocate cipher */
2987 	ret = crypt_alloc_tfms(cc, cipher_api);
2988 	if (ret < 0) {
2989 		ti->error = "Error allocating crypto tfm";
2990 		kfree(cipher_api);
2991 		return ret;
2992 	}
2993 	kfree(cipher_api);
2994 
2995 	return 0;
2996 bad_mem:
2997 	ti->error = "Cannot allocate cipher strings";
2998 	return -ENOMEM;
2999 }
3000 
3001 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3002 {
3003 	struct crypt_config *cc = ti->private;
3004 	char *ivmode = NULL, *ivopts = NULL;
3005 	int ret;
3006 
3007 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3008 	if (!cc->cipher_string) {
3009 		ti->error = "Cannot allocate cipher strings";
3010 		return -ENOMEM;
3011 	}
3012 
3013 	if (strstarts(cipher_in, "capi:"))
3014 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3015 	else
3016 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3017 	if (ret)
3018 		return ret;
3019 
3020 	/* Initialize IV */
3021 	ret = crypt_ctr_ivmode(ti, ivmode);
3022 	if (ret < 0)
3023 		return ret;
3024 
3025 	/* Initialize and set key */
3026 	ret = crypt_set_key(cc, key);
3027 	if (ret < 0) {
3028 		ti->error = "Error decoding and setting key";
3029 		return ret;
3030 	}
3031 
3032 	/* Allocate IV */
3033 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3034 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3035 		if (ret < 0) {
3036 			ti->error = "Error creating IV";
3037 			return ret;
3038 		}
3039 	}
3040 
3041 	/* Initialize IV (set keys for ESSIV etc) */
3042 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3043 		ret = cc->iv_gen_ops->init(cc);
3044 		if (ret < 0) {
3045 			ti->error = "Error initialising IV";
3046 			return ret;
3047 		}
3048 	}
3049 
3050 	/* wipe the kernel key payload copy */
3051 	if (cc->key_string)
3052 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3053 
3054 	return ret;
3055 }
3056 
3057 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3058 {
3059 	struct crypt_config *cc = ti->private;
3060 	struct dm_arg_set as;
3061 	static const struct dm_arg _args[] = {
3062 		{0, 8, "Invalid number of feature args"},
3063 	};
3064 	unsigned int opt_params, val;
3065 	const char *opt_string, *sval;
3066 	char dummy;
3067 	int ret;
3068 
3069 	/* Optional parameters */
3070 	as.argc = argc;
3071 	as.argv = argv;
3072 
3073 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3074 	if (ret)
3075 		return ret;
3076 
3077 	while (opt_params--) {
3078 		opt_string = dm_shift_arg(&as);
3079 		if (!opt_string) {
3080 			ti->error = "Not enough feature arguments";
3081 			return -EINVAL;
3082 		}
3083 
3084 		if (!strcasecmp(opt_string, "allow_discards"))
3085 			ti->num_discard_bios = 1;
3086 
3087 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3088 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3089 
3090 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3091 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3092 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3093 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3094 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3095 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3096 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3097 			if (val == 0 || val > MAX_TAG_SIZE) {
3098 				ti->error = "Invalid integrity arguments";
3099 				return -EINVAL;
3100 			}
3101 			cc->on_disk_tag_size = val;
3102 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3103 			if (!strcasecmp(sval, "aead")) {
3104 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3105 			} else  if (strcasecmp(sval, "none")) {
3106 				ti->error = "Unknown integrity profile";
3107 				return -EINVAL;
3108 			}
3109 
3110 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3111 			if (!cc->cipher_auth)
3112 				return -ENOMEM;
3113 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3114 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3115 			    cc->sector_size > 4096 ||
3116 			    (cc->sector_size & (cc->sector_size - 1))) {
3117 				ti->error = "Invalid feature value for sector_size";
3118 				return -EINVAL;
3119 			}
3120 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3121 				ti->error = "Device size is not multiple of sector_size feature";
3122 				return -EINVAL;
3123 			}
3124 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3125 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3126 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3127 		else {
3128 			ti->error = "Invalid feature arguments";
3129 			return -EINVAL;
3130 		}
3131 	}
3132 
3133 	return 0;
3134 }
3135 
3136 #ifdef CONFIG_BLK_DEV_ZONED
3137 static int crypt_report_zones(struct dm_target *ti,
3138 		struct dm_report_zones_args *args, unsigned int nr_zones)
3139 {
3140 	struct crypt_config *cc = ti->private;
3141 	sector_t sector = cc->start + dm_target_offset(ti, args->next_sector);
3142 
3143 	args->start = cc->start;
3144 	return blkdev_report_zones(cc->dev->bdev, sector, nr_zones,
3145 				   dm_report_zones_cb, args);
3146 }
3147 #else
3148 #define crypt_report_zones NULL
3149 #endif
3150 
3151 /*
3152  * Construct an encryption mapping:
3153  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3154  */
3155 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3156 {
3157 	struct crypt_config *cc;
3158 	const char *devname = dm_table_device_name(ti->table);
3159 	int key_size;
3160 	unsigned int align_mask;
3161 	unsigned long long tmpll;
3162 	int ret;
3163 	size_t iv_size_padding, additional_req_size;
3164 	char dummy;
3165 
3166 	if (argc < 5) {
3167 		ti->error = "Not enough arguments";
3168 		return -EINVAL;
3169 	}
3170 
3171 	key_size = get_key_size(&argv[1]);
3172 	if (key_size < 0) {
3173 		ti->error = "Cannot parse key size";
3174 		return -EINVAL;
3175 	}
3176 
3177 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3178 	if (!cc) {
3179 		ti->error = "Cannot allocate encryption context";
3180 		return -ENOMEM;
3181 	}
3182 	cc->key_size = key_size;
3183 	cc->sector_size = (1 << SECTOR_SHIFT);
3184 	cc->sector_shift = 0;
3185 
3186 	ti->private = cc;
3187 
3188 	spin_lock(&dm_crypt_clients_lock);
3189 	dm_crypt_clients_n++;
3190 	crypt_calculate_pages_per_client();
3191 	spin_unlock(&dm_crypt_clients_lock);
3192 
3193 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3194 	if (ret < 0)
3195 		goto bad;
3196 
3197 	/* Optional parameters need to be read before cipher constructor */
3198 	if (argc > 5) {
3199 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3200 		if (ret)
3201 			goto bad;
3202 	}
3203 
3204 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3205 	if (ret < 0)
3206 		goto bad;
3207 
3208 	if (crypt_integrity_aead(cc)) {
3209 		cc->dmreq_start = sizeof(struct aead_request);
3210 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3211 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3212 	} else {
3213 		cc->dmreq_start = sizeof(struct skcipher_request);
3214 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3215 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3216 	}
3217 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3218 
3219 	if (align_mask < CRYPTO_MINALIGN) {
3220 		/* Allocate the padding exactly */
3221 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3222 				& align_mask;
3223 	} else {
3224 		/*
3225 		 * If the cipher requires greater alignment than kmalloc
3226 		 * alignment, we don't know the exact position of the
3227 		 * initialization vector. We must assume worst case.
3228 		 */
3229 		iv_size_padding = align_mask;
3230 	}
3231 
3232 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3233 	additional_req_size = sizeof(struct dm_crypt_request) +
3234 		iv_size_padding + cc->iv_size +
3235 		cc->iv_size +
3236 		sizeof(uint64_t) +
3237 		sizeof(unsigned int);
3238 
3239 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3240 	if (ret) {
3241 		ti->error = "Cannot allocate crypt request mempool";
3242 		goto bad;
3243 	}
3244 
3245 	cc->per_bio_data_size = ti->per_io_data_size =
3246 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3247 		      ARCH_KMALLOC_MINALIGN);
3248 
3249 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3250 	if (ret) {
3251 		ti->error = "Cannot allocate page mempool";
3252 		goto bad;
3253 	}
3254 
3255 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3256 	if (ret) {
3257 		ti->error = "Cannot allocate crypt bioset";
3258 		goto bad;
3259 	}
3260 
3261 	mutex_init(&cc->bio_alloc_lock);
3262 
3263 	ret = -EINVAL;
3264 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3265 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3266 		ti->error = "Invalid iv_offset sector";
3267 		goto bad;
3268 	}
3269 	cc->iv_offset = tmpll;
3270 
3271 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3272 	if (ret) {
3273 		ti->error = "Device lookup failed";
3274 		goto bad;
3275 	}
3276 
3277 	ret = -EINVAL;
3278 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3279 		ti->error = "Invalid device sector";
3280 		goto bad;
3281 	}
3282 	cc->start = tmpll;
3283 
3284 	/*
3285 	 * For zoned block devices, we need to preserve the issuer write
3286 	 * ordering. To do so, disable write workqueues and force inline
3287 	 * encryption completion.
3288 	 */
3289 	if (bdev_is_zoned(cc->dev->bdev)) {
3290 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3291 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3292 	}
3293 
3294 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3295 		ret = crypt_integrity_ctr(cc, ti);
3296 		if (ret)
3297 			goto bad;
3298 
3299 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3300 		if (!cc->tag_pool_max_sectors)
3301 			cc->tag_pool_max_sectors = 1;
3302 
3303 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3304 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3305 		if (ret) {
3306 			ti->error = "Cannot allocate integrity tags mempool";
3307 			goto bad;
3308 		}
3309 
3310 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3311 	}
3312 
3313 	ret = -ENOMEM;
3314 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3315 	if (!cc->io_queue) {
3316 		ti->error = "Couldn't create kcryptd io queue";
3317 		goto bad;
3318 	}
3319 
3320 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3321 		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3322 						  1, devname);
3323 	else
3324 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3325 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3326 						  num_online_cpus(), devname);
3327 	if (!cc->crypt_queue) {
3328 		ti->error = "Couldn't create kcryptd queue";
3329 		goto bad;
3330 	}
3331 
3332 	spin_lock_init(&cc->write_thread_lock);
3333 	cc->write_tree = RB_ROOT;
3334 
3335 	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3336 	if (IS_ERR(cc->write_thread)) {
3337 		ret = PTR_ERR(cc->write_thread);
3338 		cc->write_thread = NULL;
3339 		ti->error = "Couldn't spawn write thread";
3340 		goto bad;
3341 	}
3342 	wake_up_process(cc->write_thread);
3343 
3344 	ti->num_flush_bios = 1;
3345 	ti->limit_swap_bios = true;
3346 
3347 	return 0;
3348 
3349 bad:
3350 	crypt_dtr(ti);
3351 	return ret;
3352 }
3353 
3354 static int crypt_map(struct dm_target *ti, struct bio *bio)
3355 {
3356 	struct dm_crypt_io *io;
3357 	struct crypt_config *cc = ti->private;
3358 
3359 	/*
3360 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3361 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3362 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3363 	 */
3364 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3365 	    bio_op(bio) == REQ_OP_DISCARD)) {
3366 		bio_set_dev(bio, cc->dev->bdev);
3367 		if (bio_sectors(bio))
3368 			bio->bi_iter.bi_sector = cc->start +
3369 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3370 		return DM_MAPIO_REMAPPED;
3371 	}
3372 
3373 	/*
3374 	 * Check if bio is too large, split as needed.
3375 	 */
3376 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3377 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3378 		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3379 
3380 	/*
3381 	 * Ensure that bio is a multiple of internal sector encryption size
3382 	 * and is aligned to this size as defined in IO hints.
3383 	 */
3384 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3385 		return DM_MAPIO_KILL;
3386 
3387 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3388 		return DM_MAPIO_KILL;
3389 
3390 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3391 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3392 
3393 	if (cc->on_disk_tag_size) {
3394 		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3395 
3396 		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3397 		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3398 				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3399 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3400 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3401 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3402 			io->integrity_metadata_from_pool = true;
3403 		}
3404 	}
3405 
3406 	if (crypt_integrity_aead(cc))
3407 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3408 	else
3409 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3410 
3411 	if (bio_data_dir(io->base_bio) == READ) {
3412 		if (kcryptd_io_read(io, GFP_NOWAIT))
3413 			kcryptd_queue_read(io);
3414 	} else
3415 		kcryptd_queue_crypt(io);
3416 
3417 	return DM_MAPIO_SUBMITTED;
3418 }
3419 
3420 static void crypt_status(struct dm_target *ti, status_type_t type,
3421 			 unsigned status_flags, char *result, unsigned maxlen)
3422 {
3423 	struct crypt_config *cc = ti->private;
3424 	unsigned i, sz = 0;
3425 	int num_feature_args = 0;
3426 
3427 	switch (type) {
3428 	case STATUSTYPE_INFO:
3429 		result[0] = '\0';
3430 		break;
3431 
3432 	case STATUSTYPE_TABLE:
3433 		DMEMIT("%s ", cc->cipher_string);
3434 
3435 		if (cc->key_size > 0) {
3436 			if (cc->key_string)
3437 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3438 			else
3439 				for (i = 0; i < cc->key_size; i++)
3440 					DMEMIT("%02x", cc->key[i]);
3441 		} else
3442 			DMEMIT("-");
3443 
3444 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3445 				cc->dev->name, (unsigned long long)cc->start);
3446 
3447 		num_feature_args += !!ti->num_discard_bios;
3448 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3449 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3450 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3451 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3452 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3453 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3454 		if (cc->on_disk_tag_size)
3455 			num_feature_args++;
3456 		if (num_feature_args) {
3457 			DMEMIT(" %d", num_feature_args);
3458 			if (ti->num_discard_bios)
3459 				DMEMIT(" allow_discards");
3460 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3461 				DMEMIT(" same_cpu_crypt");
3462 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3463 				DMEMIT(" submit_from_crypt_cpus");
3464 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3465 				DMEMIT(" no_read_workqueue");
3466 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3467 				DMEMIT(" no_write_workqueue");
3468 			if (cc->on_disk_tag_size)
3469 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3470 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3471 				DMEMIT(" sector_size:%d", cc->sector_size);
3472 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3473 				DMEMIT(" iv_large_sectors");
3474 		}
3475 
3476 		break;
3477 	}
3478 }
3479 
3480 static void crypt_postsuspend(struct dm_target *ti)
3481 {
3482 	struct crypt_config *cc = ti->private;
3483 
3484 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3485 }
3486 
3487 static int crypt_preresume(struct dm_target *ti)
3488 {
3489 	struct crypt_config *cc = ti->private;
3490 
3491 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3492 		DMERR("aborting resume - crypt key is not set.");
3493 		return -EAGAIN;
3494 	}
3495 
3496 	return 0;
3497 }
3498 
3499 static void crypt_resume(struct dm_target *ti)
3500 {
3501 	struct crypt_config *cc = ti->private;
3502 
3503 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3504 }
3505 
3506 /* Message interface
3507  *	key set <key>
3508  *	key wipe
3509  */
3510 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3511 			 char *result, unsigned maxlen)
3512 {
3513 	struct crypt_config *cc = ti->private;
3514 	int key_size, ret = -EINVAL;
3515 
3516 	if (argc < 2)
3517 		goto error;
3518 
3519 	if (!strcasecmp(argv[0], "key")) {
3520 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3521 			DMWARN("not suspended during key manipulation.");
3522 			return -EINVAL;
3523 		}
3524 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3525 			/* The key size may not be changed. */
3526 			key_size = get_key_size(&argv[2]);
3527 			if (key_size < 0 || cc->key_size != key_size) {
3528 				memset(argv[2], '0', strlen(argv[2]));
3529 				return -EINVAL;
3530 			}
3531 
3532 			ret = crypt_set_key(cc, argv[2]);
3533 			if (ret)
3534 				return ret;
3535 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3536 				ret = cc->iv_gen_ops->init(cc);
3537 			/* wipe the kernel key payload copy */
3538 			if (cc->key_string)
3539 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3540 			return ret;
3541 		}
3542 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3543 			return crypt_wipe_key(cc);
3544 	}
3545 
3546 error:
3547 	DMWARN("unrecognised message received.");
3548 	return -EINVAL;
3549 }
3550 
3551 static int crypt_iterate_devices(struct dm_target *ti,
3552 				 iterate_devices_callout_fn fn, void *data)
3553 {
3554 	struct crypt_config *cc = ti->private;
3555 
3556 	return fn(ti, cc->dev, cc->start, ti->len, data);
3557 }
3558 
3559 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3560 {
3561 	struct crypt_config *cc = ti->private;
3562 
3563 	/*
3564 	 * Unfortunate constraint that is required to avoid the potential
3565 	 * for exceeding underlying device's max_segments limits -- due to
3566 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3567 	 * bio that are not as physically contiguous as the original bio.
3568 	 */
3569 	limits->max_segment_size = PAGE_SIZE;
3570 
3571 	limits->logical_block_size =
3572 		max_t(unsigned, limits->logical_block_size, cc->sector_size);
3573 	limits->physical_block_size =
3574 		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3575 	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3576 }
3577 
3578 static struct target_type crypt_target = {
3579 	.name   = "crypt",
3580 	.version = {1, 23, 0},
3581 	.module = THIS_MODULE,
3582 	.ctr    = crypt_ctr,
3583 	.dtr    = crypt_dtr,
3584 	.features = DM_TARGET_ZONED_HM,
3585 	.report_zones = crypt_report_zones,
3586 	.map    = crypt_map,
3587 	.status = crypt_status,
3588 	.postsuspend = crypt_postsuspend,
3589 	.preresume = crypt_preresume,
3590 	.resume = crypt_resume,
3591 	.message = crypt_message,
3592 	.iterate_devices = crypt_iterate_devices,
3593 	.io_hints = crypt_io_hints,
3594 };
3595 
3596 static int __init dm_crypt_init(void)
3597 {
3598 	int r;
3599 
3600 	r = dm_register_target(&crypt_target);
3601 	if (r < 0)
3602 		DMERR("register failed %d", r);
3603 
3604 	return r;
3605 }
3606 
3607 static void __exit dm_crypt_exit(void)
3608 {
3609 	dm_unregister_target(&crypt_target);
3610 }
3611 
3612 module_init(dm_crypt_init);
3613 module_exit(dm_crypt_exit);
3614 
3615 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3616 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3617 MODULE_LICENSE("GPL");
3618