xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 84b102f5)
1 /*
2  * Copyright (C) 2003 Jana Saout <jana@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
5  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
6  *
7  * This file is released under the GPL.
8  */
9 
10 #include <linux/completion.h>
11 #include <linux/err.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/key.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/crypto.h>
21 #include <linux/workqueue.h>
22 #include <linux/kthread.h>
23 #include <linux/backing-dev.h>
24 #include <linux/atomic.h>
25 #include <linux/scatterlist.h>
26 #include <linux/rbtree.h>
27 #include <linux/ctype.h>
28 #include <asm/page.h>
29 #include <asm/unaligned.h>
30 #include <crypto/hash.h>
31 #include <crypto/md5.h>
32 #include <crypto/algapi.h>
33 #include <crypto/skcipher.h>
34 #include <crypto/aead.h>
35 #include <crypto/authenc.h>
36 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
37 #include <linux/key-type.h>
38 #include <keys/user-type.h>
39 #include <keys/encrypted-type.h>
40 
41 #include <linux/device-mapper.h>
42 
43 #define DM_MSG_PREFIX "crypt"
44 
45 /*
46  * context holding the current state of a multi-part conversion
47  */
48 struct convert_context {
49 	struct completion restart;
50 	struct bio *bio_in;
51 	struct bio *bio_out;
52 	struct bvec_iter iter_in;
53 	struct bvec_iter iter_out;
54 	u64 cc_sector;
55 	atomic_t cc_pending;
56 	union {
57 		struct skcipher_request *req;
58 		struct aead_request *req_aead;
59 	} r;
60 
61 };
62 
63 /*
64  * per bio private data
65  */
66 struct dm_crypt_io {
67 	struct crypt_config *cc;
68 	struct bio *base_bio;
69 	u8 *integrity_metadata;
70 	bool integrity_metadata_from_pool;
71 	struct work_struct work;
72 	struct tasklet_struct tasklet;
73 
74 	struct convert_context ctx;
75 
76 	atomic_t io_pending;
77 	blk_status_t error;
78 	sector_t sector;
79 
80 	struct rb_node rb_node;
81 } CRYPTO_MINALIGN_ATTR;
82 
83 struct dm_crypt_request {
84 	struct convert_context *ctx;
85 	struct scatterlist sg_in[4];
86 	struct scatterlist sg_out[4];
87 	u64 iv_sector;
88 };
89 
90 struct crypt_config;
91 
92 struct crypt_iv_operations {
93 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
94 		   const char *opts);
95 	void (*dtr)(struct crypt_config *cc);
96 	int (*init)(struct crypt_config *cc);
97 	int (*wipe)(struct crypt_config *cc);
98 	int (*generator)(struct crypt_config *cc, u8 *iv,
99 			 struct dm_crypt_request *dmreq);
100 	int (*post)(struct crypt_config *cc, u8 *iv,
101 		    struct dm_crypt_request *dmreq);
102 };
103 
104 struct iv_benbi_private {
105 	int shift;
106 };
107 
108 #define LMK_SEED_SIZE 64 /* hash + 0 */
109 struct iv_lmk_private {
110 	struct crypto_shash *hash_tfm;
111 	u8 *seed;
112 };
113 
114 #define TCW_WHITENING_SIZE 16
115 struct iv_tcw_private {
116 	struct crypto_shash *crc32_tfm;
117 	u8 *iv_seed;
118 	u8 *whitening;
119 };
120 
121 #define ELEPHANT_MAX_KEY_SIZE 32
122 struct iv_elephant_private {
123 	struct crypto_skcipher *tfm;
124 };
125 
126 /*
127  * Crypt: maps a linear range of a block device
128  * and encrypts / decrypts at the same time.
129  */
130 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
131 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
132 	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
133 	     DM_CRYPT_WRITE_INLINE };
134 
135 enum cipher_flags {
136 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cihper */
137 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
138 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
139 };
140 
141 /*
142  * The fields in here must be read only after initialization.
143  */
144 struct crypt_config {
145 	struct dm_dev *dev;
146 	sector_t start;
147 
148 	struct percpu_counter n_allocated_pages;
149 
150 	struct workqueue_struct *io_queue;
151 	struct workqueue_struct *crypt_queue;
152 
153 	spinlock_t write_thread_lock;
154 	struct task_struct *write_thread;
155 	struct rb_root write_tree;
156 
157 	char *cipher_string;
158 	char *cipher_auth;
159 	char *key_string;
160 
161 	const struct crypt_iv_operations *iv_gen_ops;
162 	union {
163 		struct iv_benbi_private benbi;
164 		struct iv_lmk_private lmk;
165 		struct iv_tcw_private tcw;
166 		struct iv_elephant_private elephant;
167 	} iv_gen_private;
168 	u64 iv_offset;
169 	unsigned int iv_size;
170 	unsigned short int sector_size;
171 	unsigned char sector_shift;
172 
173 	union {
174 		struct crypto_skcipher **tfms;
175 		struct crypto_aead **tfms_aead;
176 	} cipher_tfm;
177 	unsigned tfms_count;
178 	unsigned long cipher_flags;
179 
180 	/*
181 	 * Layout of each crypto request:
182 	 *
183 	 *   struct skcipher_request
184 	 *      context
185 	 *      padding
186 	 *   struct dm_crypt_request
187 	 *      padding
188 	 *   IV
189 	 *
190 	 * The padding is added so that dm_crypt_request and the IV are
191 	 * correctly aligned.
192 	 */
193 	unsigned int dmreq_start;
194 
195 	unsigned int per_bio_data_size;
196 
197 	unsigned long flags;
198 	unsigned int key_size;
199 	unsigned int key_parts;      /* independent parts in key buffer */
200 	unsigned int key_extra_size; /* additional keys length */
201 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
202 
203 	unsigned int integrity_tag_size;
204 	unsigned int integrity_iv_size;
205 	unsigned int on_disk_tag_size;
206 
207 	/*
208 	 * pool for per bio private data, crypto requests,
209 	 * encryption requeusts/buffer pages and integrity tags
210 	 */
211 	unsigned tag_pool_max_sectors;
212 	mempool_t tag_pool;
213 	mempool_t req_pool;
214 	mempool_t page_pool;
215 
216 	struct bio_set bs;
217 	struct mutex bio_alloc_lock;
218 
219 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
220 	u8 key[];
221 };
222 
223 #define MIN_IOS		64
224 #define MAX_TAG_SIZE	480
225 #define POOL_ENTRY_SIZE	512
226 
227 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
228 static unsigned dm_crypt_clients_n = 0;
229 static volatile unsigned long dm_crypt_pages_per_client;
230 #define DM_CRYPT_MEMORY_PERCENT			2
231 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_PAGES * 16)
232 
233 static void clone_init(struct dm_crypt_io *, struct bio *);
234 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
235 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
236 					     struct scatterlist *sg);
237 
238 static bool crypt_integrity_aead(struct crypt_config *cc);
239 
240 /*
241  * Use this to access cipher attributes that are independent of the key.
242  */
243 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
244 {
245 	return cc->cipher_tfm.tfms[0];
246 }
247 
248 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
249 {
250 	return cc->cipher_tfm.tfms_aead[0];
251 }
252 
253 /*
254  * Different IV generation algorithms:
255  *
256  * plain: the initial vector is the 32-bit little-endian version of the sector
257  *        number, padded with zeros if necessary.
258  *
259  * plain64: the initial vector is the 64-bit little-endian version of the sector
260  *        number, padded with zeros if necessary.
261  *
262  * plain64be: the initial vector is the 64-bit big-endian version of the sector
263  *        number, padded with zeros if necessary.
264  *
265  * essiv: "encrypted sector|salt initial vector", the sector number is
266  *        encrypted with the bulk cipher using a salt as key. The salt
267  *        should be derived from the bulk cipher's key via hashing.
268  *
269  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
270  *        (needed for LRW-32-AES and possible other narrow block modes)
271  *
272  * null: the initial vector is always zero.  Provides compatibility with
273  *       obsolete loop_fish2 devices.  Do not use for new devices.
274  *
275  * lmk:  Compatible implementation of the block chaining mode used
276  *       by the Loop-AES block device encryption system
277  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
278  *       It operates on full 512 byte sectors and uses CBC
279  *       with an IV derived from the sector number, the data and
280  *       optionally extra IV seed.
281  *       This means that after decryption the first block
282  *       of sector must be tweaked according to decrypted data.
283  *       Loop-AES can use three encryption schemes:
284  *         version 1: is plain aes-cbc mode
285  *         version 2: uses 64 multikey scheme with lmk IV generator
286  *         version 3: the same as version 2 with additional IV seed
287  *                   (it uses 65 keys, last key is used as IV seed)
288  *
289  * tcw:  Compatible implementation of the block chaining mode used
290  *       by the TrueCrypt device encryption system (prior to version 4.1).
291  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
292  *       It operates on full 512 byte sectors and uses CBC
293  *       with an IV derived from initial key and the sector number.
294  *       In addition, whitening value is applied on every sector, whitening
295  *       is calculated from initial key, sector number and mixed using CRC32.
296  *       Note that this encryption scheme is vulnerable to watermarking attacks
297  *       and should be used for old compatible containers access only.
298  *
299  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
300  *        The IV is encrypted little-endian byte-offset (with the same key
301  *        and cipher as the volume).
302  *
303  * elephant: The extended version of eboiv with additional Elephant diffuser
304  *           used with Bitlocker CBC mode.
305  *           This mode was used in older Windows systems
306  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
307  */
308 
309 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
310 			      struct dm_crypt_request *dmreq)
311 {
312 	memset(iv, 0, cc->iv_size);
313 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
314 
315 	return 0;
316 }
317 
318 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
319 				struct dm_crypt_request *dmreq)
320 {
321 	memset(iv, 0, cc->iv_size);
322 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
323 
324 	return 0;
325 }
326 
327 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
328 				  struct dm_crypt_request *dmreq)
329 {
330 	memset(iv, 0, cc->iv_size);
331 	/* iv_size is at least of size u64; usually it is 16 bytes */
332 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
333 
334 	return 0;
335 }
336 
337 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
338 			      struct dm_crypt_request *dmreq)
339 {
340 	/*
341 	 * ESSIV encryption of the IV is now handled by the crypto API,
342 	 * so just pass the plain sector number here.
343 	 */
344 	memset(iv, 0, cc->iv_size);
345 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
346 
347 	return 0;
348 }
349 
350 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
351 			      const char *opts)
352 {
353 	unsigned bs;
354 	int log;
355 
356 	if (crypt_integrity_aead(cc))
357 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
358 	else
359 		bs = crypto_skcipher_blocksize(any_tfm(cc));
360 	log = ilog2(bs);
361 
362 	/* we need to calculate how far we must shift the sector count
363 	 * to get the cipher block count, we use this shift in _gen */
364 
365 	if (1 << log != bs) {
366 		ti->error = "cypher blocksize is not a power of 2";
367 		return -EINVAL;
368 	}
369 
370 	if (log > 9) {
371 		ti->error = "cypher blocksize is > 512";
372 		return -EINVAL;
373 	}
374 
375 	cc->iv_gen_private.benbi.shift = 9 - log;
376 
377 	return 0;
378 }
379 
380 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
381 {
382 }
383 
384 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
385 			      struct dm_crypt_request *dmreq)
386 {
387 	__be64 val;
388 
389 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
390 
391 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
392 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
393 
394 	return 0;
395 }
396 
397 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
398 			     struct dm_crypt_request *dmreq)
399 {
400 	memset(iv, 0, cc->iv_size);
401 
402 	return 0;
403 }
404 
405 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
406 {
407 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
408 
409 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
410 		crypto_free_shash(lmk->hash_tfm);
411 	lmk->hash_tfm = NULL;
412 
413 	kfree_sensitive(lmk->seed);
414 	lmk->seed = NULL;
415 }
416 
417 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
418 			    const char *opts)
419 {
420 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
421 
422 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
423 		ti->error = "Unsupported sector size for LMK";
424 		return -EINVAL;
425 	}
426 
427 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
428 					   CRYPTO_ALG_ALLOCATES_MEMORY);
429 	if (IS_ERR(lmk->hash_tfm)) {
430 		ti->error = "Error initializing LMK hash";
431 		return PTR_ERR(lmk->hash_tfm);
432 	}
433 
434 	/* No seed in LMK version 2 */
435 	if (cc->key_parts == cc->tfms_count) {
436 		lmk->seed = NULL;
437 		return 0;
438 	}
439 
440 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
441 	if (!lmk->seed) {
442 		crypt_iv_lmk_dtr(cc);
443 		ti->error = "Error kmallocing seed storage in LMK";
444 		return -ENOMEM;
445 	}
446 
447 	return 0;
448 }
449 
450 static int crypt_iv_lmk_init(struct crypt_config *cc)
451 {
452 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
453 	int subkey_size = cc->key_size / cc->key_parts;
454 
455 	/* LMK seed is on the position of LMK_KEYS + 1 key */
456 	if (lmk->seed)
457 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
458 		       crypto_shash_digestsize(lmk->hash_tfm));
459 
460 	return 0;
461 }
462 
463 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
464 {
465 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
466 
467 	if (lmk->seed)
468 		memset(lmk->seed, 0, LMK_SEED_SIZE);
469 
470 	return 0;
471 }
472 
473 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
474 			    struct dm_crypt_request *dmreq,
475 			    u8 *data)
476 {
477 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
478 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
479 	struct md5_state md5state;
480 	__le32 buf[4];
481 	int i, r;
482 
483 	desc->tfm = lmk->hash_tfm;
484 
485 	r = crypto_shash_init(desc);
486 	if (r)
487 		return r;
488 
489 	if (lmk->seed) {
490 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
491 		if (r)
492 			return r;
493 	}
494 
495 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
496 	r = crypto_shash_update(desc, data + 16, 16 * 31);
497 	if (r)
498 		return r;
499 
500 	/* Sector is cropped to 56 bits here */
501 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
502 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
503 	buf[2] = cpu_to_le32(4024);
504 	buf[3] = 0;
505 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
506 	if (r)
507 		return r;
508 
509 	/* No MD5 padding here */
510 	r = crypto_shash_export(desc, &md5state);
511 	if (r)
512 		return r;
513 
514 	for (i = 0; i < MD5_HASH_WORDS; i++)
515 		__cpu_to_le32s(&md5state.hash[i]);
516 	memcpy(iv, &md5state.hash, cc->iv_size);
517 
518 	return 0;
519 }
520 
521 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
522 			    struct dm_crypt_request *dmreq)
523 {
524 	struct scatterlist *sg;
525 	u8 *src;
526 	int r = 0;
527 
528 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
529 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
530 		src = kmap_atomic(sg_page(sg));
531 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
532 		kunmap_atomic(src);
533 	} else
534 		memset(iv, 0, cc->iv_size);
535 
536 	return r;
537 }
538 
539 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
540 			     struct dm_crypt_request *dmreq)
541 {
542 	struct scatterlist *sg;
543 	u8 *dst;
544 	int r;
545 
546 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
547 		return 0;
548 
549 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
550 	dst = kmap_atomic(sg_page(sg));
551 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
552 
553 	/* Tweak the first block of plaintext sector */
554 	if (!r)
555 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
556 
557 	kunmap_atomic(dst);
558 	return r;
559 }
560 
561 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
562 {
563 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
564 
565 	kfree_sensitive(tcw->iv_seed);
566 	tcw->iv_seed = NULL;
567 	kfree_sensitive(tcw->whitening);
568 	tcw->whitening = NULL;
569 
570 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
571 		crypto_free_shash(tcw->crc32_tfm);
572 	tcw->crc32_tfm = NULL;
573 }
574 
575 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
576 			    const char *opts)
577 {
578 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
579 
580 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
581 		ti->error = "Unsupported sector size for TCW";
582 		return -EINVAL;
583 	}
584 
585 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
586 		ti->error = "Wrong key size for TCW";
587 		return -EINVAL;
588 	}
589 
590 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
591 					    CRYPTO_ALG_ALLOCATES_MEMORY);
592 	if (IS_ERR(tcw->crc32_tfm)) {
593 		ti->error = "Error initializing CRC32 in TCW";
594 		return PTR_ERR(tcw->crc32_tfm);
595 	}
596 
597 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
598 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
599 	if (!tcw->iv_seed || !tcw->whitening) {
600 		crypt_iv_tcw_dtr(cc);
601 		ti->error = "Error allocating seed storage in TCW";
602 		return -ENOMEM;
603 	}
604 
605 	return 0;
606 }
607 
608 static int crypt_iv_tcw_init(struct crypt_config *cc)
609 {
610 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
611 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
612 
613 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
614 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
615 	       TCW_WHITENING_SIZE);
616 
617 	return 0;
618 }
619 
620 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
621 {
622 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
623 
624 	memset(tcw->iv_seed, 0, cc->iv_size);
625 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
626 
627 	return 0;
628 }
629 
630 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
631 				  struct dm_crypt_request *dmreq,
632 				  u8 *data)
633 {
634 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
635 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
636 	u8 buf[TCW_WHITENING_SIZE];
637 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
638 	int i, r;
639 
640 	/* xor whitening with sector number */
641 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
642 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
643 
644 	/* calculate crc32 for every 32bit part and xor it */
645 	desc->tfm = tcw->crc32_tfm;
646 	for (i = 0; i < 4; i++) {
647 		r = crypto_shash_init(desc);
648 		if (r)
649 			goto out;
650 		r = crypto_shash_update(desc, &buf[i * 4], 4);
651 		if (r)
652 			goto out;
653 		r = crypto_shash_final(desc, &buf[i * 4]);
654 		if (r)
655 			goto out;
656 	}
657 	crypto_xor(&buf[0], &buf[12], 4);
658 	crypto_xor(&buf[4], &buf[8], 4);
659 
660 	/* apply whitening (8 bytes) to whole sector */
661 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
662 		crypto_xor(data + i * 8, buf, 8);
663 out:
664 	memzero_explicit(buf, sizeof(buf));
665 	return r;
666 }
667 
668 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
669 			    struct dm_crypt_request *dmreq)
670 {
671 	struct scatterlist *sg;
672 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
673 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
674 	u8 *src;
675 	int r = 0;
676 
677 	/* Remove whitening from ciphertext */
678 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
679 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
680 		src = kmap_atomic(sg_page(sg));
681 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
682 		kunmap_atomic(src);
683 	}
684 
685 	/* Calculate IV */
686 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
687 	if (cc->iv_size > 8)
688 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
689 			       cc->iv_size - 8);
690 
691 	return r;
692 }
693 
694 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
695 			     struct dm_crypt_request *dmreq)
696 {
697 	struct scatterlist *sg;
698 	u8 *dst;
699 	int r;
700 
701 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
702 		return 0;
703 
704 	/* Apply whitening on ciphertext */
705 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
706 	dst = kmap_atomic(sg_page(sg));
707 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
708 	kunmap_atomic(dst);
709 
710 	return r;
711 }
712 
713 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
714 				struct dm_crypt_request *dmreq)
715 {
716 	/* Used only for writes, there must be an additional space to store IV */
717 	get_random_bytes(iv, cc->iv_size);
718 	return 0;
719 }
720 
721 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
722 			    const char *opts)
723 {
724 	if (crypt_integrity_aead(cc)) {
725 		ti->error = "AEAD transforms not supported for EBOIV";
726 		return -EINVAL;
727 	}
728 
729 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
730 		ti->error = "Block size of EBOIV cipher does "
731 			    "not match IV size of block cipher";
732 		return -EINVAL;
733 	}
734 
735 	return 0;
736 }
737 
738 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
739 			    struct dm_crypt_request *dmreq)
740 {
741 	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
742 	struct skcipher_request *req;
743 	struct scatterlist src, dst;
744 	DECLARE_CRYPTO_WAIT(wait);
745 	int err;
746 
747 	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
748 	if (!req)
749 		return -ENOMEM;
750 
751 	memset(buf, 0, cc->iv_size);
752 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
753 
754 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
755 	sg_init_one(&dst, iv, cc->iv_size);
756 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
757 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
758 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
759 	skcipher_request_free(req);
760 
761 	return err;
762 }
763 
764 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
765 {
766 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
767 
768 	crypto_free_skcipher(elephant->tfm);
769 	elephant->tfm = NULL;
770 }
771 
772 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
773 			    const char *opts)
774 {
775 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
776 	int r;
777 
778 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
779 					      CRYPTO_ALG_ALLOCATES_MEMORY);
780 	if (IS_ERR(elephant->tfm)) {
781 		r = PTR_ERR(elephant->tfm);
782 		elephant->tfm = NULL;
783 		return r;
784 	}
785 
786 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
787 	if (r)
788 		crypt_iv_elephant_dtr(cc);
789 	return r;
790 }
791 
792 static void diffuser_disk_to_cpu(u32 *d, size_t n)
793 {
794 #ifndef __LITTLE_ENDIAN
795 	int i;
796 
797 	for (i = 0; i < n; i++)
798 		d[i] = le32_to_cpu((__le32)d[i]);
799 #endif
800 }
801 
802 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
803 {
804 #ifndef __LITTLE_ENDIAN
805 	int i;
806 
807 	for (i = 0; i < n; i++)
808 		d[i] = cpu_to_le32((u32)d[i]);
809 #endif
810 }
811 
812 static void diffuser_a_decrypt(u32 *d, size_t n)
813 {
814 	int i, i1, i2, i3;
815 
816 	for (i = 0; i < 5; i++) {
817 		i1 = 0;
818 		i2 = n - 2;
819 		i3 = n - 5;
820 
821 		while (i1 < (n - 1)) {
822 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
823 			i1++; i2++; i3++;
824 
825 			if (i3 >= n)
826 				i3 -= n;
827 
828 			d[i1] += d[i2] ^ d[i3];
829 			i1++; i2++; i3++;
830 
831 			if (i2 >= n)
832 				i2 -= n;
833 
834 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
835 			i1++; i2++; i3++;
836 
837 			d[i1] += d[i2] ^ d[i3];
838 			i1++; i2++; i3++;
839 		}
840 	}
841 }
842 
843 static void diffuser_a_encrypt(u32 *d, size_t n)
844 {
845 	int i, i1, i2, i3;
846 
847 	for (i = 0; i < 5; i++) {
848 		i1 = n - 1;
849 		i2 = n - 2 - 1;
850 		i3 = n - 5 - 1;
851 
852 		while (i1 > 0) {
853 			d[i1] -= d[i2] ^ d[i3];
854 			i1--; i2--; i3--;
855 
856 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
857 			i1--; i2--; i3--;
858 
859 			if (i2 < 0)
860 				i2 += n;
861 
862 			d[i1] -= d[i2] ^ d[i3];
863 			i1--; i2--; i3--;
864 
865 			if (i3 < 0)
866 				i3 += n;
867 
868 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
869 			i1--; i2--; i3--;
870 		}
871 	}
872 }
873 
874 static void diffuser_b_decrypt(u32 *d, size_t n)
875 {
876 	int i, i1, i2, i3;
877 
878 	for (i = 0; i < 3; i++) {
879 		i1 = 0;
880 		i2 = 2;
881 		i3 = 5;
882 
883 		while (i1 < (n - 1)) {
884 			d[i1] += d[i2] ^ d[i3];
885 			i1++; i2++; i3++;
886 
887 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
888 			i1++; i2++; i3++;
889 
890 			if (i2 >= n)
891 				i2 -= n;
892 
893 			d[i1] += d[i2] ^ d[i3];
894 			i1++; i2++; i3++;
895 
896 			if (i3 >= n)
897 				i3 -= n;
898 
899 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
900 			i1++; i2++; i3++;
901 		}
902 	}
903 }
904 
905 static void diffuser_b_encrypt(u32 *d, size_t n)
906 {
907 	int i, i1, i2, i3;
908 
909 	for (i = 0; i < 3; i++) {
910 		i1 = n - 1;
911 		i2 = 2 - 1;
912 		i3 = 5 - 1;
913 
914 		while (i1 > 0) {
915 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
916 			i1--; i2--; i3--;
917 
918 			if (i3 < 0)
919 				i3 += n;
920 
921 			d[i1] -= d[i2] ^ d[i3];
922 			i1--; i2--; i3--;
923 
924 			if (i2 < 0)
925 				i2 += n;
926 
927 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
928 			i1--; i2--; i3--;
929 
930 			d[i1] -= d[i2] ^ d[i3];
931 			i1--; i2--; i3--;
932 		}
933 	}
934 }
935 
936 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
937 {
938 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
939 	u8 *es, *ks, *data, *data2, *data_offset;
940 	struct skcipher_request *req;
941 	struct scatterlist *sg, *sg2, src, dst;
942 	DECLARE_CRYPTO_WAIT(wait);
943 	int i, r;
944 
945 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
946 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
947 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
948 
949 	if (!req || !es || !ks) {
950 		r = -ENOMEM;
951 		goto out;
952 	}
953 
954 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
955 
956 	/* E(Ks, e(s)) */
957 	sg_init_one(&src, es, 16);
958 	sg_init_one(&dst, ks, 16);
959 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
960 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
961 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
962 	if (r)
963 		goto out;
964 
965 	/* E(Ks, e'(s)) */
966 	es[15] = 0x80;
967 	sg_init_one(&dst, &ks[16], 16);
968 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
969 	if (r)
970 		goto out;
971 
972 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
973 	data = kmap_atomic(sg_page(sg));
974 	data_offset = data + sg->offset;
975 
976 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
977 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
978 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
979 		data2 = kmap_atomic(sg_page(sg2));
980 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
981 		kunmap_atomic(data2);
982 	}
983 
984 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
985 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
986 		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
987 		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
988 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
989 	}
990 
991 	for (i = 0; i < (cc->sector_size / 32); i++)
992 		crypto_xor(data_offset + i * 32, ks, 32);
993 
994 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
995 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
996 		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
997 		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
998 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
999 	}
1000 
1001 	kunmap_atomic(data);
1002 out:
1003 	kfree_sensitive(ks);
1004 	kfree_sensitive(es);
1005 	skcipher_request_free(req);
1006 	return r;
1007 }
1008 
1009 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1010 			    struct dm_crypt_request *dmreq)
1011 {
1012 	int r;
1013 
1014 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1015 		r = crypt_iv_elephant(cc, dmreq);
1016 		if (r)
1017 			return r;
1018 	}
1019 
1020 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1021 }
1022 
1023 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1024 				  struct dm_crypt_request *dmreq)
1025 {
1026 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1027 		return crypt_iv_elephant(cc, dmreq);
1028 
1029 	return 0;
1030 }
1031 
1032 static int crypt_iv_elephant_init(struct crypt_config *cc)
1033 {
1034 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1035 	int key_offset = cc->key_size - cc->key_extra_size;
1036 
1037 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1038 }
1039 
1040 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1041 {
1042 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1043 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1044 
1045 	memset(key, 0, cc->key_extra_size);
1046 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1047 }
1048 
1049 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1050 	.generator = crypt_iv_plain_gen
1051 };
1052 
1053 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1054 	.generator = crypt_iv_plain64_gen
1055 };
1056 
1057 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1058 	.generator = crypt_iv_plain64be_gen
1059 };
1060 
1061 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1062 	.generator = crypt_iv_essiv_gen
1063 };
1064 
1065 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1066 	.ctr	   = crypt_iv_benbi_ctr,
1067 	.dtr	   = crypt_iv_benbi_dtr,
1068 	.generator = crypt_iv_benbi_gen
1069 };
1070 
1071 static const struct crypt_iv_operations crypt_iv_null_ops = {
1072 	.generator = crypt_iv_null_gen
1073 };
1074 
1075 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1076 	.ctr	   = crypt_iv_lmk_ctr,
1077 	.dtr	   = crypt_iv_lmk_dtr,
1078 	.init	   = crypt_iv_lmk_init,
1079 	.wipe	   = crypt_iv_lmk_wipe,
1080 	.generator = crypt_iv_lmk_gen,
1081 	.post	   = crypt_iv_lmk_post
1082 };
1083 
1084 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1085 	.ctr	   = crypt_iv_tcw_ctr,
1086 	.dtr	   = crypt_iv_tcw_dtr,
1087 	.init	   = crypt_iv_tcw_init,
1088 	.wipe	   = crypt_iv_tcw_wipe,
1089 	.generator = crypt_iv_tcw_gen,
1090 	.post	   = crypt_iv_tcw_post
1091 };
1092 
1093 static const struct crypt_iv_operations crypt_iv_random_ops = {
1094 	.generator = crypt_iv_random_gen
1095 };
1096 
1097 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1098 	.ctr	   = crypt_iv_eboiv_ctr,
1099 	.generator = crypt_iv_eboiv_gen
1100 };
1101 
1102 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1103 	.ctr	   = crypt_iv_elephant_ctr,
1104 	.dtr	   = crypt_iv_elephant_dtr,
1105 	.init	   = crypt_iv_elephant_init,
1106 	.wipe	   = crypt_iv_elephant_wipe,
1107 	.generator = crypt_iv_elephant_gen,
1108 	.post	   = crypt_iv_elephant_post
1109 };
1110 
1111 /*
1112  * Integrity extensions
1113  */
1114 static bool crypt_integrity_aead(struct crypt_config *cc)
1115 {
1116 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1117 }
1118 
1119 static bool crypt_integrity_hmac(struct crypt_config *cc)
1120 {
1121 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1122 }
1123 
1124 /* Get sg containing data */
1125 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1126 					     struct scatterlist *sg)
1127 {
1128 	if (unlikely(crypt_integrity_aead(cc)))
1129 		return &sg[2];
1130 
1131 	return sg;
1132 }
1133 
1134 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1135 {
1136 	struct bio_integrity_payload *bip;
1137 	unsigned int tag_len;
1138 	int ret;
1139 
1140 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1141 		return 0;
1142 
1143 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1144 	if (IS_ERR(bip))
1145 		return PTR_ERR(bip);
1146 
1147 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1148 
1149 	bip->bip_iter.bi_size = tag_len;
1150 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1151 
1152 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1153 				     tag_len, offset_in_page(io->integrity_metadata));
1154 	if (unlikely(ret != tag_len))
1155 		return -ENOMEM;
1156 
1157 	return 0;
1158 }
1159 
1160 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1161 {
1162 #ifdef CONFIG_BLK_DEV_INTEGRITY
1163 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1164 	struct mapped_device *md = dm_table_get_md(ti->table);
1165 
1166 	/* From now we require underlying device with our integrity profile */
1167 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1168 		ti->error = "Integrity profile not supported.";
1169 		return -EINVAL;
1170 	}
1171 
1172 	if (bi->tag_size != cc->on_disk_tag_size ||
1173 	    bi->tuple_size != cc->on_disk_tag_size) {
1174 		ti->error = "Integrity profile tag size mismatch.";
1175 		return -EINVAL;
1176 	}
1177 	if (1 << bi->interval_exp != cc->sector_size) {
1178 		ti->error = "Integrity profile sector size mismatch.";
1179 		return -EINVAL;
1180 	}
1181 
1182 	if (crypt_integrity_aead(cc)) {
1183 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1184 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1185 		       cc->integrity_tag_size, cc->integrity_iv_size);
1186 
1187 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1188 			ti->error = "Integrity AEAD auth tag size is not supported.";
1189 			return -EINVAL;
1190 		}
1191 	} else if (cc->integrity_iv_size)
1192 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1193 		       cc->integrity_iv_size);
1194 
1195 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1196 		ti->error = "Not enough space for integrity tag in the profile.";
1197 		return -EINVAL;
1198 	}
1199 
1200 	return 0;
1201 #else
1202 	ti->error = "Integrity profile not supported.";
1203 	return -EINVAL;
1204 #endif
1205 }
1206 
1207 static void crypt_convert_init(struct crypt_config *cc,
1208 			       struct convert_context *ctx,
1209 			       struct bio *bio_out, struct bio *bio_in,
1210 			       sector_t sector)
1211 {
1212 	ctx->bio_in = bio_in;
1213 	ctx->bio_out = bio_out;
1214 	if (bio_in)
1215 		ctx->iter_in = bio_in->bi_iter;
1216 	if (bio_out)
1217 		ctx->iter_out = bio_out->bi_iter;
1218 	ctx->cc_sector = sector + cc->iv_offset;
1219 	init_completion(&ctx->restart);
1220 }
1221 
1222 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1223 					     void *req)
1224 {
1225 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1226 }
1227 
1228 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1229 {
1230 	return (void *)((char *)dmreq - cc->dmreq_start);
1231 }
1232 
1233 static u8 *iv_of_dmreq(struct crypt_config *cc,
1234 		       struct dm_crypt_request *dmreq)
1235 {
1236 	if (crypt_integrity_aead(cc))
1237 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1238 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1239 	else
1240 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1241 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1242 }
1243 
1244 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1245 		       struct dm_crypt_request *dmreq)
1246 {
1247 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1248 }
1249 
1250 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1251 		       struct dm_crypt_request *dmreq)
1252 {
1253 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1254 	return (__le64 *) ptr;
1255 }
1256 
1257 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1258 		       struct dm_crypt_request *dmreq)
1259 {
1260 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1261 		  cc->iv_size + sizeof(uint64_t);
1262 	return (unsigned int*)ptr;
1263 }
1264 
1265 static void *tag_from_dmreq(struct crypt_config *cc,
1266 				struct dm_crypt_request *dmreq)
1267 {
1268 	struct convert_context *ctx = dmreq->ctx;
1269 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1270 
1271 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1272 		cc->on_disk_tag_size];
1273 }
1274 
1275 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1276 			       struct dm_crypt_request *dmreq)
1277 {
1278 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1279 }
1280 
1281 static int crypt_convert_block_aead(struct crypt_config *cc,
1282 				     struct convert_context *ctx,
1283 				     struct aead_request *req,
1284 				     unsigned int tag_offset)
1285 {
1286 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1287 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1288 	struct dm_crypt_request *dmreq;
1289 	u8 *iv, *org_iv, *tag_iv, *tag;
1290 	__le64 *sector;
1291 	int r = 0;
1292 
1293 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1294 
1295 	/* Reject unexpected unaligned bio. */
1296 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1297 		return -EIO;
1298 
1299 	dmreq = dmreq_of_req(cc, req);
1300 	dmreq->iv_sector = ctx->cc_sector;
1301 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1302 		dmreq->iv_sector >>= cc->sector_shift;
1303 	dmreq->ctx = ctx;
1304 
1305 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1306 
1307 	sector = org_sector_of_dmreq(cc, dmreq);
1308 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1309 
1310 	iv = iv_of_dmreq(cc, dmreq);
1311 	org_iv = org_iv_of_dmreq(cc, dmreq);
1312 	tag = tag_from_dmreq(cc, dmreq);
1313 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1314 
1315 	/* AEAD request:
1316 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1317 	 *  | (authenticated) | (auth+encryption) |              |
1318 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1319 	 */
1320 	sg_init_table(dmreq->sg_in, 4);
1321 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1322 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1323 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1324 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1325 
1326 	sg_init_table(dmreq->sg_out, 4);
1327 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1328 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1329 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1330 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1331 
1332 	if (cc->iv_gen_ops) {
1333 		/* For READs use IV stored in integrity metadata */
1334 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1335 			memcpy(org_iv, tag_iv, cc->iv_size);
1336 		} else {
1337 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1338 			if (r < 0)
1339 				return r;
1340 			/* Store generated IV in integrity metadata */
1341 			if (cc->integrity_iv_size)
1342 				memcpy(tag_iv, org_iv, cc->iv_size);
1343 		}
1344 		/* Working copy of IV, to be modified in crypto API */
1345 		memcpy(iv, org_iv, cc->iv_size);
1346 	}
1347 
1348 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1349 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1350 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1351 				       cc->sector_size, iv);
1352 		r = crypto_aead_encrypt(req);
1353 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1354 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1355 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1356 	} else {
1357 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1358 				       cc->sector_size + cc->integrity_tag_size, iv);
1359 		r = crypto_aead_decrypt(req);
1360 	}
1361 
1362 	if (r == -EBADMSG) {
1363 		char b[BDEVNAME_SIZE];
1364 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1365 			    (unsigned long long)le64_to_cpu(*sector));
1366 	}
1367 
1368 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1369 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1370 
1371 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1372 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1373 
1374 	return r;
1375 }
1376 
1377 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1378 					struct convert_context *ctx,
1379 					struct skcipher_request *req,
1380 					unsigned int tag_offset)
1381 {
1382 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1383 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1384 	struct scatterlist *sg_in, *sg_out;
1385 	struct dm_crypt_request *dmreq;
1386 	u8 *iv, *org_iv, *tag_iv;
1387 	__le64 *sector;
1388 	int r = 0;
1389 
1390 	/* Reject unexpected unaligned bio. */
1391 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1392 		return -EIO;
1393 
1394 	dmreq = dmreq_of_req(cc, req);
1395 	dmreq->iv_sector = ctx->cc_sector;
1396 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1397 		dmreq->iv_sector >>= cc->sector_shift;
1398 	dmreq->ctx = ctx;
1399 
1400 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1401 
1402 	iv = iv_of_dmreq(cc, dmreq);
1403 	org_iv = org_iv_of_dmreq(cc, dmreq);
1404 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1405 
1406 	sector = org_sector_of_dmreq(cc, dmreq);
1407 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1408 
1409 	/* For skcipher we use only the first sg item */
1410 	sg_in  = &dmreq->sg_in[0];
1411 	sg_out = &dmreq->sg_out[0];
1412 
1413 	sg_init_table(sg_in, 1);
1414 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1415 
1416 	sg_init_table(sg_out, 1);
1417 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1418 
1419 	if (cc->iv_gen_ops) {
1420 		/* For READs use IV stored in integrity metadata */
1421 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1422 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1423 		} else {
1424 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1425 			if (r < 0)
1426 				return r;
1427 			/* Data can be already preprocessed in generator */
1428 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1429 				sg_in = sg_out;
1430 			/* Store generated IV in integrity metadata */
1431 			if (cc->integrity_iv_size)
1432 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1433 		}
1434 		/* Working copy of IV, to be modified in crypto API */
1435 		memcpy(iv, org_iv, cc->iv_size);
1436 	}
1437 
1438 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1439 
1440 	if (bio_data_dir(ctx->bio_in) == WRITE)
1441 		r = crypto_skcipher_encrypt(req);
1442 	else
1443 		r = crypto_skcipher_decrypt(req);
1444 
1445 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1446 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1447 
1448 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1449 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1450 
1451 	return r;
1452 }
1453 
1454 static void kcryptd_async_done(struct crypto_async_request *async_req,
1455 			       int error);
1456 
1457 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1458 				     struct convert_context *ctx)
1459 {
1460 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1461 
1462 	if (!ctx->r.req) {
1463 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1464 		if (!ctx->r.req)
1465 			return -ENOMEM;
1466 	}
1467 
1468 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1469 
1470 	/*
1471 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1472 	 * requests if driver request queue is full.
1473 	 */
1474 	skcipher_request_set_callback(ctx->r.req,
1475 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1476 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1477 
1478 	return 0;
1479 }
1480 
1481 static int crypt_alloc_req_aead(struct crypt_config *cc,
1482 				 struct convert_context *ctx)
1483 {
1484 	if (!ctx->r.req_aead) {
1485 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1486 		if (!ctx->r.req_aead)
1487 			return -ENOMEM;
1488 	}
1489 
1490 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1491 
1492 	/*
1493 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1494 	 * requests if driver request queue is full.
1495 	 */
1496 	aead_request_set_callback(ctx->r.req_aead,
1497 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1498 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1499 
1500 	return 0;
1501 }
1502 
1503 static int crypt_alloc_req(struct crypt_config *cc,
1504 			    struct convert_context *ctx)
1505 {
1506 	if (crypt_integrity_aead(cc))
1507 		return crypt_alloc_req_aead(cc, ctx);
1508 	else
1509 		return crypt_alloc_req_skcipher(cc, ctx);
1510 }
1511 
1512 static void crypt_free_req_skcipher(struct crypt_config *cc,
1513 				    struct skcipher_request *req, struct bio *base_bio)
1514 {
1515 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1516 
1517 	if ((struct skcipher_request *)(io + 1) != req)
1518 		mempool_free(req, &cc->req_pool);
1519 }
1520 
1521 static void crypt_free_req_aead(struct crypt_config *cc,
1522 				struct aead_request *req, struct bio *base_bio)
1523 {
1524 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1525 
1526 	if ((struct aead_request *)(io + 1) != req)
1527 		mempool_free(req, &cc->req_pool);
1528 }
1529 
1530 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1531 {
1532 	if (crypt_integrity_aead(cc))
1533 		crypt_free_req_aead(cc, req, base_bio);
1534 	else
1535 		crypt_free_req_skcipher(cc, req, base_bio);
1536 }
1537 
1538 /*
1539  * Encrypt / decrypt data from one bio to another one (can be the same one)
1540  */
1541 static blk_status_t crypt_convert(struct crypt_config *cc,
1542 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1543 {
1544 	unsigned int tag_offset = 0;
1545 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1546 	int r;
1547 
1548 	/*
1549 	 * if reset_pending is set we are dealing with the bio for the first time,
1550 	 * else we're continuing to work on the previous bio, so don't mess with
1551 	 * the cc_pending counter
1552 	 */
1553 	if (reset_pending)
1554 		atomic_set(&ctx->cc_pending, 1);
1555 
1556 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1557 
1558 		r = crypt_alloc_req(cc, ctx);
1559 		if (r) {
1560 			complete(&ctx->restart);
1561 			return BLK_STS_DEV_RESOURCE;
1562 		}
1563 
1564 		atomic_inc(&ctx->cc_pending);
1565 
1566 		if (crypt_integrity_aead(cc))
1567 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1568 		else
1569 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1570 
1571 		switch (r) {
1572 		/*
1573 		 * The request was queued by a crypto driver
1574 		 * but the driver request queue is full, let's wait.
1575 		 */
1576 		case -EBUSY:
1577 			if (in_interrupt()) {
1578 				if (try_wait_for_completion(&ctx->restart)) {
1579 					/*
1580 					 * we don't have to block to wait for completion,
1581 					 * so proceed
1582 					 */
1583 				} else {
1584 					/*
1585 					 * we can't wait for completion without blocking
1586 					 * exit and continue processing in a workqueue
1587 					 */
1588 					ctx->r.req = NULL;
1589 					ctx->cc_sector += sector_step;
1590 					tag_offset++;
1591 					return BLK_STS_DEV_RESOURCE;
1592 				}
1593 			} else {
1594 				wait_for_completion(&ctx->restart);
1595 			}
1596 			reinit_completion(&ctx->restart);
1597 			fallthrough;
1598 		/*
1599 		 * The request is queued and processed asynchronously,
1600 		 * completion function kcryptd_async_done() will be called.
1601 		 */
1602 		case -EINPROGRESS:
1603 			ctx->r.req = NULL;
1604 			ctx->cc_sector += sector_step;
1605 			tag_offset++;
1606 			continue;
1607 		/*
1608 		 * The request was already processed (synchronously).
1609 		 */
1610 		case 0:
1611 			atomic_dec(&ctx->cc_pending);
1612 			ctx->cc_sector += sector_step;
1613 			tag_offset++;
1614 			if (!atomic)
1615 				cond_resched();
1616 			continue;
1617 		/*
1618 		 * There was a data integrity error.
1619 		 */
1620 		case -EBADMSG:
1621 			atomic_dec(&ctx->cc_pending);
1622 			return BLK_STS_PROTECTION;
1623 		/*
1624 		 * There was an error while processing the request.
1625 		 */
1626 		default:
1627 			atomic_dec(&ctx->cc_pending);
1628 			return BLK_STS_IOERR;
1629 		}
1630 	}
1631 
1632 	return 0;
1633 }
1634 
1635 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1636 
1637 /*
1638  * Generate a new unfragmented bio with the given size
1639  * This should never violate the device limitations (but only because
1640  * max_segment_size is being constrained to PAGE_SIZE).
1641  *
1642  * This function may be called concurrently. If we allocate from the mempool
1643  * concurrently, there is a possibility of deadlock. For example, if we have
1644  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1645  * the mempool concurrently, it may deadlock in a situation where both processes
1646  * have allocated 128 pages and the mempool is exhausted.
1647  *
1648  * In order to avoid this scenario we allocate the pages under a mutex.
1649  *
1650  * In order to not degrade performance with excessive locking, we try
1651  * non-blocking allocations without a mutex first but on failure we fallback
1652  * to blocking allocations with a mutex.
1653  */
1654 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1655 {
1656 	struct crypt_config *cc = io->cc;
1657 	struct bio *clone;
1658 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1659 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1660 	unsigned i, len, remaining_size;
1661 	struct page *page;
1662 
1663 retry:
1664 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1665 		mutex_lock(&cc->bio_alloc_lock);
1666 
1667 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1668 	if (!clone)
1669 		goto out;
1670 
1671 	clone_init(io, clone);
1672 
1673 	remaining_size = size;
1674 
1675 	for (i = 0; i < nr_iovecs; i++) {
1676 		page = mempool_alloc(&cc->page_pool, gfp_mask);
1677 		if (!page) {
1678 			crypt_free_buffer_pages(cc, clone);
1679 			bio_put(clone);
1680 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1681 			goto retry;
1682 		}
1683 
1684 		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1685 
1686 		bio_add_page(clone, page, len, 0);
1687 
1688 		remaining_size -= len;
1689 	}
1690 
1691 	/* Allocate space for integrity tags */
1692 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1693 		crypt_free_buffer_pages(cc, clone);
1694 		bio_put(clone);
1695 		clone = NULL;
1696 	}
1697 out:
1698 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1699 		mutex_unlock(&cc->bio_alloc_lock);
1700 
1701 	return clone;
1702 }
1703 
1704 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1705 {
1706 	struct bio_vec *bv;
1707 	struct bvec_iter_all iter_all;
1708 
1709 	bio_for_each_segment_all(bv, clone, iter_all) {
1710 		BUG_ON(!bv->bv_page);
1711 		mempool_free(bv->bv_page, &cc->page_pool);
1712 	}
1713 }
1714 
1715 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1716 			  struct bio *bio, sector_t sector)
1717 {
1718 	io->cc = cc;
1719 	io->base_bio = bio;
1720 	io->sector = sector;
1721 	io->error = 0;
1722 	io->ctx.r.req = NULL;
1723 	io->integrity_metadata = NULL;
1724 	io->integrity_metadata_from_pool = false;
1725 	atomic_set(&io->io_pending, 0);
1726 }
1727 
1728 static void crypt_inc_pending(struct dm_crypt_io *io)
1729 {
1730 	atomic_inc(&io->io_pending);
1731 }
1732 
1733 static void kcryptd_io_bio_endio(struct work_struct *work)
1734 {
1735 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1736 	bio_endio(io->base_bio);
1737 }
1738 
1739 /*
1740  * One of the bios was finished. Check for completion of
1741  * the whole request and correctly clean up the buffer.
1742  */
1743 static void crypt_dec_pending(struct dm_crypt_io *io)
1744 {
1745 	struct crypt_config *cc = io->cc;
1746 	struct bio *base_bio = io->base_bio;
1747 	blk_status_t error = io->error;
1748 
1749 	if (!atomic_dec_and_test(&io->io_pending))
1750 		return;
1751 
1752 	if (io->ctx.r.req)
1753 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1754 
1755 	if (unlikely(io->integrity_metadata_from_pool))
1756 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1757 	else
1758 		kfree(io->integrity_metadata);
1759 
1760 	base_bio->bi_status = error;
1761 
1762 	/*
1763 	 * If we are running this function from our tasklet,
1764 	 * we can't call bio_endio() here, because it will call
1765 	 * clone_endio() from dm.c, which in turn will
1766 	 * free the current struct dm_crypt_io structure with
1767 	 * our tasklet. In this case we need to delay bio_endio()
1768 	 * execution to after the tasklet is done and dequeued.
1769 	 */
1770 	if (tasklet_trylock(&io->tasklet)) {
1771 		tasklet_unlock(&io->tasklet);
1772 		bio_endio(base_bio);
1773 		return;
1774 	}
1775 
1776 	INIT_WORK(&io->work, kcryptd_io_bio_endio);
1777 	queue_work(cc->io_queue, &io->work);
1778 }
1779 
1780 /*
1781  * kcryptd/kcryptd_io:
1782  *
1783  * Needed because it would be very unwise to do decryption in an
1784  * interrupt context.
1785  *
1786  * kcryptd performs the actual encryption or decryption.
1787  *
1788  * kcryptd_io performs the IO submission.
1789  *
1790  * They must be separated as otherwise the final stages could be
1791  * starved by new requests which can block in the first stages due
1792  * to memory allocation.
1793  *
1794  * The work is done per CPU global for all dm-crypt instances.
1795  * They should not depend on each other and do not block.
1796  */
1797 static void crypt_endio(struct bio *clone)
1798 {
1799 	struct dm_crypt_io *io = clone->bi_private;
1800 	struct crypt_config *cc = io->cc;
1801 	unsigned rw = bio_data_dir(clone);
1802 	blk_status_t error;
1803 
1804 	/*
1805 	 * free the processed pages
1806 	 */
1807 	if (rw == WRITE)
1808 		crypt_free_buffer_pages(cc, clone);
1809 
1810 	error = clone->bi_status;
1811 	bio_put(clone);
1812 
1813 	if (rw == READ && !error) {
1814 		kcryptd_queue_crypt(io);
1815 		return;
1816 	}
1817 
1818 	if (unlikely(error))
1819 		io->error = error;
1820 
1821 	crypt_dec_pending(io);
1822 }
1823 
1824 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1825 {
1826 	struct crypt_config *cc = io->cc;
1827 
1828 	clone->bi_private = io;
1829 	clone->bi_end_io  = crypt_endio;
1830 	bio_set_dev(clone, cc->dev->bdev);
1831 	clone->bi_opf	  = io->base_bio->bi_opf;
1832 }
1833 
1834 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1835 {
1836 	struct crypt_config *cc = io->cc;
1837 	struct bio *clone;
1838 
1839 	/*
1840 	 * We need the original biovec array in order to decrypt
1841 	 * the whole bio data *afterwards* -- thanks to immutable
1842 	 * biovecs we don't need to worry about the block layer
1843 	 * modifying the biovec array; so leverage bio_clone_fast().
1844 	 */
1845 	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1846 	if (!clone)
1847 		return 1;
1848 
1849 	crypt_inc_pending(io);
1850 
1851 	clone_init(io, clone);
1852 	clone->bi_iter.bi_sector = cc->start + io->sector;
1853 
1854 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1855 		crypt_dec_pending(io);
1856 		bio_put(clone);
1857 		return 1;
1858 	}
1859 
1860 	submit_bio_noacct(clone);
1861 	return 0;
1862 }
1863 
1864 static void kcryptd_io_read_work(struct work_struct *work)
1865 {
1866 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1867 
1868 	crypt_inc_pending(io);
1869 	if (kcryptd_io_read(io, GFP_NOIO))
1870 		io->error = BLK_STS_RESOURCE;
1871 	crypt_dec_pending(io);
1872 }
1873 
1874 static void kcryptd_queue_read(struct dm_crypt_io *io)
1875 {
1876 	struct crypt_config *cc = io->cc;
1877 
1878 	INIT_WORK(&io->work, kcryptd_io_read_work);
1879 	queue_work(cc->io_queue, &io->work);
1880 }
1881 
1882 static void kcryptd_io_write(struct dm_crypt_io *io)
1883 {
1884 	struct bio *clone = io->ctx.bio_out;
1885 
1886 	submit_bio_noacct(clone);
1887 }
1888 
1889 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1890 
1891 static int dmcrypt_write(void *data)
1892 {
1893 	struct crypt_config *cc = data;
1894 	struct dm_crypt_io *io;
1895 
1896 	while (1) {
1897 		struct rb_root write_tree;
1898 		struct blk_plug plug;
1899 
1900 		spin_lock_irq(&cc->write_thread_lock);
1901 continue_locked:
1902 
1903 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1904 			goto pop_from_list;
1905 
1906 		set_current_state(TASK_INTERRUPTIBLE);
1907 
1908 		spin_unlock_irq(&cc->write_thread_lock);
1909 
1910 		if (unlikely(kthread_should_stop())) {
1911 			set_current_state(TASK_RUNNING);
1912 			break;
1913 		}
1914 
1915 		schedule();
1916 
1917 		set_current_state(TASK_RUNNING);
1918 		spin_lock_irq(&cc->write_thread_lock);
1919 		goto continue_locked;
1920 
1921 pop_from_list:
1922 		write_tree = cc->write_tree;
1923 		cc->write_tree = RB_ROOT;
1924 		spin_unlock_irq(&cc->write_thread_lock);
1925 
1926 		BUG_ON(rb_parent(write_tree.rb_node));
1927 
1928 		/*
1929 		 * Note: we cannot walk the tree here with rb_next because
1930 		 * the structures may be freed when kcryptd_io_write is called.
1931 		 */
1932 		blk_start_plug(&plug);
1933 		do {
1934 			io = crypt_io_from_node(rb_first(&write_tree));
1935 			rb_erase(&io->rb_node, &write_tree);
1936 			kcryptd_io_write(io);
1937 		} while (!RB_EMPTY_ROOT(&write_tree));
1938 		blk_finish_plug(&plug);
1939 	}
1940 	return 0;
1941 }
1942 
1943 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1944 {
1945 	struct bio *clone = io->ctx.bio_out;
1946 	struct crypt_config *cc = io->cc;
1947 	unsigned long flags;
1948 	sector_t sector;
1949 	struct rb_node **rbp, *parent;
1950 
1951 	if (unlikely(io->error)) {
1952 		crypt_free_buffer_pages(cc, clone);
1953 		bio_put(clone);
1954 		crypt_dec_pending(io);
1955 		return;
1956 	}
1957 
1958 	/* crypt_convert should have filled the clone bio */
1959 	BUG_ON(io->ctx.iter_out.bi_size);
1960 
1961 	clone->bi_iter.bi_sector = cc->start + io->sector;
1962 
1963 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1964 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1965 		submit_bio_noacct(clone);
1966 		return;
1967 	}
1968 
1969 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1970 	if (RB_EMPTY_ROOT(&cc->write_tree))
1971 		wake_up_process(cc->write_thread);
1972 	rbp = &cc->write_tree.rb_node;
1973 	parent = NULL;
1974 	sector = io->sector;
1975 	while (*rbp) {
1976 		parent = *rbp;
1977 		if (sector < crypt_io_from_node(parent)->sector)
1978 			rbp = &(*rbp)->rb_left;
1979 		else
1980 			rbp = &(*rbp)->rb_right;
1981 	}
1982 	rb_link_node(&io->rb_node, parent, rbp);
1983 	rb_insert_color(&io->rb_node, &cc->write_tree);
1984 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1985 }
1986 
1987 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1988 				       struct convert_context *ctx)
1989 
1990 {
1991 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1992 		return false;
1993 
1994 	/*
1995 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1996 	 * constraints so they do not need to be issued inline by
1997 	 * kcryptd_crypt_write_convert().
1998 	 */
1999 	switch (bio_op(ctx->bio_in)) {
2000 	case REQ_OP_WRITE:
2001 	case REQ_OP_WRITE_SAME:
2002 	case REQ_OP_WRITE_ZEROES:
2003 		return true;
2004 	default:
2005 		return false;
2006 	}
2007 }
2008 
2009 static void kcryptd_crypt_write_continue(struct work_struct *work)
2010 {
2011 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2012 	struct crypt_config *cc = io->cc;
2013 	struct convert_context *ctx = &io->ctx;
2014 	int crypt_finished;
2015 	sector_t sector = io->sector;
2016 	blk_status_t r;
2017 
2018 	wait_for_completion(&ctx->restart);
2019 	reinit_completion(&ctx->restart);
2020 
2021 	r = crypt_convert(cc, &io->ctx, true, false);
2022 	if (r)
2023 		io->error = r;
2024 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2025 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2026 		/* Wait for completion signaled by kcryptd_async_done() */
2027 		wait_for_completion(&ctx->restart);
2028 		crypt_finished = 1;
2029 	}
2030 
2031 	/* Encryption was already finished, submit io now */
2032 	if (crypt_finished) {
2033 		kcryptd_crypt_write_io_submit(io, 0);
2034 		io->sector = sector;
2035 	}
2036 
2037 	crypt_dec_pending(io);
2038 }
2039 
2040 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2041 {
2042 	struct crypt_config *cc = io->cc;
2043 	struct convert_context *ctx = &io->ctx;
2044 	struct bio *clone;
2045 	int crypt_finished;
2046 	sector_t sector = io->sector;
2047 	blk_status_t r;
2048 
2049 	/*
2050 	 * Prevent io from disappearing until this function completes.
2051 	 */
2052 	crypt_inc_pending(io);
2053 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2054 
2055 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2056 	if (unlikely(!clone)) {
2057 		io->error = BLK_STS_IOERR;
2058 		goto dec;
2059 	}
2060 
2061 	io->ctx.bio_out = clone;
2062 	io->ctx.iter_out = clone->bi_iter;
2063 
2064 	sector += bio_sectors(clone);
2065 
2066 	crypt_inc_pending(io);
2067 	r = crypt_convert(cc, ctx,
2068 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2069 	/*
2070 	 * Crypto API backlogged the request, because its queue was full
2071 	 * and we're in softirq context, so continue from a workqueue
2072 	 * (TODO: is it actually possible to be in softirq in the write path?)
2073 	 */
2074 	if (r == BLK_STS_DEV_RESOURCE) {
2075 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2076 		queue_work(cc->crypt_queue, &io->work);
2077 		return;
2078 	}
2079 	if (r)
2080 		io->error = r;
2081 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2082 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2083 		/* Wait for completion signaled by kcryptd_async_done() */
2084 		wait_for_completion(&ctx->restart);
2085 		crypt_finished = 1;
2086 	}
2087 
2088 	/* Encryption was already finished, submit io now */
2089 	if (crypt_finished) {
2090 		kcryptd_crypt_write_io_submit(io, 0);
2091 		io->sector = sector;
2092 	}
2093 
2094 dec:
2095 	crypt_dec_pending(io);
2096 }
2097 
2098 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2099 {
2100 	crypt_dec_pending(io);
2101 }
2102 
2103 static void kcryptd_crypt_read_continue(struct work_struct *work)
2104 {
2105 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2106 	struct crypt_config *cc = io->cc;
2107 	blk_status_t r;
2108 
2109 	wait_for_completion(&io->ctx.restart);
2110 	reinit_completion(&io->ctx.restart);
2111 
2112 	r = crypt_convert(cc, &io->ctx, true, false);
2113 	if (r)
2114 		io->error = r;
2115 
2116 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2117 		kcryptd_crypt_read_done(io);
2118 
2119 	crypt_dec_pending(io);
2120 }
2121 
2122 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2123 {
2124 	struct crypt_config *cc = io->cc;
2125 	blk_status_t r;
2126 
2127 	crypt_inc_pending(io);
2128 
2129 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2130 			   io->sector);
2131 
2132 	r = crypt_convert(cc, &io->ctx,
2133 			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2134 	/*
2135 	 * Crypto API backlogged the request, because its queue was full
2136 	 * and we're in softirq context, so continue from a workqueue
2137 	 */
2138 	if (r == BLK_STS_DEV_RESOURCE) {
2139 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2140 		queue_work(cc->crypt_queue, &io->work);
2141 		return;
2142 	}
2143 	if (r)
2144 		io->error = r;
2145 
2146 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2147 		kcryptd_crypt_read_done(io);
2148 
2149 	crypt_dec_pending(io);
2150 }
2151 
2152 static void kcryptd_async_done(struct crypto_async_request *async_req,
2153 			       int error)
2154 {
2155 	struct dm_crypt_request *dmreq = async_req->data;
2156 	struct convert_context *ctx = dmreq->ctx;
2157 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2158 	struct crypt_config *cc = io->cc;
2159 
2160 	/*
2161 	 * A request from crypto driver backlog is going to be processed now,
2162 	 * finish the completion and continue in crypt_convert().
2163 	 * (Callback will be called for the second time for this request.)
2164 	 */
2165 	if (error == -EINPROGRESS) {
2166 		complete(&ctx->restart);
2167 		return;
2168 	}
2169 
2170 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2171 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2172 
2173 	if (error == -EBADMSG) {
2174 		char b[BDEVNAME_SIZE];
2175 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2176 			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
2177 		io->error = BLK_STS_PROTECTION;
2178 	} else if (error < 0)
2179 		io->error = BLK_STS_IOERR;
2180 
2181 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2182 
2183 	if (!atomic_dec_and_test(&ctx->cc_pending))
2184 		return;
2185 
2186 	/*
2187 	 * The request is fully completed: for inline writes, let
2188 	 * kcryptd_crypt_write_convert() do the IO submission.
2189 	 */
2190 	if (bio_data_dir(io->base_bio) == READ) {
2191 		kcryptd_crypt_read_done(io);
2192 		return;
2193 	}
2194 
2195 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2196 		complete(&ctx->restart);
2197 		return;
2198 	}
2199 
2200 	kcryptd_crypt_write_io_submit(io, 1);
2201 }
2202 
2203 static void kcryptd_crypt(struct work_struct *work)
2204 {
2205 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2206 
2207 	if (bio_data_dir(io->base_bio) == READ)
2208 		kcryptd_crypt_read_convert(io);
2209 	else
2210 		kcryptd_crypt_write_convert(io);
2211 }
2212 
2213 static void kcryptd_crypt_tasklet(unsigned long work)
2214 {
2215 	kcryptd_crypt((struct work_struct *)work);
2216 }
2217 
2218 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2219 {
2220 	struct crypt_config *cc = io->cc;
2221 
2222 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2223 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2224 		/*
2225 		 * in_irq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2226 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2227 		 * it is being executed with irqs disabled.
2228 		 */
2229 		if (in_irq() || irqs_disabled()) {
2230 			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2231 			tasklet_schedule(&io->tasklet);
2232 			return;
2233 		}
2234 
2235 		kcryptd_crypt(&io->work);
2236 		return;
2237 	}
2238 
2239 	INIT_WORK(&io->work, kcryptd_crypt);
2240 	queue_work(cc->crypt_queue, &io->work);
2241 }
2242 
2243 static void crypt_free_tfms_aead(struct crypt_config *cc)
2244 {
2245 	if (!cc->cipher_tfm.tfms_aead)
2246 		return;
2247 
2248 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2249 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2250 		cc->cipher_tfm.tfms_aead[0] = NULL;
2251 	}
2252 
2253 	kfree(cc->cipher_tfm.tfms_aead);
2254 	cc->cipher_tfm.tfms_aead = NULL;
2255 }
2256 
2257 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2258 {
2259 	unsigned i;
2260 
2261 	if (!cc->cipher_tfm.tfms)
2262 		return;
2263 
2264 	for (i = 0; i < cc->tfms_count; i++)
2265 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2266 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2267 			cc->cipher_tfm.tfms[i] = NULL;
2268 		}
2269 
2270 	kfree(cc->cipher_tfm.tfms);
2271 	cc->cipher_tfm.tfms = NULL;
2272 }
2273 
2274 static void crypt_free_tfms(struct crypt_config *cc)
2275 {
2276 	if (crypt_integrity_aead(cc))
2277 		crypt_free_tfms_aead(cc);
2278 	else
2279 		crypt_free_tfms_skcipher(cc);
2280 }
2281 
2282 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2283 {
2284 	unsigned i;
2285 	int err;
2286 
2287 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2288 				      sizeof(struct crypto_skcipher *),
2289 				      GFP_KERNEL);
2290 	if (!cc->cipher_tfm.tfms)
2291 		return -ENOMEM;
2292 
2293 	for (i = 0; i < cc->tfms_count; i++) {
2294 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2295 						CRYPTO_ALG_ALLOCATES_MEMORY);
2296 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2297 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2298 			crypt_free_tfms(cc);
2299 			return err;
2300 		}
2301 	}
2302 
2303 	/*
2304 	 * dm-crypt performance can vary greatly depending on which crypto
2305 	 * algorithm implementation is used.  Help people debug performance
2306 	 * problems by logging the ->cra_driver_name.
2307 	 */
2308 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2309 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2310 	return 0;
2311 }
2312 
2313 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2314 {
2315 	int err;
2316 
2317 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2318 	if (!cc->cipher_tfm.tfms)
2319 		return -ENOMEM;
2320 
2321 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2322 						CRYPTO_ALG_ALLOCATES_MEMORY);
2323 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2324 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2325 		crypt_free_tfms(cc);
2326 		return err;
2327 	}
2328 
2329 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2330 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2331 	return 0;
2332 }
2333 
2334 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2335 {
2336 	if (crypt_integrity_aead(cc))
2337 		return crypt_alloc_tfms_aead(cc, ciphermode);
2338 	else
2339 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2340 }
2341 
2342 static unsigned crypt_subkey_size(struct crypt_config *cc)
2343 {
2344 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2345 }
2346 
2347 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2348 {
2349 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2350 }
2351 
2352 /*
2353  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2354  * the key must be for some reason in special format.
2355  * This funcion converts cc->key to this special format.
2356  */
2357 static void crypt_copy_authenckey(char *p, const void *key,
2358 				  unsigned enckeylen, unsigned authkeylen)
2359 {
2360 	struct crypto_authenc_key_param *param;
2361 	struct rtattr *rta;
2362 
2363 	rta = (struct rtattr *)p;
2364 	param = RTA_DATA(rta);
2365 	param->enckeylen = cpu_to_be32(enckeylen);
2366 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2367 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2368 	p += RTA_SPACE(sizeof(*param));
2369 	memcpy(p, key + enckeylen, authkeylen);
2370 	p += authkeylen;
2371 	memcpy(p, key, enckeylen);
2372 }
2373 
2374 static int crypt_setkey(struct crypt_config *cc)
2375 {
2376 	unsigned subkey_size;
2377 	int err = 0, i, r;
2378 
2379 	/* Ignore extra keys (which are used for IV etc) */
2380 	subkey_size = crypt_subkey_size(cc);
2381 
2382 	if (crypt_integrity_hmac(cc)) {
2383 		if (subkey_size < cc->key_mac_size)
2384 			return -EINVAL;
2385 
2386 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2387 				      subkey_size - cc->key_mac_size,
2388 				      cc->key_mac_size);
2389 	}
2390 
2391 	for (i = 0; i < cc->tfms_count; i++) {
2392 		if (crypt_integrity_hmac(cc))
2393 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2394 				cc->authenc_key, crypt_authenckey_size(cc));
2395 		else if (crypt_integrity_aead(cc))
2396 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2397 					       cc->key + (i * subkey_size),
2398 					       subkey_size);
2399 		else
2400 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2401 						   cc->key + (i * subkey_size),
2402 						   subkey_size);
2403 		if (r)
2404 			err = r;
2405 	}
2406 
2407 	if (crypt_integrity_hmac(cc))
2408 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2409 
2410 	return err;
2411 }
2412 
2413 #ifdef CONFIG_KEYS
2414 
2415 static bool contains_whitespace(const char *str)
2416 {
2417 	while (*str)
2418 		if (isspace(*str++))
2419 			return true;
2420 	return false;
2421 }
2422 
2423 static int set_key_user(struct crypt_config *cc, struct key *key)
2424 {
2425 	const struct user_key_payload *ukp;
2426 
2427 	ukp = user_key_payload_locked(key);
2428 	if (!ukp)
2429 		return -EKEYREVOKED;
2430 
2431 	if (cc->key_size != ukp->datalen)
2432 		return -EINVAL;
2433 
2434 	memcpy(cc->key, ukp->data, cc->key_size);
2435 
2436 	return 0;
2437 }
2438 
2439 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
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 #endif /* CONFIG_ENCRYPTED_KEYS */
2456 
2457 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2458 {
2459 	char *new_key_string, *key_desc;
2460 	int ret;
2461 	struct key_type *type;
2462 	struct key *key;
2463 	int (*set_key)(struct crypt_config *cc, struct key *key);
2464 
2465 	/*
2466 	 * Reject key_string with whitespace. dm core currently lacks code for
2467 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2468 	 */
2469 	if (contains_whitespace(key_string)) {
2470 		DMERR("whitespace chars not allowed in key string");
2471 		return -EINVAL;
2472 	}
2473 
2474 	/* look for next ':' separating key_type from key_description */
2475 	key_desc = strpbrk(key_string, ":");
2476 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2477 		return -EINVAL;
2478 
2479 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2480 		type = &key_type_logon;
2481 		set_key = set_key_user;
2482 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2483 		type = &key_type_user;
2484 		set_key = set_key_user;
2485 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
2486 	} else if (!strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2487 		type = &key_type_encrypted;
2488 		set_key = set_key_encrypted;
2489 #endif
2490 	} else {
2491 		return -EINVAL;
2492 	}
2493 
2494 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2495 	if (!new_key_string)
2496 		return -ENOMEM;
2497 
2498 	key = request_key(type, key_desc + 1, NULL);
2499 	if (IS_ERR(key)) {
2500 		kfree_sensitive(new_key_string);
2501 		return PTR_ERR(key);
2502 	}
2503 
2504 	down_read(&key->sem);
2505 
2506 	ret = set_key(cc, key);
2507 	if (ret < 0) {
2508 		up_read(&key->sem);
2509 		key_put(key);
2510 		kfree_sensitive(new_key_string);
2511 		return ret;
2512 	}
2513 
2514 	up_read(&key->sem);
2515 	key_put(key);
2516 
2517 	/* clear the flag since following operations may invalidate previously valid key */
2518 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2519 
2520 	ret = crypt_setkey(cc);
2521 
2522 	if (!ret) {
2523 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2524 		kfree_sensitive(cc->key_string);
2525 		cc->key_string = new_key_string;
2526 	} else
2527 		kfree_sensitive(new_key_string);
2528 
2529 	return ret;
2530 }
2531 
2532 static int get_key_size(char **key_string)
2533 {
2534 	char *colon, dummy;
2535 	int ret;
2536 
2537 	if (*key_string[0] != ':')
2538 		return strlen(*key_string) >> 1;
2539 
2540 	/* look for next ':' in key string */
2541 	colon = strpbrk(*key_string + 1, ":");
2542 	if (!colon)
2543 		return -EINVAL;
2544 
2545 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2546 		return -EINVAL;
2547 
2548 	*key_string = colon;
2549 
2550 	/* remaining key string should be :<logon|user>:<key_desc> */
2551 
2552 	return ret;
2553 }
2554 
2555 #else
2556 
2557 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2558 {
2559 	return -EINVAL;
2560 }
2561 
2562 static int get_key_size(char **key_string)
2563 {
2564 	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2565 }
2566 
2567 #endif /* CONFIG_KEYS */
2568 
2569 static int crypt_set_key(struct crypt_config *cc, char *key)
2570 {
2571 	int r = -EINVAL;
2572 	int key_string_len = strlen(key);
2573 
2574 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2575 	if (!cc->key_size && strcmp(key, "-"))
2576 		goto out;
2577 
2578 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2579 	if (key[0] == ':') {
2580 		r = crypt_set_keyring_key(cc, key + 1);
2581 		goto out;
2582 	}
2583 
2584 	/* clear the flag since following operations may invalidate previously valid key */
2585 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2586 
2587 	/* wipe references to any kernel keyring key */
2588 	kfree_sensitive(cc->key_string);
2589 	cc->key_string = NULL;
2590 
2591 	/* Decode key from its hex representation. */
2592 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2593 		goto out;
2594 
2595 	r = crypt_setkey(cc);
2596 	if (!r)
2597 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2598 
2599 out:
2600 	/* Hex key string not needed after here, so wipe it. */
2601 	memset(key, '0', key_string_len);
2602 
2603 	return r;
2604 }
2605 
2606 static int crypt_wipe_key(struct crypt_config *cc)
2607 {
2608 	int r;
2609 
2610 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2611 	get_random_bytes(&cc->key, cc->key_size);
2612 
2613 	/* Wipe IV private keys */
2614 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2615 		r = cc->iv_gen_ops->wipe(cc);
2616 		if (r)
2617 			return r;
2618 	}
2619 
2620 	kfree_sensitive(cc->key_string);
2621 	cc->key_string = NULL;
2622 	r = crypt_setkey(cc);
2623 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2624 
2625 	return r;
2626 }
2627 
2628 static void crypt_calculate_pages_per_client(void)
2629 {
2630 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2631 
2632 	if (!dm_crypt_clients_n)
2633 		return;
2634 
2635 	pages /= dm_crypt_clients_n;
2636 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2637 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2638 	dm_crypt_pages_per_client = pages;
2639 }
2640 
2641 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2642 {
2643 	struct crypt_config *cc = pool_data;
2644 	struct page *page;
2645 
2646 	if (unlikely(percpu_counter_compare(&cc->n_allocated_pages, dm_crypt_pages_per_client) >= 0) &&
2647 	    likely(gfp_mask & __GFP_NORETRY))
2648 		return NULL;
2649 
2650 	page = alloc_page(gfp_mask);
2651 	if (likely(page != NULL))
2652 		percpu_counter_add(&cc->n_allocated_pages, 1);
2653 
2654 	return page;
2655 }
2656 
2657 static void crypt_page_free(void *page, void *pool_data)
2658 {
2659 	struct crypt_config *cc = pool_data;
2660 
2661 	__free_page(page);
2662 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2663 }
2664 
2665 static void crypt_dtr(struct dm_target *ti)
2666 {
2667 	struct crypt_config *cc = ti->private;
2668 
2669 	ti->private = NULL;
2670 
2671 	if (!cc)
2672 		return;
2673 
2674 	if (cc->write_thread)
2675 		kthread_stop(cc->write_thread);
2676 
2677 	if (cc->io_queue)
2678 		destroy_workqueue(cc->io_queue);
2679 	if (cc->crypt_queue)
2680 		destroy_workqueue(cc->crypt_queue);
2681 
2682 	crypt_free_tfms(cc);
2683 
2684 	bioset_exit(&cc->bs);
2685 
2686 	mempool_exit(&cc->page_pool);
2687 	mempool_exit(&cc->req_pool);
2688 	mempool_exit(&cc->tag_pool);
2689 
2690 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2691 	percpu_counter_destroy(&cc->n_allocated_pages);
2692 
2693 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2694 		cc->iv_gen_ops->dtr(cc);
2695 
2696 	if (cc->dev)
2697 		dm_put_device(ti, cc->dev);
2698 
2699 	kfree_sensitive(cc->cipher_string);
2700 	kfree_sensitive(cc->key_string);
2701 	kfree_sensitive(cc->cipher_auth);
2702 	kfree_sensitive(cc->authenc_key);
2703 
2704 	mutex_destroy(&cc->bio_alloc_lock);
2705 
2706 	/* Must zero key material before freeing */
2707 	kfree_sensitive(cc);
2708 
2709 	spin_lock(&dm_crypt_clients_lock);
2710 	WARN_ON(!dm_crypt_clients_n);
2711 	dm_crypt_clients_n--;
2712 	crypt_calculate_pages_per_client();
2713 	spin_unlock(&dm_crypt_clients_lock);
2714 }
2715 
2716 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2717 {
2718 	struct crypt_config *cc = ti->private;
2719 
2720 	if (crypt_integrity_aead(cc))
2721 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2722 	else
2723 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2724 
2725 	if (cc->iv_size)
2726 		/* at least a 64 bit sector number should fit in our buffer */
2727 		cc->iv_size = max(cc->iv_size,
2728 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2729 	else if (ivmode) {
2730 		DMWARN("Selected cipher does not support IVs");
2731 		ivmode = NULL;
2732 	}
2733 
2734 	/* Choose ivmode, see comments at iv code. */
2735 	if (ivmode == NULL)
2736 		cc->iv_gen_ops = NULL;
2737 	else if (strcmp(ivmode, "plain") == 0)
2738 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2739 	else if (strcmp(ivmode, "plain64") == 0)
2740 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2741 	else if (strcmp(ivmode, "plain64be") == 0)
2742 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2743 	else if (strcmp(ivmode, "essiv") == 0)
2744 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2745 	else if (strcmp(ivmode, "benbi") == 0)
2746 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2747 	else if (strcmp(ivmode, "null") == 0)
2748 		cc->iv_gen_ops = &crypt_iv_null_ops;
2749 	else if (strcmp(ivmode, "eboiv") == 0)
2750 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2751 	else if (strcmp(ivmode, "elephant") == 0) {
2752 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2753 		cc->key_parts = 2;
2754 		cc->key_extra_size = cc->key_size / 2;
2755 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2756 			return -EINVAL;
2757 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2758 	} else if (strcmp(ivmode, "lmk") == 0) {
2759 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2760 		/*
2761 		 * Version 2 and 3 is recognised according
2762 		 * to length of provided multi-key string.
2763 		 * If present (version 3), last key is used as IV seed.
2764 		 * All keys (including IV seed) are always the same size.
2765 		 */
2766 		if (cc->key_size % cc->key_parts) {
2767 			cc->key_parts++;
2768 			cc->key_extra_size = cc->key_size / cc->key_parts;
2769 		}
2770 	} else if (strcmp(ivmode, "tcw") == 0) {
2771 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2772 		cc->key_parts += 2; /* IV + whitening */
2773 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2774 	} else if (strcmp(ivmode, "random") == 0) {
2775 		cc->iv_gen_ops = &crypt_iv_random_ops;
2776 		/* Need storage space in integrity fields. */
2777 		cc->integrity_iv_size = cc->iv_size;
2778 	} else {
2779 		ti->error = "Invalid IV mode";
2780 		return -EINVAL;
2781 	}
2782 
2783 	return 0;
2784 }
2785 
2786 /*
2787  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2788  * The HMAC is needed to calculate tag size (HMAC digest size).
2789  * This should be probably done by crypto-api calls (once available...)
2790  */
2791 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2792 {
2793 	char *start, *end, *mac_alg = NULL;
2794 	struct crypto_ahash *mac;
2795 
2796 	if (!strstarts(cipher_api, "authenc("))
2797 		return 0;
2798 
2799 	start = strchr(cipher_api, '(');
2800 	end = strchr(cipher_api, ',');
2801 	if (!start || !end || ++start > end)
2802 		return -EINVAL;
2803 
2804 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2805 	if (!mac_alg)
2806 		return -ENOMEM;
2807 	strncpy(mac_alg, start, end - start);
2808 
2809 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2810 	kfree(mac_alg);
2811 
2812 	if (IS_ERR(mac))
2813 		return PTR_ERR(mac);
2814 
2815 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2816 	crypto_free_ahash(mac);
2817 
2818 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2819 	if (!cc->authenc_key)
2820 		return -ENOMEM;
2821 
2822 	return 0;
2823 }
2824 
2825 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2826 				char **ivmode, char **ivopts)
2827 {
2828 	struct crypt_config *cc = ti->private;
2829 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2830 	int ret = -EINVAL;
2831 
2832 	cc->tfms_count = 1;
2833 
2834 	/*
2835 	 * New format (capi: prefix)
2836 	 * capi:cipher_api_spec-iv:ivopts
2837 	 */
2838 	tmp = &cipher_in[strlen("capi:")];
2839 
2840 	/* Separate IV options if present, it can contain another '-' in hash name */
2841 	*ivopts = strrchr(tmp, ':');
2842 	if (*ivopts) {
2843 		**ivopts = '\0';
2844 		(*ivopts)++;
2845 	}
2846 	/* Parse IV mode */
2847 	*ivmode = strrchr(tmp, '-');
2848 	if (*ivmode) {
2849 		**ivmode = '\0';
2850 		(*ivmode)++;
2851 	}
2852 	/* The rest is crypto API spec */
2853 	cipher_api = tmp;
2854 
2855 	/* Alloc AEAD, can be used only in new format. */
2856 	if (crypt_integrity_aead(cc)) {
2857 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2858 		if (ret < 0) {
2859 			ti->error = "Invalid AEAD cipher spec";
2860 			return -ENOMEM;
2861 		}
2862 	}
2863 
2864 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2865 		cc->tfms_count = 64;
2866 
2867 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2868 		if (!*ivopts) {
2869 			ti->error = "Digest algorithm missing for ESSIV mode";
2870 			return -EINVAL;
2871 		}
2872 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2873 			       cipher_api, *ivopts);
2874 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2875 			ti->error = "Cannot allocate cipher string";
2876 			return -ENOMEM;
2877 		}
2878 		cipher_api = buf;
2879 	}
2880 
2881 	cc->key_parts = cc->tfms_count;
2882 
2883 	/* Allocate cipher */
2884 	ret = crypt_alloc_tfms(cc, cipher_api);
2885 	if (ret < 0) {
2886 		ti->error = "Error allocating crypto tfm";
2887 		return ret;
2888 	}
2889 
2890 	if (crypt_integrity_aead(cc))
2891 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2892 	else
2893 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2894 
2895 	return 0;
2896 }
2897 
2898 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2899 				char **ivmode, char **ivopts)
2900 {
2901 	struct crypt_config *cc = ti->private;
2902 	char *tmp, *cipher, *chainmode, *keycount;
2903 	char *cipher_api = NULL;
2904 	int ret = -EINVAL;
2905 	char dummy;
2906 
2907 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2908 		ti->error = "Bad cipher specification";
2909 		return -EINVAL;
2910 	}
2911 
2912 	/*
2913 	 * Legacy dm-crypt cipher specification
2914 	 * cipher[:keycount]-mode-iv:ivopts
2915 	 */
2916 	tmp = cipher_in;
2917 	keycount = strsep(&tmp, "-");
2918 	cipher = strsep(&keycount, ":");
2919 
2920 	if (!keycount)
2921 		cc->tfms_count = 1;
2922 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2923 		 !is_power_of_2(cc->tfms_count)) {
2924 		ti->error = "Bad cipher key count specification";
2925 		return -EINVAL;
2926 	}
2927 	cc->key_parts = cc->tfms_count;
2928 
2929 	chainmode = strsep(&tmp, "-");
2930 	*ivmode = strsep(&tmp, ":");
2931 	*ivopts = tmp;
2932 
2933 	/*
2934 	 * For compatibility with the original dm-crypt mapping format, if
2935 	 * only the cipher name is supplied, use cbc-plain.
2936 	 */
2937 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2938 		chainmode = "cbc";
2939 		*ivmode = "plain";
2940 	}
2941 
2942 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2943 		ti->error = "IV mechanism required";
2944 		return -EINVAL;
2945 	}
2946 
2947 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2948 	if (!cipher_api)
2949 		goto bad_mem;
2950 
2951 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2952 		if (!*ivopts) {
2953 			ti->error = "Digest algorithm missing for ESSIV mode";
2954 			kfree(cipher_api);
2955 			return -EINVAL;
2956 		}
2957 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2958 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2959 	} else {
2960 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2961 			       "%s(%s)", chainmode, cipher);
2962 	}
2963 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2964 		kfree(cipher_api);
2965 		goto bad_mem;
2966 	}
2967 
2968 	/* Allocate cipher */
2969 	ret = crypt_alloc_tfms(cc, cipher_api);
2970 	if (ret < 0) {
2971 		ti->error = "Error allocating crypto tfm";
2972 		kfree(cipher_api);
2973 		return ret;
2974 	}
2975 	kfree(cipher_api);
2976 
2977 	return 0;
2978 bad_mem:
2979 	ti->error = "Cannot allocate cipher strings";
2980 	return -ENOMEM;
2981 }
2982 
2983 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
2984 {
2985 	struct crypt_config *cc = ti->private;
2986 	char *ivmode = NULL, *ivopts = NULL;
2987 	int ret;
2988 
2989 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
2990 	if (!cc->cipher_string) {
2991 		ti->error = "Cannot allocate cipher strings";
2992 		return -ENOMEM;
2993 	}
2994 
2995 	if (strstarts(cipher_in, "capi:"))
2996 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
2997 	else
2998 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
2999 	if (ret)
3000 		return ret;
3001 
3002 	/* Initialize IV */
3003 	ret = crypt_ctr_ivmode(ti, ivmode);
3004 	if (ret < 0)
3005 		return ret;
3006 
3007 	/* Initialize and set key */
3008 	ret = crypt_set_key(cc, key);
3009 	if (ret < 0) {
3010 		ti->error = "Error decoding and setting key";
3011 		return ret;
3012 	}
3013 
3014 	/* Allocate IV */
3015 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3016 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3017 		if (ret < 0) {
3018 			ti->error = "Error creating IV";
3019 			return ret;
3020 		}
3021 	}
3022 
3023 	/* Initialize IV (set keys for ESSIV etc) */
3024 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3025 		ret = cc->iv_gen_ops->init(cc);
3026 		if (ret < 0) {
3027 			ti->error = "Error initialising IV";
3028 			return ret;
3029 		}
3030 	}
3031 
3032 	/* wipe the kernel key payload copy */
3033 	if (cc->key_string)
3034 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3035 
3036 	return ret;
3037 }
3038 
3039 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3040 {
3041 	struct crypt_config *cc = ti->private;
3042 	struct dm_arg_set as;
3043 	static const struct dm_arg _args[] = {
3044 		{0, 8, "Invalid number of feature args"},
3045 	};
3046 	unsigned int opt_params, val;
3047 	const char *opt_string, *sval;
3048 	char dummy;
3049 	int ret;
3050 
3051 	/* Optional parameters */
3052 	as.argc = argc;
3053 	as.argv = argv;
3054 
3055 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3056 	if (ret)
3057 		return ret;
3058 
3059 	while (opt_params--) {
3060 		opt_string = dm_shift_arg(&as);
3061 		if (!opt_string) {
3062 			ti->error = "Not enough feature arguments";
3063 			return -EINVAL;
3064 		}
3065 
3066 		if (!strcasecmp(opt_string, "allow_discards"))
3067 			ti->num_discard_bios = 1;
3068 
3069 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3070 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3071 
3072 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3073 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3074 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3075 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3076 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3077 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3078 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3079 			if (val == 0 || val > MAX_TAG_SIZE) {
3080 				ti->error = "Invalid integrity arguments";
3081 				return -EINVAL;
3082 			}
3083 			cc->on_disk_tag_size = val;
3084 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3085 			if (!strcasecmp(sval, "aead")) {
3086 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3087 			} else  if (strcasecmp(sval, "none")) {
3088 				ti->error = "Unknown integrity profile";
3089 				return -EINVAL;
3090 			}
3091 
3092 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3093 			if (!cc->cipher_auth)
3094 				return -ENOMEM;
3095 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3096 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3097 			    cc->sector_size > 4096 ||
3098 			    (cc->sector_size & (cc->sector_size - 1))) {
3099 				ti->error = "Invalid feature value for sector_size";
3100 				return -EINVAL;
3101 			}
3102 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3103 				ti->error = "Device size is not multiple of sector_size feature";
3104 				return -EINVAL;
3105 			}
3106 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3107 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3108 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3109 		else {
3110 			ti->error = "Invalid feature arguments";
3111 			return -EINVAL;
3112 		}
3113 	}
3114 
3115 	return 0;
3116 }
3117 
3118 #ifdef CONFIG_BLK_DEV_ZONED
3119 
3120 static int crypt_report_zones(struct dm_target *ti,
3121 		struct dm_report_zones_args *args, unsigned int nr_zones)
3122 {
3123 	struct crypt_config *cc = ti->private;
3124 	sector_t sector = cc->start + dm_target_offset(ti, args->next_sector);
3125 
3126 	args->start = cc->start;
3127 	return blkdev_report_zones(cc->dev->bdev, sector, nr_zones,
3128 				   dm_report_zones_cb, args);
3129 }
3130 
3131 #endif
3132 
3133 /*
3134  * Construct an encryption mapping:
3135  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3136  */
3137 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3138 {
3139 	struct crypt_config *cc;
3140 	const char *devname = dm_table_device_name(ti->table);
3141 	int key_size;
3142 	unsigned int align_mask;
3143 	unsigned long long tmpll;
3144 	int ret;
3145 	size_t iv_size_padding, additional_req_size;
3146 	char dummy;
3147 
3148 	if (argc < 5) {
3149 		ti->error = "Not enough arguments";
3150 		return -EINVAL;
3151 	}
3152 
3153 	key_size = get_key_size(&argv[1]);
3154 	if (key_size < 0) {
3155 		ti->error = "Cannot parse key size";
3156 		return -EINVAL;
3157 	}
3158 
3159 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3160 	if (!cc) {
3161 		ti->error = "Cannot allocate encryption context";
3162 		return -ENOMEM;
3163 	}
3164 	cc->key_size = key_size;
3165 	cc->sector_size = (1 << SECTOR_SHIFT);
3166 	cc->sector_shift = 0;
3167 
3168 	ti->private = cc;
3169 
3170 	spin_lock(&dm_crypt_clients_lock);
3171 	dm_crypt_clients_n++;
3172 	crypt_calculate_pages_per_client();
3173 	spin_unlock(&dm_crypt_clients_lock);
3174 
3175 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3176 	if (ret < 0)
3177 		goto bad;
3178 
3179 	/* Optional parameters need to be read before cipher constructor */
3180 	if (argc > 5) {
3181 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3182 		if (ret)
3183 			goto bad;
3184 	}
3185 
3186 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3187 	if (ret < 0)
3188 		goto bad;
3189 
3190 	if (crypt_integrity_aead(cc)) {
3191 		cc->dmreq_start = sizeof(struct aead_request);
3192 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3193 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3194 	} else {
3195 		cc->dmreq_start = sizeof(struct skcipher_request);
3196 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3197 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3198 	}
3199 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3200 
3201 	if (align_mask < CRYPTO_MINALIGN) {
3202 		/* Allocate the padding exactly */
3203 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3204 				& align_mask;
3205 	} else {
3206 		/*
3207 		 * If the cipher requires greater alignment than kmalloc
3208 		 * alignment, we don't know the exact position of the
3209 		 * initialization vector. We must assume worst case.
3210 		 */
3211 		iv_size_padding = align_mask;
3212 	}
3213 
3214 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3215 	additional_req_size = sizeof(struct dm_crypt_request) +
3216 		iv_size_padding + cc->iv_size +
3217 		cc->iv_size +
3218 		sizeof(uint64_t) +
3219 		sizeof(unsigned int);
3220 
3221 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3222 	if (ret) {
3223 		ti->error = "Cannot allocate crypt request mempool";
3224 		goto bad;
3225 	}
3226 
3227 	cc->per_bio_data_size = ti->per_io_data_size =
3228 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3229 		      ARCH_KMALLOC_MINALIGN);
3230 
3231 	ret = mempool_init(&cc->page_pool, BIO_MAX_PAGES, crypt_page_alloc, crypt_page_free, cc);
3232 	if (ret) {
3233 		ti->error = "Cannot allocate page mempool";
3234 		goto bad;
3235 	}
3236 
3237 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3238 	if (ret) {
3239 		ti->error = "Cannot allocate crypt bioset";
3240 		goto bad;
3241 	}
3242 
3243 	mutex_init(&cc->bio_alloc_lock);
3244 
3245 	ret = -EINVAL;
3246 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3247 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3248 		ti->error = "Invalid iv_offset sector";
3249 		goto bad;
3250 	}
3251 	cc->iv_offset = tmpll;
3252 
3253 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3254 	if (ret) {
3255 		ti->error = "Device lookup failed";
3256 		goto bad;
3257 	}
3258 
3259 	ret = -EINVAL;
3260 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3261 		ti->error = "Invalid device sector";
3262 		goto bad;
3263 	}
3264 	cc->start = tmpll;
3265 
3266 	/*
3267 	 * For zoned block devices, we need to preserve the issuer write
3268 	 * ordering. To do so, disable write workqueues and force inline
3269 	 * encryption completion.
3270 	 */
3271 	if (bdev_is_zoned(cc->dev->bdev)) {
3272 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3273 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3274 	}
3275 
3276 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3277 		ret = crypt_integrity_ctr(cc, ti);
3278 		if (ret)
3279 			goto bad;
3280 
3281 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3282 		if (!cc->tag_pool_max_sectors)
3283 			cc->tag_pool_max_sectors = 1;
3284 
3285 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3286 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3287 		if (ret) {
3288 			ti->error = "Cannot allocate integrity tags mempool";
3289 			goto bad;
3290 		}
3291 
3292 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3293 	}
3294 
3295 	ret = -ENOMEM;
3296 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3297 	if (!cc->io_queue) {
3298 		ti->error = "Couldn't create kcryptd io queue";
3299 		goto bad;
3300 	}
3301 
3302 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3303 		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3304 						  1, devname);
3305 	else
3306 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3307 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3308 						  num_online_cpus(), devname);
3309 	if (!cc->crypt_queue) {
3310 		ti->error = "Couldn't create kcryptd queue";
3311 		goto bad;
3312 	}
3313 
3314 	spin_lock_init(&cc->write_thread_lock);
3315 	cc->write_tree = RB_ROOT;
3316 
3317 	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3318 	if (IS_ERR(cc->write_thread)) {
3319 		ret = PTR_ERR(cc->write_thread);
3320 		cc->write_thread = NULL;
3321 		ti->error = "Couldn't spawn write thread";
3322 		goto bad;
3323 	}
3324 	wake_up_process(cc->write_thread);
3325 
3326 	ti->num_flush_bios = 1;
3327 
3328 	return 0;
3329 
3330 bad:
3331 	crypt_dtr(ti);
3332 	return ret;
3333 }
3334 
3335 static int crypt_map(struct dm_target *ti, struct bio *bio)
3336 {
3337 	struct dm_crypt_io *io;
3338 	struct crypt_config *cc = ti->private;
3339 
3340 	/*
3341 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3342 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3343 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3344 	 */
3345 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3346 	    bio_op(bio) == REQ_OP_DISCARD)) {
3347 		bio_set_dev(bio, cc->dev->bdev);
3348 		if (bio_sectors(bio))
3349 			bio->bi_iter.bi_sector = cc->start +
3350 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3351 		return DM_MAPIO_REMAPPED;
3352 	}
3353 
3354 	/*
3355 	 * Check if bio is too large, split as needed.
3356 	 */
3357 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_PAGES << PAGE_SHIFT)) &&
3358 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3359 		dm_accept_partial_bio(bio, ((BIO_MAX_PAGES << PAGE_SHIFT) >> SECTOR_SHIFT));
3360 
3361 	/*
3362 	 * Ensure that bio is a multiple of internal sector encryption size
3363 	 * and is aligned to this size as defined in IO hints.
3364 	 */
3365 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3366 		return DM_MAPIO_KILL;
3367 
3368 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3369 		return DM_MAPIO_KILL;
3370 
3371 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3372 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3373 
3374 	if (cc->on_disk_tag_size) {
3375 		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3376 
3377 		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3378 		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3379 				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3380 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3381 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3382 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3383 			io->integrity_metadata_from_pool = true;
3384 		}
3385 	}
3386 
3387 	if (crypt_integrity_aead(cc))
3388 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3389 	else
3390 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3391 
3392 	if (bio_data_dir(io->base_bio) == READ) {
3393 		if (kcryptd_io_read(io, GFP_NOWAIT))
3394 			kcryptd_queue_read(io);
3395 	} else
3396 		kcryptd_queue_crypt(io);
3397 
3398 	return DM_MAPIO_SUBMITTED;
3399 }
3400 
3401 static void crypt_status(struct dm_target *ti, status_type_t type,
3402 			 unsigned status_flags, char *result, unsigned maxlen)
3403 {
3404 	struct crypt_config *cc = ti->private;
3405 	unsigned i, sz = 0;
3406 	int num_feature_args = 0;
3407 
3408 	switch (type) {
3409 	case STATUSTYPE_INFO:
3410 		result[0] = '\0';
3411 		break;
3412 
3413 	case STATUSTYPE_TABLE:
3414 		DMEMIT("%s ", cc->cipher_string);
3415 
3416 		if (cc->key_size > 0) {
3417 			if (cc->key_string)
3418 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3419 			else
3420 				for (i = 0; i < cc->key_size; i++)
3421 					DMEMIT("%02x", cc->key[i]);
3422 		} else
3423 			DMEMIT("-");
3424 
3425 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3426 				cc->dev->name, (unsigned long long)cc->start);
3427 
3428 		num_feature_args += !!ti->num_discard_bios;
3429 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3430 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3431 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3432 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3433 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3434 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3435 		if (cc->on_disk_tag_size)
3436 			num_feature_args++;
3437 		if (num_feature_args) {
3438 			DMEMIT(" %d", num_feature_args);
3439 			if (ti->num_discard_bios)
3440 				DMEMIT(" allow_discards");
3441 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3442 				DMEMIT(" same_cpu_crypt");
3443 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3444 				DMEMIT(" submit_from_crypt_cpus");
3445 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3446 				DMEMIT(" no_read_workqueue");
3447 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3448 				DMEMIT(" no_write_workqueue");
3449 			if (cc->on_disk_tag_size)
3450 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3451 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3452 				DMEMIT(" sector_size:%d", cc->sector_size);
3453 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3454 				DMEMIT(" iv_large_sectors");
3455 		}
3456 
3457 		break;
3458 	}
3459 }
3460 
3461 static void crypt_postsuspend(struct dm_target *ti)
3462 {
3463 	struct crypt_config *cc = ti->private;
3464 
3465 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3466 }
3467 
3468 static int crypt_preresume(struct dm_target *ti)
3469 {
3470 	struct crypt_config *cc = ti->private;
3471 
3472 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3473 		DMERR("aborting resume - crypt key is not set.");
3474 		return -EAGAIN;
3475 	}
3476 
3477 	return 0;
3478 }
3479 
3480 static void crypt_resume(struct dm_target *ti)
3481 {
3482 	struct crypt_config *cc = ti->private;
3483 
3484 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3485 }
3486 
3487 /* Message interface
3488  *	key set <key>
3489  *	key wipe
3490  */
3491 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3492 			 char *result, unsigned maxlen)
3493 {
3494 	struct crypt_config *cc = ti->private;
3495 	int key_size, ret = -EINVAL;
3496 
3497 	if (argc < 2)
3498 		goto error;
3499 
3500 	if (!strcasecmp(argv[0], "key")) {
3501 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3502 			DMWARN("not suspended during key manipulation.");
3503 			return -EINVAL;
3504 		}
3505 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3506 			/* The key size may not be changed. */
3507 			key_size = get_key_size(&argv[2]);
3508 			if (key_size < 0 || cc->key_size != key_size) {
3509 				memset(argv[2], '0', strlen(argv[2]));
3510 				return -EINVAL;
3511 			}
3512 
3513 			ret = crypt_set_key(cc, argv[2]);
3514 			if (ret)
3515 				return ret;
3516 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3517 				ret = cc->iv_gen_ops->init(cc);
3518 			/* wipe the kernel key payload copy */
3519 			if (cc->key_string)
3520 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3521 			return ret;
3522 		}
3523 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3524 			return crypt_wipe_key(cc);
3525 	}
3526 
3527 error:
3528 	DMWARN("unrecognised message received.");
3529 	return -EINVAL;
3530 }
3531 
3532 static int crypt_iterate_devices(struct dm_target *ti,
3533 				 iterate_devices_callout_fn fn, void *data)
3534 {
3535 	struct crypt_config *cc = ti->private;
3536 
3537 	return fn(ti, cc->dev, cc->start, ti->len, data);
3538 }
3539 
3540 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3541 {
3542 	struct crypt_config *cc = ti->private;
3543 
3544 	/*
3545 	 * Unfortunate constraint that is required to avoid the potential
3546 	 * for exceeding underlying device's max_segments limits -- due to
3547 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3548 	 * bio that are not as physically contiguous as the original bio.
3549 	 */
3550 	limits->max_segment_size = PAGE_SIZE;
3551 
3552 	limits->logical_block_size =
3553 		max_t(unsigned, limits->logical_block_size, cc->sector_size);
3554 	limits->physical_block_size =
3555 		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3556 	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3557 }
3558 
3559 static struct target_type crypt_target = {
3560 	.name   = "crypt",
3561 	.version = {1, 22, 0},
3562 	.module = THIS_MODULE,
3563 	.ctr    = crypt_ctr,
3564 	.dtr    = crypt_dtr,
3565 #ifdef CONFIG_BLK_DEV_ZONED
3566 	.features = DM_TARGET_ZONED_HM,
3567 	.report_zones = crypt_report_zones,
3568 #endif
3569 	.map    = crypt_map,
3570 	.status = crypt_status,
3571 	.postsuspend = crypt_postsuspend,
3572 	.preresume = crypt_preresume,
3573 	.resume = crypt_resume,
3574 	.message = crypt_message,
3575 	.iterate_devices = crypt_iterate_devices,
3576 	.io_hints = crypt_io_hints,
3577 };
3578 
3579 static int __init dm_crypt_init(void)
3580 {
3581 	int r;
3582 
3583 	r = dm_register_target(&crypt_target);
3584 	if (r < 0)
3585 		DMERR("register failed %d", r);
3586 
3587 	return r;
3588 }
3589 
3590 static void __exit dm_crypt_exit(void)
3591 {
3592 	dm_unregister_target(&crypt_target);
3593 }
3594 
3595 module_init(dm_crypt_init);
3596 module_exit(dm_crypt_exit);
3597 
3598 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3599 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3600 MODULE_LICENSE("GPL");
3601