xref: /linux/drivers/md/dm-crypt.c (revision 136114e0abf03005e182d75761ab694648e6d388)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2003 Jana Saout <jana@saout.de>
4  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
5  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
6  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
7  *
8  * This file is released under the GPL.
9  */
10 
11 #include <linux/completion.h>
12 #include <linux/err.h>
13 #include <linux/module.h>
14 #include <linux/hex.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/key.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-integrity.h>
21 #include <linux/crc32.h>
22 #include <linux/mempool.h>
23 #include <linux/slab.h>
24 #include <linux/crypto.h>
25 #include <linux/fips.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/backing-dev.h>
29 #include <linux/atomic.h>
30 #include <linux/scatterlist.h>
31 #include <linux/rbtree.h>
32 #include <linux/ctype.h>
33 #include <asm/page.h>
34 #include <linux/unaligned.h>
35 #include <crypto/hash.h>
36 #include <crypto/md5.h>
37 #include <crypto/skcipher.h>
38 #include <crypto/aead.h>
39 #include <crypto/authenc.h>
40 #include <crypto/utils.h>
41 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
42 #include <linux/key-type.h>
43 #include <keys/user-type.h>
44 #include <keys/encrypted-type.h>
45 #include <keys/trusted-type.h>
46 
47 #include <linux/device-mapper.h>
48 
49 #include "dm-audit.h"
50 
51 #define DM_MSG_PREFIX "crypt"
52 
53 static DEFINE_IDA(workqueue_ida);
54 
55 /*
56  * context holding the current state of a multi-part conversion
57  */
58 struct convert_context {
59 	struct completion restart;
60 	struct bio *bio_in;
61 	struct bvec_iter iter_in;
62 	struct bio *bio_out;
63 	struct bvec_iter iter_out;
64 	atomic_t cc_pending;
65 	unsigned int tag_offset;
66 	u64 cc_sector;
67 	union {
68 		struct skcipher_request *req;
69 		struct aead_request *req_aead;
70 	} r;
71 	bool aead_recheck;
72 	bool aead_failed;
73 
74 };
75 
76 /*
77  * per bio private data
78  */
79 struct dm_crypt_io {
80 	struct crypt_config *cc;
81 	struct bio *base_bio;
82 	u8 *integrity_metadata;
83 	bool integrity_metadata_from_pool:1;
84 
85 	struct work_struct work;
86 
87 	struct convert_context ctx;
88 
89 	atomic_t io_pending;
90 	blk_status_t error;
91 	sector_t sector;
92 
93 	struct bvec_iter saved_bi_iter;
94 
95 	struct rb_node rb_node;
96 } CRYPTO_MINALIGN_ATTR;
97 
98 struct dm_crypt_request {
99 	struct convert_context *ctx;
100 	struct scatterlist sg_in[4];
101 	struct scatterlist sg_out[4];
102 	u64 iv_sector;
103 };
104 
105 struct crypt_config;
106 
107 struct crypt_iv_operations {
108 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
109 		   const char *opts);
110 	void (*dtr)(struct crypt_config *cc);
111 	int (*init)(struct crypt_config *cc);
112 	int (*wipe)(struct crypt_config *cc);
113 	int (*generator)(struct crypt_config *cc, u8 *iv,
114 			 struct dm_crypt_request *dmreq);
115 	int (*post)(struct crypt_config *cc, u8 *iv,
116 		    struct dm_crypt_request *dmreq);
117 };
118 
119 struct iv_benbi_private {
120 	int shift;
121 };
122 
123 #define LMK_SEED_SIZE 64 /* hash + 0 */
124 struct iv_lmk_private {
125 	u8 *seed;
126 };
127 
128 #define TCW_WHITENING_SIZE 16
129 struct iv_tcw_private {
130 	u8 *iv_seed;
131 	u8 *whitening;
132 };
133 
134 #define ELEPHANT_MAX_KEY_SIZE 32
135 struct iv_elephant_private {
136 	struct crypto_skcipher *tfm;
137 };
138 
139 /*
140  * Crypt: maps a linear range of a block device
141  * and encrypts / decrypts at the same time.
142  */
143 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
144 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
145 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
146 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
147 
148 enum cipher_flags {
149 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
150 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
151 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
152 	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
153 };
154 
155 /*
156  * The fields in here must be read only after initialization.
157  */
158 struct crypt_config {
159 	struct dm_dev *dev;
160 	sector_t start;
161 
162 	struct percpu_counter n_allocated_pages;
163 
164 	struct workqueue_struct *io_queue;
165 	struct workqueue_struct *crypt_queue;
166 
167 	spinlock_t write_thread_lock;
168 	struct task_struct *write_thread;
169 	struct rb_root write_tree;
170 
171 	char *cipher_string;
172 	char *cipher_auth;
173 	char *key_string;
174 
175 	const struct crypt_iv_operations *iv_gen_ops;
176 	union {
177 		struct iv_benbi_private benbi;
178 		struct iv_lmk_private lmk;
179 		struct iv_tcw_private tcw;
180 		struct iv_elephant_private elephant;
181 	} iv_gen_private;
182 	u64 iv_offset;
183 	unsigned int iv_size;
184 	unsigned short sector_size;
185 	unsigned char sector_shift;
186 
187 	union {
188 		struct crypto_skcipher **tfms;
189 		struct crypto_aead **tfms_aead;
190 	} cipher_tfm;
191 	unsigned int tfms_count;
192 	int workqueue_id;
193 	unsigned long cipher_flags;
194 
195 	/*
196 	 * Layout of each crypto request:
197 	 *
198 	 *   struct skcipher_request
199 	 *      context
200 	 *      padding
201 	 *   struct dm_crypt_request
202 	 *      padding
203 	 *   IV
204 	 *
205 	 * The padding is added so that dm_crypt_request and the IV are
206 	 * correctly aligned.
207 	 */
208 	unsigned int dmreq_start;
209 
210 	unsigned int per_bio_data_size;
211 
212 	unsigned long flags;
213 	unsigned int key_size;
214 	unsigned int key_parts;      /* independent parts in key buffer */
215 	unsigned int key_extra_size; /* additional keys length */
216 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
217 
218 	unsigned int integrity_tag_size;
219 	unsigned int integrity_iv_size;
220 	unsigned int used_tag_size;
221 	unsigned int tuple_size;
222 
223 	/*
224 	 * pool for per bio private data, crypto requests,
225 	 * encryption requeusts/buffer pages and integrity tags
226 	 */
227 	unsigned int tag_pool_max_sectors;
228 	mempool_t tag_pool;
229 	mempool_t req_pool;
230 	mempool_t page_pool;
231 
232 	struct bio_set bs;
233 	struct mutex bio_alloc_lock;
234 
235 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
236 	u8 key[] __counted_by(key_size);
237 };
238 
239 #define MIN_IOS		64
240 #define MAX_TAG_SIZE	480
241 #define POOL_ENTRY_SIZE	512
242 
243 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
244 static unsigned int dm_crypt_clients_n;
245 static volatile unsigned long dm_crypt_pages_per_client;
246 #define DM_CRYPT_MEMORY_PERCENT			2
247 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
248 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
249 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
250 
251 static unsigned int max_read_size = 0;
252 module_param(max_read_size, uint, 0644);
253 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
254 static unsigned int max_write_size = 0;
255 module_param(max_write_size, uint, 0644);
256 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
257 
258 static unsigned get_max_request_sectors(struct dm_target *ti, struct bio *bio, bool no_split)
259 {
260 	struct crypt_config *cc = ti->private;
261 	unsigned val, sector_align;
262 	bool wrt = op_is_write(bio_op(bio));
263 
264 	if (no_split) {
265 		val = -1;
266 	} else if (wrt) {
267 		val = min_not_zero(READ_ONCE(max_write_size),
268 				   DM_CRYPT_DEFAULT_MAX_WRITE_SIZE);
269 	} else {
270 		val = min_not_zero(READ_ONCE(max_read_size),
271 				   DM_CRYPT_DEFAULT_MAX_READ_SIZE);
272 	}
273 
274 	if (wrt || cc->used_tag_size)
275 		val = min(val, BIO_MAX_VECS << PAGE_SHIFT);
276 
277 	sector_align = max(bdev_logical_block_size(cc->dev->bdev),
278 			   (unsigned)cc->sector_size);
279 	val = round_down(val, sector_align);
280 	if (unlikely(!val))
281 		val = sector_align;
282 	return val >> SECTOR_SHIFT;
283 }
284 
285 static void crypt_endio(struct bio *clone);
286 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
287 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
288 					     struct scatterlist *sg);
289 
290 static bool crypt_integrity_aead(struct crypt_config *cc);
291 
292 /*
293  * Use this to access cipher attributes that are independent of the key.
294  */
295 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
296 {
297 	return cc->cipher_tfm.tfms[0];
298 }
299 
300 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
301 {
302 	return cc->cipher_tfm.tfms_aead[0];
303 }
304 
305 /*
306  * Different IV generation algorithms:
307  *
308  * plain: the initial vector is the 32-bit little-endian version of the sector
309  *        number, padded with zeros if necessary.
310  *
311  * plain64: the initial vector is the 64-bit little-endian version of the sector
312  *        number, padded with zeros if necessary.
313  *
314  * plain64be: the initial vector is the 64-bit big-endian version of the sector
315  *        number, padded with zeros if necessary.
316  *
317  * essiv: "encrypted sector|salt initial vector", the sector number is
318  *        encrypted with the bulk cipher using a salt as key. The salt
319  *        should be derived from the bulk cipher's key via hashing.
320  *
321  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
322  *        (needed for LRW-32-AES and possible other narrow block modes)
323  *
324  * null: the initial vector is always zero.  Provides compatibility with
325  *       obsolete loop_fish2 devices.  Do not use for new devices.
326  *
327  * lmk:  Compatible implementation of the block chaining mode used
328  *       by the Loop-AES block device encryption system
329  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
330  *       It operates on full 512 byte sectors and uses CBC
331  *       with an IV derived from the sector number, the data and
332  *       optionally extra IV seed.
333  *       This means that after decryption the first block
334  *       of sector must be tweaked according to decrypted data.
335  *       Loop-AES can use three encryption schemes:
336  *         version 1: is plain aes-cbc mode
337  *         version 2: uses 64 multikey scheme with lmk IV generator
338  *         version 3: the same as version 2 with additional IV seed
339  *                   (it uses 65 keys, last key is used as IV seed)
340  *
341  * tcw:  Compatible implementation of the block chaining mode used
342  *       by the TrueCrypt device encryption system (prior to version 4.1).
343  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
344  *       It operates on full 512 byte sectors and uses CBC
345  *       with an IV derived from initial key and the sector number.
346  *       In addition, whitening value is applied on every sector, whitening
347  *       is calculated from initial key, sector number and mixed using CRC32.
348  *       Note that this encryption scheme is vulnerable to watermarking attacks
349  *       and should be used for old compatible containers access only.
350  *
351  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
352  *        The IV is encrypted little-endian byte-offset (with the same key
353  *        and cipher as the volume).
354  *
355  * elephant: The extended version of eboiv with additional Elephant diffuser
356  *           used with Bitlocker CBC mode.
357  *           This mode was used in older Windows systems
358  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
359  */
360 
361 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
362 			      struct dm_crypt_request *dmreq)
363 {
364 	memset(iv, 0, cc->iv_size);
365 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
366 
367 	return 0;
368 }
369 
370 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
371 				struct dm_crypt_request *dmreq)
372 {
373 	memset(iv, 0, cc->iv_size);
374 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
375 
376 	return 0;
377 }
378 
379 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
380 				  struct dm_crypt_request *dmreq)
381 {
382 	memset(iv, 0, cc->iv_size);
383 	/* iv_size is at least of size u64; usually it is 16 bytes */
384 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
385 
386 	return 0;
387 }
388 
389 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
390 			      struct dm_crypt_request *dmreq)
391 {
392 	/*
393 	 * ESSIV encryption of the IV is now handled by the crypto API,
394 	 * so just pass the plain sector number here.
395 	 */
396 	memset(iv, 0, cc->iv_size);
397 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
398 
399 	return 0;
400 }
401 
402 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
403 			      const char *opts)
404 {
405 	unsigned int bs;
406 	int log;
407 
408 	if (crypt_integrity_aead(cc))
409 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
410 	else
411 		bs = crypto_skcipher_blocksize(any_tfm(cc));
412 	log = ilog2(bs);
413 
414 	/*
415 	 * We need to calculate how far we must shift the sector count
416 	 * to get the cipher block count, we use this shift in _gen.
417 	 */
418 	if (1 << log != bs) {
419 		ti->error = "cypher blocksize is not a power of 2";
420 		return -EINVAL;
421 	}
422 
423 	if (log > 9) {
424 		ti->error = "cypher blocksize is > 512";
425 		return -EINVAL;
426 	}
427 
428 	cc->iv_gen_private.benbi.shift = 9 - log;
429 
430 	return 0;
431 }
432 
433 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
434 {
435 }
436 
437 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
438 			      struct dm_crypt_request *dmreq)
439 {
440 	__be64 val;
441 
442 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
443 
444 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
445 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
446 
447 	return 0;
448 }
449 
450 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
451 			     struct dm_crypt_request *dmreq)
452 {
453 	memset(iv, 0, cc->iv_size);
454 
455 	return 0;
456 }
457 
458 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
459 {
460 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
461 
462 	kfree_sensitive(lmk->seed);
463 	lmk->seed = NULL;
464 }
465 
466 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
467 			    const char *opts)
468 {
469 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
470 
471 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
472 		ti->error = "Unsupported sector size for LMK";
473 		return -EINVAL;
474 	}
475 
476 	if (fips_enabled) {
477 		ti->error = "LMK support is disabled due to FIPS";
478 		/* ... because it uses MD5. */
479 		return -EINVAL;
480 	}
481 
482 	/* No seed in LMK version 2 */
483 	if (cc->key_parts == cc->tfms_count) {
484 		lmk->seed = NULL;
485 		return 0;
486 	}
487 
488 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
489 	if (!lmk->seed) {
490 		ti->error = "Error kmallocing seed storage in LMK";
491 		return -ENOMEM;
492 	}
493 
494 	return 0;
495 }
496 
497 static int crypt_iv_lmk_init(struct crypt_config *cc)
498 {
499 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
500 	int subkey_size = cc->key_size / cc->key_parts;
501 
502 	/* LMK seed is on the position of LMK_KEYS + 1 key */
503 	if (lmk->seed)
504 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
505 		       MD5_DIGEST_SIZE);
506 
507 	return 0;
508 }
509 
510 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
511 {
512 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
513 
514 	if (lmk->seed)
515 		memset(lmk->seed, 0, LMK_SEED_SIZE);
516 
517 	return 0;
518 }
519 
520 static void crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
521 			     struct dm_crypt_request *dmreq, u8 *data)
522 {
523 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
524 	struct md5_ctx ctx;
525 	__le32 buf[4];
526 
527 	md5_init(&ctx);
528 
529 	if (lmk->seed)
530 		md5_update(&ctx, lmk->seed, LMK_SEED_SIZE);
531 
532 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
533 	md5_update(&ctx, data + 16, 16 * 31);
534 
535 	/* Sector is cropped to 56 bits here */
536 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
537 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
538 	buf[2] = cpu_to_le32(4024);
539 	buf[3] = 0;
540 	md5_update(&ctx, (u8 *)buf, sizeof(buf));
541 
542 	/* No MD5 padding here */
543 	cpu_to_le32_array(ctx.state.h, ARRAY_SIZE(ctx.state.h));
544 	memcpy(iv, ctx.state.h, cc->iv_size);
545 }
546 
547 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
548 			    struct dm_crypt_request *dmreq)
549 {
550 	struct scatterlist *sg;
551 	u8 *src;
552 
553 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
554 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
555 		src = kmap_local_page(sg_page(sg));
556 		crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
557 		kunmap_local(src);
558 	} else
559 		memset(iv, 0, cc->iv_size);
560 	return 0;
561 }
562 
563 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
564 			     struct dm_crypt_request *dmreq)
565 {
566 	struct scatterlist *sg;
567 	u8 *dst;
568 
569 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
570 		return 0;
571 
572 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
573 	dst = kmap_local_page(sg_page(sg));
574 	crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
575 
576 	/* Tweak the first block of plaintext sector */
577 	crypto_xor(dst + sg->offset, iv, cc->iv_size);
578 
579 	kunmap_local(dst);
580 	return 0;
581 }
582 
583 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
584 {
585 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
586 
587 	kfree_sensitive(tcw->iv_seed);
588 	tcw->iv_seed = NULL;
589 	kfree_sensitive(tcw->whitening);
590 	tcw->whitening = NULL;
591 }
592 
593 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
594 			    const char *opts)
595 {
596 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
597 
598 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
599 		ti->error = "Unsupported sector size for TCW";
600 		return -EINVAL;
601 	}
602 
603 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
604 		ti->error = "Wrong key size for TCW";
605 		return -EINVAL;
606 	}
607 
608 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
609 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
610 	if (!tcw->iv_seed || !tcw->whitening) {
611 		crypt_iv_tcw_dtr(cc);
612 		ti->error = "Error allocating seed storage in TCW";
613 		return -ENOMEM;
614 	}
615 
616 	return 0;
617 }
618 
619 static int crypt_iv_tcw_init(struct crypt_config *cc)
620 {
621 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
622 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
623 
624 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
625 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
626 	       TCW_WHITENING_SIZE);
627 
628 	return 0;
629 }
630 
631 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
632 {
633 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
634 
635 	memset(tcw->iv_seed, 0, cc->iv_size);
636 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
637 
638 	return 0;
639 }
640 
641 static void crypt_iv_tcw_whitening(struct crypt_config *cc,
642 				   struct dm_crypt_request *dmreq, u8 *data)
643 {
644 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
645 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
646 	u8 buf[TCW_WHITENING_SIZE];
647 	int i;
648 
649 	/* xor whitening with sector number */
650 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
651 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
652 
653 	/* calculate crc32 for every 32bit part and xor it */
654 	for (i = 0; i < 4; i++)
655 		put_unaligned_le32(crc32(0, &buf[i * 4], 4), &buf[i * 4]);
656 	crypto_xor(&buf[0], &buf[12], 4);
657 	crypto_xor(&buf[4], &buf[8], 4);
658 
659 	/* apply whitening (8 bytes) to whole sector */
660 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
661 		crypto_xor(data + i * 8, buf, 8);
662 	memzero_explicit(buf, sizeof(buf));
663 }
664 
665 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
666 			    struct dm_crypt_request *dmreq)
667 {
668 	struct scatterlist *sg;
669 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
670 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
671 	u8 *src;
672 
673 	/* Remove whitening from ciphertext */
674 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
675 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
676 		src = kmap_local_page(sg_page(sg));
677 		crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
678 		kunmap_local(src);
679 	}
680 
681 	/* Calculate IV */
682 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
683 	if (cc->iv_size > 8)
684 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
685 			       cc->iv_size - 8);
686 
687 	return 0;
688 }
689 
690 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
691 			     struct dm_crypt_request *dmreq)
692 {
693 	struct scatterlist *sg;
694 	u8 *dst;
695 
696 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
697 		return 0;
698 
699 	/* Apply whitening on ciphertext */
700 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
701 	dst = kmap_local_page(sg_page(sg));
702 	crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
703 	kunmap_local(dst);
704 
705 	return 0;
706 }
707 
708 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
709 				struct dm_crypt_request *dmreq)
710 {
711 	/* Used only for writes, there must be an additional space to store IV */
712 	get_random_bytes(iv, cc->iv_size);
713 	return 0;
714 }
715 
716 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
717 			    const char *opts)
718 {
719 	if (crypt_integrity_aead(cc)) {
720 		ti->error = "AEAD transforms not supported for EBOIV";
721 		return -EINVAL;
722 	}
723 
724 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
725 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
726 		return -EINVAL;
727 	}
728 
729 	return 0;
730 }
731 
732 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
733 			    struct dm_crypt_request *dmreq)
734 {
735 	struct crypto_skcipher *tfm = any_tfm(cc);
736 	struct skcipher_request *req;
737 	struct scatterlist src, dst;
738 	DECLARE_CRYPTO_WAIT(wait);
739 	unsigned int reqsize;
740 	int err;
741 	u8 *buf;
742 
743 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
744 	reqsize = ALIGN(reqsize, __alignof__(__le64));
745 
746 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
747 	if (!req)
748 		return -ENOMEM;
749 
750 	skcipher_request_set_tfm(req, tfm);
751 
752 	buf = (u8 *)req + reqsize;
753 	memset(buf, 0, cc->iv_size);
754 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
755 
756 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
757 	sg_init_one(&dst, iv, cc->iv_size);
758 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
759 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
760 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
761 	kfree_sensitive(req);
762 
763 	return err;
764 }
765 
766 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
767 {
768 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
769 
770 	crypto_free_skcipher(elephant->tfm);
771 	elephant->tfm = NULL;
772 }
773 
774 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
775 			    const char *opts)
776 {
777 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
778 	int r;
779 
780 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
781 					      CRYPTO_ALG_ALLOCATES_MEMORY);
782 	if (IS_ERR(elephant->tfm)) {
783 		r = PTR_ERR(elephant->tfm);
784 		elephant->tfm = NULL;
785 		return r;
786 	}
787 
788 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
789 	if (r)
790 		crypt_iv_elephant_dtr(cc);
791 	return r;
792 }
793 
794 static void diffuser_disk_to_cpu(u32 *d, size_t n)
795 {
796 #ifndef __LITTLE_ENDIAN
797 	int i;
798 
799 	for (i = 0; i < n; i++)
800 		d[i] = le32_to_cpu((__le32)d[i]);
801 #endif
802 }
803 
804 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
805 {
806 #ifndef __LITTLE_ENDIAN
807 	int i;
808 
809 	for (i = 0; i < n; i++)
810 		d[i] = cpu_to_le32((u32)d[i]);
811 #endif
812 }
813 
814 static void diffuser_a_decrypt(u32 *d, size_t n)
815 {
816 	int i, i1, i2, i3;
817 
818 	for (i = 0; i < 5; i++) {
819 		i1 = 0;
820 		i2 = n - 2;
821 		i3 = n - 5;
822 
823 		while (i1 < (n - 1)) {
824 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
825 			i1++; i2++; i3++;
826 
827 			if (i3 >= n)
828 				i3 -= n;
829 
830 			d[i1] += d[i2] ^ d[i3];
831 			i1++; i2++; i3++;
832 
833 			if (i2 >= n)
834 				i2 -= n;
835 
836 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
837 			i1++; i2++; i3++;
838 
839 			d[i1] += d[i2] ^ d[i3];
840 			i1++; i2++; i3++;
841 		}
842 	}
843 }
844 
845 static void diffuser_a_encrypt(u32 *d, size_t n)
846 {
847 	int i, i1, i2, i3;
848 
849 	for (i = 0; i < 5; i++) {
850 		i1 = n - 1;
851 		i2 = n - 2 - 1;
852 		i3 = n - 5 - 1;
853 
854 		while (i1 > 0) {
855 			d[i1] -= d[i2] ^ d[i3];
856 			i1--; i2--; i3--;
857 
858 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
859 			i1--; i2--; i3--;
860 
861 			if (i2 < 0)
862 				i2 += n;
863 
864 			d[i1] -= d[i2] ^ d[i3];
865 			i1--; i2--; i3--;
866 
867 			if (i3 < 0)
868 				i3 += n;
869 
870 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
871 			i1--; i2--; i3--;
872 		}
873 	}
874 }
875 
876 static void diffuser_b_decrypt(u32 *d, size_t n)
877 {
878 	int i, i1, i2, i3;
879 
880 	for (i = 0; i < 3; i++) {
881 		i1 = 0;
882 		i2 = 2;
883 		i3 = 5;
884 
885 		while (i1 < (n - 1)) {
886 			d[i1] += d[i2] ^ d[i3];
887 			i1++; i2++; i3++;
888 
889 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
890 			i1++; i2++; i3++;
891 
892 			if (i2 >= n)
893 				i2 -= n;
894 
895 			d[i1] += d[i2] ^ d[i3];
896 			i1++; i2++; i3++;
897 
898 			if (i3 >= n)
899 				i3 -= n;
900 
901 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
902 			i1++; i2++; i3++;
903 		}
904 	}
905 }
906 
907 static void diffuser_b_encrypt(u32 *d, size_t n)
908 {
909 	int i, i1, i2, i3;
910 
911 	for (i = 0; i < 3; i++) {
912 		i1 = n - 1;
913 		i2 = 2 - 1;
914 		i3 = 5 - 1;
915 
916 		while (i1 > 0) {
917 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
918 			i1--; i2--; i3--;
919 
920 			if (i3 < 0)
921 				i3 += n;
922 
923 			d[i1] -= d[i2] ^ d[i3];
924 			i1--; i2--; i3--;
925 
926 			if (i2 < 0)
927 				i2 += n;
928 
929 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
930 			i1--; i2--; i3--;
931 
932 			d[i1] -= d[i2] ^ d[i3];
933 			i1--; i2--; i3--;
934 		}
935 	}
936 }
937 
938 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
939 {
940 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
941 	u8 *es, *ks, *data, *data2, *data_offset;
942 	struct skcipher_request *req;
943 	struct scatterlist *sg, *sg2, src, dst;
944 	DECLARE_CRYPTO_WAIT(wait);
945 	int i, r;
946 
947 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
948 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
949 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
950 
951 	if (!req || !es || !ks) {
952 		r = -ENOMEM;
953 		goto out;
954 	}
955 
956 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
957 
958 	/* E(Ks, e(s)) */
959 	sg_init_one(&src, es, 16);
960 	sg_init_one(&dst, ks, 16);
961 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
962 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
963 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
964 	if (r)
965 		goto out;
966 
967 	/* E(Ks, e'(s)) */
968 	es[15] = 0x80;
969 	sg_init_one(&dst, &ks[16], 16);
970 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
971 	if (r)
972 		goto out;
973 
974 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
975 	data = kmap_local_page(sg_page(sg));
976 	data_offset = data + sg->offset;
977 
978 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
979 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
980 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
981 		data2 = kmap_local_page(sg_page(sg2));
982 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
983 		kunmap_local(data2);
984 	}
985 
986 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
987 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
988 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
989 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
990 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
991 	}
992 
993 	for (i = 0; i < (cc->sector_size / 32); i++)
994 		crypto_xor(data_offset + i * 32, ks, 32);
995 
996 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
997 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
998 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
999 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1000 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1001 	}
1002 
1003 	kunmap_local(data);
1004 out:
1005 	kfree_sensitive(ks);
1006 	kfree_sensitive(es);
1007 	skcipher_request_free(req);
1008 	return r;
1009 }
1010 
1011 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1012 			    struct dm_crypt_request *dmreq)
1013 {
1014 	int r;
1015 
1016 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1017 		r = crypt_iv_elephant(cc, dmreq);
1018 		if (r)
1019 			return r;
1020 	}
1021 
1022 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1023 }
1024 
1025 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1026 				  struct dm_crypt_request *dmreq)
1027 {
1028 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1029 		return crypt_iv_elephant(cc, dmreq);
1030 
1031 	return 0;
1032 }
1033 
1034 static int crypt_iv_elephant_init(struct crypt_config *cc)
1035 {
1036 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1037 	int key_offset = cc->key_size - cc->key_extra_size;
1038 
1039 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1040 }
1041 
1042 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1043 {
1044 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1045 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1046 
1047 	memset(key, 0, cc->key_extra_size);
1048 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1049 }
1050 
1051 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1052 	.generator = crypt_iv_plain_gen
1053 };
1054 
1055 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1056 	.generator = crypt_iv_plain64_gen
1057 };
1058 
1059 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1060 	.generator = crypt_iv_plain64be_gen
1061 };
1062 
1063 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1064 	.generator = crypt_iv_essiv_gen
1065 };
1066 
1067 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1068 	.ctr	   = crypt_iv_benbi_ctr,
1069 	.dtr	   = crypt_iv_benbi_dtr,
1070 	.generator = crypt_iv_benbi_gen
1071 };
1072 
1073 static const struct crypt_iv_operations crypt_iv_null_ops = {
1074 	.generator = crypt_iv_null_gen
1075 };
1076 
1077 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1078 	.ctr	   = crypt_iv_lmk_ctr,
1079 	.dtr	   = crypt_iv_lmk_dtr,
1080 	.init	   = crypt_iv_lmk_init,
1081 	.wipe	   = crypt_iv_lmk_wipe,
1082 	.generator = crypt_iv_lmk_gen,
1083 	.post	   = crypt_iv_lmk_post
1084 };
1085 
1086 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1087 	.ctr	   = crypt_iv_tcw_ctr,
1088 	.dtr	   = crypt_iv_tcw_dtr,
1089 	.init	   = crypt_iv_tcw_init,
1090 	.wipe	   = crypt_iv_tcw_wipe,
1091 	.generator = crypt_iv_tcw_gen,
1092 	.post	   = crypt_iv_tcw_post
1093 };
1094 
1095 static const struct crypt_iv_operations crypt_iv_random_ops = {
1096 	.generator = crypt_iv_random_gen
1097 };
1098 
1099 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1100 	.ctr	   = crypt_iv_eboiv_ctr,
1101 	.generator = crypt_iv_eboiv_gen
1102 };
1103 
1104 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1105 	.ctr	   = crypt_iv_elephant_ctr,
1106 	.dtr	   = crypt_iv_elephant_dtr,
1107 	.init	   = crypt_iv_elephant_init,
1108 	.wipe	   = crypt_iv_elephant_wipe,
1109 	.generator = crypt_iv_elephant_gen,
1110 	.post	   = crypt_iv_elephant_post
1111 };
1112 
1113 /*
1114  * Integrity extensions
1115  */
1116 static bool crypt_integrity_aead(struct crypt_config *cc)
1117 {
1118 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1119 }
1120 
1121 static bool crypt_integrity_hmac(struct crypt_config *cc)
1122 {
1123 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1124 }
1125 
1126 /* Get sg containing data */
1127 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1128 					     struct scatterlist *sg)
1129 {
1130 	if (unlikely(crypt_integrity_aead(cc)))
1131 		return &sg[2];
1132 
1133 	return sg;
1134 }
1135 
1136 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1137 {
1138 	struct bio_integrity_payload *bip;
1139 	unsigned int tag_len;
1140 	int ret;
1141 
1142 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1143 		return 0;
1144 
1145 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1146 	if (IS_ERR(bip))
1147 		return PTR_ERR(bip);
1148 
1149 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1150 
1151 	bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
1152 
1153 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1154 				     tag_len, offset_in_page(io->integrity_metadata));
1155 	if (unlikely(ret != tag_len))
1156 		return -ENOMEM;
1157 
1158 	return 0;
1159 }
1160 
1161 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1162 {
1163 #ifdef CONFIG_BLK_DEV_INTEGRITY
1164 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1165 	struct mapped_device *md = dm_table_get_md(ti->table);
1166 
1167 	/* We require an underlying device with non-PI metadata */
1168 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1169 		ti->error = "Integrity profile not supported.";
1170 		return -EINVAL;
1171 	}
1172 
1173 	if (bi->metadata_size < cc->used_tag_size) {
1174 		ti->error = "Integrity profile tag size mismatch.";
1175 		return -EINVAL;
1176 	}
1177 	cc->tuple_size = bi->metadata_size;
1178 	if (1 << bi->interval_exp != cc->sector_size) {
1179 		ti->error = "Integrity profile sector size mismatch.";
1180 		return -EINVAL;
1181 	}
1182 
1183 	if (crypt_integrity_aead(cc)) {
1184 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1185 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1186 		       cc->integrity_tag_size, cc->integrity_iv_size);
1187 
1188 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1189 			ti->error = "Integrity AEAD auth tag size is not supported.";
1190 			return -EINVAL;
1191 		}
1192 	} else if (cc->integrity_iv_size)
1193 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1194 		       cc->integrity_iv_size);
1195 
1196 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1197 		ti->error = "Not enough space for integrity tag in the profile.";
1198 		return -EINVAL;
1199 	}
1200 
1201 	return 0;
1202 #else
1203 	ti->error = "Integrity profile not supported.";
1204 	return -EINVAL;
1205 #endif
1206 }
1207 
1208 static void crypt_convert_init(struct crypt_config *cc,
1209 			       struct convert_context *ctx,
1210 			       struct bio *bio_out, struct bio *bio_in,
1211 			       sector_t sector)
1212 {
1213 	ctx->bio_in = bio_in;
1214 	ctx->bio_out = bio_out;
1215 	if (bio_in)
1216 		ctx->iter_in = bio_in->bi_iter;
1217 	if (bio_out)
1218 		ctx->iter_out = bio_out->bi_iter;
1219 	ctx->cc_sector = sector + cc->iv_offset;
1220 	ctx->tag_offset = 0;
1221 	init_completion(&ctx->restart);
1222 }
1223 
1224 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1225 					     void *req)
1226 {
1227 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1228 }
1229 
1230 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1231 {
1232 	return (void *)((char *)dmreq - cc->dmreq_start);
1233 }
1234 
1235 static u8 *iv_of_dmreq(struct crypt_config *cc,
1236 		       struct dm_crypt_request *dmreq)
1237 {
1238 	if (crypt_integrity_aead(cc))
1239 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1240 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1241 	else
1242 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1243 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1244 }
1245 
1246 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1247 		       struct dm_crypt_request *dmreq)
1248 {
1249 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1250 }
1251 
1252 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1253 		       struct dm_crypt_request *dmreq)
1254 {
1255 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1256 
1257 	return (__le64 *) ptr;
1258 }
1259 
1260 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1261 		       struct dm_crypt_request *dmreq)
1262 {
1263 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1264 		  cc->iv_size + sizeof(uint64_t);
1265 
1266 	return (unsigned int *)ptr;
1267 }
1268 
1269 static void *tag_from_dmreq(struct crypt_config *cc,
1270 				struct dm_crypt_request *dmreq)
1271 {
1272 	struct convert_context *ctx = dmreq->ctx;
1273 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1274 
1275 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1276 		cc->tuple_size];
1277 }
1278 
1279 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1280 			       struct dm_crypt_request *dmreq)
1281 {
1282 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1283 }
1284 
1285 static int crypt_convert_block_aead(struct crypt_config *cc,
1286 				     struct convert_context *ctx,
1287 				     struct aead_request *req,
1288 				     unsigned int tag_offset)
1289 {
1290 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1291 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1292 	struct dm_crypt_request *dmreq;
1293 	u8 *iv, *org_iv, *tag_iv, *tag;
1294 	__le64 *sector;
1295 	int r = 0;
1296 
1297 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1298 
1299 	/* Reject unexpected unaligned bio. */
1300 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1301 		return -EIO;
1302 
1303 	dmreq = dmreq_of_req(cc, req);
1304 	dmreq->iv_sector = ctx->cc_sector;
1305 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1306 		dmreq->iv_sector >>= cc->sector_shift;
1307 	dmreq->ctx = ctx;
1308 
1309 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1310 
1311 	sector = org_sector_of_dmreq(cc, dmreq);
1312 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1313 
1314 	iv = iv_of_dmreq(cc, dmreq);
1315 	org_iv = org_iv_of_dmreq(cc, dmreq);
1316 	tag = tag_from_dmreq(cc, dmreq);
1317 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1318 
1319 	/* AEAD request:
1320 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1321 	 *  | (authenticated) | (auth+encryption) |              |
1322 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1323 	 */
1324 	sg_init_table(dmreq->sg_in, 4);
1325 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1326 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1327 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1328 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1329 
1330 	sg_init_table(dmreq->sg_out, 4);
1331 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1332 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1333 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1334 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1335 
1336 	if (cc->iv_gen_ops) {
1337 		/* For READs use IV stored in integrity metadata */
1338 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1339 			memcpy(org_iv, tag_iv, cc->iv_size);
1340 		} else {
1341 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1342 			if (r < 0)
1343 				return r;
1344 			/* Store generated IV in integrity metadata */
1345 			if (cc->integrity_iv_size)
1346 				memcpy(tag_iv, org_iv, cc->iv_size);
1347 		}
1348 		/* Working copy of IV, to be modified in crypto API */
1349 		memcpy(iv, org_iv, cc->iv_size);
1350 	}
1351 
1352 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1353 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1354 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1355 				       cc->sector_size, iv);
1356 		r = crypto_aead_encrypt(req);
1357 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1358 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1359 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1360 	} else {
1361 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1362 				       cc->sector_size + cc->integrity_tag_size, iv);
1363 		r = crypto_aead_decrypt(req);
1364 	}
1365 
1366 	if (r == -EBADMSG) {
1367 		sector_t s = le64_to_cpu(*sector);
1368 
1369 		ctx->aead_failed = true;
1370 		if (ctx->aead_recheck) {
1371 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1372 				    ctx->bio_in->bi_bdev, s);
1373 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1374 					 ctx->bio_in, s, 0);
1375 		}
1376 	}
1377 
1378 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1379 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1380 
1381 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1382 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1383 
1384 	return r;
1385 }
1386 
1387 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1388 					struct convert_context *ctx,
1389 					struct skcipher_request *req,
1390 					unsigned int tag_offset)
1391 {
1392 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1393 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1394 	struct scatterlist *sg_in, *sg_out;
1395 	struct dm_crypt_request *dmreq;
1396 	u8 *iv, *org_iv, *tag_iv;
1397 	__le64 *sector;
1398 	int r = 0;
1399 
1400 	/* Reject unexpected unaligned bio. */
1401 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1402 		return -EIO;
1403 
1404 	dmreq = dmreq_of_req(cc, req);
1405 	dmreq->iv_sector = ctx->cc_sector;
1406 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1407 		dmreq->iv_sector >>= cc->sector_shift;
1408 	dmreq->ctx = ctx;
1409 
1410 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1411 
1412 	iv = iv_of_dmreq(cc, dmreq);
1413 	org_iv = org_iv_of_dmreq(cc, dmreq);
1414 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1415 
1416 	sector = org_sector_of_dmreq(cc, dmreq);
1417 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1418 
1419 	/* For skcipher we use only the first sg item */
1420 	sg_in  = &dmreq->sg_in[0];
1421 	sg_out = &dmreq->sg_out[0];
1422 
1423 	sg_init_table(sg_in, 1);
1424 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1425 
1426 	sg_init_table(sg_out, 1);
1427 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1428 
1429 	if (cc->iv_gen_ops) {
1430 		/* For READs use IV stored in integrity metadata */
1431 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1432 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1433 		} else {
1434 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1435 			if (r < 0)
1436 				return r;
1437 			/* Data can be already preprocessed in generator */
1438 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1439 				sg_in = sg_out;
1440 			/* Store generated IV in integrity metadata */
1441 			if (cc->integrity_iv_size)
1442 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1443 		}
1444 		/* Working copy of IV, to be modified in crypto API */
1445 		memcpy(iv, org_iv, cc->iv_size);
1446 	}
1447 
1448 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1449 
1450 	if (bio_data_dir(ctx->bio_in) == WRITE)
1451 		r = crypto_skcipher_encrypt(req);
1452 	else
1453 		r = crypto_skcipher_decrypt(req);
1454 
1455 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1456 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1457 
1458 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1459 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1460 
1461 	return r;
1462 }
1463 
1464 static void kcryptd_async_done(void *async_req, int error);
1465 
1466 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1467 				     struct convert_context *ctx)
1468 {
1469 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1470 
1471 	if (!ctx->r.req) {
1472 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1473 		if (!ctx->r.req)
1474 			return -ENOMEM;
1475 	}
1476 
1477 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1478 
1479 	/*
1480 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1481 	 * requests if driver request queue is full.
1482 	 */
1483 	skcipher_request_set_callback(ctx->r.req,
1484 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1485 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1486 
1487 	return 0;
1488 }
1489 
1490 static int crypt_alloc_req_aead(struct crypt_config *cc,
1491 				 struct convert_context *ctx)
1492 {
1493 	if (!ctx->r.req_aead) {
1494 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1495 		if (!ctx->r.req_aead)
1496 			return -ENOMEM;
1497 	}
1498 
1499 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1500 
1501 	/*
1502 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1503 	 * requests if driver request queue is full.
1504 	 */
1505 	aead_request_set_callback(ctx->r.req_aead,
1506 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1507 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1508 
1509 	return 0;
1510 }
1511 
1512 static int crypt_alloc_req(struct crypt_config *cc,
1513 			    struct convert_context *ctx)
1514 {
1515 	if (crypt_integrity_aead(cc))
1516 		return crypt_alloc_req_aead(cc, ctx);
1517 	else
1518 		return crypt_alloc_req_skcipher(cc, ctx);
1519 }
1520 
1521 static void crypt_free_req_skcipher(struct crypt_config *cc,
1522 				    struct skcipher_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 skcipher_request *)(io + 1) != req)
1527 		mempool_free(req, &cc->req_pool);
1528 }
1529 
1530 static void crypt_free_req_aead(struct crypt_config *cc,
1531 				struct aead_request *req, struct bio *base_bio)
1532 {
1533 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1534 
1535 	if ((struct aead_request *)(io + 1) != req)
1536 		mempool_free(req, &cc->req_pool);
1537 }
1538 
1539 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1540 {
1541 	if (crypt_integrity_aead(cc))
1542 		crypt_free_req_aead(cc, req, base_bio);
1543 	else
1544 		crypt_free_req_skcipher(cc, req, base_bio);
1545 }
1546 
1547 /*
1548  * Encrypt / decrypt data from one bio to another one (can be the same one)
1549  */
1550 static blk_status_t crypt_convert(struct crypt_config *cc,
1551 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1552 {
1553 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1554 	int r;
1555 
1556 	/*
1557 	 * if reset_pending is set we are dealing with the bio for the first time,
1558 	 * else we're continuing to work on the previous bio, so don't mess with
1559 	 * the cc_pending counter
1560 	 */
1561 	if (reset_pending)
1562 		atomic_set(&ctx->cc_pending, 1);
1563 
1564 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1565 
1566 		r = crypt_alloc_req(cc, ctx);
1567 		if (r) {
1568 			complete(&ctx->restart);
1569 			return BLK_STS_DEV_RESOURCE;
1570 		}
1571 
1572 		atomic_inc(&ctx->cc_pending);
1573 
1574 		if (crypt_integrity_aead(cc))
1575 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, ctx->tag_offset);
1576 		else
1577 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->tag_offset);
1578 
1579 		switch (r) {
1580 		/*
1581 		 * The request was queued by a crypto driver
1582 		 * but the driver request queue is full, let's wait.
1583 		 */
1584 		case -EBUSY:
1585 			if (in_interrupt()) {
1586 				if (try_wait_for_completion(&ctx->restart)) {
1587 					/*
1588 					 * we don't have to block to wait for completion,
1589 					 * so proceed
1590 					 */
1591 				} else {
1592 					/*
1593 					 * we can't wait for completion without blocking
1594 					 * exit and continue processing in a workqueue
1595 					 */
1596 					ctx->r.req = NULL;
1597 					ctx->tag_offset++;
1598 					ctx->cc_sector += sector_step;
1599 					return BLK_STS_DEV_RESOURCE;
1600 				}
1601 			} else {
1602 				wait_for_completion(&ctx->restart);
1603 			}
1604 			reinit_completion(&ctx->restart);
1605 			fallthrough;
1606 		/*
1607 		 * The request is queued and processed asynchronously,
1608 		 * completion function kcryptd_async_done() will be called.
1609 		 */
1610 		case -EINPROGRESS:
1611 			ctx->r.req = NULL;
1612 			ctx->tag_offset++;
1613 			ctx->cc_sector += sector_step;
1614 			continue;
1615 		/*
1616 		 * The request was already processed (synchronously).
1617 		 */
1618 		case 0:
1619 			atomic_dec(&ctx->cc_pending);
1620 			ctx->cc_sector += sector_step;
1621 			ctx->tag_offset++;
1622 			if (!atomic)
1623 				cond_resched();
1624 			continue;
1625 		/*
1626 		 * There was a data integrity error.
1627 		 */
1628 		case -EBADMSG:
1629 			atomic_dec(&ctx->cc_pending);
1630 			return BLK_STS_PROTECTION;
1631 		/*
1632 		 * There was an error while processing the request.
1633 		 */
1634 		default:
1635 			atomic_dec(&ctx->cc_pending);
1636 			return BLK_STS_IOERR;
1637 		}
1638 	}
1639 
1640 	return 0;
1641 }
1642 
1643 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1644 
1645 /*
1646  * Generate a new unfragmented bio with the given size
1647  * This should never violate the device limitations (but if it did then block
1648  * core should split the bio as needed).
1649  *
1650  * This function may be called concurrently. If we allocate from the mempool
1651  * concurrently, there is a possibility of deadlock. For example, if we have
1652  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1653  * the mempool concurrently, it may deadlock in a situation where both processes
1654  * have allocated 128 pages and the mempool is exhausted.
1655  *
1656  * In order to avoid this scenario we allocate the pages under a mutex.
1657  *
1658  * In order to not degrade performance with excessive locking, we try
1659  * non-blocking allocations without a mutex first but on failure we fallback
1660  * to blocking allocations with a mutex.
1661  *
1662  * In order to reduce allocation overhead, we try to allocate compound pages in
1663  * the first pass. If they are not available, we fall back to the mempool.
1664  */
1665 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1666 {
1667 	struct crypt_config *cc = io->cc;
1668 	struct bio *clone;
1669 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1670 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1671 	unsigned int remaining_size;
1672 	unsigned int order = MAX_PAGE_ORDER;
1673 
1674 retry:
1675 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1676 		mutex_lock(&cc->bio_alloc_lock);
1677 
1678 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1679 				 GFP_NOIO, &cc->bs);
1680 	clone->bi_private = io;
1681 	clone->bi_end_io = crypt_endio;
1682 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1683 	clone->bi_iter.bi_sector = cc->start + io->sector;
1684 
1685 	remaining_size = size;
1686 
1687 	while (remaining_size) {
1688 		struct page *pages;
1689 		unsigned size_to_add;
1690 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1691 		order = min(order, remaining_order);
1692 
1693 		while (order > 0) {
1694 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1695 					(1 << order) > dm_crypt_pages_per_client))
1696 				goto decrease_order;
1697 			pages = alloc_pages(gfp_mask
1698 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1699 				order);
1700 			if (likely(pages != NULL)) {
1701 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1702 				goto have_pages;
1703 			}
1704 decrease_order:
1705 			order--;
1706 		}
1707 
1708 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1709 		if (!pages) {
1710 			crypt_free_buffer_pages(cc, clone);
1711 			bio_put(clone);
1712 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1713 			order = 0;
1714 			goto retry;
1715 		}
1716 
1717 have_pages:
1718 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1719 		__bio_add_page(clone, pages, size_to_add, 0);
1720 		remaining_size -= size_to_add;
1721 	}
1722 
1723 	/* Allocate space for integrity tags */
1724 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1725 		crypt_free_buffer_pages(cc, clone);
1726 		bio_put(clone);
1727 		clone = NULL;
1728 	}
1729 
1730 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1731 		mutex_unlock(&cc->bio_alloc_lock);
1732 
1733 	return clone;
1734 }
1735 
1736 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1737 {
1738 	struct folio_iter fi;
1739 
1740 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1741 		bio_for_each_folio_all(fi, clone) {
1742 			if (folio_test_large(fi.folio)) {
1743 				percpu_counter_sub(&cc->n_allocated_pages,
1744 						folio_nr_pages(fi.folio));
1745 				folio_put(fi.folio);
1746 			} else {
1747 				mempool_free(&fi.folio->page, &cc->page_pool);
1748 			}
1749 		}
1750 	}
1751 }
1752 
1753 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1754 			  struct bio *bio, sector_t sector)
1755 {
1756 	io->cc = cc;
1757 	io->base_bio = bio;
1758 	io->sector = sector;
1759 	io->error = 0;
1760 	io->ctx.aead_recheck = false;
1761 	io->ctx.aead_failed = false;
1762 	io->ctx.r.req = NULL;
1763 	io->integrity_metadata = NULL;
1764 	io->integrity_metadata_from_pool = false;
1765 	atomic_set(&io->io_pending, 0);
1766 }
1767 
1768 static void crypt_inc_pending(struct dm_crypt_io *io)
1769 {
1770 	atomic_inc(&io->io_pending);
1771 }
1772 
1773 static void kcryptd_queue_read(struct dm_crypt_io *io);
1774 
1775 /*
1776  * One of the bios was finished. Check for completion of
1777  * the whole request and correctly clean up the buffer.
1778  */
1779 static void crypt_dec_pending(struct dm_crypt_io *io)
1780 {
1781 	struct crypt_config *cc = io->cc;
1782 	struct bio *base_bio = io->base_bio;
1783 	blk_status_t error = io->error;
1784 
1785 	if (!atomic_dec_and_test(&io->io_pending))
1786 		return;
1787 
1788 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1789 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1790 		io->ctx.aead_recheck = true;
1791 		io->ctx.aead_failed = false;
1792 		io->error = 0;
1793 		kcryptd_queue_read(io);
1794 		return;
1795 	}
1796 
1797 	if (io->ctx.r.req)
1798 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1799 
1800 	if (unlikely(io->integrity_metadata_from_pool))
1801 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1802 	else
1803 		kfree(io->integrity_metadata);
1804 
1805 	base_bio->bi_status = error;
1806 
1807 	bio_endio(base_bio);
1808 }
1809 
1810 /*
1811  * kcryptd/kcryptd_io:
1812  *
1813  * Needed because it would be very unwise to do decryption in an
1814  * interrupt context.
1815  *
1816  * kcryptd performs the actual encryption or decryption.
1817  *
1818  * kcryptd_io performs the IO submission.
1819  *
1820  * They must be separated as otherwise the final stages could be
1821  * starved by new requests which can block in the first stages due
1822  * to memory allocation.
1823  *
1824  * The work is done per CPU global for all dm-crypt instances.
1825  * They should not depend on each other and do not block.
1826  */
1827 static void crypt_endio(struct bio *clone)
1828 {
1829 	struct dm_crypt_io *io = clone->bi_private;
1830 	struct crypt_config *cc = io->cc;
1831 	unsigned int rw = bio_data_dir(clone);
1832 	blk_status_t error = clone->bi_status;
1833 
1834 	if (io->ctx.aead_recheck && !error) {
1835 		kcryptd_queue_crypt(io);
1836 		return;
1837 	}
1838 
1839 	/*
1840 	 * free the processed pages
1841 	 */
1842 	if (rw == WRITE || io->ctx.aead_recheck)
1843 		crypt_free_buffer_pages(cc, clone);
1844 
1845 	bio_put(clone);
1846 
1847 	if (rw == READ && !error) {
1848 		kcryptd_queue_crypt(io);
1849 		return;
1850 	}
1851 
1852 	if (unlikely(error))
1853 		io->error = error;
1854 
1855 	crypt_dec_pending(io);
1856 }
1857 
1858 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1859 
1860 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1861 {
1862 	struct crypt_config *cc = io->cc;
1863 	struct bio *clone;
1864 
1865 	if (io->ctx.aead_recheck) {
1866 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1867 			return 1;
1868 		crypt_inc_pending(io);
1869 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1870 		if (unlikely(!clone)) {
1871 			crypt_dec_pending(io);
1872 			return 1;
1873 		}
1874 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1875 		io->saved_bi_iter = clone->bi_iter;
1876 		dm_submit_bio_remap(io->base_bio, clone);
1877 		return 0;
1878 	}
1879 
1880 	/*
1881 	 * We need the original biovec array in order to decrypt the whole bio
1882 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1883 	 * worry about the block layer modifying the biovec array; so leverage
1884 	 * bio_alloc_clone().
1885 	 */
1886 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1887 	if (!clone)
1888 		return 1;
1889 
1890 	clone->bi_iter.bi_sector = cc->start + io->sector;
1891 	clone->bi_private = io;
1892 	clone->bi_end_io = crypt_endio;
1893 
1894 	crypt_inc_pending(io);
1895 
1896 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1897 		crypt_dec_pending(io);
1898 		bio_put(clone);
1899 		return 1;
1900 	}
1901 
1902 	dm_submit_bio_remap(io->base_bio, clone);
1903 	return 0;
1904 }
1905 
1906 static void kcryptd_io_read_work(struct work_struct *work)
1907 {
1908 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1909 
1910 	crypt_inc_pending(io);
1911 	if (kcryptd_io_read(io, GFP_NOIO))
1912 		io->error = BLK_STS_RESOURCE;
1913 	crypt_dec_pending(io);
1914 }
1915 
1916 static void kcryptd_queue_read(struct dm_crypt_io *io)
1917 {
1918 	struct crypt_config *cc = io->cc;
1919 
1920 	INIT_WORK(&io->work, kcryptd_io_read_work);
1921 	queue_work(cc->io_queue, &io->work);
1922 }
1923 
1924 static void kcryptd_io_write(struct dm_crypt_io *io)
1925 {
1926 	struct bio *clone = io->ctx.bio_out;
1927 
1928 	dm_submit_bio_remap(io->base_bio, clone);
1929 }
1930 
1931 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1932 
1933 static int dmcrypt_write(void *data)
1934 {
1935 	struct crypt_config *cc = data;
1936 	struct dm_crypt_io *io;
1937 
1938 	while (1) {
1939 		struct rb_root write_tree;
1940 		struct blk_plug plug;
1941 
1942 		spin_lock_irq(&cc->write_thread_lock);
1943 continue_locked:
1944 
1945 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1946 			goto pop_from_list;
1947 
1948 		set_current_state(TASK_INTERRUPTIBLE);
1949 
1950 		spin_unlock_irq(&cc->write_thread_lock);
1951 
1952 		if (unlikely(kthread_should_stop())) {
1953 			set_current_state(TASK_RUNNING);
1954 			break;
1955 		}
1956 
1957 		schedule();
1958 
1959 		spin_lock_irq(&cc->write_thread_lock);
1960 		goto continue_locked;
1961 
1962 pop_from_list:
1963 		write_tree = cc->write_tree;
1964 		cc->write_tree = RB_ROOT;
1965 		spin_unlock_irq(&cc->write_thread_lock);
1966 
1967 		BUG_ON(rb_parent(write_tree.rb_node));
1968 
1969 		/*
1970 		 * Note: we cannot walk the tree here with rb_next because
1971 		 * the structures may be freed when kcryptd_io_write is called.
1972 		 */
1973 		blk_start_plug(&plug);
1974 		do {
1975 			io = crypt_io_from_node(rb_first(&write_tree));
1976 			rb_erase(&io->rb_node, &write_tree);
1977 			kcryptd_io_write(io);
1978 			cond_resched();
1979 		} while (!RB_EMPTY_ROOT(&write_tree));
1980 		blk_finish_plug(&plug);
1981 	}
1982 	return 0;
1983 }
1984 
1985 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1986 {
1987 	struct bio *clone = io->ctx.bio_out;
1988 	struct crypt_config *cc = io->cc;
1989 	unsigned long flags;
1990 	sector_t sector;
1991 	struct rb_node **rbp, *parent;
1992 
1993 	if (unlikely(io->error)) {
1994 		crypt_free_buffer_pages(cc, clone);
1995 		bio_put(clone);
1996 		crypt_dec_pending(io);
1997 		return;
1998 	}
1999 
2000 	/* crypt_convert should have filled the clone bio */
2001 	BUG_ON(io->ctx.iter_out.bi_size);
2002 
2003 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2004 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2005 		dm_submit_bio_remap(io->base_bio, clone);
2006 		return;
2007 	}
2008 
2009 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2010 	if (RB_EMPTY_ROOT(&cc->write_tree))
2011 		wake_up_process(cc->write_thread);
2012 	rbp = &cc->write_tree.rb_node;
2013 	parent = NULL;
2014 	sector = io->sector;
2015 	while (*rbp) {
2016 		parent = *rbp;
2017 		if (sector < crypt_io_from_node(parent)->sector)
2018 			rbp = &(*rbp)->rb_left;
2019 		else
2020 			rbp = &(*rbp)->rb_right;
2021 	}
2022 	rb_link_node(&io->rb_node, parent, rbp);
2023 	rb_insert_color(&io->rb_node, &cc->write_tree);
2024 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2025 }
2026 
2027 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2028 				       struct convert_context *ctx)
2029 
2030 {
2031 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2032 		return false;
2033 
2034 	/*
2035 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2036 	 * constraints so they do not need to be issued inline by
2037 	 * kcryptd_crypt_write_convert().
2038 	 */
2039 	switch (bio_op(ctx->bio_in)) {
2040 	case REQ_OP_WRITE:
2041 	case REQ_OP_WRITE_ZEROES:
2042 		return true;
2043 	default:
2044 		return false;
2045 	}
2046 }
2047 
2048 static void kcryptd_crypt_write_continue(struct work_struct *work)
2049 {
2050 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2051 	struct crypt_config *cc = io->cc;
2052 	struct convert_context *ctx = &io->ctx;
2053 	int crypt_finished;
2054 	blk_status_t r;
2055 
2056 	wait_for_completion(&ctx->restart);
2057 	reinit_completion(&ctx->restart);
2058 
2059 	r = crypt_convert(cc, &io->ctx, false, false);
2060 	if (r)
2061 		io->error = r;
2062 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2063 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2064 		/* Wait for completion signaled by kcryptd_async_done() */
2065 		wait_for_completion(&ctx->restart);
2066 		crypt_finished = 1;
2067 	}
2068 
2069 	/* Encryption was already finished, submit io now */
2070 	if (crypt_finished)
2071 		kcryptd_crypt_write_io_submit(io, 0);
2072 
2073 	crypt_dec_pending(io);
2074 }
2075 
2076 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2077 {
2078 	struct crypt_config *cc = io->cc;
2079 	struct convert_context *ctx = &io->ctx;
2080 	struct bio *clone;
2081 	int crypt_finished;
2082 	blk_status_t r;
2083 
2084 	/*
2085 	 * Prevent io from disappearing until this function completes.
2086 	 */
2087 	crypt_inc_pending(io);
2088 	crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector);
2089 
2090 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2091 	if (unlikely(!clone)) {
2092 		io->error = BLK_STS_IOERR;
2093 		goto dec;
2094 	}
2095 
2096 	io->ctx.bio_out = clone;
2097 	io->ctx.iter_out = clone->bi_iter;
2098 
2099 	if (crypt_integrity_aead(cc)) {
2100 		bio_copy_data(clone, io->base_bio);
2101 		io->ctx.bio_in = clone;
2102 		io->ctx.iter_in = clone->bi_iter;
2103 	}
2104 
2105 	crypt_inc_pending(io);
2106 	r = crypt_convert(cc, ctx,
2107 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2108 	/*
2109 	 * Crypto API backlogged the request, because its queue was full
2110 	 * and we're in softirq context, so continue from a workqueue
2111 	 * (TODO: is it actually possible to be in softirq in the write path?)
2112 	 */
2113 	if (r == BLK_STS_DEV_RESOURCE) {
2114 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2115 		queue_work(cc->crypt_queue, &io->work);
2116 		return;
2117 	}
2118 	if (r)
2119 		io->error = r;
2120 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2121 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2122 		/* Wait for completion signaled by kcryptd_async_done() */
2123 		wait_for_completion(&ctx->restart);
2124 		crypt_finished = 1;
2125 	}
2126 
2127 	/* Encryption was already finished, submit io now */
2128 	if (crypt_finished)
2129 		kcryptd_crypt_write_io_submit(io, 0);
2130 
2131 dec:
2132 	crypt_dec_pending(io);
2133 }
2134 
2135 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2136 {
2137 	if (io->ctx.aead_recheck) {
2138 		if (!io->error) {
2139 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2140 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2141 		}
2142 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2143 		bio_put(io->ctx.bio_in);
2144 	}
2145 	crypt_dec_pending(io);
2146 }
2147 
2148 static void kcryptd_crypt_read_continue(struct work_struct *work)
2149 {
2150 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2151 	struct crypt_config *cc = io->cc;
2152 	blk_status_t r;
2153 
2154 	wait_for_completion(&io->ctx.restart);
2155 	reinit_completion(&io->ctx.restart);
2156 
2157 	r = crypt_convert(cc, &io->ctx, false, false);
2158 	if (r)
2159 		io->error = r;
2160 
2161 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2162 		kcryptd_crypt_read_done(io);
2163 
2164 	crypt_dec_pending(io);
2165 }
2166 
2167 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2168 {
2169 	struct crypt_config *cc = io->cc;
2170 	blk_status_t r;
2171 
2172 	crypt_inc_pending(io);
2173 
2174 	if (io->ctx.aead_recheck) {
2175 		r = crypt_convert(cc, &io->ctx,
2176 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2177 	} else {
2178 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2179 				   io->sector);
2180 
2181 		r = crypt_convert(cc, &io->ctx,
2182 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2183 	}
2184 	/*
2185 	 * Crypto API backlogged the request, because its queue was full
2186 	 * and we're in softirq context, so continue from a workqueue
2187 	 */
2188 	if (r == BLK_STS_DEV_RESOURCE) {
2189 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2190 		queue_work(cc->crypt_queue, &io->work);
2191 		return;
2192 	}
2193 	if (r)
2194 		io->error = r;
2195 
2196 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2197 		kcryptd_crypt_read_done(io);
2198 
2199 	crypt_dec_pending(io);
2200 }
2201 
2202 static void kcryptd_async_done(void *data, int error)
2203 {
2204 	struct dm_crypt_request *dmreq = data;
2205 	struct convert_context *ctx = dmreq->ctx;
2206 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2207 	struct crypt_config *cc = io->cc;
2208 
2209 	/*
2210 	 * A request from crypto driver backlog is going to be processed now,
2211 	 * finish the completion and continue in crypt_convert().
2212 	 * (Callback will be called for the second time for this request.)
2213 	 */
2214 	if (error == -EINPROGRESS) {
2215 		complete(&ctx->restart);
2216 		return;
2217 	}
2218 
2219 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2220 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2221 
2222 	if (error == -EBADMSG) {
2223 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2224 
2225 		ctx->aead_failed = true;
2226 		if (ctx->aead_recheck) {
2227 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2228 				    ctx->bio_in->bi_bdev, s);
2229 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2230 					 ctx->bio_in, s, 0);
2231 		}
2232 		io->error = BLK_STS_PROTECTION;
2233 	} else if (error < 0)
2234 		io->error = BLK_STS_IOERR;
2235 
2236 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2237 
2238 	if (!atomic_dec_and_test(&ctx->cc_pending))
2239 		return;
2240 
2241 	/*
2242 	 * The request is fully completed: for inline writes, let
2243 	 * kcryptd_crypt_write_convert() do the IO submission.
2244 	 */
2245 	if (bio_data_dir(io->base_bio) == READ) {
2246 		kcryptd_crypt_read_done(io);
2247 		return;
2248 	}
2249 
2250 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2251 		complete(&ctx->restart);
2252 		return;
2253 	}
2254 
2255 	kcryptd_crypt_write_io_submit(io, 1);
2256 }
2257 
2258 static void kcryptd_crypt(struct work_struct *work)
2259 {
2260 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2261 
2262 	if (bio_data_dir(io->base_bio) == READ)
2263 		kcryptd_crypt_read_convert(io);
2264 	else
2265 		kcryptd_crypt_write_convert(io);
2266 }
2267 
2268 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2269 {
2270 	struct crypt_config *cc = io->cc;
2271 
2272 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2273 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2274 		/*
2275 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2276 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2277 		 * it is being executed with irqs disabled.
2278 		 */
2279 		if (in_hardirq() || irqs_disabled()) {
2280 			INIT_WORK(&io->work, kcryptd_crypt);
2281 			queue_work(system_bh_wq, &io->work);
2282 			return;
2283 		} else {
2284 			kcryptd_crypt(&io->work);
2285 			return;
2286 		}
2287 	}
2288 
2289 	INIT_WORK(&io->work, kcryptd_crypt);
2290 	queue_work(cc->crypt_queue, &io->work);
2291 }
2292 
2293 static void crypt_free_tfms_aead(struct crypt_config *cc)
2294 {
2295 	if (!cc->cipher_tfm.tfms_aead)
2296 		return;
2297 
2298 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2299 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2300 		cc->cipher_tfm.tfms_aead[0] = NULL;
2301 	}
2302 
2303 	kfree(cc->cipher_tfm.tfms_aead);
2304 	cc->cipher_tfm.tfms_aead = NULL;
2305 }
2306 
2307 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2308 {
2309 	unsigned int i;
2310 
2311 	if (!cc->cipher_tfm.tfms)
2312 		return;
2313 
2314 	for (i = 0; i < cc->tfms_count; i++)
2315 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2316 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2317 			cc->cipher_tfm.tfms[i] = NULL;
2318 		}
2319 
2320 	kfree(cc->cipher_tfm.tfms);
2321 	cc->cipher_tfm.tfms = NULL;
2322 }
2323 
2324 static void crypt_free_tfms(struct crypt_config *cc)
2325 {
2326 	if (crypt_integrity_aead(cc))
2327 		crypt_free_tfms_aead(cc);
2328 	else
2329 		crypt_free_tfms_skcipher(cc);
2330 }
2331 
2332 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2333 {
2334 	unsigned int i;
2335 	int err;
2336 
2337 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2338 				      sizeof(struct crypto_skcipher *),
2339 				      GFP_KERNEL);
2340 	if (!cc->cipher_tfm.tfms)
2341 		return -ENOMEM;
2342 
2343 	for (i = 0; i < cc->tfms_count; i++) {
2344 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2345 						CRYPTO_ALG_ALLOCATES_MEMORY);
2346 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2347 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2348 			crypt_free_tfms(cc);
2349 			return err;
2350 		}
2351 	}
2352 
2353 	/*
2354 	 * dm-crypt performance can vary greatly depending on which crypto
2355 	 * algorithm implementation is used.  Help people debug performance
2356 	 * problems by logging the ->cra_driver_name.
2357 	 */
2358 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2359 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2360 	return 0;
2361 }
2362 
2363 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2364 {
2365 	int err;
2366 
2367 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2368 	if (!cc->cipher_tfm.tfms)
2369 		return -ENOMEM;
2370 
2371 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2372 						CRYPTO_ALG_ALLOCATES_MEMORY);
2373 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2374 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2375 		crypt_free_tfms(cc);
2376 		return err;
2377 	}
2378 
2379 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2380 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2381 	return 0;
2382 }
2383 
2384 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2385 {
2386 	if (crypt_integrity_aead(cc))
2387 		return crypt_alloc_tfms_aead(cc, ciphermode);
2388 	else
2389 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2390 }
2391 
2392 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2393 {
2394 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2395 }
2396 
2397 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2398 {
2399 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2400 }
2401 
2402 /*
2403  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2404  * the key must be for some reason in special format.
2405  * This funcion converts cc->key to this special format.
2406  */
2407 static void crypt_copy_authenckey(char *p, const void *key,
2408 				  unsigned int enckeylen, unsigned int authkeylen)
2409 {
2410 	struct crypto_authenc_key_param *param;
2411 	struct rtattr *rta;
2412 
2413 	rta = (struct rtattr *)p;
2414 	param = RTA_DATA(rta);
2415 	param->enckeylen = cpu_to_be32(enckeylen);
2416 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2417 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2418 	p += RTA_SPACE(sizeof(*param));
2419 	memcpy(p, key + enckeylen, authkeylen);
2420 	p += authkeylen;
2421 	memcpy(p, key, enckeylen);
2422 }
2423 
2424 static int crypt_setkey(struct crypt_config *cc)
2425 {
2426 	unsigned int subkey_size;
2427 	int err = 0, i, r;
2428 
2429 	/* Ignore extra keys (which are used for IV etc) */
2430 	subkey_size = crypt_subkey_size(cc);
2431 
2432 	if (crypt_integrity_hmac(cc)) {
2433 		if (subkey_size < cc->key_mac_size)
2434 			return -EINVAL;
2435 
2436 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2437 				      subkey_size - cc->key_mac_size,
2438 				      cc->key_mac_size);
2439 	}
2440 
2441 	for (i = 0; i < cc->tfms_count; i++) {
2442 		if (crypt_integrity_hmac(cc))
2443 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2444 				cc->authenc_key, crypt_authenckey_size(cc));
2445 		else if (crypt_integrity_aead(cc))
2446 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2447 					       cc->key + (i * subkey_size),
2448 					       subkey_size);
2449 		else
2450 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2451 						   cc->key + (i * subkey_size),
2452 						   subkey_size);
2453 		if (r)
2454 			err = r;
2455 	}
2456 
2457 	if (crypt_integrity_hmac(cc))
2458 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2459 
2460 	return err;
2461 }
2462 
2463 #ifdef CONFIG_KEYS
2464 
2465 static bool contains_whitespace(const char *str)
2466 {
2467 	while (*str)
2468 		if (isspace(*str++))
2469 			return true;
2470 	return false;
2471 }
2472 
2473 static int set_key_user(struct crypt_config *cc, struct key *key)
2474 {
2475 	const struct user_key_payload *ukp;
2476 
2477 	ukp = user_key_payload_locked(key);
2478 	if (!ukp)
2479 		return -EKEYREVOKED;
2480 
2481 	if (cc->key_size != ukp->datalen)
2482 		return -EINVAL;
2483 
2484 	memcpy(cc->key, ukp->data, cc->key_size);
2485 
2486 	return 0;
2487 }
2488 
2489 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2490 {
2491 	const struct encrypted_key_payload *ekp;
2492 
2493 	ekp = key->payload.data[0];
2494 	if (!ekp)
2495 		return -EKEYREVOKED;
2496 
2497 	if (cc->key_size != ekp->decrypted_datalen)
2498 		return -EINVAL;
2499 
2500 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2501 
2502 	return 0;
2503 }
2504 
2505 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2506 {
2507 	const struct trusted_key_payload *tkp;
2508 
2509 	tkp = key->payload.data[0];
2510 	if (!tkp)
2511 		return -EKEYREVOKED;
2512 
2513 	if (cc->key_size != tkp->key_len)
2514 		return -EINVAL;
2515 
2516 	memcpy(cc->key, tkp->key, cc->key_size);
2517 
2518 	return 0;
2519 }
2520 
2521 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2522 {
2523 	char *new_key_string, *key_desc;
2524 	int ret;
2525 	struct key_type *type;
2526 	struct key *key;
2527 	int (*set_key)(struct crypt_config *cc, struct key *key);
2528 
2529 	/*
2530 	 * Reject key_string with whitespace. dm core currently lacks code for
2531 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2532 	 */
2533 	if (contains_whitespace(key_string)) {
2534 		DMERR("whitespace chars not allowed in key string");
2535 		return -EINVAL;
2536 	}
2537 
2538 	/* look for next ':' separating key_type from key_description */
2539 	key_desc = strchr(key_string, ':');
2540 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2541 		return -EINVAL;
2542 
2543 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2544 		type = &key_type_logon;
2545 		set_key = set_key_user;
2546 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2547 		type = &key_type_user;
2548 		set_key = set_key_user;
2549 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2550 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2551 		type = &key_type_encrypted;
2552 		set_key = set_key_encrypted;
2553 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2554 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2555 		type = &key_type_trusted;
2556 		set_key = set_key_trusted;
2557 	} else {
2558 		return -EINVAL;
2559 	}
2560 
2561 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2562 	if (!new_key_string)
2563 		return -ENOMEM;
2564 
2565 	key = request_key(type, key_desc + 1, NULL);
2566 	if (IS_ERR(key)) {
2567 		ret = PTR_ERR(key);
2568 		goto free_new_key_string;
2569 	}
2570 
2571 	down_read(&key->sem);
2572 	ret = set_key(cc, key);
2573 	up_read(&key->sem);
2574 	key_put(key);
2575 	if (ret < 0)
2576 		goto free_new_key_string;
2577 
2578 	/* clear the flag since following operations may invalidate previously valid key */
2579 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2580 
2581 	ret = crypt_setkey(cc);
2582 	if (ret)
2583 		goto free_new_key_string;
2584 
2585 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2586 	kfree_sensitive(cc->key_string);
2587 	cc->key_string = new_key_string;
2588 	return 0;
2589 
2590 free_new_key_string:
2591 	kfree_sensitive(new_key_string);
2592 	return ret;
2593 }
2594 
2595 static int get_key_size(char **key_string)
2596 {
2597 	char *colon, dummy;
2598 	int ret;
2599 
2600 	if (*key_string[0] != ':')
2601 		return strlen(*key_string) >> 1;
2602 
2603 	/* look for next ':' in key string */
2604 	colon = strpbrk(*key_string + 1, ":");
2605 	if (!colon)
2606 		return -EINVAL;
2607 
2608 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2609 		return -EINVAL;
2610 
2611 	*key_string = colon;
2612 
2613 	/* remaining key string should be :<logon|user>:<key_desc> */
2614 
2615 	return ret;
2616 }
2617 
2618 #else
2619 
2620 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2621 {
2622 	return -EINVAL;
2623 }
2624 
2625 static int get_key_size(char **key_string)
2626 {
2627 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2628 }
2629 
2630 #endif /* CONFIG_KEYS */
2631 
2632 static int crypt_set_key(struct crypt_config *cc, char *key)
2633 {
2634 	int r = -EINVAL;
2635 	int key_string_len = strlen(key);
2636 
2637 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2638 	if (!cc->key_size && strcmp(key, "-"))
2639 		goto out;
2640 
2641 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2642 	if (key[0] == ':') {
2643 		r = crypt_set_keyring_key(cc, key + 1);
2644 		goto out;
2645 	}
2646 
2647 	/* clear the flag since following operations may invalidate previously valid key */
2648 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2649 
2650 	/* wipe references to any kernel keyring key */
2651 	kfree_sensitive(cc->key_string);
2652 	cc->key_string = NULL;
2653 
2654 	/* Decode key from its hex representation. */
2655 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2656 		goto out;
2657 
2658 	r = crypt_setkey(cc);
2659 	if (!r)
2660 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2661 
2662 out:
2663 	/* Hex key string not needed after here, so wipe it. */
2664 	memset(key, '0', key_string_len);
2665 
2666 	return r;
2667 }
2668 
2669 static int crypt_wipe_key(struct crypt_config *cc)
2670 {
2671 	int r;
2672 
2673 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2674 	get_random_bytes(&cc->key, cc->key_size);
2675 
2676 	/* Wipe IV private keys */
2677 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2678 		r = cc->iv_gen_ops->wipe(cc);
2679 		if (r)
2680 			return r;
2681 	}
2682 
2683 	kfree_sensitive(cc->key_string);
2684 	cc->key_string = NULL;
2685 	r = crypt_setkey(cc);
2686 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2687 
2688 	return r;
2689 }
2690 
2691 static void crypt_calculate_pages_per_client(void)
2692 {
2693 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2694 
2695 	if (!dm_crypt_clients_n)
2696 		return;
2697 
2698 	pages /= dm_crypt_clients_n;
2699 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2700 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2701 	dm_crypt_pages_per_client = pages;
2702 }
2703 
2704 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2705 {
2706 	struct crypt_config *cc = pool_data;
2707 	struct page *page;
2708 
2709 	/*
2710 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2711 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2712 	 * but avoids potential spinlock contention of an exact result.
2713 	 */
2714 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2715 	    likely(gfp_mask & __GFP_NORETRY))
2716 		return NULL;
2717 
2718 	page = alloc_page(gfp_mask);
2719 	if (likely(page != NULL))
2720 		percpu_counter_add(&cc->n_allocated_pages, 1);
2721 
2722 	return page;
2723 }
2724 
2725 static void crypt_page_free(void *page, void *pool_data)
2726 {
2727 	struct crypt_config *cc = pool_data;
2728 
2729 	__free_page(page);
2730 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2731 }
2732 
2733 static void crypt_dtr(struct dm_target *ti)
2734 {
2735 	struct crypt_config *cc = ti->private;
2736 
2737 	ti->private = NULL;
2738 
2739 	if (!cc)
2740 		return;
2741 
2742 	if (cc->write_thread)
2743 		kthread_stop(cc->write_thread);
2744 
2745 	if (cc->io_queue)
2746 		destroy_workqueue(cc->io_queue);
2747 	if (cc->crypt_queue)
2748 		destroy_workqueue(cc->crypt_queue);
2749 
2750 	if (cc->workqueue_id)
2751 		ida_free(&workqueue_ida, cc->workqueue_id);
2752 
2753 	crypt_free_tfms(cc);
2754 
2755 	bioset_exit(&cc->bs);
2756 
2757 	mempool_exit(&cc->page_pool);
2758 	mempool_exit(&cc->req_pool);
2759 	mempool_exit(&cc->tag_pool);
2760 
2761 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2762 	percpu_counter_destroy(&cc->n_allocated_pages);
2763 
2764 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2765 		cc->iv_gen_ops->dtr(cc);
2766 
2767 	if (cc->dev)
2768 		dm_put_device(ti, cc->dev);
2769 
2770 	kfree_sensitive(cc->cipher_string);
2771 	kfree_sensitive(cc->key_string);
2772 	kfree_sensitive(cc->cipher_auth);
2773 	kfree_sensitive(cc->authenc_key);
2774 
2775 	mutex_destroy(&cc->bio_alloc_lock);
2776 
2777 	/* Must zero key material before freeing */
2778 	kfree_sensitive(cc);
2779 
2780 	spin_lock(&dm_crypt_clients_lock);
2781 	WARN_ON(!dm_crypt_clients_n);
2782 	dm_crypt_clients_n--;
2783 	crypt_calculate_pages_per_client();
2784 	spin_unlock(&dm_crypt_clients_lock);
2785 
2786 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2787 }
2788 
2789 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2790 {
2791 	struct crypt_config *cc = ti->private;
2792 
2793 	if (crypt_integrity_aead(cc))
2794 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2795 	else
2796 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2797 
2798 	if (cc->iv_size)
2799 		/* at least a 64 bit sector number should fit in our buffer */
2800 		cc->iv_size = max(cc->iv_size,
2801 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2802 	else if (ivmode) {
2803 		DMWARN("Selected cipher does not support IVs");
2804 		ivmode = NULL;
2805 	}
2806 
2807 	/* Choose ivmode, see comments at iv code. */
2808 	if (ivmode == NULL)
2809 		cc->iv_gen_ops = NULL;
2810 	else if (strcmp(ivmode, "plain") == 0)
2811 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2812 	else if (strcmp(ivmode, "plain64") == 0)
2813 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2814 	else if (strcmp(ivmode, "plain64be") == 0)
2815 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2816 	else if (strcmp(ivmode, "essiv") == 0)
2817 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2818 	else if (strcmp(ivmode, "benbi") == 0)
2819 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2820 	else if (strcmp(ivmode, "null") == 0)
2821 		cc->iv_gen_ops = &crypt_iv_null_ops;
2822 	else if (strcmp(ivmode, "eboiv") == 0)
2823 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2824 	else if (strcmp(ivmode, "elephant") == 0) {
2825 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2826 		cc->key_parts = 2;
2827 		cc->key_extra_size = cc->key_size / 2;
2828 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2829 			return -EINVAL;
2830 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2831 	} else if (strcmp(ivmode, "lmk") == 0) {
2832 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2833 		/*
2834 		 * Version 2 and 3 is recognised according
2835 		 * to length of provided multi-key string.
2836 		 * If present (version 3), last key is used as IV seed.
2837 		 * All keys (including IV seed) are always the same size.
2838 		 */
2839 		if (cc->key_size % cc->key_parts) {
2840 			cc->key_parts++;
2841 			cc->key_extra_size = cc->key_size / cc->key_parts;
2842 		}
2843 	} else if (strcmp(ivmode, "tcw") == 0) {
2844 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2845 		cc->key_parts += 2; /* IV + whitening */
2846 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2847 	} else if (strcmp(ivmode, "random") == 0) {
2848 		cc->iv_gen_ops = &crypt_iv_random_ops;
2849 		/* Need storage space in integrity fields. */
2850 		cc->integrity_iv_size = cc->iv_size;
2851 	} else {
2852 		ti->error = "Invalid IV mode";
2853 		return -EINVAL;
2854 	}
2855 
2856 	return 0;
2857 }
2858 
2859 /*
2860  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2861  * The HMAC is needed to calculate tag size (HMAC digest size).
2862  * This should be probably done by crypto-api calls (once available...)
2863  */
2864 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2865 {
2866 	char *start, *end, *mac_alg = NULL;
2867 	struct crypto_ahash *mac;
2868 
2869 	if (!strstarts(cipher_api, "authenc("))
2870 		return 0;
2871 
2872 	start = strchr(cipher_api, '(');
2873 	end = strchr(cipher_api, ',');
2874 	if (!start || !end || ++start > end)
2875 		return -EINVAL;
2876 
2877 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2878 	if (!mac_alg)
2879 		return -ENOMEM;
2880 
2881 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2882 	kfree(mac_alg);
2883 
2884 	if (IS_ERR(mac))
2885 		return PTR_ERR(mac);
2886 
2887 	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2888 		cc->key_mac_size = crypto_ahash_digestsize(mac);
2889 	crypto_free_ahash(mac);
2890 
2891 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2892 	if (!cc->authenc_key)
2893 		return -ENOMEM;
2894 
2895 	return 0;
2896 }
2897 
2898 static int crypt_ctr_cipher_new(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_api, buf[CRYPTO_MAX_ALG_NAME];
2903 	int ret = -EINVAL;
2904 
2905 	cc->tfms_count = 1;
2906 
2907 	/*
2908 	 * New format (capi: prefix)
2909 	 * capi:cipher_api_spec-iv:ivopts
2910 	 */
2911 	tmp = &cipher_in[strlen("capi:")];
2912 
2913 	/* Separate IV options if present, it can contain another '-' in hash name */
2914 	*ivopts = strrchr(tmp, ':');
2915 	if (*ivopts) {
2916 		**ivopts = '\0';
2917 		(*ivopts)++;
2918 	}
2919 	/* Parse IV mode */
2920 	*ivmode = strrchr(tmp, '-');
2921 	if (*ivmode) {
2922 		**ivmode = '\0';
2923 		(*ivmode)++;
2924 	}
2925 	/* The rest is crypto API spec */
2926 	cipher_api = tmp;
2927 
2928 	/* Alloc AEAD, can be used only in new format. */
2929 	if (crypt_integrity_aead(cc)) {
2930 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2931 		if (ret < 0) {
2932 			ti->error = "Invalid AEAD cipher spec";
2933 			return ret;
2934 		}
2935 	}
2936 
2937 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2938 		cc->tfms_count = 64;
2939 
2940 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2941 		if (!*ivopts) {
2942 			ti->error = "Digest algorithm missing for ESSIV mode";
2943 			return -EINVAL;
2944 		}
2945 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2946 			       cipher_api, *ivopts);
2947 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2948 			ti->error = "Cannot allocate cipher string";
2949 			return -ENOMEM;
2950 		}
2951 		cipher_api = buf;
2952 	}
2953 
2954 	cc->key_parts = cc->tfms_count;
2955 
2956 	/* Allocate cipher */
2957 	ret = crypt_alloc_tfms(cc, cipher_api);
2958 	if (ret < 0) {
2959 		ti->error = "Error allocating crypto tfm";
2960 		return ret;
2961 	}
2962 
2963 	if (crypt_integrity_aead(cc))
2964 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2965 	else
2966 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2967 
2968 	return 0;
2969 }
2970 
2971 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2972 				char **ivmode, char **ivopts)
2973 {
2974 	struct crypt_config *cc = ti->private;
2975 	char *tmp, *cipher, *chainmode, *keycount;
2976 	char *cipher_api = NULL;
2977 	int ret = -EINVAL;
2978 	char dummy;
2979 
2980 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2981 		ti->error = "Bad cipher specification";
2982 		return -EINVAL;
2983 	}
2984 
2985 	/*
2986 	 * Legacy dm-crypt cipher specification
2987 	 * cipher[:keycount]-mode-iv:ivopts
2988 	 */
2989 	tmp = cipher_in;
2990 	keycount = strsep(&tmp, "-");
2991 	cipher = strsep(&keycount, ":");
2992 
2993 	if (!keycount)
2994 		cc->tfms_count = 1;
2995 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2996 		 !is_power_of_2(cc->tfms_count)) {
2997 		ti->error = "Bad cipher key count specification";
2998 		return -EINVAL;
2999 	}
3000 	cc->key_parts = cc->tfms_count;
3001 
3002 	chainmode = strsep(&tmp, "-");
3003 	*ivmode = strsep(&tmp, ":");
3004 	*ivopts = tmp;
3005 
3006 	/*
3007 	 * For compatibility with the original dm-crypt mapping format, if
3008 	 * only the cipher name is supplied, use cbc-plain.
3009 	 */
3010 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3011 		chainmode = "cbc";
3012 		*ivmode = "plain";
3013 	}
3014 
3015 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3016 		ti->error = "IV mechanism required";
3017 		return -EINVAL;
3018 	}
3019 
3020 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3021 	if (!cipher_api)
3022 		goto bad_mem;
3023 
3024 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3025 		if (!*ivopts) {
3026 			ti->error = "Digest algorithm missing for ESSIV mode";
3027 			kfree(cipher_api);
3028 			return -EINVAL;
3029 		}
3030 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3031 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3032 	} else {
3033 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3034 			       "%s(%s)", chainmode, cipher);
3035 	}
3036 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3037 		kfree(cipher_api);
3038 		goto bad_mem;
3039 	}
3040 
3041 	/* Allocate cipher */
3042 	ret = crypt_alloc_tfms(cc, cipher_api);
3043 	if (ret < 0) {
3044 		ti->error = "Error allocating crypto tfm";
3045 		kfree(cipher_api);
3046 		return ret;
3047 	}
3048 	kfree(cipher_api);
3049 
3050 	return 0;
3051 bad_mem:
3052 	ti->error = "Cannot allocate cipher strings";
3053 	return -ENOMEM;
3054 }
3055 
3056 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3057 {
3058 	struct crypt_config *cc = ti->private;
3059 	char *ivmode = NULL, *ivopts = NULL;
3060 	int ret;
3061 
3062 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3063 	if (!cc->cipher_string) {
3064 		ti->error = "Cannot allocate cipher strings";
3065 		return -ENOMEM;
3066 	}
3067 
3068 	if (strstarts(cipher_in, "capi:"))
3069 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3070 	else
3071 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3072 	if (ret)
3073 		return ret;
3074 
3075 	/* Initialize IV */
3076 	ret = crypt_ctr_ivmode(ti, ivmode);
3077 	if (ret < 0)
3078 		return ret;
3079 
3080 	/* Initialize and set key */
3081 	ret = crypt_set_key(cc, key);
3082 	if (ret < 0) {
3083 		ti->error = "Error decoding and setting key";
3084 		return ret;
3085 	}
3086 
3087 	/* Allocate IV */
3088 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3089 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3090 		if (ret < 0) {
3091 			ti->error = "Error creating IV";
3092 			return ret;
3093 		}
3094 	}
3095 
3096 	/* Initialize IV (set keys for ESSIV etc) */
3097 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3098 		ret = cc->iv_gen_ops->init(cc);
3099 		if (ret < 0) {
3100 			ti->error = "Error initialising IV";
3101 			return ret;
3102 		}
3103 	}
3104 
3105 	/* wipe the kernel key payload copy */
3106 	if (cc->key_string)
3107 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3108 
3109 	return ret;
3110 }
3111 
3112 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3113 {
3114 	struct crypt_config *cc = ti->private;
3115 	struct dm_arg_set as;
3116 	static const struct dm_arg _args[] = {
3117 		{0, 9, "Invalid number of feature args"},
3118 	};
3119 	unsigned int opt_params, val;
3120 	const char *opt_string, *sval;
3121 	char dummy;
3122 	int ret;
3123 
3124 	/* Optional parameters */
3125 	as.argc = argc;
3126 	as.argv = argv;
3127 
3128 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3129 	if (ret)
3130 		return ret;
3131 
3132 	while (opt_params--) {
3133 		opt_string = dm_shift_arg(&as);
3134 		if (!opt_string) {
3135 			ti->error = "Not enough feature arguments";
3136 			return -EINVAL;
3137 		}
3138 
3139 		if (!strcasecmp(opt_string, "allow_discards"))
3140 			ti->num_discard_bios = 1;
3141 
3142 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3143 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3144 		else if (!strcasecmp(opt_string, "high_priority"))
3145 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3146 
3147 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3148 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3149 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3150 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3151 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3152 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3153 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3154 			if (val == 0 || val > MAX_TAG_SIZE) {
3155 				ti->error = "Invalid integrity arguments";
3156 				return -EINVAL;
3157 			}
3158 			cc->used_tag_size = val;
3159 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3160 			if (!strcasecmp(sval, "aead")) {
3161 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3162 			} else if (strcasecmp(sval, "none")) {
3163 				ti->error = "Unknown integrity profile";
3164 				return -EINVAL;
3165 			}
3166 
3167 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3168 			if (!cc->cipher_auth)
3169 				return -ENOMEM;
3170 		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3171 			if (!val) {
3172 				ti->error = "Invalid integrity_key_size argument";
3173 				return -EINVAL;
3174 			}
3175 			cc->key_mac_size = val;
3176 			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3177 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3178 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3179 			    cc->sector_size > 4096 ||
3180 			    (cc->sector_size & (cc->sector_size - 1))) {
3181 				ti->error = "Invalid feature value for sector_size";
3182 				return -EINVAL;
3183 			}
3184 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3185 				ti->error = "Device size is not multiple of sector_size feature";
3186 				return -EINVAL;
3187 			}
3188 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3189 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3190 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3191 		else {
3192 			ti->error = "Invalid feature arguments";
3193 			return -EINVAL;
3194 		}
3195 	}
3196 
3197 	return 0;
3198 }
3199 
3200 #ifdef CONFIG_BLK_DEV_ZONED
3201 static int crypt_report_zones(struct dm_target *ti,
3202 		struct dm_report_zones_args *args, unsigned int nr_zones)
3203 {
3204 	struct crypt_config *cc = ti->private;
3205 
3206 	return dm_report_zones(cc->dev->bdev, cc->start,
3207 			cc->start + dm_target_offset(ti, args->next_sector),
3208 			args, nr_zones);
3209 }
3210 #else
3211 #define crypt_report_zones NULL
3212 #endif
3213 
3214 /*
3215  * Construct an encryption mapping:
3216  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3217  */
3218 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3219 {
3220 	struct crypt_config *cc;
3221 	const char *devname = dm_table_device_name(ti->table);
3222 	int key_size, wq_id;
3223 	unsigned int align_mask;
3224 	unsigned int common_wq_flags;
3225 	unsigned long long tmpll;
3226 	int ret;
3227 	size_t iv_size_padding, additional_req_size;
3228 	char dummy;
3229 
3230 	if (argc < 5) {
3231 		ti->error = "Not enough arguments";
3232 		return -EINVAL;
3233 	}
3234 
3235 	key_size = get_key_size(&argv[1]);
3236 	if (key_size < 0) {
3237 		ti->error = "Cannot parse key size";
3238 		return -EINVAL;
3239 	}
3240 
3241 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3242 	if (!cc) {
3243 		ti->error = "Cannot allocate encryption context";
3244 		return -ENOMEM;
3245 	}
3246 	cc->key_size = key_size;
3247 	cc->sector_size = (1 << SECTOR_SHIFT);
3248 	cc->sector_shift = 0;
3249 
3250 	ti->private = cc;
3251 
3252 	spin_lock(&dm_crypt_clients_lock);
3253 	dm_crypt_clients_n++;
3254 	crypt_calculate_pages_per_client();
3255 	spin_unlock(&dm_crypt_clients_lock);
3256 
3257 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3258 	if (ret < 0)
3259 		goto bad;
3260 
3261 	/* Optional parameters need to be read before cipher constructor */
3262 	if (argc > 5) {
3263 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3264 		if (ret)
3265 			goto bad;
3266 	}
3267 
3268 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3269 	if (ret < 0)
3270 		goto bad;
3271 
3272 	if (crypt_integrity_aead(cc)) {
3273 		cc->dmreq_start = sizeof(struct aead_request);
3274 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3275 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3276 	} else {
3277 		cc->dmreq_start = sizeof(struct skcipher_request);
3278 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3279 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3280 	}
3281 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3282 
3283 	if (align_mask < CRYPTO_MINALIGN) {
3284 		/* Allocate the padding exactly */
3285 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3286 				& align_mask;
3287 	} else {
3288 		/*
3289 		 * If the cipher requires greater alignment than kmalloc
3290 		 * alignment, we don't know the exact position of the
3291 		 * initialization vector. We must assume worst case.
3292 		 */
3293 		iv_size_padding = align_mask;
3294 	}
3295 
3296 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3297 	additional_req_size = sizeof(struct dm_crypt_request) +
3298 		iv_size_padding + cc->iv_size +
3299 		cc->iv_size +
3300 		sizeof(uint64_t) +
3301 		sizeof(unsigned int);
3302 
3303 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3304 	if (ret) {
3305 		ti->error = "Cannot allocate crypt request mempool";
3306 		goto bad;
3307 	}
3308 
3309 	cc->per_bio_data_size = ti->per_io_data_size =
3310 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3311 		      ARCH_DMA_MINALIGN);
3312 
3313 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3314 	if (ret) {
3315 		ti->error = "Cannot allocate page mempool";
3316 		goto bad;
3317 	}
3318 
3319 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3320 	if (ret) {
3321 		ti->error = "Cannot allocate crypt bioset";
3322 		goto bad;
3323 	}
3324 
3325 	mutex_init(&cc->bio_alloc_lock);
3326 
3327 	ret = -EINVAL;
3328 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3329 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3330 		ti->error = "Invalid iv_offset sector";
3331 		goto bad;
3332 	}
3333 	cc->iv_offset = tmpll;
3334 
3335 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3336 	if (ret) {
3337 		ti->error = "Device lookup failed";
3338 		goto bad;
3339 	}
3340 
3341 	ret = -EINVAL;
3342 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3343 		ti->error = "Invalid device sector";
3344 		goto bad;
3345 	}
3346 	cc->start = tmpll;
3347 
3348 	if (bdev_is_zoned(cc->dev->bdev)) {
3349 		/*
3350 		 * For zoned block devices, we need to preserve the issuer write
3351 		 * ordering. To do so, disable write workqueues and force inline
3352 		 * encryption completion.
3353 		 */
3354 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3355 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3356 
3357 		/*
3358 		 * All zone append writes to a zone of a zoned block device will
3359 		 * have the same BIO sector, the start of the zone. When the
3360 		 * cypher IV mode uses sector values, all data targeting a
3361 		 * zone will be encrypted using the first sector numbers of the
3362 		 * zone. This will not result in write errors but will
3363 		 * cause most reads to fail as reads will use the sector values
3364 		 * for the actual data locations, resulting in IV mismatch.
3365 		 * To avoid this problem, ask DM core to emulate zone append
3366 		 * operations with regular writes.
3367 		 */
3368 		DMDEBUG("Zone append operations will be emulated");
3369 		ti->emulate_zone_append = true;
3370 	}
3371 
3372 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3373 		ret = crypt_integrity_ctr(cc, ti);
3374 		if (ret)
3375 			goto bad;
3376 
3377 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3378 		if (!cc->tag_pool_max_sectors)
3379 			cc->tag_pool_max_sectors = 1;
3380 
3381 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3382 			cc->tag_pool_max_sectors * cc->tuple_size);
3383 		if (ret) {
3384 			ti->error = "Cannot allocate integrity tags mempool";
3385 			goto bad;
3386 		}
3387 
3388 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3389 	}
3390 
3391 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3392 	if (wq_id < 0) {
3393 		ti->error = "Couldn't get workqueue id";
3394 		ret = wq_id;
3395 		goto bad;
3396 	}
3397 	cc->workqueue_id = wq_id;
3398 
3399 	ret = -ENOMEM;
3400 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3401 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3402 		common_wq_flags |= WQ_HIGHPRI;
3403 
3404 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d",
3405 				       common_wq_flags | WQ_PERCPU, 1,
3406 				       devname, wq_id);
3407 	if (!cc->io_queue) {
3408 		ti->error = "Couldn't create kcryptd io queue";
3409 		goto bad;
3410 	}
3411 
3412 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3413 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3414 						  common_wq_flags | WQ_CPU_INTENSIVE | WQ_PERCPU,
3415 						  1, devname, wq_id);
3416 	} else {
3417 		/*
3418 		 * While crypt_queue is certainly CPU intensive, the use of
3419 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3420 		 */
3421 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3422 						  common_wq_flags | WQ_UNBOUND,
3423 						  num_online_cpus(), devname, wq_id);
3424 	}
3425 	if (!cc->crypt_queue) {
3426 		ti->error = "Couldn't create kcryptd queue";
3427 		goto bad;
3428 	}
3429 
3430 	spin_lock_init(&cc->write_thread_lock);
3431 	cc->write_tree = RB_ROOT;
3432 
3433 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3434 	if (IS_ERR(cc->write_thread)) {
3435 		ret = PTR_ERR(cc->write_thread);
3436 		cc->write_thread = NULL;
3437 		ti->error = "Couldn't spawn write thread";
3438 		goto bad;
3439 	}
3440 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3441 		set_user_nice(cc->write_thread, MIN_NICE);
3442 
3443 	ti->num_flush_bios = 1;
3444 	ti->limit_swap_bios = true;
3445 	ti->accounts_remapped_io = true;
3446 
3447 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3448 	return 0;
3449 
3450 bad:
3451 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3452 	crypt_dtr(ti);
3453 	return ret;
3454 }
3455 
3456 static int crypt_map(struct dm_target *ti, struct bio *bio)
3457 {
3458 	struct dm_crypt_io *io;
3459 	struct crypt_config *cc = ti->private;
3460 	unsigned max_sectors;
3461 	bool no_split;
3462 
3463 	/*
3464 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3465 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3466 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3467 	 */
3468 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3469 	    bio_op(bio) == REQ_OP_DISCARD)) {
3470 		bio_set_dev(bio, cc->dev->bdev);
3471 		if (bio_sectors(bio))
3472 			bio->bi_iter.bi_sector = cc->start +
3473 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3474 		return DM_MAPIO_REMAPPED;
3475 	}
3476 
3477 	/*
3478 	 * Check if bio is too large, split as needed.
3479 	 *
3480 	 * For zoned devices, splitting write operations creates the
3481 	 * risk of deadlocking queue freeze operations with zone write
3482 	 * plugging BIO work when the reminder of a split BIO is
3483 	 * issued. So always allow the entire BIO to proceed.
3484 	 */
3485 	no_split = (ti->emulate_zone_append && op_is_write(bio_op(bio))) ||
3486 		   (bio->bi_opf & REQ_ATOMIC);
3487 	max_sectors = get_max_request_sectors(ti, bio, no_split);
3488 	if (unlikely(bio_sectors(bio) > max_sectors)) {
3489 		if (unlikely(no_split))
3490 			return DM_MAPIO_KILL;
3491 		dm_accept_partial_bio(bio, max_sectors);
3492 	}
3493 
3494 	/*
3495 	 * Ensure that bio is a multiple of internal sector encryption size
3496 	 * and is aligned to this size as defined in IO hints.
3497 	 */
3498 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3499 		return DM_MAPIO_KILL;
3500 
3501 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3502 		return DM_MAPIO_KILL;
3503 
3504 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3505 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3506 
3507 	if (cc->tuple_size) {
3508 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3509 
3510 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3511 			io->integrity_metadata = NULL;
3512 		else
3513 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3514 
3515 		if (unlikely(!io->integrity_metadata)) {
3516 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3517 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3518 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3519 			io->integrity_metadata_from_pool = true;
3520 		}
3521 	}
3522 
3523 	if (crypt_integrity_aead(cc))
3524 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3525 	else
3526 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3527 
3528 	if (bio_data_dir(io->base_bio) == READ) {
3529 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3530 			kcryptd_queue_read(io);
3531 	} else
3532 		kcryptd_queue_crypt(io);
3533 
3534 	return DM_MAPIO_SUBMITTED;
3535 }
3536 
3537 static char hex2asc(unsigned char c)
3538 {
3539 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3540 }
3541 
3542 static void crypt_status(struct dm_target *ti, status_type_t type,
3543 			 unsigned int status_flags, char *result, unsigned int maxlen)
3544 {
3545 	struct crypt_config *cc = ti->private;
3546 	unsigned int i, sz = 0;
3547 	int num_feature_args = 0;
3548 
3549 	switch (type) {
3550 	case STATUSTYPE_INFO:
3551 		result[0] = '\0';
3552 		break;
3553 
3554 	case STATUSTYPE_TABLE:
3555 		DMEMIT("%s ", cc->cipher_string);
3556 
3557 		if (cc->key_size > 0) {
3558 			if (cc->key_string)
3559 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3560 			else {
3561 				for (i = 0; i < cc->key_size; i++) {
3562 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3563 					       hex2asc(cc->key[i] & 0xf));
3564 				}
3565 			}
3566 		} else
3567 			DMEMIT("-");
3568 
3569 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3570 				cc->dev->name, (unsigned long long)cc->start);
3571 
3572 		num_feature_args += !!ti->num_discard_bios;
3573 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3574 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3575 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3576 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3577 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3578 		num_feature_args += !!cc->used_tag_size;
3579 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3580 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3581 		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3582 		if (num_feature_args) {
3583 			DMEMIT(" %d", num_feature_args);
3584 			if (ti->num_discard_bios)
3585 				DMEMIT(" allow_discards");
3586 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3587 				DMEMIT(" same_cpu_crypt");
3588 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3589 				DMEMIT(" high_priority");
3590 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3591 				DMEMIT(" submit_from_crypt_cpus");
3592 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3593 				DMEMIT(" no_read_workqueue");
3594 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3595 				DMEMIT(" no_write_workqueue");
3596 			if (cc->used_tag_size)
3597 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3598 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3599 				DMEMIT(" sector_size:%d", cc->sector_size);
3600 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3601 				DMEMIT(" iv_large_sectors");
3602 			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3603 				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3604 		}
3605 		break;
3606 
3607 	case STATUSTYPE_IMA:
3608 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3609 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3610 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3611 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3612 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3613 		       'y' : 'n');
3614 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3615 		       'y' : 'n');
3616 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3617 		       'y' : 'n');
3618 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3619 		       'y' : 'n');
3620 
3621 		if (cc->used_tag_size)
3622 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3623 			       cc->used_tag_size, cc->cipher_auth);
3624 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3625 			DMEMIT(",sector_size=%d", cc->sector_size);
3626 		if (cc->cipher_string)
3627 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3628 
3629 		DMEMIT(",key_size=%u", cc->key_size);
3630 		DMEMIT(",key_parts=%u", cc->key_parts);
3631 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3632 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3633 		DMEMIT(";");
3634 		break;
3635 	}
3636 }
3637 
3638 static void crypt_postsuspend(struct dm_target *ti)
3639 {
3640 	struct crypt_config *cc = ti->private;
3641 
3642 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3643 }
3644 
3645 static int crypt_preresume(struct dm_target *ti)
3646 {
3647 	struct crypt_config *cc = ti->private;
3648 
3649 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3650 		DMERR("aborting resume - crypt key is not set.");
3651 		return -EAGAIN;
3652 	}
3653 
3654 	return 0;
3655 }
3656 
3657 static void crypt_resume(struct dm_target *ti)
3658 {
3659 	struct crypt_config *cc = ti->private;
3660 
3661 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3662 }
3663 
3664 /* Message interface
3665  *	key set <key>
3666  *	key wipe
3667  */
3668 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3669 			 char *result, unsigned int maxlen)
3670 {
3671 	struct crypt_config *cc = ti->private;
3672 	int key_size, ret = -EINVAL;
3673 
3674 	if (argc < 2)
3675 		goto error;
3676 
3677 	if (!strcasecmp(argv[0], "key")) {
3678 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3679 			DMWARN("not suspended during key manipulation.");
3680 			return -EINVAL;
3681 		}
3682 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3683 			/* The key size may not be changed. */
3684 			key_size = get_key_size(&argv[2]);
3685 			if (key_size < 0 || cc->key_size != key_size) {
3686 				memset(argv[2], '0', strlen(argv[2]));
3687 				return -EINVAL;
3688 			}
3689 
3690 			ret = crypt_set_key(cc, argv[2]);
3691 			if (ret)
3692 				return ret;
3693 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3694 				ret = cc->iv_gen_ops->init(cc);
3695 			/* wipe the kernel key payload copy */
3696 			if (cc->key_string)
3697 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3698 			return ret;
3699 		}
3700 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3701 			return crypt_wipe_key(cc);
3702 	}
3703 
3704 error:
3705 	DMWARN("unrecognised message received.");
3706 	return -EINVAL;
3707 }
3708 
3709 static int crypt_iterate_devices(struct dm_target *ti,
3710 				 iterate_devices_callout_fn fn, void *data)
3711 {
3712 	struct crypt_config *cc = ti->private;
3713 
3714 	return fn(ti, cc->dev, cc->start, ti->len, data);
3715 }
3716 
3717 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3718 {
3719 	struct crypt_config *cc = ti->private;
3720 
3721 	limits->logical_block_size =
3722 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3723 	limits->physical_block_size =
3724 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3725 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3726 	limits->dma_alignment = limits->logical_block_size - 1;
3727 
3728 	/*
3729 	 * For zoned dm-crypt targets, there will be no internal splitting of
3730 	 * write BIOs to avoid exceeding BIO_MAX_VECS vectors per BIO. But
3731 	 * without respecting this limit, crypt_alloc_buffer() will trigger a
3732 	 * BUG(). Avoid this by forcing DM core to split write BIOs to this
3733 	 * limit.
3734 	 */
3735 	if (ti->emulate_zone_append)
3736 		limits->max_hw_sectors = min(limits->max_hw_sectors,
3737 					     BIO_MAX_VECS << PAGE_SECTORS_SHIFT);
3738 
3739 	limits->atomic_write_hw_unit_max = min(limits->atomic_write_hw_unit_max,
3740 					       BIO_MAX_VECS << PAGE_SHIFT);
3741 	limits->atomic_write_hw_max = min(limits->atomic_write_hw_max,
3742 					  BIO_MAX_VECS << PAGE_SHIFT);
3743 }
3744 
3745 static struct target_type crypt_target = {
3746 	.name   = "crypt",
3747 	.version = {1, 29, 0},
3748 	.module = THIS_MODULE,
3749 	.ctr    = crypt_ctr,
3750 	.dtr    = crypt_dtr,
3751 	.features = DM_TARGET_ZONED_HM | DM_TARGET_ATOMIC_WRITES,
3752 	.report_zones = crypt_report_zones,
3753 	.map    = crypt_map,
3754 	.status = crypt_status,
3755 	.postsuspend = crypt_postsuspend,
3756 	.preresume = crypt_preresume,
3757 	.resume = crypt_resume,
3758 	.message = crypt_message,
3759 	.iterate_devices = crypt_iterate_devices,
3760 	.io_hints = crypt_io_hints,
3761 };
3762 module_dm(crypt);
3763 
3764 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3765 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3766 MODULE_LICENSE("GPL");
3767