xref: /linux/drivers/md/dm-crypt.c (revision 4b4a8d9560d987f4df17b7248ab1c8146138d0f5)
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/aes.h>
36 #include <crypto/hash.h>
37 #include <crypto/md5.h>
38 #include <crypto/skcipher.h>
39 #include <crypto/aead.h>
40 #include <crypto/authenc.h>
41 #include <crypto/utils.h>
42 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
43 #include <linux/key-type.h>
44 #include <keys/user-type.h>
45 #include <keys/encrypted-type.h>
46 #include <keys/trusted-type.h>
47 
48 #include <linux/device-mapper.h>
49 
50 #include "dm-audit.h"
51 
52 #define DM_MSG_PREFIX "crypt"
53 
54 static DEFINE_IDA(workqueue_ida);
55 
56 /*
57  * context holding the current state of a multi-part conversion
58  */
59 struct convert_context {
60 	struct completion restart;
61 	struct bio *bio_in;
62 	struct bvec_iter iter_in;
63 	struct bio *bio_out;
64 	struct bvec_iter iter_out;
65 	atomic_t cc_pending;
66 	unsigned int tag_offset;
67 	u64 cc_sector;
68 	union {
69 		struct skcipher_request *req;
70 		struct aead_request *req_aead;
71 	} r;
72 	bool aead_recheck;
73 	bool aead_failed;
74 
75 };
76 
77 /*
78  * per bio private data
79  */
80 struct dm_crypt_io {
81 	struct crypt_config *cc;
82 	struct bio *base_bio;
83 	u8 *integrity_metadata;
84 	bool integrity_metadata_from_pool:1;
85 
86 	struct work_struct work;
87 
88 	struct convert_context ctx;
89 
90 	atomic_t io_pending;
91 	blk_status_t error;
92 	sector_t sector;
93 
94 	struct bvec_iter saved_bi_iter;
95 
96 	struct rb_node rb_node;
97 } CRYPTO_MINALIGN_ATTR;
98 
99 struct dm_crypt_request {
100 	struct convert_context *ctx;
101 	struct scatterlist sg_in[4];
102 	struct scatterlist sg_out[4];
103 	u64 iv_sector;
104 };
105 
106 struct crypt_config;
107 
108 struct crypt_iv_operations {
109 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
110 		   const char *opts);
111 	void (*dtr)(struct crypt_config *cc);
112 	int (*init)(struct crypt_config *cc);
113 	void (*wipe)(struct crypt_config *cc);
114 	int (*generator)(struct crypt_config *cc, u8 *iv,
115 			 struct dm_crypt_request *dmreq);
116 	int (*post)(struct crypt_config *cc, u8 *iv,
117 		    struct dm_crypt_request *dmreq);
118 };
119 
120 struct iv_benbi_private {
121 	int shift;
122 };
123 
124 #define LMK_SEED_SIZE 64 /* hash + 0 */
125 struct iv_lmk_private {
126 	u8 *seed;
127 };
128 
129 #define TCW_WHITENING_SIZE 16
130 struct iv_tcw_private {
131 	u8 *iv_seed;
132 	u8 *whitening;
133 };
134 
135 #define ELEPHANT_MAX_KEY_SIZE 32
136 struct iv_elephant_private {
137 	struct aes_enckey *key;
138 };
139 
140 /*
141  * Crypt: maps a linear range of a block device
142  * and encrypts / decrypts at the same time.
143  */
144 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
145 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
146 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
147 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
148 
149 enum cipher_flags {
150 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
151 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
152 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
153 	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
154 };
155 
156 /*
157  * The fields in here must be read only after initialization.
158  */
159 struct crypt_config {
160 	struct dm_dev *dev;
161 	sector_t start;
162 
163 	struct percpu_counter n_allocated_pages;
164 
165 	struct workqueue_struct *io_queue;
166 	struct workqueue_struct *crypt_queue;
167 
168 	spinlock_t write_thread_lock;
169 	struct task_struct *write_thread;
170 	struct rb_root write_tree;
171 
172 	char *cipher_string;
173 	char *cipher_auth;
174 	char *key_string;
175 
176 	const struct crypt_iv_operations *iv_gen_ops;
177 	union {
178 		struct iv_benbi_private benbi;
179 		struct iv_lmk_private lmk;
180 		struct iv_tcw_private tcw;
181 		struct iv_elephant_private elephant;
182 	} iv_gen_private;
183 	u64 iv_offset;
184 	unsigned int iv_size;
185 	unsigned short sector_size;
186 	unsigned char sector_shift;
187 
188 	union {
189 		struct crypto_skcipher **tfms;
190 		struct crypto_aead **tfms_aead;
191 	} cipher_tfm;
192 	unsigned int tfms_count;
193 	int workqueue_id;
194 	unsigned long cipher_flags;
195 
196 	/*
197 	 * Layout of each crypto request:
198 	 *
199 	 *   struct skcipher_request
200 	 *      context
201 	 *      padding
202 	 *   struct dm_crypt_request
203 	 *      padding
204 	 *   IV
205 	 *
206 	 * The padding is added so that dm_crypt_request and the IV are
207 	 * correctly aligned.
208 	 */
209 	unsigned int dmreq_start;
210 
211 	unsigned int per_bio_data_size;
212 
213 	unsigned long flags;
214 	unsigned int key_size;
215 	unsigned int key_parts;      /* independent parts in key buffer */
216 	unsigned int key_extra_size; /* additional keys length */
217 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
218 
219 	unsigned int integrity_tag_size;
220 	unsigned int integrity_iv_size;
221 	unsigned int used_tag_size;
222 	unsigned int tuple_size;
223 
224 	/*
225 	 * pool for per bio private data, crypto requests,
226 	 * encryption requeusts/buffer pages and integrity tags
227 	 */
228 	unsigned int tag_pool_max_sectors;
229 	mempool_t tag_pool;
230 	mempool_t req_pool;
231 	mempool_t page_pool;
232 
233 	struct bio_set bs;
234 	struct mutex bio_alloc_lock;
235 
236 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
237 	u8 key[] __counted_by(key_size);
238 };
239 
240 #define MIN_IOS		64
241 #define MAX_TAG_SIZE	480
242 #define POOL_ENTRY_SIZE	512
243 
244 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
245 static unsigned int dm_crypt_clients_n;
246 static volatile unsigned long dm_crypt_pages_per_client;
247 #define DM_CRYPT_MEMORY_PERCENT			2
248 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
249 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
250 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
251 
252 static unsigned int max_read_size = 0;
253 module_param(max_read_size, uint, 0644);
254 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
255 static unsigned int max_write_size = 0;
256 module_param(max_write_size, uint, 0644);
257 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
258 
259 static unsigned get_max_request_sectors(struct dm_target *ti, struct bio *bio, bool no_split)
260 {
261 	struct crypt_config *cc = ti->private;
262 	unsigned val, sector_align;
263 	bool wrt = op_is_write(bio_op(bio));
264 
265 	if (no_split) {
266 		val = -1;
267 	} else if (wrt) {
268 		val = min_not_zero(READ_ONCE(max_write_size),
269 				   DM_CRYPT_DEFAULT_MAX_WRITE_SIZE);
270 	} else {
271 		val = min_not_zero(READ_ONCE(max_read_size),
272 				   DM_CRYPT_DEFAULT_MAX_READ_SIZE);
273 	}
274 
275 	if (wrt || cc->used_tag_size)
276 		val = min(val, BIO_MAX_VECS << PAGE_SHIFT);
277 
278 	sector_align = max(bdev_logical_block_size(cc->dev->bdev),
279 			   (unsigned)cc->sector_size);
280 	val = round_down(val, sector_align);
281 	if (unlikely(!val))
282 		val = sector_align;
283 	return val >> SECTOR_SHIFT;
284 }
285 
286 static void crypt_endio(struct bio *clone);
287 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
288 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
289 					     struct scatterlist *sg);
290 
291 static bool crypt_integrity_aead(struct crypt_config *cc);
292 
293 /*
294  * Use this to access cipher attributes that are independent of the key.
295  */
296 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
297 {
298 	return cc->cipher_tfm.tfms[0];
299 }
300 
301 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
302 {
303 	return cc->cipher_tfm.tfms_aead[0];
304 }
305 
306 /*
307  * Different IV generation algorithms:
308  *
309  * plain: the initial vector is the 32-bit little-endian version of the sector
310  *        number, padded with zeros if necessary.
311  *
312  * plain64: the initial vector is the 64-bit little-endian version of the sector
313  *        number, padded with zeros if necessary.
314  *
315  * plain64be: the initial vector is the 64-bit big-endian version of the sector
316  *        number, padded with zeros if necessary.
317  *
318  * essiv: "encrypted sector|salt initial vector", the sector number is
319  *        encrypted with the bulk cipher using a salt as key. The salt
320  *        should be derived from the bulk cipher's key via hashing.
321  *
322  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
323  *        (needed for LRW-32-AES and possible other narrow block modes)
324  *
325  * null: the initial vector is always zero.  Provides compatibility with
326  *       obsolete loop_fish2 devices.  Do not use for new devices.
327  *
328  * lmk:  Compatible implementation of the block chaining mode used
329  *       by the Loop-AES block device encryption system
330  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
331  *       It operates on full 512 byte sectors and uses CBC
332  *       with an IV derived from the sector number, the data and
333  *       optionally extra IV seed.
334  *       This means that after decryption the first block
335  *       of sector must be tweaked according to decrypted data.
336  *       Loop-AES can use three encryption schemes:
337  *         version 1: is plain aes-cbc mode
338  *         version 2: uses 64 multikey scheme with lmk IV generator
339  *         version 3: the same as version 2 with additional IV seed
340  *                   (it uses 65 keys, last key is used as IV seed)
341  *
342  * tcw:  Compatible implementation of the block chaining mode used
343  *       by the TrueCrypt device encryption system (prior to version 4.1).
344  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
345  *       It operates on full 512 byte sectors and uses CBC
346  *       with an IV derived from initial key and the sector number.
347  *       In addition, whitening value is applied on every sector, whitening
348  *       is calculated from initial key, sector number and mixed using CRC32.
349  *       Note that this encryption scheme is vulnerable to watermarking attacks
350  *       and should be used for old compatible containers access only.
351  *
352  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
353  *        The IV is encrypted little-endian byte-offset (with the same key
354  *        and cipher as the volume).
355  *
356  * elephant: The extended version of eboiv with additional Elephant diffuser
357  *           used with Bitlocker CBC mode.
358  *           This mode was used in older Windows systems
359  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
360  */
361 
362 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
363 			      struct dm_crypt_request *dmreq)
364 {
365 	memset(iv, 0, cc->iv_size);
366 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
367 
368 	return 0;
369 }
370 
371 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
372 				struct dm_crypt_request *dmreq)
373 {
374 	memset(iv, 0, cc->iv_size);
375 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
376 
377 	return 0;
378 }
379 
380 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
381 				  struct dm_crypt_request *dmreq)
382 {
383 	memset(iv, 0, cc->iv_size);
384 	/* iv_size is at least of size u64; usually it is 16 bytes */
385 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
386 
387 	return 0;
388 }
389 
390 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
391 			      struct dm_crypt_request *dmreq)
392 {
393 	/*
394 	 * ESSIV encryption of the IV is now handled by the crypto API,
395 	 * so just pass the plain sector number here.
396 	 */
397 	memset(iv, 0, cc->iv_size);
398 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
399 
400 	return 0;
401 }
402 
403 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
404 			      const char *opts)
405 {
406 	unsigned int bs;
407 	int log;
408 
409 	if (crypt_integrity_aead(cc))
410 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
411 	else
412 		bs = crypto_skcipher_blocksize(any_tfm(cc));
413 	log = ilog2(bs);
414 
415 	/*
416 	 * We need to calculate how far we must shift the sector count
417 	 * to get the cipher block count, we use this shift in _gen.
418 	 */
419 	if (1 << log != bs) {
420 		ti->error = "cypher blocksize is not a power of 2";
421 		return -EINVAL;
422 	}
423 
424 	if (log > 9) {
425 		ti->error = "cypher blocksize is > 512";
426 		return -EINVAL;
427 	}
428 
429 	cc->iv_gen_private.benbi.shift = 9 - log;
430 
431 	return 0;
432 }
433 
434 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
435 {
436 }
437 
438 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
439 			      struct dm_crypt_request *dmreq)
440 {
441 	__be64 val;
442 
443 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
444 
445 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
446 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
447 
448 	return 0;
449 }
450 
451 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
452 			     struct dm_crypt_request *dmreq)
453 {
454 	memset(iv, 0, cc->iv_size);
455 
456 	return 0;
457 }
458 
459 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
460 {
461 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
462 
463 	kfree_sensitive(lmk->seed);
464 	lmk->seed = NULL;
465 }
466 
467 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
468 			    const char *opts)
469 {
470 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
471 
472 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
473 		ti->error = "Unsupported sector size for LMK";
474 		return -EINVAL;
475 	}
476 
477 	if (fips_enabled) {
478 		ti->error = "LMK support is disabled due to FIPS";
479 		/* ... because it uses MD5. */
480 		return -EINVAL;
481 	}
482 
483 	/* No seed in LMK version 2 */
484 	if (cc->key_parts == cc->tfms_count) {
485 		lmk->seed = NULL;
486 		return 0;
487 	}
488 
489 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
490 	if (!lmk->seed) {
491 		ti->error = "Error kmallocing seed storage in LMK";
492 		return -ENOMEM;
493 	}
494 
495 	return 0;
496 }
497 
498 static int crypt_iv_lmk_init(struct crypt_config *cc)
499 {
500 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
501 	int subkey_size = cc->key_size / cc->key_parts;
502 
503 	/* LMK seed is on the position of LMK_KEYS + 1 key */
504 	if (lmk->seed)
505 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
506 		       MD5_DIGEST_SIZE);
507 
508 	return 0;
509 }
510 
511 static void crypt_iv_lmk_wipe(struct crypt_config *cc)
512 {
513 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
514 
515 	if (lmk->seed)
516 		memset(lmk->seed, 0, LMK_SEED_SIZE);
517 }
518 
519 static void crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
520 			     struct dm_crypt_request *dmreq, u8 *data)
521 {
522 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
523 	struct md5_ctx ctx;
524 	__le32 buf[4];
525 
526 	md5_init(&ctx);
527 
528 	if (lmk->seed)
529 		md5_update(&ctx, lmk->seed, LMK_SEED_SIZE);
530 
531 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
532 	md5_update(&ctx, data + 16, 16 * 31);
533 
534 	/* Sector is cropped to 56 bits here */
535 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
536 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
537 	buf[2] = cpu_to_le32(4024);
538 	buf[3] = 0;
539 	md5_update(&ctx, (u8 *)buf, sizeof(buf));
540 
541 	/* No MD5 padding here */
542 	cpu_to_le32_array(ctx.state.h, ARRAY_SIZE(ctx.state.h));
543 	memcpy(iv, ctx.state.h, cc->iv_size);
544 }
545 
546 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
547 			    struct dm_crypt_request *dmreq)
548 {
549 	struct scatterlist *sg;
550 	u8 *src;
551 
552 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
553 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
554 		src = kmap_local_page(sg_page(sg));
555 		crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
556 		kunmap_local(src);
557 	} else
558 		memset(iv, 0, cc->iv_size);
559 	return 0;
560 }
561 
562 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
563 			     struct dm_crypt_request *dmreq)
564 {
565 	struct scatterlist *sg;
566 	u8 *dst;
567 
568 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
569 		return 0;
570 
571 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
572 	dst = kmap_local_page(sg_page(sg));
573 	crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
574 
575 	/* Tweak the first block of plaintext sector */
576 	crypto_xor(dst + sg->offset, iv, cc->iv_size);
577 
578 	kunmap_local(dst);
579 	return 0;
580 }
581 
582 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
583 {
584 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
585 
586 	kfree_sensitive(tcw->iv_seed);
587 	tcw->iv_seed = NULL;
588 	kfree_sensitive(tcw->whitening);
589 	tcw->whitening = NULL;
590 }
591 
592 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
593 			    const char *opts)
594 {
595 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
596 
597 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
598 		ti->error = "Unsupported sector size for TCW";
599 		return -EINVAL;
600 	}
601 
602 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
603 		ti->error = "Wrong key size for TCW";
604 		return -EINVAL;
605 	}
606 
607 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
608 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
609 	if (!tcw->iv_seed || !tcw->whitening) {
610 		crypt_iv_tcw_dtr(cc);
611 		ti->error = "Error allocating seed storage in TCW";
612 		return -ENOMEM;
613 	}
614 
615 	return 0;
616 }
617 
618 static int crypt_iv_tcw_init(struct crypt_config *cc)
619 {
620 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
621 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
622 
623 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
624 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
625 	       TCW_WHITENING_SIZE);
626 
627 	return 0;
628 }
629 
630 static void crypt_iv_tcw_wipe(struct crypt_config *cc)
631 {
632 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
633 
634 	memset(tcw->iv_seed, 0, cc->iv_size);
635 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
636 }
637 
638 static void crypt_iv_tcw_whitening(struct crypt_config *cc,
639 				   struct dm_crypt_request *dmreq, u8 *data)
640 {
641 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
642 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
643 	u8 buf[TCW_WHITENING_SIZE];
644 	int i;
645 
646 	/* xor whitening with sector number */
647 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
648 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
649 
650 	/* calculate crc32 for every 32bit part and xor it */
651 	for (i = 0; i < 4; i++)
652 		put_unaligned_le32(crc32(0, &buf[i * 4], 4), &buf[i * 4]);
653 	crypto_xor(&buf[0], &buf[12], 4);
654 	crypto_xor(&buf[4], &buf[8], 4);
655 
656 	/* apply whitening (8 bytes) to whole sector */
657 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
658 		crypto_xor(data + i * 8, buf, 8);
659 	memzero_explicit(buf, sizeof(buf));
660 }
661 
662 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
663 			    struct dm_crypt_request *dmreq)
664 {
665 	struct scatterlist *sg;
666 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
667 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
668 	u8 *src;
669 
670 	/* Remove whitening from ciphertext */
671 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
672 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
673 		src = kmap_local_page(sg_page(sg));
674 		crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
675 		kunmap_local(src);
676 	}
677 
678 	/* Calculate IV */
679 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
680 	if (cc->iv_size > 8)
681 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
682 			       cc->iv_size - 8);
683 
684 	return 0;
685 }
686 
687 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
688 			     struct dm_crypt_request *dmreq)
689 {
690 	struct scatterlist *sg;
691 	u8 *dst;
692 
693 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
694 		return 0;
695 
696 	/* Apply whitening on ciphertext */
697 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
698 	dst = kmap_local_page(sg_page(sg));
699 	crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
700 	kunmap_local(dst);
701 
702 	return 0;
703 }
704 
705 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
706 				struct dm_crypt_request *dmreq)
707 {
708 	/* Used only for writes, there must be an additional space to store IV */
709 	get_random_bytes(iv, cc->iv_size);
710 	return 0;
711 }
712 
713 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
714 			    const char *opts)
715 {
716 	if (crypt_integrity_aead(cc)) {
717 		ti->error = "AEAD transforms not supported for EBOIV";
718 		return -EINVAL;
719 	}
720 
721 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
722 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
723 		return -EINVAL;
724 	}
725 
726 	return 0;
727 }
728 
729 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
730 			    struct dm_crypt_request *dmreq)
731 {
732 	struct crypto_skcipher *tfm = any_tfm(cc);
733 	struct skcipher_request *req;
734 	struct scatterlist src, dst;
735 	DECLARE_CRYPTO_WAIT(wait);
736 	unsigned int reqsize;
737 	int err;
738 	u8 *buf;
739 
740 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
741 	reqsize = ALIGN(reqsize, __alignof__(__le64));
742 
743 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
744 	if (!req)
745 		return -ENOMEM;
746 
747 	skcipher_request_set_tfm(req, tfm);
748 
749 	buf = (u8 *)req + reqsize;
750 	memset(buf, 0, cc->iv_size);
751 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
752 
753 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
754 	sg_init_one(&dst, iv, cc->iv_size);
755 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
756 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
757 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
758 	kfree_sensitive(req);
759 
760 	return err;
761 }
762 
763 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
764 {
765 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
766 
767 	kfree_sensitive(elephant->key);
768 	elephant->key = NULL;
769 }
770 
771 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
772 			    const char *opts)
773 {
774 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
775 	int r;
776 
777 	elephant->key = kmalloc_obj(*elephant->key);
778 	if (!elephant->key)
779 		return -ENOMEM;
780 
781 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
782 	if (r)
783 		crypt_iv_elephant_dtr(cc);
784 	return r;
785 }
786 
787 static void diffuser_disk_to_cpu(u32 *d, size_t n)
788 {
789 #ifndef __LITTLE_ENDIAN
790 	int i;
791 
792 	for (i = 0; i < n; i++)
793 		d[i] = le32_to_cpu((__le32)d[i]);
794 #endif
795 }
796 
797 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
798 {
799 #ifndef __LITTLE_ENDIAN
800 	int i;
801 
802 	for (i = 0; i < n; i++)
803 		d[i] = cpu_to_le32((u32)d[i]);
804 #endif
805 }
806 
807 static void diffuser_a_decrypt(u32 *d, size_t n)
808 {
809 	int i, i1, i2, i3;
810 
811 	for (i = 0; i < 5; i++) {
812 		i1 = 0;
813 		i2 = n - 2;
814 		i3 = n - 5;
815 
816 		while (i1 < (n - 1)) {
817 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
818 			i1++; i2++; i3++;
819 
820 			if (i3 >= n)
821 				i3 -= n;
822 
823 			d[i1] += d[i2] ^ d[i3];
824 			i1++; i2++; i3++;
825 
826 			if (i2 >= n)
827 				i2 -= n;
828 
829 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
830 			i1++; i2++; i3++;
831 
832 			d[i1] += d[i2] ^ d[i3];
833 			i1++; i2++; i3++;
834 		}
835 	}
836 }
837 
838 static void diffuser_a_encrypt(u32 *d, size_t n)
839 {
840 	int i, i1, i2, i3;
841 
842 	for (i = 0; i < 5; i++) {
843 		i1 = n - 1;
844 		i2 = n - 2 - 1;
845 		i3 = n - 5 - 1;
846 
847 		while (i1 > 0) {
848 			d[i1] -= d[i2] ^ d[i3];
849 			i1--; i2--; i3--;
850 
851 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
852 			i1--; i2--; i3--;
853 
854 			if (i2 < 0)
855 				i2 += n;
856 
857 			d[i1] -= d[i2] ^ d[i3];
858 			i1--; i2--; i3--;
859 
860 			if (i3 < 0)
861 				i3 += n;
862 
863 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
864 			i1--; i2--; i3--;
865 		}
866 	}
867 }
868 
869 static void diffuser_b_decrypt(u32 *d, size_t n)
870 {
871 	int i, i1, i2, i3;
872 
873 	for (i = 0; i < 3; i++) {
874 		i1 = 0;
875 		i2 = 2;
876 		i3 = 5;
877 
878 		while (i1 < (n - 1)) {
879 			d[i1] += d[i2] ^ d[i3];
880 			i1++; i2++; i3++;
881 
882 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
883 			i1++; i2++; i3++;
884 
885 			if (i2 >= n)
886 				i2 -= n;
887 
888 			d[i1] += d[i2] ^ d[i3];
889 			i1++; i2++; i3++;
890 
891 			if (i3 >= n)
892 				i3 -= n;
893 
894 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
895 			i1++; i2++; i3++;
896 		}
897 	}
898 }
899 
900 static void diffuser_b_encrypt(u32 *d, size_t n)
901 {
902 	int i, i1, i2, i3;
903 
904 	for (i = 0; i < 3; i++) {
905 		i1 = n - 1;
906 		i2 = 2 - 1;
907 		i3 = 5 - 1;
908 
909 		while (i1 > 0) {
910 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
911 			i1--; i2--; i3--;
912 
913 			if (i3 < 0)
914 				i3 += n;
915 
916 			d[i1] -= d[i2] ^ d[i3];
917 			i1--; i2--; i3--;
918 
919 			if (i2 < 0)
920 				i2 += n;
921 
922 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
923 			i1--; i2--; i3--;
924 
925 			d[i1] -= d[i2] ^ d[i3];
926 			i1--; i2--; i3--;
927 		}
928 	}
929 }
930 
931 static void crypt_iv_elephant(struct crypt_config *cc,
932 			      struct dm_crypt_request *dmreq)
933 {
934 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
935 	u8 *data, *data2, *data_offset;
936 	struct scatterlist *sg, *sg2;
937 	union {
938 		__le64 w[2];
939 		u8 b[16];
940 	} es;
941 	u8 ks[32] __aligned(__alignof(long)); /* Elephant sector key */
942 	int i;
943 
944 	es.w[0] = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
945 	es.w[1] = 0;
946 
947 	/* E(Ks, e(s)) */
948 	aes_encrypt(elephant->key, &ks[0], es.b);
949 
950 	/* E(Ks, e'(s)) */
951 	es.b[15] = 0x80;
952 	aes_encrypt(elephant->key, &ks[16], es.b);
953 
954 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
955 	data = kmap_local_page(sg_page(sg));
956 	data_offset = data + sg->offset;
957 
958 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
959 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
960 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
961 		data2 = kmap_local_page(sg_page(sg2));
962 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
963 		kunmap_local(data2);
964 	}
965 
966 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
967 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
968 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
969 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
970 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
971 	}
972 
973 	for (i = 0; i < (cc->sector_size / 32); i++)
974 		crypto_xor(data_offset + i * 32, ks, 32);
975 
976 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
977 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
978 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
979 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
980 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
981 	}
982 
983 	kunmap_local(data);
984 	memzero_explicit(ks, sizeof(ks));
985 	memzero_explicit(&es, sizeof(es));
986 }
987 
988 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
989 			    struct dm_crypt_request *dmreq)
990 {
991 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
992 		crypt_iv_elephant(cc, dmreq);
993 
994 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
995 }
996 
997 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
998 				  struct dm_crypt_request *dmreq)
999 {
1000 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1001 		crypt_iv_elephant(cc, dmreq);
1002 
1003 	return 0;
1004 }
1005 
1006 static int crypt_iv_elephant_init(struct crypt_config *cc)
1007 {
1008 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1009 	int key_offset = cc->key_size - cc->key_extra_size;
1010 
1011 	return aes_prepareenckey(elephant->key, &cc->key[key_offset], cc->key_extra_size);
1012 }
1013 
1014 static void crypt_iv_elephant_wipe(struct crypt_config *cc)
1015 {
1016 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1017 
1018 	memzero_explicit(elephant->key, sizeof(*elephant->key));
1019 }
1020 
1021 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1022 	.generator = crypt_iv_plain_gen
1023 };
1024 
1025 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1026 	.generator = crypt_iv_plain64_gen
1027 };
1028 
1029 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1030 	.generator = crypt_iv_plain64be_gen
1031 };
1032 
1033 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1034 	.generator = crypt_iv_essiv_gen
1035 };
1036 
1037 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1038 	.ctr	   = crypt_iv_benbi_ctr,
1039 	.dtr	   = crypt_iv_benbi_dtr,
1040 	.generator = crypt_iv_benbi_gen
1041 };
1042 
1043 static const struct crypt_iv_operations crypt_iv_null_ops = {
1044 	.generator = crypt_iv_null_gen
1045 };
1046 
1047 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1048 	.ctr	   = crypt_iv_lmk_ctr,
1049 	.dtr	   = crypt_iv_lmk_dtr,
1050 	.init	   = crypt_iv_lmk_init,
1051 	.wipe	   = crypt_iv_lmk_wipe,
1052 	.generator = crypt_iv_lmk_gen,
1053 	.post	   = crypt_iv_lmk_post
1054 };
1055 
1056 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1057 	.ctr	   = crypt_iv_tcw_ctr,
1058 	.dtr	   = crypt_iv_tcw_dtr,
1059 	.init	   = crypt_iv_tcw_init,
1060 	.wipe	   = crypt_iv_tcw_wipe,
1061 	.generator = crypt_iv_tcw_gen,
1062 	.post	   = crypt_iv_tcw_post
1063 };
1064 
1065 static const struct crypt_iv_operations crypt_iv_random_ops = {
1066 	.generator = crypt_iv_random_gen
1067 };
1068 
1069 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1070 	.ctr	   = crypt_iv_eboiv_ctr,
1071 	.generator = crypt_iv_eboiv_gen
1072 };
1073 
1074 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1075 	.ctr	   = crypt_iv_elephant_ctr,
1076 	.dtr	   = crypt_iv_elephant_dtr,
1077 	.init	   = crypt_iv_elephant_init,
1078 	.wipe	   = crypt_iv_elephant_wipe,
1079 	.generator = crypt_iv_elephant_gen,
1080 	.post	   = crypt_iv_elephant_post
1081 };
1082 
1083 /*
1084  * Integrity extensions
1085  */
1086 static bool crypt_integrity_aead(struct crypt_config *cc)
1087 {
1088 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1089 }
1090 
1091 static bool crypt_integrity_hmac(struct crypt_config *cc)
1092 {
1093 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1094 }
1095 
1096 /* Get sg containing data */
1097 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1098 					     struct scatterlist *sg)
1099 {
1100 	if (unlikely(crypt_integrity_aead(cc)))
1101 		return &sg[2];
1102 
1103 	return sg;
1104 }
1105 
1106 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1107 {
1108 	struct bio_integrity_payload *bip;
1109 	unsigned int tag_len;
1110 	int ret;
1111 
1112 	if (!bio_sectors(bio) || !io->cc->tuple_size)
1113 		return 0;
1114 
1115 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1116 	if (IS_ERR(bip))
1117 		return PTR_ERR(bip);
1118 
1119 	tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1120 
1121 	bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
1122 
1123 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1124 				     tag_len, offset_in_page(io->integrity_metadata));
1125 	if (unlikely(ret != tag_len))
1126 		return -ENOMEM;
1127 
1128 	return 0;
1129 }
1130 
1131 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1132 {
1133 #ifdef CONFIG_BLK_DEV_INTEGRITY
1134 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1135 	struct mapped_device *md = dm_table_get_md(ti->table);
1136 
1137 	/* We require an underlying device with non-PI metadata */
1138 	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1139 		ti->error = "Integrity profile not supported.";
1140 		return -EINVAL;
1141 	}
1142 
1143 	if (bi->metadata_size < cc->used_tag_size) {
1144 		ti->error = "Integrity profile tag size mismatch.";
1145 		return -EINVAL;
1146 	}
1147 	cc->tuple_size = bi->metadata_size;
1148 	if (1 << bi->interval_exp != cc->sector_size) {
1149 		ti->error = "Integrity profile sector size mismatch.";
1150 		return -EINVAL;
1151 	}
1152 
1153 	if (crypt_integrity_aead(cc)) {
1154 		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1155 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1156 		       cc->integrity_tag_size, cc->integrity_iv_size);
1157 
1158 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1159 			ti->error = "Integrity AEAD auth tag size is not supported.";
1160 			return -EINVAL;
1161 		}
1162 	} else if (cc->integrity_iv_size)
1163 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1164 		       cc->integrity_iv_size);
1165 
1166 	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1167 		ti->error = "Not enough space for integrity tag in the profile.";
1168 		return -EINVAL;
1169 	}
1170 
1171 	return 0;
1172 #else
1173 	ti->error = "Integrity profile not supported.";
1174 	return -EINVAL;
1175 #endif
1176 }
1177 
1178 static void crypt_convert_init(struct crypt_config *cc,
1179 			       struct convert_context *ctx,
1180 			       struct bio *bio_out, struct bio *bio_in,
1181 			       sector_t sector)
1182 {
1183 	ctx->bio_in = bio_in;
1184 	ctx->bio_out = bio_out;
1185 	if (bio_in)
1186 		ctx->iter_in = bio_in->bi_iter;
1187 	if (bio_out)
1188 		ctx->iter_out = bio_out->bi_iter;
1189 	ctx->cc_sector = sector + cc->iv_offset;
1190 	ctx->tag_offset = 0;
1191 	init_completion(&ctx->restart);
1192 }
1193 
1194 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1195 					     void *req)
1196 {
1197 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1198 }
1199 
1200 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1201 {
1202 	return (void *)((char *)dmreq - cc->dmreq_start);
1203 }
1204 
1205 static u8 *iv_of_dmreq(struct crypt_config *cc,
1206 		       struct dm_crypt_request *dmreq)
1207 {
1208 	if (crypt_integrity_aead(cc))
1209 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1210 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1211 	else
1212 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1213 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1214 }
1215 
1216 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1217 		       struct dm_crypt_request *dmreq)
1218 {
1219 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1220 }
1221 
1222 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1223 		       struct dm_crypt_request *dmreq)
1224 {
1225 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1226 
1227 	return (__le64 *) ptr;
1228 }
1229 
1230 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1231 		       struct dm_crypt_request *dmreq)
1232 {
1233 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1234 		  cc->iv_size + sizeof(uint64_t);
1235 
1236 	return (unsigned int *)ptr;
1237 }
1238 
1239 static void *tag_from_dmreq(struct crypt_config *cc,
1240 				struct dm_crypt_request *dmreq)
1241 {
1242 	struct convert_context *ctx = dmreq->ctx;
1243 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1244 
1245 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1246 		cc->tuple_size];
1247 }
1248 
1249 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1250 			       struct dm_crypt_request *dmreq)
1251 {
1252 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1253 }
1254 
1255 static int crypt_convert_block_aead(struct crypt_config *cc,
1256 				     struct convert_context *ctx,
1257 				     struct aead_request *req,
1258 				     unsigned int tag_offset)
1259 {
1260 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1261 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1262 	struct dm_crypt_request *dmreq;
1263 	u8 *iv, *org_iv, *tag_iv, *tag;
1264 	__le64 *sector;
1265 	int r = 0;
1266 
1267 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1268 
1269 	/* Reject unexpected unaligned bio. */
1270 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1271 		return -EIO;
1272 
1273 	dmreq = dmreq_of_req(cc, req);
1274 	dmreq->iv_sector = ctx->cc_sector;
1275 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1276 		dmreq->iv_sector >>= cc->sector_shift;
1277 	dmreq->ctx = ctx;
1278 
1279 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1280 
1281 	sector = org_sector_of_dmreq(cc, dmreq);
1282 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1283 
1284 	iv = iv_of_dmreq(cc, dmreq);
1285 	org_iv = org_iv_of_dmreq(cc, dmreq);
1286 	tag = tag_from_dmreq(cc, dmreq);
1287 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1288 
1289 	/* AEAD request:
1290 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1291 	 *  | (authenticated) | (auth+encryption) |              |
1292 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1293 	 */
1294 	sg_init_table(dmreq->sg_in, 4);
1295 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1296 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1297 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1298 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1299 
1300 	sg_init_table(dmreq->sg_out, 4);
1301 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1302 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1303 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1304 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1305 
1306 	if (cc->iv_gen_ops) {
1307 		/* For READs use IV stored in integrity metadata */
1308 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1309 			memcpy(org_iv, tag_iv, cc->iv_size);
1310 		} else {
1311 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1312 			if (r < 0)
1313 				return r;
1314 			/* Store generated IV in integrity metadata */
1315 			if (cc->integrity_iv_size)
1316 				memcpy(tag_iv, org_iv, cc->iv_size);
1317 		}
1318 		/* Working copy of IV, to be modified in crypto API */
1319 		memcpy(iv, org_iv, cc->iv_size);
1320 	}
1321 
1322 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1323 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1324 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1325 				       cc->sector_size, iv);
1326 		r = crypto_aead_encrypt(req);
1327 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1328 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1329 			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1330 	} else {
1331 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1332 				       cc->sector_size + cc->integrity_tag_size, iv);
1333 		r = crypto_aead_decrypt(req);
1334 	}
1335 
1336 	if (r == -EBADMSG) {
1337 		sector_t s = le64_to_cpu(*sector);
1338 
1339 		ctx->aead_failed = true;
1340 		if (ctx->aead_recheck) {
1341 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1342 				    ctx->bio_in->bi_bdev, s);
1343 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1344 					 ctx->bio_in, s, 0);
1345 		}
1346 	}
1347 
1348 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1349 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1350 
1351 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1352 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1353 
1354 	return r;
1355 }
1356 
1357 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1358 					struct convert_context *ctx,
1359 					struct skcipher_request *req,
1360 					unsigned int tag_offset)
1361 {
1362 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1363 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1364 	struct scatterlist *sg_in, *sg_out;
1365 	struct dm_crypt_request *dmreq;
1366 	u8 *iv, *org_iv, *tag_iv;
1367 	__le64 *sector;
1368 	int r = 0;
1369 
1370 	/* Reject unexpected unaligned bio. */
1371 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1372 		return -EIO;
1373 
1374 	dmreq = dmreq_of_req(cc, req);
1375 	dmreq->iv_sector = ctx->cc_sector;
1376 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1377 		dmreq->iv_sector >>= cc->sector_shift;
1378 	dmreq->ctx = ctx;
1379 
1380 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1381 
1382 	iv = iv_of_dmreq(cc, dmreq);
1383 	org_iv = org_iv_of_dmreq(cc, dmreq);
1384 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1385 
1386 	sector = org_sector_of_dmreq(cc, dmreq);
1387 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1388 
1389 	/* For skcipher we use only the first sg item */
1390 	sg_in  = &dmreq->sg_in[0];
1391 	sg_out = &dmreq->sg_out[0];
1392 
1393 	sg_init_table(sg_in, 1);
1394 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1395 
1396 	sg_init_table(sg_out, 1);
1397 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1398 
1399 	if (cc->iv_gen_ops) {
1400 		/* For READs use IV stored in integrity metadata */
1401 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1402 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1403 		} else {
1404 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1405 			if (r < 0)
1406 				return r;
1407 			/* Data can be already preprocessed in generator */
1408 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1409 				sg_in = sg_out;
1410 			/* Store generated IV in integrity metadata */
1411 			if (cc->integrity_iv_size)
1412 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1413 		}
1414 		/* Working copy of IV, to be modified in crypto API */
1415 		memcpy(iv, org_iv, cc->iv_size);
1416 	}
1417 
1418 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1419 
1420 	if (bio_data_dir(ctx->bio_in) == WRITE)
1421 		r = crypto_skcipher_encrypt(req);
1422 	else
1423 		r = crypto_skcipher_decrypt(req);
1424 
1425 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1426 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1427 
1428 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1429 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1430 
1431 	return r;
1432 }
1433 
1434 static void kcryptd_async_done(void *async_req, int error);
1435 
1436 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1437 				     struct convert_context *ctx)
1438 {
1439 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1440 
1441 	if (!ctx->r.req) {
1442 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1443 		if (!ctx->r.req)
1444 			return -ENOMEM;
1445 	}
1446 
1447 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1448 
1449 	/*
1450 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1451 	 * requests if driver request queue is full.
1452 	 */
1453 	skcipher_request_set_callback(ctx->r.req,
1454 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1455 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1456 
1457 	return 0;
1458 }
1459 
1460 static int crypt_alloc_req_aead(struct crypt_config *cc,
1461 				 struct convert_context *ctx)
1462 {
1463 	if (!ctx->r.req_aead) {
1464 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1465 		if (!ctx->r.req_aead)
1466 			return -ENOMEM;
1467 	}
1468 
1469 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1470 
1471 	/*
1472 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1473 	 * requests if driver request queue is full.
1474 	 */
1475 	aead_request_set_callback(ctx->r.req_aead,
1476 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1477 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1478 
1479 	return 0;
1480 }
1481 
1482 static int crypt_alloc_req(struct crypt_config *cc,
1483 			    struct convert_context *ctx)
1484 {
1485 	if (crypt_integrity_aead(cc))
1486 		return crypt_alloc_req_aead(cc, ctx);
1487 	else
1488 		return crypt_alloc_req_skcipher(cc, ctx);
1489 }
1490 
1491 static void crypt_free_req_skcipher(struct crypt_config *cc,
1492 				    struct skcipher_request *req, struct bio *base_bio)
1493 {
1494 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1495 
1496 	if ((struct skcipher_request *)(io + 1) != req)
1497 		mempool_free(req, &cc->req_pool);
1498 }
1499 
1500 static void crypt_free_req_aead(struct crypt_config *cc,
1501 				struct aead_request *req, struct bio *base_bio)
1502 {
1503 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1504 
1505 	if ((struct aead_request *)(io + 1) != req)
1506 		mempool_free(req, &cc->req_pool);
1507 }
1508 
1509 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1510 {
1511 	if (crypt_integrity_aead(cc))
1512 		crypt_free_req_aead(cc, req, base_bio);
1513 	else
1514 		crypt_free_req_skcipher(cc, req, base_bio);
1515 }
1516 
1517 /*
1518  * Encrypt / decrypt data from one bio to another one (can be the same one)
1519  */
1520 static blk_status_t crypt_convert(struct crypt_config *cc,
1521 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1522 {
1523 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1524 	int r;
1525 
1526 	/*
1527 	 * if reset_pending is set we are dealing with the bio for the first time,
1528 	 * else we're continuing to work on the previous bio, so don't mess with
1529 	 * the cc_pending counter
1530 	 */
1531 	if (reset_pending)
1532 		atomic_set(&ctx->cc_pending, 1);
1533 
1534 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1535 
1536 		r = crypt_alloc_req(cc, ctx);
1537 		if (r) {
1538 			complete(&ctx->restart);
1539 			return BLK_STS_DEV_RESOURCE;
1540 		}
1541 
1542 		atomic_inc(&ctx->cc_pending);
1543 
1544 		if (crypt_integrity_aead(cc))
1545 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, ctx->tag_offset);
1546 		else
1547 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->tag_offset);
1548 
1549 		switch (r) {
1550 		/*
1551 		 * The request was queued by a crypto driver
1552 		 * but the driver request queue is full, let's wait.
1553 		 */
1554 		case -EBUSY:
1555 			if (in_interrupt()) {
1556 				if (try_wait_for_completion(&ctx->restart)) {
1557 					/*
1558 					 * we don't have to block to wait for completion,
1559 					 * so proceed
1560 					 */
1561 				} else {
1562 					/*
1563 					 * we can't wait for completion without blocking
1564 					 * exit and continue processing in a workqueue
1565 					 */
1566 					ctx->r.req = NULL;
1567 					ctx->tag_offset++;
1568 					ctx->cc_sector += sector_step;
1569 					return BLK_STS_DEV_RESOURCE;
1570 				}
1571 			} else {
1572 				wait_for_completion(&ctx->restart);
1573 			}
1574 			reinit_completion(&ctx->restart);
1575 			fallthrough;
1576 		/*
1577 		 * The request is queued and processed asynchronously,
1578 		 * completion function kcryptd_async_done() will be called.
1579 		 */
1580 		case -EINPROGRESS:
1581 			ctx->r.req = NULL;
1582 			ctx->tag_offset++;
1583 			ctx->cc_sector += sector_step;
1584 			continue;
1585 		/*
1586 		 * The request was already processed (synchronously).
1587 		 */
1588 		case 0:
1589 			atomic_dec(&ctx->cc_pending);
1590 			ctx->cc_sector += sector_step;
1591 			ctx->tag_offset++;
1592 			if (!atomic)
1593 				cond_resched();
1594 			continue;
1595 		/*
1596 		 * There was a data integrity error.
1597 		 */
1598 		case -EBADMSG:
1599 			atomic_dec(&ctx->cc_pending);
1600 			return BLK_STS_PROTECTION;
1601 		/*
1602 		 * There was an error while processing the request.
1603 		 */
1604 		default:
1605 			atomic_dec(&ctx->cc_pending);
1606 			return BLK_STS_IOERR;
1607 		}
1608 	}
1609 
1610 	return 0;
1611 }
1612 
1613 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1614 
1615 /*
1616  * Generate a new unfragmented bio with the given size
1617  * This should never violate the device limitations (but if it did then block
1618  * core should split the bio as needed).
1619  *
1620  * This function may be called concurrently. If we allocate from the mempool
1621  * concurrently, there is a possibility of deadlock. For example, if we have
1622  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1623  * the mempool concurrently, it may deadlock in a situation where both processes
1624  * have allocated 128 pages and the mempool is exhausted.
1625  *
1626  * In order to avoid this scenario we allocate the pages under a mutex.
1627  *
1628  * In order to not degrade performance with excessive locking, we try
1629  * non-blocking allocations without a mutex first but on failure we fallback
1630  * to blocking allocations with a mutex.
1631  *
1632  * In order to reduce allocation overhead, we try to allocate compound pages in
1633  * the first pass. If they are not available, we fall back to the mempool.
1634  */
1635 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1636 {
1637 	struct crypt_config *cc = io->cc;
1638 	struct bio *clone;
1639 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1640 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1641 	unsigned int remaining_size;
1642 	unsigned int order = MAX_PAGE_ORDER;
1643 
1644 retry:
1645 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1646 		mutex_lock(&cc->bio_alloc_lock);
1647 
1648 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1649 				 GFP_NOIO, &cc->bs);
1650 	clone->bi_private = io;
1651 	clone->bi_end_io = crypt_endio;
1652 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1653 	clone->bi_iter.bi_sector = cc->start + io->sector;
1654 
1655 	remaining_size = size;
1656 
1657 	while (remaining_size) {
1658 		struct page *pages;
1659 		unsigned size_to_add;
1660 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1661 		order = min(order, remaining_order);
1662 
1663 		while (order > 0) {
1664 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1665 					(1 << order) > dm_crypt_pages_per_client))
1666 				goto decrease_order;
1667 			pages = alloc_pages(gfp_mask
1668 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1669 				order);
1670 			if (likely(pages != NULL)) {
1671 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1672 				goto have_pages;
1673 			}
1674 decrease_order:
1675 			order--;
1676 		}
1677 
1678 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1679 		if (!pages) {
1680 			crypt_free_buffer_pages(cc, clone);
1681 			bio_put(clone);
1682 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1683 			order = 0;
1684 			goto retry;
1685 		}
1686 
1687 have_pages:
1688 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1689 		__bio_add_page(clone, pages, size_to_add, 0);
1690 		remaining_size -= size_to_add;
1691 	}
1692 
1693 	/* Allocate space for integrity tags */
1694 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1695 		crypt_free_buffer_pages(cc, clone);
1696 		bio_put(clone);
1697 		clone = NULL;
1698 	}
1699 
1700 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1701 		mutex_unlock(&cc->bio_alloc_lock);
1702 
1703 	return clone;
1704 }
1705 
1706 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1707 {
1708 	struct folio_iter fi;
1709 
1710 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1711 		bio_for_each_folio_all(fi, clone) {
1712 			if (folio_test_large(fi.folio)) {
1713 				percpu_counter_sub(&cc->n_allocated_pages,
1714 						folio_nr_pages(fi.folio));
1715 				folio_put(fi.folio);
1716 			} else {
1717 				mempool_free(&fi.folio->page, &cc->page_pool);
1718 			}
1719 		}
1720 	}
1721 }
1722 
1723 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1724 			  struct bio *bio, sector_t sector)
1725 {
1726 	io->cc = cc;
1727 	io->base_bio = bio;
1728 	io->sector = sector;
1729 	io->error = 0;
1730 	io->ctx.aead_recheck = false;
1731 	io->ctx.aead_failed = false;
1732 	io->ctx.r.req = NULL;
1733 	io->integrity_metadata = NULL;
1734 	io->integrity_metadata_from_pool = false;
1735 	atomic_set(&io->io_pending, 0);
1736 }
1737 
1738 static void crypt_inc_pending(struct dm_crypt_io *io)
1739 {
1740 	atomic_inc(&io->io_pending);
1741 }
1742 
1743 static void kcryptd_queue_read(struct dm_crypt_io *io);
1744 
1745 /*
1746  * One of the bios was finished. Check for completion of
1747  * the whole request and correctly clean up the buffer.
1748  */
1749 static void crypt_dec_pending(struct dm_crypt_io *io)
1750 {
1751 	struct crypt_config *cc = io->cc;
1752 	struct bio *base_bio = io->base_bio;
1753 	blk_status_t error = io->error;
1754 
1755 	if (!atomic_dec_and_test(&io->io_pending))
1756 		return;
1757 
1758 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1759 	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1760 		io->ctx.aead_recheck = true;
1761 		io->ctx.aead_failed = false;
1762 		io->error = 0;
1763 		kcryptd_queue_read(io);
1764 		return;
1765 	}
1766 
1767 	if (io->ctx.r.req)
1768 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1769 
1770 	if (unlikely(io->integrity_metadata_from_pool))
1771 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1772 	else
1773 		kfree(io->integrity_metadata);
1774 
1775 	base_bio->bi_status = error;
1776 
1777 	bio_endio(base_bio);
1778 }
1779 
1780 /*
1781  * kcryptd/kcryptd_io:
1782  *
1783  * Needed because it would be very unwise to do decryption in an
1784  * interrupt context.
1785  *
1786  * kcryptd performs the actual encryption or decryption.
1787  *
1788  * kcryptd_io performs the IO submission.
1789  *
1790  * They must be separated as otherwise the final stages could be
1791  * starved by new requests which can block in the first stages due
1792  * to memory allocation.
1793  *
1794  * The work is done per CPU global for all dm-crypt instances.
1795  * They should not depend on each other and do not block.
1796  */
1797 static void crypt_endio(struct bio *clone)
1798 {
1799 	struct dm_crypt_io *io = clone->bi_private;
1800 	struct crypt_config *cc = io->cc;
1801 	unsigned int rw = bio_data_dir(clone);
1802 	blk_status_t error = clone->bi_status;
1803 
1804 	if (io->ctx.aead_recheck && !error) {
1805 		kcryptd_queue_crypt(io);
1806 		return;
1807 	}
1808 
1809 	/*
1810 	 * free the processed pages
1811 	 */
1812 	if (rw == WRITE || io->ctx.aead_recheck)
1813 		crypt_free_buffer_pages(cc, clone);
1814 
1815 	bio_put(clone);
1816 
1817 	if (rw == READ && !error) {
1818 		kcryptd_queue_crypt(io);
1819 		return;
1820 	}
1821 
1822 	if (unlikely(error))
1823 		io->error = error;
1824 
1825 	crypt_dec_pending(io);
1826 }
1827 
1828 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1829 
1830 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1831 {
1832 	struct crypt_config *cc = io->cc;
1833 	struct bio *clone;
1834 
1835 	if (io->ctx.aead_recheck) {
1836 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1837 			return 1;
1838 		crypt_inc_pending(io);
1839 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1840 		if (unlikely(!clone)) {
1841 			crypt_dec_pending(io);
1842 			return 1;
1843 		}
1844 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1845 		io->saved_bi_iter = clone->bi_iter;
1846 		dm_submit_bio_remap(io->base_bio, clone);
1847 		return 0;
1848 	}
1849 
1850 	/*
1851 	 * We need the original biovec array in order to decrypt the whole bio
1852 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1853 	 * worry about the block layer modifying the biovec array; so leverage
1854 	 * bio_alloc_clone().
1855 	 */
1856 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1857 	if (!clone)
1858 		return 1;
1859 
1860 	clone->bi_iter.bi_sector = cc->start + io->sector;
1861 	clone->bi_private = io;
1862 	clone->bi_end_io = crypt_endio;
1863 
1864 	crypt_inc_pending(io);
1865 
1866 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1867 		crypt_dec_pending(io);
1868 		bio_put(clone);
1869 		return 1;
1870 	}
1871 
1872 	dm_submit_bio_remap(io->base_bio, clone);
1873 	return 0;
1874 }
1875 
1876 static void kcryptd_io_read_work(struct work_struct *work)
1877 {
1878 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1879 
1880 	crypt_inc_pending(io);
1881 	if (kcryptd_io_read(io, GFP_NOIO))
1882 		io->error = BLK_STS_RESOURCE;
1883 	crypt_dec_pending(io);
1884 }
1885 
1886 static void kcryptd_queue_read(struct dm_crypt_io *io)
1887 {
1888 	struct crypt_config *cc = io->cc;
1889 
1890 	INIT_WORK(&io->work, kcryptd_io_read_work);
1891 	queue_work(cc->io_queue, &io->work);
1892 }
1893 
1894 static void kcryptd_io_write(struct dm_crypt_io *io)
1895 {
1896 	struct bio *clone = io->ctx.bio_out;
1897 
1898 	dm_submit_bio_remap(io->base_bio, clone);
1899 }
1900 
1901 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1902 
1903 static int dmcrypt_write(void *data)
1904 {
1905 	struct crypt_config *cc = data;
1906 	struct dm_crypt_io *io;
1907 
1908 	while (1) {
1909 		struct rb_root write_tree;
1910 		struct blk_plug plug;
1911 
1912 		spin_lock_irq(&cc->write_thread_lock);
1913 continue_locked:
1914 
1915 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1916 			goto pop_from_list;
1917 
1918 		set_current_state(TASK_INTERRUPTIBLE);
1919 
1920 		spin_unlock_irq(&cc->write_thread_lock);
1921 
1922 		if (unlikely(kthread_should_stop())) {
1923 			set_current_state(TASK_RUNNING);
1924 			break;
1925 		}
1926 
1927 		schedule();
1928 
1929 		spin_lock_irq(&cc->write_thread_lock);
1930 		goto continue_locked;
1931 
1932 pop_from_list:
1933 		write_tree = cc->write_tree;
1934 		cc->write_tree = RB_ROOT;
1935 		spin_unlock_irq(&cc->write_thread_lock);
1936 
1937 		BUG_ON(rb_parent(write_tree.rb_node));
1938 
1939 		/*
1940 		 * Note: we cannot walk the tree here with rb_next because
1941 		 * the structures may be freed when kcryptd_io_write is called.
1942 		 */
1943 		blk_start_plug(&plug);
1944 		do {
1945 			io = crypt_io_from_node(rb_first(&write_tree));
1946 			rb_erase(&io->rb_node, &write_tree);
1947 			kcryptd_io_write(io);
1948 			cond_resched();
1949 		} while (!RB_EMPTY_ROOT(&write_tree));
1950 		blk_finish_plug(&plug);
1951 	}
1952 	return 0;
1953 }
1954 
1955 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1956 {
1957 	struct bio *clone = io->ctx.bio_out;
1958 	struct crypt_config *cc = io->cc;
1959 	unsigned long flags;
1960 	sector_t sector;
1961 	struct rb_node **rbp, *parent;
1962 
1963 	if (unlikely(io->error)) {
1964 		crypt_free_buffer_pages(cc, clone);
1965 		bio_put(clone);
1966 		crypt_dec_pending(io);
1967 		return;
1968 	}
1969 
1970 	/* crypt_convert should have filled the clone bio */
1971 	BUG_ON(io->ctx.iter_out.bi_size);
1972 
1973 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1974 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1975 		dm_submit_bio_remap(io->base_bio, clone);
1976 		return;
1977 	}
1978 
1979 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1980 	if (RB_EMPTY_ROOT(&cc->write_tree))
1981 		wake_up_process(cc->write_thread);
1982 	rbp = &cc->write_tree.rb_node;
1983 	parent = NULL;
1984 	sector = io->sector;
1985 	while (*rbp) {
1986 		parent = *rbp;
1987 		if (sector < crypt_io_from_node(parent)->sector)
1988 			rbp = &(*rbp)->rb_left;
1989 		else
1990 			rbp = &(*rbp)->rb_right;
1991 	}
1992 	rb_link_node(&io->rb_node, parent, rbp);
1993 	rb_insert_color(&io->rb_node, &cc->write_tree);
1994 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1995 }
1996 
1997 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1998 				       struct convert_context *ctx)
1999 
2000 {
2001 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2002 		return false;
2003 
2004 	/*
2005 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2006 	 * constraints so they do not need to be issued inline by
2007 	 * kcryptd_crypt_write_convert().
2008 	 */
2009 	switch (bio_op(ctx->bio_in)) {
2010 	case REQ_OP_WRITE:
2011 	case REQ_OP_WRITE_ZEROES:
2012 		return true;
2013 	default:
2014 		return false;
2015 	}
2016 }
2017 
2018 static void kcryptd_crypt_write_continue(struct work_struct *work)
2019 {
2020 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2021 	struct crypt_config *cc = io->cc;
2022 	struct convert_context *ctx = &io->ctx;
2023 	int crypt_finished;
2024 	blk_status_t r;
2025 
2026 	wait_for_completion(&ctx->restart);
2027 	reinit_completion(&ctx->restart);
2028 
2029 	r = crypt_convert(cc, &io->ctx, false, false);
2030 	if (r)
2031 		io->error = r;
2032 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2033 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2034 		/* Wait for completion signaled by kcryptd_async_done() */
2035 		wait_for_completion(&ctx->restart);
2036 		crypt_finished = 1;
2037 	}
2038 
2039 	/* Encryption was already finished, submit io now */
2040 	if (crypt_finished)
2041 		kcryptd_crypt_write_io_submit(io, 0);
2042 
2043 	crypt_dec_pending(io);
2044 }
2045 
2046 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2047 {
2048 	struct crypt_config *cc = io->cc;
2049 	struct convert_context *ctx = &io->ctx;
2050 	struct bio *clone;
2051 	int crypt_finished;
2052 	blk_status_t r;
2053 
2054 	/*
2055 	 * Prevent io from disappearing until this function completes.
2056 	 */
2057 	crypt_inc_pending(io);
2058 	crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector);
2059 
2060 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2061 	if (unlikely(!clone)) {
2062 		io->error = BLK_STS_IOERR;
2063 		goto dec;
2064 	}
2065 
2066 	io->ctx.bio_out = clone;
2067 	io->ctx.iter_out = clone->bi_iter;
2068 
2069 	if (crypt_integrity_aead(cc)) {
2070 		bio_copy_data(clone, io->base_bio);
2071 		io->ctx.bio_in = clone;
2072 		io->ctx.iter_in = clone->bi_iter;
2073 	}
2074 
2075 	crypt_inc_pending(io);
2076 	r = crypt_convert(cc, ctx,
2077 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2078 	/*
2079 	 * Crypto API backlogged the request, because its queue was full
2080 	 * and we're in softirq context, so continue from a workqueue
2081 	 * (TODO: is it actually possible to be in softirq in the write path?)
2082 	 */
2083 	if (r == BLK_STS_DEV_RESOURCE) {
2084 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2085 		queue_work(cc->crypt_queue, &io->work);
2086 		return;
2087 	}
2088 	if (r)
2089 		io->error = r;
2090 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2091 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2092 		/* Wait for completion signaled by kcryptd_async_done() */
2093 		wait_for_completion(&ctx->restart);
2094 		crypt_finished = 1;
2095 	}
2096 
2097 	/* Encryption was already finished, submit io now */
2098 	if (crypt_finished)
2099 		kcryptd_crypt_write_io_submit(io, 0);
2100 
2101 dec:
2102 	crypt_dec_pending(io);
2103 }
2104 
2105 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2106 {
2107 	if (io->ctx.aead_recheck) {
2108 		if (!io->error) {
2109 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2110 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2111 		}
2112 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2113 		bio_put(io->ctx.bio_in);
2114 	}
2115 	crypt_dec_pending(io);
2116 }
2117 
2118 static void kcryptd_crypt_read_continue(struct work_struct *work)
2119 {
2120 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2121 	struct crypt_config *cc = io->cc;
2122 	blk_status_t r;
2123 
2124 	wait_for_completion(&io->ctx.restart);
2125 	reinit_completion(&io->ctx.restart);
2126 
2127 	r = crypt_convert(cc, &io->ctx, false, false);
2128 	if (r)
2129 		io->error = r;
2130 
2131 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2132 		kcryptd_crypt_read_done(io);
2133 
2134 	crypt_dec_pending(io);
2135 }
2136 
2137 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2138 {
2139 	struct crypt_config *cc = io->cc;
2140 	blk_status_t r;
2141 
2142 	crypt_inc_pending(io);
2143 
2144 	if (io->ctx.aead_recheck) {
2145 		r = crypt_convert(cc, &io->ctx,
2146 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2147 	} else {
2148 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2149 				   io->sector);
2150 
2151 		r = crypt_convert(cc, &io->ctx,
2152 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2153 	}
2154 	/*
2155 	 * Crypto API backlogged the request, because its queue was full
2156 	 * and we're in softirq context, so continue from a workqueue
2157 	 */
2158 	if (r == BLK_STS_DEV_RESOURCE) {
2159 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2160 		queue_work(cc->crypt_queue, &io->work);
2161 		return;
2162 	}
2163 	if (r)
2164 		io->error = r;
2165 
2166 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2167 		kcryptd_crypt_read_done(io);
2168 
2169 	crypt_dec_pending(io);
2170 }
2171 
2172 static void kcryptd_async_done(void *data, int error)
2173 {
2174 	struct dm_crypt_request *dmreq = data;
2175 	struct convert_context *ctx = dmreq->ctx;
2176 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2177 	struct crypt_config *cc = io->cc;
2178 
2179 	/*
2180 	 * A request from crypto driver backlog is going to be processed now,
2181 	 * finish the completion and continue in crypt_convert().
2182 	 * (Callback will be called for the second time for this request.)
2183 	 */
2184 	if (error == -EINPROGRESS) {
2185 		complete(&ctx->restart);
2186 		return;
2187 	}
2188 
2189 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2190 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2191 
2192 	if (error == -EBADMSG) {
2193 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2194 
2195 		ctx->aead_failed = true;
2196 		if (ctx->aead_recheck) {
2197 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2198 				    ctx->bio_in->bi_bdev, s);
2199 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2200 					 ctx->bio_in, s, 0);
2201 		}
2202 		io->error = BLK_STS_PROTECTION;
2203 	} else if (error < 0)
2204 		io->error = BLK_STS_IOERR;
2205 
2206 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2207 
2208 	if (!atomic_dec_and_test(&ctx->cc_pending))
2209 		return;
2210 
2211 	/*
2212 	 * The request is fully completed: for inline writes, let
2213 	 * kcryptd_crypt_write_convert() do the IO submission.
2214 	 */
2215 	if (bio_data_dir(io->base_bio) == READ) {
2216 		kcryptd_crypt_read_done(io);
2217 		return;
2218 	}
2219 
2220 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2221 		complete(&ctx->restart);
2222 		return;
2223 	}
2224 
2225 	kcryptd_crypt_write_io_submit(io, 1);
2226 }
2227 
2228 static void kcryptd_crypt(struct work_struct *work)
2229 {
2230 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2231 
2232 	if (bio_data_dir(io->base_bio) == READ)
2233 		kcryptd_crypt_read_convert(io);
2234 	else
2235 		kcryptd_crypt_write_convert(io);
2236 }
2237 
2238 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2239 {
2240 	struct crypt_config *cc = io->cc;
2241 
2242 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2243 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2244 		/*
2245 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2246 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2247 		 * it is being executed with irqs disabled.
2248 		 */
2249 		if (in_hardirq() || irqs_disabled()) {
2250 			INIT_WORK(&io->work, kcryptd_crypt);
2251 			queue_work(system_bh_wq, &io->work);
2252 			return;
2253 		} else {
2254 			kcryptd_crypt(&io->work);
2255 			return;
2256 		}
2257 	}
2258 
2259 	INIT_WORK(&io->work, kcryptd_crypt);
2260 	queue_work(cc->crypt_queue, &io->work);
2261 }
2262 
2263 static void crypt_free_tfms_aead(struct crypt_config *cc)
2264 {
2265 	if (!cc->cipher_tfm.tfms_aead)
2266 		return;
2267 
2268 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2269 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2270 		cc->cipher_tfm.tfms_aead[0] = NULL;
2271 	}
2272 
2273 	kfree(cc->cipher_tfm.tfms_aead);
2274 	cc->cipher_tfm.tfms_aead = NULL;
2275 }
2276 
2277 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2278 {
2279 	unsigned int i;
2280 
2281 	if (!cc->cipher_tfm.tfms)
2282 		return;
2283 
2284 	for (i = 0; i < cc->tfms_count; i++)
2285 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2286 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2287 			cc->cipher_tfm.tfms[i] = NULL;
2288 		}
2289 
2290 	kfree(cc->cipher_tfm.tfms);
2291 	cc->cipher_tfm.tfms = NULL;
2292 }
2293 
2294 static void crypt_free_tfms(struct crypt_config *cc)
2295 {
2296 	if (crypt_integrity_aead(cc))
2297 		crypt_free_tfms_aead(cc);
2298 	else
2299 		crypt_free_tfms_skcipher(cc);
2300 }
2301 
2302 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2303 {
2304 	unsigned int i;
2305 	int err;
2306 
2307 	cc->cipher_tfm.tfms = kzalloc_objs(struct crypto_skcipher *,
2308 					   cc->tfms_count);
2309 	if (!cc->cipher_tfm.tfms)
2310 		return -ENOMEM;
2311 
2312 	for (i = 0; i < cc->tfms_count; i++) {
2313 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2314 						CRYPTO_ALG_ALLOCATES_MEMORY);
2315 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2316 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2317 			crypt_free_tfms(cc);
2318 			return err;
2319 		}
2320 	}
2321 
2322 	/*
2323 	 * dm-crypt performance can vary greatly depending on which crypto
2324 	 * algorithm implementation is used.  Help people debug performance
2325 	 * problems by logging the ->cra_driver_name.
2326 	 */
2327 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2328 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2329 	return 0;
2330 }
2331 
2332 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2333 {
2334 	int err;
2335 
2336 	cc->cipher_tfm.tfms = kmalloc_obj(struct crypto_skcipher *);
2337 	if (!cc->cipher_tfm.tfms)
2338 		return -ENOMEM;
2339 
2340 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2341 						CRYPTO_ALG_ALLOCATES_MEMORY);
2342 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2343 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2344 		crypt_free_tfms(cc);
2345 		return err;
2346 	}
2347 
2348 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2349 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2350 	return 0;
2351 }
2352 
2353 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2354 {
2355 	if (crypt_integrity_aead(cc))
2356 		return crypt_alloc_tfms_aead(cc, ciphermode);
2357 	else
2358 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2359 }
2360 
2361 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2362 {
2363 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2364 }
2365 
2366 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2367 {
2368 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2369 }
2370 
2371 /*
2372  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2373  * the key must be for some reason in special format.
2374  * This funcion converts cc->key to this special format.
2375  */
2376 static void crypt_copy_authenckey(char *p, const void *key,
2377 				  unsigned int enckeylen, unsigned int authkeylen)
2378 {
2379 	struct crypto_authenc_key_param *param;
2380 	struct rtattr *rta;
2381 
2382 	rta = (struct rtattr *)p;
2383 	param = RTA_DATA(rta);
2384 	param->enckeylen = cpu_to_be32(enckeylen);
2385 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2386 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2387 	p += RTA_SPACE(sizeof(*param));
2388 	memcpy(p, key + enckeylen, authkeylen);
2389 	p += authkeylen;
2390 	memcpy(p, key, enckeylen);
2391 }
2392 
2393 static int crypt_setkey(struct crypt_config *cc)
2394 {
2395 	unsigned int subkey_size;
2396 	int err = 0, i, r;
2397 
2398 	/* Ignore extra keys (which are used for IV etc) */
2399 	subkey_size = crypt_subkey_size(cc);
2400 
2401 	if (crypt_integrity_hmac(cc)) {
2402 		if (subkey_size < cc->key_mac_size)
2403 			return -EINVAL;
2404 
2405 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2406 				      subkey_size - cc->key_mac_size,
2407 				      cc->key_mac_size);
2408 	}
2409 
2410 	for (i = 0; i < cc->tfms_count; i++) {
2411 		if (crypt_integrity_hmac(cc))
2412 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2413 				cc->authenc_key, crypt_authenckey_size(cc));
2414 		else if (crypt_integrity_aead(cc))
2415 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2416 					       cc->key + (i * subkey_size),
2417 					       subkey_size);
2418 		else
2419 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2420 						   cc->key + (i * subkey_size),
2421 						   subkey_size);
2422 		if (r)
2423 			err = r;
2424 	}
2425 
2426 	if (crypt_integrity_hmac(cc))
2427 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2428 
2429 	return err;
2430 }
2431 
2432 #ifdef CONFIG_KEYS
2433 
2434 static bool contains_whitespace(const char *str)
2435 {
2436 	while (*str)
2437 		if (isspace(*str++))
2438 			return true;
2439 	return false;
2440 }
2441 
2442 static int set_key_user(struct crypt_config *cc, struct key *key)
2443 {
2444 	const struct user_key_payload *ukp;
2445 
2446 	ukp = user_key_payload_locked(key);
2447 	if (!ukp)
2448 		return -EKEYREVOKED;
2449 
2450 	if (cc->key_size != ukp->datalen)
2451 		return -EINVAL;
2452 
2453 	memcpy(cc->key, ukp->data, cc->key_size);
2454 
2455 	return 0;
2456 }
2457 
2458 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2459 {
2460 	const struct encrypted_key_payload *ekp;
2461 
2462 	ekp = key->payload.data[0];
2463 	if (!ekp)
2464 		return -EKEYREVOKED;
2465 
2466 	if (cc->key_size != ekp->decrypted_datalen)
2467 		return -EINVAL;
2468 
2469 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2470 
2471 	return 0;
2472 }
2473 
2474 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2475 {
2476 	const struct trusted_key_payload *tkp;
2477 
2478 	tkp = key->payload.data[0];
2479 	if (!tkp)
2480 		return -EKEYREVOKED;
2481 
2482 	if (cc->key_size != tkp->key_len)
2483 		return -EINVAL;
2484 
2485 	memcpy(cc->key, tkp->key, cc->key_size);
2486 
2487 	return 0;
2488 }
2489 
2490 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2491 {
2492 	char *new_key_string, *key_desc;
2493 	int ret;
2494 	struct key_type *type;
2495 	struct key *key;
2496 	int (*set_key)(struct crypt_config *cc, struct key *key);
2497 
2498 	/*
2499 	 * Reject key_string with whitespace. dm core currently lacks code for
2500 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2501 	 */
2502 	if (contains_whitespace(key_string)) {
2503 		DMERR("whitespace chars not allowed in key string");
2504 		return -EINVAL;
2505 	}
2506 
2507 	/* look for next ':' separating key_type from key_description */
2508 	key_desc = strchr(key_string, ':');
2509 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2510 		return -EINVAL;
2511 
2512 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2513 		type = &key_type_logon;
2514 		set_key = set_key_user;
2515 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2516 		type = &key_type_user;
2517 		set_key = set_key_user;
2518 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2519 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2520 		type = &key_type_encrypted;
2521 		set_key = set_key_encrypted;
2522 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2523 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2524 		type = &key_type_trusted;
2525 		set_key = set_key_trusted;
2526 	} else {
2527 		return -EINVAL;
2528 	}
2529 
2530 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2531 	if (!new_key_string)
2532 		return -ENOMEM;
2533 
2534 	key = request_key(type, key_desc + 1, NULL);
2535 	if (IS_ERR(key)) {
2536 		ret = PTR_ERR(key);
2537 		goto free_new_key_string;
2538 	}
2539 
2540 	down_read(&key->sem);
2541 	ret = set_key(cc, key);
2542 	up_read(&key->sem);
2543 	key_put(key);
2544 	if (ret < 0)
2545 		goto free_new_key_string;
2546 
2547 	/* clear the flag since following operations may invalidate previously valid key */
2548 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2549 
2550 	ret = crypt_setkey(cc);
2551 	if (ret)
2552 		goto free_new_key_string;
2553 
2554 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2555 	kfree_sensitive(cc->key_string);
2556 	cc->key_string = new_key_string;
2557 	return 0;
2558 
2559 free_new_key_string:
2560 	kfree_sensitive(new_key_string);
2561 	return ret;
2562 }
2563 
2564 static int get_key_size(char **key_string)
2565 {
2566 	char *colon, dummy;
2567 	int ret;
2568 
2569 	if (*key_string[0] != ':')
2570 		return strlen(*key_string) >> 1;
2571 
2572 	/* look for next ':' in key string */
2573 	colon = strpbrk(*key_string + 1, ":");
2574 	if (!colon)
2575 		return -EINVAL;
2576 
2577 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2578 		return -EINVAL;
2579 
2580 	*key_string = colon;
2581 
2582 	/* remaining key string should be :<logon|user>:<key_desc> */
2583 
2584 	return ret;
2585 }
2586 
2587 #else
2588 
2589 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2590 {
2591 	return -EINVAL;
2592 }
2593 
2594 static int get_key_size(char **key_string)
2595 {
2596 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2597 }
2598 
2599 #endif /* CONFIG_KEYS */
2600 
2601 static int crypt_set_key(struct crypt_config *cc, char *key)
2602 {
2603 	int r = -EINVAL;
2604 	int key_string_len = strlen(key);
2605 
2606 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2607 	if (!cc->key_size && strcmp(key, "-"))
2608 		goto out;
2609 
2610 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2611 	if (key[0] == ':') {
2612 		r = crypt_set_keyring_key(cc, key + 1);
2613 		goto out;
2614 	}
2615 
2616 	/* clear the flag since following operations may invalidate previously valid key */
2617 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2618 
2619 	/* wipe references to any kernel keyring key */
2620 	kfree_sensitive(cc->key_string);
2621 	cc->key_string = NULL;
2622 
2623 	/* Decode key from its hex representation. */
2624 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2625 		goto out;
2626 
2627 	r = crypt_setkey(cc);
2628 	if (!r)
2629 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2630 
2631 out:
2632 	/* Hex key string not needed after here, so wipe it. */
2633 	memset(key, '0', key_string_len);
2634 
2635 	return r;
2636 }
2637 
2638 static int crypt_wipe_key(struct crypt_config *cc)
2639 {
2640 	int r;
2641 
2642 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2643 	get_random_bytes(&cc->key, cc->key_size);
2644 
2645 	/* Wipe IV private keys */
2646 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe)
2647 		cc->iv_gen_ops->wipe(cc);
2648 
2649 	kfree_sensitive(cc->key_string);
2650 	cc->key_string = NULL;
2651 	r = crypt_setkey(cc);
2652 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2653 
2654 	return r;
2655 }
2656 
2657 static void crypt_calculate_pages_per_client(void)
2658 {
2659 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2660 
2661 	if (!dm_crypt_clients_n)
2662 		return;
2663 
2664 	pages /= dm_crypt_clients_n;
2665 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2666 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2667 	dm_crypt_pages_per_client = pages;
2668 }
2669 
2670 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2671 {
2672 	struct crypt_config *cc = pool_data;
2673 	struct page *page;
2674 
2675 	/*
2676 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2677 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2678 	 * but avoids potential spinlock contention of an exact result.
2679 	 */
2680 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2681 	    likely(gfp_mask & __GFP_NORETRY))
2682 		return NULL;
2683 
2684 	page = alloc_page(gfp_mask);
2685 	if (likely(page != NULL))
2686 		percpu_counter_add(&cc->n_allocated_pages, 1);
2687 
2688 	return page;
2689 }
2690 
2691 static void crypt_page_free(void *page, void *pool_data)
2692 {
2693 	struct crypt_config *cc = pool_data;
2694 
2695 	__free_page(page);
2696 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2697 }
2698 
2699 static void crypt_dtr(struct dm_target *ti)
2700 {
2701 	struct crypt_config *cc = ti->private;
2702 
2703 	ti->private = NULL;
2704 
2705 	if (!cc)
2706 		return;
2707 
2708 	if (cc->write_thread)
2709 		kthread_stop(cc->write_thread);
2710 
2711 	if (cc->io_queue)
2712 		destroy_workqueue(cc->io_queue);
2713 	if (cc->crypt_queue)
2714 		destroy_workqueue(cc->crypt_queue);
2715 
2716 	if (cc->workqueue_id)
2717 		ida_free(&workqueue_ida, cc->workqueue_id);
2718 
2719 	crypt_free_tfms(cc);
2720 
2721 	bioset_exit(&cc->bs);
2722 
2723 	mempool_exit(&cc->page_pool);
2724 	mempool_exit(&cc->req_pool);
2725 	mempool_exit(&cc->tag_pool);
2726 
2727 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2728 	percpu_counter_destroy(&cc->n_allocated_pages);
2729 
2730 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2731 		cc->iv_gen_ops->dtr(cc);
2732 
2733 	if (cc->dev)
2734 		dm_put_device(ti, cc->dev);
2735 
2736 	kfree_sensitive(cc->cipher_string);
2737 	kfree_sensitive(cc->key_string);
2738 	kfree_sensitive(cc->cipher_auth);
2739 	kfree_sensitive(cc->authenc_key);
2740 
2741 	mutex_destroy(&cc->bio_alloc_lock);
2742 
2743 	/* Must zero key material before freeing */
2744 	kfree_sensitive(cc);
2745 
2746 	spin_lock(&dm_crypt_clients_lock);
2747 	WARN_ON(!dm_crypt_clients_n);
2748 	dm_crypt_clients_n--;
2749 	crypt_calculate_pages_per_client();
2750 	spin_unlock(&dm_crypt_clients_lock);
2751 
2752 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2753 }
2754 
2755 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2756 {
2757 	struct crypt_config *cc = ti->private;
2758 
2759 	if (crypt_integrity_aead(cc))
2760 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2761 	else
2762 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2763 
2764 	if (cc->iv_size)
2765 		/* at least a 64 bit sector number should fit in our buffer */
2766 		cc->iv_size = max(cc->iv_size,
2767 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2768 	else if (ivmode) {
2769 		DMWARN("Selected cipher does not support IVs");
2770 		ivmode = NULL;
2771 	}
2772 
2773 	/* Choose ivmode, see comments at iv code. */
2774 	if (ivmode == NULL)
2775 		cc->iv_gen_ops = NULL;
2776 	else if (strcmp(ivmode, "plain") == 0)
2777 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2778 	else if (strcmp(ivmode, "plain64") == 0)
2779 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2780 	else if (strcmp(ivmode, "plain64be") == 0)
2781 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2782 	else if (strcmp(ivmode, "essiv") == 0)
2783 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2784 	else if (strcmp(ivmode, "benbi") == 0)
2785 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2786 	else if (strcmp(ivmode, "null") == 0)
2787 		cc->iv_gen_ops = &crypt_iv_null_ops;
2788 	else if (strcmp(ivmode, "eboiv") == 0)
2789 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2790 	else if (strcmp(ivmode, "elephant") == 0) {
2791 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2792 		cc->key_parts = 2;
2793 		cc->key_extra_size = cc->key_size / 2;
2794 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2795 			return -EINVAL;
2796 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2797 	} else if (strcmp(ivmode, "lmk") == 0) {
2798 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2799 		/*
2800 		 * Version 2 and 3 is recognised according
2801 		 * to length of provided multi-key string.
2802 		 * If present (version 3), last key is used as IV seed.
2803 		 * All keys (including IV seed) are always the same size.
2804 		 */
2805 		if (cc->key_size % cc->key_parts) {
2806 			cc->key_parts++;
2807 			cc->key_extra_size = cc->key_size / cc->key_parts;
2808 		}
2809 	} else if (strcmp(ivmode, "tcw") == 0) {
2810 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2811 		cc->key_parts += 2; /* IV + whitening */
2812 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2813 	} else if (strcmp(ivmode, "random") == 0) {
2814 		cc->iv_gen_ops = &crypt_iv_random_ops;
2815 		/* Need storage space in integrity fields. */
2816 		cc->integrity_iv_size = cc->iv_size;
2817 	} else {
2818 		ti->error = "Invalid IV mode";
2819 		return -EINVAL;
2820 	}
2821 
2822 	return 0;
2823 }
2824 
2825 /*
2826  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2827  * The HMAC is needed to calculate tag size (HMAC digest size).
2828  * This should be probably done by crypto-api calls (once available...)
2829  */
2830 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2831 {
2832 	char *start, *end, *mac_alg = NULL;
2833 	struct crypto_ahash *mac;
2834 
2835 	if (!strstarts(cipher_api, "authenc("))
2836 		return 0;
2837 
2838 	start = strchr(cipher_api, '(');
2839 	end = strchr(cipher_api, ',');
2840 	if (!start || !end || ++start > end)
2841 		return -EINVAL;
2842 
2843 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2844 	if (!mac_alg)
2845 		return -ENOMEM;
2846 
2847 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2848 	kfree(mac_alg);
2849 
2850 	if (IS_ERR(mac))
2851 		return PTR_ERR(mac);
2852 
2853 	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2854 		cc->key_mac_size = crypto_ahash_digestsize(mac);
2855 	crypto_free_ahash(mac);
2856 
2857 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2858 	if (!cc->authenc_key)
2859 		return -ENOMEM;
2860 
2861 	return 0;
2862 }
2863 
2864 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2865 				char **ivmode, char **ivopts)
2866 {
2867 	struct crypt_config *cc = ti->private;
2868 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2869 	int ret = -EINVAL;
2870 
2871 	cc->tfms_count = 1;
2872 
2873 	/*
2874 	 * New format (capi: prefix)
2875 	 * capi:cipher_api_spec-iv:ivopts
2876 	 */
2877 	tmp = &cipher_in[strlen("capi:")];
2878 
2879 	/* Separate IV options if present, it can contain another '-' in hash name */
2880 	*ivopts = strrchr(tmp, ':');
2881 	if (*ivopts) {
2882 		**ivopts = '\0';
2883 		(*ivopts)++;
2884 	}
2885 	/* Parse IV mode */
2886 	*ivmode = strrchr(tmp, '-');
2887 	if (*ivmode) {
2888 		**ivmode = '\0';
2889 		(*ivmode)++;
2890 	}
2891 	/* The rest is crypto API spec */
2892 	cipher_api = tmp;
2893 
2894 	/* Alloc AEAD, can be used only in new format. */
2895 	if (crypt_integrity_aead(cc)) {
2896 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2897 		if (ret < 0) {
2898 			ti->error = "Invalid AEAD cipher spec";
2899 			return ret;
2900 		}
2901 	}
2902 
2903 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2904 		cc->tfms_count = 64;
2905 
2906 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2907 		if (!*ivopts) {
2908 			ti->error = "Digest algorithm missing for ESSIV mode";
2909 			return -EINVAL;
2910 		}
2911 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2912 			       cipher_api, *ivopts);
2913 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2914 			ti->error = "Cannot allocate cipher string";
2915 			return -ENOMEM;
2916 		}
2917 		cipher_api = buf;
2918 	}
2919 
2920 	cc->key_parts = cc->tfms_count;
2921 
2922 	/* Allocate cipher */
2923 	ret = crypt_alloc_tfms(cc, cipher_api);
2924 	if (ret < 0) {
2925 		ti->error = "Error allocating crypto tfm";
2926 		return ret;
2927 	}
2928 
2929 	if (crypt_integrity_aead(cc))
2930 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2931 	else
2932 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2933 
2934 	return 0;
2935 }
2936 
2937 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2938 				char **ivmode, char **ivopts)
2939 {
2940 	struct crypt_config *cc = ti->private;
2941 	char *tmp, *cipher, *chainmode, *keycount;
2942 	char *cipher_api = NULL;
2943 	int ret = -EINVAL;
2944 	char dummy;
2945 
2946 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2947 		ti->error = "Bad cipher specification";
2948 		return -EINVAL;
2949 	}
2950 
2951 	/*
2952 	 * Legacy dm-crypt cipher specification
2953 	 * cipher[:keycount]-mode-iv:ivopts
2954 	 */
2955 	tmp = cipher_in;
2956 	keycount = strsep(&tmp, "-");
2957 	cipher = strsep(&keycount, ":");
2958 
2959 	if (!keycount)
2960 		cc->tfms_count = 1;
2961 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2962 		 !is_power_of_2(cc->tfms_count)) {
2963 		ti->error = "Bad cipher key count specification";
2964 		return -EINVAL;
2965 	}
2966 	cc->key_parts = cc->tfms_count;
2967 
2968 	chainmode = strsep(&tmp, "-");
2969 	*ivmode = strsep(&tmp, ":");
2970 	*ivopts = tmp;
2971 
2972 	/*
2973 	 * For compatibility with the original dm-crypt mapping format, if
2974 	 * only the cipher name is supplied, use cbc-plain.
2975 	 */
2976 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2977 		chainmode = "cbc";
2978 		*ivmode = "plain";
2979 	}
2980 
2981 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2982 		ti->error = "IV mechanism required";
2983 		return -EINVAL;
2984 	}
2985 
2986 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2987 	if (!cipher_api)
2988 		goto bad_mem;
2989 
2990 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2991 		if (!*ivopts) {
2992 			ti->error = "Digest algorithm missing for ESSIV mode";
2993 			kfree(cipher_api);
2994 			return -EINVAL;
2995 		}
2996 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2997 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2998 	} else {
2999 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3000 			       "%s(%s)", chainmode, cipher);
3001 	}
3002 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3003 		kfree(cipher_api);
3004 		goto bad_mem;
3005 	}
3006 
3007 	/* Allocate cipher */
3008 	ret = crypt_alloc_tfms(cc, cipher_api);
3009 	if (ret < 0) {
3010 		ti->error = "Error allocating crypto tfm";
3011 		kfree(cipher_api);
3012 		return ret;
3013 	}
3014 	kfree(cipher_api);
3015 
3016 	return 0;
3017 bad_mem:
3018 	ti->error = "Cannot allocate cipher strings";
3019 	return -ENOMEM;
3020 }
3021 
3022 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3023 {
3024 	struct crypt_config *cc = ti->private;
3025 	char *ivmode = NULL, *ivopts = NULL;
3026 	int ret;
3027 
3028 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3029 	if (!cc->cipher_string) {
3030 		ti->error = "Cannot allocate cipher strings";
3031 		return -ENOMEM;
3032 	}
3033 
3034 	if (strstarts(cipher_in, "capi:"))
3035 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3036 	else
3037 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3038 	if (ret)
3039 		return ret;
3040 
3041 	/* Initialize IV */
3042 	ret = crypt_ctr_ivmode(ti, ivmode);
3043 	if (ret < 0)
3044 		return ret;
3045 
3046 	/* Initialize and set key */
3047 	ret = crypt_set_key(cc, key);
3048 	if (ret < 0) {
3049 		ti->error = "Error decoding and setting key";
3050 		return ret;
3051 	}
3052 
3053 	/* Allocate IV */
3054 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3055 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3056 		if (ret < 0) {
3057 			ti->error = "Error creating IV";
3058 			return ret;
3059 		}
3060 	}
3061 
3062 	/* Initialize IV (set keys for ESSIV etc) */
3063 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3064 		ret = cc->iv_gen_ops->init(cc);
3065 		if (ret < 0) {
3066 			ti->error = "Error initialising IV";
3067 			return ret;
3068 		}
3069 	}
3070 
3071 	/* wipe the kernel key payload copy */
3072 	if (cc->key_string)
3073 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3074 
3075 	return ret;
3076 }
3077 
3078 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3079 {
3080 	struct crypt_config *cc = ti->private;
3081 	struct dm_arg_set as;
3082 	static const struct dm_arg _args[] = {
3083 		{0, 9, "Invalid number of feature args"},
3084 	};
3085 	unsigned int opt_params, val;
3086 	const char *opt_string, *sval;
3087 	char dummy;
3088 	int ret;
3089 
3090 	/* Optional parameters */
3091 	as.argc = argc;
3092 	as.argv = argv;
3093 
3094 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3095 	if (ret)
3096 		return ret;
3097 
3098 	while (opt_params--) {
3099 		opt_string = dm_shift_arg(&as);
3100 		if (!opt_string) {
3101 			ti->error = "Not enough feature arguments";
3102 			return -EINVAL;
3103 		}
3104 
3105 		if (!strcasecmp(opt_string, "allow_discards"))
3106 			ti->num_discard_bios = 1;
3107 
3108 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3109 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3110 		else if (!strcasecmp(opt_string, "high_priority"))
3111 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3112 
3113 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3114 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3115 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3116 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3117 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3118 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3119 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3120 			if (val == 0 || val > MAX_TAG_SIZE) {
3121 				ti->error = "Invalid integrity arguments";
3122 				return -EINVAL;
3123 			}
3124 			cc->used_tag_size = val;
3125 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3126 			if (!strcasecmp(sval, "aead")) {
3127 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3128 			} else if (strcasecmp(sval, "none")) {
3129 				ti->error = "Unknown integrity profile";
3130 				return -EINVAL;
3131 			}
3132 
3133 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3134 			if (!cc->cipher_auth)
3135 				return -ENOMEM;
3136 		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3137 			if (!val) {
3138 				ti->error = "Invalid integrity_key_size argument";
3139 				return -EINVAL;
3140 			}
3141 			cc->key_mac_size = val;
3142 			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3143 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3144 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3145 			    cc->sector_size > 4096 ||
3146 			    (cc->sector_size & (cc->sector_size - 1))) {
3147 				ti->error = "Invalid feature value for sector_size";
3148 				return -EINVAL;
3149 			}
3150 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3151 				ti->error = "Device size is not multiple of sector_size feature";
3152 				return -EINVAL;
3153 			}
3154 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3155 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3156 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3157 		else {
3158 			ti->error = "Invalid feature arguments";
3159 			return -EINVAL;
3160 		}
3161 	}
3162 
3163 	return 0;
3164 }
3165 
3166 #ifdef CONFIG_BLK_DEV_ZONED
3167 static int crypt_report_zones(struct dm_target *ti,
3168 		struct dm_report_zones_args *args, unsigned int nr_zones)
3169 {
3170 	struct crypt_config *cc = ti->private;
3171 
3172 	return dm_report_zones(cc->dev->bdev, cc->start,
3173 			cc->start + dm_target_offset(ti, args->next_sector),
3174 			args, nr_zones);
3175 }
3176 #else
3177 #define crypt_report_zones NULL
3178 #endif
3179 
3180 /*
3181  * Construct an encryption mapping:
3182  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3183  */
3184 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3185 {
3186 	struct crypt_config *cc;
3187 	const char *devname = dm_table_device_name(ti->table);
3188 	int key_size, wq_id;
3189 	unsigned int align_mask;
3190 	unsigned int common_wq_flags;
3191 	unsigned long long tmpll;
3192 	int ret;
3193 	size_t iv_size_padding, additional_req_size;
3194 	char dummy;
3195 
3196 	if (argc < 5) {
3197 		ti->error = "Not enough arguments";
3198 		return -EINVAL;
3199 	}
3200 
3201 	key_size = get_key_size(&argv[1]);
3202 	if (key_size < 0) {
3203 		ti->error = "Cannot parse key size";
3204 		return -EINVAL;
3205 	}
3206 
3207 	cc = kzalloc_flex(*cc, key, key_size);
3208 	if (!cc) {
3209 		ti->error = "Cannot allocate encryption context";
3210 		return -ENOMEM;
3211 	}
3212 	cc->key_size = key_size;
3213 	cc->sector_size = (1 << SECTOR_SHIFT);
3214 	cc->sector_shift = 0;
3215 
3216 	ti->private = cc;
3217 
3218 	spin_lock(&dm_crypt_clients_lock);
3219 	dm_crypt_clients_n++;
3220 	crypt_calculate_pages_per_client();
3221 	spin_unlock(&dm_crypt_clients_lock);
3222 
3223 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3224 	if (ret < 0)
3225 		goto bad;
3226 
3227 	/* Optional parameters need to be read before cipher constructor */
3228 	if (argc > 5) {
3229 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3230 		if (ret)
3231 			goto bad;
3232 	}
3233 
3234 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3235 	if (ret < 0)
3236 		goto bad;
3237 
3238 	if (crypt_integrity_aead(cc)) {
3239 		cc->dmreq_start = sizeof(struct aead_request);
3240 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3241 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3242 	} else {
3243 		cc->dmreq_start = sizeof(struct skcipher_request);
3244 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3245 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3246 	}
3247 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3248 
3249 	if (align_mask < CRYPTO_MINALIGN) {
3250 		/* Allocate the padding exactly */
3251 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3252 				& align_mask;
3253 	} else {
3254 		/*
3255 		 * If the cipher requires greater alignment than kmalloc
3256 		 * alignment, we don't know the exact position of the
3257 		 * initialization vector. We must assume worst case.
3258 		 */
3259 		iv_size_padding = align_mask;
3260 	}
3261 
3262 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3263 	additional_req_size = sizeof(struct dm_crypt_request) +
3264 		iv_size_padding + cc->iv_size +
3265 		cc->iv_size +
3266 		sizeof(uint64_t) +
3267 		sizeof(unsigned int);
3268 
3269 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3270 	if (ret) {
3271 		ti->error = "Cannot allocate crypt request mempool";
3272 		goto bad;
3273 	}
3274 
3275 	cc->per_bio_data_size = ti->per_io_data_size =
3276 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3277 		      ARCH_DMA_MINALIGN);
3278 
3279 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3280 	if (ret) {
3281 		ti->error = "Cannot allocate page mempool";
3282 		goto bad;
3283 	}
3284 
3285 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3286 	if (ret) {
3287 		ti->error = "Cannot allocate crypt bioset";
3288 		goto bad;
3289 	}
3290 
3291 	mutex_init(&cc->bio_alloc_lock);
3292 
3293 	ret = -EINVAL;
3294 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3295 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3296 		ti->error = "Invalid iv_offset sector";
3297 		goto bad;
3298 	}
3299 	cc->iv_offset = tmpll;
3300 
3301 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3302 	if (ret) {
3303 		ti->error = "Device lookup failed";
3304 		goto bad;
3305 	}
3306 
3307 	ret = -EINVAL;
3308 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3309 		ti->error = "Invalid device sector";
3310 		goto bad;
3311 	}
3312 	cc->start = tmpll;
3313 
3314 	if (bdev_is_zoned(cc->dev->bdev)) {
3315 		/*
3316 		 * For zoned block devices, we need to preserve the issuer write
3317 		 * ordering. To do so, disable write workqueues and force inline
3318 		 * encryption completion.
3319 		 */
3320 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3321 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3322 
3323 		/*
3324 		 * All zone append writes to a zone of a zoned block device will
3325 		 * have the same BIO sector, the start of the zone. When the
3326 		 * cypher IV mode uses sector values, all data targeting a
3327 		 * zone will be encrypted using the first sector numbers of the
3328 		 * zone. This will not result in write errors but will
3329 		 * cause most reads to fail as reads will use the sector values
3330 		 * for the actual data locations, resulting in IV mismatch.
3331 		 * To avoid this problem, ask DM core to emulate zone append
3332 		 * operations with regular writes.
3333 		 */
3334 		DMDEBUG("Zone append operations will be emulated");
3335 		ti->emulate_zone_append = true;
3336 	}
3337 
3338 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3339 		ret = crypt_integrity_ctr(cc, ti);
3340 		if (ret)
3341 			goto bad;
3342 
3343 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3344 		if (!cc->tag_pool_max_sectors)
3345 			cc->tag_pool_max_sectors = 1;
3346 
3347 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3348 			cc->tag_pool_max_sectors * cc->tuple_size);
3349 		if (ret) {
3350 			ti->error = "Cannot allocate integrity tags mempool";
3351 			goto bad;
3352 		}
3353 
3354 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3355 	}
3356 
3357 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3358 	if (wq_id < 0) {
3359 		ti->error = "Couldn't get workqueue id";
3360 		ret = wq_id;
3361 		goto bad;
3362 	}
3363 	cc->workqueue_id = wq_id;
3364 
3365 	ret = -ENOMEM;
3366 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3367 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3368 		common_wq_flags |= WQ_HIGHPRI;
3369 
3370 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d",
3371 				       common_wq_flags | WQ_PERCPU, 1,
3372 				       devname, wq_id);
3373 	if (!cc->io_queue) {
3374 		ti->error = "Couldn't create kcryptd io queue";
3375 		goto bad;
3376 	}
3377 
3378 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3379 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3380 						  common_wq_flags | WQ_CPU_INTENSIVE | WQ_PERCPU,
3381 						  1, devname, wq_id);
3382 	} else {
3383 		/*
3384 		 * While crypt_queue is certainly CPU intensive, the use of
3385 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3386 		 */
3387 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3388 						  common_wq_flags | WQ_UNBOUND,
3389 						  num_online_cpus(), devname, wq_id);
3390 	}
3391 	if (!cc->crypt_queue) {
3392 		ti->error = "Couldn't create kcryptd queue";
3393 		goto bad;
3394 	}
3395 
3396 	spin_lock_init(&cc->write_thread_lock);
3397 	cc->write_tree = RB_ROOT;
3398 
3399 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3400 	if (IS_ERR(cc->write_thread)) {
3401 		ret = PTR_ERR(cc->write_thread);
3402 		cc->write_thread = NULL;
3403 		ti->error = "Couldn't spawn write thread";
3404 		goto bad;
3405 	}
3406 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3407 		set_user_nice(cc->write_thread, MIN_NICE);
3408 
3409 	ti->num_flush_bios = 1;
3410 	ti->limit_swap_bios = true;
3411 	ti->accounts_remapped_io = true;
3412 
3413 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3414 	return 0;
3415 
3416 bad:
3417 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3418 	crypt_dtr(ti);
3419 	return ret;
3420 }
3421 
3422 static int crypt_map(struct dm_target *ti, struct bio *bio)
3423 {
3424 	struct dm_crypt_io *io;
3425 	struct crypt_config *cc = ti->private;
3426 	unsigned max_sectors;
3427 	bool no_split;
3428 
3429 	/*
3430 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3431 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3432 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3433 	 */
3434 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3435 	    bio_op(bio) == REQ_OP_DISCARD)) {
3436 		bio_set_dev(bio, cc->dev->bdev);
3437 		if (bio_sectors(bio))
3438 			bio->bi_iter.bi_sector = cc->start +
3439 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3440 		return DM_MAPIO_REMAPPED;
3441 	}
3442 
3443 	/*
3444 	 * Check if bio is too large, split as needed.
3445 	 *
3446 	 * For zoned devices, splitting write operations creates the
3447 	 * risk of deadlocking queue freeze operations with zone write
3448 	 * plugging BIO work when the reminder of a split BIO is
3449 	 * issued. So always allow the entire BIO to proceed.
3450 	 */
3451 	no_split = (ti->emulate_zone_append && op_is_write(bio_op(bio))) ||
3452 		   (bio->bi_opf & REQ_ATOMIC);
3453 	max_sectors = get_max_request_sectors(ti, bio, no_split);
3454 	if (unlikely(bio_sectors(bio) > max_sectors)) {
3455 		if (unlikely(no_split))
3456 			return DM_MAPIO_KILL;
3457 		dm_accept_partial_bio(bio, max_sectors);
3458 	}
3459 
3460 	/*
3461 	 * Ensure that bio is a multiple of internal sector encryption size
3462 	 * and is aligned to this size as defined in IO hints.
3463 	 */
3464 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3465 		return DM_MAPIO_KILL;
3466 
3467 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3468 		return DM_MAPIO_KILL;
3469 
3470 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3471 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3472 
3473 	if (cc->tuple_size) {
3474 		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3475 
3476 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3477 			io->integrity_metadata = NULL;
3478 		else
3479 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3480 
3481 		if (unlikely(!io->integrity_metadata)) {
3482 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3483 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3484 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3485 			io->integrity_metadata_from_pool = true;
3486 		}
3487 	}
3488 
3489 	if (crypt_integrity_aead(cc))
3490 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3491 	else
3492 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3493 
3494 	if (bio_data_dir(io->base_bio) == READ) {
3495 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3496 			kcryptd_queue_read(io);
3497 	} else
3498 		kcryptd_queue_crypt(io);
3499 
3500 	return DM_MAPIO_SUBMITTED;
3501 }
3502 
3503 static char hex2asc(unsigned char c)
3504 {
3505 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3506 }
3507 
3508 static void crypt_status(struct dm_target *ti, status_type_t type,
3509 			 unsigned int status_flags, char *result, unsigned int maxlen)
3510 {
3511 	struct crypt_config *cc = ti->private;
3512 	unsigned int i, sz = 0;
3513 	int num_feature_args = 0;
3514 
3515 	switch (type) {
3516 	case STATUSTYPE_INFO:
3517 		result[0] = '\0';
3518 		break;
3519 
3520 	case STATUSTYPE_TABLE:
3521 		DMEMIT("%s ", cc->cipher_string);
3522 
3523 		if (cc->key_size > 0) {
3524 			if (cc->key_string)
3525 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3526 			else {
3527 				for (i = 0; i < cc->key_size; i++) {
3528 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3529 					       hex2asc(cc->key[i] & 0xf));
3530 				}
3531 			}
3532 		} else
3533 			DMEMIT("-");
3534 
3535 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3536 				cc->dev->name, (unsigned long long)cc->start);
3537 
3538 		num_feature_args += !!ti->num_discard_bios;
3539 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3540 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3541 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3542 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3543 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3544 		num_feature_args += !!cc->used_tag_size;
3545 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3546 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3547 		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3548 		if (num_feature_args) {
3549 			DMEMIT(" %d", num_feature_args);
3550 			if (ti->num_discard_bios)
3551 				DMEMIT(" allow_discards");
3552 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3553 				DMEMIT(" same_cpu_crypt");
3554 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3555 				DMEMIT(" high_priority");
3556 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3557 				DMEMIT(" submit_from_crypt_cpus");
3558 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3559 				DMEMIT(" no_read_workqueue");
3560 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3561 				DMEMIT(" no_write_workqueue");
3562 			if (cc->used_tag_size)
3563 				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3564 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3565 				DMEMIT(" sector_size:%d", cc->sector_size);
3566 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3567 				DMEMIT(" iv_large_sectors");
3568 			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3569 				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3570 		}
3571 		break;
3572 
3573 	case STATUSTYPE_IMA:
3574 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3575 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3576 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3577 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3578 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3579 		       'y' : 'n');
3580 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3581 		       'y' : 'n');
3582 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3583 		       'y' : 'n');
3584 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3585 		       'y' : 'n');
3586 
3587 		if (cc->used_tag_size)
3588 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3589 			       cc->used_tag_size, cc->cipher_auth);
3590 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3591 			DMEMIT(",sector_size=%d", cc->sector_size);
3592 		if (cc->cipher_string)
3593 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3594 
3595 		DMEMIT(",key_size=%u", cc->key_size);
3596 		DMEMIT(",key_parts=%u", cc->key_parts);
3597 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3598 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3599 		DMEMIT(";");
3600 		break;
3601 	}
3602 }
3603 
3604 static void crypt_postsuspend(struct dm_target *ti)
3605 {
3606 	struct crypt_config *cc = ti->private;
3607 
3608 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3609 }
3610 
3611 static int crypt_preresume(struct dm_target *ti)
3612 {
3613 	struct crypt_config *cc = ti->private;
3614 
3615 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3616 		DMERR("aborting resume - crypt key is not set.");
3617 		return -EAGAIN;
3618 	}
3619 
3620 	return 0;
3621 }
3622 
3623 static void crypt_resume(struct dm_target *ti)
3624 {
3625 	struct crypt_config *cc = ti->private;
3626 
3627 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3628 }
3629 
3630 /* Message interface
3631  *	key set <key>
3632  *	key wipe
3633  */
3634 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3635 			 char *result, unsigned int maxlen)
3636 {
3637 	struct crypt_config *cc = ti->private;
3638 	int key_size, ret = -EINVAL;
3639 
3640 	if (argc < 2)
3641 		goto error;
3642 
3643 	if (!strcasecmp(argv[0], "key")) {
3644 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3645 			DMWARN("not suspended during key manipulation.");
3646 			return -EINVAL;
3647 		}
3648 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3649 			/* The key size may not be changed. */
3650 			key_size = get_key_size(&argv[2]);
3651 			if (key_size < 0 || cc->key_size != key_size) {
3652 				memset(argv[2], '0', strlen(argv[2]));
3653 				return -EINVAL;
3654 			}
3655 
3656 			ret = crypt_set_key(cc, argv[2]);
3657 			if (ret)
3658 				return ret;
3659 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3660 				ret = cc->iv_gen_ops->init(cc);
3661 			/* wipe the kernel key payload copy */
3662 			if (cc->key_string)
3663 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3664 			return ret;
3665 		}
3666 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3667 			return crypt_wipe_key(cc);
3668 	}
3669 
3670 error:
3671 	DMWARN("unrecognised message received.");
3672 	return -EINVAL;
3673 }
3674 
3675 static int crypt_iterate_devices(struct dm_target *ti,
3676 				 iterate_devices_callout_fn fn, void *data)
3677 {
3678 	struct crypt_config *cc = ti->private;
3679 
3680 	return fn(ti, cc->dev, cc->start, ti->len, data);
3681 }
3682 
3683 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3684 {
3685 	struct crypt_config *cc = ti->private;
3686 
3687 	limits->logical_block_size =
3688 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3689 	limits->physical_block_size =
3690 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3691 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3692 	limits->dma_alignment = limits->logical_block_size - 1;
3693 
3694 	/*
3695 	 * For zoned dm-crypt targets, there will be no internal splitting of
3696 	 * write BIOs to avoid exceeding BIO_MAX_VECS vectors per BIO. But
3697 	 * without respecting this limit, crypt_alloc_buffer() will trigger a
3698 	 * BUG(). Avoid this by forcing DM core to split write BIOs to this
3699 	 * limit.
3700 	 */
3701 	if (ti->emulate_zone_append)
3702 		limits->max_hw_sectors = min(limits->max_hw_sectors,
3703 					     BIO_MAX_VECS << PAGE_SECTORS_SHIFT);
3704 
3705 	limits->atomic_write_hw_unit_max = min(limits->atomic_write_hw_unit_max,
3706 					       BIO_MAX_VECS << PAGE_SHIFT);
3707 	limits->atomic_write_hw_max = min(limits->atomic_write_hw_max,
3708 					  BIO_MAX_VECS << PAGE_SHIFT);
3709 }
3710 
3711 static struct target_type crypt_target = {
3712 	.name   = "crypt",
3713 	.version = {1, 29, 0},
3714 	.module = THIS_MODULE,
3715 	.ctr    = crypt_ctr,
3716 	.dtr    = crypt_dtr,
3717 	.features = DM_TARGET_ZONED_HM | DM_TARGET_ATOMIC_WRITES,
3718 	.report_zones = crypt_report_zones,
3719 	.map    = crypt_map,
3720 	.status = crypt_status,
3721 	.postsuspend = crypt_postsuspend,
3722 	.preresume = crypt_preresume,
3723 	.resume = crypt_resume,
3724 	.message = crypt_message,
3725 	.iterate_devices = crypt_iterate_devices,
3726 	.io_hints = crypt_io_hints,
3727 };
3728 module_dm(crypt);
3729 
3730 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3731 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3732 MODULE_LICENSE("GPL");
3733