2 * Copyright (C) 2003 Jana Saout <jana@saout.de>
3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
5 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
7 * This file is released under the GPL.
10 #include <linux/completion.h>
11 #include <linux/err.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/key.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/crypto.h>
21 #include <linux/workqueue.h>
22 #include <linux/kthread.h>
23 #include <linux/backing-dev.h>
24 #include <linux/atomic.h>
25 #include <linux/scatterlist.h>
26 #include <linux/rbtree.h>
27 #include <linux/ctype.h>
29 #include <asm/unaligned.h>
30 #include <crypto/hash.h>
31 #include <crypto/md5.h>
32 #include <crypto/algapi.h>
33 #include <crypto/skcipher.h>
34 #include <crypto/aead.h>
35 #include <crypto/authenc.h>
36 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
37 #include <linux/key-type.h>
38 #include <keys/user-type.h>
39 #include <keys/encrypted-type.h>
41 #include <linux/device-mapper.h>
43 #define DM_MSG_PREFIX "crypt"
46 * context holding the current state of a multi-part conversion
48 struct convert_context {
49 struct completion restart;
52 struct bvec_iter iter_in;
53 struct bvec_iter iter_out;
57 struct skcipher_request *req;
58 struct aead_request *req_aead;
64 * per bio private data
67 struct crypt_config *cc;
69 u8 *integrity_metadata;
70 bool integrity_metadata_from_pool;
71 struct work_struct work;
72 struct tasklet_struct tasklet;
74 struct convert_context ctx;
80 struct rb_node rb_node;
81 } CRYPTO_MINALIGN_ATTR;
83 struct dm_crypt_request {
84 struct convert_context *ctx;
85 struct scatterlist sg_in[4];
86 struct scatterlist sg_out[4];
92 struct crypt_iv_operations {
93 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
95 void (*dtr)(struct crypt_config *cc);
96 int (*init)(struct crypt_config *cc);
97 int (*wipe)(struct crypt_config *cc);
98 int (*generator)(struct crypt_config *cc, u8 *iv,
99 struct dm_crypt_request *dmreq);
100 int (*post)(struct crypt_config *cc, u8 *iv,
101 struct dm_crypt_request *dmreq);
104 struct iv_benbi_private {
108 #define LMK_SEED_SIZE 64 /* hash + 0 */
109 struct iv_lmk_private {
110 struct crypto_shash *hash_tfm;
114 #define TCW_WHITENING_SIZE 16
115 struct iv_tcw_private {
116 struct crypto_shash *crc32_tfm;
121 #define ELEPHANT_MAX_KEY_SIZE 32
122 struct iv_elephant_private {
123 struct crypto_skcipher *tfm;
127 * Crypt: maps a linear range of a block device
128 * and encrypts / decrypts at the same time.
130 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
131 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
132 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
133 DM_CRYPT_WRITE_INLINE };
136 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cihper */
137 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
138 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
142 * The fields in here must be read only after initialization.
144 struct crypt_config {
148 struct percpu_counter n_allocated_pages;
150 struct workqueue_struct *io_queue;
151 struct workqueue_struct *crypt_queue;
153 spinlock_t write_thread_lock;
154 struct task_struct *write_thread;
155 struct rb_root write_tree;
161 const struct crypt_iv_operations *iv_gen_ops;
163 struct iv_benbi_private benbi;
164 struct iv_lmk_private lmk;
165 struct iv_tcw_private tcw;
166 struct iv_elephant_private elephant;
169 unsigned int iv_size;
170 unsigned short int sector_size;
171 unsigned char sector_shift;
174 struct crypto_skcipher **tfms;
175 struct crypto_aead **tfms_aead;
178 unsigned long cipher_flags;
181 * Layout of each crypto request:
183 * struct skcipher_request
186 * struct dm_crypt_request
190 * The padding is added so that dm_crypt_request and the IV are
193 unsigned int dmreq_start;
195 unsigned int per_bio_data_size;
198 unsigned int key_size;
199 unsigned int key_parts; /* independent parts in key buffer */
200 unsigned int key_extra_size; /* additional keys length */
201 unsigned int key_mac_size; /* MAC key size for authenc(...) */
203 unsigned int integrity_tag_size;
204 unsigned int integrity_iv_size;
205 unsigned int on_disk_tag_size;
208 * pool for per bio private data, crypto requests,
209 * encryption requeusts/buffer pages and integrity tags
211 unsigned tag_pool_max_sectors;
217 struct mutex bio_alloc_lock;
219 u8 *authenc_key; /* space for keys in authenc() format (if used) */
224 #define MAX_TAG_SIZE 480
225 #define POOL_ENTRY_SIZE 512
227 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
228 static unsigned dm_crypt_clients_n = 0;
229 static volatile unsigned long dm_crypt_pages_per_client;
230 #define DM_CRYPT_MEMORY_PERCENT 2
231 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_PAGES * 16)
233 static void clone_init(struct dm_crypt_io *, struct bio *);
234 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
235 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
236 struct scatterlist *sg);
238 static bool crypt_integrity_aead(struct crypt_config *cc);
241 * Use this to access cipher attributes that are independent of the key.
243 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
245 return cc->cipher_tfm.tfms[0];
248 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
250 return cc->cipher_tfm.tfms_aead[0];
254 * Different IV generation algorithms:
256 * plain: the initial vector is the 32-bit little-endian version of the sector
257 * number, padded with zeros if necessary.
259 * plain64: the initial vector is the 64-bit little-endian version of the sector
260 * number, padded with zeros if necessary.
262 * plain64be: the initial vector is the 64-bit big-endian version of the sector
263 * number, padded with zeros if necessary.
265 * essiv: "encrypted sector|salt initial vector", the sector number is
266 * encrypted with the bulk cipher using a salt as key. The salt
267 * should be derived from the bulk cipher's key via hashing.
269 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
270 * (needed for LRW-32-AES and possible other narrow block modes)
272 * null: the initial vector is always zero. Provides compatibility with
273 * obsolete loop_fish2 devices. Do not use for new devices.
275 * lmk: Compatible implementation of the block chaining mode used
276 * by the Loop-AES block device encryption system
277 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
278 * It operates on full 512 byte sectors and uses CBC
279 * with an IV derived from the sector number, the data and
280 * optionally extra IV seed.
281 * This means that after decryption the first block
282 * of sector must be tweaked according to decrypted data.
283 * Loop-AES can use three encryption schemes:
284 * version 1: is plain aes-cbc mode
285 * version 2: uses 64 multikey scheme with lmk IV generator
286 * version 3: the same as version 2 with additional IV seed
287 * (it uses 65 keys, last key is used as IV seed)
289 * tcw: Compatible implementation of the block chaining mode used
290 * by the TrueCrypt device encryption system (prior to version 4.1).
291 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
292 * It operates on full 512 byte sectors and uses CBC
293 * with an IV derived from initial key and the sector number.
294 * In addition, whitening value is applied on every sector, whitening
295 * is calculated from initial key, sector number and mixed using CRC32.
296 * Note that this encryption scheme is vulnerable to watermarking attacks
297 * and should be used for old compatible containers access only.
299 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
300 * The IV is encrypted little-endian byte-offset (with the same key
301 * and cipher as the volume).
303 * elephant: The extended version of eboiv with additional Elephant diffuser
304 * used with Bitlocker CBC mode.
305 * This mode was used in older Windows systems
306 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
309 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
310 struct dm_crypt_request *dmreq)
312 memset(iv, 0, cc->iv_size);
313 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
318 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
319 struct dm_crypt_request *dmreq)
321 memset(iv, 0, cc->iv_size);
322 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
327 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
328 struct dm_crypt_request *dmreq)
330 memset(iv, 0, cc->iv_size);
331 /* iv_size is at least of size u64; usually it is 16 bytes */
332 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
337 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
338 struct dm_crypt_request *dmreq)
341 * ESSIV encryption of the IV is now handled by the crypto API,
342 * so just pass the plain sector number here.
344 memset(iv, 0, cc->iv_size);
345 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
350 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
356 if (crypt_integrity_aead(cc))
357 bs = crypto_aead_blocksize(any_tfm_aead(cc));
359 bs = crypto_skcipher_blocksize(any_tfm(cc));
362 /* we need to calculate how far we must shift the sector count
363 * to get the cipher block count, we use this shift in _gen */
365 if (1 << log != bs) {
366 ti->error = "cypher blocksize is not a power of 2";
371 ti->error = "cypher blocksize is > 512";
375 cc->iv_gen_private.benbi.shift = 9 - log;
380 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
384 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
385 struct dm_crypt_request *dmreq)
389 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
391 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
392 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
397 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
398 struct dm_crypt_request *dmreq)
400 memset(iv, 0, cc->iv_size);
405 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
407 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
409 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
410 crypto_free_shash(lmk->hash_tfm);
411 lmk->hash_tfm = NULL;
413 kfree_sensitive(lmk->seed);
417 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
420 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
422 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
423 ti->error = "Unsupported sector size for LMK";
427 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
428 CRYPTO_ALG_ALLOCATES_MEMORY);
429 if (IS_ERR(lmk->hash_tfm)) {
430 ti->error = "Error initializing LMK hash";
431 return PTR_ERR(lmk->hash_tfm);
434 /* No seed in LMK version 2 */
435 if (cc->key_parts == cc->tfms_count) {
440 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
442 crypt_iv_lmk_dtr(cc);
443 ti->error = "Error kmallocing seed storage in LMK";
450 static int crypt_iv_lmk_init(struct crypt_config *cc)
452 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
453 int subkey_size = cc->key_size / cc->key_parts;
455 /* LMK seed is on the position of LMK_KEYS + 1 key */
457 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
458 crypto_shash_digestsize(lmk->hash_tfm));
463 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
465 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
468 memset(lmk->seed, 0, LMK_SEED_SIZE);
473 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
474 struct dm_crypt_request *dmreq,
477 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
478 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
479 struct md5_state md5state;
483 desc->tfm = lmk->hash_tfm;
485 r = crypto_shash_init(desc);
490 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
495 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
496 r = crypto_shash_update(desc, data + 16, 16 * 31);
500 /* Sector is cropped to 56 bits here */
501 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
502 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
503 buf[2] = cpu_to_le32(4024);
505 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
509 /* No MD5 padding here */
510 r = crypto_shash_export(desc, &md5state);
514 for (i = 0; i < MD5_HASH_WORDS; i++)
515 __cpu_to_le32s(&md5state.hash[i]);
516 memcpy(iv, &md5state.hash, cc->iv_size);
521 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
522 struct dm_crypt_request *dmreq)
524 struct scatterlist *sg;
528 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
529 sg = crypt_get_sg_data(cc, dmreq->sg_in);
530 src = kmap_atomic(sg_page(sg));
531 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
534 memset(iv, 0, cc->iv_size);
539 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
540 struct dm_crypt_request *dmreq)
542 struct scatterlist *sg;
546 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
549 sg = crypt_get_sg_data(cc, dmreq->sg_out);
550 dst = kmap_atomic(sg_page(sg));
551 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
553 /* Tweak the first block of plaintext sector */
555 crypto_xor(dst + sg->offset, iv, cc->iv_size);
561 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
563 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
565 kfree_sensitive(tcw->iv_seed);
567 kfree_sensitive(tcw->whitening);
568 tcw->whitening = NULL;
570 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
571 crypto_free_shash(tcw->crc32_tfm);
572 tcw->crc32_tfm = NULL;
575 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
578 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
580 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
581 ti->error = "Unsupported sector size for TCW";
585 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
586 ti->error = "Wrong key size for TCW";
590 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
591 CRYPTO_ALG_ALLOCATES_MEMORY);
592 if (IS_ERR(tcw->crc32_tfm)) {
593 ti->error = "Error initializing CRC32 in TCW";
594 return PTR_ERR(tcw->crc32_tfm);
597 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
598 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
599 if (!tcw->iv_seed || !tcw->whitening) {
600 crypt_iv_tcw_dtr(cc);
601 ti->error = "Error allocating seed storage in TCW";
608 static int crypt_iv_tcw_init(struct crypt_config *cc)
610 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
611 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
613 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
614 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
620 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
622 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
624 memset(tcw->iv_seed, 0, cc->iv_size);
625 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
630 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
631 struct dm_crypt_request *dmreq,
634 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
635 __le64 sector = cpu_to_le64(dmreq->iv_sector);
636 u8 buf[TCW_WHITENING_SIZE];
637 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
640 /* xor whitening with sector number */
641 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
642 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
644 /* calculate crc32 for every 32bit part and xor it */
645 desc->tfm = tcw->crc32_tfm;
646 for (i = 0; i < 4; i++) {
647 r = crypto_shash_init(desc);
650 r = crypto_shash_update(desc, &buf[i * 4], 4);
653 r = crypto_shash_final(desc, &buf[i * 4]);
657 crypto_xor(&buf[0], &buf[12], 4);
658 crypto_xor(&buf[4], &buf[8], 4);
660 /* apply whitening (8 bytes) to whole sector */
661 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
662 crypto_xor(data + i * 8, buf, 8);
664 memzero_explicit(buf, sizeof(buf));
668 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
669 struct dm_crypt_request *dmreq)
671 struct scatterlist *sg;
672 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
673 __le64 sector = cpu_to_le64(dmreq->iv_sector);
677 /* Remove whitening from ciphertext */
678 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
679 sg = crypt_get_sg_data(cc, dmreq->sg_in);
680 src = kmap_atomic(sg_page(sg));
681 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
686 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
688 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
694 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
695 struct dm_crypt_request *dmreq)
697 struct scatterlist *sg;
701 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
704 /* Apply whitening on ciphertext */
705 sg = crypt_get_sg_data(cc, dmreq->sg_out);
706 dst = kmap_atomic(sg_page(sg));
707 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
713 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
714 struct dm_crypt_request *dmreq)
716 /* Used only for writes, there must be an additional space to store IV */
717 get_random_bytes(iv, cc->iv_size);
721 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
724 if (crypt_integrity_aead(cc)) {
725 ti->error = "AEAD transforms not supported for EBOIV";
729 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
730 ti->error = "Block size of EBOIV cipher does "
731 "not match IV size of block cipher";
738 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
739 struct dm_crypt_request *dmreq)
741 u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
742 struct skcipher_request *req;
743 struct scatterlist src, dst;
744 DECLARE_CRYPTO_WAIT(wait);
747 req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
751 memset(buf, 0, cc->iv_size);
752 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
754 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
755 sg_init_one(&dst, iv, cc->iv_size);
756 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
757 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
758 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
759 skcipher_request_free(req);
764 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
766 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
768 crypto_free_skcipher(elephant->tfm);
769 elephant->tfm = NULL;
772 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
775 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
778 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
779 CRYPTO_ALG_ALLOCATES_MEMORY);
780 if (IS_ERR(elephant->tfm)) {
781 r = PTR_ERR(elephant->tfm);
782 elephant->tfm = NULL;
786 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
788 crypt_iv_elephant_dtr(cc);
792 static void diffuser_disk_to_cpu(u32 *d, size_t n)
794 #ifndef __LITTLE_ENDIAN
797 for (i = 0; i < n; i++)
798 d[i] = le32_to_cpu((__le32)d[i]);
802 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
804 #ifndef __LITTLE_ENDIAN
807 for (i = 0; i < n; i++)
808 d[i] = cpu_to_le32((u32)d[i]);
812 static void diffuser_a_decrypt(u32 *d, size_t n)
816 for (i = 0; i < 5; i++) {
821 while (i1 < (n - 1)) {
822 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
828 d[i1] += d[i2] ^ d[i3];
834 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
837 d[i1] += d[i2] ^ d[i3];
843 static void diffuser_a_encrypt(u32 *d, size_t n)
847 for (i = 0; i < 5; i++) {
853 d[i1] -= d[i2] ^ d[i3];
856 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
862 d[i1] -= d[i2] ^ d[i3];
868 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
874 static void diffuser_b_decrypt(u32 *d, size_t n)
878 for (i = 0; i < 3; i++) {
883 while (i1 < (n - 1)) {
884 d[i1] += d[i2] ^ d[i3];
887 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
893 d[i1] += d[i2] ^ d[i3];
899 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
905 static void diffuser_b_encrypt(u32 *d, size_t n)
909 for (i = 0; i < 3; i++) {
915 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
921 d[i1] -= d[i2] ^ d[i3];
927 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
930 d[i1] -= d[i2] ^ d[i3];
936 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
938 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
939 u8 *es, *ks, *data, *data2, *data_offset;
940 struct skcipher_request *req;
941 struct scatterlist *sg, *sg2, src, dst;
942 DECLARE_CRYPTO_WAIT(wait);
945 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
946 es = kzalloc(16, GFP_NOIO); /* Key for AES */
947 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
949 if (!req || !es || !ks) {
954 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
957 sg_init_one(&src, es, 16);
958 sg_init_one(&dst, ks, 16);
959 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
960 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
961 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
967 sg_init_one(&dst, &ks[16], 16);
968 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
972 sg = crypt_get_sg_data(cc, dmreq->sg_out);
973 data = kmap_atomic(sg_page(sg));
974 data_offset = data + sg->offset;
976 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
977 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
978 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
979 data2 = kmap_atomic(sg_page(sg2));
980 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
981 kunmap_atomic(data2);
984 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
985 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
986 diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
987 diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
988 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
991 for (i = 0; i < (cc->sector_size / 32); i++)
992 crypto_xor(data_offset + i * 32, ks, 32);
994 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
995 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
996 diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
997 diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
998 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1001 kunmap_atomic(data);
1003 kfree_sensitive(ks);
1004 kfree_sensitive(es);
1005 skcipher_request_free(req);
1009 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1010 struct dm_crypt_request *dmreq)
1014 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1015 r = crypt_iv_elephant(cc, dmreq);
1020 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1023 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1024 struct dm_crypt_request *dmreq)
1026 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1027 return crypt_iv_elephant(cc, dmreq);
1032 static int crypt_iv_elephant_init(struct crypt_config *cc)
1034 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1035 int key_offset = cc->key_size - cc->key_extra_size;
1037 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1040 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1042 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1043 u8 key[ELEPHANT_MAX_KEY_SIZE];
1045 memset(key, 0, cc->key_extra_size);
1046 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1049 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1050 .generator = crypt_iv_plain_gen
1053 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1054 .generator = crypt_iv_plain64_gen
1057 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1058 .generator = crypt_iv_plain64be_gen
1061 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1062 .generator = crypt_iv_essiv_gen
1065 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1066 .ctr = crypt_iv_benbi_ctr,
1067 .dtr = crypt_iv_benbi_dtr,
1068 .generator = crypt_iv_benbi_gen
1071 static const struct crypt_iv_operations crypt_iv_null_ops = {
1072 .generator = crypt_iv_null_gen
1075 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1076 .ctr = crypt_iv_lmk_ctr,
1077 .dtr = crypt_iv_lmk_dtr,
1078 .init = crypt_iv_lmk_init,
1079 .wipe = crypt_iv_lmk_wipe,
1080 .generator = crypt_iv_lmk_gen,
1081 .post = crypt_iv_lmk_post
1084 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1085 .ctr = crypt_iv_tcw_ctr,
1086 .dtr = crypt_iv_tcw_dtr,
1087 .init = crypt_iv_tcw_init,
1088 .wipe = crypt_iv_tcw_wipe,
1089 .generator = crypt_iv_tcw_gen,
1090 .post = crypt_iv_tcw_post
1093 static struct crypt_iv_operations crypt_iv_random_ops = {
1094 .generator = crypt_iv_random_gen
1097 static struct crypt_iv_operations crypt_iv_eboiv_ops = {
1098 .ctr = crypt_iv_eboiv_ctr,
1099 .generator = crypt_iv_eboiv_gen
1102 static struct crypt_iv_operations crypt_iv_elephant_ops = {
1103 .ctr = crypt_iv_elephant_ctr,
1104 .dtr = crypt_iv_elephant_dtr,
1105 .init = crypt_iv_elephant_init,
1106 .wipe = crypt_iv_elephant_wipe,
1107 .generator = crypt_iv_elephant_gen,
1108 .post = crypt_iv_elephant_post
1112 * Integrity extensions
1114 static bool crypt_integrity_aead(struct crypt_config *cc)
1116 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1119 static bool crypt_integrity_hmac(struct crypt_config *cc)
1121 return crypt_integrity_aead(cc) && cc->key_mac_size;
1124 /* Get sg containing data */
1125 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1126 struct scatterlist *sg)
1128 if (unlikely(crypt_integrity_aead(cc)))
1134 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1136 struct bio_integrity_payload *bip;
1137 unsigned int tag_len;
1140 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1143 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1145 return PTR_ERR(bip);
1147 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1149 bip->bip_iter.bi_size = tag_len;
1150 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1152 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1153 tag_len, offset_in_page(io->integrity_metadata));
1154 if (unlikely(ret != tag_len))
1160 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1162 #ifdef CONFIG_BLK_DEV_INTEGRITY
1163 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1164 struct mapped_device *md = dm_table_get_md(ti->table);
1166 /* From now we require underlying device with our integrity profile */
1167 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1168 ti->error = "Integrity profile not supported.";
1172 if (bi->tag_size != cc->on_disk_tag_size ||
1173 bi->tuple_size != cc->on_disk_tag_size) {
1174 ti->error = "Integrity profile tag size mismatch.";
1177 if (1 << bi->interval_exp != cc->sector_size) {
1178 ti->error = "Integrity profile sector size mismatch.";
1182 if (crypt_integrity_aead(cc)) {
1183 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1184 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1185 cc->integrity_tag_size, cc->integrity_iv_size);
1187 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1188 ti->error = "Integrity AEAD auth tag size is not supported.";
1191 } else if (cc->integrity_iv_size)
1192 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1193 cc->integrity_iv_size);
1195 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1196 ti->error = "Not enough space for integrity tag in the profile.";
1202 ti->error = "Integrity profile not supported.";
1207 static void crypt_convert_init(struct crypt_config *cc,
1208 struct convert_context *ctx,
1209 struct bio *bio_out, struct bio *bio_in,
1212 ctx->bio_in = bio_in;
1213 ctx->bio_out = bio_out;
1215 ctx->iter_in = bio_in->bi_iter;
1217 ctx->iter_out = bio_out->bi_iter;
1218 ctx->cc_sector = sector + cc->iv_offset;
1219 init_completion(&ctx->restart);
1222 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1225 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1228 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1230 return (void *)((char *)dmreq - cc->dmreq_start);
1233 static u8 *iv_of_dmreq(struct crypt_config *cc,
1234 struct dm_crypt_request *dmreq)
1236 if (crypt_integrity_aead(cc))
1237 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1238 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1240 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1241 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1244 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1245 struct dm_crypt_request *dmreq)
1247 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1250 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1251 struct dm_crypt_request *dmreq)
1253 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1254 return (__le64 *) ptr;
1257 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1258 struct dm_crypt_request *dmreq)
1260 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1261 cc->iv_size + sizeof(uint64_t);
1262 return (unsigned int*)ptr;
1265 static void *tag_from_dmreq(struct crypt_config *cc,
1266 struct dm_crypt_request *dmreq)
1268 struct convert_context *ctx = dmreq->ctx;
1269 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1271 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1272 cc->on_disk_tag_size];
1275 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1276 struct dm_crypt_request *dmreq)
1278 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1281 static int crypt_convert_block_aead(struct crypt_config *cc,
1282 struct convert_context *ctx,
1283 struct aead_request *req,
1284 unsigned int tag_offset)
1286 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1287 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1288 struct dm_crypt_request *dmreq;
1289 u8 *iv, *org_iv, *tag_iv, *tag;
1293 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1295 /* Reject unexpected unaligned bio. */
1296 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1299 dmreq = dmreq_of_req(cc, req);
1300 dmreq->iv_sector = ctx->cc_sector;
1301 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1302 dmreq->iv_sector >>= cc->sector_shift;
1305 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1307 sector = org_sector_of_dmreq(cc, dmreq);
1308 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1310 iv = iv_of_dmreq(cc, dmreq);
1311 org_iv = org_iv_of_dmreq(cc, dmreq);
1312 tag = tag_from_dmreq(cc, dmreq);
1313 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1316 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1317 * | (authenticated) | (auth+encryption) | |
1318 * | sector_LE | IV | sector in/out | tag in/out |
1320 sg_init_table(dmreq->sg_in, 4);
1321 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1322 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1323 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1324 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1326 sg_init_table(dmreq->sg_out, 4);
1327 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1328 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1329 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1330 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1332 if (cc->iv_gen_ops) {
1333 /* For READs use IV stored in integrity metadata */
1334 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1335 memcpy(org_iv, tag_iv, cc->iv_size);
1337 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1340 /* Store generated IV in integrity metadata */
1341 if (cc->integrity_iv_size)
1342 memcpy(tag_iv, org_iv, cc->iv_size);
1344 /* Working copy of IV, to be modified in crypto API */
1345 memcpy(iv, org_iv, cc->iv_size);
1348 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1349 if (bio_data_dir(ctx->bio_in) == WRITE) {
1350 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1351 cc->sector_size, iv);
1352 r = crypto_aead_encrypt(req);
1353 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1354 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1355 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1357 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1358 cc->sector_size + cc->integrity_tag_size, iv);
1359 r = crypto_aead_decrypt(req);
1362 if (r == -EBADMSG) {
1363 char b[BDEVNAME_SIZE];
1364 DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1365 (unsigned long long)le64_to_cpu(*sector));
1368 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1369 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1371 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1372 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1377 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1378 struct convert_context *ctx,
1379 struct skcipher_request *req,
1380 unsigned int tag_offset)
1382 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1383 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1384 struct scatterlist *sg_in, *sg_out;
1385 struct dm_crypt_request *dmreq;
1386 u8 *iv, *org_iv, *tag_iv;
1390 /* Reject unexpected unaligned bio. */
1391 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1394 dmreq = dmreq_of_req(cc, req);
1395 dmreq->iv_sector = ctx->cc_sector;
1396 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1397 dmreq->iv_sector >>= cc->sector_shift;
1400 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1402 iv = iv_of_dmreq(cc, dmreq);
1403 org_iv = org_iv_of_dmreq(cc, dmreq);
1404 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1406 sector = org_sector_of_dmreq(cc, dmreq);
1407 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1409 /* For skcipher we use only the first sg item */
1410 sg_in = &dmreq->sg_in[0];
1411 sg_out = &dmreq->sg_out[0];
1413 sg_init_table(sg_in, 1);
1414 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1416 sg_init_table(sg_out, 1);
1417 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1419 if (cc->iv_gen_ops) {
1420 /* For READs use IV stored in integrity metadata */
1421 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1422 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1424 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1427 /* Data can be already preprocessed in generator */
1428 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1430 /* Store generated IV in integrity metadata */
1431 if (cc->integrity_iv_size)
1432 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1434 /* Working copy of IV, to be modified in crypto API */
1435 memcpy(iv, org_iv, cc->iv_size);
1438 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1440 if (bio_data_dir(ctx->bio_in) == WRITE)
1441 r = crypto_skcipher_encrypt(req);
1443 r = crypto_skcipher_decrypt(req);
1445 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1446 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1448 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1449 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1454 static void kcryptd_async_done(struct crypto_async_request *async_req,
1457 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1458 struct convert_context *ctx)
1460 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1463 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1468 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1471 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1472 * requests if driver request queue is full.
1474 skcipher_request_set_callback(ctx->r.req,
1475 CRYPTO_TFM_REQ_MAY_BACKLOG,
1476 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1481 static int crypt_alloc_req_aead(struct crypt_config *cc,
1482 struct convert_context *ctx)
1484 if (!ctx->r.req_aead) {
1485 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1486 if (!ctx->r.req_aead)
1490 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1493 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1494 * requests if driver request queue is full.
1496 aead_request_set_callback(ctx->r.req_aead,
1497 CRYPTO_TFM_REQ_MAY_BACKLOG,
1498 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1503 static int crypt_alloc_req(struct crypt_config *cc,
1504 struct convert_context *ctx)
1506 if (crypt_integrity_aead(cc))
1507 return crypt_alloc_req_aead(cc, ctx);
1509 return crypt_alloc_req_skcipher(cc, ctx);
1512 static void crypt_free_req_skcipher(struct crypt_config *cc,
1513 struct skcipher_request *req, struct bio *base_bio)
1515 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1517 if ((struct skcipher_request *)(io + 1) != req)
1518 mempool_free(req, &cc->req_pool);
1521 static void crypt_free_req_aead(struct crypt_config *cc,
1522 struct aead_request *req, struct bio *base_bio)
1524 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1526 if ((struct aead_request *)(io + 1) != req)
1527 mempool_free(req, &cc->req_pool);
1530 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1532 if (crypt_integrity_aead(cc))
1533 crypt_free_req_aead(cc, req, base_bio);
1535 crypt_free_req_skcipher(cc, req, base_bio);
1539 * Encrypt / decrypt data from one bio to another one (can be the same one)
1541 static blk_status_t crypt_convert(struct crypt_config *cc,
1542 struct convert_context *ctx, bool atomic, bool reset_pending)
1544 unsigned int tag_offset = 0;
1545 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1549 * if reset_pending is set we are dealing with the bio for the first time,
1550 * else we're continuing to work on the previous bio, so don't mess with
1551 * the cc_pending counter
1554 atomic_set(&ctx->cc_pending, 1);
1556 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1558 r = crypt_alloc_req(cc, ctx);
1560 complete(&ctx->restart);
1561 return BLK_STS_DEV_RESOURCE;
1564 atomic_inc(&ctx->cc_pending);
1566 if (crypt_integrity_aead(cc))
1567 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1569 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1573 * The request was queued by a crypto driver
1574 * but the driver request queue is full, let's wait.
1577 if (in_interrupt()) {
1578 if (try_wait_for_completion(&ctx->restart)) {
1580 * we don't have to block to wait for completion,
1585 * we can't wait for completion without blocking
1586 * exit and continue processing in a workqueue
1589 ctx->cc_sector += sector_step;
1591 return BLK_STS_DEV_RESOURCE;
1594 wait_for_completion(&ctx->restart);
1596 reinit_completion(&ctx->restart);
1599 * The request is queued and processed asynchronously,
1600 * completion function kcryptd_async_done() will be called.
1604 ctx->cc_sector += sector_step;
1608 * The request was already processed (synchronously).
1611 atomic_dec(&ctx->cc_pending);
1612 ctx->cc_sector += sector_step;
1618 * There was a data integrity error.
1621 atomic_dec(&ctx->cc_pending);
1622 return BLK_STS_PROTECTION;
1624 * There was an error while processing the request.
1627 atomic_dec(&ctx->cc_pending);
1628 return BLK_STS_IOERR;
1635 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1638 * Generate a new unfragmented bio with the given size
1639 * This should never violate the device limitations (but only because
1640 * max_segment_size is being constrained to PAGE_SIZE).
1642 * This function may be called concurrently. If we allocate from the mempool
1643 * concurrently, there is a possibility of deadlock. For example, if we have
1644 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1645 * the mempool concurrently, it may deadlock in a situation where both processes
1646 * have allocated 128 pages and the mempool is exhausted.
1648 * In order to avoid this scenario we allocate the pages under a mutex.
1650 * In order to not degrade performance with excessive locking, we try
1651 * non-blocking allocations without a mutex first but on failure we fallback
1652 * to blocking allocations with a mutex.
1654 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1656 struct crypt_config *cc = io->cc;
1658 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1659 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1660 unsigned i, len, remaining_size;
1664 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1665 mutex_lock(&cc->bio_alloc_lock);
1667 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1671 clone_init(io, clone);
1673 remaining_size = size;
1675 for (i = 0; i < nr_iovecs; i++) {
1676 page = mempool_alloc(&cc->page_pool, gfp_mask);
1678 crypt_free_buffer_pages(cc, clone);
1680 gfp_mask |= __GFP_DIRECT_RECLAIM;
1684 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1686 bio_add_page(clone, page, len, 0);
1688 remaining_size -= len;
1691 /* Allocate space for integrity tags */
1692 if (dm_crypt_integrity_io_alloc(io, clone)) {
1693 crypt_free_buffer_pages(cc, clone);
1698 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1699 mutex_unlock(&cc->bio_alloc_lock);
1704 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1707 struct bvec_iter_all iter_all;
1709 bio_for_each_segment_all(bv, clone, iter_all) {
1710 BUG_ON(!bv->bv_page);
1711 mempool_free(bv->bv_page, &cc->page_pool);
1715 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1716 struct bio *bio, sector_t sector)
1720 io->sector = sector;
1722 io->ctx.r.req = NULL;
1723 io->integrity_metadata = NULL;
1724 io->integrity_metadata_from_pool = false;
1725 atomic_set(&io->io_pending, 0);
1728 static void crypt_inc_pending(struct dm_crypt_io *io)
1730 atomic_inc(&io->io_pending);
1733 static void kcryptd_io_bio_endio(struct work_struct *work)
1735 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1736 bio_endio(io->base_bio);
1740 * One of the bios was finished. Check for completion of
1741 * the whole request and correctly clean up the buffer.
1743 static void crypt_dec_pending(struct dm_crypt_io *io)
1745 struct crypt_config *cc = io->cc;
1746 struct bio *base_bio = io->base_bio;
1747 blk_status_t error = io->error;
1749 if (!atomic_dec_and_test(&io->io_pending))
1753 crypt_free_req(cc, io->ctx.r.req, base_bio);
1755 if (unlikely(io->integrity_metadata_from_pool))
1756 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1758 kfree(io->integrity_metadata);
1760 base_bio->bi_status = error;
1763 * If we are running this function from our tasklet,
1764 * we can't call bio_endio() here, because it will call
1765 * clone_endio() from dm.c, which in turn will
1766 * free the current struct dm_crypt_io structure with
1767 * our tasklet. In this case we need to delay bio_endio()
1768 * execution to after the tasklet is done and dequeued.
1770 if (tasklet_trylock(&io->tasklet)) {
1771 tasklet_unlock(&io->tasklet);
1772 bio_endio(base_bio);
1776 INIT_WORK(&io->work, kcryptd_io_bio_endio);
1777 queue_work(cc->io_queue, &io->work);
1781 * kcryptd/kcryptd_io:
1783 * Needed because it would be very unwise to do decryption in an
1784 * interrupt context.
1786 * kcryptd performs the actual encryption or decryption.
1788 * kcryptd_io performs the IO submission.
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.
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.
1797 static void crypt_endio(struct bio *clone)
1799 struct dm_crypt_io *io = clone->bi_private;
1800 struct crypt_config *cc = io->cc;
1801 unsigned rw = bio_data_dir(clone);
1805 * free the processed pages
1808 crypt_free_buffer_pages(cc, clone);
1810 error = clone->bi_status;
1813 if (rw == READ && !error) {
1814 kcryptd_queue_crypt(io);
1818 if (unlikely(error))
1821 crypt_dec_pending(io);
1824 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1826 struct crypt_config *cc = io->cc;
1828 clone->bi_private = io;
1829 clone->bi_end_io = crypt_endio;
1830 bio_set_dev(clone, cc->dev->bdev);
1831 clone->bi_opf = io->base_bio->bi_opf;
1834 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1836 struct crypt_config *cc = io->cc;
1840 * We need the original biovec array in order to decrypt
1841 * the whole bio data *afterwards* -- thanks to immutable
1842 * biovecs we don't need to worry about the block layer
1843 * modifying the biovec array; so leverage bio_clone_fast().
1845 clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1849 crypt_inc_pending(io);
1851 clone_init(io, clone);
1852 clone->bi_iter.bi_sector = cc->start + io->sector;
1854 if (dm_crypt_integrity_io_alloc(io, clone)) {
1855 crypt_dec_pending(io);
1860 submit_bio_noacct(clone);
1864 static void kcryptd_io_read_work(struct work_struct *work)
1866 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1868 crypt_inc_pending(io);
1869 if (kcryptd_io_read(io, GFP_NOIO))
1870 io->error = BLK_STS_RESOURCE;
1871 crypt_dec_pending(io);
1874 static void kcryptd_queue_read(struct dm_crypt_io *io)
1876 struct crypt_config *cc = io->cc;
1878 INIT_WORK(&io->work, kcryptd_io_read_work);
1879 queue_work(cc->io_queue, &io->work);
1882 static void kcryptd_io_write(struct dm_crypt_io *io)
1884 struct bio *clone = io->ctx.bio_out;
1886 submit_bio_noacct(clone);
1889 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1891 static int dmcrypt_write(void *data)
1893 struct crypt_config *cc = data;
1894 struct dm_crypt_io *io;
1897 struct rb_root write_tree;
1898 struct blk_plug plug;
1900 spin_lock_irq(&cc->write_thread_lock);
1903 if (!RB_EMPTY_ROOT(&cc->write_tree))
1906 set_current_state(TASK_INTERRUPTIBLE);
1908 spin_unlock_irq(&cc->write_thread_lock);
1910 if (unlikely(kthread_should_stop())) {
1911 set_current_state(TASK_RUNNING);
1917 set_current_state(TASK_RUNNING);
1918 spin_lock_irq(&cc->write_thread_lock);
1919 goto continue_locked;
1922 write_tree = cc->write_tree;
1923 cc->write_tree = RB_ROOT;
1924 spin_unlock_irq(&cc->write_thread_lock);
1926 BUG_ON(rb_parent(write_tree.rb_node));
1929 * Note: we cannot walk the tree here with rb_next because
1930 * the structures may be freed when kcryptd_io_write is called.
1932 blk_start_plug(&plug);
1934 io = crypt_io_from_node(rb_first(&write_tree));
1935 rb_erase(&io->rb_node, &write_tree);
1936 kcryptd_io_write(io);
1937 } while (!RB_EMPTY_ROOT(&write_tree));
1938 blk_finish_plug(&plug);
1943 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1945 struct bio *clone = io->ctx.bio_out;
1946 struct crypt_config *cc = io->cc;
1947 unsigned long flags;
1949 struct rb_node **rbp, *parent;
1951 if (unlikely(io->error)) {
1952 crypt_free_buffer_pages(cc, clone);
1954 crypt_dec_pending(io);
1958 /* crypt_convert should have filled the clone bio */
1959 BUG_ON(io->ctx.iter_out.bi_size);
1961 clone->bi_iter.bi_sector = cc->start + io->sector;
1963 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1964 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1965 submit_bio_noacct(clone);
1969 spin_lock_irqsave(&cc->write_thread_lock, flags);
1970 if (RB_EMPTY_ROOT(&cc->write_tree))
1971 wake_up_process(cc->write_thread);
1972 rbp = &cc->write_tree.rb_node;
1974 sector = io->sector;
1977 if (sector < crypt_io_from_node(parent)->sector)
1978 rbp = &(*rbp)->rb_left;
1980 rbp = &(*rbp)->rb_right;
1982 rb_link_node(&io->rb_node, parent, rbp);
1983 rb_insert_color(&io->rb_node, &cc->write_tree);
1984 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1987 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1988 struct convert_context *ctx)
1991 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1995 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1996 * constraints so they do not need to be issued inline by
1997 * kcryptd_crypt_write_convert().
1999 switch (bio_op(ctx->bio_in)) {
2001 case REQ_OP_WRITE_SAME:
2002 case REQ_OP_WRITE_ZEROES:
2009 static void kcryptd_crypt_write_continue(struct work_struct *work)
2011 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2012 struct crypt_config *cc = io->cc;
2013 struct convert_context *ctx = &io->ctx;
2015 sector_t sector = io->sector;
2018 wait_for_completion(&ctx->restart);
2019 reinit_completion(&ctx->restart);
2021 r = crypt_convert(cc, &io->ctx, true, false);
2024 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2025 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2026 /* Wait for completion signaled by kcryptd_async_done() */
2027 wait_for_completion(&ctx->restart);
2031 /* Encryption was already finished, submit io now */
2032 if (crypt_finished) {
2033 kcryptd_crypt_write_io_submit(io, 0);
2034 io->sector = sector;
2037 crypt_dec_pending(io);
2040 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2042 struct crypt_config *cc = io->cc;
2043 struct convert_context *ctx = &io->ctx;
2046 sector_t sector = io->sector;
2050 * Prevent io from disappearing until this function completes.
2052 crypt_inc_pending(io);
2053 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2055 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2056 if (unlikely(!clone)) {
2057 io->error = BLK_STS_IOERR;
2061 io->ctx.bio_out = clone;
2062 io->ctx.iter_out = clone->bi_iter;
2064 sector += bio_sectors(clone);
2066 crypt_inc_pending(io);
2067 r = crypt_convert(cc, ctx,
2068 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2070 * Crypto API backlogged the request, because its queue was full
2071 * and we're in softirq context, so continue from a workqueue
2072 * (TODO: is it actually possible to be in softirq in the write path?)
2074 if (r == BLK_STS_DEV_RESOURCE) {
2075 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2076 queue_work(cc->crypt_queue, &io->work);
2081 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2082 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2083 /* Wait for completion signaled by kcryptd_async_done() */
2084 wait_for_completion(&ctx->restart);
2088 /* Encryption was already finished, submit io now */
2089 if (crypt_finished) {
2090 kcryptd_crypt_write_io_submit(io, 0);
2091 io->sector = sector;
2095 crypt_dec_pending(io);
2098 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2100 crypt_dec_pending(io);
2103 static void kcryptd_crypt_read_continue(struct work_struct *work)
2105 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2106 struct crypt_config *cc = io->cc;
2109 wait_for_completion(&io->ctx.restart);
2110 reinit_completion(&io->ctx.restart);
2112 r = crypt_convert(cc, &io->ctx, true, false);
2116 if (atomic_dec_and_test(&io->ctx.cc_pending))
2117 kcryptd_crypt_read_done(io);
2119 crypt_dec_pending(io);
2122 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2124 struct crypt_config *cc = io->cc;
2127 crypt_inc_pending(io);
2129 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2132 r = crypt_convert(cc, &io->ctx,
2133 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2135 * Crypto API backlogged the request, because its queue was full
2136 * and we're in softirq context, so continue from a workqueue
2138 if (r == BLK_STS_DEV_RESOURCE) {
2139 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2140 queue_work(cc->crypt_queue, &io->work);
2146 if (atomic_dec_and_test(&io->ctx.cc_pending))
2147 kcryptd_crypt_read_done(io);
2149 crypt_dec_pending(io);
2152 static void kcryptd_async_done(struct crypto_async_request *async_req,
2155 struct dm_crypt_request *dmreq = async_req->data;
2156 struct convert_context *ctx = dmreq->ctx;
2157 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2158 struct crypt_config *cc = io->cc;
2161 * A request from crypto driver backlog is going to be processed now,
2162 * finish the completion and continue in crypt_convert().
2163 * (Callback will be called for the second time for this request.)
2165 if (error == -EINPROGRESS) {
2166 complete(&ctx->restart);
2170 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2171 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2173 if (error == -EBADMSG) {
2174 char b[BDEVNAME_SIZE];
2175 DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2176 (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
2177 io->error = BLK_STS_PROTECTION;
2178 } else if (error < 0)
2179 io->error = BLK_STS_IOERR;
2181 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2183 if (!atomic_dec_and_test(&ctx->cc_pending))
2187 * The request is fully completed: for inline writes, let
2188 * kcryptd_crypt_write_convert() do the IO submission.
2190 if (bio_data_dir(io->base_bio) == READ) {
2191 kcryptd_crypt_read_done(io);
2195 if (kcryptd_crypt_write_inline(cc, ctx)) {
2196 complete(&ctx->restart);
2200 kcryptd_crypt_write_io_submit(io, 1);
2203 static void kcryptd_crypt(struct work_struct *work)
2205 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2207 if (bio_data_dir(io->base_bio) == READ)
2208 kcryptd_crypt_read_convert(io);
2210 kcryptd_crypt_write_convert(io);
2213 static void kcryptd_crypt_tasklet(unsigned long work)
2215 kcryptd_crypt((struct work_struct *)work);
2218 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2220 struct crypt_config *cc = io->cc;
2222 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2223 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2225 * in_irq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2226 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2227 * it is being executed with irqs disabled.
2229 if (in_irq() || irqs_disabled()) {
2230 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2231 tasklet_schedule(&io->tasklet);
2235 kcryptd_crypt(&io->work);
2239 INIT_WORK(&io->work, kcryptd_crypt);
2240 queue_work(cc->crypt_queue, &io->work);
2243 static void crypt_free_tfms_aead(struct crypt_config *cc)
2245 if (!cc->cipher_tfm.tfms_aead)
2248 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2249 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2250 cc->cipher_tfm.tfms_aead[0] = NULL;
2253 kfree(cc->cipher_tfm.tfms_aead);
2254 cc->cipher_tfm.tfms_aead = NULL;
2257 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2261 if (!cc->cipher_tfm.tfms)
2264 for (i = 0; i < cc->tfms_count; i++)
2265 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2266 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2267 cc->cipher_tfm.tfms[i] = NULL;
2270 kfree(cc->cipher_tfm.tfms);
2271 cc->cipher_tfm.tfms = NULL;
2274 static void crypt_free_tfms(struct crypt_config *cc)
2276 if (crypt_integrity_aead(cc))
2277 crypt_free_tfms_aead(cc);
2279 crypt_free_tfms_skcipher(cc);
2282 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2287 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2288 sizeof(struct crypto_skcipher *),
2290 if (!cc->cipher_tfm.tfms)
2293 for (i = 0; i < cc->tfms_count; i++) {
2294 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2295 CRYPTO_ALG_ALLOCATES_MEMORY);
2296 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2297 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2298 crypt_free_tfms(cc);
2304 * dm-crypt performance can vary greatly depending on which crypto
2305 * algorithm implementation is used. Help people debug performance
2306 * problems by logging the ->cra_driver_name.
2308 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2309 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2313 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2317 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2318 if (!cc->cipher_tfm.tfms)
2321 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2322 CRYPTO_ALG_ALLOCATES_MEMORY);
2323 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2324 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2325 crypt_free_tfms(cc);
2329 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2330 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2334 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2336 if (crypt_integrity_aead(cc))
2337 return crypt_alloc_tfms_aead(cc, ciphermode);
2339 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2342 static unsigned crypt_subkey_size(struct crypt_config *cc)
2344 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2347 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2349 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2353 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2354 * the key must be for some reason in special format.
2355 * This funcion converts cc->key to this special format.
2357 static void crypt_copy_authenckey(char *p, const void *key,
2358 unsigned enckeylen, unsigned authkeylen)
2360 struct crypto_authenc_key_param *param;
2363 rta = (struct rtattr *)p;
2364 param = RTA_DATA(rta);
2365 param->enckeylen = cpu_to_be32(enckeylen);
2366 rta->rta_len = RTA_LENGTH(sizeof(*param));
2367 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2368 p += RTA_SPACE(sizeof(*param));
2369 memcpy(p, key + enckeylen, authkeylen);
2371 memcpy(p, key, enckeylen);
2374 static int crypt_setkey(struct crypt_config *cc)
2376 unsigned subkey_size;
2379 /* Ignore extra keys (which are used for IV etc) */
2380 subkey_size = crypt_subkey_size(cc);
2382 if (crypt_integrity_hmac(cc)) {
2383 if (subkey_size < cc->key_mac_size)
2386 crypt_copy_authenckey(cc->authenc_key, cc->key,
2387 subkey_size - cc->key_mac_size,
2391 for (i = 0; i < cc->tfms_count; i++) {
2392 if (crypt_integrity_hmac(cc))
2393 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2394 cc->authenc_key, crypt_authenckey_size(cc));
2395 else if (crypt_integrity_aead(cc))
2396 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2397 cc->key + (i * subkey_size),
2400 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2401 cc->key + (i * subkey_size),
2407 if (crypt_integrity_hmac(cc))
2408 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2415 static bool contains_whitespace(const char *str)
2418 if (isspace(*str++))
2423 static int set_key_user(struct crypt_config *cc, struct key *key)
2425 const struct user_key_payload *ukp;
2427 ukp = user_key_payload_locked(key);
2429 return -EKEYREVOKED;
2431 if (cc->key_size != ukp->datalen)
2434 memcpy(cc->key, ukp->data, cc->key_size);
2439 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
2440 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2442 const struct encrypted_key_payload *ekp;
2444 ekp = key->payload.data[0];
2446 return -EKEYREVOKED;
2448 if (cc->key_size != ekp->decrypted_datalen)
2451 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2455 #endif /* CONFIG_ENCRYPTED_KEYS */
2457 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2459 char *new_key_string, *key_desc;
2461 struct key_type *type;
2463 int (*set_key)(struct crypt_config *cc, struct key *key);
2466 * Reject key_string with whitespace. dm core currently lacks code for
2467 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2469 if (contains_whitespace(key_string)) {
2470 DMERR("whitespace chars not allowed in key string");
2474 /* look for next ':' separating key_type from key_description */
2475 key_desc = strpbrk(key_string, ":");
2476 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2479 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2480 type = &key_type_logon;
2481 set_key = set_key_user;
2482 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2483 type = &key_type_user;
2484 set_key = set_key_user;
2485 #if defined(CONFIG_ENCRYPTED_KEYS) || defined(CONFIG_ENCRYPTED_KEYS_MODULE)
2486 } else if (!strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2487 type = &key_type_encrypted;
2488 set_key = set_key_encrypted;
2494 new_key_string = kstrdup(key_string, GFP_KERNEL);
2495 if (!new_key_string)
2498 key = request_key(type, key_desc + 1, NULL);
2500 kfree_sensitive(new_key_string);
2501 return PTR_ERR(key);
2504 down_read(&key->sem);
2506 ret = set_key(cc, key);
2510 kfree_sensitive(new_key_string);
2517 /* clear the flag since following operations may invalidate previously valid key */
2518 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2520 ret = crypt_setkey(cc);
2523 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2524 kfree_sensitive(cc->key_string);
2525 cc->key_string = new_key_string;
2527 kfree_sensitive(new_key_string);
2532 static int get_key_size(char **key_string)
2537 if (*key_string[0] != ':')
2538 return strlen(*key_string) >> 1;
2540 /* look for next ':' in key string */
2541 colon = strpbrk(*key_string + 1, ":");
2545 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2548 *key_string = colon;
2550 /* remaining key string should be :<logon|user>:<key_desc> */
2557 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2562 static int get_key_size(char **key_string)
2564 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2567 #endif /* CONFIG_KEYS */
2569 static int crypt_set_key(struct crypt_config *cc, char *key)
2572 int key_string_len = strlen(key);
2574 /* Hyphen (which gives a key_size of zero) means there is no key. */
2575 if (!cc->key_size && strcmp(key, "-"))
2578 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2579 if (key[0] == ':') {
2580 r = crypt_set_keyring_key(cc, key + 1);
2584 /* clear the flag since following operations may invalidate previously valid key */
2585 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2587 /* wipe references to any kernel keyring key */
2588 kfree_sensitive(cc->key_string);
2589 cc->key_string = NULL;
2591 /* Decode key from its hex representation. */
2592 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2595 r = crypt_setkey(cc);
2597 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2600 /* Hex key string not needed after here, so wipe it. */
2601 memset(key, '0', key_string_len);
2606 static int crypt_wipe_key(struct crypt_config *cc)
2610 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2611 get_random_bytes(&cc->key, cc->key_size);
2613 /* Wipe IV private keys */
2614 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2615 r = cc->iv_gen_ops->wipe(cc);
2620 kfree_sensitive(cc->key_string);
2621 cc->key_string = NULL;
2622 r = crypt_setkey(cc);
2623 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2628 static void crypt_calculate_pages_per_client(void)
2630 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2632 if (!dm_crypt_clients_n)
2635 pages /= dm_crypt_clients_n;
2636 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2637 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2638 dm_crypt_pages_per_client = pages;
2641 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2643 struct crypt_config *cc = pool_data;
2647 * Note, percpu_counter_read_positive() may over (and under) estimate
2648 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2649 * but avoids potential spinlock contention of an exact result.
2651 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2652 likely(gfp_mask & __GFP_NORETRY))
2655 page = alloc_page(gfp_mask);
2656 if (likely(page != NULL))
2657 percpu_counter_add(&cc->n_allocated_pages, 1);
2662 static void crypt_page_free(void *page, void *pool_data)
2664 struct crypt_config *cc = pool_data;
2667 percpu_counter_sub(&cc->n_allocated_pages, 1);
2670 static void crypt_dtr(struct dm_target *ti)
2672 struct crypt_config *cc = ti->private;
2679 if (cc->write_thread)
2680 kthread_stop(cc->write_thread);
2683 destroy_workqueue(cc->io_queue);
2684 if (cc->crypt_queue)
2685 destroy_workqueue(cc->crypt_queue);
2687 crypt_free_tfms(cc);
2689 bioset_exit(&cc->bs);
2691 mempool_exit(&cc->page_pool);
2692 mempool_exit(&cc->req_pool);
2693 mempool_exit(&cc->tag_pool);
2695 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2696 percpu_counter_destroy(&cc->n_allocated_pages);
2698 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2699 cc->iv_gen_ops->dtr(cc);
2702 dm_put_device(ti, cc->dev);
2704 kfree_sensitive(cc->cipher_string);
2705 kfree_sensitive(cc->key_string);
2706 kfree_sensitive(cc->cipher_auth);
2707 kfree_sensitive(cc->authenc_key);
2709 mutex_destroy(&cc->bio_alloc_lock);
2711 /* Must zero key material before freeing */
2712 kfree_sensitive(cc);
2714 spin_lock(&dm_crypt_clients_lock);
2715 WARN_ON(!dm_crypt_clients_n);
2716 dm_crypt_clients_n--;
2717 crypt_calculate_pages_per_client();
2718 spin_unlock(&dm_crypt_clients_lock);
2721 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2723 struct crypt_config *cc = ti->private;
2725 if (crypt_integrity_aead(cc))
2726 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2728 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2731 /* at least a 64 bit sector number should fit in our buffer */
2732 cc->iv_size = max(cc->iv_size,
2733 (unsigned int)(sizeof(u64) / sizeof(u8)));
2735 DMWARN("Selected cipher does not support IVs");
2739 /* Choose ivmode, see comments at iv code. */
2741 cc->iv_gen_ops = NULL;
2742 else if (strcmp(ivmode, "plain") == 0)
2743 cc->iv_gen_ops = &crypt_iv_plain_ops;
2744 else if (strcmp(ivmode, "plain64") == 0)
2745 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2746 else if (strcmp(ivmode, "plain64be") == 0)
2747 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2748 else if (strcmp(ivmode, "essiv") == 0)
2749 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2750 else if (strcmp(ivmode, "benbi") == 0)
2751 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2752 else if (strcmp(ivmode, "null") == 0)
2753 cc->iv_gen_ops = &crypt_iv_null_ops;
2754 else if (strcmp(ivmode, "eboiv") == 0)
2755 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2756 else if (strcmp(ivmode, "elephant") == 0) {
2757 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2759 cc->key_extra_size = cc->key_size / 2;
2760 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2762 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2763 } else if (strcmp(ivmode, "lmk") == 0) {
2764 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2766 * Version 2 and 3 is recognised according
2767 * to length of provided multi-key string.
2768 * If present (version 3), last key is used as IV seed.
2769 * All keys (including IV seed) are always the same size.
2771 if (cc->key_size % cc->key_parts) {
2773 cc->key_extra_size = cc->key_size / cc->key_parts;
2775 } else if (strcmp(ivmode, "tcw") == 0) {
2776 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2777 cc->key_parts += 2; /* IV + whitening */
2778 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2779 } else if (strcmp(ivmode, "random") == 0) {
2780 cc->iv_gen_ops = &crypt_iv_random_ops;
2781 /* Need storage space in integrity fields. */
2782 cc->integrity_iv_size = cc->iv_size;
2784 ti->error = "Invalid IV mode";
2792 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2793 * The HMAC is needed to calculate tag size (HMAC digest size).
2794 * This should be probably done by crypto-api calls (once available...)
2796 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2798 char *start, *end, *mac_alg = NULL;
2799 struct crypto_ahash *mac;
2801 if (!strstarts(cipher_api, "authenc("))
2804 start = strchr(cipher_api, '(');
2805 end = strchr(cipher_api, ',');
2806 if (!start || !end || ++start > end)
2809 mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2812 strncpy(mac_alg, start, end - start);
2814 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2818 return PTR_ERR(mac);
2820 cc->key_mac_size = crypto_ahash_digestsize(mac);
2821 crypto_free_ahash(mac);
2823 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2824 if (!cc->authenc_key)
2830 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2831 char **ivmode, char **ivopts)
2833 struct crypt_config *cc = ti->private;
2834 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2840 * New format (capi: prefix)
2841 * capi:cipher_api_spec-iv:ivopts
2843 tmp = &cipher_in[strlen("capi:")];
2845 /* Separate IV options if present, it can contain another '-' in hash name */
2846 *ivopts = strrchr(tmp, ':');
2852 *ivmode = strrchr(tmp, '-');
2857 /* The rest is crypto API spec */
2860 /* Alloc AEAD, can be used only in new format. */
2861 if (crypt_integrity_aead(cc)) {
2862 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2864 ti->error = "Invalid AEAD cipher spec";
2869 if (*ivmode && !strcmp(*ivmode, "lmk"))
2870 cc->tfms_count = 64;
2872 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2874 ti->error = "Digest algorithm missing for ESSIV mode";
2877 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2878 cipher_api, *ivopts);
2879 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2880 ti->error = "Cannot allocate cipher string";
2886 cc->key_parts = cc->tfms_count;
2888 /* Allocate cipher */
2889 ret = crypt_alloc_tfms(cc, cipher_api);
2891 ti->error = "Error allocating crypto tfm";
2895 if (crypt_integrity_aead(cc))
2896 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2898 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2903 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2904 char **ivmode, char **ivopts)
2906 struct crypt_config *cc = ti->private;
2907 char *tmp, *cipher, *chainmode, *keycount;
2908 char *cipher_api = NULL;
2912 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2913 ti->error = "Bad cipher specification";
2918 * Legacy dm-crypt cipher specification
2919 * cipher[:keycount]-mode-iv:ivopts
2922 keycount = strsep(&tmp, "-");
2923 cipher = strsep(&keycount, ":");
2927 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2928 !is_power_of_2(cc->tfms_count)) {
2929 ti->error = "Bad cipher key count specification";
2932 cc->key_parts = cc->tfms_count;
2934 chainmode = strsep(&tmp, "-");
2935 *ivmode = strsep(&tmp, ":");
2939 * For compatibility with the original dm-crypt mapping format, if
2940 * only the cipher name is supplied, use cbc-plain.
2942 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2947 if (strcmp(chainmode, "ecb") && !*ivmode) {
2948 ti->error = "IV mechanism required";
2952 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2956 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2958 ti->error = "Digest algorithm missing for ESSIV mode";
2962 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2963 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2965 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2966 "%s(%s)", chainmode, cipher);
2968 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2973 /* Allocate cipher */
2974 ret = crypt_alloc_tfms(cc, cipher_api);
2976 ti->error = "Error allocating crypto tfm";
2984 ti->error = "Cannot allocate cipher strings";
2988 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
2990 struct crypt_config *cc = ti->private;
2991 char *ivmode = NULL, *ivopts = NULL;
2994 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
2995 if (!cc->cipher_string) {
2996 ti->error = "Cannot allocate cipher strings";
3000 if (strstarts(cipher_in, "capi:"))
3001 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3003 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3008 ret = crypt_ctr_ivmode(ti, ivmode);
3012 /* Initialize and set key */
3013 ret = crypt_set_key(cc, key);
3015 ti->error = "Error decoding and setting key";
3020 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3021 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3023 ti->error = "Error creating IV";
3028 /* Initialize IV (set keys for ESSIV etc) */
3029 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3030 ret = cc->iv_gen_ops->init(cc);
3032 ti->error = "Error initialising IV";
3037 /* wipe the kernel key payload copy */
3039 memset(cc->key, 0, cc->key_size * sizeof(u8));
3044 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3046 struct crypt_config *cc = ti->private;
3047 struct dm_arg_set as;
3048 static const struct dm_arg _args[] = {
3049 {0, 8, "Invalid number of feature args"},
3051 unsigned int opt_params, val;
3052 const char *opt_string, *sval;
3056 /* Optional parameters */
3060 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3064 while (opt_params--) {
3065 opt_string = dm_shift_arg(&as);
3067 ti->error = "Not enough feature arguments";
3071 if (!strcasecmp(opt_string, "allow_discards"))
3072 ti->num_discard_bios = 1;
3074 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3075 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3077 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3078 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3079 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3080 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3081 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3082 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3083 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3084 if (val == 0 || val > MAX_TAG_SIZE) {
3085 ti->error = "Invalid integrity arguments";
3088 cc->on_disk_tag_size = val;
3089 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3090 if (!strcasecmp(sval, "aead")) {
3091 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3092 } else if (strcasecmp(sval, "none")) {
3093 ti->error = "Unknown integrity profile";
3097 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3098 if (!cc->cipher_auth)
3100 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3101 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3102 cc->sector_size > 4096 ||
3103 (cc->sector_size & (cc->sector_size - 1))) {
3104 ti->error = "Invalid feature value for sector_size";
3107 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3108 ti->error = "Device size is not multiple of sector_size feature";
3111 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3112 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3113 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3115 ti->error = "Invalid feature arguments";
3123 #ifdef CONFIG_BLK_DEV_ZONED
3125 static int crypt_report_zones(struct dm_target *ti,
3126 struct dm_report_zones_args *args, unsigned int nr_zones)
3128 struct crypt_config *cc = ti->private;
3129 sector_t sector = cc->start + dm_target_offset(ti, args->next_sector);
3131 args->start = cc->start;
3132 return blkdev_report_zones(cc->dev->bdev, sector, nr_zones,
3133 dm_report_zones_cb, args);
3139 * Construct an encryption mapping:
3140 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3142 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3144 struct crypt_config *cc;
3145 const char *devname = dm_table_device_name(ti->table);
3147 unsigned int align_mask;
3148 unsigned long long tmpll;
3150 size_t iv_size_padding, additional_req_size;
3154 ti->error = "Not enough arguments";
3158 key_size = get_key_size(&argv[1]);
3160 ti->error = "Cannot parse key size";
3164 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3166 ti->error = "Cannot allocate encryption context";
3169 cc->key_size = key_size;
3170 cc->sector_size = (1 << SECTOR_SHIFT);
3171 cc->sector_shift = 0;
3175 spin_lock(&dm_crypt_clients_lock);
3176 dm_crypt_clients_n++;
3177 crypt_calculate_pages_per_client();
3178 spin_unlock(&dm_crypt_clients_lock);
3180 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3184 /* Optional parameters need to be read before cipher constructor */
3186 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3191 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3195 if (crypt_integrity_aead(cc)) {
3196 cc->dmreq_start = sizeof(struct aead_request);
3197 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3198 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3200 cc->dmreq_start = sizeof(struct skcipher_request);
3201 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3202 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3204 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3206 if (align_mask < CRYPTO_MINALIGN) {
3207 /* Allocate the padding exactly */
3208 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3212 * If the cipher requires greater alignment than kmalloc
3213 * alignment, we don't know the exact position of the
3214 * initialization vector. We must assume worst case.
3216 iv_size_padding = align_mask;
3219 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3220 additional_req_size = sizeof(struct dm_crypt_request) +
3221 iv_size_padding + cc->iv_size +
3224 sizeof(unsigned int);
3226 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3228 ti->error = "Cannot allocate crypt request mempool";
3232 cc->per_bio_data_size = ti->per_io_data_size =
3233 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3234 ARCH_KMALLOC_MINALIGN);
3236 ret = mempool_init(&cc->page_pool, BIO_MAX_PAGES, crypt_page_alloc, crypt_page_free, cc);
3238 ti->error = "Cannot allocate page mempool";
3242 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3244 ti->error = "Cannot allocate crypt bioset";
3248 mutex_init(&cc->bio_alloc_lock);
3251 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3252 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3253 ti->error = "Invalid iv_offset sector";
3256 cc->iv_offset = tmpll;
3258 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3260 ti->error = "Device lookup failed";
3265 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3266 ti->error = "Invalid device sector";
3272 * For zoned block devices, we need to preserve the issuer write
3273 * ordering. To do so, disable write workqueues and force inline
3274 * encryption completion.
3276 if (bdev_is_zoned(cc->dev->bdev)) {
3277 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3278 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3281 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3282 ret = crypt_integrity_ctr(cc, ti);
3286 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3287 if (!cc->tag_pool_max_sectors)
3288 cc->tag_pool_max_sectors = 1;
3290 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3291 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3293 ti->error = "Cannot allocate integrity tags mempool";
3297 cc->tag_pool_max_sectors <<= cc->sector_shift;
3301 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3302 if (!cc->io_queue) {
3303 ti->error = "Couldn't create kcryptd io queue";
3307 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3308 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3311 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3312 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3313 num_online_cpus(), devname);
3314 if (!cc->crypt_queue) {
3315 ti->error = "Couldn't create kcryptd queue";
3319 spin_lock_init(&cc->write_thread_lock);
3320 cc->write_tree = RB_ROOT;
3322 cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3323 if (IS_ERR(cc->write_thread)) {
3324 ret = PTR_ERR(cc->write_thread);
3325 cc->write_thread = NULL;
3326 ti->error = "Couldn't spawn write thread";
3329 wake_up_process(cc->write_thread);
3331 ti->num_flush_bios = 1;
3332 ti->limit_swap_bios = true;
3341 static int crypt_map(struct dm_target *ti, struct bio *bio)
3343 struct dm_crypt_io *io;
3344 struct crypt_config *cc = ti->private;
3347 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3348 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3349 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3351 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3352 bio_op(bio) == REQ_OP_DISCARD)) {
3353 bio_set_dev(bio, cc->dev->bdev);
3354 if (bio_sectors(bio))
3355 bio->bi_iter.bi_sector = cc->start +
3356 dm_target_offset(ti, bio->bi_iter.bi_sector);
3357 return DM_MAPIO_REMAPPED;
3361 * Check if bio is too large, split as needed.
3363 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_PAGES << PAGE_SHIFT)) &&
3364 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3365 dm_accept_partial_bio(bio, ((BIO_MAX_PAGES << PAGE_SHIFT) >> SECTOR_SHIFT));
3368 * Ensure that bio is a multiple of internal sector encryption size
3369 * and is aligned to this size as defined in IO hints.
3371 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3372 return DM_MAPIO_KILL;
3374 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3375 return DM_MAPIO_KILL;
3377 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3378 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3380 if (cc->on_disk_tag_size) {
3381 unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3383 if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3384 unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3385 GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3386 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3387 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3388 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3389 io->integrity_metadata_from_pool = true;
3393 if (crypt_integrity_aead(cc))
3394 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3396 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3398 if (bio_data_dir(io->base_bio) == READ) {
3399 if (kcryptd_io_read(io, GFP_NOWAIT))
3400 kcryptd_queue_read(io);
3402 kcryptd_queue_crypt(io);
3404 return DM_MAPIO_SUBMITTED;
3407 static char hex2asc(unsigned char c)
3409 return c + '0' + ((unsigned)(9 - c) >> 4 & 0x27);
3412 static void crypt_status(struct dm_target *ti, status_type_t type,
3413 unsigned status_flags, char *result, unsigned maxlen)
3415 struct crypt_config *cc = ti->private;
3417 int num_feature_args = 0;
3420 case STATUSTYPE_INFO:
3424 case STATUSTYPE_TABLE:
3425 DMEMIT("%s ", cc->cipher_string);
3427 if (cc->key_size > 0) {
3429 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3431 for (i = 0; i < cc->key_size; i++) {
3432 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3433 hex2asc(cc->key[i] & 0xf));
3439 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3440 cc->dev->name, (unsigned long long)cc->start);
3442 num_feature_args += !!ti->num_discard_bios;
3443 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3444 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3445 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3446 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3447 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3448 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3449 if (cc->on_disk_tag_size)
3451 if (num_feature_args) {
3452 DMEMIT(" %d", num_feature_args);
3453 if (ti->num_discard_bios)
3454 DMEMIT(" allow_discards");
3455 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3456 DMEMIT(" same_cpu_crypt");
3457 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3458 DMEMIT(" submit_from_crypt_cpus");
3459 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3460 DMEMIT(" no_read_workqueue");
3461 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3462 DMEMIT(" no_write_workqueue");
3463 if (cc->on_disk_tag_size)
3464 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3465 if (cc->sector_size != (1 << SECTOR_SHIFT))
3466 DMEMIT(" sector_size:%d", cc->sector_size);
3467 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3468 DMEMIT(" iv_large_sectors");
3475 static void crypt_postsuspend(struct dm_target *ti)
3477 struct crypt_config *cc = ti->private;
3479 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3482 static int crypt_preresume(struct dm_target *ti)
3484 struct crypt_config *cc = ti->private;
3486 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3487 DMERR("aborting resume - crypt key is not set.");
3494 static void crypt_resume(struct dm_target *ti)
3496 struct crypt_config *cc = ti->private;
3498 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3501 /* Message interface
3505 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3506 char *result, unsigned maxlen)
3508 struct crypt_config *cc = ti->private;
3509 int key_size, ret = -EINVAL;
3514 if (!strcasecmp(argv[0], "key")) {
3515 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3516 DMWARN("not suspended during key manipulation.");
3519 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3520 /* The key size may not be changed. */
3521 key_size = get_key_size(&argv[2]);
3522 if (key_size < 0 || cc->key_size != key_size) {
3523 memset(argv[2], '0', strlen(argv[2]));
3527 ret = crypt_set_key(cc, argv[2]);
3530 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3531 ret = cc->iv_gen_ops->init(cc);
3532 /* wipe the kernel key payload copy */
3534 memset(cc->key, 0, cc->key_size * sizeof(u8));
3537 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3538 return crypt_wipe_key(cc);
3542 DMWARN("unrecognised message received.");
3546 static int crypt_iterate_devices(struct dm_target *ti,
3547 iterate_devices_callout_fn fn, void *data)
3549 struct crypt_config *cc = ti->private;
3551 return fn(ti, cc->dev, cc->start, ti->len, data);
3554 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3556 struct crypt_config *cc = ti->private;
3559 * Unfortunate constraint that is required to avoid the potential
3560 * for exceeding underlying device's max_segments limits -- due to
3561 * crypt_alloc_buffer() possibly allocating pages for the encryption
3562 * bio that are not as physically contiguous as the original bio.
3564 limits->max_segment_size = PAGE_SIZE;
3566 limits->logical_block_size =
3567 max_t(unsigned, limits->logical_block_size, cc->sector_size);
3568 limits->physical_block_size =
3569 max_t(unsigned, limits->physical_block_size, cc->sector_size);
3570 limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3573 static struct target_type crypt_target = {
3575 .version = {1, 22, 0},
3576 .module = THIS_MODULE,
3579 #ifdef CONFIG_BLK_DEV_ZONED
3580 .features = DM_TARGET_ZONED_HM,
3581 .report_zones = crypt_report_zones,
3584 .status = crypt_status,
3585 .postsuspend = crypt_postsuspend,
3586 .preresume = crypt_preresume,
3587 .resume = crypt_resume,
3588 .message = crypt_message,
3589 .iterate_devices = crypt_iterate_devices,
3590 .io_hints = crypt_io_hints,
3593 static int __init dm_crypt_init(void)
3597 r = dm_register_target(&crypt_target);
3599 DMERR("register failed %d", r);
3604 static void __exit dm_crypt_exit(void)
3606 dm_unregister_target(&crypt_target);
3609 module_init(dm_crypt_init);
3610 module_exit(dm_crypt_exit);
3612 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3613 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3614 MODULE_LICENSE("GPL");