1 // SPDX-License-Identifier: GPL-2.0-only
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>
8 * This file is released under the GPL.
11 #include <linux/completion.h>
12 #include <linux/err.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/key.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-integrity.h>
20 #include <linux/mempool.h>
21 #include <linux/slab.h>
22 #include <linux/crypto.h>
23 #include <linux/workqueue.h>
24 #include <linux/kthread.h>
25 #include <linux/backing-dev.h>
26 #include <linux/atomic.h>
27 #include <linux/scatterlist.h>
28 #include <linux/rbtree.h>
29 #include <linux/ctype.h>
31 #include <asm/unaligned.h>
32 #include <crypto/hash.h>
33 #include <crypto/md5.h>
34 #include <crypto/algapi.h>
35 #include <crypto/skcipher.h>
36 #include <crypto/aead.h>
37 #include <crypto/authenc.h>
38 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
39 #include <linux/key-type.h>
40 #include <keys/user-type.h>
41 #include <keys/encrypted-type.h>
42 #include <keys/trusted-type.h>
44 #include <linux/device-mapper.h>
48 #define DM_MSG_PREFIX "crypt"
51 * context holding the current state of a multi-part conversion
53 struct convert_context {
54 struct completion restart;
57 struct bvec_iter iter_in;
58 struct bvec_iter iter_out;
62 struct skcipher_request *req;
63 struct aead_request *req_aead;
69 * per bio private data
72 struct crypt_config *cc;
74 u8 *integrity_metadata;
75 bool integrity_metadata_from_pool;
76 struct work_struct work;
77 struct tasklet_struct tasklet;
79 struct convert_context ctx;
85 struct rb_node rb_node;
86 } CRYPTO_MINALIGN_ATTR;
88 struct dm_crypt_request {
89 struct convert_context *ctx;
90 struct scatterlist sg_in[4];
91 struct scatterlist sg_out[4];
97 struct crypt_iv_operations {
98 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
100 void (*dtr)(struct crypt_config *cc);
101 int (*init)(struct crypt_config *cc);
102 int (*wipe)(struct crypt_config *cc);
103 int (*generator)(struct crypt_config *cc, u8 *iv,
104 struct dm_crypt_request *dmreq);
105 int (*post)(struct crypt_config *cc, u8 *iv,
106 struct dm_crypt_request *dmreq);
109 struct iv_benbi_private {
113 #define LMK_SEED_SIZE 64 /* hash + 0 */
114 struct iv_lmk_private {
115 struct crypto_shash *hash_tfm;
119 #define TCW_WHITENING_SIZE 16
120 struct iv_tcw_private {
121 struct crypto_shash *crc32_tfm;
126 #define ELEPHANT_MAX_KEY_SIZE 32
127 struct iv_elephant_private {
128 struct crypto_skcipher *tfm;
132 * Crypt: maps a linear range of a block device
133 * and encrypts / decrypts at the same time.
135 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
136 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
137 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
138 DM_CRYPT_WRITE_INLINE };
141 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
142 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
143 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
147 * The fields in here must be read only after initialization.
149 struct crypt_config {
153 struct percpu_counter n_allocated_pages;
155 struct workqueue_struct *io_queue;
156 struct workqueue_struct *crypt_queue;
158 spinlock_t write_thread_lock;
159 struct task_struct *write_thread;
160 struct rb_root write_tree;
166 const struct crypt_iv_operations *iv_gen_ops;
168 struct iv_benbi_private benbi;
169 struct iv_lmk_private lmk;
170 struct iv_tcw_private tcw;
171 struct iv_elephant_private elephant;
174 unsigned int iv_size;
175 unsigned short int sector_size;
176 unsigned char sector_shift;
179 struct crypto_skcipher **tfms;
180 struct crypto_aead **tfms_aead;
183 unsigned long cipher_flags;
186 * Layout of each crypto request:
188 * struct skcipher_request
191 * struct dm_crypt_request
195 * The padding is added so that dm_crypt_request and the IV are
198 unsigned int dmreq_start;
200 unsigned int per_bio_data_size;
203 unsigned int key_size;
204 unsigned int key_parts; /* independent parts in key buffer */
205 unsigned int key_extra_size; /* additional keys length */
206 unsigned int key_mac_size; /* MAC key size for authenc(...) */
208 unsigned int integrity_tag_size;
209 unsigned int integrity_iv_size;
210 unsigned int on_disk_tag_size;
213 * pool for per bio private data, crypto requests,
214 * encryption requeusts/buffer pages and integrity tags
216 unsigned tag_pool_max_sectors;
222 struct mutex bio_alloc_lock;
224 u8 *authenc_key; /* space for keys in authenc() format (if used) */
229 #define MAX_TAG_SIZE 480
230 #define POOL_ENTRY_SIZE 512
232 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
233 static unsigned dm_crypt_clients_n = 0;
234 static volatile unsigned long dm_crypt_pages_per_client;
235 #define DM_CRYPT_MEMORY_PERCENT 2
236 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
238 static void crypt_endio(struct bio *clone);
239 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
240 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
241 struct scatterlist *sg);
243 static bool crypt_integrity_aead(struct crypt_config *cc);
246 * Use this to access cipher attributes that are independent of the key.
248 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
250 return cc->cipher_tfm.tfms[0];
253 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
255 return cc->cipher_tfm.tfms_aead[0];
259 * Different IV generation algorithms:
261 * plain: the initial vector is the 32-bit little-endian version of the sector
262 * number, padded with zeros if necessary.
264 * plain64: the initial vector is the 64-bit little-endian version of the sector
265 * number, padded with zeros if necessary.
267 * plain64be: the initial vector is the 64-bit big-endian version of the sector
268 * number, padded with zeros if necessary.
270 * essiv: "encrypted sector|salt initial vector", the sector number is
271 * encrypted with the bulk cipher using a salt as key. The salt
272 * should be derived from the bulk cipher's key via hashing.
274 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
275 * (needed for LRW-32-AES and possible other narrow block modes)
277 * null: the initial vector is always zero. Provides compatibility with
278 * obsolete loop_fish2 devices. Do not use for new devices.
280 * lmk: Compatible implementation of the block chaining mode used
281 * by the Loop-AES block device encryption system
282 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
283 * It operates on full 512 byte sectors and uses CBC
284 * with an IV derived from the sector number, the data and
285 * optionally extra IV seed.
286 * This means that after decryption the first block
287 * of sector must be tweaked according to decrypted data.
288 * Loop-AES can use three encryption schemes:
289 * version 1: is plain aes-cbc mode
290 * version 2: uses 64 multikey scheme with lmk IV generator
291 * version 3: the same as version 2 with additional IV seed
292 * (it uses 65 keys, last key is used as IV seed)
294 * tcw: Compatible implementation of the block chaining mode used
295 * by the TrueCrypt device encryption system (prior to version 4.1).
296 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
297 * It operates on full 512 byte sectors and uses CBC
298 * with an IV derived from initial key and the sector number.
299 * In addition, whitening value is applied on every sector, whitening
300 * is calculated from initial key, sector number and mixed using CRC32.
301 * Note that this encryption scheme is vulnerable to watermarking attacks
302 * and should be used for old compatible containers access only.
304 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
305 * The IV is encrypted little-endian byte-offset (with the same key
306 * and cipher as the volume).
308 * elephant: The extended version of eboiv with additional Elephant diffuser
309 * used with Bitlocker CBC mode.
310 * This mode was used in older Windows systems
311 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
314 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
315 struct dm_crypt_request *dmreq)
317 memset(iv, 0, cc->iv_size);
318 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
323 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
324 struct dm_crypt_request *dmreq)
326 memset(iv, 0, cc->iv_size);
327 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
332 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
333 struct dm_crypt_request *dmreq)
335 memset(iv, 0, cc->iv_size);
336 /* iv_size is at least of size u64; usually it is 16 bytes */
337 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
342 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
343 struct dm_crypt_request *dmreq)
346 * ESSIV encryption of the IV is now handled by the crypto API,
347 * so just pass the plain sector number here.
349 memset(iv, 0, cc->iv_size);
350 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
355 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
361 if (crypt_integrity_aead(cc))
362 bs = crypto_aead_blocksize(any_tfm_aead(cc));
364 bs = crypto_skcipher_blocksize(any_tfm(cc));
367 /* we need to calculate how far we must shift the sector count
368 * to get the cipher block count, we use this shift in _gen */
370 if (1 << log != bs) {
371 ti->error = "cypher blocksize is not a power of 2";
376 ti->error = "cypher blocksize is > 512";
380 cc->iv_gen_private.benbi.shift = 9 - log;
385 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
389 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
390 struct dm_crypt_request *dmreq)
394 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
396 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
397 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
402 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
403 struct dm_crypt_request *dmreq)
405 memset(iv, 0, cc->iv_size);
410 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
412 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
414 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
415 crypto_free_shash(lmk->hash_tfm);
416 lmk->hash_tfm = NULL;
418 kfree_sensitive(lmk->seed);
422 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
425 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
427 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
428 ti->error = "Unsupported sector size for LMK";
432 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
433 CRYPTO_ALG_ALLOCATES_MEMORY);
434 if (IS_ERR(lmk->hash_tfm)) {
435 ti->error = "Error initializing LMK hash";
436 return PTR_ERR(lmk->hash_tfm);
439 /* No seed in LMK version 2 */
440 if (cc->key_parts == cc->tfms_count) {
445 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
447 crypt_iv_lmk_dtr(cc);
448 ti->error = "Error kmallocing seed storage in LMK";
455 static int crypt_iv_lmk_init(struct crypt_config *cc)
457 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
458 int subkey_size = cc->key_size / cc->key_parts;
460 /* LMK seed is on the position of LMK_KEYS + 1 key */
462 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
463 crypto_shash_digestsize(lmk->hash_tfm));
468 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
470 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
473 memset(lmk->seed, 0, LMK_SEED_SIZE);
478 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
479 struct dm_crypt_request *dmreq,
482 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
483 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
484 struct md5_state md5state;
488 desc->tfm = lmk->hash_tfm;
490 r = crypto_shash_init(desc);
495 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
500 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
501 r = crypto_shash_update(desc, data + 16, 16 * 31);
505 /* Sector is cropped to 56 bits here */
506 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
507 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
508 buf[2] = cpu_to_le32(4024);
510 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
514 /* No MD5 padding here */
515 r = crypto_shash_export(desc, &md5state);
519 for (i = 0; i < MD5_HASH_WORDS; i++)
520 __cpu_to_le32s(&md5state.hash[i]);
521 memcpy(iv, &md5state.hash, cc->iv_size);
526 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
527 struct dm_crypt_request *dmreq)
529 struct scatterlist *sg;
533 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
534 sg = crypt_get_sg_data(cc, dmreq->sg_in);
535 src = kmap_atomic(sg_page(sg));
536 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
539 memset(iv, 0, cc->iv_size);
544 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
545 struct dm_crypt_request *dmreq)
547 struct scatterlist *sg;
551 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
554 sg = crypt_get_sg_data(cc, dmreq->sg_out);
555 dst = kmap_atomic(sg_page(sg));
556 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
558 /* Tweak the first block of plaintext sector */
560 crypto_xor(dst + sg->offset, iv, cc->iv_size);
566 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
568 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
570 kfree_sensitive(tcw->iv_seed);
572 kfree_sensitive(tcw->whitening);
573 tcw->whitening = NULL;
575 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
576 crypto_free_shash(tcw->crc32_tfm);
577 tcw->crc32_tfm = NULL;
580 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
583 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
585 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
586 ti->error = "Unsupported sector size for TCW";
590 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
591 ti->error = "Wrong key size for TCW";
595 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
596 CRYPTO_ALG_ALLOCATES_MEMORY);
597 if (IS_ERR(tcw->crc32_tfm)) {
598 ti->error = "Error initializing CRC32 in TCW";
599 return PTR_ERR(tcw->crc32_tfm);
602 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
603 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
604 if (!tcw->iv_seed || !tcw->whitening) {
605 crypt_iv_tcw_dtr(cc);
606 ti->error = "Error allocating seed storage in TCW";
613 static int crypt_iv_tcw_init(struct crypt_config *cc)
615 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
616 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
618 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
619 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
625 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
627 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
629 memset(tcw->iv_seed, 0, cc->iv_size);
630 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
635 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
636 struct dm_crypt_request *dmreq,
639 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
640 __le64 sector = cpu_to_le64(dmreq->iv_sector);
641 u8 buf[TCW_WHITENING_SIZE];
642 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
645 /* xor whitening with sector number */
646 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
647 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
649 /* calculate crc32 for every 32bit part and xor it */
650 desc->tfm = tcw->crc32_tfm;
651 for (i = 0; i < 4; i++) {
652 r = crypto_shash_init(desc);
655 r = crypto_shash_update(desc, &buf[i * 4], 4);
658 r = crypto_shash_final(desc, &buf[i * 4]);
662 crypto_xor(&buf[0], &buf[12], 4);
663 crypto_xor(&buf[4], &buf[8], 4);
665 /* apply whitening (8 bytes) to whole sector */
666 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
667 crypto_xor(data + i * 8, buf, 8);
669 memzero_explicit(buf, sizeof(buf));
673 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
674 struct dm_crypt_request *dmreq)
676 struct scatterlist *sg;
677 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
678 __le64 sector = cpu_to_le64(dmreq->iv_sector);
682 /* Remove whitening from ciphertext */
683 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
684 sg = crypt_get_sg_data(cc, dmreq->sg_in);
685 src = kmap_atomic(sg_page(sg));
686 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
691 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
693 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
699 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
700 struct dm_crypt_request *dmreq)
702 struct scatterlist *sg;
706 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
709 /* Apply whitening on ciphertext */
710 sg = crypt_get_sg_data(cc, dmreq->sg_out);
711 dst = kmap_atomic(sg_page(sg));
712 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
718 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
719 struct dm_crypt_request *dmreq)
721 /* Used only for writes, there must be an additional space to store IV */
722 get_random_bytes(iv, cc->iv_size);
726 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
729 if (crypt_integrity_aead(cc)) {
730 ti->error = "AEAD transforms not supported for EBOIV";
734 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
735 ti->error = "Block size of EBOIV cipher does "
736 "not match IV size of block cipher";
743 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
744 struct dm_crypt_request *dmreq)
746 u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
747 struct skcipher_request *req;
748 struct scatterlist src, dst;
749 DECLARE_CRYPTO_WAIT(wait);
752 req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
756 memset(buf, 0, cc->iv_size);
757 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
759 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
760 sg_init_one(&dst, iv, cc->iv_size);
761 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
762 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
763 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
764 skcipher_request_free(req);
769 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
771 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
773 crypto_free_skcipher(elephant->tfm);
774 elephant->tfm = NULL;
777 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
780 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
783 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
784 CRYPTO_ALG_ALLOCATES_MEMORY);
785 if (IS_ERR(elephant->tfm)) {
786 r = PTR_ERR(elephant->tfm);
787 elephant->tfm = NULL;
791 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
793 crypt_iv_elephant_dtr(cc);
797 static void diffuser_disk_to_cpu(u32 *d, size_t n)
799 #ifndef __LITTLE_ENDIAN
802 for (i = 0; i < n; i++)
803 d[i] = le32_to_cpu((__le32)d[i]);
807 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
809 #ifndef __LITTLE_ENDIAN
812 for (i = 0; i < n; i++)
813 d[i] = cpu_to_le32((u32)d[i]);
817 static void diffuser_a_decrypt(u32 *d, size_t n)
821 for (i = 0; i < 5; i++) {
826 while (i1 < (n - 1)) {
827 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
833 d[i1] += d[i2] ^ d[i3];
839 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
842 d[i1] += d[i2] ^ d[i3];
848 static void diffuser_a_encrypt(u32 *d, size_t n)
852 for (i = 0; i < 5; i++) {
858 d[i1] -= d[i2] ^ d[i3];
861 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
867 d[i1] -= d[i2] ^ d[i3];
873 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
879 static void diffuser_b_decrypt(u32 *d, size_t n)
883 for (i = 0; i < 3; i++) {
888 while (i1 < (n - 1)) {
889 d[i1] += d[i2] ^ d[i3];
892 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
898 d[i1] += d[i2] ^ d[i3];
904 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
910 static void diffuser_b_encrypt(u32 *d, size_t n)
914 for (i = 0; i < 3; i++) {
920 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
926 d[i1] -= d[i2] ^ d[i3];
932 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
935 d[i1] -= d[i2] ^ d[i3];
941 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
943 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
944 u8 *es, *ks, *data, *data2, *data_offset;
945 struct skcipher_request *req;
946 struct scatterlist *sg, *sg2, src, dst;
947 DECLARE_CRYPTO_WAIT(wait);
950 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
951 es = kzalloc(16, GFP_NOIO); /* Key for AES */
952 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
954 if (!req || !es || !ks) {
959 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
962 sg_init_one(&src, es, 16);
963 sg_init_one(&dst, ks, 16);
964 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
965 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
966 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
972 sg_init_one(&dst, &ks[16], 16);
973 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
977 sg = crypt_get_sg_data(cc, dmreq->sg_out);
978 data = kmap_atomic(sg_page(sg));
979 data_offset = data + sg->offset;
981 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
982 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
983 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
984 data2 = kmap_atomic(sg_page(sg2));
985 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
986 kunmap_atomic(data2);
989 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
990 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
991 diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
992 diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
993 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
996 for (i = 0; i < (cc->sector_size / 32); i++)
997 crypto_xor(data_offset + i * 32, ks, 32);
999 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1000 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
1001 diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1002 diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1003 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1006 kunmap_atomic(data);
1008 kfree_sensitive(ks);
1009 kfree_sensitive(es);
1010 skcipher_request_free(req);
1014 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1015 struct dm_crypt_request *dmreq)
1019 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1020 r = crypt_iv_elephant(cc, dmreq);
1025 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1028 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1029 struct dm_crypt_request *dmreq)
1031 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1032 return crypt_iv_elephant(cc, dmreq);
1037 static int crypt_iv_elephant_init(struct crypt_config *cc)
1039 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1040 int key_offset = cc->key_size - cc->key_extra_size;
1042 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1045 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1047 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1048 u8 key[ELEPHANT_MAX_KEY_SIZE];
1050 memset(key, 0, cc->key_extra_size);
1051 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1054 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1055 .generator = crypt_iv_plain_gen
1058 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1059 .generator = crypt_iv_plain64_gen
1062 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1063 .generator = crypt_iv_plain64be_gen
1066 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1067 .generator = crypt_iv_essiv_gen
1070 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1071 .ctr = crypt_iv_benbi_ctr,
1072 .dtr = crypt_iv_benbi_dtr,
1073 .generator = crypt_iv_benbi_gen
1076 static const struct crypt_iv_operations crypt_iv_null_ops = {
1077 .generator = crypt_iv_null_gen
1080 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1081 .ctr = crypt_iv_lmk_ctr,
1082 .dtr = crypt_iv_lmk_dtr,
1083 .init = crypt_iv_lmk_init,
1084 .wipe = crypt_iv_lmk_wipe,
1085 .generator = crypt_iv_lmk_gen,
1086 .post = crypt_iv_lmk_post
1089 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1090 .ctr = crypt_iv_tcw_ctr,
1091 .dtr = crypt_iv_tcw_dtr,
1092 .init = crypt_iv_tcw_init,
1093 .wipe = crypt_iv_tcw_wipe,
1094 .generator = crypt_iv_tcw_gen,
1095 .post = crypt_iv_tcw_post
1098 static const struct crypt_iv_operations crypt_iv_random_ops = {
1099 .generator = crypt_iv_random_gen
1102 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1103 .ctr = crypt_iv_eboiv_ctr,
1104 .generator = crypt_iv_eboiv_gen
1107 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1108 .ctr = crypt_iv_elephant_ctr,
1109 .dtr = crypt_iv_elephant_dtr,
1110 .init = crypt_iv_elephant_init,
1111 .wipe = crypt_iv_elephant_wipe,
1112 .generator = crypt_iv_elephant_gen,
1113 .post = crypt_iv_elephant_post
1117 * Integrity extensions
1119 static bool crypt_integrity_aead(struct crypt_config *cc)
1121 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1124 static bool crypt_integrity_hmac(struct crypt_config *cc)
1126 return crypt_integrity_aead(cc) && cc->key_mac_size;
1129 /* Get sg containing data */
1130 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1131 struct scatterlist *sg)
1133 if (unlikely(crypt_integrity_aead(cc)))
1139 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1141 struct bio_integrity_payload *bip;
1142 unsigned int tag_len;
1145 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1148 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1150 return PTR_ERR(bip);
1152 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1154 bip->bip_iter.bi_size = tag_len;
1155 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1157 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1158 tag_len, offset_in_page(io->integrity_metadata));
1159 if (unlikely(ret != tag_len))
1165 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1167 #ifdef CONFIG_BLK_DEV_INTEGRITY
1168 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1169 struct mapped_device *md = dm_table_get_md(ti->table);
1171 /* From now we require underlying device with our integrity profile */
1172 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1173 ti->error = "Integrity profile not supported.";
1177 if (bi->tag_size != cc->on_disk_tag_size ||
1178 bi->tuple_size != cc->on_disk_tag_size) {
1179 ti->error = "Integrity profile tag size mismatch.";
1182 if (1 << bi->interval_exp != cc->sector_size) {
1183 ti->error = "Integrity profile sector size mismatch.";
1187 if (crypt_integrity_aead(cc)) {
1188 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1189 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1190 cc->integrity_tag_size, cc->integrity_iv_size);
1192 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1193 ti->error = "Integrity AEAD auth tag size is not supported.";
1196 } else if (cc->integrity_iv_size)
1197 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1198 cc->integrity_iv_size);
1200 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1201 ti->error = "Not enough space for integrity tag in the profile.";
1207 ti->error = "Integrity profile not supported.";
1212 static void crypt_convert_init(struct crypt_config *cc,
1213 struct convert_context *ctx,
1214 struct bio *bio_out, struct bio *bio_in,
1217 ctx->bio_in = bio_in;
1218 ctx->bio_out = bio_out;
1220 ctx->iter_in = bio_in->bi_iter;
1222 ctx->iter_out = bio_out->bi_iter;
1223 ctx->cc_sector = sector + cc->iv_offset;
1224 init_completion(&ctx->restart);
1227 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1230 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1233 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1235 return (void *)((char *)dmreq - cc->dmreq_start);
1238 static u8 *iv_of_dmreq(struct crypt_config *cc,
1239 struct dm_crypt_request *dmreq)
1241 if (crypt_integrity_aead(cc))
1242 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1243 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1245 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1246 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1249 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1250 struct dm_crypt_request *dmreq)
1252 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1255 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1256 struct dm_crypt_request *dmreq)
1258 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1259 return (__le64 *) ptr;
1262 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1263 struct dm_crypt_request *dmreq)
1265 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1266 cc->iv_size + sizeof(uint64_t);
1267 return (unsigned int*)ptr;
1270 static void *tag_from_dmreq(struct crypt_config *cc,
1271 struct dm_crypt_request *dmreq)
1273 struct convert_context *ctx = dmreq->ctx;
1274 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1276 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1277 cc->on_disk_tag_size];
1280 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1281 struct dm_crypt_request *dmreq)
1283 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1286 static int crypt_convert_block_aead(struct crypt_config *cc,
1287 struct convert_context *ctx,
1288 struct aead_request *req,
1289 unsigned int tag_offset)
1291 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1292 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1293 struct dm_crypt_request *dmreq;
1294 u8 *iv, *org_iv, *tag_iv, *tag;
1298 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1300 /* Reject unexpected unaligned bio. */
1301 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1304 dmreq = dmreq_of_req(cc, req);
1305 dmreq->iv_sector = ctx->cc_sector;
1306 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1307 dmreq->iv_sector >>= cc->sector_shift;
1310 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1312 sector = org_sector_of_dmreq(cc, dmreq);
1313 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1315 iv = iv_of_dmreq(cc, dmreq);
1316 org_iv = org_iv_of_dmreq(cc, dmreq);
1317 tag = tag_from_dmreq(cc, dmreq);
1318 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1321 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1322 * | (authenticated) | (auth+encryption) | |
1323 * | sector_LE | IV | sector in/out | tag in/out |
1325 sg_init_table(dmreq->sg_in, 4);
1326 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1327 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1328 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1329 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1331 sg_init_table(dmreq->sg_out, 4);
1332 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1333 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1334 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1335 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1337 if (cc->iv_gen_ops) {
1338 /* For READs use IV stored in integrity metadata */
1339 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1340 memcpy(org_iv, tag_iv, cc->iv_size);
1342 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1345 /* Store generated IV in integrity metadata */
1346 if (cc->integrity_iv_size)
1347 memcpy(tag_iv, org_iv, cc->iv_size);
1349 /* Working copy of IV, to be modified in crypto API */
1350 memcpy(iv, org_iv, cc->iv_size);
1353 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1354 if (bio_data_dir(ctx->bio_in) == WRITE) {
1355 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1356 cc->sector_size, iv);
1357 r = crypto_aead_encrypt(req);
1358 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1359 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1360 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1362 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1363 cc->sector_size + cc->integrity_tag_size, iv);
1364 r = crypto_aead_decrypt(req);
1367 if (r == -EBADMSG) {
1368 sector_t s = le64_to_cpu(*sector);
1370 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1371 ctx->bio_in->bi_bdev, s);
1372 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1376 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1377 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1379 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1380 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1385 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1386 struct convert_context *ctx,
1387 struct skcipher_request *req,
1388 unsigned int tag_offset)
1390 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1391 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1392 struct scatterlist *sg_in, *sg_out;
1393 struct dm_crypt_request *dmreq;
1394 u8 *iv, *org_iv, *tag_iv;
1398 /* Reject unexpected unaligned bio. */
1399 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1402 dmreq = dmreq_of_req(cc, req);
1403 dmreq->iv_sector = ctx->cc_sector;
1404 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1405 dmreq->iv_sector >>= cc->sector_shift;
1408 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1410 iv = iv_of_dmreq(cc, dmreq);
1411 org_iv = org_iv_of_dmreq(cc, dmreq);
1412 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1414 sector = org_sector_of_dmreq(cc, dmreq);
1415 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1417 /* For skcipher we use only the first sg item */
1418 sg_in = &dmreq->sg_in[0];
1419 sg_out = &dmreq->sg_out[0];
1421 sg_init_table(sg_in, 1);
1422 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1424 sg_init_table(sg_out, 1);
1425 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1427 if (cc->iv_gen_ops) {
1428 /* For READs use IV stored in integrity metadata */
1429 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1430 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1432 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1435 /* Data can be already preprocessed in generator */
1436 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1438 /* Store generated IV in integrity metadata */
1439 if (cc->integrity_iv_size)
1440 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1442 /* Working copy of IV, to be modified in crypto API */
1443 memcpy(iv, org_iv, cc->iv_size);
1446 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1448 if (bio_data_dir(ctx->bio_in) == WRITE)
1449 r = crypto_skcipher_encrypt(req);
1451 r = crypto_skcipher_decrypt(req);
1453 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1454 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1456 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1457 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1462 static void kcryptd_async_done(struct crypto_async_request *async_req,
1465 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1466 struct convert_context *ctx)
1468 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1471 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1476 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1479 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1480 * requests if driver request queue is full.
1482 skcipher_request_set_callback(ctx->r.req,
1483 CRYPTO_TFM_REQ_MAY_BACKLOG,
1484 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1489 static int crypt_alloc_req_aead(struct crypt_config *cc,
1490 struct convert_context *ctx)
1492 if (!ctx->r.req_aead) {
1493 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1494 if (!ctx->r.req_aead)
1498 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1501 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1502 * requests if driver request queue is full.
1504 aead_request_set_callback(ctx->r.req_aead,
1505 CRYPTO_TFM_REQ_MAY_BACKLOG,
1506 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1511 static int crypt_alloc_req(struct crypt_config *cc,
1512 struct convert_context *ctx)
1514 if (crypt_integrity_aead(cc))
1515 return crypt_alloc_req_aead(cc, ctx);
1517 return crypt_alloc_req_skcipher(cc, ctx);
1520 static void crypt_free_req_skcipher(struct crypt_config *cc,
1521 struct skcipher_request *req, struct bio *base_bio)
1523 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1525 if ((struct skcipher_request *)(io + 1) != req)
1526 mempool_free(req, &cc->req_pool);
1529 static void crypt_free_req_aead(struct crypt_config *cc,
1530 struct aead_request *req, struct bio *base_bio)
1532 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1534 if ((struct aead_request *)(io + 1) != req)
1535 mempool_free(req, &cc->req_pool);
1538 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1540 if (crypt_integrity_aead(cc))
1541 crypt_free_req_aead(cc, req, base_bio);
1543 crypt_free_req_skcipher(cc, req, base_bio);
1547 * Encrypt / decrypt data from one bio to another one (can be the same one)
1549 static blk_status_t crypt_convert(struct crypt_config *cc,
1550 struct convert_context *ctx, bool atomic, bool reset_pending)
1552 unsigned int tag_offset = 0;
1553 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1557 * if reset_pending is set we are dealing with the bio for the first time,
1558 * else we're continuing to work on the previous bio, so don't mess with
1559 * the cc_pending counter
1562 atomic_set(&ctx->cc_pending, 1);
1564 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1566 r = crypt_alloc_req(cc, ctx);
1568 complete(&ctx->restart);
1569 return BLK_STS_DEV_RESOURCE;
1572 atomic_inc(&ctx->cc_pending);
1574 if (crypt_integrity_aead(cc))
1575 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1577 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1581 * The request was queued by a crypto driver
1582 * but the driver request queue is full, let's wait.
1585 if (in_interrupt()) {
1586 if (try_wait_for_completion(&ctx->restart)) {
1588 * we don't have to block to wait for completion,
1593 * we can't wait for completion without blocking
1594 * exit and continue processing in a workqueue
1597 ctx->cc_sector += sector_step;
1599 return BLK_STS_DEV_RESOURCE;
1602 wait_for_completion(&ctx->restart);
1604 reinit_completion(&ctx->restart);
1607 * The request is queued and processed asynchronously,
1608 * completion function kcryptd_async_done() will be called.
1612 ctx->cc_sector += sector_step;
1616 * The request was already processed (synchronously).
1619 atomic_dec(&ctx->cc_pending);
1620 ctx->cc_sector += sector_step;
1626 * There was a data integrity error.
1629 atomic_dec(&ctx->cc_pending);
1630 return BLK_STS_PROTECTION;
1632 * There was an error while processing the request.
1635 atomic_dec(&ctx->cc_pending);
1636 return BLK_STS_IOERR;
1643 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1646 * Generate a new unfragmented bio with the given size
1647 * This should never violate the device limitations (but only because
1648 * max_segment_size is being constrained to PAGE_SIZE).
1650 * This function may be called concurrently. If we allocate from the mempool
1651 * concurrently, there is a possibility of deadlock. For example, if we have
1652 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1653 * the mempool concurrently, it may deadlock in a situation where both processes
1654 * have allocated 128 pages and the mempool is exhausted.
1656 * In order to avoid this scenario we allocate the pages under a mutex.
1658 * In order to not degrade performance with excessive locking, we try
1659 * non-blocking allocations without a mutex first but on failure we fallback
1660 * to blocking allocations with a mutex.
1662 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1664 struct crypt_config *cc = io->cc;
1666 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1667 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1668 unsigned i, len, remaining_size;
1672 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1673 mutex_lock(&cc->bio_alloc_lock);
1675 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1677 clone->bi_private = io;
1678 clone->bi_end_io = crypt_endio;
1680 remaining_size = size;
1682 for (i = 0; i < nr_iovecs; i++) {
1683 page = mempool_alloc(&cc->page_pool, gfp_mask);
1685 crypt_free_buffer_pages(cc, clone);
1687 gfp_mask |= __GFP_DIRECT_RECLAIM;
1691 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1693 bio_add_page(clone, page, len, 0);
1695 remaining_size -= len;
1698 /* Allocate space for integrity tags */
1699 if (dm_crypt_integrity_io_alloc(io, clone)) {
1700 crypt_free_buffer_pages(cc, clone);
1705 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1706 mutex_unlock(&cc->bio_alloc_lock);
1711 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1714 struct bvec_iter_all iter_all;
1716 bio_for_each_segment_all(bv, clone, iter_all) {
1717 BUG_ON(!bv->bv_page);
1718 mempool_free(bv->bv_page, &cc->page_pool);
1722 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1723 struct bio *bio, sector_t sector)
1727 io->sector = sector;
1729 io->ctx.r.req = NULL;
1730 io->integrity_metadata = NULL;
1731 io->integrity_metadata_from_pool = false;
1732 atomic_set(&io->io_pending, 0);
1735 static void crypt_inc_pending(struct dm_crypt_io *io)
1737 atomic_inc(&io->io_pending);
1740 static void kcryptd_io_bio_endio(struct work_struct *work)
1742 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1743 bio_endio(io->base_bio);
1747 * One of the bios was finished. Check for completion of
1748 * the whole request and correctly clean up the buffer.
1750 static void crypt_dec_pending(struct dm_crypt_io *io)
1752 struct crypt_config *cc = io->cc;
1753 struct bio *base_bio = io->base_bio;
1754 blk_status_t error = io->error;
1756 if (!atomic_dec_and_test(&io->io_pending))
1760 crypt_free_req(cc, io->ctx.r.req, base_bio);
1762 if (unlikely(io->integrity_metadata_from_pool))
1763 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1765 kfree(io->integrity_metadata);
1767 base_bio->bi_status = error;
1770 * If we are running this function from our tasklet,
1771 * we can't call bio_endio() here, because it will call
1772 * clone_endio() from dm.c, which in turn will
1773 * free the current struct dm_crypt_io structure with
1774 * our tasklet. In this case we need to delay bio_endio()
1775 * execution to after the tasklet is done and dequeued.
1777 if (tasklet_trylock(&io->tasklet)) {
1778 tasklet_unlock(&io->tasklet);
1779 bio_endio(base_bio);
1783 INIT_WORK(&io->work, kcryptd_io_bio_endio);
1784 queue_work(cc->io_queue, &io->work);
1788 * kcryptd/kcryptd_io:
1790 * Needed because it would be very unwise to do decryption in an
1791 * interrupt context.
1793 * kcryptd performs the actual encryption or decryption.
1795 * kcryptd_io performs the IO submission.
1797 * They must be separated as otherwise the final stages could be
1798 * starved by new requests which can block in the first stages due
1799 * to memory allocation.
1801 * The work is done per CPU global for all dm-crypt instances.
1802 * They should not depend on each other and do not block.
1804 static void crypt_endio(struct bio *clone)
1806 struct dm_crypt_io *io = clone->bi_private;
1807 struct crypt_config *cc = io->cc;
1808 unsigned rw = bio_data_dir(clone);
1812 * free the processed pages
1815 crypt_free_buffer_pages(cc, clone);
1817 error = clone->bi_status;
1820 if (rw == READ && !error) {
1821 kcryptd_queue_crypt(io);
1825 if (unlikely(error))
1828 crypt_dec_pending(io);
1831 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1833 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1835 struct crypt_config *cc = io->cc;
1839 * We need the original biovec array in order to decrypt the whole bio
1840 * data *afterwards* -- thanks to immutable biovecs we don't need to
1841 * worry about the block layer modifying the biovec array; so leverage
1842 * bio_alloc_clone().
1844 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1847 clone->bi_private = io;
1848 clone->bi_end_io = crypt_endio;
1850 crypt_inc_pending(io);
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 dm_submit_bio_remap(io->base_bio, 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 dm_submit_bio_remap(io->base_bio, 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 dm_submit_bio_remap(io->base_bio, 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_ZEROES:
2008 static void kcryptd_crypt_write_continue(struct work_struct *work)
2010 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2011 struct crypt_config *cc = io->cc;
2012 struct convert_context *ctx = &io->ctx;
2014 sector_t sector = io->sector;
2017 wait_for_completion(&ctx->restart);
2018 reinit_completion(&ctx->restart);
2020 r = crypt_convert(cc, &io->ctx, true, false);
2023 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2024 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2025 /* Wait for completion signaled by kcryptd_async_done() */
2026 wait_for_completion(&ctx->restart);
2030 /* Encryption was already finished, submit io now */
2031 if (crypt_finished) {
2032 kcryptd_crypt_write_io_submit(io, 0);
2033 io->sector = sector;
2036 crypt_dec_pending(io);
2039 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2041 struct crypt_config *cc = io->cc;
2042 struct convert_context *ctx = &io->ctx;
2045 sector_t sector = io->sector;
2049 * Prevent io from disappearing until this function completes.
2051 crypt_inc_pending(io);
2052 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2054 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2055 if (unlikely(!clone)) {
2056 io->error = BLK_STS_IOERR;
2060 io->ctx.bio_out = clone;
2061 io->ctx.iter_out = clone->bi_iter;
2063 sector += bio_sectors(clone);
2065 crypt_inc_pending(io);
2066 r = crypt_convert(cc, ctx,
2067 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2069 * Crypto API backlogged the request, because its queue was full
2070 * and we're in softirq context, so continue from a workqueue
2071 * (TODO: is it actually possible to be in softirq in the write path?)
2073 if (r == BLK_STS_DEV_RESOURCE) {
2074 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2075 queue_work(cc->crypt_queue, &io->work);
2080 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2081 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2082 /* Wait for completion signaled by kcryptd_async_done() */
2083 wait_for_completion(&ctx->restart);
2087 /* Encryption was already finished, submit io now */
2088 if (crypt_finished) {
2089 kcryptd_crypt_write_io_submit(io, 0);
2090 io->sector = sector;
2094 crypt_dec_pending(io);
2097 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2099 crypt_dec_pending(io);
2102 static void kcryptd_crypt_read_continue(struct work_struct *work)
2104 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2105 struct crypt_config *cc = io->cc;
2108 wait_for_completion(&io->ctx.restart);
2109 reinit_completion(&io->ctx.restart);
2111 r = crypt_convert(cc, &io->ctx, true, false);
2115 if (atomic_dec_and_test(&io->ctx.cc_pending))
2116 kcryptd_crypt_read_done(io);
2118 crypt_dec_pending(io);
2121 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2123 struct crypt_config *cc = io->cc;
2126 crypt_inc_pending(io);
2128 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2131 r = crypt_convert(cc, &io->ctx,
2132 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2134 * Crypto API backlogged the request, because its queue was full
2135 * and we're in softirq context, so continue from a workqueue
2137 if (r == BLK_STS_DEV_RESOURCE) {
2138 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2139 queue_work(cc->crypt_queue, &io->work);
2145 if (atomic_dec_and_test(&io->ctx.cc_pending))
2146 kcryptd_crypt_read_done(io);
2148 crypt_dec_pending(io);
2151 static void kcryptd_async_done(struct crypto_async_request *async_req,
2154 struct dm_crypt_request *dmreq = async_req->data;
2155 struct convert_context *ctx = dmreq->ctx;
2156 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2157 struct crypt_config *cc = io->cc;
2160 * A request from crypto driver backlog is going to be processed now,
2161 * finish the completion and continue in crypt_convert().
2162 * (Callback will be called for the second time for this request.)
2164 if (error == -EINPROGRESS) {
2165 complete(&ctx->restart);
2169 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2170 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2172 if (error == -EBADMSG) {
2173 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2175 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2176 ctx->bio_in->bi_bdev, s);
2177 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2179 io->error = BLK_STS_PROTECTION;
2180 } else if (error < 0)
2181 io->error = BLK_STS_IOERR;
2183 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2185 if (!atomic_dec_and_test(&ctx->cc_pending))
2189 * The request is fully completed: for inline writes, let
2190 * kcryptd_crypt_write_convert() do the IO submission.
2192 if (bio_data_dir(io->base_bio) == READ) {
2193 kcryptd_crypt_read_done(io);
2197 if (kcryptd_crypt_write_inline(cc, ctx)) {
2198 complete(&ctx->restart);
2202 kcryptd_crypt_write_io_submit(io, 1);
2205 static void kcryptd_crypt(struct work_struct *work)
2207 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2209 if (bio_data_dir(io->base_bio) == READ)
2210 kcryptd_crypt_read_convert(io);
2212 kcryptd_crypt_write_convert(io);
2215 static void kcryptd_crypt_tasklet(unsigned long work)
2217 kcryptd_crypt((struct work_struct *)work);
2220 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2222 struct crypt_config *cc = io->cc;
2224 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2225 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2227 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2228 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2229 * it is being executed with irqs disabled.
2231 if (in_hardirq() || irqs_disabled()) {
2232 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2233 tasklet_schedule(&io->tasklet);
2237 kcryptd_crypt(&io->work);
2241 INIT_WORK(&io->work, kcryptd_crypt);
2242 queue_work(cc->crypt_queue, &io->work);
2245 static void crypt_free_tfms_aead(struct crypt_config *cc)
2247 if (!cc->cipher_tfm.tfms_aead)
2250 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2251 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2252 cc->cipher_tfm.tfms_aead[0] = NULL;
2255 kfree(cc->cipher_tfm.tfms_aead);
2256 cc->cipher_tfm.tfms_aead = NULL;
2259 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2263 if (!cc->cipher_tfm.tfms)
2266 for (i = 0; i < cc->tfms_count; i++)
2267 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2268 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2269 cc->cipher_tfm.tfms[i] = NULL;
2272 kfree(cc->cipher_tfm.tfms);
2273 cc->cipher_tfm.tfms = NULL;
2276 static void crypt_free_tfms(struct crypt_config *cc)
2278 if (crypt_integrity_aead(cc))
2279 crypt_free_tfms_aead(cc);
2281 crypt_free_tfms_skcipher(cc);
2284 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2289 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2290 sizeof(struct crypto_skcipher *),
2292 if (!cc->cipher_tfm.tfms)
2295 for (i = 0; i < cc->tfms_count; i++) {
2296 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2297 CRYPTO_ALG_ALLOCATES_MEMORY);
2298 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2299 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2300 crypt_free_tfms(cc);
2306 * dm-crypt performance can vary greatly depending on which crypto
2307 * algorithm implementation is used. Help people debug performance
2308 * problems by logging the ->cra_driver_name.
2310 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2311 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2315 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2319 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2320 if (!cc->cipher_tfm.tfms)
2323 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2324 CRYPTO_ALG_ALLOCATES_MEMORY);
2325 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2326 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2327 crypt_free_tfms(cc);
2331 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2332 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2336 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2338 if (crypt_integrity_aead(cc))
2339 return crypt_alloc_tfms_aead(cc, ciphermode);
2341 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2344 static unsigned crypt_subkey_size(struct crypt_config *cc)
2346 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2349 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2351 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2355 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2356 * the key must be for some reason in special format.
2357 * This funcion converts cc->key to this special format.
2359 static void crypt_copy_authenckey(char *p, const void *key,
2360 unsigned enckeylen, unsigned authkeylen)
2362 struct crypto_authenc_key_param *param;
2365 rta = (struct rtattr *)p;
2366 param = RTA_DATA(rta);
2367 param->enckeylen = cpu_to_be32(enckeylen);
2368 rta->rta_len = RTA_LENGTH(sizeof(*param));
2369 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2370 p += RTA_SPACE(sizeof(*param));
2371 memcpy(p, key + enckeylen, authkeylen);
2373 memcpy(p, key, enckeylen);
2376 static int crypt_setkey(struct crypt_config *cc)
2378 unsigned subkey_size;
2381 /* Ignore extra keys (which are used for IV etc) */
2382 subkey_size = crypt_subkey_size(cc);
2384 if (crypt_integrity_hmac(cc)) {
2385 if (subkey_size < cc->key_mac_size)
2388 crypt_copy_authenckey(cc->authenc_key, cc->key,
2389 subkey_size - cc->key_mac_size,
2393 for (i = 0; i < cc->tfms_count; i++) {
2394 if (crypt_integrity_hmac(cc))
2395 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2396 cc->authenc_key, crypt_authenckey_size(cc));
2397 else if (crypt_integrity_aead(cc))
2398 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2399 cc->key + (i * subkey_size),
2402 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2403 cc->key + (i * subkey_size),
2409 if (crypt_integrity_hmac(cc))
2410 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2417 static bool contains_whitespace(const char *str)
2420 if (isspace(*str++))
2425 static int set_key_user(struct crypt_config *cc, struct key *key)
2427 const struct user_key_payload *ukp;
2429 ukp = user_key_payload_locked(key);
2431 return -EKEYREVOKED;
2433 if (cc->key_size != ukp->datalen)
2436 memcpy(cc->key, ukp->data, cc->key_size);
2441 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2443 const struct encrypted_key_payload *ekp;
2445 ekp = key->payload.data[0];
2447 return -EKEYREVOKED;
2449 if (cc->key_size != ekp->decrypted_datalen)
2452 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2457 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2459 const struct trusted_key_payload *tkp;
2461 tkp = key->payload.data[0];
2463 return -EKEYREVOKED;
2465 if (cc->key_size != tkp->key_len)
2468 memcpy(cc->key, tkp->key, cc->key_size);
2473 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2475 char *new_key_string, *key_desc;
2477 struct key_type *type;
2479 int (*set_key)(struct crypt_config *cc, struct key *key);
2482 * Reject key_string with whitespace. dm core currently lacks code for
2483 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2485 if (contains_whitespace(key_string)) {
2486 DMERR("whitespace chars not allowed in key string");
2490 /* look for next ':' separating key_type from key_description */
2491 key_desc = strchr(key_string, ':');
2492 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2495 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2496 type = &key_type_logon;
2497 set_key = set_key_user;
2498 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2499 type = &key_type_user;
2500 set_key = set_key_user;
2501 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2502 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2503 type = &key_type_encrypted;
2504 set_key = set_key_encrypted;
2505 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2506 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2507 type = &key_type_trusted;
2508 set_key = set_key_trusted;
2513 new_key_string = kstrdup(key_string, GFP_KERNEL);
2514 if (!new_key_string)
2517 key = request_key(type, key_desc + 1, NULL);
2519 kfree_sensitive(new_key_string);
2520 return PTR_ERR(key);
2523 down_read(&key->sem);
2525 ret = set_key(cc, key);
2529 kfree_sensitive(new_key_string);
2536 /* clear the flag since following operations may invalidate previously valid key */
2537 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2539 ret = crypt_setkey(cc);
2542 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2543 kfree_sensitive(cc->key_string);
2544 cc->key_string = new_key_string;
2546 kfree_sensitive(new_key_string);
2551 static int get_key_size(char **key_string)
2556 if (*key_string[0] != ':')
2557 return strlen(*key_string) >> 1;
2559 /* look for next ':' in key string */
2560 colon = strpbrk(*key_string + 1, ":");
2564 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2567 *key_string = colon;
2569 /* remaining key string should be :<logon|user>:<key_desc> */
2576 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2581 static int get_key_size(char **key_string)
2583 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2586 #endif /* CONFIG_KEYS */
2588 static int crypt_set_key(struct crypt_config *cc, char *key)
2591 int key_string_len = strlen(key);
2593 /* Hyphen (which gives a key_size of zero) means there is no key. */
2594 if (!cc->key_size && strcmp(key, "-"))
2597 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2598 if (key[0] == ':') {
2599 r = crypt_set_keyring_key(cc, key + 1);
2603 /* clear the flag since following operations may invalidate previously valid key */
2604 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2606 /* wipe references to any kernel keyring key */
2607 kfree_sensitive(cc->key_string);
2608 cc->key_string = NULL;
2610 /* Decode key from its hex representation. */
2611 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2614 r = crypt_setkey(cc);
2616 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2619 /* Hex key string not needed after here, so wipe it. */
2620 memset(key, '0', key_string_len);
2625 static int crypt_wipe_key(struct crypt_config *cc)
2629 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2630 get_random_bytes(&cc->key, cc->key_size);
2632 /* Wipe IV private keys */
2633 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2634 r = cc->iv_gen_ops->wipe(cc);
2639 kfree_sensitive(cc->key_string);
2640 cc->key_string = NULL;
2641 r = crypt_setkey(cc);
2642 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2647 static void crypt_calculate_pages_per_client(void)
2649 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2651 if (!dm_crypt_clients_n)
2654 pages /= dm_crypt_clients_n;
2655 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2656 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2657 dm_crypt_pages_per_client = pages;
2660 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2662 struct crypt_config *cc = pool_data;
2666 * Note, percpu_counter_read_positive() may over (and under) estimate
2667 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2668 * but avoids potential spinlock contention of an exact result.
2670 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2671 likely(gfp_mask & __GFP_NORETRY))
2674 page = alloc_page(gfp_mask);
2675 if (likely(page != NULL))
2676 percpu_counter_add(&cc->n_allocated_pages, 1);
2681 static void crypt_page_free(void *page, void *pool_data)
2683 struct crypt_config *cc = pool_data;
2686 percpu_counter_sub(&cc->n_allocated_pages, 1);
2689 static void crypt_dtr(struct dm_target *ti)
2691 struct crypt_config *cc = ti->private;
2698 if (cc->write_thread)
2699 kthread_stop(cc->write_thread);
2702 destroy_workqueue(cc->io_queue);
2703 if (cc->crypt_queue)
2704 destroy_workqueue(cc->crypt_queue);
2706 crypt_free_tfms(cc);
2708 bioset_exit(&cc->bs);
2710 mempool_exit(&cc->page_pool);
2711 mempool_exit(&cc->req_pool);
2712 mempool_exit(&cc->tag_pool);
2714 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2715 percpu_counter_destroy(&cc->n_allocated_pages);
2717 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2718 cc->iv_gen_ops->dtr(cc);
2721 dm_put_device(ti, cc->dev);
2723 kfree_sensitive(cc->cipher_string);
2724 kfree_sensitive(cc->key_string);
2725 kfree_sensitive(cc->cipher_auth);
2726 kfree_sensitive(cc->authenc_key);
2728 mutex_destroy(&cc->bio_alloc_lock);
2730 /* Must zero key material before freeing */
2731 kfree_sensitive(cc);
2733 spin_lock(&dm_crypt_clients_lock);
2734 WARN_ON(!dm_crypt_clients_n);
2735 dm_crypt_clients_n--;
2736 crypt_calculate_pages_per_client();
2737 spin_unlock(&dm_crypt_clients_lock);
2739 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2742 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2744 struct crypt_config *cc = ti->private;
2746 if (crypt_integrity_aead(cc))
2747 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2749 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2752 /* at least a 64 bit sector number should fit in our buffer */
2753 cc->iv_size = max(cc->iv_size,
2754 (unsigned int)(sizeof(u64) / sizeof(u8)));
2756 DMWARN("Selected cipher does not support IVs");
2760 /* Choose ivmode, see comments at iv code. */
2762 cc->iv_gen_ops = NULL;
2763 else if (strcmp(ivmode, "plain") == 0)
2764 cc->iv_gen_ops = &crypt_iv_plain_ops;
2765 else if (strcmp(ivmode, "plain64") == 0)
2766 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2767 else if (strcmp(ivmode, "plain64be") == 0)
2768 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2769 else if (strcmp(ivmode, "essiv") == 0)
2770 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2771 else if (strcmp(ivmode, "benbi") == 0)
2772 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2773 else if (strcmp(ivmode, "null") == 0)
2774 cc->iv_gen_ops = &crypt_iv_null_ops;
2775 else if (strcmp(ivmode, "eboiv") == 0)
2776 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2777 else if (strcmp(ivmode, "elephant") == 0) {
2778 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2780 cc->key_extra_size = cc->key_size / 2;
2781 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2783 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2784 } else if (strcmp(ivmode, "lmk") == 0) {
2785 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2787 * Version 2 and 3 is recognised according
2788 * to length of provided multi-key string.
2789 * If present (version 3), last key is used as IV seed.
2790 * All keys (including IV seed) are always the same size.
2792 if (cc->key_size % cc->key_parts) {
2794 cc->key_extra_size = cc->key_size / cc->key_parts;
2796 } else if (strcmp(ivmode, "tcw") == 0) {
2797 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2798 cc->key_parts += 2; /* IV + whitening */
2799 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2800 } else if (strcmp(ivmode, "random") == 0) {
2801 cc->iv_gen_ops = &crypt_iv_random_ops;
2802 /* Need storage space in integrity fields. */
2803 cc->integrity_iv_size = cc->iv_size;
2805 ti->error = "Invalid IV mode";
2813 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2814 * The HMAC is needed to calculate tag size (HMAC digest size).
2815 * This should be probably done by crypto-api calls (once available...)
2817 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2819 char *start, *end, *mac_alg = NULL;
2820 struct crypto_ahash *mac;
2822 if (!strstarts(cipher_api, "authenc("))
2825 start = strchr(cipher_api, '(');
2826 end = strchr(cipher_api, ',');
2827 if (!start || !end || ++start > end)
2830 mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2833 strncpy(mac_alg, start, end - start);
2835 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2839 return PTR_ERR(mac);
2841 cc->key_mac_size = crypto_ahash_digestsize(mac);
2842 crypto_free_ahash(mac);
2844 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2845 if (!cc->authenc_key)
2851 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2852 char **ivmode, char **ivopts)
2854 struct crypt_config *cc = ti->private;
2855 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2861 * New format (capi: prefix)
2862 * capi:cipher_api_spec-iv:ivopts
2864 tmp = &cipher_in[strlen("capi:")];
2866 /* Separate IV options if present, it can contain another '-' in hash name */
2867 *ivopts = strrchr(tmp, ':');
2873 *ivmode = strrchr(tmp, '-');
2878 /* The rest is crypto API spec */
2881 /* Alloc AEAD, can be used only in new format. */
2882 if (crypt_integrity_aead(cc)) {
2883 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2885 ti->error = "Invalid AEAD cipher spec";
2890 if (*ivmode && !strcmp(*ivmode, "lmk"))
2891 cc->tfms_count = 64;
2893 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2895 ti->error = "Digest algorithm missing for ESSIV mode";
2898 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2899 cipher_api, *ivopts);
2900 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2901 ti->error = "Cannot allocate cipher string";
2907 cc->key_parts = cc->tfms_count;
2909 /* Allocate cipher */
2910 ret = crypt_alloc_tfms(cc, cipher_api);
2912 ti->error = "Error allocating crypto tfm";
2916 if (crypt_integrity_aead(cc))
2917 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2919 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2924 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2925 char **ivmode, char **ivopts)
2927 struct crypt_config *cc = ti->private;
2928 char *tmp, *cipher, *chainmode, *keycount;
2929 char *cipher_api = NULL;
2933 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2934 ti->error = "Bad cipher specification";
2939 * Legacy dm-crypt cipher specification
2940 * cipher[:keycount]-mode-iv:ivopts
2943 keycount = strsep(&tmp, "-");
2944 cipher = strsep(&keycount, ":");
2948 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2949 !is_power_of_2(cc->tfms_count)) {
2950 ti->error = "Bad cipher key count specification";
2953 cc->key_parts = cc->tfms_count;
2955 chainmode = strsep(&tmp, "-");
2956 *ivmode = strsep(&tmp, ":");
2960 * For compatibility with the original dm-crypt mapping format, if
2961 * only the cipher name is supplied, use cbc-plain.
2963 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2968 if (strcmp(chainmode, "ecb") && !*ivmode) {
2969 ti->error = "IV mechanism required";
2973 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2977 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2979 ti->error = "Digest algorithm missing for ESSIV mode";
2983 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2984 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2986 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2987 "%s(%s)", chainmode, cipher);
2989 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2994 /* Allocate cipher */
2995 ret = crypt_alloc_tfms(cc, cipher_api);
2997 ti->error = "Error allocating crypto tfm";
3005 ti->error = "Cannot allocate cipher strings";
3009 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3011 struct crypt_config *cc = ti->private;
3012 char *ivmode = NULL, *ivopts = NULL;
3015 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3016 if (!cc->cipher_string) {
3017 ti->error = "Cannot allocate cipher strings";
3021 if (strstarts(cipher_in, "capi:"))
3022 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3024 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3029 ret = crypt_ctr_ivmode(ti, ivmode);
3033 /* Initialize and set key */
3034 ret = crypt_set_key(cc, key);
3036 ti->error = "Error decoding and setting key";
3041 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3042 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3044 ti->error = "Error creating IV";
3049 /* Initialize IV (set keys for ESSIV etc) */
3050 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3051 ret = cc->iv_gen_ops->init(cc);
3053 ti->error = "Error initialising IV";
3058 /* wipe the kernel key payload copy */
3060 memset(cc->key, 0, cc->key_size * sizeof(u8));
3065 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3067 struct crypt_config *cc = ti->private;
3068 struct dm_arg_set as;
3069 static const struct dm_arg _args[] = {
3070 {0, 8, "Invalid number of feature args"},
3072 unsigned int opt_params, val;
3073 const char *opt_string, *sval;
3077 /* Optional parameters */
3081 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3085 while (opt_params--) {
3086 opt_string = dm_shift_arg(&as);
3088 ti->error = "Not enough feature arguments";
3092 if (!strcasecmp(opt_string, "allow_discards"))
3093 ti->num_discard_bios = 1;
3095 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3096 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3098 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3099 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3100 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3101 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3102 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3103 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3104 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3105 if (val == 0 || val > MAX_TAG_SIZE) {
3106 ti->error = "Invalid integrity arguments";
3109 cc->on_disk_tag_size = val;
3110 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3111 if (!strcasecmp(sval, "aead")) {
3112 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3113 } else if (strcasecmp(sval, "none")) {
3114 ti->error = "Unknown integrity profile";
3118 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3119 if (!cc->cipher_auth)
3121 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3122 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3123 cc->sector_size > 4096 ||
3124 (cc->sector_size & (cc->sector_size - 1))) {
3125 ti->error = "Invalid feature value for sector_size";
3128 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3129 ti->error = "Device size is not multiple of sector_size feature";
3132 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3133 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3134 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3136 ti->error = "Invalid feature arguments";
3144 #ifdef CONFIG_BLK_DEV_ZONED
3145 static int crypt_report_zones(struct dm_target *ti,
3146 struct dm_report_zones_args *args, unsigned int nr_zones)
3148 struct crypt_config *cc = ti->private;
3150 return dm_report_zones(cc->dev->bdev, cc->start,
3151 cc->start + dm_target_offset(ti, args->next_sector),
3155 #define crypt_report_zones NULL
3159 * Construct an encryption mapping:
3160 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3162 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3164 struct crypt_config *cc;
3165 const char *devname = dm_table_device_name(ti->table);
3167 unsigned int align_mask;
3168 unsigned long long tmpll;
3170 size_t iv_size_padding, additional_req_size;
3174 ti->error = "Not enough arguments";
3178 key_size = get_key_size(&argv[1]);
3180 ti->error = "Cannot parse key size";
3184 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3186 ti->error = "Cannot allocate encryption context";
3189 cc->key_size = key_size;
3190 cc->sector_size = (1 << SECTOR_SHIFT);
3191 cc->sector_shift = 0;
3195 spin_lock(&dm_crypt_clients_lock);
3196 dm_crypt_clients_n++;
3197 crypt_calculate_pages_per_client();
3198 spin_unlock(&dm_crypt_clients_lock);
3200 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3204 /* Optional parameters need to be read before cipher constructor */
3206 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3211 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3215 if (crypt_integrity_aead(cc)) {
3216 cc->dmreq_start = sizeof(struct aead_request);
3217 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3218 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3220 cc->dmreq_start = sizeof(struct skcipher_request);
3221 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3222 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3224 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3226 if (align_mask < CRYPTO_MINALIGN) {
3227 /* Allocate the padding exactly */
3228 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3232 * If the cipher requires greater alignment than kmalloc
3233 * alignment, we don't know the exact position of the
3234 * initialization vector. We must assume worst case.
3236 iv_size_padding = align_mask;
3239 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3240 additional_req_size = sizeof(struct dm_crypt_request) +
3241 iv_size_padding + cc->iv_size +
3244 sizeof(unsigned int);
3246 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3248 ti->error = "Cannot allocate crypt request mempool";
3252 cc->per_bio_data_size = ti->per_io_data_size =
3253 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3254 ARCH_KMALLOC_MINALIGN);
3256 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3258 ti->error = "Cannot allocate page mempool";
3262 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3264 ti->error = "Cannot allocate crypt bioset";
3268 mutex_init(&cc->bio_alloc_lock);
3271 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3272 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3273 ti->error = "Invalid iv_offset sector";
3276 cc->iv_offset = tmpll;
3278 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3280 ti->error = "Device lookup failed";
3285 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3286 ti->error = "Invalid device sector";
3291 if (bdev_is_zoned(cc->dev->bdev)) {
3293 * For zoned block devices, we need to preserve the issuer write
3294 * ordering. To do so, disable write workqueues and force inline
3295 * encryption completion.
3297 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3298 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3301 * All zone append writes to a zone of a zoned block device will
3302 * have the same BIO sector, the start of the zone. When the
3303 * cypher IV mode uses sector values, all data targeting a
3304 * zone will be encrypted using the first sector numbers of the
3305 * zone. This will not result in write errors but will
3306 * cause most reads to fail as reads will use the sector values
3307 * for the actual data locations, resulting in IV mismatch.
3308 * To avoid this problem, ask DM core to emulate zone append
3309 * operations with regular writes.
3311 DMDEBUG("Zone append operations will be emulated");
3312 ti->emulate_zone_append = true;
3315 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3316 ret = crypt_integrity_ctr(cc, ti);
3320 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3321 if (!cc->tag_pool_max_sectors)
3322 cc->tag_pool_max_sectors = 1;
3324 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3325 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3327 ti->error = "Cannot allocate integrity tags mempool";
3331 cc->tag_pool_max_sectors <<= cc->sector_shift;
3335 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3336 if (!cc->io_queue) {
3337 ti->error = "Couldn't create kcryptd io queue";
3341 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3342 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3345 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3346 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3347 num_online_cpus(), devname);
3348 if (!cc->crypt_queue) {
3349 ti->error = "Couldn't create kcryptd queue";
3353 spin_lock_init(&cc->write_thread_lock);
3354 cc->write_tree = RB_ROOT;
3356 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3357 if (IS_ERR(cc->write_thread)) {
3358 ret = PTR_ERR(cc->write_thread);
3359 cc->write_thread = NULL;
3360 ti->error = "Couldn't spawn write thread";
3364 ti->num_flush_bios = 1;
3365 ti->limit_swap_bios = true;
3366 ti->accounts_remapped_io = true;
3368 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3372 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3377 static int crypt_map(struct dm_target *ti, struct bio *bio)
3379 struct dm_crypt_io *io;
3380 struct crypt_config *cc = ti->private;
3383 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3384 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3385 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3387 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3388 bio_op(bio) == REQ_OP_DISCARD)) {
3389 bio_set_dev(bio, cc->dev->bdev);
3390 if (bio_sectors(bio))
3391 bio->bi_iter.bi_sector = cc->start +
3392 dm_target_offset(ti, bio->bi_iter.bi_sector);
3393 return DM_MAPIO_REMAPPED;
3397 * Check if bio is too large, split as needed.
3399 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3400 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3401 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3404 * Ensure that bio is a multiple of internal sector encryption size
3405 * and is aligned to this size as defined in IO hints.
3407 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3408 return DM_MAPIO_KILL;
3410 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3411 return DM_MAPIO_KILL;
3413 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3414 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3416 if (cc->on_disk_tag_size) {
3417 unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3419 if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3420 unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3421 GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3422 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3423 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3424 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3425 io->integrity_metadata_from_pool = true;
3429 if (crypt_integrity_aead(cc))
3430 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3432 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3434 if (bio_data_dir(io->base_bio) == READ) {
3435 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3436 kcryptd_queue_read(io);
3438 kcryptd_queue_crypt(io);
3440 return DM_MAPIO_SUBMITTED;
3443 static char hex2asc(unsigned char c)
3445 return c + '0' + ((unsigned)(9 - c) >> 4 & 0x27);
3448 static void crypt_status(struct dm_target *ti, status_type_t type,
3449 unsigned status_flags, char *result, unsigned maxlen)
3451 struct crypt_config *cc = ti->private;
3453 int num_feature_args = 0;
3456 case STATUSTYPE_INFO:
3460 case STATUSTYPE_TABLE:
3461 DMEMIT("%s ", cc->cipher_string);
3463 if (cc->key_size > 0) {
3465 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3467 for (i = 0; i < cc->key_size; i++) {
3468 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3469 hex2asc(cc->key[i] & 0xf));
3475 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3476 cc->dev->name, (unsigned long long)cc->start);
3478 num_feature_args += !!ti->num_discard_bios;
3479 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3480 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3481 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3482 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3483 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3484 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3485 if (cc->on_disk_tag_size)
3487 if (num_feature_args) {
3488 DMEMIT(" %d", num_feature_args);
3489 if (ti->num_discard_bios)
3490 DMEMIT(" allow_discards");
3491 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3492 DMEMIT(" same_cpu_crypt");
3493 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3494 DMEMIT(" submit_from_crypt_cpus");
3495 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3496 DMEMIT(" no_read_workqueue");
3497 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3498 DMEMIT(" no_write_workqueue");
3499 if (cc->on_disk_tag_size)
3500 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3501 if (cc->sector_size != (1 << SECTOR_SHIFT))
3502 DMEMIT(" sector_size:%d", cc->sector_size);
3503 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3504 DMEMIT(" iv_large_sectors");
3508 case STATUSTYPE_IMA:
3509 DMEMIT_TARGET_NAME_VERSION(ti->type);
3510 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3511 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3512 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3514 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3516 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3518 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3521 if (cc->on_disk_tag_size)
3522 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3523 cc->on_disk_tag_size, cc->cipher_auth);
3524 if (cc->sector_size != (1 << SECTOR_SHIFT))
3525 DMEMIT(",sector_size=%d", cc->sector_size);
3526 if (cc->cipher_string)
3527 DMEMIT(",cipher_string=%s", cc->cipher_string);
3529 DMEMIT(",key_size=%u", cc->key_size);
3530 DMEMIT(",key_parts=%u", cc->key_parts);
3531 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3532 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3538 static void crypt_postsuspend(struct dm_target *ti)
3540 struct crypt_config *cc = ti->private;
3542 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3545 static int crypt_preresume(struct dm_target *ti)
3547 struct crypt_config *cc = ti->private;
3549 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3550 DMERR("aborting resume - crypt key is not set.");
3557 static void crypt_resume(struct dm_target *ti)
3559 struct crypt_config *cc = ti->private;
3561 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3564 /* Message interface
3568 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3569 char *result, unsigned maxlen)
3571 struct crypt_config *cc = ti->private;
3572 int key_size, ret = -EINVAL;
3577 if (!strcasecmp(argv[0], "key")) {
3578 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3579 DMWARN("not suspended during key manipulation.");
3582 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3583 /* The key size may not be changed. */
3584 key_size = get_key_size(&argv[2]);
3585 if (key_size < 0 || cc->key_size != key_size) {
3586 memset(argv[2], '0', strlen(argv[2]));
3590 ret = crypt_set_key(cc, argv[2]);
3593 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3594 ret = cc->iv_gen_ops->init(cc);
3595 /* wipe the kernel key payload copy */
3597 memset(cc->key, 0, cc->key_size * sizeof(u8));
3600 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3601 return crypt_wipe_key(cc);
3605 DMWARN("unrecognised message received.");
3609 static int crypt_iterate_devices(struct dm_target *ti,
3610 iterate_devices_callout_fn fn, void *data)
3612 struct crypt_config *cc = ti->private;
3614 return fn(ti, cc->dev, cc->start, ti->len, data);
3617 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3619 struct crypt_config *cc = ti->private;
3622 * Unfortunate constraint that is required to avoid the potential
3623 * for exceeding underlying device's max_segments limits -- due to
3624 * crypt_alloc_buffer() possibly allocating pages for the encryption
3625 * bio that are not as physically contiguous as the original bio.
3627 limits->max_segment_size = PAGE_SIZE;
3629 limits->logical_block_size =
3630 max_t(unsigned, limits->logical_block_size, cc->sector_size);
3631 limits->physical_block_size =
3632 max_t(unsigned, limits->physical_block_size, cc->sector_size);
3633 limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3634 limits->dma_alignment = limits->logical_block_size - 1;
3637 static struct target_type crypt_target = {
3639 .version = {1, 24, 0},
3640 .module = THIS_MODULE,
3643 .features = DM_TARGET_ZONED_HM,
3644 .report_zones = crypt_report_zones,
3646 .status = crypt_status,
3647 .postsuspend = crypt_postsuspend,
3648 .preresume = crypt_preresume,
3649 .resume = crypt_resume,
3650 .message = crypt_message,
3651 .iterate_devices = crypt_iterate_devices,
3652 .io_hints = crypt_io_hints,
3655 static int __init dm_crypt_init(void)
3659 r = dm_register_target(&crypt_target);
3661 DMERR("register failed %d", r);
3666 static void __exit dm_crypt_exit(void)
3668 dm_unregister_target(&crypt_target);
3671 module_init(dm_crypt_init);
3672 module_exit(dm_crypt_exit);
3674 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3675 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3676 MODULE_LICENSE("GPL");