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/blk-integrity.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/crypto.h>
22 #include <linux/workqueue.h>
23 #include <linux/kthread.h>
24 #include <linux/backing-dev.h>
25 #include <linux/atomic.h>
26 #include <linux/scatterlist.h>
27 #include <linux/rbtree.h>
28 #include <linux/ctype.h>
30 #include <asm/unaligned.h>
31 #include <crypto/hash.h>
32 #include <crypto/md5.h>
33 #include <crypto/algapi.h>
34 #include <crypto/skcipher.h>
35 #include <crypto/aead.h>
36 #include <crypto/authenc.h>
37 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
38 #include <linux/key-type.h>
39 #include <keys/user-type.h>
40 #include <keys/encrypted-type.h>
41 #include <keys/trusted-type.h>
43 #include <linux/device-mapper.h>
47 #define DM_MSG_PREFIX "crypt"
50 * context holding the current state of a multi-part conversion
52 struct convert_context {
53 struct completion restart;
56 struct bvec_iter iter_in;
57 struct bvec_iter iter_out;
61 struct skcipher_request *req;
62 struct aead_request *req_aead;
68 * per bio private data
71 struct crypt_config *cc;
73 u8 *integrity_metadata;
74 bool integrity_metadata_from_pool;
75 struct work_struct work;
76 struct tasklet_struct tasklet;
78 struct convert_context ctx;
84 struct rb_node rb_node;
85 } CRYPTO_MINALIGN_ATTR;
87 struct dm_crypt_request {
88 struct convert_context *ctx;
89 struct scatterlist sg_in[4];
90 struct scatterlist sg_out[4];
96 struct crypt_iv_operations {
97 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
99 void (*dtr)(struct crypt_config *cc);
100 int (*init)(struct crypt_config *cc);
101 int (*wipe)(struct crypt_config *cc);
102 int (*generator)(struct crypt_config *cc, u8 *iv,
103 struct dm_crypt_request *dmreq);
104 int (*post)(struct crypt_config *cc, u8 *iv,
105 struct dm_crypt_request *dmreq);
108 struct iv_benbi_private {
112 #define LMK_SEED_SIZE 64 /* hash + 0 */
113 struct iv_lmk_private {
114 struct crypto_shash *hash_tfm;
118 #define TCW_WHITENING_SIZE 16
119 struct iv_tcw_private {
120 struct crypto_shash *crc32_tfm;
125 #define ELEPHANT_MAX_KEY_SIZE 32
126 struct iv_elephant_private {
127 struct crypto_skcipher *tfm;
131 * Crypt: maps a linear range of a block device
132 * and encrypts / decrypts at the same time.
134 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
135 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
136 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
137 DM_CRYPT_WRITE_INLINE };
140 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
141 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
142 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
146 * The fields in here must be read only after initialization.
148 struct crypt_config {
152 struct percpu_counter n_allocated_pages;
154 struct workqueue_struct *io_queue;
155 struct workqueue_struct *crypt_queue;
157 spinlock_t write_thread_lock;
158 struct task_struct *write_thread;
159 struct rb_root write_tree;
165 const struct crypt_iv_operations *iv_gen_ops;
167 struct iv_benbi_private benbi;
168 struct iv_lmk_private lmk;
169 struct iv_tcw_private tcw;
170 struct iv_elephant_private elephant;
173 unsigned int iv_size;
174 unsigned short int sector_size;
175 unsigned char sector_shift;
178 struct crypto_skcipher **tfms;
179 struct crypto_aead **tfms_aead;
182 unsigned long cipher_flags;
185 * Layout of each crypto request:
187 * struct skcipher_request
190 * struct dm_crypt_request
194 * The padding is added so that dm_crypt_request and the IV are
197 unsigned int dmreq_start;
199 unsigned int per_bio_data_size;
202 unsigned int key_size;
203 unsigned int key_parts; /* independent parts in key buffer */
204 unsigned int key_extra_size; /* additional keys length */
205 unsigned int key_mac_size; /* MAC key size for authenc(...) */
207 unsigned int integrity_tag_size;
208 unsigned int integrity_iv_size;
209 unsigned int on_disk_tag_size;
212 * pool for per bio private data, crypto requests,
213 * encryption requeusts/buffer pages and integrity tags
215 unsigned tag_pool_max_sectors;
221 struct mutex bio_alloc_lock;
223 u8 *authenc_key; /* space for keys in authenc() format (if used) */
228 #define MAX_TAG_SIZE 480
229 #define POOL_ENTRY_SIZE 512
231 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
232 static unsigned dm_crypt_clients_n = 0;
233 static volatile unsigned long dm_crypt_pages_per_client;
234 #define DM_CRYPT_MEMORY_PERCENT 2
235 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
237 static void crypt_endio(struct bio *clone);
238 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
239 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
240 struct scatterlist *sg);
242 static bool crypt_integrity_aead(struct crypt_config *cc);
245 * Use this to access cipher attributes that are independent of the key.
247 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
249 return cc->cipher_tfm.tfms[0];
252 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
254 return cc->cipher_tfm.tfms_aead[0];
258 * Different IV generation algorithms:
260 * plain: the initial vector is the 32-bit little-endian version of the sector
261 * number, padded with zeros if necessary.
263 * plain64: the initial vector is the 64-bit little-endian version of the sector
264 * number, padded with zeros if necessary.
266 * plain64be: the initial vector is the 64-bit big-endian version of the sector
267 * number, padded with zeros if necessary.
269 * essiv: "encrypted sector|salt initial vector", the sector number is
270 * encrypted with the bulk cipher using a salt as key. The salt
271 * should be derived from the bulk cipher's key via hashing.
273 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
274 * (needed for LRW-32-AES and possible other narrow block modes)
276 * null: the initial vector is always zero. Provides compatibility with
277 * obsolete loop_fish2 devices. Do not use for new devices.
279 * lmk: Compatible implementation of the block chaining mode used
280 * by the Loop-AES block device encryption system
281 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
282 * It operates on full 512 byte sectors and uses CBC
283 * with an IV derived from the sector number, the data and
284 * optionally extra IV seed.
285 * This means that after decryption the first block
286 * of sector must be tweaked according to decrypted data.
287 * Loop-AES can use three encryption schemes:
288 * version 1: is plain aes-cbc mode
289 * version 2: uses 64 multikey scheme with lmk IV generator
290 * version 3: the same as version 2 with additional IV seed
291 * (it uses 65 keys, last key is used as IV seed)
293 * tcw: Compatible implementation of the block chaining mode used
294 * by the TrueCrypt device encryption system (prior to version 4.1).
295 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
296 * It operates on full 512 byte sectors and uses CBC
297 * with an IV derived from initial key and the sector number.
298 * In addition, whitening value is applied on every sector, whitening
299 * is calculated from initial key, sector number and mixed using CRC32.
300 * Note that this encryption scheme is vulnerable to watermarking attacks
301 * and should be used for old compatible containers access only.
303 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
304 * The IV is encrypted little-endian byte-offset (with the same key
305 * and cipher as the volume).
307 * elephant: The extended version of eboiv with additional Elephant diffuser
308 * used with Bitlocker CBC mode.
309 * This mode was used in older Windows systems
310 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
313 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
314 struct dm_crypt_request *dmreq)
316 memset(iv, 0, cc->iv_size);
317 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
322 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
323 struct dm_crypt_request *dmreq)
325 memset(iv, 0, cc->iv_size);
326 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
331 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
332 struct dm_crypt_request *dmreq)
334 memset(iv, 0, cc->iv_size);
335 /* iv_size is at least of size u64; usually it is 16 bytes */
336 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
341 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
342 struct dm_crypt_request *dmreq)
345 * ESSIV encryption of the IV is now handled by the crypto API,
346 * so just pass the plain sector number here.
348 memset(iv, 0, cc->iv_size);
349 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
354 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
360 if (crypt_integrity_aead(cc))
361 bs = crypto_aead_blocksize(any_tfm_aead(cc));
363 bs = crypto_skcipher_blocksize(any_tfm(cc));
366 /* we need to calculate how far we must shift the sector count
367 * to get the cipher block count, we use this shift in _gen */
369 if (1 << log != bs) {
370 ti->error = "cypher blocksize is not a power of 2";
375 ti->error = "cypher blocksize is > 512";
379 cc->iv_gen_private.benbi.shift = 9 - log;
384 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
388 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
389 struct dm_crypt_request *dmreq)
393 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
395 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
396 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
401 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
402 struct dm_crypt_request *dmreq)
404 memset(iv, 0, cc->iv_size);
409 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
411 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
413 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
414 crypto_free_shash(lmk->hash_tfm);
415 lmk->hash_tfm = NULL;
417 kfree_sensitive(lmk->seed);
421 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
424 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
426 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
427 ti->error = "Unsupported sector size for LMK";
431 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
432 CRYPTO_ALG_ALLOCATES_MEMORY);
433 if (IS_ERR(lmk->hash_tfm)) {
434 ti->error = "Error initializing LMK hash";
435 return PTR_ERR(lmk->hash_tfm);
438 /* No seed in LMK version 2 */
439 if (cc->key_parts == cc->tfms_count) {
444 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
446 crypt_iv_lmk_dtr(cc);
447 ti->error = "Error kmallocing seed storage in LMK";
454 static int crypt_iv_lmk_init(struct crypt_config *cc)
456 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
457 int subkey_size = cc->key_size / cc->key_parts;
459 /* LMK seed is on the position of LMK_KEYS + 1 key */
461 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
462 crypto_shash_digestsize(lmk->hash_tfm));
467 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
469 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
472 memset(lmk->seed, 0, LMK_SEED_SIZE);
477 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
478 struct dm_crypt_request *dmreq,
481 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
482 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
483 struct md5_state md5state;
487 desc->tfm = lmk->hash_tfm;
489 r = crypto_shash_init(desc);
494 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
499 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
500 r = crypto_shash_update(desc, data + 16, 16 * 31);
504 /* Sector is cropped to 56 bits here */
505 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
506 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
507 buf[2] = cpu_to_le32(4024);
509 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
513 /* No MD5 padding here */
514 r = crypto_shash_export(desc, &md5state);
518 for (i = 0; i < MD5_HASH_WORDS; i++)
519 __cpu_to_le32s(&md5state.hash[i]);
520 memcpy(iv, &md5state.hash, cc->iv_size);
525 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
526 struct dm_crypt_request *dmreq)
528 struct scatterlist *sg;
532 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
533 sg = crypt_get_sg_data(cc, dmreq->sg_in);
534 src = kmap_atomic(sg_page(sg));
535 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
538 memset(iv, 0, cc->iv_size);
543 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
544 struct dm_crypt_request *dmreq)
546 struct scatterlist *sg;
550 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
553 sg = crypt_get_sg_data(cc, dmreq->sg_out);
554 dst = kmap_atomic(sg_page(sg));
555 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
557 /* Tweak the first block of plaintext sector */
559 crypto_xor(dst + sg->offset, iv, cc->iv_size);
565 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
567 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
569 kfree_sensitive(tcw->iv_seed);
571 kfree_sensitive(tcw->whitening);
572 tcw->whitening = NULL;
574 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
575 crypto_free_shash(tcw->crc32_tfm);
576 tcw->crc32_tfm = NULL;
579 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
582 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
584 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
585 ti->error = "Unsupported sector size for TCW";
589 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
590 ti->error = "Wrong key size for TCW";
594 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
595 CRYPTO_ALG_ALLOCATES_MEMORY);
596 if (IS_ERR(tcw->crc32_tfm)) {
597 ti->error = "Error initializing CRC32 in TCW";
598 return PTR_ERR(tcw->crc32_tfm);
601 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
602 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
603 if (!tcw->iv_seed || !tcw->whitening) {
604 crypt_iv_tcw_dtr(cc);
605 ti->error = "Error allocating seed storage in TCW";
612 static int crypt_iv_tcw_init(struct crypt_config *cc)
614 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
615 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
617 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
618 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
624 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
626 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
628 memset(tcw->iv_seed, 0, cc->iv_size);
629 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
634 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
635 struct dm_crypt_request *dmreq,
638 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
639 __le64 sector = cpu_to_le64(dmreq->iv_sector);
640 u8 buf[TCW_WHITENING_SIZE];
641 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
644 /* xor whitening with sector number */
645 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
646 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
648 /* calculate crc32 for every 32bit part and xor it */
649 desc->tfm = tcw->crc32_tfm;
650 for (i = 0; i < 4; i++) {
651 r = crypto_shash_init(desc);
654 r = crypto_shash_update(desc, &buf[i * 4], 4);
657 r = crypto_shash_final(desc, &buf[i * 4]);
661 crypto_xor(&buf[0], &buf[12], 4);
662 crypto_xor(&buf[4], &buf[8], 4);
664 /* apply whitening (8 bytes) to whole sector */
665 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
666 crypto_xor(data + i * 8, buf, 8);
668 memzero_explicit(buf, sizeof(buf));
672 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
673 struct dm_crypt_request *dmreq)
675 struct scatterlist *sg;
676 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
677 __le64 sector = cpu_to_le64(dmreq->iv_sector);
681 /* Remove whitening from ciphertext */
682 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
683 sg = crypt_get_sg_data(cc, dmreq->sg_in);
684 src = kmap_atomic(sg_page(sg));
685 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
690 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
692 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
698 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
699 struct dm_crypt_request *dmreq)
701 struct scatterlist *sg;
705 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
708 /* Apply whitening on ciphertext */
709 sg = crypt_get_sg_data(cc, dmreq->sg_out);
710 dst = kmap_atomic(sg_page(sg));
711 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
717 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
718 struct dm_crypt_request *dmreq)
720 /* Used only for writes, there must be an additional space to store IV */
721 get_random_bytes(iv, cc->iv_size);
725 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
728 if (crypt_integrity_aead(cc)) {
729 ti->error = "AEAD transforms not supported for EBOIV";
733 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
734 ti->error = "Block size of EBOIV cipher does "
735 "not match IV size of block cipher";
742 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
743 struct dm_crypt_request *dmreq)
745 u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
746 struct skcipher_request *req;
747 struct scatterlist src, dst;
748 DECLARE_CRYPTO_WAIT(wait);
751 req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
755 memset(buf, 0, cc->iv_size);
756 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
758 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
759 sg_init_one(&dst, iv, cc->iv_size);
760 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
761 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
762 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
763 skcipher_request_free(req);
768 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
770 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
772 crypto_free_skcipher(elephant->tfm);
773 elephant->tfm = NULL;
776 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
779 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
782 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
783 CRYPTO_ALG_ALLOCATES_MEMORY);
784 if (IS_ERR(elephant->tfm)) {
785 r = PTR_ERR(elephant->tfm);
786 elephant->tfm = NULL;
790 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
792 crypt_iv_elephant_dtr(cc);
796 static void diffuser_disk_to_cpu(u32 *d, size_t n)
798 #ifndef __LITTLE_ENDIAN
801 for (i = 0; i < n; i++)
802 d[i] = le32_to_cpu((__le32)d[i]);
806 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
808 #ifndef __LITTLE_ENDIAN
811 for (i = 0; i < n; i++)
812 d[i] = cpu_to_le32((u32)d[i]);
816 static void diffuser_a_decrypt(u32 *d, size_t n)
820 for (i = 0; i < 5; i++) {
825 while (i1 < (n - 1)) {
826 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
832 d[i1] += d[i2] ^ d[i3];
838 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
841 d[i1] += d[i2] ^ d[i3];
847 static void diffuser_a_encrypt(u32 *d, size_t n)
851 for (i = 0; i < 5; i++) {
857 d[i1] -= d[i2] ^ d[i3];
860 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
866 d[i1] -= d[i2] ^ d[i3];
872 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
878 static void diffuser_b_decrypt(u32 *d, size_t n)
882 for (i = 0; i < 3; i++) {
887 while (i1 < (n - 1)) {
888 d[i1] += d[i2] ^ d[i3];
891 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
897 d[i1] += d[i2] ^ d[i3];
903 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
909 static void diffuser_b_encrypt(u32 *d, size_t n)
913 for (i = 0; i < 3; i++) {
919 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
925 d[i1] -= d[i2] ^ d[i3];
931 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
934 d[i1] -= d[i2] ^ d[i3];
940 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
942 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
943 u8 *es, *ks, *data, *data2, *data_offset;
944 struct skcipher_request *req;
945 struct scatterlist *sg, *sg2, src, dst;
946 DECLARE_CRYPTO_WAIT(wait);
949 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
950 es = kzalloc(16, GFP_NOIO); /* Key for AES */
951 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
953 if (!req || !es || !ks) {
958 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
961 sg_init_one(&src, es, 16);
962 sg_init_one(&dst, ks, 16);
963 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
964 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
965 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
971 sg_init_one(&dst, &ks[16], 16);
972 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
976 sg = crypt_get_sg_data(cc, dmreq->sg_out);
977 data = kmap_atomic(sg_page(sg));
978 data_offset = data + sg->offset;
980 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
981 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
982 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
983 data2 = kmap_atomic(sg_page(sg2));
984 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
985 kunmap_atomic(data2);
988 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
989 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
990 diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
991 diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
992 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
995 for (i = 0; i < (cc->sector_size / 32); i++)
996 crypto_xor(data_offset + i * 32, ks, 32);
998 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
999 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
1000 diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1001 diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1002 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1005 kunmap_atomic(data);
1007 kfree_sensitive(ks);
1008 kfree_sensitive(es);
1009 skcipher_request_free(req);
1013 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1014 struct dm_crypt_request *dmreq)
1018 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1019 r = crypt_iv_elephant(cc, dmreq);
1024 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1027 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1028 struct dm_crypt_request *dmreq)
1030 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1031 return crypt_iv_elephant(cc, dmreq);
1036 static int crypt_iv_elephant_init(struct crypt_config *cc)
1038 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1039 int key_offset = cc->key_size - cc->key_extra_size;
1041 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1044 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1046 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1047 u8 key[ELEPHANT_MAX_KEY_SIZE];
1049 memset(key, 0, cc->key_extra_size);
1050 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1053 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1054 .generator = crypt_iv_plain_gen
1057 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1058 .generator = crypt_iv_plain64_gen
1061 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1062 .generator = crypt_iv_plain64be_gen
1065 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1066 .generator = crypt_iv_essiv_gen
1069 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1070 .ctr = crypt_iv_benbi_ctr,
1071 .dtr = crypt_iv_benbi_dtr,
1072 .generator = crypt_iv_benbi_gen
1075 static const struct crypt_iv_operations crypt_iv_null_ops = {
1076 .generator = crypt_iv_null_gen
1079 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1080 .ctr = crypt_iv_lmk_ctr,
1081 .dtr = crypt_iv_lmk_dtr,
1082 .init = crypt_iv_lmk_init,
1083 .wipe = crypt_iv_lmk_wipe,
1084 .generator = crypt_iv_lmk_gen,
1085 .post = crypt_iv_lmk_post
1088 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1089 .ctr = crypt_iv_tcw_ctr,
1090 .dtr = crypt_iv_tcw_dtr,
1091 .init = crypt_iv_tcw_init,
1092 .wipe = crypt_iv_tcw_wipe,
1093 .generator = crypt_iv_tcw_gen,
1094 .post = crypt_iv_tcw_post
1097 static const struct crypt_iv_operations crypt_iv_random_ops = {
1098 .generator = crypt_iv_random_gen
1101 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1102 .ctr = crypt_iv_eboiv_ctr,
1103 .generator = crypt_iv_eboiv_gen
1106 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1107 .ctr = crypt_iv_elephant_ctr,
1108 .dtr = crypt_iv_elephant_dtr,
1109 .init = crypt_iv_elephant_init,
1110 .wipe = crypt_iv_elephant_wipe,
1111 .generator = crypt_iv_elephant_gen,
1112 .post = crypt_iv_elephant_post
1116 * Integrity extensions
1118 static bool crypt_integrity_aead(struct crypt_config *cc)
1120 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1123 static bool crypt_integrity_hmac(struct crypt_config *cc)
1125 return crypt_integrity_aead(cc) && cc->key_mac_size;
1128 /* Get sg containing data */
1129 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1130 struct scatterlist *sg)
1132 if (unlikely(crypt_integrity_aead(cc)))
1138 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1140 struct bio_integrity_payload *bip;
1141 unsigned int tag_len;
1144 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1147 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1149 return PTR_ERR(bip);
1151 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1153 bip->bip_iter.bi_size = tag_len;
1154 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1156 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1157 tag_len, offset_in_page(io->integrity_metadata));
1158 if (unlikely(ret != tag_len))
1164 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1166 #ifdef CONFIG_BLK_DEV_INTEGRITY
1167 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1168 struct mapped_device *md = dm_table_get_md(ti->table);
1170 /* From now we require underlying device with our integrity profile */
1171 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1172 ti->error = "Integrity profile not supported.";
1176 if (bi->tag_size != cc->on_disk_tag_size ||
1177 bi->tuple_size != cc->on_disk_tag_size) {
1178 ti->error = "Integrity profile tag size mismatch.";
1181 if (1 << bi->interval_exp != cc->sector_size) {
1182 ti->error = "Integrity profile sector size mismatch.";
1186 if (crypt_integrity_aead(cc)) {
1187 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1188 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1189 cc->integrity_tag_size, cc->integrity_iv_size);
1191 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1192 ti->error = "Integrity AEAD auth tag size is not supported.";
1195 } else if (cc->integrity_iv_size)
1196 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1197 cc->integrity_iv_size);
1199 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1200 ti->error = "Not enough space for integrity tag in the profile.";
1206 ti->error = "Integrity profile not supported.";
1211 static void crypt_convert_init(struct crypt_config *cc,
1212 struct convert_context *ctx,
1213 struct bio *bio_out, struct bio *bio_in,
1216 ctx->bio_in = bio_in;
1217 ctx->bio_out = bio_out;
1219 ctx->iter_in = bio_in->bi_iter;
1221 ctx->iter_out = bio_out->bi_iter;
1222 ctx->cc_sector = sector + cc->iv_offset;
1223 init_completion(&ctx->restart);
1226 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1229 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1232 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1234 return (void *)((char *)dmreq - cc->dmreq_start);
1237 static u8 *iv_of_dmreq(struct crypt_config *cc,
1238 struct dm_crypt_request *dmreq)
1240 if (crypt_integrity_aead(cc))
1241 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1242 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1244 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1245 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1248 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1249 struct dm_crypt_request *dmreq)
1251 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1254 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1255 struct dm_crypt_request *dmreq)
1257 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1258 return (__le64 *) ptr;
1261 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1262 struct dm_crypt_request *dmreq)
1264 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1265 cc->iv_size + sizeof(uint64_t);
1266 return (unsigned int*)ptr;
1269 static void *tag_from_dmreq(struct crypt_config *cc,
1270 struct dm_crypt_request *dmreq)
1272 struct convert_context *ctx = dmreq->ctx;
1273 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1275 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1276 cc->on_disk_tag_size];
1279 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1280 struct dm_crypt_request *dmreq)
1282 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1285 static int crypt_convert_block_aead(struct crypt_config *cc,
1286 struct convert_context *ctx,
1287 struct aead_request *req,
1288 unsigned int tag_offset)
1290 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1291 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1292 struct dm_crypt_request *dmreq;
1293 u8 *iv, *org_iv, *tag_iv, *tag;
1297 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1299 /* Reject unexpected unaligned bio. */
1300 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1303 dmreq = dmreq_of_req(cc, req);
1304 dmreq->iv_sector = ctx->cc_sector;
1305 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1306 dmreq->iv_sector >>= cc->sector_shift;
1309 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1311 sector = org_sector_of_dmreq(cc, dmreq);
1312 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1314 iv = iv_of_dmreq(cc, dmreq);
1315 org_iv = org_iv_of_dmreq(cc, dmreq);
1316 tag = tag_from_dmreq(cc, dmreq);
1317 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1320 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1321 * | (authenticated) | (auth+encryption) | |
1322 * | sector_LE | IV | sector in/out | tag in/out |
1324 sg_init_table(dmreq->sg_in, 4);
1325 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1326 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1327 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1328 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1330 sg_init_table(dmreq->sg_out, 4);
1331 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1332 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1333 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1334 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1336 if (cc->iv_gen_ops) {
1337 /* For READs use IV stored in integrity metadata */
1338 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1339 memcpy(org_iv, tag_iv, cc->iv_size);
1341 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1344 /* Store generated IV in integrity metadata */
1345 if (cc->integrity_iv_size)
1346 memcpy(tag_iv, org_iv, cc->iv_size);
1348 /* Working copy of IV, to be modified in crypto API */
1349 memcpy(iv, org_iv, cc->iv_size);
1352 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1353 if (bio_data_dir(ctx->bio_in) == WRITE) {
1354 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1355 cc->sector_size, iv);
1356 r = crypto_aead_encrypt(req);
1357 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1358 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1359 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1361 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1362 cc->sector_size + cc->integrity_tag_size, iv);
1363 r = crypto_aead_decrypt(req);
1366 if (r == -EBADMSG) {
1367 sector_t s = le64_to_cpu(*sector);
1369 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1370 ctx->bio_in->bi_bdev, s);
1371 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1375 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1376 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1378 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1379 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1384 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1385 struct convert_context *ctx,
1386 struct skcipher_request *req,
1387 unsigned int tag_offset)
1389 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1390 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1391 struct scatterlist *sg_in, *sg_out;
1392 struct dm_crypt_request *dmreq;
1393 u8 *iv, *org_iv, *tag_iv;
1397 /* Reject unexpected unaligned bio. */
1398 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1401 dmreq = dmreq_of_req(cc, req);
1402 dmreq->iv_sector = ctx->cc_sector;
1403 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1404 dmreq->iv_sector >>= cc->sector_shift;
1407 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1409 iv = iv_of_dmreq(cc, dmreq);
1410 org_iv = org_iv_of_dmreq(cc, dmreq);
1411 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1413 sector = org_sector_of_dmreq(cc, dmreq);
1414 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1416 /* For skcipher we use only the first sg item */
1417 sg_in = &dmreq->sg_in[0];
1418 sg_out = &dmreq->sg_out[0];
1420 sg_init_table(sg_in, 1);
1421 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1423 sg_init_table(sg_out, 1);
1424 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1426 if (cc->iv_gen_ops) {
1427 /* For READs use IV stored in integrity metadata */
1428 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1429 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1431 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1434 /* Data can be already preprocessed in generator */
1435 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1437 /* Store generated IV in integrity metadata */
1438 if (cc->integrity_iv_size)
1439 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1441 /* Working copy of IV, to be modified in crypto API */
1442 memcpy(iv, org_iv, cc->iv_size);
1445 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1447 if (bio_data_dir(ctx->bio_in) == WRITE)
1448 r = crypto_skcipher_encrypt(req);
1450 r = crypto_skcipher_decrypt(req);
1452 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1453 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1455 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1456 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1461 static void kcryptd_async_done(struct crypto_async_request *async_req,
1464 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1465 struct convert_context *ctx)
1467 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1470 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1475 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1478 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1479 * requests if driver request queue is full.
1481 skcipher_request_set_callback(ctx->r.req,
1482 CRYPTO_TFM_REQ_MAY_BACKLOG,
1483 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1488 static int crypt_alloc_req_aead(struct crypt_config *cc,
1489 struct convert_context *ctx)
1491 if (!ctx->r.req_aead) {
1492 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1493 if (!ctx->r.req_aead)
1497 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1500 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1501 * requests if driver request queue is full.
1503 aead_request_set_callback(ctx->r.req_aead,
1504 CRYPTO_TFM_REQ_MAY_BACKLOG,
1505 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1510 static int crypt_alloc_req(struct crypt_config *cc,
1511 struct convert_context *ctx)
1513 if (crypt_integrity_aead(cc))
1514 return crypt_alloc_req_aead(cc, ctx);
1516 return crypt_alloc_req_skcipher(cc, ctx);
1519 static void crypt_free_req_skcipher(struct crypt_config *cc,
1520 struct skcipher_request *req, struct bio *base_bio)
1522 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1524 if ((struct skcipher_request *)(io + 1) != req)
1525 mempool_free(req, &cc->req_pool);
1528 static void crypt_free_req_aead(struct crypt_config *cc,
1529 struct aead_request *req, struct bio *base_bio)
1531 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1533 if ((struct aead_request *)(io + 1) != req)
1534 mempool_free(req, &cc->req_pool);
1537 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1539 if (crypt_integrity_aead(cc))
1540 crypt_free_req_aead(cc, req, base_bio);
1542 crypt_free_req_skcipher(cc, req, base_bio);
1546 * Encrypt / decrypt data from one bio to another one (can be the same one)
1548 static blk_status_t crypt_convert(struct crypt_config *cc,
1549 struct convert_context *ctx, bool atomic, bool reset_pending)
1551 unsigned int tag_offset = 0;
1552 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1556 * if reset_pending is set we are dealing with the bio for the first time,
1557 * else we're continuing to work on the previous bio, so don't mess with
1558 * the cc_pending counter
1561 atomic_set(&ctx->cc_pending, 1);
1563 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1565 r = crypt_alloc_req(cc, ctx);
1567 complete(&ctx->restart);
1568 return BLK_STS_DEV_RESOURCE;
1571 atomic_inc(&ctx->cc_pending);
1573 if (crypt_integrity_aead(cc))
1574 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1576 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1580 * The request was queued by a crypto driver
1581 * but the driver request queue is full, let's wait.
1584 if (in_interrupt()) {
1585 if (try_wait_for_completion(&ctx->restart)) {
1587 * we don't have to block to wait for completion,
1592 * we can't wait for completion without blocking
1593 * exit and continue processing in a workqueue
1596 ctx->cc_sector += sector_step;
1598 return BLK_STS_DEV_RESOURCE;
1601 wait_for_completion(&ctx->restart);
1603 reinit_completion(&ctx->restart);
1606 * The request is queued and processed asynchronously,
1607 * completion function kcryptd_async_done() will be called.
1611 ctx->cc_sector += sector_step;
1615 * The request was already processed (synchronously).
1618 atomic_dec(&ctx->cc_pending);
1619 ctx->cc_sector += sector_step;
1625 * There was a data integrity error.
1628 atomic_dec(&ctx->cc_pending);
1629 return BLK_STS_PROTECTION;
1631 * There was an error while processing the request.
1634 atomic_dec(&ctx->cc_pending);
1635 return BLK_STS_IOERR;
1642 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1645 * Generate a new unfragmented bio with the given size
1646 * This should never violate the device limitations (but only because
1647 * max_segment_size is being constrained to PAGE_SIZE).
1649 * This function may be called concurrently. If we allocate from the mempool
1650 * concurrently, there is a possibility of deadlock. For example, if we have
1651 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1652 * the mempool concurrently, it may deadlock in a situation where both processes
1653 * have allocated 128 pages and the mempool is exhausted.
1655 * In order to avoid this scenario we allocate the pages under a mutex.
1657 * In order to not degrade performance with excessive locking, we try
1658 * non-blocking allocations without a mutex first but on failure we fallback
1659 * to blocking allocations with a mutex.
1661 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1663 struct crypt_config *cc = io->cc;
1665 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1666 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1667 unsigned i, len, remaining_size;
1671 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1672 mutex_lock(&cc->bio_alloc_lock);
1674 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1676 clone->bi_private = io;
1677 clone->bi_end_io = crypt_endio;
1679 remaining_size = size;
1681 for (i = 0; i < nr_iovecs; i++) {
1682 page = mempool_alloc(&cc->page_pool, gfp_mask);
1684 crypt_free_buffer_pages(cc, clone);
1686 gfp_mask |= __GFP_DIRECT_RECLAIM;
1690 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1692 bio_add_page(clone, page, len, 0);
1694 remaining_size -= len;
1697 /* Allocate space for integrity tags */
1698 if (dm_crypt_integrity_io_alloc(io, clone)) {
1699 crypt_free_buffer_pages(cc, clone);
1704 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1705 mutex_unlock(&cc->bio_alloc_lock);
1710 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1713 struct bvec_iter_all iter_all;
1715 bio_for_each_segment_all(bv, clone, iter_all) {
1716 BUG_ON(!bv->bv_page);
1717 mempool_free(bv->bv_page, &cc->page_pool);
1721 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1722 struct bio *bio, sector_t sector)
1726 io->sector = sector;
1728 io->ctx.r.req = NULL;
1729 io->integrity_metadata = NULL;
1730 io->integrity_metadata_from_pool = false;
1731 atomic_set(&io->io_pending, 0);
1734 static void crypt_inc_pending(struct dm_crypt_io *io)
1736 atomic_inc(&io->io_pending);
1739 static void kcryptd_io_bio_endio(struct work_struct *work)
1741 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1742 bio_endio(io->base_bio);
1746 * One of the bios was finished. Check for completion of
1747 * the whole request and correctly clean up the buffer.
1749 static void crypt_dec_pending(struct dm_crypt_io *io)
1751 struct crypt_config *cc = io->cc;
1752 struct bio *base_bio = io->base_bio;
1753 blk_status_t error = io->error;
1755 if (!atomic_dec_and_test(&io->io_pending))
1759 crypt_free_req(cc, io->ctx.r.req, base_bio);
1761 if (unlikely(io->integrity_metadata_from_pool))
1762 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1764 kfree(io->integrity_metadata);
1766 base_bio->bi_status = error;
1769 * If we are running this function from our tasklet,
1770 * we can't call bio_endio() here, because it will call
1771 * clone_endio() from dm.c, which in turn will
1772 * free the current struct dm_crypt_io structure with
1773 * our tasklet. In this case we need to delay bio_endio()
1774 * execution to after the tasklet is done and dequeued.
1776 if (tasklet_trylock(&io->tasklet)) {
1777 tasklet_unlock(&io->tasklet);
1778 bio_endio(base_bio);
1782 INIT_WORK(&io->work, kcryptd_io_bio_endio);
1783 queue_work(cc->io_queue, &io->work);
1787 * kcryptd/kcryptd_io:
1789 * Needed because it would be very unwise to do decryption in an
1790 * interrupt context.
1792 * kcryptd performs the actual encryption or decryption.
1794 * kcryptd_io performs the IO submission.
1796 * They must be separated as otherwise the final stages could be
1797 * starved by new requests which can block in the first stages due
1798 * to memory allocation.
1800 * The work is done per CPU global for all dm-crypt instances.
1801 * They should not depend on each other and do not block.
1803 static void crypt_endio(struct bio *clone)
1805 struct dm_crypt_io *io = clone->bi_private;
1806 struct crypt_config *cc = io->cc;
1807 unsigned rw = bio_data_dir(clone);
1811 * free the processed pages
1814 crypt_free_buffer_pages(cc, clone);
1816 error = clone->bi_status;
1819 if (rw == READ && !error) {
1820 kcryptd_queue_crypt(io);
1824 if (unlikely(error))
1827 crypt_dec_pending(io);
1830 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1832 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1834 struct crypt_config *cc = io->cc;
1838 * We need the original biovec array in order to decrypt the whole bio
1839 * data *afterwards* -- thanks to immutable biovecs we don't need to
1840 * worry about the block layer modifying the biovec array; so leverage
1841 * bio_alloc_clone().
1843 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1846 clone->bi_private = io;
1847 clone->bi_end_io = crypt_endio;
1849 crypt_inc_pending(io);
1851 clone->bi_iter.bi_sector = cc->start + io->sector;
1853 if (dm_crypt_integrity_io_alloc(io, clone)) {
1854 crypt_dec_pending(io);
1859 dm_submit_bio_remap(io->base_bio, clone);
1863 static void kcryptd_io_read_work(struct work_struct *work)
1865 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1867 crypt_inc_pending(io);
1868 if (kcryptd_io_read(io, GFP_NOIO))
1869 io->error = BLK_STS_RESOURCE;
1870 crypt_dec_pending(io);
1873 static void kcryptd_queue_read(struct dm_crypt_io *io)
1875 struct crypt_config *cc = io->cc;
1877 INIT_WORK(&io->work, kcryptd_io_read_work);
1878 queue_work(cc->io_queue, &io->work);
1881 static void kcryptd_io_write(struct dm_crypt_io *io)
1883 struct bio *clone = io->ctx.bio_out;
1885 dm_submit_bio_remap(io->base_bio, clone);
1888 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1890 static int dmcrypt_write(void *data)
1892 struct crypt_config *cc = data;
1893 struct dm_crypt_io *io;
1896 struct rb_root write_tree;
1897 struct blk_plug plug;
1899 spin_lock_irq(&cc->write_thread_lock);
1902 if (!RB_EMPTY_ROOT(&cc->write_tree))
1905 set_current_state(TASK_INTERRUPTIBLE);
1907 spin_unlock_irq(&cc->write_thread_lock);
1909 if (unlikely(kthread_should_stop())) {
1910 set_current_state(TASK_RUNNING);
1916 set_current_state(TASK_RUNNING);
1917 spin_lock_irq(&cc->write_thread_lock);
1918 goto continue_locked;
1921 write_tree = cc->write_tree;
1922 cc->write_tree = RB_ROOT;
1923 spin_unlock_irq(&cc->write_thread_lock);
1925 BUG_ON(rb_parent(write_tree.rb_node));
1928 * Note: we cannot walk the tree here with rb_next because
1929 * the structures may be freed when kcryptd_io_write is called.
1931 blk_start_plug(&plug);
1933 io = crypt_io_from_node(rb_first(&write_tree));
1934 rb_erase(&io->rb_node, &write_tree);
1935 kcryptd_io_write(io);
1936 } while (!RB_EMPTY_ROOT(&write_tree));
1937 blk_finish_plug(&plug);
1942 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1944 struct bio *clone = io->ctx.bio_out;
1945 struct crypt_config *cc = io->cc;
1946 unsigned long flags;
1948 struct rb_node **rbp, *parent;
1950 if (unlikely(io->error)) {
1951 crypt_free_buffer_pages(cc, clone);
1953 crypt_dec_pending(io);
1957 /* crypt_convert should have filled the clone bio */
1958 BUG_ON(io->ctx.iter_out.bi_size);
1960 clone->bi_iter.bi_sector = cc->start + io->sector;
1962 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1963 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1964 dm_submit_bio_remap(io->base_bio, clone);
1968 spin_lock_irqsave(&cc->write_thread_lock, flags);
1969 if (RB_EMPTY_ROOT(&cc->write_tree))
1970 wake_up_process(cc->write_thread);
1971 rbp = &cc->write_tree.rb_node;
1973 sector = io->sector;
1976 if (sector < crypt_io_from_node(parent)->sector)
1977 rbp = &(*rbp)->rb_left;
1979 rbp = &(*rbp)->rb_right;
1981 rb_link_node(&io->rb_node, parent, rbp);
1982 rb_insert_color(&io->rb_node, &cc->write_tree);
1983 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1986 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1987 struct convert_context *ctx)
1990 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1994 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1995 * constraints so they do not need to be issued inline by
1996 * kcryptd_crypt_write_convert().
1998 switch (bio_op(ctx->bio_in)) {
2000 case REQ_OP_WRITE_ZEROES:
2007 static void kcryptd_crypt_write_continue(struct work_struct *work)
2009 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2010 struct crypt_config *cc = io->cc;
2011 struct convert_context *ctx = &io->ctx;
2013 sector_t sector = io->sector;
2016 wait_for_completion(&ctx->restart);
2017 reinit_completion(&ctx->restart);
2019 r = crypt_convert(cc, &io->ctx, true, false);
2022 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2023 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2024 /* Wait for completion signaled by kcryptd_async_done() */
2025 wait_for_completion(&ctx->restart);
2029 /* Encryption was already finished, submit io now */
2030 if (crypt_finished) {
2031 kcryptd_crypt_write_io_submit(io, 0);
2032 io->sector = sector;
2035 crypt_dec_pending(io);
2038 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2040 struct crypt_config *cc = io->cc;
2041 struct convert_context *ctx = &io->ctx;
2044 sector_t sector = io->sector;
2048 * Prevent io from disappearing until this function completes.
2050 crypt_inc_pending(io);
2051 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2053 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2054 if (unlikely(!clone)) {
2055 io->error = BLK_STS_IOERR;
2059 io->ctx.bio_out = clone;
2060 io->ctx.iter_out = clone->bi_iter;
2062 sector += bio_sectors(clone);
2064 crypt_inc_pending(io);
2065 r = crypt_convert(cc, ctx,
2066 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2068 * Crypto API backlogged the request, because its queue was full
2069 * and we're in softirq context, so continue from a workqueue
2070 * (TODO: is it actually possible to be in softirq in the write path?)
2072 if (r == BLK_STS_DEV_RESOURCE) {
2073 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2074 queue_work(cc->crypt_queue, &io->work);
2079 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2080 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2081 /* Wait for completion signaled by kcryptd_async_done() */
2082 wait_for_completion(&ctx->restart);
2086 /* Encryption was already finished, submit io now */
2087 if (crypt_finished) {
2088 kcryptd_crypt_write_io_submit(io, 0);
2089 io->sector = sector;
2093 crypt_dec_pending(io);
2096 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2098 crypt_dec_pending(io);
2101 static void kcryptd_crypt_read_continue(struct work_struct *work)
2103 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2104 struct crypt_config *cc = io->cc;
2107 wait_for_completion(&io->ctx.restart);
2108 reinit_completion(&io->ctx.restart);
2110 r = crypt_convert(cc, &io->ctx, true, false);
2114 if (atomic_dec_and_test(&io->ctx.cc_pending))
2115 kcryptd_crypt_read_done(io);
2117 crypt_dec_pending(io);
2120 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2122 struct crypt_config *cc = io->cc;
2125 crypt_inc_pending(io);
2127 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2130 r = crypt_convert(cc, &io->ctx,
2131 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2133 * Crypto API backlogged the request, because its queue was full
2134 * and we're in softirq context, so continue from a workqueue
2136 if (r == BLK_STS_DEV_RESOURCE) {
2137 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2138 queue_work(cc->crypt_queue, &io->work);
2144 if (atomic_dec_and_test(&io->ctx.cc_pending))
2145 kcryptd_crypt_read_done(io);
2147 crypt_dec_pending(io);
2150 static void kcryptd_async_done(struct crypto_async_request *async_req,
2153 struct dm_crypt_request *dmreq = async_req->data;
2154 struct convert_context *ctx = dmreq->ctx;
2155 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2156 struct crypt_config *cc = io->cc;
2159 * A request from crypto driver backlog is going to be processed now,
2160 * finish the completion and continue in crypt_convert().
2161 * (Callback will be called for the second time for this request.)
2163 if (error == -EINPROGRESS) {
2164 complete(&ctx->restart);
2168 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2169 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2171 if (error == -EBADMSG) {
2172 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2174 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2175 ctx->bio_in->bi_bdev, s);
2176 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2178 io->error = BLK_STS_PROTECTION;
2179 } else if (error < 0)
2180 io->error = BLK_STS_IOERR;
2182 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2184 if (!atomic_dec_and_test(&ctx->cc_pending))
2188 * The request is fully completed: for inline writes, let
2189 * kcryptd_crypt_write_convert() do the IO submission.
2191 if (bio_data_dir(io->base_bio) == READ) {
2192 kcryptd_crypt_read_done(io);
2196 if (kcryptd_crypt_write_inline(cc, ctx)) {
2197 complete(&ctx->restart);
2201 kcryptd_crypt_write_io_submit(io, 1);
2204 static void kcryptd_crypt(struct work_struct *work)
2206 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2208 if (bio_data_dir(io->base_bio) == READ)
2209 kcryptd_crypt_read_convert(io);
2211 kcryptd_crypt_write_convert(io);
2214 static void kcryptd_crypt_tasklet(unsigned long work)
2216 kcryptd_crypt((struct work_struct *)work);
2219 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2221 struct crypt_config *cc = io->cc;
2223 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2224 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2226 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2227 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2228 * it is being executed with irqs disabled.
2230 if (in_hardirq() || irqs_disabled()) {
2231 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2232 tasklet_schedule(&io->tasklet);
2236 kcryptd_crypt(&io->work);
2240 INIT_WORK(&io->work, kcryptd_crypt);
2241 queue_work(cc->crypt_queue, &io->work);
2244 static void crypt_free_tfms_aead(struct crypt_config *cc)
2246 if (!cc->cipher_tfm.tfms_aead)
2249 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2250 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2251 cc->cipher_tfm.tfms_aead[0] = NULL;
2254 kfree(cc->cipher_tfm.tfms_aead);
2255 cc->cipher_tfm.tfms_aead = NULL;
2258 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2262 if (!cc->cipher_tfm.tfms)
2265 for (i = 0; i < cc->tfms_count; i++)
2266 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2267 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2268 cc->cipher_tfm.tfms[i] = NULL;
2271 kfree(cc->cipher_tfm.tfms);
2272 cc->cipher_tfm.tfms = NULL;
2275 static void crypt_free_tfms(struct crypt_config *cc)
2277 if (crypt_integrity_aead(cc))
2278 crypt_free_tfms_aead(cc);
2280 crypt_free_tfms_skcipher(cc);
2283 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2288 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2289 sizeof(struct crypto_skcipher *),
2291 if (!cc->cipher_tfm.tfms)
2294 for (i = 0; i < cc->tfms_count; i++) {
2295 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2296 CRYPTO_ALG_ALLOCATES_MEMORY);
2297 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2298 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2299 crypt_free_tfms(cc);
2305 * dm-crypt performance can vary greatly depending on which crypto
2306 * algorithm implementation is used. Help people debug performance
2307 * problems by logging the ->cra_driver_name.
2309 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2310 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2314 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2318 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2319 if (!cc->cipher_tfm.tfms)
2322 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2323 CRYPTO_ALG_ALLOCATES_MEMORY);
2324 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2325 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2326 crypt_free_tfms(cc);
2330 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2331 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2335 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2337 if (crypt_integrity_aead(cc))
2338 return crypt_alloc_tfms_aead(cc, ciphermode);
2340 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2343 static unsigned crypt_subkey_size(struct crypt_config *cc)
2345 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2348 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2350 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2354 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2355 * the key must be for some reason in special format.
2356 * This funcion converts cc->key to this special format.
2358 static void crypt_copy_authenckey(char *p, const void *key,
2359 unsigned enckeylen, unsigned authkeylen)
2361 struct crypto_authenc_key_param *param;
2364 rta = (struct rtattr *)p;
2365 param = RTA_DATA(rta);
2366 param->enckeylen = cpu_to_be32(enckeylen);
2367 rta->rta_len = RTA_LENGTH(sizeof(*param));
2368 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2369 p += RTA_SPACE(sizeof(*param));
2370 memcpy(p, key + enckeylen, authkeylen);
2372 memcpy(p, key, enckeylen);
2375 static int crypt_setkey(struct crypt_config *cc)
2377 unsigned subkey_size;
2380 /* Ignore extra keys (which are used for IV etc) */
2381 subkey_size = crypt_subkey_size(cc);
2383 if (crypt_integrity_hmac(cc)) {
2384 if (subkey_size < cc->key_mac_size)
2387 crypt_copy_authenckey(cc->authenc_key, cc->key,
2388 subkey_size - cc->key_mac_size,
2392 for (i = 0; i < cc->tfms_count; i++) {
2393 if (crypt_integrity_hmac(cc))
2394 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2395 cc->authenc_key, crypt_authenckey_size(cc));
2396 else if (crypt_integrity_aead(cc))
2397 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2398 cc->key + (i * subkey_size),
2401 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2402 cc->key + (i * subkey_size),
2408 if (crypt_integrity_hmac(cc))
2409 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2416 static bool contains_whitespace(const char *str)
2419 if (isspace(*str++))
2424 static int set_key_user(struct crypt_config *cc, struct key *key)
2426 const struct user_key_payload *ukp;
2428 ukp = user_key_payload_locked(key);
2430 return -EKEYREVOKED;
2432 if (cc->key_size != ukp->datalen)
2435 memcpy(cc->key, ukp->data, cc->key_size);
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);
2456 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2458 const struct trusted_key_payload *tkp;
2460 tkp = key->payload.data[0];
2462 return -EKEYREVOKED;
2464 if (cc->key_size != tkp->key_len)
2467 memcpy(cc->key, tkp->key, cc->key_size);
2472 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2474 char *new_key_string, *key_desc;
2476 struct key_type *type;
2478 int (*set_key)(struct crypt_config *cc, struct key *key);
2481 * Reject key_string with whitespace. dm core currently lacks code for
2482 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2484 if (contains_whitespace(key_string)) {
2485 DMERR("whitespace chars not allowed in key string");
2489 /* look for next ':' separating key_type from key_description */
2490 key_desc = strpbrk(key_string, ":");
2491 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2494 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2495 type = &key_type_logon;
2496 set_key = set_key_user;
2497 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2498 type = &key_type_user;
2499 set_key = set_key_user;
2500 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2501 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2502 type = &key_type_encrypted;
2503 set_key = set_key_encrypted;
2504 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2505 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2506 type = &key_type_trusted;
2507 set_key = set_key_trusted;
2512 new_key_string = kstrdup(key_string, GFP_KERNEL);
2513 if (!new_key_string)
2516 key = request_key(type, key_desc + 1, NULL);
2518 kfree_sensitive(new_key_string);
2519 return PTR_ERR(key);
2522 down_read(&key->sem);
2524 ret = set_key(cc, key);
2528 kfree_sensitive(new_key_string);
2535 /* clear the flag since following operations may invalidate previously valid key */
2536 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2538 ret = crypt_setkey(cc);
2541 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2542 kfree_sensitive(cc->key_string);
2543 cc->key_string = new_key_string;
2545 kfree_sensitive(new_key_string);
2550 static int get_key_size(char **key_string)
2555 if (*key_string[0] != ':')
2556 return strlen(*key_string) >> 1;
2558 /* look for next ':' in key string */
2559 colon = strpbrk(*key_string + 1, ":");
2563 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2566 *key_string = colon;
2568 /* remaining key string should be :<logon|user>:<key_desc> */
2575 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2580 static int get_key_size(char **key_string)
2582 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2585 #endif /* CONFIG_KEYS */
2587 static int crypt_set_key(struct crypt_config *cc, char *key)
2590 int key_string_len = strlen(key);
2592 /* Hyphen (which gives a key_size of zero) means there is no key. */
2593 if (!cc->key_size && strcmp(key, "-"))
2596 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2597 if (key[0] == ':') {
2598 r = crypt_set_keyring_key(cc, key + 1);
2602 /* clear the flag since following operations may invalidate previously valid key */
2603 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2605 /* wipe references to any kernel keyring key */
2606 kfree_sensitive(cc->key_string);
2607 cc->key_string = NULL;
2609 /* Decode key from its hex representation. */
2610 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2613 r = crypt_setkey(cc);
2615 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2618 /* Hex key string not needed after here, so wipe it. */
2619 memset(key, '0', key_string_len);
2624 static int crypt_wipe_key(struct crypt_config *cc)
2628 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2629 get_random_bytes(&cc->key, cc->key_size);
2631 /* Wipe IV private keys */
2632 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2633 r = cc->iv_gen_ops->wipe(cc);
2638 kfree_sensitive(cc->key_string);
2639 cc->key_string = NULL;
2640 r = crypt_setkey(cc);
2641 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2646 static void crypt_calculate_pages_per_client(void)
2648 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2650 if (!dm_crypt_clients_n)
2653 pages /= dm_crypt_clients_n;
2654 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2655 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2656 dm_crypt_pages_per_client = pages;
2659 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2661 struct crypt_config *cc = pool_data;
2665 * Note, percpu_counter_read_positive() may over (and under) estimate
2666 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2667 * but avoids potential spinlock contention of an exact result.
2669 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2670 likely(gfp_mask & __GFP_NORETRY))
2673 page = alloc_page(gfp_mask);
2674 if (likely(page != NULL))
2675 percpu_counter_add(&cc->n_allocated_pages, 1);
2680 static void crypt_page_free(void *page, void *pool_data)
2682 struct crypt_config *cc = pool_data;
2685 percpu_counter_sub(&cc->n_allocated_pages, 1);
2688 static void crypt_dtr(struct dm_target *ti)
2690 struct crypt_config *cc = ti->private;
2697 if (cc->write_thread)
2698 kthread_stop(cc->write_thread);
2701 destroy_workqueue(cc->io_queue);
2702 if (cc->crypt_queue)
2703 destroy_workqueue(cc->crypt_queue);
2705 crypt_free_tfms(cc);
2707 bioset_exit(&cc->bs);
2709 mempool_exit(&cc->page_pool);
2710 mempool_exit(&cc->req_pool);
2711 mempool_exit(&cc->tag_pool);
2713 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2714 percpu_counter_destroy(&cc->n_allocated_pages);
2716 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2717 cc->iv_gen_ops->dtr(cc);
2720 dm_put_device(ti, cc->dev);
2722 kfree_sensitive(cc->cipher_string);
2723 kfree_sensitive(cc->key_string);
2724 kfree_sensitive(cc->cipher_auth);
2725 kfree_sensitive(cc->authenc_key);
2727 mutex_destroy(&cc->bio_alloc_lock);
2729 /* Must zero key material before freeing */
2730 kfree_sensitive(cc);
2732 spin_lock(&dm_crypt_clients_lock);
2733 WARN_ON(!dm_crypt_clients_n);
2734 dm_crypt_clients_n--;
2735 crypt_calculate_pages_per_client();
2736 spin_unlock(&dm_crypt_clients_lock);
2738 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2741 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2743 struct crypt_config *cc = ti->private;
2745 if (crypt_integrity_aead(cc))
2746 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2748 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2751 /* at least a 64 bit sector number should fit in our buffer */
2752 cc->iv_size = max(cc->iv_size,
2753 (unsigned int)(sizeof(u64) / sizeof(u8)));
2755 DMWARN("Selected cipher does not support IVs");
2759 /* Choose ivmode, see comments at iv code. */
2761 cc->iv_gen_ops = NULL;
2762 else if (strcmp(ivmode, "plain") == 0)
2763 cc->iv_gen_ops = &crypt_iv_plain_ops;
2764 else if (strcmp(ivmode, "plain64") == 0)
2765 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2766 else if (strcmp(ivmode, "plain64be") == 0)
2767 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2768 else if (strcmp(ivmode, "essiv") == 0)
2769 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2770 else if (strcmp(ivmode, "benbi") == 0)
2771 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2772 else if (strcmp(ivmode, "null") == 0)
2773 cc->iv_gen_ops = &crypt_iv_null_ops;
2774 else if (strcmp(ivmode, "eboiv") == 0)
2775 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2776 else if (strcmp(ivmode, "elephant") == 0) {
2777 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2779 cc->key_extra_size = cc->key_size / 2;
2780 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2782 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2783 } else if (strcmp(ivmode, "lmk") == 0) {
2784 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2786 * Version 2 and 3 is recognised according
2787 * to length of provided multi-key string.
2788 * If present (version 3), last key is used as IV seed.
2789 * All keys (including IV seed) are always the same size.
2791 if (cc->key_size % cc->key_parts) {
2793 cc->key_extra_size = cc->key_size / cc->key_parts;
2795 } else if (strcmp(ivmode, "tcw") == 0) {
2796 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2797 cc->key_parts += 2; /* IV + whitening */
2798 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2799 } else if (strcmp(ivmode, "random") == 0) {
2800 cc->iv_gen_ops = &crypt_iv_random_ops;
2801 /* Need storage space in integrity fields. */
2802 cc->integrity_iv_size = cc->iv_size;
2804 ti->error = "Invalid IV mode";
2812 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2813 * The HMAC is needed to calculate tag size (HMAC digest size).
2814 * This should be probably done by crypto-api calls (once available...)
2816 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2818 char *start, *end, *mac_alg = NULL;
2819 struct crypto_ahash *mac;
2821 if (!strstarts(cipher_api, "authenc("))
2824 start = strchr(cipher_api, '(');
2825 end = strchr(cipher_api, ',');
2826 if (!start || !end || ++start > end)
2829 mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2832 strncpy(mac_alg, start, end - start);
2834 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2838 return PTR_ERR(mac);
2840 cc->key_mac_size = crypto_ahash_digestsize(mac);
2841 crypto_free_ahash(mac);
2843 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2844 if (!cc->authenc_key)
2850 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2851 char **ivmode, char **ivopts)
2853 struct crypt_config *cc = ti->private;
2854 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2860 * New format (capi: prefix)
2861 * capi:cipher_api_spec-iv:ivopts
2863 tmp = &cipher_in[strlen("capi:")];
2865 /* Separate IV options if present, it can contain another '-' in hash name */
2866 *ivopts = strrchr(tmp, ':');
2872 *ivmode = strrchr(tmp, '-');
2877 /* The rest is crypto API spec */
2880 /* Alloc AEAD, can be used only in new format. */
2881 if (crypt_integrity_aead(cc)) {
2882 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2884 ti->error = "Invalid AEAD cipher spec";
2889 if (*ivmode && !strcmp(*ivmode, "lmk"))
2890 cc->tfms_count = 64;
2892 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2894 ti->error = "Digest algorithm missing for ESSIV mode";
2897 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2898 cipher_api, *ivopts);
2899 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2900 ti->error = "Cannot allocate cipher string";
2906 cc->key_parts = cc->tfms_count;
2908 /* Allocate cipher */
2909 ret = crypt_alloc_tfms(cc, cipher_api);
2911 ti->error = "Error allocating crypto tfm";
2915 if (crypt_integrity_aead(cc))
2916 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2918 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2923 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2924 char **ivmode, char **ivopts)
2926 struct crypt_config *cc = ti->private;
2927 char *tmp, *cipher, *chainmode, *keycount;
2928 char *cipher_api = NULL;
2932 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2933 ti->error = "Bad cipher specification";
2938 * Legacy dm-crypt cipher specification
2939 * cipher[:keycount]-mode-iv:ivopts
2942 keycount = strsep(&tmp, "-");
2943 cipher = strsep(&keycount, ":");
2947 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2948 !is_power_of_2(cc->tfms_count)) {
2949 ti->error = "Bad cipher key count specification";
2952 cc->key_parts = cc->tfms_count;
2954 chainmode = strsep(&tmp, "-");
2955 *ivmode = strsep(&tmp, ":");
2959 * For compatibility with the original dm-crypt mapping format, if
2960 * only the cipher name is supplied, use cbc-plain.
2962 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2967 if (strcmp(chainmode, "ecb") && !*ivmode) {
2968 ti->error = "IV mechanism required";
2972 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2976 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2978 ti->error = "Digest algorithm missing for ESSIV mode";
2982 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2983 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2985 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2986 "%s(%s)", chainmode, cipher);
2988 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2993 /* Allocate cipher */
2994 ret = crypt_alloc_tfms(cc, cipher_api);
2996 ti->error = "Error allocating crypto tfm";
3004 ti->error = "Cannot allocate cipher strings";
3008 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3010 struct crypt_config *cc = ti->private;
3011 char *ivmode = NULL, *ivopts = NULL;
3014 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3015 if (!cc->cipher_string) {
3016 ti->error = "Cannot allocate cipher strings";
3020 if (strstarts(cipher_in, "capi:"))
3021 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3023 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3028 ret = crypt_ctr_ivmode(ti, ivmode);
3032 /* Initialize and set key */
3033 ret = crypt_set_key(cc, key);
3035 ti->error = "Error decoding and setting key";
3040 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3041 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3043 ti->error = "Error creating IV";
3048 /* Initialize IV (set keys for ESSIV etc) */
3049 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3050 ret = cc->iv_gen_ops->init(cc);
3052 ti->error = "Error initialising IV";
3057 /* wipe the kernel key payload copy */
3059 memset(cc->key, 0, cc->key_size * sizeof(u8));
3064 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3066 struct crypt_config *cc = ti->private;
3067 struct dm_arg_set as;
3068 static const struct dm_arg _args[] = {
3069 {0, 8, "Invalid number of feature args"},
3071 unsigned int opt_params, val;
3072 const char *opt_string, *sval;
3076 /* Optional parameters */
3080 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3084 while (opt_params--) {
3085 opt_string = dm_shift_arg(&as);
3087 ti->error = "Not enough feature arguments";
3091 if (!strcasecmp(opt_string, "allow_discards"))
3092 ti->num_discard_bios = 1;
3094 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3095 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3097 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3098 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3099 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3100 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3101 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3102 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3103 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3104 if (val == 0 || val > MAX_TAG_SIZE) {
3105 ti->error = "Invalid integrity arguments";
3108 cc->on_disk_tag_size = val;
3109 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3110 if (!strcasecmp(sval, "aead")) {
3111 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3112 } else if (strcasecmp(sval, "none")) {
3113 ti->error = "Unknown integrity profile";
3117 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3118 if (!cc->cipher_auth)
3120 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3121 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3122 cc->sector_size > 4096 ||
3123 (cc->sector_size & (cc->sector_size - 1))) {
3124 ti->error = "Invalid feature value for sector_size";
3127 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3128 ti->error = "Device size is not multiple of sector_size feature";
3131 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3132 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3133 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3135 ti->error = "Invalid feature arguments";
3143 #ifdef CONFIG_BLK_DEV_ZONED
3144 static int crypt_report_zones(struct dm_target *ti,
3145 struct dm_report_zones_args *args, unsigned int nr_zones)
3147 struct crypt_config *cc = ti->private;
3149 return dm_report_zones(cc->dev->bdev, cc->start,
3150 cc->start + dm_target_offset(ti, args->next_sector),
3154 #define crypt_report_zones NULL
3158 * Construct an encryption mapping:
3159 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3161 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3163 struct crypt_config *cc;
3164 const char *devname = dm_table_device_name(ti->table);
3166 unsigned int align_mask;
3167 unsigned long long tmpll;
3169 size_t iv_size_padding, additional_req_size;
3173 ti->error = "Not enough arguments";
3177 key_size = get_key_size(&argv[1]);
3179 ti->error = "Cannot parse key size";
3183 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3185 ti->error = "Cannot allocate encryption context";
3188 cc->key_size = key_size;
3189 cc->sector_size = (1 << SECTOR_SHIFT);
3190 cc->sector_shift = 0;
3194 spin_lock(&dm_crypt_clients_lock);
3195 dm_crypt_clients_n++;
3196 crypt_calculate_pages_per_client();
3197 spin_unlock(&dm_crypt_clients_lock);
3199 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3203 /* Optional parameters need to be read before cipher constructor */
3205 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3210 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3214 if (crypt_integrity_aead(cc)) {
3215 cc->dmreq_start = sizeof(struct aead_request);
3216 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3217 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3219 cc->dmreq_start = sizeof(struct skcipher_request);
3220 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3221 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3223 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3225 if (align_mask < CRYPTO_MINALIGN) {
3226 /* Allocate the padding exactly */
3227 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3231 * If the cipher requires greater alignment than kmalloc
3232 * alignment, we don't know the exact position of the
3233 * initialization vector. We must assume worst case.
3235 iv_size_padding = align_mask;
3238 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3239 additional_req_size = sizeof(struct dm_crypt_request) +
3240 iv_size_padding + cc->iv_size +
3243 sizeof(unsigned int);
3245 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3247 ti->error = "Cannot allocate crypt request mempool";
3251 cc->per_bio_data_size = ti->per_io_data_size =
3252 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3253 ARCH_KMALLOC_MINALIGN);
3255 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3257 ti->error = "Cannot allocate page mempool";
3261 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3263 ti->error = "Cannot allocate crypt bioset";
3267 mutex_init(&cc->bio_alloc_lock);
3270 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3271 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3272 ti->error = "Invalid iv_offset sector";
3275 cc->iv_offset = tmpll;
3277 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3279 ti->error = "Device lookup failed";
3284 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3285 ti->error = "Invalid device sector";
3290 if (bdev_is_zoned(cc->dev->bdev)) {
3292 * For zoned block devices, we need to preserve the issuer write
3293 * ordering. To do so, disable write workqueues and force inline
3294 * encryption completion.
3296 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3297 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3300 * All zone append writes to a zone of a zoned block device will
3301 * have the same BIO sector, the start of the zone. When the
3302 * cypher IV mode uses sector values, all data targeting a
3303 * zone will be encrypted using the first sector numbers of the
3304 * zone. This will not result in write errors but will
3305 * cause most reads to fail as reads will use the sector values
3306 * for the actual data locations, resulting in IV mismatch.
3307 * To avoid this problem, ask DM core to emulate zone append
3308 * operations with regular writes.
3310 DMDEBUG("Zone append operations will be emulated");
3311 ti->emulate_zone_append = true;
3314 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3315 ret = crypt_integrity_ctr(cc, ti);
3319 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3320 if (!cc->tag_pool_max_sectors)
3321 cc->tag_pool_max_sectors = 1;
3323 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3324 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3326 ti->error = "Cannot allocate integrity tags mempool";
3330 cc->tag_pool_max_sectors <<= cc->sector_shift;
3334 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3335 if (!cc->io_queue) {
3336 ti->error = "Couldn't create kcryptd io queue";
3340 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3341 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3344 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3345 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3346 num_online_cpus(), devname);
3347 if (!cc->crypt_queue) {
3348 ti->error = "Couldn't create kcryptd queue";
3352 spin_lock_init(&cc->write_thread_lock);
3353 cc->write_tree = RB_ROOT;
3355 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3356 if (IS_ERR(cc->write_thread)) {
3357 ret = PTR_ERR(cc->write_thread);
3358 cc->write_thread = NULL;
3359 ti->error = "Couldn't spawn write thread";
3363 ti->num_flush_bios = 1;
3364 ti->limit_swap_bios = true;
3365 ti->accounts_remapped_io = true;
3367 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3371 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3376 static int crypt_map(struct dm_target *ti, struct bio *bio)
3378 struct dm_crypt_io *io;
3379 struct crypt_config *cc = ti->private;
3382 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3383 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3384 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3386 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3387 bio_op(bio) == REQ_OP_DISCARD)) {
3388 bio_set_dev(bio, cc->dev->bdev);
3389 if (bio_sectors(bio))
3390 bio->bi_iter.bi_sector = cc->start +
3391 dm_target_offset(ti, bio->bi_iter.bi_sector);
3392 return DM_MAPIO_REMAPPED;
3396 * Check if bio is too large, split as needed.
3398 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3399 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3400 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3403 * Ensure that bio is a multiple of internal sector encryption size
3404 * and is aligned to this size as defined in IO hints.
3406 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3407 return DM_MAPIO_KILL;
3409 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3410 return DM_MAPIO_KILL;
3412 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3413 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3415 if (cc->on_disk_tag_size) {
3416 unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3418 if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3419 unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3420 GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3421 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3422 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3423 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3424 io->integrity_metadata_from_pool = true;
3428 if (crypt_integrity_aead(cc))
3429 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3431 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3433 if (bio_data_dir(io->base_bio) == READ) {
3434 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3435 kcryptd_queue_read(io);
3437 kcryptd_queue_crypt(io);
3439 return DM_MAPIO_SUBMITTED;
3442 static char hex2asc(unsigned char c)
3444 return c + '0' + ((unsigned)(9 - c) >> 4 & 0x27);
3447 static void crypt_status(struct dm_target *ti, status_type_t type,
3448 unsigned status_flags, char *result, unsigned maxlen)
3450 struct crypt_config *cc = ti->private;
3452 int num_feature_args = 0;
3455 case STATUSTYPE_INFO:
3459 case STATUSTYPE_TABLE:
3460 DMEMIT("%s ", cc->cipher_string);
3462 if (cc->key_size > 0) {
3464 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3466 for (i = 0; i < cc->key_size; i++) {
3467 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3468 hex2asc(cc->key[i] & 0xf));
3474 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3475 cc->dev->name, (unsigned long long)cc->start);
3477 num_feature_args += !!ti->num_discard_bios;
3478 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3479 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3480 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3481 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3482 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3483 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3484 if (cc->on_disk_tag_size)
3486 if (num_feature_args) {
3487 DMEMIT(" %d", num_feature_args);
3488 if (ti->num_discard_bios)
3489 DMEMIT(" allow_discards");
3490 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3491 DMEMIT(" same_cpu_crypt");
3492 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3493 DMEMIT(" submit_from_crypt_cpus");
3494 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3495 DMEMIT(" no_read_workqueue");
3496 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3497 DMEMIT(" no_write_workqueue");
3498 if (cc->on_disk_tag_size)
3499 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3500 if (cc->sector_size != (1 << SECTOR_SHIFT))
3501 DMEMIT(" sector_size:%d", cc->sector_size);
3502 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3503 DMEMIT(" iv_large_sectors");
3507 case STATUSTYPE_IMA:
3508 DMEMIT_TARGET_NAME_VERSION(ti->type);
3509 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3510 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3511 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3513 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3515 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3517 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3520 if (cc->on_disk_tag_size)
3521 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3522 cc->on_disk_tag_size, cc->cipher_auth);
3523 if (cc->sector_size != (1 << SECTOR_SHIFT))
3524 DMEMIT(",sector_size=%d", cc->sector_size);
3525 if (cc->cipher_string)
3526 DMEMIT(",cipher_string=%s", cc->cipher_string);
3528 DMEMIT(",key_size=%u", cc->key_size);
3529 DMEMIT(",key_parts=%u", cc->key_parts);
3530 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3531 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3537 static void crypt_postsuspend(struct dm_target *ti)
3539 struct crypt_config *cc = ti->private;
3541 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3544 static int crypt_preresume(struct dm_target *ti)
3546 struct crypt_config *cc = ti->private;
3548 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3549 DMERR("aborting resume - crypt key is not set.");
3556 static void crypt_resume(struct dm_target *ti)
3558 struct crypt_config *cc = ti->private;
3560 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3563 /* Message interface
3567 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3568 char *result, unsigned maxlen)
3570 struct crypt_config *cc = ti->private;
3571 int key_size, ret = -EINVAL;
3576 if (!strcasecmp(argv[0], "key")) {
3577 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3578 DMWARN("not suspended during key manipulation.");
3581 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3582 /* The key size may not be changed. */
3583 key_size = get_key_size(&argv[2]);
3584 if (key_size < 0 || cc->key_size != key_size) {
3585 memset(argv[2], '0', strlen(argv[2]));
3589 ret = crypt_set_key(cc, argv[2]);
3592 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3593 ret = cc->iv_gen_ops->init(cc);
3594 /* wipe the kernel key payload copy */
3596 memset(cc->key, 0, cc->key_size * sizeof(u8));
3599 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3600 return crypt_wipe_key(cc);
3604 DMWARN("unrecognised message received.");
3608 static int crypt_iterate_devices(struct dm_target *ti,
3609 iterate_devices_callout_fn fn, void *data)
3611 struct crypt_config *cc = ti->private;
3613 return fn(ti, cc->dev, cc->start, ti->len, data);
3616 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3618 struct crypt_config *cc = ti->private;
3621 * Unfortunate constraint that is required to avoid the potential
3622 * for exceeding underlying device's max_segments limits -- due to
3623 * crypt_alloc_buffer() possibly allocating pages for the encryption
3624 * bio that are not as physically contiguous as the original bio.
3626 limits->max_segment_size = PAGE_SIZE;
3628 limits->logical_block_size =
3629 max_t(unsigned, limits->logical_block_size, cc->sector_size);
3630 limits->physical_block_size =
3631 max_t(unsigned, limits->physical_block_size, cc->sector_size);
3632 limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3635 static struct target_type crypt_target = {
3637 .version = {1, 24, 0},
3638 .module = THIS_MODULE,
3641 .features = DM_TARGET_ZONED_HM,
3642 .report_zones = crypt_report_zones,
3644 .status = crypt_status,
3645 .postsuspend = crypt_postsuspend,
3646 .preresume = crypt_preresume,
3647 .resume = crypt_resume,
3648 .message = crypt_message,
3649 .iterate_devices = crypt_iterate_devices,
3650 .io_hints = crypt_io_hints,
3653 static int __init dm_crypt_init(void)
3657 r = dm_register_target(&crypt_target);
3659 DMERR("register failed %d", r);
3664 static void __exit dm_crypt_exit(void)
3666 dm_unregister_target(&crypt_target);
3669 module_init(dm_crypt_init);
3670 module_exit(dm_crypt_exit);
3672 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3673 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3674 MODULE_LICENSE("GPL");