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/skcipher.h>
35 #include <crypto/aead.h>
36 #include <crypto/authenc.h>
37 #include <crypto/utils.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:1;
78 struct work_struct work;
79 struct tasklet_struct tasklet;
81 struct convert_context ctx;
87 struct rb_node rb_node;
88 } CRYPTO_MINALIGN_ATTR;
90 struct dm_crypt_request {
91 struct convert_context *ctx;
92 struct scatterlist sg_in[4];
93 struct scatterlist sg_out[4];
99 struct crypt_iv_operations {
100 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
102 void (*dtr)(struct crypt_config *cc);
103 int (*init)(struct crypt_config *cc);
104 int (*wipe)(struct crypt_config *cc);
105 int (*generator)(struct crypt_config *cc, u8 *iv,
106 struct dm_crypt_request *dmreq);
107 int (*post)(struct crypt_config *cc, u8 *iv,
108 struct dm_crypt_request *dmreq);
111 struct iv_benbi_private {
115 #define LMK_SEED_SIZE 64 /* hash + 0 */
116 struct iv_lmk_private {
117 struct crypto_shash *hash_tfm;
121 #define TCW_WHITENING_SIZE 16
122 struct iv_tcw_private {
123 struct crypto_shash *crc32_tfm;
128 #define ELEPHANT_MAX_KEY_SIZE 32
129 struct iv_elephant_private {
130 struct crypto_skcipher *tfm;
134 * Crypt: maps a linear range of a block device
135 * and encrypts / decrypts at the same time.
137 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
138 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
139 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
140 DM_CRYPT_WRITE_INLINE };
143 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
144 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
145 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
149 * The fields in here must be read only after initialization.
151 struct crypt_config {
155 struct percpu_counter n_allocated_pages;
157 struct workqueue_struct *io_queue;
158 struct workqueue_struct *crypt_queue;
160 spinlock_t write_thread_lock;
161 struct task_struct *write_thread;
162 struct rb_root write_tree;
168 const struct crypt_iv_operations *iv_gen_ops;
170 struct iv_benbi_private benbi;
171 struct iv_lmk_private lmk;
172 struct iv_tcw_private tcw;
173 struct iv_elephant_private elephant;
176 unsigned int iv_size;
177 unsigned short sector_size;
178 unsigned char sector_shift;
181 struct crypto_skcipher **tfms;
182 struct crypto_aead **tfms_aead;
184 unsigned int tfms_count;
185 unsigned long cipher_flags;
188 * Layout of each crypto request:
190 * struct skcipher_request
193 * struct dm_crypt_request
197 * The padding is added so that dm_crypt_request and the IV are
200 unsigned int dmreq_start;
202 unsigned int per_bio_data_size;
205 unsigned int key_size;
206 unsigned int key_parts; /* independent parts in key buffer */
207 unsigned int key_extra_size; /* additional keys length */
208 unsigned int key_mac_size; /* MAC key size for authenc(...) */
210 unsigned int integrity_tag_size;
211 unsigned int integrity_iv_size;
212 unsigned int on_disk_tag_size;
215 * pool for per bio private data, crypto requests,
216 * encryption requeusts/buffer pages and integrity tags
218 unsigned int tag_pool_max_sectors;
224 struct mutex bio_alloc_lock;
226 u8 *authenc_key; /* space for keys in authenc() format (if used) */
231 #define MAX_TAG_SIZE 480
232 #define POOL_ENTRY_SIZE 512
234 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
235 static unsigned int dm_crypt_clients_n;
236 static volatile unsigned long dm_crypt_pages_per_client;
237 #define DM_CRYPT_MEMORY_PERCENT 2
238 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
240 static void crypt_endio(struct bio *clone);
241 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
242 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
243 struct scatterlist *sg);
245 static bool crypt_integrity_aead(struct crypt_config *cc);
248 * Use this to access cipher attributes that are independent of the key.
250 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
252 return cc->cipher_tfm.tfms[0];
255 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
257 return cc->cipher_tfm.tfms_aead[0];
261 * Different IV generation algorithms:
263 * plain: the initial vector is the 32-bit little-endian version of the sector
264 * number, padded with zeros if necessary.
266 * plain64: the initial vector is the 64-bit little-endian version of the sector
267 * number, padded with zeros if necessary.
269 * plain64be: the initial vector is the 64-bit big-endian version of the sector
270 * number, padded with zeros if necessary.
272 * essiv: "encrypted sector|salt initial vector", the sector number is
273 * encrypted with the bulk cipher using a salt as key. The salt
274 * should be derived from the bulk cipher's key via hashing.
276 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
277 * (needed for LRW-32-AES and possible other narrow block modes)
279 * null: the initial vector is always zero. Provides compatibility with
280 * obsolete loop_fish2 devices. Do not use for new devices.
282 * lmk: Compatible implementation of the block chaining mode used
283 * by the Loop-AES block device encryption system
284 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
285 * It operates on full 512 byte sectors and uses CBC
286 * with an IV derived from the sector number, the data and
287 * optionally extra IV seed.
288 * This means that after decryption the first block
289 * of sector must be tweaked according to decrypted data.
290 * Loop-AES can use three encryption schemes:
291 * version 1: is plain aes-cbc mode
292 * version 2: uses 64 multikey scheme with lmk IV generator
293 * version 3: the same as version 2 with additional IV seed
294 * (it uses 65 keys, last key is used as IV seed)
296 * tcw: Compatible implementation of the block chaining mode used
297 * by the TrueCrypt device encryption system (prior to version 4.1).
298 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
299 * It operates on full 512 byte sectors and uses CBC
300 * with an IV derived from initial key and the sector number.
301 * In addition, whitening value is applied on every sector, whitening
302 * is calculated from initial key, sector number and mixed using CRC32.
303 * Note that this encryption scheme is vulnerable to watermarking attacks
304 * and should be used for old compatible containers access only.
306 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
307 * The IV is encrypted little-endian byte-offset (with the same key
308 * and cipher as the volume).
310 * elephant: The extended version of eboiv with additional Elephant diffuser
311 * used with Bitlocker CBC mode.
312 * This mode was used in older Windows systems
313 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
316 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
317 struct dm_crypt_request *dmreq)
319 memset(iv, 0, cc->iv_size);
320 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
325 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
326 struct dm_crypt_request *dmreq)
328 memset(iv, 0, cc->iv_size);
329 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
334 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
335 struct dm_crypt_request *dmreq)
337 memset(iv, 0, cc->iv_size);
338 /* iv_size is at least of size u64; usually it is 16 bytes */
339 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
344 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
345 struct dm_crypt_request *dmreq)
348 * ESSIV encryption of the IV is now handled by the crypto API,
349 * so just pass the plain sector number here.
351 memset(iv, 0, cc->iv_size);
352 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
357 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
363 if (crypt_integrity_aead(cc))
364 bs = crypto_aead_blocksize(any_tfm_aead(cc));
366 bs = crypto_skcipher_blocksize(any_tfm(cc));
370 * We need to calculate how far we must shift the sector count
371 * to get the cipher block count, we use this shift in _gen.
373 if (1 << log != bs) {
374 ti->error = "cypher blocksize is not a power of 2";
379 ti->error = "cypher blocksize is > 512";
383 cc->iv_gen_private.benbi.shift = 9 - log;
388 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
392 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
393 struct dm_crypt_request *dmreq)
397 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
399 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
400 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
405 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
406 struct dm_crypt_request *dmreq)
408 memset(iv, 0, cc->iv_size);
413 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
415 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
417 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
418 crypto_free_shash(lmk->hash_tfm);
419 lmk->hash_tfm = NULL;
421 kfree_sensitive(lmk->seed);
425 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
428 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
430 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
431 ti->error = "Unsupported sector size for LMK";
435 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
436 CRYPTO_ALG_ALLOCATES_MEMORY);
437 if (IS_ERR(lmk->hash_tfm)) {
438 ti->error = "Error initializing LMK hash";
439 return PTR_ERR(lmk->hash_tfm);
442 /* No seed in LMK version 2 */
443 if (cc->key_parts == cc->tfms_count) {
448 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
450 crypt_iv_lmk_dtr(cc);
451 ti->error = "Error kmallocing seed storage in LMK";
458 static int crypt_iv_lmk_init(struct crypt_config *cc)
460 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
461 int subkey_size = cc->key_size / cc->key_parts;
463 /* LMK seed is on the position of LMK_KEYS + 1 key */
465 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
466 crypto_shash_digestsize(lmk->hash_tfm));
471 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
473 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
476 memset(lmk->seed, 0, LMK_SEED_SIZE);
481 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
482 struct dm_crypt_request *dmreq,
485 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
486 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
487 struct md5_state md5state;
491 desc->tfm = lmk->hash_tfm;
493 r = crypto_shash_init(desc);
498 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
503 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
504 r = crypto_shash_update(desc, data + 16, 16 * 31);
508 /* Sector is cropped to 56 bits here */
509 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
510 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
511 buf[2] = cpu_to_le32(4024);
513 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
517 /* No MD5 padding here */
518 r = crypto_shash_export(desc, &md5state);
522 for (i = 0; i < MD5_HASH_WORDS; i++)
523 __cpu_to_le32s(&md5state.hash[i]);
524 memcpy(iv, &md5state.hash, cc->iv_size);
529 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
530 struct dm_crypt_request *dmreq)
532 struct scatterlist *sg;
536 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
537 sg = crypt_get_sg_data(cc, dmreq->sg_in);
538 src = kmap_local_page(sg_page(sg));
539 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
542 memset(iv, 0, cc->iv_size);
547 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
548 struct dm_crypt_request *dmreq)
550 struct scatterlist *sg;
554 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
557 sg = crypt_get_sg_data(cc, dmreq->sg_out);
558 dst = kmap_local_page(sg_page(sg));
559 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
561 /* Tweak the first block of plaintext sector */
563 crypto_xor(dst + sg->offset, iv, cc->iv_size);
569 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
571 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
573 kfree_sensitive(tcw->iv_seed);
575 kfree_sensitive(tcw->whitening);
576 tcw->whitening = NULL;
578 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
579 crypto_free_shash(tcw->crc32_tfm);
580 tcw->crc32_tfm = NULL;
583 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
586 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
588 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
589 ti->error = "Unsupported sector size for TCW";
593 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
594 ti->error = "Wrong key size for TCW";
598 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
599 CRYPTO_ALG_ALLOCATES_MEMORY);
600 if (IS_ERR(tcw->crc32_tfm)) {
601 ti->error = "Error initializing CRC32 in TCW";
602 return PTR_ERR(tcw->crc32_tfm);
605 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
606 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
607 if (!tcw->iv_seed || !tcw->whitening) {
608 crypt_iv_tcw_dtr(cc);
609 ti->error = "Error allocating seed storage in TCW";
616 static int crypt_iv_tcw_init(struct crypt_config *cc)
618 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
619 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
621 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
622 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
628 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
630 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
632 memset(tcw->iv_seed, 0, cc->iv_size);
633 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
638 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
639 struct dm_crypt_request *dmreq,
642 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
643 __le64 sector = cpu_to_le64(dmreq->iv_sector);
644 u8 buf[TCW_WHITENING_SIZE];
645 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
648 /* xor whitening with sector number */
649 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
650 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
652 /* calculate crc32 for every 32bit part and xor it */
653 desc->tfm = tcw->crc32_tfm;
654 for (i = 0; i < 4; i++) {
655 r = crypto_shash_init(desc);
658 r = crypto_shash_update(desc, &buf[i * 4], 4);
661 r = crypto_shash_final(desc, &buf[i * 4]);
665 crypto_xor(&buf[0], &buf[12], 4);
666 crypto_xor(&buf[4], &buf[8], 4);
668 /* apply whitening (8 bytes) to whole sector */
669 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
670 crypto_xor(data + i * 8, buf, 8);
672 memzero_explicit(buf, sizeof(buf));
676 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
677 struct dm_crypt_request *dmreq)
679 struct scatterlist *sg;
680 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
681 __le64 sector = cpu_to_le64(dmreq->iv_sector);
685 /* Remove whitening from ciphertext */
686 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
687 sg = crypt_get_sg_data(cc, dmreq->sg_in);
688 src = kmap_local_page(sg_page(sg));
689 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
694 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
696 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
702 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
703 struct dm_crypt_request *dmreq)
705 struct scatterlist *sg;
709 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
712 /* Apply whitening on ciphertext */
713 sg = crypt_get_sg_data(cc, dmreq->sg_out);
714 dst = kmap_local_page(sg_page(sg));
715 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
721 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
722 struct dm_crypt_request *dmreq)
724 /* Used only for writes, there must be an additional space to store IV */
725 get_random_bytes(iv, cc->iv_size);
729 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
732 if (crypt_integrity_aead(cc)) {
733 ti->error = "AEAD transforms not supported for EBOIV";
737 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
738 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
745 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
746 struct dm_crypt_request *dmreq)
748 struct crypto_skcipher *tfm = any_tfm(cc);
749 struct skcipher_request *req;
750 struct scatterlist src, dst;
751 DECLARE_CRYPTO_WAIT(wait);
752 unsigned int reqsize;
756 reqsize = ALIGN(crypto_skcipher_reqsize(tfm), __alignof__(__le64));
758 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
762 skcipher_request_set_tfm(req, tfm);
764 buf = (u8 *)req + reqsize;
765 memset(buf, 0, cc->iv_size);
766 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
768 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
769 sg_init_one(&dst, iv, cc->iv_size);
770 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
771 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
772 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
773 kfree_sensitive(req);
778 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
780 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
782 crypto_free_skcipher(elephant->tfm);
783 elephant->tfm = NULL;
786 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
789 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
792 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
793 CRYPTO_ALG_ALLOCATES_MEMORY);
794 if (IS_ERR(elephant->tfm)) {
795 r = PTR_ERR(elephant->tfm);
796 elephant->tfm = NULL;
800 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
802 crypt_iv_elephant_dtr(cc);
806 static void diffuser_disk_to_cpu(u32 *d, size_t n)
808 #ifndef __LITTLE_ENDIAN
811 for (i = 0; i < n; i++)
812 d[i] = le32_to_cpu((__le32)d[i]);
816 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
818 #ifndef __LITTLE_ENDIAN
821 for (i = 0; i < n; i++)
822 d[i] = cpu_to_le32((u32)d[i]);
826 static void diffuser_a_decrypt(u32 *d, size_t n)
830 for (i = 0; i < 5; i++) {
835 while (i1 < (n - 1)) {
836 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
842 d[i1] += d[i2] ^ d[i3];
848 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
851 d[i1] += d[i2] ^ d[i3];
857 static void diffuser_a_encrypt(u32 *d, size_t n)
861 for (i = 0; i < 5; i++) {
867 d[i1] -= d[i2] ^ d[i3];
870 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
876 d[i1] -= d[i2] ^ d[i3];
882 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
888 static void diffuser_b_decrypt(u32 *d, size_t n)
892 for (i = 0; i < 3; i++) {
897 while (i1 < (n - 1)) {
898 d[i1] += d[i2] ^ d[i3];
901 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
907 d[i1] += d[i2] ^ d[i3];
913 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
919 static void diffuser_b_encrypt(u32 *d, size_t n)
923 for (i = 0; i < 3; i++) {
929 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
935 d[i1] -= d[i2] ^ d[i3];
941 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
944 d[i1] -= d[i2] ^ d[i3];
950 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
952 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
953 u8 *es, *ks, *data, *data2, *data_offset;
954 struct skcipher_request *req;
955 struct scatterlist *sg, *sg2, src, dst;
956 DECLARE_CRYPTO_WAIT(wait);
959 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
960 es = kzalloc(16, GFP_NOIO); /* Key for AES */
961 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
963 if (!req || !es || !ks) {
968 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
971 sg_init_one(&src, es, 16);
972 sg_init_one(&dst, ks, 16);
973 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
974 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
975 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
981 sg_init_one(&dst, &ks[16], 16);
982 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
986 sg = crypt_get_sg_data(cc, dmreq->sg_out);
987 data = kmap_local_page(sg_page(sg));
988 data_offset = data + sg->offset;
990 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
991 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
992 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
993 data2 = kmap_local_page(sg_page(sg2));
994 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
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_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1001 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1002 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1005 for (i = 0; i < (cc->sector_size / 32); i++)
1006 crypto_xor(data_offset + i * 32, ks, 32);
1008 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1009 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1010 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1011 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1012 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1017 kfree_sensitive(ks);
1018 kfree_sensitive(es);
1019 skcipher_request_free(req);
1023 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1024 struct dm_crypt_request *dmreq)
1028 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1029 r = crypt_iv_elephant(cc, dmreq);
1034 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1037 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1038 struct dm_crypt_request *dmreq)
1040 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1041 return crypt_iv_elephant(cc, dmreq);
1046 static int crypt_iv_elephant_init(struct crypt_config *cc)
1048 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1049 int key_offset = cc->key_size - cc->key_extra_size;
1051 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1054 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1056 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1057 u8 key[ELEPHANT_MAX_KEY_SIZE];
1059 memset(key, 0, cc->key_extra_size);
1060 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1063 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1064 .generator = crypt_iv_plain_gen
1067 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1068 .generator = crypt_iv_plain64_gen
1071 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1072 .generator = crypt_iv_plain64be_gen
1075 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1076 .generator = crypt_iv_essiv_gen
1079 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1080 .ctr = crypt_iv_benbi_ctr,
1081 .dtr = crypt_iv_benbi_dtr,
1082 .generator = crypt_iv_benbi_gen
1085 static const struct crypt_iv_operations crypt_iv_null_ops = {
1086 .generator = crypt_iv_null_gen
1089 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1090 .ctr = crypt_iv_lmk_ctr,
1091 .dtr = crypt_iv_lmk_dtr,
1092 .init = crypt_iv_lmk_init,
1093 .wipe = crypt_iv_lmk_wipe,
1094 .generator = crypt_iv_lmk_gen,
1095 .post = crypt_iv_lmk_post
1098 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1099 .ctr = crypt_iv_tcw_ctr,
1100 .dtr = crypt_iv_tcw_dtr,
1101 .init = crypt_iv_tcw_init,
1102 .wipe = crypt_iv_tcw_wipe,
1103 .generator = crypt_iv_tcw_gen,
1104 .post = crypt_iv_tcw_post
1107 static const struct crypt_iv_operations crypt_iv_random_ops = {
1108 .generator = crypt_iv_random_gen
1111 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1112 .ctr = crypt_iv_eboiv_ctr,
1113 .generator = crypt_iv_eboiv_gen
1116 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1117 .ctr = crypt_iv_elephant_ctr,
1118 .dtr = crypt_iv_elephant_dtr,
1119 .init = crypt_iv_elephant_init,
1120 .wipe = crypt_iv_elephant_wipe,
1121 .generator = crypt_iv_elephant_gen,
1122 .post = crypt_iv_elephant_post
1126 * Integrity extensions
1128 static bool crypt_integrity_aead(struct crypt_config *cc)
1130 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1133 static bool crypt_integrity_hmac(struct crypt_config *cc)
1135 return crypt_integrity_aead(cc) && cc->key_mac_size;
1138 /* Get sg containing data */
1139 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1140 struct scatterlist *sg)
1142 if (unlikely(crypt_integrity_aead(cc)))
1148 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1150 struct bio_integrity_payload *bip;
1151 unsigned int tag_len;
1154 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1157 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1159 return PTR_ERR(bip);
1161 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1163 bip->bip_iter.bi_size = tag_len;
1164 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1166 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1167 tag_len, offset_in_page(io->integrity_metadata));
1168 if (unlikely(ret != tag_len))
1174 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1176 #ifdef CONFIG_BLK_DEV_INTEGRITY
1177 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1178 struct mapped_device *md = dm_table_get_md(ti->table);
1180 /* From now we require underlying device with our integrity profile */
1181 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1182 ti->error = "Integrity profile not supported.";
1186 if (bi->tag_size != cc->on_disk_tag_size ||
1187 bi->tuple_size != cc->on_disk_tag_size) {
1188 ti->error = "Integrity profile tag size mismatch.";
1191 if (1 << bi->interval_exp != cc->sector_size) {
1192 ti->error = "Integrity profile sector size mismatch.";
1196 if (crypt_integrity_aead(cc)) {
1197 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1198 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1199 cc->integrity_tag_size, cc->integrity_iv_size);
1201 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1202 ti->error = "Integrity AEAD auth tag size is not supported.";
1205 } else if (cc->integrity_iv_size)
1206 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1207 cc->integrity_iv_size);
1209 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1210 ti->error = "Not enough space for integrity tag in the profile.";
1216 ti->error = "Integrity profile not supported.";
1221 static void crypt_convert_init(struct crypt_config *cc,
1222 struct convert_context *ctx,
1223 struct bio *bio_out, struct bio *bio_in,
1226 ctx->bio_in = bio_in;
1227 ctx->bio_out = bio_out;
1229 ctx->iter_in = bio_in->bi_iter;
1231 ctx->iter_out = bio_out->bi_iter;
1232 ctx->cc_sector = sector + cc->iv_offset;
1233 init_completion(&ctx->restart);
1236 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1239 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1242 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1244 return (void *)((char *)dmreq - cc->dmreq_start);
1247 static u8 *iv_of_dmreq(struct crypt_config *cc,
1248 struct dm_crypt_request *dmreq)
1250 if (crypt_integrity_aead(cc))
1251 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1252 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1254 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1255 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1258 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1259 struct dm_crypt_request *dmreq)
1261 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1264 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1265 struct dm_crypt_request *dmreq)
1267 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1269 return (__le64 *) ptr;
1272 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1273 struct dm_crypt_request *dmreq)
1275 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1276 cc->iv_size + sizeof(uint64_t);
1278 return (unsigned int *)ptr;
1281 static void *tag_from_dmreq(struct crypt_config *cc,
1282 struct dm_crypt_request *dmreq)
1284 struct convert_context *ctx = dmreq->ctx;
1285 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1287 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1288 cc->on_disk_tag_size];
1291 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1292 struct dm_crypt_request *dmreq)
1294 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1297 static int crypt_convert_block_aead(struct crypt_config *cc,
1298 struct convert_context *ctx,
1299 struct aead_request *req,
1300 unsigned int tag_offset)
1302 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1303 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1304 struct dm_crypt_request *dmreq;
1305 u8 *iv, *org_iv, *tag_iv, *tag;
1309 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1311 /* Reject unexpected unaligned bio. */
1312 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1315 dmreq = dmreq_of_req(cc, req);
1316 dmreq->iv_sector = ctx->cc_sector;
1317 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1318 dmreq->iv_sector >>= cc->sector_shift;
1321 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1323 sector = org_sector_of_dmreq(cc, dmreq);
1324 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1326 iv = iv_of_dmreq(cc, dmreq);
1327 org_iv = org_iv_of_dmreq(cc, dmreq);
1328 tag = tag_from_dmreq(cc, dmreq);
1329 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1332 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1333 * | (authenticated) | (auth+encryption) | |
1334 * | sector_LE | IV | sector in/out | tag in/out |
1336 sg_init_table(dmreq->sg_in, 4);
1337 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1338 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1339 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1340 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1342 sg_init_table(dmreq->sg_out, 4);
1343 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1344 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1345 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1346 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1348 if (cc->iv_gen_ops) {
1349 /* For READs use IV stored in integrity metadata */
1350 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1351 memcpy(org_iv, tag_iv, cc->iv_size);
1353 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1356 /* Store generated IV in integrity metadata */
1357 if (cc->integrity_iv_size)
1358 memcpy(tag_iv, org_iv, cc->iv_size);
1360 /* Working copy of IV, to be modified in crypto API */
1361 memcpy(iv, org_iv, cc->iv_size);
1364 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1365 if (bio_data_dir(ctx->bio_in) == WRITE) {
1366 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1367 cc->sector_size, iv);
1368 r = crypto_aead_encrypt(req);
1369 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1370 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1371 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1373 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1374 cc->sector_size + cc->integrity_tag_size, iv);
1375 r = crypto_aead_decrypt(req);
1378 if (r == -EBADMSG) {
1379 sector_t s = le64_to_cpu(*sector);
1381 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1382 ctx->bio_in->bi_bdev, s);
1383 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1387 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1388 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1390 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1391 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1396 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1397 struct convert_context *ctx,
1398 struct skcipher_request *req,
1399 unsigned int tag_offset)
1401 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1402 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1403 struct scatterlist *sg_in, *sg_out;
1404 struct dm_crypt_request *dmreq;
1405 u8 *iv, *org_iv, *tag_iv;
1409 /* Reject unexpected unaligned bio. */
1410 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1413 dmreq = dmreq_of_req(cc, req);
1414 dmreq->iv_sector = ctx->cc_sector;
1415 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1416 dmreq->iv_sector >>= cc->sector_shift;
1419 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1421 iv = iv_of_dmreq(cc, dmreq);
1422 org_iv = org_iv_of_dmreq(cc, dmreq);
1423 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1425 sector = org_sector_of_dmreq(cc, dmreq);
1426 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1428 /* For skcipher we use only the first sg item */
1429 sg_in = &dmreq->sg_in[0];
1430 sg_out = &dmreq->sg_out[0];
1432 sg_init_table(sg_in, 1);
1433 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1435 sg_init_table(sg_out, 1);
1436 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1438 if (cc->iv_gen_ops) {
1439 /* For READs use IV stored in integrity metadata */
1440 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1441 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1443 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1446 /* Data can be already preprocessed in generator */
1447 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1449 /* Store generated IV in integrity metadata */
1450 if (cc->integrity_iv_size)
1451 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1453 /* Working copy of IV, to be modified in crypto API */
1454 memcpy(iv, org_iv, cc->iv_size);
1457 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1459 if (bio_data_dir(ctx->bio_in) == WRITE)
1460 r = crypto_skcipher_encrypt(req);
1462 r = crypto_skcipher_decrypt(req);
1464 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1465 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1467 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1468 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1473 static void kcryptd_async_done(void *async_req, int error);
1475 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1476 struct convert_context *ctx)
1478 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1481 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1486 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1489 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1490 * requests if driver request queue is full.
1492 skcipher_request_set_callback(ctx->r.req,
1493 CRYPTO_TFM_REQ_MAY_BACKLOG,
1494 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1499 static int crypt_alloc_req_aead(struct crypt_config *cc,
1500 struct convert_context *ctx)
1502 if (!ctx->r.req_aead) {
1503 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1504 if (!ctx->r.req_aead)
1508 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1511 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1512 * requests if driver request queue is full.
1514 aead_request_set_callback(ctx->r.req_aead,
1515 CRYPTO_TFM_REQ_MAY_BACKLOG,
1516 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1521 static int crypt_alloc_req(struct crypt_config *cc,
1522 struct convert_context *ctx)
1524 if (crypt_integrity_aead(cc))
1525 return crypt_alloc_req_aead(cc, ctx);
1527 return crypt_alloc_req_skcipher(cc, ctx);
1530 static void crypt_free_req_skcipher(struct crypt_config *cc,
1531 struct skcipher_request *req, struct bio *base_bio)
1533 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1535 if ((struct skcipher_request *)(io + 1) != req)
1536 mempool_free(req, &cc->req_pool);
1539 static void crypt_free_req_aead(struct crypt_config *cc,
1540 struct aead_request *req, struct bio *base_bio)
1542 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1544 if ((struct aead_request *)(io + 1) != req)
1545 mempool_free(req, &cc->req_pool);
1548 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1550 if (crypt_integrity_aead(cc))
1551 crypt_free_req_aead(cc, req, base_bio);
1553 crypt_free_req_skcipher(cc, req, base_bio);
1557 * Encrypt / decrypt data from one bio to another one (can be the same one)
1559 static blk_status_t crypt_convert(struct crypt_config *cc,
1560 struct convert_context *ctx, bool atomic, bool reset_pending)
1562 unsigned int tag_offset = 0;
1563 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1567 * if reset_pending is set we are dealing with the bio for the first time,
1568 * else we're continuing to work on the previous bio, so don't mess with
1569 * the cc_pending counter
1572 atomic_set(&ctx->cc_pending, 1);
1574 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1576 r = crypt_alloc_req(cc, ctx);
1578 complete(&ctx->restart);
1579 return BLK_STS_DEV_RESOURCE;
1582 atomic_inc(&ctx->cc_pending);
1584 if (crypt_integrity_aead(cc))
1585 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1587 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1591 * The request was queued by a crypto driver
1592 * but the driver request queue is full, let's wait.
1595 if (in_interrupt()) {
1596 if (try_wait_for_completion(&ctx->restart)) {
1598 * we don't have to block to wait for completion,
1603 * we can't wait for completion without blocking
1604 * exit and continue processing in a workqueue
1607 ctx->cc_sector += sector_step;
1609 return BLK_STS_DEV_RESOURCE;
1612 wait_for_completion(&ctx->restart);
1614 reinit_completion(&ctx->restart);
1617 * The request is queued and processed asynchronously,
1618 * completion function kcryptd_async_done() will be called.
1622 ctx->cc_sector += sector_step;
1626 * The request was already processed (synchronously).
1629 atomic_dec(&ctx->cc_pending);
1630 ctx->cc_sector += sector_step;
1636 * There was a data integrity error.
1639 atomic_dec(&ctx->cc_pending);
1640 return BLK_STS_PROTECTION;
1642 * There was an error while processing the request.
1645 atomic_dec(&ctx->cc_pending);
1646 return BLK_STS_IOERR;
1653 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1656 * Generate a new unfragmented bio with the given size
1657 * This should never violate the device limitations (but only because
1658 * max_segment_size is being constrained to PAGE_SIZE).
1660 * This function may be called concurrently. If we allocate from the mempool
1661 * concurrently, there is a possibility of deadlock. For example, if we have
1662 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1663 * the mempool concurrently, it may deadlock in a situation where both processes
1664 * have allocated 128 pages and the mempool is exhausted.
1666 * In order to avoid this scenario we allocate the pages under a mutex.
1668 * In order to not degrade performance with excessive locking, we try
1669 * non-blocking allocations without a mutex first but on failure we fallback
1670 * to blocking allocations with a mutex.
1672 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1674 struct crypt_config *cc = io->cc;
1676 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1677 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1678 unsigned int i, len, remaining_size;
1682 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1683 mutex_lock(&cc->bio_alloc_lock);
1685 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1687 clone->bi_private = io;
1688 clone->bi_end_io = crypt_endio;
1690 remaining_size = size;
1692 for (i = 0; i < nr_iovecs; i++) {
1693 page = mempool_alloc(&cc->page_pool, gfp_mask);
1695 crypt_free_buffer_pages(cc, clone);
1697 gfp_mask |= __GFP_DIRECT_RECLAIM;
1701 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1703 bio_add_page(clone, page, len, 0);
1705 remaining_size -= len;
1708 /* Allocate space for integrity tags */
1709 if (dm_crypt_integrity_io_alloc(io, clone)) {
1710 crypt_free_buffer_pages(cc, clone);
1715 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1716 mutex_unlock(&cc->bio_alloc_lock);
1721 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1724 struct bvec_iter_all iter_all;
1726 bio_for_each_segment_all(bv, clone, iter_all) {
1727 BUG_ON(!bv->bv_page);
1728 mempool_free(bv->bv_page, &cc->page_pool);
1732 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1733 struct bio *bio, sector_t sector)
1737 io->sector = sector;
1739 io->ctx.r.req = NULL;
1740 io->integrity_metadata = NULL;
1741 io->integrity_metadata_from_pool = false;
1742 io->in_tasklet = false;
1743 atomic_set(&io->io_pending, 0);
1746 static void crypt_inc_pending(struct dm_crypt_io *io)
1748 atomic_inc(&io->io_pending);
1751 static void kcryptd_io_bio_endio(struct work_struct *work)
1753 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1755 bio_endio(io->base_bio);
1759 * One of the bios was finished. Check for completion of
1760 * the whole request and correctly clean up the buffer.
1762 static void crypt_dec_pending(struct dm_crypt_io *io)
1764 struct crypt_config *cc = io->cc;
1765 struct bio *base_bio = io->base_bio;
1766 blk_status_t error = io->error;
1768 if (!atomic_dec_and_test(&io->io_pending))
1772 crypt_free_req(cc, io->ctx.r.req, base_bio);
1774 if (unlikely(io->integrity_metadata_from_pool))
1775 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1777 kfree(io->integrity_metadata);
1779 base_bio->bi_status = error;
1782 * If we are running this function from our tasklet,
1783 * we can't call bio_endio() here, because it will call
1784 * clone_endio() from dm.c, which in turn will
1785 * free the current struct dm_crypt_io structure with
1786 * our tasklet. In this case we need to delay bio_endio()
1787 * execution to after the tasklet is done and dequeued.
1789 if (io->in_tasklet) {
1790 INIT_WORK(&io->work, kcryptd_io_bio_endio);
1791 queue_work(cc->io_queue, &io->work);
1795 bio_endio(base_bio);
1799 * kcryptd/kcryptd_io:
1801 * Needed because it would be very unwise to do decryption in an
1802 * interrupt context.
1804 * kcryptd performs the actual encryption or decryption.
1806 * kcryptd_io performs the IO submission.
1808 * They must be separated as otherwise the final stages could be
1809 * starved by new requests which can block in the first stages due
1810 * to memory allocation.
1812 * The work is done per CPU global for all dm-crypt instances.
1813 * They should not depend on each other and do not block.
1815 static void crypt_endio(struct bio *clone)
1817 struct dm_crypt_io *io = clone->bi_private;
1818 struct crypt_config *cc = io->cc;
1819 unsigned int rw = bio_data_dir(clone);
1823 * free the processed pages
1826 crypt_free_buffer_pages(cc, clone);
1828 error = clone->bi_status;
1831 if (rw == READ && !error) {
1832 kcryptd_queue_crypt(io);
1836 if (unlikely(error))
1839 crypt_dec_pending(io);
1842 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1844 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1846 struct crypt_config *cc = io->cc;
1850 * We need the original biovec array in order to decrypt the whole bio
1851 * data *afterwards* -- thanks to immutable biovecs we don't need to
1852 * worry about the block layer modifying the biovec array; so leverage
1853 * bio_alloc_clone().
1855 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1858 clone->bi_private = io;
1859 clone->bi_end_io = crypt_endio;
1861 crypt_inc_pending(io);
1863 clone->bi_iter.bi_sector = cc->start + io->sector;
1865 if (dm_crypt_integrity_io_alloc(io, clone)) {
1866 crypt_dec_pending(io);
1871 dm_submit_bio_remap(io->base_bio, clone);
1875 static void kcryptd_io_read_work(struct work_struct *work)
1877 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1879 crypt_inc_pending(io);
1880 if (kcryptd_io_read(io, GFP_NOIO))
1881 io->error = BLK_STS_RESOURCE;
1882 crypt_dec_pending(io);
1885 static void kcryptd_queue_read(struct dm_crypt_io *io)
1887 struct crypt_config *cc = io->cc;
1889 INIT_WORK(&io->work, kcryptd_io_read_work);
1890 queue_work(cc->io_queue, &io->work);
1893 static void kcryptd_io_write(struct dm_crypt_io *io)
1895 struct bio *clone = io->ctx.bio_out;
1897 dm_submit_bio_remap(io->base_bio, clone);
1900 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1902 static int dmcrypt_write(void *data)
1904 struct crypt_config *cc = data;
1905 struct dm_crypt_io *io;
1908 struct rb_root write_tree;
1909 struct blk_plug plug;
1911 spin_lock_irq(&cc->write_thread_lock);
1914 if (!RB_EMPTY_ROOT(&cc->write_tree))
1917 set_current_state(TASK_INTERRUPTIBLE);
1919 spin_unlock_irq(&cc->write_thread_lock);
1921 if (unlikely(kthread_should_stop())) {
1922 set_current_state(TASK_RUNNING);
1928 set_current_state(TASK_RUNNING);
1929 spin_lock_irq(&cc->write_thread_lock);
1930 goto continue_locked;
1933 write_tree = cc->write_tree;
1934 cc->write_tree = RB_ROOT;
1935 spin_unlock_irq(&cc->write_thread_lock);
1937 BUG_ON(rb_parent(write_tree.rb_node));
1940 * Note: we cannot walk the tree here with rb_next because
1941 * the structures may be freed when kcryptd_io_write is called.
1943 blk_start_plug(&plug);
1945 io = crypt_io_from_node(rb_first(&write_tree));
1946 rb_erase(&io->rb_node, &write_tree);
1947 kcryptd_io_write(io);
1949 } while (!RB_EMPTY_ROOT(&write_tree));
1950 blk_finish_plug(&plug);
1955 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1957 struct bio *clone = io->ctx.bio_out;
1958 struct crypt_config *cc = io->cc;
1959 unsigned long flags;
1961 struct rb_node **rbp, *parent;
1963 if (unlikely(io->error)) {
1964 crypt_free_buffer_pages(cc, clone);
1966 crypt_dec_pending(io);
1970 /* crypt_convert should have filled the clone bio */
1971 BUG_ON(io->ctx.iter_out.bi_size);
1973 clone->bi_iter.bi_sector = cc->start + io->sector;
1975 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1976 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1977 dm_submit_bio_remap(io->base_bio, clone);
1981 spin_lock_irqsave(&cc->write_thread_lock, flags);
1982 if (RB_EMPTY_ROOT(&cc->write_tree))
1983 wake_up_process(cc->write_thread);
1984 rbp = &cc->write_tree.rb_node;
1986 sector = io->sector;
1989 if (sector < crypt_io_from_node(parent)->sector)
1990 rbp = &(*rbp)->rb_left;
1992 rbp = &(*rbp)->rb_right;
1994 rb_link_node(&io->rb_node, parent, rbp);
1995 rb_insert_color(&io->rb_node, &cc->write_tree);
1996 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1999 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2000 struct convert_context *ctx)
2003 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2007 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2008 * constraints so they do not need to be issued inline by
2009 * kcryptd_crypt_write_convert().
2011 switch (bio_op(ctx->bio_in)) {
2013 case REQ_OP_WRITE_ZEROES:
2020 static void kcryptd_crypt_write_continue(struct work_struct *work)
2022 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2023 struct crypt_config *cc = io->cc;
2024 struct convert_context *ctx = &io->ctx;
2026 sector_t sector = io->sector;
2029 wait_for_completion(&ctx->restart);
2030 reinit_completion(&ctx->restart);
2032 r = crypt_convert(cc, &io->ctx, true, false);
2035 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2036 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2037 /* Wait for completion signaled by kcryptd_async_done() */
2038 wait_for_completion(&ctx->restart);
2042 /* Encryption was already finished, submit io now */
2043 if (crypt_finished) {
2044 kcryptd_crypt_write_io_submit(io, 0);
2045 io->sector = sector;
2048 crypt_dec_pending(io);
2051 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2053 struct crypt_config *cc = io->cc;
2054 struct convert_context *ctx = &io->ctx;
2057 sector_t sector = io->sector;
2061 * Prevent io from disappearing until this function completes.
2063 crypt_inc_pending(io);
2064 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2066 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2067 if (unlikely(!clone)) {
2068 io->error = BLK_STS_IOERR;
2072 io->ctx.bio_out = clone;
2073 io->ctx.iter_out = clone->bi_iter;
2075 sector += bio_sectors(clone);
2077 crypt_inc_pending(io);
2078 r = crypt_convert(cc, ctx,
2079 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2081 * Crypto API backlogged the request, because its queue was full
2082 * and we're in softirq context, so continue from a workqueue
2083 * (TODO: is it actually possible to be in softirq in the write path?)
2085 if (r == BLK_STS_DEV_RESOURCE) {
2086 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2087 queue_work(cc->crypt_queue, &io->work);
2092 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2093 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2094 /* Wait for completion signaled by kcryptd_async_done() */
2095 wait_for_completion(&ctx->restart);
2099 /* Encryption was already finished, submit io now */
2100 if (crypt_finished) {
2101 kcryptd_crypt_write_io_submit(io, 0);
2102 io->sector = sector;
2106 crypt_dec_pending(io);
2109 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2111 crypt_dec_pending(io);
2114 static void kcryptd_crypt_read_continue(struct work_struct *work)
2116 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2117 struct crypt_config *cc = io->cc;
2120 wait_for_completion(&io->ctx.restart);
2121 reinit_completion(&io->ctx.restart);
2123 r = crypt_convert(cc, &io->ctx, true, false);
2127 if (atomic_dec_and_test(&io->ctx.cc_pending))
2128 kcryptd_crypt_read_done(io);
2130 crypt_dec_pending(io);
2133 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2135 struct crypt_config *cc = io->cc;
2138 crypt_inc_pending(io);
2140 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2143 r = crypt_convert(cc, &io->ctx,
2144 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2146 * Crypto API backlogged the request, because its queue was full
2147 * and we're in softirq context, so continue from a workqueue
2149 if (r == BLK_STS_DEV_RESOURCE) {
2150 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2151 queue_work(cc->crypt_queue, &io->work);
2157 if (atomic_dec_and_test(&io->ctx.cc_pending))
2158 kcryptd_crypt_read_done(io);
2160 crypt_dec_pending(io);
2163 static void kcryptd_async_done(void *data, int error)
2165 struct dm_crypt_request *dmreq = data;
2166 struct convert_context *ctx = dmreq->ctx;
2167 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2168 struct crypt_config *cc = io->cc;
2171 * A request from crypto driver backlog is going to be processed now,
2172 * finish the completion and continue in crypt_convert().
2173 * (Callback will be called for the second time for this request.)
2175 if (error == -EINPROGRESS) {
2176 complete(&ctx->restart);
2180 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2181 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2183 if (error == -EBADMSG) {
2184 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2186 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2187 ctx->bio_in->bi_bdev, s);
2188 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2190 io->error = BLK_STS_PROTECTION;
2191 } else if (error < 0)
2192 io->error = BLK_STS_IOERR;
2194 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2196 if (!atomic_dec_and_test(&ctx->cc_pending))
2200 * The request is fully completed: for inline writes, let
2201 * kcryptd_crypt_write_convert() do the IO submission.
2203 if (bio_data_dir(io->base_bio) == READ) {
2204 kcryptd_crypt_read_done(io);
2208 if (kcryptd_crypt_write_inline(cc, ctx)) {
2209 complete(&ctx->restart);
2213 kcryptd_crypt_write_io_submit(io, 1);
2216 static void kcryptd_crypt(struct work_struct *work)
2218 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2220 if (bio_data_dir(io->base_bio) == READ)
2221 kcryptd_crypt_read_convert(io);
2223 kcryptd_crypt_write_convert(io);
2226 static void kcryptd_crypt_tasklet(unsigned long work)
2228 kcryptd_crypt((struct work_struct *)work);
2231 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2233 struct crypt_config *cc = io->cc;
2235 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2236 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2238 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2239 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2240 * it is being executed with irqs disabled.
2242 if (in_hardirq() || irqs_disabled()) {
2243 io->in_tasklet = true;
2244 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2245 tasklet_schedule(&io->tasklet);
2249 kcryptd_crypt(&io->work);
2253 INIT_WORK(&io->work, kcryptd_crypt);
2254 queue_work(cc->crypt_queue, &io->work);
2257 static void crypt_free_tfms_aead(struct crypt_config *cc)
2259 if (!cc->cipher_tfm.tfms_aead)
2262 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2263 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2264 cc->cipher_tfm.tfms_aead[0] = NULL;
2267 kfree(cc->cipher_tfm.tfms_aead);
2268 cc->cipher_tfm.tfms_aead = NULL;
2271 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2275 if (!cc->cipher_tfm.tfms)
2278 for (i = 0; i < cc->tfms_count; i++)
2279 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2280 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2281 cc->cipher_tfm.tfms[i] = NULL;
2284 kfree(cc->cipher_tfm.tfms);
2285 cc->cipher_tfm.tfms = NULL;
2288 static void crypt_free_tfms(struct crypt_config *cc)
2290 if (crypt_integrity_aead(cc))
2291 crypt_free_tfms_aead(cc);
2293 crypt_free_tfms_skcipher(cc);
2296 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2301 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2302 sizeof(struct crypto_skcipher *),
2304 if (!cc->cipher_tfm.tfms)
2307 for (i = 0; i < cc->tfms_count; i++) {
2308 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2309 CRYPTO_ALG_ALLOCATES_MEMORY);
2310 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2311 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2312 crypt_free_tfms(cc);
2318 * dm-crypt performance can vary greatly depending on which crypto
2319 * algorithm implementation is used. Help people debug performance
2320 * problems by logging the ->cra_driver_name.
2322 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2323 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2327 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2331 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2332 if (!cc->cipher_tfm.tfms)
2335 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2336 CRYPTO_ALG_ALLOCATES_MEMORY);
2337 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2338 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2339 crypt_free_tfms(cc);
2343 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2344 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2348 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2350 if (crypt_integrity_aead(cc))
2351 return crypt_alloc_tfms_aead(cc, ciphermode);
2353 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2356 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2358 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2361 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2363 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2367 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2368 * the key must be for some reason in special format.
2369 * This funcion converts cc->key to this special format.
2371 static void crypt_copy_authenckey(char *p, const void *key,
2372 unsigned int enckeylen, unsigned int authkeylen)
2374 struct crypto_authenc_key_param *param;
2377 rta = (struct rtattr *)p;
2378 param = RTA_DATA(rta);
2379 param->enckeylen = cpu_to_be32(enckeylen);
2380 rta->rta_len = RTA_LENGTH(sizeof(*param));
2381 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2382 p += RTA_SPACE(sizeof(*param));
2383 memcpy(p, key + enckeylen, authkeylen);
2385 memcpy(p, key, enckeylen);
2388 static int crypt_setkey(struct crypt_config *cc)
2390 unsigned int subkey_size;
2393 /* Ignore extra keys (which are used for IV etc) */
2394 subkey_size = crypt_subkey_size(cc);
2396 if (crypt_integrity_hmac(cc)) {
2397 if (subkey_size < cc->key_mac_size)
2400 crypt_copy_authenckey(cc->authenc_key, cc->key,
2401 subkey_size - cc->key_mac_size,
2405 for (i = 0; i < cc->tfms_count; i++) {
2406 if (crypt_integrity_hmac(cc))
2407 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2408 cc->authenc_key, crypt_authenckey_size(cc));
2409 else if (crypt_integrity_aead(cc))
2410 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2411 cc->key + (i * subkey_size),
2414 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2415 cc->key + (i * subkey_size),
2421 if (crypt_integrity_hmac(cc))
2422 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2429 static bool contains_whitespace(const char *str)
2432 if (isspace(*str++))
2437 static int set_key_user(struct crypt_config *cc, struct key *key)
2439 const struct user_key_payload *ukp;
2441 ukp = user_key_payload_locked(key);
2443 return -EKEYREVOKED;
2445 if (cc->key_size != ukp->datalen)
2448 memcpy(cc->key, ukp->data, cc->key_size);
2453 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2455 const struct encrypted_key_payload *ekp;
2457 ekp = key->payload.data[0];
2459 return -EKEYREVOKED;
2461 if (cc->key_size != ekp->decrypted_datalen)
2464 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2469 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2471 const struct trusted_key_payload *tkp;
2473 tkp = key->payload.data[0];
2475 return -EKEYREVOKED;
2477 if (cc->key_size != tkp->key_len)
2480 memcpy(cc->key, tkp->key, cc->key_size);
2485 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2487 char *new_key_string, *key_desc;
2489 struct key_type *type;
2491 int (*set_key)(struct crypt_config *cc, struct key *key);
2494 * Reject key_string with whitespace. dm core currently lacks code for
2495 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2497 if (contains_whitespace(key_string)) {
2498 DMERR("whitespace chars not allowed in key string");
2502 /* look for next ':' separating key_type from key_description */
2503 key_desc = strchr(key_string, ':');
2504 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2507 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2508 type = &key_type_logon;
2509 set_key = set_key_user;
2510 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2511 type = &key_type_user;
2512 set_key = set_key_user;
2513 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2514 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2515 type = &key_type_encrypted;
2516 set_key = set_key_encrypted;
2517 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2518 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2519 type = &key_type_trusted;
2520 set_key = set_key_trusted;
2525 new_key_string = kstrdup(key_string, GFP_KERNEL);
2526 if (!new_key_string)
2529 key = request_key(type, key_desc + 1, NULL);
2531 kfree_sensitive(new_key_string);
2532 return PTR_ERR(key);
2535 down_read(&key->sem);
2537 ret = set_key(cc, key);
2541 kfree_sensitive(new_key_string);
2548 /* clear the flag since following operations may invalidate previously valid key */
2549 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2551 ret = crypt_setkey(cc);
2554 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2555 kfree_sensitive(cc->key_string);
2556 cc->key_string = new_key_string;
2558 kfree_sensitive(new_key_string);
2563 static int get_key_size(char **key_string)
2568 if (*key_string[0] != ':')
2569 return strlen(*key_string) >> 1;
2571 /* look for next ':' in key string */
2572 colon = strpbrk(*key_string + 1, ":");
2576 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2579 *key_string = colon;
2581 /* remaining key string should be :<logon|user>:<key_desc> */
2588 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2593 static int get_key_size(char **key_string)
2595 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2598 #endif /* CONFIG_KEYS */
2600 static int crypt_set_key(struct crypt_config *cc, char *key)
2603 int key_string_len = strlen(key);
2605 /* Hyphen (which gives a key_size of zero) means there is no key. */
2606 if (!cc->key_size && strcmp(key, "-"))
2609 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2610 if (key[0] == ':') {
2611 r = crypt_set_keyring_key(cc, key + 1);
2615 /* clear the flag since following operations may invalidate previously valid key */
2616 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2618 /* wipe references to any kernel keyring key */
2619 kfree_sensitive(cc->key_string);
2620 cc->key_string = NULL;
2622 /* Decode key from its hex representation. */
2623 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2626 r = crypt_setkey(cc);
2628 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2631 /* Hex key string not needed after here, so wipe it. */
2632 memset(key, '0', key_string_len);
2637 static int crypt_wipe_key(struct crypt_config *cc)
2641 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2642 get_random_bytes(&cc->key, cc->key_size);
2644 /* Wipe IV private keys */
2645 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2646 r = cc->iv_gen_ops->wipe(cc);
2651 kfree_sensitive(cc->key_string);
2652 cc->key_string = NULL;
2653 r = crypt_setkey(cc);
2654 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2659 static void crypt_calculate_pages_per_client(void)
2661 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2663 if (!dm_crypt_clients_n)
2666 pages /= dm_crypt_clients_n;
2667 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2668 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2669 dm_crypt_pages_per_client = pages;
2672 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2674 struct crypt_config *cc = pool_data;
2678 * Note, percpu_counter_read_positive() may over (and under) estimate
2679 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2680 * but avoids potential spinlock contention of an exact result.
2682 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2683 likely(gfp_mask & __GFP_NORETRY))
2686 page = alloc_page(gfp_mask);
2687 if (likely(page != NULL))
2688 percpu_counter_add(&cc->n_allocated_pages, 1);
2693 static void crypt_page_free(void *page, void *pool_data)
2695 struct crypt_config *cc = pool_data;
2698 percpu_counter_sub(&cc->n_allocated_pages, 1);
2701 static void crypt_dtr(struct dm_target *ti)
2703 struct crypt_config *cc = ti->private;
2710 if (cc->write_thread)
2711 kthread_stop(cc->write_thread);
2714 destroy_workqueue(cc->io_queue);
2715 if (cc->crypt_queue)
2716 destroy_workqueue(cc->crypt_queue);
2718 crypt_free_tfms(cc);
2720 bioset_exit(&cc->bs);
2722 mempool_exit(&cc->page_pool);
2723 mempool_exit(&cc->req_pool);
2724 mempool_exit(&cc->tag_pool);
2726 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2727 percpu_counter_destroy(&cc->n_allocated_pages);
2729 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2730 cc->iv_gen_ops->dtr(cc);
2733 dm_put_device(ti, cc->dev);
2735 kfree_sensitive(cc->cipher_string);
2736 kfree_sensitive(cc->key_string);
2737 kfree_sensitive(cc->cipher_auth);
2738 kfree_sensitive(cc->authenc_key);
2740 mutex_destroy(&cc->bio_alloc_lock);
2742 /* Must zero key material before freeing */
2743 kfree_sensitive(cc);
2745 spin_lock(&dm_crypt_clients_lock);
2746 WARN_ON(!dm_crypt_clients_n);
2747 dm_crypt_clients_n--;
2748 crypt_calculate_pages_per_client();
2749 spin_unlock(&dm_crypt_clients_lock);
2751 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2754 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2756 struct crypt_config *cc = ti->private;
2758 if (crypt_integrity_aead(cc))
2759 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2761 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2764 /* at least a 64 bit sector number should fit in our buffer */
2765 cc->iv_size = max(cc->iv_size,
2766 (unsigned int)(sizeof(u64) / sizeof(u8)));
2768 DMWARN("Selected cipher does not support IVs");
2772 /* Choose ivmode, see comments at iv code. */
2774 cc->iv_gen_ops = NULL;
2775 else if (strcmp(ivmode, "plain") == 0)
2776 cc->iv_gen_ops = &crypt_iv_plain_ops;
2777 else if (strcmp(ivmode, "plain64") == 0)
2778 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2779 else if (strcmp(ivmode, "plain64be") == 0)
2780 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2781 else if (strcmp(ivmode, "essiv") == 0)
2782 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2783 else if (strcmp(ivmode, "benbi") == 0)
2784 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2785 else if (strcmp(ivmode, "null") == 0)
2786 cc->iv_gen_ops = &crypt_iv_null_ops;
2787 else if (strcmp(ivmode, "eboiv") == 0)
2788 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2789 else if (strcmp(ivmode, "elephant") == 0) {
2790 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2792 cc->key_extra_size = cc->key_size / 2;
2793 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2795 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2796 } else if (strcmp(ivmode, "lmk") == 0) {
2797 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2799 * Version 2 and 3 is recognised according
2800 * to length of provided multi-key string.
2801 * If present (version 3), last key is used as IV seed.
2802 * All keys (including IV seed) are always the same size.
2804 if (cc->key_size % cc->key_parts) {
2806 cc->key_extra_size = cc->key_size / cc->key_parts;
2808 } else if (strcmp(ivmode, "tcw") == 0) {
2809 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2810 cc->key_parts += 2; /* IV + whitening */
2811 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2812 } else if (strcmp(ivmode, "random") == 0) {
2813 cc->iv_gen_ops = &crypt_iv_random_ops;
2814 /* Need storage space in integrity fields. */
2815 cc->integrity_iv_size = cc->iv_size;
2817 ti->error = "Invalid IV mode";
2825 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2826 * The HMAC is needed to calculate tag size (HMAC digest size).
2827 * This should be probably done by crypto-api calls (once available...)
2829 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2831 char *start, *end, *mac_alg = NULL;
2832 struct crypto_ahash *mac;
2834 if (!strstarts(cipher_api, "authenc("))
2837 start = strchr(cipher_api, '(');
2838 end = strchr(cipher_api, ',');
2839 if (!start || !end || ++start > end)
2842 mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2845 strncpy(mac_alg, start, end - start);
2847 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2851 return PTR_ERR(mac);
2853 cc->key_mac_size = crypto_ahash_digestsize(mac);
2854 crypto_free_ahash(mac);
2856 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2857 if (!cc->authenc_key)
2863 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2864 char **ivmode, char **ivopts)
2866 struct crypt_config *cc = ti->private;
2867 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2873 * New format (capi: prefix)
2874 * capi:cipher_api_spec-iv:ivopts
2876 tmp = &cipher_in[strlen("capi:")];
2878 /* Separate IV options if present, it can contain another '-' in hash name */
2879 *ivopts = strrchr(tmp, ':');
2885 *ivmode = strrchr(tmp, '-');
2890 /* The rest is crypto API spec */
2893 /* Alloc AEAD, can be used only in new format. */
2894 if (crypt_integrity_aead(cc)) {
2895 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2897 ti->error = "Invalid AEAD cipher spec";
2902 if (*ivmode && !strcmp(*ivmode, "lmk"))
2903 cc->tfms_count = 64;
2905 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2907 ti->error = "Digest algorithm missing for ESSIV mode";
2910 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2911 cipher_api, *ivopts);
2912 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2913 ti->error = "Cannot allocate cipher string";
2919 cc->key_parts = cc->tfms_count;
2921 /* Allocate cipher */
2922 ret = crypt_alloc_tfms(cc, cipher_api);
2924 ti->error = "Error allocating crypto tfm";
2928 if (crypt_integrity_aead(cc))
2929 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2931 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2936 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2937 char **ivmode, char **ivopts)
2939 struct crypt_config *cc = ti->private;
2940 char *tmp, *cipher, *chainmode, *keycount;
2941 char *cipher_api = NULL;
2945 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2946 ti->error = "Bad cipher specification";
2951 * Legacy dm-crypt cipher specification
2952 * cipher[:keycount]-mode-iv:ivopts
2955 keycount = strsep(&tmp, "-");
2956 cipher = strsep(&keycount, ":");
2960 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2961 !is_power_of_2(cc->tfms_count)) {
2962 ti->error = "Bad cipher key count specification";
2965 cc->key_parts = cc->tfms_count;
2967 chainmode = strsep(&tmp, "-");
2968 *ivmode = strsep(&tmp, ":");
2972 * For compatibility with the original dm-crypt mapping format, if
2973 * only the cipher name is supplied, use cbc-plain.
2975 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2980 if (strcmp(chainmode, "ecb") && !*ivmode) {
2981 ti->error = "IV mechanism required";
2985 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2989 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2991 ti->error = "Digest algorithm missing for ESSIV mode";
2995 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2996 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2998 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2999 "%s(%s)", chainmode, cipher);
3001 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3006 /* Allocate cipher */
3007 ret = crypt_alloc_tfms(cc, cipher_api);
3009 ti->error = "Error allocating crypto tfm";
3017 ti->error = "Cannot allocate cipher strings";
3021 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3023 struct crypt_config *cc = ti->private;
3024 char *ivmode = NULL, *ivopts = NULL;
3027 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3028 if (!cc->cipher_string) {
3029 ti->error = "Cannot allocate cipher strings";
3033 if (strstarts(cipher_in, "capi:"))
3034 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3036 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3041 ret = crypt_ctr_ivmode(ti, ivmode);
3045 /* Initialize and set key */
3046 ret = crypt_set_key(cc, key);
3048 ti->error = "Error decoding and setting key";
3053 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3054 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3056 ti->error = "Error creating IV";
3061 /* Initialize IV (set keys for ESSIV etc) */
3062 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3063 ret = cc->iv_gen_ops->init(cc);
3065 ti->error = "Error initialising IV";
3070 /* wipe the kernel key payload copy */
3072 memset(cc->key, 0, cc->key_size * sizeof(u8));
3077 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3079 struct crypt_config *cc = ti->private;
3080 struct dm_arg_set as;
3081 static const struct dm_arg _args[] = {
3082 {0, 8, "Invalid number of feature args"},
3084 unsigned int opt_params, val;
3085 const char *opt_string, *sval;
3089 /* Optional parameters */
3093 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3097 while (opt_params--) {
3098 opt_string = dm_shift_arg(&as);
3100 ti->error = "Not enough feature arguments";
3104 if (!strcasecmp(opt_string, "allow_discards"))
3105 ti->num_discard_bios = 1;
3107 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3108 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3110 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3111 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3112 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3113 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3114 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3115 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3116 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3117 if (val == 0 || val > MAX_TAG_SIZE) {
3118 ti->error = "Invalid integrity arguments";
3121 cc->on_disk_tag_size = val;
3122 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3123 if (!strcasecmp(sval, "aead")) {
3124 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3125 } else if (strcasecmp(sval, "none")) {
3126 ti->error = "Unknown integrity profile";
3130 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3131 if (!cc->cipher_auth)
3133 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3134 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3135 cc->sector_size > 4096 ||
3136 (cc->sector_size & (cc->sector_size - 1))) {
3137 ti->error = "Invalid feature value for sector_size";
3140 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3141 ti->error = "Device size is not multiple of sector_size feature";
3144 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3145 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3146 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3148 ti->error = "Invalid feature arguments";
3156 #ifdef CONFIG_BLK_DEV_ZONED
3157 static int crypt_report_zones(struct dm_target *ti,
3158 struct dm_report_zones_args *args, unsigned int nr_zones)
3160 struct crypt_config *cc = ti->private;
3162 return dm_report_zones(cc->dev->bdev, cc->start,
3163 cc->start + dm_target_offset(ti, args->next_sector),
3167 #define crypt_report_zones NULL
3171 * Construct an encryption mapping:
3172 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3174 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3176 struct crypt_config *cc;
3177 const char *devname = dm_table_device_name(ti->table);
3179 unsigned int align_mask;
3180 unsigned long long tmpll;
3182 size_t iv_size_padding, additional_req_size;
3186 ti->error = "Not enough arguments";
3190 key_size = get_key_size(&argv[1]);
3192 ti->error = "Cannot parse key size";
3196 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3198 ti->error = "Cannot allocate encryption context";
3201 cc->key_size = key_size;
3202 cc->sector_size = (1 << SECTOR_SHIFT);
3203 cc->sector_shift = 0;
3207 spin_lock(&dm_crypt_clients_lock);
3208 dm_crypt_clients_n++;
3209 crypt_calculate_pages_per_client();
3210 spin_unlock(&dm_crypt_clients_lock);
3212 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3216 /* Optional parameters need to be read before cipher constructor */
3218 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3223 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3227 if (crypt_integrity_aead(cc)) {
3228 cc->dmreq_start = sizeof(struct aead_request);
3229 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3230 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3232 cc->dmreq_start = sizeof(struct skcipher_request);
3233 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3234 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3236 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3238 if (align_mask < CRYPTO_MINALIGN) {
3239 /* Allocate the padding exactly */
3240 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3244 * If the cipher requires greater alignment than kmalloc
3245 * alignment, we don't know the exact position of the
3246 * initialization vector. We must assume worst case.
3248 iv_size_padding = align_mask;
3251 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3252 additional_req_size = sizeof(struct dm_crypt_request) +
3253 iv_size_padding + cc->iv_size +
3256 sizeof(unsigned int);
3258 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3260 ti->error = "Cannot allocate crypt request mempool";
3264 cc->per_bio_data_size = ti->per_io_data_size =
3265 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3266 ARCH_KMALLOC_MINALIGN);
3268 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3270 ti->error = "Cannot allocate page mempool";
3274 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3276 ti->error = "Cannot allocate crypt bioset";
3280 mutex_init(&cc->bio_alloc_lock);
3283 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3284 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3285 ti->error = "Invalid iv_offset sector";
3288 cc->iv_offset = tmpll;
3290 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3292 ti->error = "Device lookup failed";
3297 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3298 ti->error = "Invalid device sector";
3303 if (bdev_is_zoned(cc->dev->bdev)) {
3305 * For zoned block devices, we need to preserve the issuer write
3306 * ordering. To do so, disable write workqueues and force inline
3307 * encryption completion.
3309 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3310 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3313 * All zone append writes to a zone of a zoned block device will
3314 * have the same BIO sector, the start of the zone. When the
3315 * cypher IV mode uses sector values, all data targeting a
3316 * zone will be encrypted using the first sector numbers of the
3317 * zone. This will not result in write errors but will
3318 * cause most reads to fail as reads will use the sector values
3319 * for the actual data locations, resulting in IV mismatch.
3320 * To avoid this problem, ask DM core to emulate zone append
3321 * operations with regular writes.
3323 DMDEBUG("Zone append operations will be emulated");
3324 ti->emulate_zone_append = true;
3327 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3328 ret = crypt_integrity_ctr(cc, ti);
3332 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3333 if (!cc->tag_pool_max_sectors)
3334 cc->tag_pool_max_sectors = 1;
3336 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3337 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3339 ti->error = "Cannot allocate integrity tags mempool";
3343 cc->tag_pool_max_sectors <<= cc->sector_shift;
3347 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3348 if (!cc->io_queue) {
3349 ti->error = "Couldn't create kcryptd io queue";
3353 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3354 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3357 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3358 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3359 num_online_cpus(), devname);
3360 if (!cc->crypt_queue) {
3361 ti->error = "Couldn't create kcryptd queue";
3365 spin_lock_init(&cc->write_thread_lock);
3366 cc->write_tree = RB_ROOT;
3368 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3369 if (IS_ERR(cc->write_thread)) {
3370 ret = PTR_ERR(cc->write_thread);
3371 cc->write_thread = NULL;
3372 ti->error = "Couldn't spawn write thread";
3376 ti->num_flush_bios = 1;
3377 ti->limit_swap_bios = true;
3378 ti->accounts_remapped_io = true;
3380 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3384 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3389 static int crypt_map(struct dm_target *ti, struct bio *bio)
3391 struct dm_crypt_io *io;
3392 struct crypt_config *cc = ti->private;
3395 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3396 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3397 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3399 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3400 bio_op(bio) == REQ_OP_DISCARD)) {
3401 bio_set_dev(bio, cc->dev->bdev);
3402 if (bio_sectors(bio))
3403 bio->bi_iter.bi_sector = cc->start +
3404 dm_target_offset(ti, bio->bi_iter.bi_sector);
3405 return DM_MAPIO_REMAPPED;
3409 * Check if bio is too large, split as needed.
3411 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3412 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3413 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3416 * Ensure that bio is a multiple of internal sector encryption size
3417 * and is aligned to this size as defined in IO hints.
3419 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3420 return DM_MAPIO_KILL;
3422 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3423 return DM_MAPIO_KILL;
3425 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3426 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3428 if (cc->on_disk_tag_size) {
3429 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3431 if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3432 io->integrity_metadata = NULL;
3434 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3436 if (unlikely(!io->integrity_metadata)) {
3437 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3438 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3439 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3440 io->integrity_metadata_from_pool = true;
3444 if (crypt_integrity_aead(cc))
3445 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3447 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3449 if (bio_data_dir(io->base_bio) == READ) {
3450 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3451 kcryptd_queue_read(io);
3453 kcryptd_queue_crypt(io);
3455 return DM_MAPIO_SUBMITTED;
3458 static char hex2asc(unsigned char c)
3460 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3463 static void crypt_status(struct dm_target *ti, status_type_t type,
3464 unsigned int status_flags, char *result, unsigned int maxlen)
3466 struct crypt_config *cc = ti->private;
3467 unsigned int i, sz = 0;
3468 int num_feature_args = 0;
3471 case STATUSTYPE_INFO:
3475 case STATUSTYPE_TABLE:
3476 DMEMIT("%s ", cc->cipher_string);
3478 if (cc->key_size > 0) {
3480 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3482 for (i = 0; i < cc->key_size; i++) {
3483 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3484 hex2asc(cc->key[i] & 0xf));
3490 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3491 cc->dev->name, (unsigned long long)cc->start);
3493 num_feature_args += !!ti->num_discard_bios;
3494 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3495 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3496 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3497 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3498 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3499 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3500 if (cc->on_disk_tag_size)
3502 if (num_feature_args) {
3503 DMEMIT(" %d", num_feature_args);
3504 if (ti->num_discard_bios)
3505 DMEMIT(" allow_discards");
3506 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3507 DMEMIT(" same_cpu_crypt");
3508 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3509 DMEMIT(" submit_from_crypt_cpus");
3510 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3511 DMEMIT(" no_read_workqueue");
3512 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3513 DMEMIT(" no_write_workqueue");
3514 if (cc->on_disk_tag_size)
3515 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3516 if (cc->sector_size != (1 << SECTOR_SHIFT))
3517 DMEMIT(" sector_size:%d", cc->sector_size);
3518 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3519 DMEMIT(" iv_large_sectors");
3523 case STATUSTYPE_IMA:
3524 DMEMIT_TARGET_NAME_VERSION(ti->type);
3525 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3526 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3527 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3529 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3531 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3533 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3536 if (cc->on_disk_tag_size)
3537 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3538 cc->on_disk_tag_size, cc->cipher_auth);
3539 if (cc->sector_size != (1 << SECTOR_SHIFT))
3540 DMEMIT(",sector_size=%d", cc->sector_size);
3541 if (cc->cipher_string)
3542 DMEMIT(",cipher_string=%s", cc->cipher_string);
3544 DMEMIT(",key_size=%u", cc->key_size);
3545 DMEMIT(",key_parts=%u", cc->key_parts);
3546 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3547 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3553 static void crypt_postsuspend(struct dm_target *ti)
3555 struct crypt_config *cc = ti->private;
3557 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3560 static int crypt_preresume(struct dm_target *ti)
3562 struct crypt_config *cc = ti->private;
3564 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3565 DMERR("aborting resume - crypt key is not set.");
3572 static void crypt_resume(struct dm_target *ti)
3574 struct crypt_config *cc = ti->private;
3576 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3579 /* Message interface
3583 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3584 char *result, unsigned int maxlen)
3586 struct crypt_config *cc = ti->private;
3587 int key_size, ret = -EINVAL;
3592 if (!strcasecmp(argv[0], "key")) {
3593 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3594 DMWARN("not suspended during key manipulation.");
3597 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3598 /* The key size may not be changed. */
3599 key_size = get_key_size(&argv[2]);
3600 if (key_size < 0 || cc->key_size != key_size) {
3601 memset(argv[2], '0', strlen(argv[2]));
3605 ret = crypt_set_key(cc, argv[2]);
3608 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3609 ret = cc->iv_gen_ops->init(cc);
3610 /* wipe the kernel key payload copy */
3612 memset(cc->key, 0, cc->key_size * sizeof(u8));
3615 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3616 return crypt_wipe_key(cc);
3620 DMWARN("unrecognised message received.");
3624 static int crypt_iterate_devices(struct dm_target *ti,
3625 iterate_devices_callout_fn fn, void *data)
3627 struct crypt_config *cc = ti->private;
3629 return fn(ti, cc->dev, cc->start, ti->len, data);
3632 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3634 struct crypt_config *cc = ti->private;
3637 * Unfortunate constraint that is required to avoid the potential
3638 * for exceeding underlying device's max_segments limits -- due to
3639 * crypt_alloc_buffer() possibly allocating pages for the encryption
3640 * bio that are not as physically contiguous as the original bio.
3642 limits->max_segment_size = PAGE_SIZE;
3644 limits->logical_block_size =
3645 max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3646 limits->physical_block_size =
3647 max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3648 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3649 limits->dma_alignment = limits->logical_block_size - 1;
3652 static struct target_type crypt_target = {
3654 .version = {1, 24, 0},
3655 .module = THIS_MODULE,
3658 .features = DM_TARGET_ZONED_HM,
3659 .report_zones = crypt_report_zones,
3661 .status = crypt_status,
3662 .postsuspend = crypt_postsuspend,
3663 .preresume = crypt_preresume,
3664 .resume = crypt_resume,
3665 .message = crypt_message,
3666 .iterate_devices = crypt_iterate_devices,
3667 .io_hints = crypt_io_hints,
3671 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3672 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3673 MODULE_LICENSE("GPL");