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 = sizeof(*req) + crypto_skcipher_reqsize(tfm);
757 reqsize = ALIGN(reqsize, __alignof__(__le64));
759 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
763 skcipher_request_set_tfm(req, tfm);
765 buf = (u8 *)req + reqsize;
766 memset(buf, 0, cc->iv_size);
767 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
769 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
770 sg_init_one(&dst, iv, cc->iv_size);
771 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
772 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
773 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
774 kfree_sensitive(req);
779 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
781 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
783 crypto_free_skcipher(elephant->tfm);
784 elephant->tfm = NULL;
787 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
790 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
793 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
794 CRYPTO_ALG_ALLOCATES_MEMORY);
795 if (IS_ERR(elephant->tfm)) {
796 r = PTR_ERR(elephant->tfm);
797 elephant->tfm = NULL;
801 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
803 crypt_iv_elephant_dtr(cc);
807 static void diffuser_disk_to_cpu(u32 *d, size_t n)
809 #ifndef __LITTLE_ENDIAN
812 for (i = 0; i < n; i++)
813 d[i] = le32_to_cpu((__le32)d[i]);
817 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
819 #ifndef __LITTLE_ENDIAN
822 for (i = 0; i < n; i++)
823 d[i] = cpu_to_le32((u32)d[i]);
827 static void diffuser_a_decrypt(u32 *d, size_t n)
831 for (i = 0; i < 5; i++) {
836 while (i1 < (n - 1)) {
837 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
843 d[i1] += d[i2] ^ d[i3];
849 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
852 d[i1] += d[i2] ^ d[i3];
858 static void diffuser_a_encrypt(u32 *d, size_t n)
862 for (i = 0; i < 5; i++) {
868 d[i1] -= d[i2] ^ d[i3];
871 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
877 d[i1] -= d[i2] ^ d[i3];
883 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
889 static void diffuser_b_decrypt(u32 *d, size_t n)
893 for (i = 0; i < 3; i++) {
898 while (i1 < (n - 1)) {
899 d[i1] += d[i2] ^ d[i3];
902 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
908 d[i1] += d[i2] ^ d[i3];
914 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
920 static void diffuser_b_encrypt(u32 *d, size_t n)
924 for (i = 0; i < 3; i++) {
930 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
936 d[i1] -= d[i2] ^ d[i3];
942 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
945 d[i1] -= d[i2] ^ d[i3];
951 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
953 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
954 u8 *es, *ks, *data, *data2, *data_offset;
955 struct skcipher_request *req;
956 struct scatterlist *sg, *sg2, src, dst;
957 DECLARE_CRYPTO_WAIT(wait);
960 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
961 es = kzalloc(16, GFP_NOIO); /* Key for AES */
962 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
964 if (!req || !es || !ks) {
969 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
972 sg_init_one(&src, es, 16);
973 sg_init_one(&dst, ks, 16);
974 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
975 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
976 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
982 sg_init_one(&dst, &ks[16], 16);
983 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
987 sg = crypt_get_sg_data(cc, dmreq->sg_out);
988 data = kmap_local_page(sg_page(sg));
989 data_offset = data + sg->offset;
991 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
992 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
993 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
994 data2 = kmap_local_page(sg_page(sg2));
995 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
999 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
1000 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1001 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1002 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1003 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1006 for (i = 0; i < (cc->sector_size / 32); i++)
1007 crypto_xor(data_offset + i * 32, ks, 32);
1009 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1010 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1011 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1012 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1013 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1018 kfree_sensitive(ks);
1019 kfree_sensitive(es);
1020 skcipher_request_free(req);
1024 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1025 struct dm_crypt_request *dmreq)
1029 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1030 r = crypt_iv_elephant(cc, dmreq);
1035 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1038 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1039 struct dm_crypt_request *dmreq)
1041 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1042 return crypt_iv_elephant(cc, dmreq);
1047 static int crypt_iv_elephant_init(struct crypt_config *cc)
1049 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1050 int key_offset = cc->key_size - cc->key_extra_size;
1052 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1055 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1057 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1058 u8 key[ELEPHANT_MAX_KEY_SIZE];
1060 memset(key, 0, cc->key_extra_size);
1061 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1064 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1065 .generator = crypt_iv_plain_gen
1068 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1069 .generator = crypt_iv_plain64_gen
1072 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1073 .generator = crypt_iv_plain64be_gen
1076 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1077 .generator = crypt_iv_essiv_gen
1080 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1081 .ctr = crypt_iv_benbi_ctr,
1082 .dtr = crypt_iv_benbi_dtr,
1083 .generator = crypt_iv_benbi_gen
1086 static const struct crypt_iv_operations crypt_iv_null_ops = {
1087 .generator = crypt_iv_null_gen
1090 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1091 .ctr = crypt_iv_lmk_ctr,
1092 .dtr = crypt_iv_lmk_dtr,
1093 .init = crypt_iv_lmk_init,
1094 .wipe = crypt_iv_lmk_wipe,
1095 .generator = crypt_iv_lmk_gen,
1096 .post = crypt_iv_lmk_post
1099 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1100 .ctr = crypt_iv_tcw_ctr,
1101 .dtr = crypt_iv_tcw_dtr,
1102 .init = crypt_iv_tcw_init,
1103 .wipe = crypt_iv_tcw_wipe,
1104 .generator = crypt_iv_tcw_gen,
1105 .post = crypt_iv_tcw_post
1108 static const struct crypt_iv_operations crypt_iv_random_ops = {
1109 .generator = crypt_iv_random_gen
1112 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1113 .ctr = crypt_iv_eboiv_ctr,
1114 .generator = crypt_iv_eboiv_gen
1117 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1118 .ctr = crypt_iv_elephant_ctr,
1119 .dtr = crypt_iv_elephant_dtr,
1120 .init = crypt_iv_elephant_init,
1121 .wipe = crypt_iv_elephant_wipe,
1122 .generator = crypt_iv_elephant_gen,
1123 .post = crypt_iv_elephant_post
1127 * Integrity extensions
1129 static bool crypt_integrity_aead(struct crypt_config *cc)
1131 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1134 static bool crypt_integrity_hmac(struct crypt_config *cc)
1136 return crypt_integrity_aead(cc) && cc->key_mac_size;
1139 /* Get sg containing data */
1140 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1141 struct scatterlist *sg)
1143 if (unlikely(crypt_integrity_aead(cc)))
1149 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1151 struct bio_integrity_payload *bip;
1152 unsigned int tag_len;
1155 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1158 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1160 return PTR_ERR(bip);
1162 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
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 * In order to reduce allocation overhead, we try to allocate compound pages in
1673 * the first pass. If they are not available, we fall back to the mempool.
1675 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1677 struct crypt_config *cc = io->cc;
1679 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1680 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1681 unsigned int remaining_size;
1682 unsigned int order = MAX_ORDER - 1;
1685 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1686 mutex_lock(&cc->bio_alloc_lock);
1688 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1690 clone->bi_private = io;
1691 clone->bi_end_io = crypt_endio;
1693 remaining_size = size;
1695 while (remaining_size) {
1697 unsigned size_to_add;
1698 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1699 order = min(order, remaining_order);
1702 pages = alloc_pages(gfp_mask
1703 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1705 if (likely(pages != NULL))
1710 pages = mempool_alloc(&cc->page_pool, gfp_mask);
1712 crypt_free_buffer_pages(cc, clone);
1714 gfp_mask |= __GFP_DIRECT_RECLAIM;
1720 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1721 __bio_add_page(clone, pages, size_to_add, 0);
1722 remaining_size -= size_to_add;
1725 /* Allocate space for integrity tags */
1726 if (dm_crypt_integrity_io_alloc(io, clone)) {
1727 crypt_free_buffer_pages(cc, clone);
1732 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1733 mutex_unlock(&cc->bio_alloc_lock);
1738 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1740 struct folio_iter fi;
1742 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1743 bio_for_each_folio_all(fi, clone) {
1744 if (folio_test_large(fi.folio))
1745 folio_put(fi.folio);
1747 mempool_free(&fi.folio->page, &cc->page_pool);
1752 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1753 struct bio *bio, sector_t sector)
1757 io->sector = sector;
1759 io->ctx.r.req = NULL;
1760 io->integrity_metadata = NULL;
1761 io->integrity_metadata_from_pool = false;
1762 io->in_tasklet = false;
1763 atomic_set(&io->io_pending, 0);
1766 static void crypt_inc_pending(struct dm_crypt_io *io)
1768 atomic_inc(&io->io_pending);
1771 static void kcryptd_io_bio_endio(struct work_struct *work)
1773 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1775 bio_endio(io->base_bio);
1779 * One of the bios was finished. Check for completion of
1780 * the whole request and correctly clean up the buffer.
1782 static void crypt_dec_pending(struct dm_crypt_io *io)
1784 struct crypt_config *cc = io->cc;
1785 struct bio *base_bio = io->base_bio;
1786 blk_status_t error = io->error;
1788 if (!atomic_dec_and_test(&io->io_pending))
1792 crypt_free_req(cc, io->ctx.r.req, base_bio);
1794 if (unlikely(io->integrity_metadata_from_pool))
1795 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1797 kfree(io->integrity_metadata);
1799 base_bio->bi_status = error;
1802 * If we are running this function from our tasklet,
1803 * we can't call bio_endio() here, because it will call
1804 * clone_endio() from dm.c, which in turn will
1805 * free the current struct dm_crypt_io structure with
1806 * our tasklet. In this case we need to delay bio_endio()
1807 * execution to after the tasklet is done and dequeued.
1809 if (io->in_tasklet) {
1810 INIT_WORK(&io->work, kcryptd_io_bio_endio);
1811 queue_work(cc->io_queue, &io->work);
1815 bio_endio(base_bio);
1819 * kcryptd/kcryptd_io:
1821 * Needed because it would be very unwise to do decryption in an
1822 * interrupt context.
1824 * kcryptd performs the actual encryption or decryption.
1826 * kcryptd_io performs the IO submission.
1828 * They must be separated as otherwise the final stages could be
1829 * starved by new requests which can block in the first stages due
1830 * to memory allocation.
1832 * The work is done per CPU global for all dm-crypt instances.
1833 * They should not depend on each other and do not block.
1835 static void crypt_endio(struct bio *clone)
1837 struct dm_crypt_io *io = clone->bi_private;
1838 struct crypt_config *cc = io->cc;
1839 unsigned int rw = bio_data_dir(clone);
1843 * free the processed pages
1846 crypt_free_buffer_pages(cc, clone);
1848 error = clone->bi_status;
1851 if (rw == READ && !error) {
1852 kcryptd_queue_crypt(io);
1856 if (unlikely(error))
1859 crypt_dec_pending(io);
1862 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1864 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1866 struct crypt_config *cc = io->cc;
1870 * We need the original biovec array in order to decrypt the whole bio
1871 * data *afterwards* -- thanks to immutable biovecs we don't need to
1872 * worry about the block layer modifying the biovec array; so leverage
1873 * bio_alloc_clone().
1875 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1878 clone->bi_private = io;
1879 clone->bi_end_io = crypt_endio;
1881 crypt_inc_pending(io);
1883 clone->bi_iter.bi_sector = cc->start + io->sector;
1885 if (dm_crypt_integrity_io_alloc(io, clone)) {
1886 crypt_dec_pending(io);
1891 dm_submit_bio_remap(io->base_bio, clone);
1895 static void kcryptd_io_read_work(struct work_struct *work)
1897 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1899 crypt_inc_pending(io);
1900 if (kcryptd_io_read(io, GFP_NOIO))
1901 io->error = BLK_STS_RESOURCE;
1902 crypt_dec_pending(io);
1905 static void kcryptd_queue_read(struct dm_crypt_io *io)
1907 struct crypt_config *cc = io->cc;
1909 INIT_WORK(&io->work, kcryptd_io_read_work);
1910 queue_work(cc->io_queue, &io->work);
1913 static void kcryptd_io_write(struct dm_crypt_io *io)
1915 struct bio *clone = io->ctx.bio_out;
1917 dm_submit_bio_remap(io->base_bio, clone);
1920 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1922 static int dmcrypt_write(void *data)
1924 struct crypt_config *cc = data;
1925 struct dm_crypt_io *io;
1928 struct rb_root write_tree;
1929 struct blk_plug plug;
1931 spin_lock_irq(&cc->write_thread_lock);
1934 if (!RB_EMPTY_ROOT(&cc->write_tree))
1937 set_current_state(TASK_INTERRUPTIBLE);
1939 spin_unlock_irq(&cc->write_thread_lock);
1941 if (unlikely(kthread_should_stop())) {
1942 set_current_state(TASK_RUNNING);
1948 set_current_state(TASK_RUNNING);
1949 spin_lock_irq(&cc->write_thread_lock);
1950 goto continue_locked;
1953 write_tree = cc->write_tree;
1954 cc->write_tree = RB_ROOT;
1955 spin_unlock_irq(&cc->write_thread_lock);
1957 BUG_ON(rb_parent(write_tree.rb_node));
1960 * Note: we cannot walk the tree here with rb_next because
1961 * the structures may be freed when kcryptd_io_write is called.
1963 blk_start_plug(&plug);
1965 io = crypt_io_from_node(rb_first(&write_tree));
1966 rb_erase(&io->rb_node, &write_tree);
1967 kcryptd_io_write(io);
1969 } while (!RB_EMPTY_ROOT(&write_tree));
1970 blk_finish_plug(&plug);
1975 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1977 struct bio *clone = io->ctx.bio_out;
1978 struct crypt_config *cc = io->cc;
1979 unsigned long flags;
1981 struct rb_node **rbp, *parent;
1983 if (unlikely(io->error)) {
1984 crypt_free_buffer_pages(cc, clone);
1986 crypt_dec_pending(io);
1990 /* crypt_convert should have filled the clone bio */
1991 BUG_ON(io->ctx.iter_out.bi_size);
1993 clone->bi_iter.bi_sector = cc->start + io->sector;
1995 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1996 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1997 dm_submit_bio_remap(io->base_bio, clone);
2001 spin_lock_irqsave(&cc->write_thread_lock, flags);
2002 if (RB_EMPTY_ROOT(&cc->write_tree))
2003 wake_up_process(cc->write_thread);
2004 rbp = &cc->write_tree.rb_node;
2006 sector = io->sector;
2009 if (sector < crypt_io_from_node(parent)->sector)
2010 rbp = &(*rbp)->rb_left;
2012 rbp = &(*rbp)->rb_right;
2014 rb_link_node(&io->rb_node, parent, rbp);
2015 rb_insert_color(&io->rb_node, &cc->write_tree);
2016 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2019 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2020 struct convert_context *ctx)
2023 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2027 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2028 * constraints so they do not need to be issued inline by
2029 * kcryptd_crypt_write_convert().
2031 switch (bio_op(ctx->bio_in)) {
2033 case REQ_OP_WRITE_ZEROES:
2040 static void kcryptd_crypt_write_continue(struct work_struct *work)
2042 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2043 struct crypt_config *cc = io->cc;
2044 struct convert_context *ctx = &io->ctx;
2046 sector_t sector = io->sector;
2049 wait_for_completion(&ctx->restart);
2050 reinit_completion(&ctx->restart);
2052 r = crypt_convert(cc, &io->ctx, true, false);
2055 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2056 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2057 /* Wait for completion signaled by kcryptd_async_done() */
2058 wait_for_completion(&ctx->restart);
2062 /* Encryption was already finished, submit io now */
2063 if (crypt_finished) {
2064 kcryptd_crypt_write_io_submit(io, 0);
2065 io->sector = sector;
2068 crypt_dec_pending(io);
2071 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2073 struct crypt_config *cc = io->cc;
2074 struct convert_context *ctx = &io->ctx;
2077 sector_t sector = io->sector;
2081 * Prevent io from disappearing until this function completes.
2083 crypt_inc_pending(io);
2084 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2086 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2087 if (unlikely(!clone)) {
2088 io->error = BLK_STS_IOERR;
2092 io->ctx.bio_out = clone;
2093 io->ctx.iter_out = clone->bi_iter;
2095 sector += bio_sectors(clone);
2097 crypt_inc_pending(io);
2098 r = crypt_convert(cc, ctx,
2099 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2101 * Crypto API backlogged the request, because its queue was full
2102 * and we're in softirq context, so continue from a workqueue
2103 * (TODO: is it actually possible to be in softirq in the write path?)
2105 if (r == BLK_STS_DEV_RESOURCE) {
2106 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2107 queue_work(cc->crypt_queue, &io->work);
2112 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2113 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2114 /* Wait for completion signaled by kcryptd_async_done() */
2115 wait_for_completion(&ctx->restart);
2119 /* Encryption was already finished, submit io now */
2120 if (crypt_finished) {
2121 kcryptd_crypt_write_io_submit(io, 0);
2122 io->sector = sector;
2126 crypt_dec_pending(io);
2129 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2131 crypt_dec_pending(io);
2134 static void kcryptd_crypt_read_continue(struct work_struct *work)
2136 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2137 struct crypt_config *cc = io->cc;
2140 wait_for_completion(&io->ctx.restart);
2141 reinit_completion(&io->ctx.restart);
2143 r = crypt_convert(cc, &io->ctx, true, false);
2147 if (atomic_dec_and_test(&io->ctx.cc_pending))
2148 kcryptd_crypt_read_done(io);
2150 crypt_dec_pending(io);
2153 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2155 struct crypt_config *cc = io->cc;
2158 crypt_inc_pending(io);
2160 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2163 r = crypt_convert(cc, &io->ctx,
2164 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2166 * Crypto API backlogged the request, because its queue was full
2167 * and we're in softirq context, so continue from a workqueue
2169 if (r == BLK_STS_DEV_RESOURCE) {
2170 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2171 queue_work(cc->crypt_queue, &io->work);
2177 if (atomic_dec_and_test(&io->ctx.cc_pending))
2178 kcryptd_crypt_read_done(io);
2180 crypt_dec_pending(io);
2183 static void kcryptd_async_done(void *data, int error)
2185 struct dm_crypt_request *dmreq = data;
2186 struct convert_context *ctx = dmreq->ctx;
2187 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2188 struct crypt_config *cc = io->cc;
2191 * A request from crypto driver backlog is going to be processed now,
2192 * finish the completion and continue in crypt_convert().
2193 * (Callback will be called for the second time for this request.)
2195 if (error == -EINPROGRESS) {
2196 complete(&ctx->restart);
2200 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2201 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2203 if (error == -EBADMSG) {
2204 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2206 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2207 ctx->bio_in->bi_bdev, s);
2208 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2210 io->error = BLK_STS_PROTECTION;
2211 } else if (error < 0)
2212 io->error = BLK_STS_IOERR;
2214 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2216 if (!atomic_dec_and_test(&ctx->cc_pending))
2220 * The request is fully completed: for inline writes, let
2221 * kcryptd_crypt_write_convert() do the IO submission.
2223 if (bio_data_dir(io->base_bio) == READ) {
2224 kcryptd_crypt_read_done(io);
2228 if (kcryptd_crypt_write_inline(cc, ctx)) {
2229 complete(&ctx->restart);
2233 kcryptd_crypt_write_io_submit(io, 1);
2236 static void kcryptd_crypt(struct work_struct *work)
2238 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2240 if (bio_data_dir(io->base_bio) == READ)
2241 kcryptd_crypt_read_convert(io);
2243 kcryptd_crypt_write_convert(io);
2246 static void kcryptd_crypt_tasklet(unsigned long work)
2248 kcryptd_crypt((struct work_struct *)work);
2251 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2253 struct crypt_config *cc = io->cc;
2255 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2256 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2258 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2259 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2260 * it is being executed with irqs disabled.
2262 if (in_hardirq() || irqs_disabled()) {
2263 io->in_tasklet = true;
2264 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2265 tasklet_schedule(&io->tasklet);
2269 kcryptd_crypt(&io->work);
2273 INIT_WORK(&io->work, kcryptd_crypt);
2274 queue_work(cc->crypt_queue, &io->work);
2277 static void crypt_free_tfms_aead(struct crypt_config *cc)
2279 if (!cc->cipher_tfm.tfms_aead)
2282 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2283 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2284 cc->cipher_tfm.tfms_aead[0] = NULL;
2287 kfree(cc->cipher_tfm.tfms_aead);
2288 cc->cipher_tfm.tfms_aead = NULL;
2291 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2295 if (!cc->cipher_tfm.tfms)
2298 for (i = 0; i < cc->tfms_count; i++)
2299 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2300 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2301 cc->cipher_tfm.tfms[i] = NULL;
2304 kfree(cc->cipher_tfm.tfms);
2305 cc->cipher_tfm.tfms = NULL;
2308 static void crypt_free_tfms(struct crypt_config *cc)
2310 if (crypt_integrity_aead(cc))
2311 crypt_free_tfms_aead(cc);
2313 crypt_free_tfms_skcipher(cc);
2316 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2321 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2322 sizeof(struct crypto_skcipher *),
2324 if (!cc->cipher_tfm.tfms)
2327 for (i = 0; i < cc->tfms_count; i++) {
2328 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2329 CRYPTO_ALG_ALLOCATES_MEMORY);
2330 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2331 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2332 crypt_free_tfms(cc);
2338 * dm-crypt performance can vary greatly depending on which crypto
2339 * algorithm implementation is used. Help people debug performance
2340 * problems by logging the ->cra_driver_name.
2342 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2343 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2347 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2351 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2352 if (!cc->cipher_tfm.tfms)
2355 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2356 CRYPTO_ALG_ALLOCATES_MEMORY);
2357 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2358 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2359 crypt_free_tfms(cc);
2363 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2364 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2368 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2370 if (crypt_integrity_aead(cc))
2371 return crypt_alloc_tfms_aead(cc, ciphermode);
2373 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2376 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2378 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2381 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2383 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2387 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2388 * the key must be for some reason in special format.
2389 * This funcion converts cc->key to this special format.
2391 static void crypt_copy_authenckey(char *p, const void *key,
2392 unsigned int enckeylen, unsigned int authkeylen)
2394 struct crypto_authenc_key_param *param;
2397 rta = (struct rtattr *)p;
2398 param = RTA_DATA(rta);
2399 param->enckeylen = cpu_to_be32(enckeylen);
2400 rta->rta_len = RTA_LENGTH(sizeof(*param));
2401 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2402 p += RTA_SPACE(sizeof(*param));
2403 memcpy(p, key + enckeylen, authkeylen);
2405 memcpy(p, key, enckeylen);
2408 static int crypt_setkey(struct crypt_config *cc)
2410 unsigned int subkey_size;
2413 /* Ignore extra keys (which are used for IV etc) */
2414 subkey_size = crypt_subkey_size(cc);
2416 if (crypt_integrity_hmac(cc)) {
2417 if (subkey_size < cc->key_mac_size)
2420 crypt_copy_authenckey(cc->authenc_key, cc->key,
2421 subkey_size - cc->key_mac_size,
2425 for (i = 0; i < cc->tfms_count; i++) {
2426 if (crypt_integrity_hmac(cc))
2427 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2428 cc->authenc_key, crypt_authenckey_size(cc));
2429 else if (crypt_integrity_aead(cc))
2430 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2431 cc->key + (i * subkey_size),
2434 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2435 cc->key + (i * subkey_size),
2441 if (crypt_integrity_hmac(cc))
2442 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2449 static bool contains_whitespace(const char *str)
2452 if (isspace(*str++))
2457 static int set_key_user(struct crypt_config *cc, struct key *key)
2459 const struct user_key_payload *ukp;
2461 ukp = user_key_payload_locked(key);
2463 return -EKEYREVOKED;
2465 if (cc->key_size != ukp->datalen)
2468 memcpy(cc->key, ukp->data, cc->key_size);
2473 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2475 const struct encrypted_key_payload *ekp;
2477 ekp = key->payload.data[0];
2479 return -EKEYREVOKED;
2481 if (cc->key_size != ekp->decrypted_datalen)
2484 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2489 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2491 const struct trusted_key_payload *tkp;
2493 tkp = key->payload.data[0];
2495 return -EKEYREVOKED;
2497 if (cc->key_size != tkp->key_len)
2500 memcpy(cc->key, tkp->key, cc->key_size);
2505 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2507 char *new_key_string, *key_desc;
2509 struct key_type *type;
2511 int (*set_key)(struct crypt_config *cc, struct key *key);
2514 * Reject key_string with whitespace. dm core currently lacks code for
2515 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2517 if (contains_whitespace(key_string)) {
2518 DMERR("whitespace chars not allowed in key string");
2522 /* look for next ':' separating key_type from key_description */
2523 key_desc = strchr(key_string, ':');
2524 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2527 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2528 type = &key_type_logon;
2529 set_key = set_key_user;
2530 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2531 type = &key_type_user;
2532 set_key = set_key_user;
2533 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2534 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2535 type = &key_type_encrypted;
2536 set_key = set_key_encrypted;
2537 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2538 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2539 type = &key_type_trusted;
2540 set_key = set_key_trusted;
2545 new_key_string = kstrdup(key_string, GFP_KERNEL);
2546 if (!new_key_string)
2549 key = request_key(type, key_desc + 1, NULL);
2551 kfree_sensitive(new_key_string);
2552 return PTR_ERR(key);
2555 down_read(&key->sem);
2557 ret = set_key(cc, key);
2561 kfree_sensitive(new_key_string);
2568 /* clear the flag since following operations may invalidate previously valid key */
2569 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2571 ret = crypt_setkey(cc);
2574 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2575 kfree_sensitive(cc->key_string);
2576 cc->key_string = new_key_string;
2578 kfree_sensitive(new_key_string);
2583 static int get_key_size(char **key_string)
2588 if (*key_string[0] != ':')
2589 return strlen(*key_string) >> 1;
2591 /* look for next ':' in key string */
2592 colon = strpbrk(*key_string + 1, ":");
2596 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2599 *key_string = colon;
2601 /* remaining key string should be :<logon|user>:<key_desc> */
2608 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2613 static int get_key_size(char **key_string)
2615 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2618 #endif /* CONFIG_KEYS */
2620 static int crypt_set_key(struct crypt_config *cc, char *key)
2623 int key_string_len = strlen(key);
2625 /* Hyphen (which gives a key_size of zero) means there is no key. */
2626 if (!cc->key_size && strcmp(key, "-"))
2629 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2630 if (key[0] == ':') {
2631 r = crypt_set_keyring_key(cc, key + 1);
2635 /* clear the flag since following operations may invalidate previously valid key */
2636 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2638 /* wipe references to any kernel keyring key */
2639 kfree_sensitive(cc->key_string);
2640 cc->key_string = NULL;
2642 /* Decode key from its hex representation. */
2643 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2646 r = crypt_setkey(cc);
2648 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2651 /* Hex key string not needed after here, so wipe it. */
2652 memset(key, '0', key_string_len);
2657 static int crypt_wipe_key(struct crypt_config *cc)
2661 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2662 get_random_bytes(&cc->key, cc->key_size);
2664 /* Wipe IV private keys */
2665 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2666 r = cc->iv_gen_ops->wipe(cc);
2671 kfree_sensitive(cc->key_string);
2672 cc->key_string = NULL;
2673 r = crypt_setkey(cc);
2674 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2679 static void crypt_calculate_pages_per_client(void)
2681 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2683 if (!dm_crypt_clients_n)
2686 pages /= dm_crypt_clients_n;
2687 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2688 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2689 dm_crypt_pages_per_client = pages;
2692 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2694 struct crypt_config *cc = pool_data;
2698 * Note, percpu_counter_read_positive() may over (and under) estimate
2699 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2700 * but avoids potential spinlock contention of an exact result.
2702 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2703 likely(gfp_mask & __GFP_NORETRY))
2706 page = alloc_page(gfp_mask);
2707 if (likely(page != NULL))
2708 percpu_counter_add(&cc->n_allocated_pages, 1);
2713 static void crypt_page_free(void *page, void *pool_data)
2715 struct crypt_config *cc = pool_data;
2718 percpu_counter_sub(&cc->n_allocated_pages, 1);
2721 static void crypt_dtr(struct dm_target *ti)
2723 struct crypt_config *cc = ti->private;
2730 if (cc->write_thread)
2731 kthread_stop(cc->write_thread);
2734 destroy_workqueue(cc->io_queue);
2735 if (cc->crypt_queue)
2736 destroy_workqueue(cc->crypt_queue);
2738 crypt_free_tfms(cc);
2740 bioset_exit(&cc->bs);
2742 mempool_exit(&cc->page_pool);
2743 mempool_exit(&cc->req_pool);
2744 mempool_exit(&cc->tag_pool);
2746 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2747 percpu_counter_destroy(&cc->n_allocated_pages);
2749 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2750 cc->iv_gen_ops->dtr(cc);
2753 dm_put_device(ti, cc->dev);
2755 kfree_sensitive(cc->cipher_string);
2756 kfree_sensitive(cc->key_string);
2757 kfree_sensitive(cc->cipher_auth);
2758 kfree_sensitive(cc->authenc_key);
2760 mutex_destroy(&cc->bio_alloc_lock);
2762 /* Must zero key material before freeing */
2763 kfree_sensitive(cc);
2765 spin_lock(&dm_crypt_clients_lock);
2766 WARN_ON(!dm_crypt_clients_n);
2767 dm_crypt_clients_n--;
2768 crypt_calculate_pages_per_client();
2769 spin_unlock(&dm_crypt_clients_lock);
2771 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2774 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2776 struct crypt_config *cc = ti->private;
2778 if (crypt_integrity_aead(cc))
2779 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2781 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2784 /* at least a 64 bit sector number should fit in our buffer */
2785 cc->iv_size = max(cc->iv_size,
2786 (unsigned int)(sizeof(u64) / sizeof(u8)));
2788 DMWARN("Selected cipher does not support IVs");
2792 /* Choose ivmode, see comments at iv code. */
2794 cc->iv_gen_ops = NULL;
2795 else if (strcmp(ivmode, "plain") == 0)
2796 cc->iv_gen_ops = &crypt_iv_plain_ops;
2797 else if (strcmp(ivmode, "plain64") == 0)
2798 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2799 else if (strcmp(ivmode, "plain64be") == 0)
2800 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2801 else if (strcmp(ivmode, "essiv") == 0)
2802 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2803 else if (strcmp(ivmode, "benbi") == 0)
2804 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2805 else if (strcmp(ivmode, "null") == 0)
2806 cc->iv_gen_ops = &crypt_iv_null_ops;
2807 else if (strcmp(ivmode, "eboiv") == 0)
2808 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2809 else if (strcmp(ivmode, "elephant") == 0) {
2810 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2812 cc->key_extra_size = cc->key_size / 2;
2813 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2815 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2816 } else if (strcmp(ivmode, "lmk") == 0) {
2817 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2819 * Version 2 and 3 is recognised according
2820 * to length of provided multi-key string.
2821 * If present (version 3), last key is used as IV seed.
2822 * All keys (including IV seed) are always the same size.
2824 if (cc->key_size % cc->key_parts) {
2826 cc->key_extra_size = cc->key_size / cc->key_parts;
2828 } else if (strcmp(ivmode, "tcw") == 0) {
2829 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2830 cc->key_parts += 2; /* IV + whitening */
2831 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2832 } else if (strcmp(ivmode, "random") == 0) {
2833 cc->iv_gen_ops = &crypt_iv_random_ops;
2834 /* Need storage space in integrity fields. */
2835 cc->integrity_iv_size = cc->iv_size;
2837 ti->error = "Invalid IV mode";
2845 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2846 * The HMAC is needed to calculate tag size (HMAC digest size).
2847 * This should be probably done by crypto-api calls (once available...)
2849 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2851 char *start, *end, *mac_alg = NULL;
2852 struct crypto_ahash *mac;
2854 if (!strstarts(cipher_api, "authenc("))
2857 start = strchr(cipher_api, '(');
2858 end = strchr(cipher_api, ',');
2859 if (!start || !end || ++start > end)
2862 mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2865 strncpy(mac_alg, start, end - start);
2867 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2871 return PTR_ERR(mac);
2873 cc->key_mac_size = crypto_ahash_digestsize(mac);
2874 crypto_free_ahash(mac);
2876 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2877 if (!cc->authenc_key)
2883 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2884 char **ivmode, char **ivopts)
2886 struct crypt_config *cc = ti->private;
2887 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2893 * New format (capi: prefix)
2894 * capi:cipher_api_spec-iv:ivopts
2896 tmp = &cipher_in[strlen("capi:")];
2898 /* Separate IV options if present, it can contain another '-' in hash name */
2899 *ivopts = strrchr(tmp, ':');
2905 *ivmode = strrchr(tmp, '-');
2910 /* The rest is crypto API spec */
2913 /* Alloc AEAD, can be used only in new format. */
2914 if (crypt_integrity_aead(cc)) {
2915 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2917 ti->error = "Invalid AEAD cipher spec";
2922 if (*ivmode && !strcmp(*ivmode, "lmk"))
2923 cc->tfms_count = 64;
2925 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2927 ti->error = "Digest algorithm missing for ESSIV mode";
2930 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2931 cipher_api, *ivopts);
2932 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2933 ti->error = "Cannot allocate cipher string";
2939 cc->key_parts = cc->tfms_count;
2941 /* Allocate cipher */
2942 ret = crypt_alloc_tfms(cc, cipher_api);
2944 ti->error = "Error allocating crypto tfm";
2948 if (crypt_integrity_aead(cc))
2949 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2951 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2956 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2957 char **ivmode, char **ivopts)
2959 struct crypt_config *cc = ti->private;
2960 char *tmp, *cipher, *chainmode, *keycount;
2961 char *cipher_api = NULL;
2965 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2966 ti->error = "Bad cipher specification";
2971 * Legacy dm-crypt cipher specification
2972 * cipher[:keycount]-mode-iv:ivopts
2975 keycount = strsep(&tmp, "-");
2976 cipher = strsep(&keycount, ":");
2980 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2981 !is_power_of_2(cc->tfms_count)) {
2982 ti->error = "Bad cipher key count specification";
2985 cc->key_parts = cc->tfms_count;
2987 chainmode = strsep(&tmp, "-");
2988 *ivmode = strsep(&tmp, ":");
2992 * For compatibility with the original dm-crypt mapping format, if
2993 * only the cipher name is supplied, use cbc-plain.
2995 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3000 if (strcmp(chainmode, "ecb") && !*ivmode) {
3001 ti->error = "IV mechanism required";
3005 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3009 if (*ivmode && !strcmp(*ivmode, "essiv")) {
3011 ti->error = "Digest algorithm missing for ESSIV mode";
3015 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3016 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3018 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3019 "%s(%s)", chainmode, cipher);
3021 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3026 /* Allocate cipher */
3027 ret = crypt_alloc_tfms(cc, cipher_api);
3029 ti->error = "Error allocating crypto tfm";
3037 ti->error = "Cannot allocate cipher strings";
3041 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3043 struct crypt_config *cc = ti->private;
3044 char *ivmode = NULL, *ivopts = NULL;
3047 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3048 if (!cc->cipher_string) {
3049 ti->error = "Cannot allocate cipher strings";
3053 if (strstarts(cipher_in, "capi:"))
3054 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3056 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3061 ret = crypt_ctr_ivmode(ti, ivmode);
3065 /* Initialize and set key */
3066 ret = crypt_set_key(cc, key);
3068 ti->error = "Error decoding and setting key";
3073 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3074 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3076 ti->error = "Error creating IV";
3081 /* Initialize IV (set keys for ESSIV etc) */
3082 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3083 ret = cc->iv_gen_ops->init(cc);
3085 ti->error = "Error initialising IV";
3090 /* wipe the kernel key payload copy */
3092 memset(cc->key, 0, cc->key_size * sizeof(u8));
3097 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3099 struct crypt_config *cc = ti->private;
3100 struct dm_arg_set as;
3101 static const struct dm_arg _args[] = {
3102 {0, 8, "Invalid number of feature args"},
3104 unsigned int opt_params, val;
3105 const char *opt_string, *sval;
3109 /* Optional parameters */
3113 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3117 while (opt_params--) {
3118 opt_string = dm_shift_arg(&as);
3120 ti->error = "Not enough feature arguments";
3124 if (!strcasecmp(opt_string, "allow_discards"))
3125 ti->num_discard_bios = 1;
3127 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3128 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3130 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3131 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3132 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3133 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3134 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3135 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3136 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3137 if (val == 0 || val > MAX_TAG_SIZE) {
3138 ti->error = "Invalid integrity arguments";
3141 cc->on_disk_tag_size = val;
3142 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3143 if (!strcasecmp(sval, "aead")) {
3144 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3145 } else if (strcasecmp(sval, "none")) {
3146 ti->error = "Unknown integrity profile";
3150 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3151 if (!cc->cipher_auth)
3153 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3154 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3155 cc->sector_size > 4096 ||
3156 (cc->sector_size & (cc->sector_size - 1))) {
3157 ti->error = "Invalid feature value for sector_size";
3160 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3161 ti->error = "Device size is not multiple of sector_size feature";
3164 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3165 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3166 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3168 ti->error = "Invalid feature arguments";
3176 #ifdef CONFIG_BLK_DEV_ZONED
3177 static int crypt_report_zones(struct dm_target *ti,
3178 struct dm_report_zones_args *args, unsigned int nr_zones)
3180 struct crypt_config *cc = ti->private;
3182 return dm_report_zones(cc->dev->bdev, cc->start,
3183 cc->start + dm_target_offset(ti, args->next_sector),
3187 #define crypt_report_zones NULL
3191 * Construct an encryption mapping:
3192 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3194 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3196 struct crypt_config *cc;
3197 const char *devname = dm_table_device_name(ti->table);
3199 unsigned int align_mask;
3200 unsigned long long tmpll;
3202 size_t iv_size_padding, additional_req_size;
3206 ti->error = "Not enough arguments";
3210 key_size = get_key_size(&argv[1]);
3212 ti->error = "Cannot parse key size";
3216 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3218 ti->error = "Cannot allocate encryption context";
3221 cc->key_size = key_size;
3222 cc->sector_size = (1 << SECTOR_SHIFT);
3223 cc->sector_shift = 0;
3227 spin_lock(&dm_crypt_clients_lock);
3228 dm_crypt_clients_n++;
3229 crypt_calculate_pages_per_client();
3230 spin_unlock(&dm_crypt_clients_lock);
3232 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3236 /* Optional parameters need to be read before cipher constructor */
3238 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3243 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3247 if (crypt_integrity_aead(cc)) {
3248 cc->dmreq_start = sizeof(struct aead_request);
3249 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3250 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3252 cc->dmreq_start = sizeof(struct skcipher_request);
3253 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3254 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3256 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3258 if (align_mask < CRYPTO_MINALIGN) {
3259 /* Allocate the padding exactly */
3260 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3264 * If the cipher requires greater alignment than kmalloc
3265 * alignment, we don't know the exact position of the
3266 * initialization vector. We must assume worst case.
3268 iv_size_padding = align_mask;
3271 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3272 additional_req_size = sizeof(struct dm_crypt_request) +
3273 iv_size_padding + cc->iv_size +
3276 sizeof(unsigned int);
3278 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3280 ti->error = "Cannot allocate crypt request mempool";
3284 cc->per_bio_data_size = ti->per_io_data_size =
3285 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3288 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3290 ti->error = "Cannot allocate page mempool";
3294 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3296 ti->error = "Cannot allocate crypt bioset";
3300 mutex_init(&cc->bio_alloc_lock);
3303 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3304 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3305 ti->error = "Invalid iv_offset sector";
3308 cc->iv_offset = tmpll;
3310 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3312 ti->error = "Device lookup failed";
3317 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3318 ti->error = "Invalid device sector";
3323 if (bdev_is_zoned(cc->dev->bdev)) {
3325 * For zoned block devices, we need to preserve the issuer write
3326 * ordering. To do so, disable write workqueues and force inline
3327 * encryption completion.
3329 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3330 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3333 * All zone append writes to a zone of a zoned block device will
3334 * have the same BIO sector, the start of the zone. When the
3335 * cypher IV mode uses sector values, all data targeting a
3336 * zone will be encrypted using the first sector numbers of the
3337 * zone. This will not result in write errors but will
3338 * cause most reads to fail as reads will use the sector values
3339 * for the actual data locations, resulting in IV mismatch.
3340 * To avoid this problem, ask DM core to emulate zone append
3341 * operations with regular writes.
3343 DMDEBUG("Zone append operations will be emulated");
3344 ti->emulate_zone_append = true;
3347 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3348 ret = crypt_integrity_ctr(cc, ti);
3352 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3353 if (!cc->tag_pool_max_sectors)
3354 cc->tag_pool_max_sectors = 1;
3356 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3357 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3359 ti->error = "Cannot allocate integrity tags mempool";
3363 cc->tag_pool_max_sectors <<= cc->sector_shift;
3367 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3368 if (!cc->io_queue) {
3369 ti->error = "Couldn't create kcryptd io queue";
3373 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3374 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3377 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3378 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3379 num_online_cpus(), devname);
3380 if (!cc->crypt_queue) {
3381 ti->error = "Couldn't create kcryptd queue";
3385 spin_lock_init(&cc->write_thread_lock);
3386 cc->write_tree = RB_ROOT;
3388 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3389 if (IS_ERR(cc->write_thread)) {
3390 ret = PTR_ERR(cc->write_thread);
3391 cc->write_thread = NULL;
3392 ti->error = "Couldn't spawn write thread";
3396 ti->num_flush_bios = 1;
3397 ti->limit_swap_bios = true;
3398 ti->accounts_remapped_io = true;
3400 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3404 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3409 static int crypt_map(struct dm_target *ti, struct bio *bio)
3411 struct dm_crypt_io *io;
3412 struct crypt_config *cc = ti->private;
3415 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3416 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3417 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3419 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3420 bio_op(bio) == REQ_OP_DISCARD)) {
3421 bio_set_dev(bio, cc->dev->bdev);
3422 if (bio_sectors(bio))
3423 bio->bi_iter.bi_sector = cc->start +
3424 dm_target_offset(ti, bio->bi_iter.bi_sector);
3425 return DM_MAPIO_REMAPPED;
3429 * Check if bio is too large, split as needed.
3431 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3432 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3433 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3436 * Ensure that bio is a multiple of internal sector encryption size
3437 * and is aligned to this size as defined in IO hints.
3439 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3440 return DM_MAPIO_KILL;
3442 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3443 return DM_MAPIO_KILL;
3445 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3446 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3448 if (cc->on_disk_tag_size) {
3449 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3451 if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3452 io->integrity_metadata = NULL;
3454 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3456 if (unlikely(!io->integrity_metadata)) {
3457 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3458 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3459 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3460 io->integrity_metadata_from_pool = true;
3464 if (crypt_integrity_aead(cc))
3465 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3467 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3469 if (bio_data_dir(io->base_bio) == READ) {
3470 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3471 kcryptd_queue_read(io);
3473 kcryptd_queue_crypt(io);
3475 return DM_MAPIO_SUBMITTED;
3478 static char hex2asc(unsigned char c)
3480 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3483 static void crypt_status(struct dm_target *ti, status_type_t type,
3484 unsigned int status_flags, char *result, unsigned int maxlen)
3486 struct crypt_config *cc = ti->private;
3487 unsigned int i, sz = 0;
3488 int num_feature_args = 0;
3491 case STATUSTYPE_INFO:
3495 case STATUSTYPE_TABLE:
3496 DMEMIT("%s ", cc->cipher_string);
3498 if (cc->key_size > 0) {
3500 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3502 for (i = 0; i < cc->key_size; i++) {
3503 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3504 hex2asc(cc->key[i] & 0xf));
3510 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3511 cc->dev->name, (unsigned long long)cc->start);
3513 num_feature_args += !!ti->num_discard_bios;
3514 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3515 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3516 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3517 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3518 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3519 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3520 if (cc->on_disk_tag_size)
3522 if (num_feature_args) {
3523 DMEMIT(" %d", num_feature_args);
3524 if (ti->num_discard_bios)
3525 DMEMIT(" allow_discards");
3526 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3527 DMEMIT(" same_cpu_crypt");
3528 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3529 DMEMIT(" submit_from_crypt_cpus");
3530 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3531 DMEMIT(" no_read_workqueue");
3532 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3533 DMEMIT(" no_write_workqueue");
3534 if (cc->on_disk_tag_size)
3535 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3536 if (cc->sector_size != (1 << SECTOR_SHIFT))
3537 DMEMIT(" sector_size:%d", cc->sector_size);
3538 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3539 DMEMIT(" iv_large_sectors");
3543 case STATUSTYPE_IMA:
3544 DMEMIT_TARGET_NAME_VERSION(ti->type);
3545 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3546 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3547 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3549 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3551 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3553 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3556 if (cc->on_disk_tag_size)
3557 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3558 cc->on_disk_tag_size, cc->cipher_auth);
3559 if (cc->sector_size != (1 << SECTOR_SHIFT))
3560 DMEMIT(",sector_size=%d", cc->sector_size);
3561 if (cc->cipher_string)
3562 DMEMIT(",cipher_string=%s", cc->cipher_string);
3564 DMEMIT(",key_size=%u", cc->key_size);
3565 DMEMIT(",key_parts=%u", cc->key_parts);
3566 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3567 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3573 static void crypt_postsuspend(struct dm_target *ti)
3575 struct crypt_config *cc = ti->private;
3577 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3580 static int crypt_preresume(struct dm_target *ti)
3582 struct crypt_config *cc = ti->private;
3584 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3585 DMERR("aborting resume - crypt key is not set.");
3592 static void crypt_resume(struct dm_target *ti)
3594 struct crypt_config *cc = ti->private;
3596 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3599 /* Message interface
3603 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3604 char *result, unsigned int maxlen)
3606 struct crypt_config *cc = ti->private;
3607 int key_size, ret = -EINVAL;
3612 if (!strcasecmp(argv[0], "key")) {
3613 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3614 DMWARN("not suspended during key manipulation.");
3617 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3618 /* The key size may not be changed. */
3619 key_size = get_key_size(&argv[2]);
3620 if (key_size < 0 || cc->key_size != key_size) {
3621 memset(argv[2], '0', strlen(argv[2]));
3625 ret = crypt_set_key(cc, argv[2]);
3628 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3629 ret = cc->iv_gen_ops->init(cc);
3630 /* wipe the kernel key payload copy */
3632 memset(cc->key, 0, cc->key_size * sizeof(u8));
3635 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3636 return crypt_wipe_key(cc);
3640 DMWARN("unrecognised message received.");
3644 static int crypt_iterate_devices(struct dm_target *ti,
3645 iterate_devices_callout_fn fn, void *data)
3647 struct crypt_config *cc = ti->private;
3649 return fn(ti, cc->dev, cc->start, ti->len, data);
3652 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3654 struct crypt_config *cc = ti->private;
3657 * Unfortunate constraint that is required to avoid the potential
3658 * for exceeding underlying device's max_segments limits -- due to
3659 * crypt_alloc_buffer() possibly allocating pages for the encryption
3660 * bio that are not as physically contiguous as the original bio.
3662 limits->max_segment_size = PAGE_SIZE;
3664 limits->logical_block_size =
3665 max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3666 limits->physical_block_size =
3667 max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3668 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3669 limits->dma_alignment = limits->logical_block_size - 1;
3672 static struct target_type crypt_target = {
3674 .version = {1, 24, 0},
3675 .module = THIS_MODULE,
3678 .features = DM_TARGET_ZONED_HM,
3679 .report_zones = crypt_report_zones,
3681 .status = crypt_status,
3682 .postsuspend = crypt_postsuspend,
3683 .preresume = crypt_preresume,
3684 .resume = crypt_resume,
3685 .message = crypt_message,
3686 .iterate_devices = crypt_iterate_devices,
3687 .io_hints = crypt_io_hints,
3691 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3692 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3693 MODULE_LICENSE("GPL");