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;
56 struct bvec_iter iter_in;
58 struct bvec_iter iter_out;
62 struct skcipher_request *req;
63 struct aead_request *req_aead;
71 * per bio private data
74 struct crypt_config *cc;
76 u8 *integrity_metadata;
77 bool integrity_metadata_from_pool:1;
79 struct work_struct work;
81 struct convert_context ctx;
87 struct bvec_iter saved_bi_iter;
89 struct rb_node rb_node;
90 } CRYPTO_MINALIGN_ATTR;
92 struct dm_crypt_request {
93 struct convert_context *ctx;
94 struct scatterlist sg_in[4];
95 struct scatterlist sg_out[4];
101 struct crypt_iv_operations {
102 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
104 void (*dtr)(struct crypt_config *cc);
105 int (*init)(struct crypt_config *cc);
106 int (*wipe)(struct crypt_config *cc);
107 int (*generator)(struct crypt_config *cc, u8 *iv,
108 struct dm_crypt_request *dmreq);
109 int (*post)(struct crypt_config *cc, u8 *iv,
110 struct dm_crypt_request *dmreq);
113 struct iv_benbi_private {
117 #define LMK_SEED_SIZE 64 /* hash + 0 */
118 struct iv_lmk_private {
119 struct crypto_shash *hash_tfm;
123 #define TCW_WHITENING_SIZE 16
124 struct iv_tcw_private {
125 struct crypto_shash *crc32_tfm;
130 #define ELEPHANT_MAX_KEY_SIZE 32
131 struct iv_elephant_private {
132 struct crypto_skcipher *tfm;
136 * Crypt: maps a linear range of a block device
137 * and encrypts / decrypts at the same time.
139 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
140 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
141 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
142 DM_CRYPT_WRITE_INLINE };
145 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
146 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
147 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
151 * The fields in here must be read only after initialization.
153 struct crypt_config {
157 struct percpu_counter n_allocated_pages;
159 struct workqueue_struct *io_queue;
160 struct workqueue_struct *crypt_queue;
162 spinlock_t write_thread_lock;
163 struct task_struct *write_thread;
164 struct rb_root write_tree;
170 const struct crypt_iv_operations *iv_gen_ops;
172 struct iv_benbi_private benbi;
173 struct iv_lmk_private lmk;
174 struct iv_tcw_private tcw;
175 struct iv_elephant_private elephant;
178 unsigned int iv_size;
179 unsigned short sector_size;
180 unsigned char sector_shift;
183 struct crypto_skcipher **tfms;
184 struct crypto_aead **tfms_aead;
186 unsigned int tfms_count;
187 unsigned long cipher_flags;
190 * Layout of each crypto request:
192 * struct skcipher_request
195 * struct dm_crypt_request
199 * The padding is added so that dm_crypt_request and the IV are
202 unsigned int dmreq_start;
204 unsigned int per_bio_data_size;
207 unsigned int key_size;
208 unsigned int key_parts; /* independent parts in key buffer */
209 unsigned int key_extra_size; /* additional keys length */
210 unsigned int key_mac_size; /* MAC key size for authenc(...) */
212 unsigned int integrity_tag_size;
213 unsigned int integrity_iv_size;
214 unsigned int on_disk_tag_size;
217 * pool for per bio private data, crypto requests,
218 * encryption requeusts/buffer pages and integrity tags
220 unsigned int tag_pool_max_sectors;
226 struct mutex bio_alloc_lock;
228 u8 *authenc_key; /* space for keys in authenc() format (if used) */
229 u8 key[] __counted_by(key_size);
233 #define MAX_TAG_SIZE 480
234 #define POOL_ENTRY_SIZE 512
236 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
237 static unsigned int dm_crypt_clients_n;
238 static volatile unsigned long dm_crypt_pages_per_client;
239 #define DM_CRYPT_MEMORY_PERCENT 2
240 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
242 static void crypt_endio(struct bio *clone);
243 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
244 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
245 struct scatterlist *sg);
247 static bool crypt_integrity_aead(struct crypt_config *cc);
250 * Use this to access cipher attributes that are independent of the key.
252 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
254 return cc->cipher_tfm.tfms[0];
257 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
259 return cc->cipher_tfm.tfms_aead[0];
263 * Different IV generation algorithms:
265 * plain: the initial vector is the 32-bit little-endian version of the sector
266 * number, padded with zeros if necessary.
268 * plain64: the initial vector is the 64-bit little-endian version of the sector
269 * number, padded with zeros if necessary.
271 * plain64be: the initial vector is the 64-bit big-endian version of the sector
272 * number, padded with zeros if necessary.
274 * essiv: "encrypted sector|salt initial vector", the sector number is
275 * encrypted with the bulk cipher using a salt as key. The salt
276 * should be derived from the bulk cipher's key via hashing.
278 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
279 * (needed for LRW-32-AES and possible other narrow block modes)
281 * null: the initial vector is always zero. Provides compatibility with
282 * obsolete loop_fish2 devices. Do not use for new devices.
284 * lmk: Compatible implementation of the block chaining mode used
285 * by the Loop-AES block device encryption system
286 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
287 * It operates on full 512 byte sectors and uses CBC
288 * with an IV derived from the sector number, the data and
289 * optionally extra IV seed.
290 * This means that after decryption the first block
291 * of sector must be tweaked according to decrypted data.
292 * Loop-AES can use three encryption schemes:
293 * version 1: is plain aes-cbc mode
294 * version 2: uses 64 multikey scheme with lmk IV generator
295 * version 3: the same as version 2 with additional IV seed
296 * (it uses 65 keys, last key is used as IV seed)
298 * tcw: Compatible implementation of the block chaining mode used
299 * by the TrueCrypt device encryption system (prior to version 4.1).
300 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
301 * It operates on full 512 byte sectors and uses CBC
302 * with an IV derived from initial key and the sector number.
303 * In addition, whitening value is applied on every sector, whitening
304 * is calculated from initial key, sector number and mixed using CRC32.
305 * Note that this encryption scheme is vulnerable to watermarking attacks
306 * and should be used for old compatible containers access only.
308 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
309 * The IV is encrypted little-endian byte-offset (with the same key
310 * and cipher as the volume).
312 * elephant: The extended version of eboiv with additional Elephant diffuser
313 * used with Bitlocker CBC mode.
314 * This mode was used in older Windows systems
315 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
318 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
319 struct dm_crypt_request *dmreq)
321 memset(iv, 0, cc->iv_size);
322 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
327 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
328 struct dm_crypt_request *dmreq)
330 memset(iv, 0, cc->iv_size);
331 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
336 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
337 struct dm_crypt_request *dmreq)
339 memset(iv, 0, cc->iv_size);
340 /* iv_size is at least of size u64; usually it is 16 bytes */
341 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
346 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
347 struct dm_crypt_request *dmreq)
350 * ESSIV encryption of the IV is now handled by the crypto API,
351 * so just pass the plain sector number here.
353 memset(iv, 0, cc->iv_size);
354 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
359 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
365 if (crypt_integrity_aead(cc))
366 bs = crypto_aead_blocksize(any_tfm_aead(cc));
368 bs = crypto_skcipher_blocksize(any_tfm(cc));
372 * We need to calculate how far we must shift the sector count
373 * to get the cipher block count, we use this shift in _gen.
375 if (1 << log != bs) {
376 ti->error = "cypher blocksize is not a power of 2";
381 ti->error = "cypher blocksize is > 512";
385 cc->iv_gen_private.benbi.shift = 9 - log;
390 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
394 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
395 struct dm_crypt_request *dmreq)
399 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
401 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
402 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
407 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
408 struct dm_crypt_request *dmreq)
410 memset(iv, 0, cc->iv_size);
415 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
417 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
419 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
420 crypto_free_shash(lmk->hash_tfm);
421 lmk->hash_tfm = NULL;
423 kfree_sensitive(lmk->seed);
427 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
430 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
432 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
433 ti->error = "Unsupported sector size for LMK";
437 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
438 CRYPTO_ALG_ALLOCATES_MEMORY);
439 if (IS_ERR(lmk->hash_tfm)) {
440 ti->error = "Error initializing LMK hash";
441 return PTR_ERR(lmk->hash_tfm);
444 /* No seed in LMK version 2 */
445 if (cc->key_parts == cc->tfms_count) {
450 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
452 crypt_iv_lmk_dtr(cc);
453 ti->error = "Error kmallocing seed storage in LMK";
460 static int crypt_iv_lmk_init(struct crypt_config *cc)
462 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
463 int subkey_size = cc->key_size / cc->key_parts;
465 /* LMK seed is on the position of LMK_KEYS + 1 key */
467 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
468 crypto_shash_digestsize(lmk->hash_tfm));
473 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
475 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
478 memset(lmk->seed, 0, LMK_SEED_SIZE);
483 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
484 struct dm_crypt_request *dmreq,
487 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
488 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
489 struct md5_state md5state;
493 desc->tfm = lmk->hash_tfm;
495 r = crypto_shash_init(desc);
500 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
505 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
506 r = crypto_shash_update(desc, data + 16, 16 * 31);
510 /* Sector is cropped to 56 bits here */
511 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
512 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
513 buf[2] = cpu_to_le32(4024);
515 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
519 /* No MD5 padding here */
520 r = crypto_shash_export(desc, &md5state);
524 for (i = 0; i < MD5_HASH_WORDS; i++)
525 __cpu_to_le32s(&md5state.hash[i]);
526 memcpy(iv, &md5state.hash, cc->iv_size);
531 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
532 struct dm_crypt_request *dmreq)
534 struct scatterlist *sg;
538 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
539 sg = crypt_get_sg_data(cc, dmreq->sg_in);
540 src = kmap_local_page(sg_page(sg));
541 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
544 memset(iv, 0, cc->iv_size);
549 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
550 struct dm_crypt_request *dmreq)
552 struct scatterlist *sg;
556 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
559 sg = crypt_get_sg_data(cc, dmreq->sg_out);
560 dst = kmap_local_page(sg_page(sg));
561 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
563 /* Tweak the first block of plaintext sector */
565 crypto_xor(dst + sg->offset, iv, cc->iv_size);
571 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
573 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
575 kfree_sensitive(tcw->iv_seed);
577 kfree_sensitive(tcw->whitening);
578 tcw->whitening = NULL;
580 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
581 crypto_free_shash(tcw->crc32_tfm);
582 tcw->crc32_tfm = NULL;
585 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
588 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
590 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
591 ti->error = "Unsupported sector size for TCW";
595 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
596 ti->error = "Wrong key size for TCW";
600 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
601 CRYPTO_ALG_ALLOCATES_MEMORY);
602 if (IS_ERR(tcw->crc32_tfm)) {
603 ti->error = "Error initializing CRC32 in TCW";
604 return PTR_ERR(tcw->crc32_tfm);
607 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
608 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
609 if (!tcw->iv_seed || !tcw->whitening) {
610 crypt_iv_tcw_dtr(cc);
611 ti->error = "Error allocating seed storage in TCW";
618 static int crypt_iv_tcw_init(struct crypt_config *cc)
620 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
621 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
623 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
624 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
630 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
632 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
634 memset(tcw->iv_seed, 0, cc->iv_size);
635 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
640 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
641 struct dm_crypt_request *dmreq,
644 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
645 __le64 sector = cpu_to_le64(dmreq->iv_sector);
646 u8 buf[TCW_WHITENING_SIZE];
647 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
650 /* xor whitening with sector number */
651 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
652 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
654 /* calculate crc32 for every 32bit part and xor it */
655 desc->tfm = tcw->crc32_tfm;
656 for (i = 0; i < 4; i++) {
657 r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
661 crypto_xor(&buf[0], &buf[12], 4);
662 crypto_xor(&buf[4], &buf[8], 4);
664 /* apply whitening (8 bytes) to whole sector */
665 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
666 crypto_xor(data + i * 8, buf, 8);
668 memzero_explicit(buf, sizeof(buf));
672 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
673 struct dm_crypt_request *dmreq)
675 struct scatterlist *sg;
676 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
677 __le64 sector = cpu_to_le64(dmreq->iv_sector);
681 /* Remove whitening from ciphertext */
682 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
683 sg = crypt_get_sg_data(cc, dmreq->sg_in);
684 src = kmap_local_page(sg_page(sg));
685 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
690 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
692 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
698 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
699 struct dm_crypt_request *dmreq)
701 struct scatterlist *sg;
705 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
708 /* Apply whitening on ciphertext */
709 sg = crypt_get_sg_data(cc, dmreq->sg_out);
710 dst = kmap_local_page(sg_page(sg));
711 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
717 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
718 struct dm_crypt_request *dmreq)
720 /* Used only for writes, there must be an additional space to store IV */
721 get_random_bytes(iv, cc->iv_size);
725 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
728 if (crypt_integrity_aead(cc)) {
729 ti->error = "AEAD transforms not supported for EBOIV";
733 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
734 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
741 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
742 struct dm_crypt_request *dmreq)
744 struct crypto_skcipher *tfm = any_tfm(cc);
745 struct skcipher_request *req;
746 struct scatterlist src, dst;
747 DECLARE_CRYPTO_WAIT(wait);
748 unsigned int reqsize;
752 reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
753 reqsize = ALIGN(reqsize, __alignof__(__le64));
755 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
759 skcipher_request_set_tfm(req, tfm);
761 buf = (u8 *)req + reqsize;
762 memset(buf, 0, cc->iv_size);
763 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
765 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
766 sg_init_one(&dst, iv, cc->iv_size);
767 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
768 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
769 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
770 kfree_sensitive(req);
775 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
777 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
779 crypto_free_skcipher(elephant->tfm);
780 elephant->tfm = NULL;
783 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
786 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
789 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
790 CRYPTO_ALG_ALLOCATES_MEMORY);
791 if (IS_ERR(elephant->tfm)) {
792 r = PTR_ERR(elephant->tfm);
793 elephant->tfm = NULL;
797 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
799 crypt_iv_elephant_dtr(cc);
803 static void diffuser_disk_to_cpu(u32 *d, size_t n)
805 #ifndef __LITTLE_ENDIAN
808 for (i = 0; i < n; i++)
809 d[i] = le32_to_cpu((__le32)d[i]);
813 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
815 #ifndef __LITTLE_ENDIAN
818 for (i = 0; i < n; i++)
819 d[i] = cpu_to_le32((u32)d[i]);
823 static void diffuser_a_decrypt(u32 *d, size_t n)
827 for (i = 0; i < 5; i++) {
832 while (i1 < (n - 1)) {
833 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
839 d[i1] += d[i2] ^ d[i3];
845 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
848 d[i1] += d[i2] ^ d[i3];
854 static void diffuser_a_encrypt(u32 *d, size_t n)
858 for (i = 0; i < 5; i++) {
864 d[i1] -= d[i2] ^ d[i3];
867 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
873 d[i1] -= d[i2] ^ d[i3];
879 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
885 static void diffuser_b_decrypt(u32 *d, size_t n)
889 for (i = 0; i < 3; i++) {
894 while (i1 < (n - 1)) {
895 d[i1] += d[i2] ^ d[i3];
898 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
904 d[i1] += d[i2] ^ d[i3];
910 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
916 static void diffuser_b_encrypt(u32 *d, size_t n)
920 for (i = 0; i < 3; i++) {
926 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
932 d[i1] -= d[i2] ^ d[i3];
938 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
941 d[i1] -= d[i2] ^ d[i3];
947 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
949 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
950 u8 *es, *ks, *data, *data2, *data_offset;
951 struct skcipher_request *req;
952 struct scatterlist *sg, *sg2, src, dst;
953 DECLARE_CRYPTO_WAIT(wait);
956 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
957 es = kzalloc(16, GFP_NOIO); /* Key for AES */
958 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
960 if (!req || !es || !ks) {
965 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
968 sg_init_one(&src, es, 16);
969 sg_init_one(&dst, ks, 16);
970 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
971 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
972 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
978 sg_init_one(&dst, &ks[16], 16);
979 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
983 sg = crypt_get_sg_data(cc, dmreq->sg_out);
984 data = kmap_local_page(sg_page(sg));
985 data_offset = data + sg->offset;
987 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
988 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
989 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
990 data2 = kmap_local_page(sg_page(sg2));
991 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
995 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
996 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
997 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
998 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
999 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1002 for (i = 0; i < (cc->sector_size / 32); i++)
1003 crypto_xor(data_offset + i * 32, ks, 32);
1005 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1006 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1007 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1008 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1009 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1014 kfree_sensitive(ks);
1015 kfree_sensitive(es);
1016 skcipher_request_free(req);
1020 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1021 struct dm_crypt_request *dmreq)
1025 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1026 r = crypt_iv_elephant(cc, dmreq);
1031 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1034 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1035 struct dm_crypt_request *dmreq)
1037 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1038 return crypt_iv_elephant(cc, dmreq);
1043 static int crypt_iv_elephant_init(struct crypt_config *cc)
1045 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1046 int key_offset = cc->key_size - cc->key_extra_size;
1048 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1051 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1053 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1054 u8 key[ELEPHANT_MAX_KEY_SIZE];
1056 memset(key, 0, cc->key_extra_size);
1057 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1060 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1061 .generator = crypt_iv_plain_gen
1064 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1065 .generator = crypt_iv_plain64_gen
1068 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1069 .generator = crypt_iv_plain64be_gen
1072 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1073 .generator = crypt_iv_essiv_gen
1076 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1077 .ctr = crypt_iv_benbi_ctr,
1078 .dtr = crypt_iv_benbi_dtr,
1079 .generator = crypt_iv_benbi_gen
1082 static const struct crypt_iv_operations crypt_iv_null_ops = {
1083 .generator = crypt_iv_null_gen
1086 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1087 .ctr = crypt_iv_lmk_ctr,
1088 .dtr = crypt_iv_lmk_dtr,
1089 .init = crypt_iv_lmk_init,
1090 .wipe = crypt_iv_lmk_wipe,
1091 .generator = crypt_iv_lmk_gen,
1092 .post = crypt_iv_lmk_post
1095 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1096 .ctr = crypt_iv_tcw_ctr,
1097 .dtr = crypt_iv_tcw_dtr,
1098 .init = crypt_iv_tcw_init,
1099 .wipe = crypt_iv_tcw_wipe,
1100 .generator = crypt_iv_tcw_gen,
1101 .post = crypt_iv_tcw_post
1104 static const struct crypt_iv_operations crypt_iv_random_ops = {
1105 .generator = crypt_iv_random_gen
1108 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1109 .ctr = crypt_iv_eboiv_ctr,
1110 .generator = crypt_iv_eboiv_gen
1113 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1114 .ctr = crypt_iv_elephant_ctr,
1115 .dtr = crypt_iv_elephant_dtr,
1116 .init = crypt_iv_elephant_init,
1117 .wipe = crypt_iv_elephant_wipe,
1118 .generator = crypt_iv_elephant_gen,
1119 .post = crypt_iv_elephant_post
1123 * Integrity extensions
1125 static bool crypt_integrity_aead(struct crypt_config *cc)
1127 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1130 static bool crypt_integrity_hmac(struct crypt_config *cc)
1132 return crypt_integrity_aead(cc) && cc->key_mac_size;
1135 /* Get sg containing data */
1136 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1137 struct scatterlist *sg)
1139 if (unlikely(crypt_integrity_aead(cc)))
1145 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1147 struct bio_integrity_payload *bip;
1148 unsigned int tag_len;
1151 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1154 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1156 return PTR_ERR(bip);
1158 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1160 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1162 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1163 tag_len, offset_in_page(io->integrity_metadata));
1164 if (unlikely(ret != tag_len))
1170 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1172 #ifdef CONFIG_BLK_DEV_INTEGRITY
1173 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1174 struct mapped_device *md = dm_table_get_md(ti->table);
1176 /* From now we require underlying device with our integrity profile */
1177 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1178 ti->error = "Integrity profile not supported.";
1182 if (bi->tag_size != cc->on_disk_tag_size ||
1183 bi->tuple_size != cc->on_disk_tag_size) {
1184 ti->error = "Integrity profile tag size mismatch.";
1187 if (1 << bi->interval_exp != cc->sector_size) {
1188 ti->error = "Integrity profile sector size mismatch.";
1192 if (crypt_integrity_aead(cc)) {
1193 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1194 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1195 cc->integrity_tag_size, cc->integrity_iv_size);
1197 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1198 ti->error = "Integrity AEAD auth tag size is not supported.";
1201 } else if (cc->integrity_iv_size)
1202 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1203 cc->integrity_iv_size);
1205 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1206 ti->error = "Not enough space for integrity tag in the profile.";
1212 ti->error = "Integrity profile not supported.";
1217 static void crypt_convert_init(struct crypt_config *cc,
1218 struct convert_context *ctx,
1219 struct bio *bio_out, struct bio *bio_in,
1222 ctx->bio_in = bio_in;
1223 ctx->bio_out = bio_out;
1225 ctx->iter_in = bio_in->bi_iter;
1227 ctx->iter_out = bio_out->bi_iter;
1228 ctx->cc_sector = sector + cc->iv_offset;
1229 init_completion(&ctx->restart);
1232 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1235 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1238 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1240 return (void *)((char *)dmreq - cc->dmreq_start);
1243 static u8 *iv_of_dmreq(struct crypt_config *cc,
1244 struct dm_crypt_request *dmreq)
1246 if (crypt_integrity_aead(cc))
1247 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1248 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1250 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1251 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1254 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1255 struct dm_crypt_request *dmreq)
1257 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1260 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1261 struct dm_crypt_request *dmreq)
1263 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1265 return (__le64 *) ptr;
1268 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1269 struct dm_crypt_request *dmreq)
1271 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1272 cc->iv_size + sizeof(uint64_t);
1274 return (unsigned int *)ptr;
1277 static void *tag_from_dmreq(struct crypt_config *cc,
1278 struct dm_crypt_request *dmreq)
1280 struct convert_context *ctx = dmreq->ctx;
1281 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1283 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1284 cc->on_disk_tag_size];
1287 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1288 struct dm_crypt_request *dmreq)
1290 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1293 static int crypt_convert_block_aead(struct crypt_config *cc,
1294 struct convert_context *ctx,
1295 struct aead_request *req,
1296 unsigned int tag_offset)
1298 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1299 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1300 struct dm_crypt_request *dmreq;
1301 u8 *iv, *org_iv, *tag_iv, *tag;
1305 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1307 /* Reject unexpected unaligned bio. */
1308 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1311 dmreq = dmreq_of_req(cc, req);
1312 dmreq->iv_sector = ctx->cc_sector;
1313 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1314 dmreq->iv_sector >>= cc->sector_shift;
1317 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1319 sector = org_sector_of_dmreq(cc, dmreq);
1320 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1322 iv = iv_of_dmreq(cc, dmreq);
1323 org_iv = org_iv_of_dmreq(cc, dmreq);
1324 tag = tag_from_dmreq(cc, dmreq);
1325 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1328 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1329 * | (authenticated) | (auth+encryption) | |
1330 * | sector_LE | IV | sector in/out | tag in/out |
1332 sg_init_table(dmreq->sg_in, 4);
1333 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1334 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1335 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1336 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1338 sg_init_table(dmreq->sg_out, 4);
1339 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1340 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1341 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1342 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1344 if (cc->iv_gen_ops) {
1345 /* For READs use IV stored in integrity metadata */
1346 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1347 memcpy(org_iv, tag_iv, cc->iv_size);
1349 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1352 /* Store generated IV in integrity metadata */
1353 if (cc->integrity_iv_size)
1354 memcpy(tag_iv, org_iv, cc->iv_size);
1356 /* Working copy of IV, to be modified in crypto API */
1357 memcpy(iv, org_iv, cc->iv_size);
1360 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1361 if (bio_data_dir(ctx->bio_in) == WRITE) {
1362 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1363 cc->sector_size, iv);
1364 r = crypto_aead_encrypt(req);
1365 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1366 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1367 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1369 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1370 cc->sector_size + cc->integrity_tag_size, iv);
1371 r = crypto_aead_decrypt(req);
1374 if (r == -EBADMSG) {
1375 sector_t s = le64_to_cpu(*sector);
1377 ctx->aead_failed = true;
1378 if (ctx->aead_recheck) {
1379 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1380 ctx->bio_in->bi_bdev, s);
1381 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1386 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1387 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1389 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1390 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1395 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1396 struct convert_context *ctx,
1397 struct skcipher_request *req,
1398 unsigned int tag_offset)
1400 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1401 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1402 struct scatterlist *sg_in, *sg_out;
1403 struct dm_crypt_request *dmreq;
1404 u8 *iv, *org_iv, *tag_iv;
1408 /* Reject unexpected unaligned bio. */
1409 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1412 dmreq = dmreq_of_req(cc, req);
1413 dmreq->iv_sector = ctx->cc_sector;
1414 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1415 dmreq->iv_sector >>= cc->sector_shift;
1418 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1420 iv = iv_of_dmreq(cc, dmreq);
1421 org_iv = org_iv_of_dmreq(cc, dmreq);
1422 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1424 sector = org_sector_of_dmreq(cc, dmreq);
1425 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1427 /* For skcipher we use only the first sg item */
1428 sg_in = &dmreq->sg_in[0];
1429 sg_out = &dmreq->sg_out[0];
1431 sg_init_table(sg_in, 1);
1432 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1434 sg_init_table(sg_out, 1);
1435 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1437 if (cc->iv_gen_ops) {
1438 /* For READs use IV stored in integrity metadata */
1439 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1440 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1442 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1445 /* Data can be already preprocessed in generator */
1446 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1448 /* Store generated IV in integrity metadata */
1449 if (cc->integrity_iv_size)
1450 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1452 /* Working copy of IV, to be modified in crypto API */
1453 memcpy(iv, org_iv, cc->iv_size);
1456 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1458 if (bio_data_dir(ctx->bio_in) == WRITE)
1459 r = crypto_skcipher_encrypt(req);
1461 r = crypto_skcipher_decrypt(req);
1463 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1464 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1466 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1467 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1472 static void kcryptd_async_done(void *async_req, int error);
1474 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1475 struct convert_context *ctx)
1477 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1480 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1485 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1488 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1489 * requests if driver request queue is full.
1491 skcipher_request_set_callback(ctx->r.req,
1492 CRYPTO_TFM_REQ_MAY_BACKLOG,
1493 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1498 static int crypt_alloc_req_aead(struct crypt_config *cc,
1499 struct convert_context *ctx)
1501 if (!ctx->r.req_aead) {
1502 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1503 if (!ctx->r.req_aead)
1507 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1510 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1511 * requests if driver request queue is full.
1513 aead_request_set_callback(ctx->r.req_aead,
1514 CRYPTO_TFM_REQ_MAY_BACKLOG,
1515 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1520 static int crypt_alloc_req(struct crypt_config *cc,
1521 struct convert_context *ctx)
1523 if (crypt_integrity_aead(cc))
1524 return crypt_alloc_req_aead(cc, ctx);
1526 return crypt_alloc_req_skcipher(cc, ctx);
1529 static void crypt_free_req_skcipher(struct crypt_config *cc,
1530 struct skcipher_request *req, struct bio *base_bio)
1532 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1534 if ((struct skcipher_request *)(io + 1) != req)
1535 mempool_free(req, &cc->req_pool);
1538 static void crypt_free_req_aead(struct crypt_config *cc,
1539 struct aead_request *req, struct bio *base_bio)
1541 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1543 if ((struct aead_request *)(io + 1) != req)
1544 mempool_free(req, &cc->req_pool);
1547 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1549 if (crypt_integrity_aead(cc))
1550 crypt_free_req_aead(cc, req, base_bio);
1552 crypt_free_req_skcipher(cc, req, base_bio);
1556 * Encrypt / decrypt data from one bio to another one (can be the same one)
1558 static blk_status_t crypt_convert(struct crypt_config *cc,
1559 struct convert_context *ctx, bool atomic, bool reset_pending)
1561 unsigned int tag_offset = 0;
1562 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1566 * if reset_pending is set we are dealing with the bio for the first time,
1567 * else we're continuing to work on the previous bio, so don't mess with
1568 * the cc_pending counter
1571 atomic_set(&ctx->cc_pending, 1);
1573 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1575 r = crypt_alloc_req(cc, ctx);
1577 complete(&ctx->restart);
1578 return BLK_STS_DEV_RESOURCE;
1581 atomic_inc(&ctx->cc_pending);
1583 if (crypt_integrity_aead(cc))
1584 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1586 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1590 * The request was queued by a crypto driver
1591 * but the driver request queue is full, let's wait.
1594 if (in_interrupt()) {
1595 if (try_wait_for_completion(&ctx->restart)) {
1597 * we don't have to block to wait for completion,
1602 * we can't wait for completion without blocking
1603 * exit and continue processing in a workqueue
1606 ctx->cc_sector += sector_step;
1608 return BLK_STS_DEV_RESOURCE;
1611 wait_for_completion(&ctx->restart);
1613 reinit_completion(&ctx->restart);
1616 * The request is queued and processed asynchronously,
1617 * completion function kcryptd_async_done() will be called.
1621 ctx->cc_sector += sector_step;
1625 * The request was already processed (synchronously).
1628 atomic_dec(&ctx->cc_pending);
1629 ctx->cc_sector += sector_step;
1635 * There was a data integrity error.
1638 atomic_dec(&ctx->cc_pending);
1639 return BLK_STS_PROTECTION;
1641 * There was an error while processing the request.
1644 atomic_dec(&ctx->cc_pending);
1645 return BLK_STS_IOERR;
1652 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1655 * Generate a new unfragmented bio with the given size
1656 * This should never violate the device limitations (but only because
1657 * max_segment_size is being constrained to PAGE_SIZE).
1659 * This function may be called concurrently. If we allocate from the mempool
1660 * concurrently, there is a possibility of deadlock. For example, if we have
1661 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1662 * the mempool concurrently, it may deadlock in a situation where both processes
1663 * have allocated 128 pages and the mempool is exhausted.
1665 * In order to avoid this scenario we allocate the pages under a mutex.
1667 * In order to not degrade performance with excessive locking, we try
1668 * non-blocking allocations without a mutex first but on failure we fallback
1669 * to blocking allocations with a mutex.
1671 * In order to reduce allocation overhead, we try to allocate compound pages in
1672 * the first pass. If they are not available, we fall back to the mempool.
1674 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1676 struct crypt_config *cc = io->cc;
1678 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1679 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1680 unsigned int remaining_size;
1681 unsigned int order = MAX_PAGE_ORDER;
1684 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1685 mutex_lock(&cc->bio_alloc_lock);
1687 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1689 clone->bi_private = io;
1690 clone->bi_end_io = crypt_endio;
1691 clone->bi_ioprio = io->base_bio->bi_ioprio;
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 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1703 (1 << order) > dm_crypt_pages_per_client))
1704 goto decrease_order;
1705 pages = alloc_pages(gfp_mask
1706 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1708 if (likely(pages != NULL)) {
1709 percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1716 pages = mempool_alloc(&cc->page_pool, gfp_mask);
1718 crypt_free_buffer_pages(cc, clone);
1720 gfp_mask |= __GFP_DIRECT_RECLAIM;
1726 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1727 __bio_add_page(clone, pages, size_to_add, 0);
1728 remaining_size -= size_to_add;
1731 /* Allocate space for integrity tags */
1732 if (dm_crypt_integrity_io_alloc(io, clone)) {
1733 crypt_free_buffer_pages(cc, clone);
1738 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1739 mutex_unlock(&cc->bio_alloc_lock);
1744 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1746 struct folio_iter fi;
1748 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1749 bio_for_each_folio_all(fi, clone) {
1750 if (folio_test_large(fi.folio)) {
1751 percpu_counter_sub(&cc->n_allocated_pages,
1752 1 << folio_order(fi.folio));
1753 folio_put(fi.folio);
1755 mempool_free(&fi.folio->page, &cc->page_pool);
1761 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1762 struct bio *bio, sector_t sector)
1766 io->sector = sector;
1768 io->ctx.aead_recheck = false;
1769 io->ctx.aead_failed = false;
1770 io->ctx.r.req = NULL;
1771 io->integrity_metadata = NULL;
1772 io->integrity_metadata_from_pool = false;
1773 atomic_set(&io->io_pending, 0);
1776 static void crypt_inc_pending(struct dm_crypt_io *io)
1778 atomic_inc(&io->io_pending);
1781 static void kcryptd_queue_read(struct dm_crypt_io *io);
1784 * One of the bios was finished. Check for completion of
1785 * the whole request and correctly clean up the buffer.
1787 static void crypt_dec_pending(struct dm_crypt_io *io)
1789 struct crypt_config *cc = io->cc;
1790 struct bio *base_bio = io->base_bio;
1791 blk_status_t error = io->error;
1793 if (!atomic_dec_and_test(&io->io_pending))
1796 if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1797 cc->on_disk_tag_size && bio_data_dir(base_bio) == READ) {
1798 io->ctx.aead_recheck = true;
1799 io->ctx.aead_failed = false;
1801 kcryptd_queue_read(io);
1806 crypt_free_req(cc, io->ctx.r.req, base_bio);
1808 if (unlikely(io->integrity_metadata_from_pool))
1809 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1811 kfree(io->integrity_metadata);
1813 base_bio->bi_status = error;
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);
1840 blk_status_t error = clone->bi_status;
1842 if (io->ctx.aead_recheck && !error) {
1843 kcryptd_queue_crypt(io);
1848 * free the processed pages
1850 if (rw == WRITE || io->ctx.aead_recheck)
1851 crypt_free_buffer_pages(cc, clone);
1855 if (rw == READ && !error) {
1856 kcryptd_queue_crypt(io);
1860 if (unlikely(error))
1863 crypt_dec_pending(io);
1866 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1868 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1870 struct crypt_config *cc = io->cc;
1873 if (io->ctx.aead_recheck) {
1874 if (!(gfp & __GFP_DIRECT_RECLAIM))
1876 crypt_inc_pending(io);
1877 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1878 if (unlikely(!clone)) {
1879 crypt_dec_pending(io);
1882 clone->bi_iter.bi_sector = cc->start + io->sector;
1883 crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1884 io->saved_bi_iter = clone->bi_iter;
1885 dm_submit_bio_remap(io->base_bio, clone);
1890 * We need the original biovec array in order to decrypt the whole bio
1891 * data *afterwards* -- thanks to immutable biovecs we don't need to
1892 * worry about the block layer modifying the biovec array; so leverage
1893 * bio_alloc_clone().
1895 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1898 clone->bi_private = io;
1899 clone->bi_end_io = crypt_endio;
1901 crypt_inc_pending(io);
1903 clone->bi_iter.bi_sector = cc->start + io->sector;
1905 if (dm_crypt_integrity_io_alloc(io, clone)) {
1906 crypt_dec_pending(io);
1911 dm_submit_bio_remap(io->base_bio, clone);
1915 static void kcryptd_io_read_work(struct work_struct *work)
1917 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1919 crypt_inc_pending(io);
1920 if (kcryptd_io_read(io, GFP_NOIO))
1921 io->error = BLK_STS_RESOURCE;
1922 crypt_dec_pending(io);
1925 static void kcryptd_queue_read(struct dm_crypt_io *io)
1927 struct crypt_config *cc = io->cc;
1929 INIT_WORK(&io->work, kcryptd_io_read_work);
1930 queue_work(cc->io_queue, &io->work);
1933 static void kcryptd_io_write(struct dm_crypt_io *io)
1935 struct bio *clone = io->ctx.bio_out;
1937 dm_submit_bio_remap(io->base_bio, clone);
1940 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1942 static int dmcrypt_write(void *data)
1944 struct crypt_config *cc = data;
1945 struct dm_crypt_io *io;
1948 struct rb_root write_tree;
1949 struct blk_plug plug;
1951 spin_lock_irq(&cc->write_thread_lock);
1954 if (!RB_EMPTY_ROOT(&cc->write_tree))
1957 set_current_state(TASK_INTERRUPTIBLE);
1959 spin_unlock_irq(&cc->write_thread_lock);
1961 if (unlikely(kthread_should_stop())) {
1962 set_current_state(TASK_RUNNING);
1968 spin_lock_irq(&cc->write_thread_lock);
1969 goto continue_locked;
1972 write_tree = cc->write_tree;
1973 cc->write_tree = RB_ROOT;
1974 spin_unlock_irq(&cc->write_thread_lock);
1976 BUG_ON(rb_parent(write_tree.rb_node));
1979 * Note: we cannot walk the tree here with rb_next because
1980 * the structures may be freed when kcryptd_io_write is called.
1982 blk_start_plug(&plug);
1984 io = crypt_io_from_node(rb_first(&write_tree));
1985 rb_erase(&io->rb_node, &write_tree);
1986 kcryptd_io_write(io);
1988 } while (!RB_EMPTY_ROOT(&write_tree));
1989 blk_finish_plug(&plug);
1994 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1996 struct bio *clone = io->ctx.bio_out;
1997 struct crypt_config *cc = io->cc;
1998 unsigned long flags;
2000 struct rb_node **rbp, *parent;
2002 if (unlikely(io->error)) {
2003 crypt_free_buffer_pages(cc, clone);
2005 crypt_dec_pending(io);
2009 /* crypt_convert should have filled the clone bio */
2010 BUG_ON(io->ctx.iter_out.bi_size);
2012 clone->bi_iter.bi_sector = cc->start + io->sector;
2014 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2015 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2016 dm_submit_bio_remap(io->base_bio, clone);
2020 spin_lock_irqsave(&cc->write_thread_lock, flags);
2021 if (RB_EMPTY_ROOT(&cc->write_tree))
2022 wake_up_process(cc->write_thread);
2023 rbp = &cc->write_tree.rb_node;
2025 sector = io->sector;
2028 if (sector < crypt_io_from_node(parent)->sector)
2029 rbp = &(*rbp)->rb_left;
2031 rbp = &(*rbp)->rb_right;
2033 rb_link_node(&io->rb_node, parent, rbp);
2034 rb_insert_color(&io->rb_node, &cc->write_tree);
2035 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2038 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2039 struct convert_context *ctx)
2042 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2046 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2047 * constraints so they do not need to be issued inline by
2048 * kcryptd_crypt_write_convert().
2050 switch (bio_op(ctx->bio_in)) {
2052 case REQ_OP_WRITE_ZEROES:
2059 static void kcryptd_crypt_write_continue(struct work_struct *work)
2061 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2062 struct crypt_config *cc = io->cc;
2063 struct convert_context *ctx = &io->ctx;
2065 sector_t sector = io->sector;
2068 wait_for_completion(&ctx->restart);
2069 reinit_completion(&ctx->restart);
2071 r = crypt_convert(cc, &io->ctx, true, false);
2074 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2075 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2076 /* Wait for completion signaled by kcryptd_async_done() */
2077 wait_for_completion(&ctx->restart);
2081 /* Encryption was already finished, submit io now */
2082 if (crypt_finished) {
2083 kcryptd_crypt_write_io_submit(io, 0);
2084 io->sector = sector;
2087 crypt_dec_pending(io);
2090 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2092 struct crypt_config *cc = io->cc;
2093 struct convert_context *ctx = &io->ctx;
2096 sector_t sector = io->sector;
2100 * Prevent io from disappearing until this function completes.
2102 crypt_inc_pending(io);
2103 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2105 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2106 if (unlikely(!clone)) {
2107 io->error = BLK_STS_IOERR;
2111 io->ctx.bio_out = clone;
2112 io->ctx.iter_out = clone->bi_iter;
2114 if (crypt_integrity_aead(cc)) {
2115 bio_copy_data(clone, io->base_bio);
2116 io->ctx.bio_in = clone;
2117 io->ctx.iter_in = clone->bi_iter;
2120 sector += bio_sectors(clone);
2122 crypt_inc_pending(io);
2123 r = crypt_convert(cc, ctx,
2124 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2126 * Crypto API backlogged the request, because its queue was full
2127 * and we're in softirq context, so continue from a workqueue
2128 * (TODO: is it actually possible to be in softirq in the write path?)
2130 if (r == BLK_STS_DEV_RESOURCE) {
2131 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2132 queue_work(cc->crypt_queue, &io->work);
2137 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2138 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2139 /* Wait for completion signaled by kcryptd_async_done() */
2140 wait_for_completion(&ctx->restart);
2144 /* Encryption was already finished, submit io now */
2145 if (crypt_finished) {
2146 kcryptd_crypt_write_io_submit(io, 0);
2147 io->sector = sector;
2151 crypt_dec_pending(io);
2154 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2156 if (io->ctx.aead_recheck) {
2158 io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2159 bio_copy_data(io->base_bio, io->ctx.bio_in);
2161 crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2162 bio_put(io->ctx.bio_in);
2164 crypt_dec_pending(io);
2167 static void kcryptd_crypt_read_continue(struct work_struct *work)
2169 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2170 struct crypt_config *cc = io->cc;
2173 wait_for_completion(&io->ctx.restart);
2174 reinit_completion(&io->ctx.restart);
2176 r = crypt_convert(cc, &io->ctx, true, false);
2180 if (atomic_dec_and_test(&io->ctx.cc_pending))
2181 kcryptd_crypt_read_done(io);
2183 crypt_dec_pending(io);
2186 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2188 struct crypt_config *cc = io->cc;
2191 crypt_inc_pending(io);
2193 if (io->ctx.aead_recheck) {
2194 io->ctx.cc_sector = io->sector + cc->iv_offset;
2195 r = crypt_convert(cc, &io->ctx,
2196 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2198 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2201 r = crypt_convert(cc, &io->ctx,
2202 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2205 * Crypto API backlogged the request, because its queue was full
2206 * and we're in softirq context, so continue from a workqueue
2208 if (r == BLK_STS_DEV_RESOURCE) {
2209 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2210 queue_work(cc->crypt_queue, &io->work);
2216 if (atomic_dec_and_test(&io->ctx.cc_pending))
2217 kcryptd_crypt_read_done(io);
2219 crypt_dec_pending(io);
2222 static void kcryptd_async_done(void *data, int error)
2224 struct dm_crypt_request *dmreq = data;
2225 struct convert_context *ctx = dmreq->ctx;
2226 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2227 struct crypt_config *cc = io->cc;
2230 * A request from crypto driver backlog is going to be processed now,
2231 * finish the completion and continue in crypt_convert().
2232 * (Callback will be called for the second time for this request.)
2234 if (error == -EINPROGRESS) {
2235 complete(&ctx->restart);
2239 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2240 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2242 if (error == -EBADMSG) {
2243 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2245 ctx->aead_failed = true;
2246 if (ctx->aead_recheck) {
2247 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2248 ctx->bio_in->bi_bdev, s);
2249 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2252 io->error = BLK_STS_PROTECTION;
2253 } else if (error < 0)
2254 io->error = BLK_STS_IOERR;
2256 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2258 if (!atomic_dec_and_test(&ctx->cc_pending))
2262 * The request is fully completed: for inline writes, let
2263 * kcryptd_crypt_write_convert() do the IO submission.
2265 if (bio_data_dir(io->base_bio) == READ) {
2266 kcryptd_crypt_read_done(io);
2270 if (kcryptd_crypt_write_inline(cc, ctx)) {
2271 complete(&ctx->restart);
2275 kcryptd_crypt_write_io_submit(io, 1);
2278 static void kcryptd_crypt(struct work_struct *work)
2280 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2282 if (bio_data_dir(io->base_bio) == READ)
2283 kcryptd_crypt_read_convert(io);
2285 kcryptd_crypt_write_convert(io);
2288 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2290 struct crypt_config *cc = io->cc;
2292 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2293 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2295 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2296 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2297 * it is being executed with irqs disabled.
2299 if (in_hardirq() || irqs_disabled()) {
2300 INIT_WORK(&io->work, kcryptd_crypt);
2301 queue_work(system_bh_wq, &io->work);
2304 kcryptd_crypt(&io->work);
2309 INIT_WORK(&io->work, kcryptd_crypt);
2310 queue_work(cc->crypt_queue, &io->work);
2313 static void crypt_free_tfms_aead(struct crypt_config *cc)
2315 if (!cc->cipher_tfm.tfms_aead)
2318 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2319 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2320 cc->cipher_tfm.tfms_aead[0] = NULL;
2323 kfree(cc->cipher_tfm.tfms_aead);
2324 cc->cipher_tfm.tfms_aead = NULL;
2327 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2331 if (!cc->cipher_tfm.tfms)
2334 for (i = 0; i < cc->tfms_count; i++)
2335 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2336 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2337 cc->cipher_tfm.tfms[i] = NULL;
2340 kfree(cc->cipher_tfm.tfms);
2341 cc->cipher_tfm.tfms = NULL;
2344 static void crypt_free_tfms(struct crypt_config *cc)
2346 if (crypt_integrity_aead(cc))
2347 crypt_free_tfms_aead(cc);
2349 crypt_free_tfms_skcipher(cc);
2352 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2357 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2358 sizeof(struct crypto_skcipher *),
2360 if (!cc->cipher_tfm.tfms)
2363 for (i = 0; i < cc->tfms_count; i++) {
2364 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2365 CRYPTO_ALG_ALLOCATES_MEMORY);
2366 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2367 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2368 crypt_free_tfms(cc);
2374 * dm-crypt performance can vary greatly depending on which crypto
2375 * algorithm implementation is used. Help people debug performance
2376 * problems by logging the ->cra_driver_name.
2378 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2379 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2383 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2387 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2388 if (!cc->cipher_tfm.tfms)
2391 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2392 CRYPTO_ALG_ALLOCATES_MEMORY);
2393 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2394 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2395 crypt_free_tfms(cc);
2399 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2400 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2404 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2406 if (crypt_integrity_aead(cc))
2407 return crypt_alloc_tfms_aead(cc, ciphermode);
2409 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2412 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2414 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2417 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2419 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2423 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2424 * the key must be for some reason in special format.
2425 * This funcion converts cc->key to this special format.
2427 static void crypt_copy_authenckey(char *p, const void *key,
2428 unsigned int enckeylen, unsigned int authkeylen)
2430 struct crypto_authenc_key_param *param;
2433 rta = (struct rtattr *)p;
2434 param = RTA_DATA(rta);
2435 param->enckeylen = cpu_to_be32(enckeylen);
2436 rta->rta_len = RTA_LENGTH(sizeof(*param));
2437 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2438 p += RTA_SPACE(sizeof(*param));
2439 memcpy(p, key + enckeylen, authkeylen);
2441 memcpy(p, key, enckeylen);
2444 static int crypt_setkey(struct crypt_config *cc)
2446 unsigned int subkey_size;
2449 /* Ignore extra keys (which are used for IV etc) */
2450 subkey_size = crypt_subkey_size(cc);
2452 if (crypt_integrity_hmac(cc)) {
2453 if (subkey_size < cc->key_mac_size)
2456 crypt_copy_authenckey(cc->authenc_key, cc->key,
2457 subkey_size - cc->key_mac_size,
2461 for (i = 0; i < cc->tfms_count; i++) {
2462 if (crypt_integrity_hmac(cc))
2463 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2464 cc->authenc_key, crypt_authenckey_size(cc));
2465 else if (crypt_integrity_aead(cc))
2466 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2467 cc->key + (i * subkey_size),
2470 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2471 cc->key + (i * subkey_size),
2477 if (crypt_integrity_hmac(cc))
2478 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2485 static bool contains_whitespace(const char *str)
2488 if (isspace(*str++))
2493 static int set_key_user(struct crypt_config *cc, struct key *key)
2495 const struct user_key_payload *ukp;
2497 ukp = user_key_payload_locked(key);
2499 return -EKEYREVOKED;
2501 if (cc->key_size != ukp->datalen)
2504 memcpy(cc->key, ukp->data, cc->key_size);
2509 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2511 const struct encrypted_key_payload *ekp;
2513 ekp = key->payload.data[0];
2515 return -EKEYREVOKED;
2517 if (cc->key_size != ekp->decrypted_datalen)
2520 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2525 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2527 const struct trusted_key_payload *tkp;
2529 tkp = key->payload.data[0];
2531 return -EKEYREVOKED;
2533 if (cc->key_size != tkp->key_len)
2536 memcpy(cc->key, tkp->key, cc->key_size);
2541 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2543 char *new_key_string, *key_desc;
2545 struct key_type *type;
2547 int (*set_key)(struct crypt_config *cc, struct key *key);
2550 * Reject key_string with whitespace. dm core currently lacks code for
2551 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2553 if (contains_whitespace(key_string)) {
2554 DMERR("whitespace chars not allowed in key string");
2558 /* look for next ':' separating key_type from key_description */
2559 key_desc = strchr(key_string, ':');
2560 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2563 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2564 type = &key_type_logon;
2565 set_key = set_key_user;
2566 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2567 type = &key_type_user;
2568 set_key = set_key_user;
2569 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2570 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2571 type = &key_type_encrypted;
2572 set_key = set_key_encrypted;
2573 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2574 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2575 type = &key_type_trusted;
2576 set_key = set_key_trusted;
2581 new_key_string = kstrdup(key_string, GFP_KERNEL);
2582 if (!new_key_string)
2585 key = request_key(type, key_desc + 1, NULL);
2587 kfree_sensitive(new_key_string);
2588 return PTR_ERR(key);
2591 down_read(&key->sem);
2593 ret = set_key(cc, key);
2597 kfree_sensitive(new_key_string);
2604 /* clear the flag since following operations may invalidate previously valid key */
2605 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2607 ret = crypt_setkey(cc);
2610 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2611 kfree_sensitive(cc->key_string);
2612 cc->key_string = new_key_string;
2614 kfree_sensitive(new_key_string);
2619 static int get_key_size(char **key_string)
2624 if (*key_string[0] != ':')
2625 return strlen(*key_string) >> 1;
2627 /* look for next ':' in key string */
2628 colon = strpbrk(*key_string + 1, ":");
2632 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2635 *key_string = colon;
2637 /* remaining key string should be :<logon|user>:<key_desc> */
2644 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2649 static int get_key_size(char **key_string)
2651 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2654 #endif /* CONFIG_KEYS */
2656 static int crypt_set_key(struct crypt_config *cc, char *key)
2659 int key_string_len = strlen(key);
2661 /* Hyphen (which gives a key_size of zero) means there is no key. */
2662 if (!cc->key_size && strcmp(key, "-"))
2665 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2666 if (key[0] == ':') {
2667 r = crypt_set_keyring_key(cc, key + 1);
2671 /* clear the flag since following operations may invalidate previously valid key */
2672 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2674 /* wipe references to any kernel keyring key */
2675 kfree_sensitive(cc->key_string);
2676 cc->key_string = NULL;
2678 /* Decode key from its hex representation. */
2679 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2682 r = crypt_setkey(cc);
2684 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2687 /* Hex key string not needed after here, so wipe it. */
2688 memset(key, '0', key_string_len);
2693 static int crypt_wipe_key(struct crypt_config *cc)
2697 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2698 get_random_bytes(&cc->key, cc->key_size);
2700 /* Wipe IV private keys */
2701 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2702 r = cc->iv_gen_ops->wipe(cc);
2707 kfree_sensitive(cc->key_string);
2708 cc->key_string = NULL;
2709 r = crypt_setkey(cc);
2710 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2715 static void crypt_calculate_pages_per_client(void)
2717 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2719 if (!dm_crypt_clients_n)
2722 pages /= dm_crypt_clients_n;
2723 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2724 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2725 dm_crypt_pages_per_client = pages;
2728 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2730 struct crypt_config *cc = pool_data;
2734 * Note, percpu_counter_read_positive() may over (and under) estimate
2735 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2736 * but avoids potential spinlock contention of an exact result.
2738 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2739 likely(gfp_mask & __GFP_NORETRY))
2742 page = alloc_page(gfp_mask);
2743 if (likely(page != NULL))
2744 percpu_counter_add(&cc->n_allocated_pages, 1);
2749 static void crypt_page_free(void *page, void *pool_data)
2751 struct crypt_config *cc = pool_data;
2754 percpu_counter_sub(&cc->n_allocated_pages, 1);
2757 static void crypt_dtr(struct dm_target *ti)
2759 struct crypt_config *cc = ti->private;
2766 if (cc->write_thread)
2767 kthread_stop(cc->write_thread);
2770 destroy_workqueue(cc->io_queue);
2771 if (cc->crypt_queue)
2772 destroy_workqueue(cc->crypt_queue);
2774 crypt_free_tfms(cc);
2776 bioset_exit(&cc->bs);
2778 mempool_exit(&cc->page_pool);
2779 mempool_exit(&cc->req_pool);
2780 mempool_exit(&cc->tag_pool);
2782 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2783 percpu_counter_destroy(&cc->n_allocated_pages);
2785 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2786 cc->iv_gen_ops->dtr(cc);
2789 dm_put_device(ti, cc->dev);
2791 kfree_sensitive(cc->cipher_string);
2792 kfree_sensitive(cc->key_string);
2793 kfree_sensitive(cc->cipher_auth);
2794 kfree_sensitive(cc->authenc_key);
2796 mutex_destroy(&cc->bio_alloc_lock);
2798 /* Must zero key material before freeing */
2799 kfree_sensitive(cc);
2801 spin_lock(&dm_crypt_clients_lock);
2802 WARN_ON(!dm_crypt_clients_n);
2803 dm_crypt_clients_n--;
2804 crypt_calculate_pages_per_client();
2805 spin_unlock(&dm_crypt_clients_lock);
2807 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2810 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2812 struct crypt_config *cc = ti->private;
2814 if (crypt_integrity_aead(cc))
2815 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2817 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2820 /* at least a 64 bit sector number should fit in our buffer */
2821 cc->iv_size = max(cc->iv_size,
2822 (unsigned int)(sizeof(u64) / sizeof(u8)));
2824 DMWARN("Selected cipher does not support IVs");
2828 /* Choose ivmode, see comments at iv code. */
2830 cc->iv_gen_ops = NULL;
2831 else if (strcmp(ivmode, "plain") == 0)
2832 cc->iv_gen_ops = &crypt_iv_plain_ops;
2833 else if (strcmp(ivmode, "plain64") == 0)
2834 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2835 else if (strcmp(ivmode, "plain64be") == 0)
2836 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2837 else if (strcmp(ivmode, "essiv") == 0)
2838 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2839 else if (strcmp(ivmode, "benbi") == 0)
2840 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2841 else if (strcmp(ivmode, "null") == 0)
2842 cc->iv_gen_ops = &crypt_iv_null_ops;
2843 else if (strcmp(ivmode, "eboiv") == 0)
2844 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2845 else if (strcmp(ivmode, "elephant") == 0) {
2846 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2848 cc->key_extra_size = cc->key_size / 2;
2849 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2851 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2852 } else if (strcmp(ivmode, "lmk") == 0) {
2853 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2855 * Version 2 and 3 is recognised according
2856 * to length of provided multi-key string.
2857 * If present (version 3), last key is used as IV seed.
2858 * All keys (including IV seed) are always the same size.
2860 if (cc->key_size % cc->key_parts) {
2862 cc->key_extra_size = cc->key_size / cc->key_parts;
2864 } else if (strcmp(ivmode, "tcw") == 0) {
2865 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2866 cc->key_parts += 2; /* IV + whitening */
2867 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2868 } else if (strcmp(ivmode, "random") == 0) {
2869 cc->iv_gen_ops = &crypt_iv_random_ops;
2870 /* Need storage space in integrity fields. */
2871 cc->integrity_iv_size = cc->iv_size;
2873 ti->error = "Invalid IV mode";
2881 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2882 * The HMAC is needed to calculate tag size (HMAC digest size).
2883 * This should be probably done by crypto-api calls (once available...)
2885 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2887 char *start, *end, *mac_alg = NULL;
2888 struct crypto_ahash *mac;
2890 if (!strstarts(cipher_api, "authenc("))
2893 start = strchr(cipher_api, '(');
2894 end = strchr(cipher_api, ',');
2895 if (!start || !end || ++start > end)
2898 mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2902 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2906 return PTR_ERR(mac);
2908 cc->key_mac_size = crypto_ahash_digestsize(mac);
2909 crypto_free_ahash(mac);
2911 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2912 if (!cc->authenc_key)
2918 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2919 char **ivmode, char **ivopts)
2921 struct crypt_config *cc = ti->private;
2922 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2928 * New format (capi: prefix)
2929 * capi:cipher_api_spec-iv:ivopts
2931 tmp = &cipher_in[strlen("capi:")];
2933 /* Separate IV options if present, it can contain another '-' in hash name */
2934 *ivopts = strrchr(tmp, ':');
2940 *ivmode = strrchr(tmp, '-');
2945 /* The rest is crypto API spec */
2948 /* Alloc AEAD, can be used only in new format. */
2949 if (crypt_integrity_aead(cc)) {
2950 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2952 ti->error = "Invalid AEAD cipher spec";
2957 if (*ivmode && !strcmp(*ivmode, "lmk"))
2958 cc->tfms_count = 64;
2960 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2962 ti->error = "Digest algorithm missing for ESSIV mode";
2965 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2966 cipher_api, *ivopts);
2967 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2968 ti->error = "Cannot allocate cipher string";
2974 cc->key_parts = cc->tfms_count;
2976 /* Allocate cipher */
2977 ret = crypt_alloc_tfms(cc, cipher_api);
2979 ti->error = "Error allocating crypto tfm";
2983 if (crypt_integrity_aead(cc))
2984 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2986 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2991 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2992 char **ivmode, char **ivopts)
2994 struct crypt_config *cc = ti->private;
2995 char *tmp, *cipher, *chainmode, *keycount;
2996 char *cipher_api = NULL;
3000 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
3001 ti->error = "Bad cipher specification";
3006 * Legacy dm-crypt cipher specification
3007 * cipher[:keycount]-mode-iv:ivopts
3010 keycount = strsep(&tmp, "-");
3011 cipher = strsep(&keycount, ":");
3015 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3016 !is_power_of_2(cc->tfms_count)) {
3017 ti->error = "Bad cipher key count specification";
3020 cc->key_parts = cc->tfms_count;
3022 chainmode = strsep(&tmp, "-");
3023 *ivmode = strsep(&tmp, ":");
3027 * For compatibility with the original dm-crypt mapping format, if
3028 * only the cipher name is supplied, use cbc-plain.
3030 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3035 if (strcmp(chainmode, "ecb") && !*ivmode) {
3036 ti->error = "IV mechanism required";
3040 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3044 if (*ivmode && !strcmp(*ivmode, "essiv")) {
3046 ti->error = "Digest algorithm missing for ESSIV mode";
3050 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3051 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3053 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3054 "%s(%s)", chainmode, cipher);
3056 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3061 /* Allocate cipher */
3062 ret = crypt_alloc_tfms(cc, cipher_api);
3064 ti->error = "Error allocating crypto tfm";
3072 ti->error = "Cannot allocate cipher strings";
3076 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3078 struct crypt_config *cc = ti->private;
3079 char *ivmode = NULL, *ivopts = NULL;
3082 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3083 if (!cc->cipher_string) {
3084 ti->error = "Cannot allocate cipher strings";
3088 if (strstarts(cipher_in, "capi:"))
3089 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3091 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3096 ret = crypt_ctr_ivmode(ti, ivmode);
3100 /* Initialize and set key */
3101 ret = crypt_set_key(cc, key);
3103 ti->error = "Error decoding and setting key";
3108 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3109 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3111 ti->error = "Error creating IV";
3116 /* Initialize IV (set keys for ESSIV etc) */
3117 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3118 ret = cc->iv_gen_ops->init(cc);
3120 ti->error = "Error initialising IV";
3125 /* wipe the kernel key payload copy */
3127 memset(cc->key, 0, cc->key_size * sizeof(u8));
3132 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3134 struct crypt_config *cc = ti->private;
3135 struct dm_arg_set as;
3136 static const struct dm_arg _args[] = {
3137 {0, 8, "Invalid number of feature args"},
3139 unsigned int opt_params, val;
3140 const char *opt_string, *sval;
3144 /* Optional parameters */
3148 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3152 while (opt_params--) {
3153 opt_string = dm_shift_arg(&as);
3155 ti->error = "Not enough feature arguments";
3159 if (!strcasecmp(opt_string, "allow_discards"))
3160 ti->num_discard_bios = 1;
3162 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3163 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3165 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3166 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3167 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3168 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3169 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3170 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3171 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3172 if (val == 0 || val > MAX_TAG_SIZE) {
3173 ti->error = "Invalid integrity arguments";
3176 cc->on_disk_tag_size = val;
3177 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3178 if (!strcasecmp(sval, "aead")) {
3179 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3180 } else if (strcasecmp(sval, "none")) {
3181 ti->error = "Unknown integrity profile";
3185 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3186 if (!cc->cipher_auth)
3188 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3189 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3190 cc->sector_size > 4096 ||
3191 (cc->sector_size & (cc->sector_size - 1))) {
3192 ti->error = "Invalid feature value for sector_size";
3195 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3196 ti->error = "Device size is not multiple of sector_size feature";
3199 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3200 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3201 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3203 ti->error = "Invalid feature arguments";
3211 #ifdef CONFIG_BLK_DEV_ZONED
3212 static int crypt_report_zones(struct dm_target *ti,
3213 struct dm_report_zones_args *args, unsigned int nr_zones)
3215 struct crypt_config *cc = ti->private;
3217 return dm_report_zones(cc->dev->bdev, cc->start,
3218 cc->start + dm_target_offset(ti, args->next_sector),
3222 #define crypt_report_zones NULL
3226 * Construct an encryption mapping:
3227 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3229 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3231 struct crypt_config *cc;
3232 const char *devname = dm_table_device_name(ti->table);
3234 unsigned int align_mask;
3235 unsigned long long tmpll;
3237 size_t iv_size_padding, additional_req_size;
3241 ti->error = "Not enough arguments";
3245 key_size = get_key_size(&argv[1]);
3247 ti->error = "Cannot parse key size";
3251 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3253 ti->error = "Cannot allocate encryption context";
3256 cc->key_size = key_size;
3257 cc->sector_size = (1 << SECTOR_SHIFT);
3258 cc->sector_shift = 0;
3262 spin_lock(&dm_crypt_clients_lock);
3263 dm_crypt_clients_n++;
3264 crypt_calculate_pages_per_client();
3265 spin_unlock(&dm_crypt_clients_lock);
3267 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3271 /* Optional parameters need to be read before cipher constructor */
3273 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3278 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3282 if (crypt_integrity_aead(cc)) {
3283 cc->dmreq_start = sizeof(struct aead_request);
3284 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3285 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3287 cc->dmreq_start = sizeof(struct skcipher_request);
3288 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3289 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3291 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3293 if (align_mask < CRYPTO_MINALIGN) {
3294 /* Allocate the padding exactly */
3295 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3299 * If the cipher requires greater alignment than kmalloc
3300 * alignment, we don't know the exact position of the
3301 * initialization vector. We must assume worst case.
3303 iv_size_padding = align_mask;
3306 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3307 additional_req_size = sizeof(struct dm_crypt_request) +
3308 iv_size_padding + cc->iv_size +
3311 sizeof(unsigned int);
3313 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3315 ti->error = "Cannot allocate crypt request mempool";
3319 cc->per_bio_data_size = ti->per_io_data_size =
3320 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3323 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3325 ti->error = "Cannot allocate page mempool";
3329 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3331 ti->error = "Cannot allocate crypt bioset";
3335 mutex_init(&cc->bio_alloc_lock);
3338 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3339 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3340 ti->error = "Invalid iv_offset sector";
3343 cc->iv_offset = tmpll;
3345 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3347 ti->error = "Device lookup failed";
3352 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3353 ti->error = "Invalid device sector";
3358 if (bdev_is_zoned(cc->dev->bdev)) {
3360 * For zoned block devices, we need to preserve the issuer write
3361 * ordering. To do so, disable write workqueues and force inline
3362 * encryption completion.
3364 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3365 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3368 * All zone append writes to a zone of a zoned block device will
3369 * have the same BIO sector, the start of the zone. When the
3370 * cypher IV mode uses sector values, all data targeting a
3371 * zone will be encrypted using the first sector numbers of the
3372 * zone. This will not result in write errors but will
3373 * cause most reads to fail as reads will use the sector values
3374 * for the actual data locations, resulting in IV mismatch.
3375 * To avoid this problem, ask DM core to emulate zone append
3376 * operations with regular writes.
3378 DMDEBUG("Zone append operations will be emulated");
3379 ti->emulate_zone_append = true;
3382 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3383 ret = crypt_integrity_ctr(cc, ti);
3387 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3388 if (!cc->tag_pool_max_sectors)
3389 cc->tag_pool_max_sectors = 1;
3391 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3392 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3394 ti->error = "Cannot allocate integrity tags mempool";
3398 cc->tag_pool_max_sectors <<= cc->sector_shift;
3402 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3403 if (!cc->io_queue) {
3404 ti->error = "Couldn't create kcryptd io queue";
3408 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3409 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3412 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3413 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3414 num_online_cpus(), devname);
3415 if (!cc->crypt_queue) {
3416 ti->error = "Couldn't create kcryptd queue";
3420 spin_lock_init(&cc->write_thread_lock);
3421 cc->write_tree = RB_ROOT;
3423 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3424 if (IS_ERR(cc->write_thread)) {
3425 ret = PTR_ERR(cc->write_thread);
3426 cc->write_thread = NULL;
3427 ti->error = "Couldn't spawn write thread";
3431 ti->num_flush_bios = 1;
3432 ti->limit_swap_bios = true;
3433 ti->accounts_remapped_io = true;
3435 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3439 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3444 static int crypt_map(struct dm_target *ti, struct bio *bio)
3446 struct dm_crypt_io *io;
3447 struct crypt_config *cc = ti->private;
3450 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3451 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3452 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3454 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3455 bio_op(bio) == REQ_OP_DISCARD)) {
3456 bio_set_dev(bio, cc->dev->bdev);
3457 if (bio_sectors(bio))
3458 bio->bi_iter.bi_sector = cc->start +
3459 dm_target_offset(ti, bio->bi_iter.bi_sector);
3460 return DM_MAPIO_REMAPPED;
3464 * Check if bio is too large, split as needed.
3466 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3467 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3468 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3471 * Ensure that bio is a multiple of internal sector encryption size
3472 * and is aligned to this size as defined in IO hints.
3474 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3475 return DM_MAPIO_KILL;
3477 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3478 return DM_MAPIO_KILL;
3480 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3481 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3483 if (cc->on_disk_tag_size) {
3484 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3486 if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3487 io->integrity_metadata = NULL;
3489 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3491 if (unlikely(!io->integrity_metadata)) {
3492 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3493 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3494 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3495 io->integrity_metadata_from_pool = true;
3499 if (crypt_integrity_aead(cc))
3500 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3502 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3504 if (bio_data_dir(io->base_bio) == READ) {
3505 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3506 kcryptd_queue_read(io);
3508 kcryptd_queue_crypt(io);
3510 return DM_MAPIO_SUBMITTED;
3513 static char hex2asc(unsigned char c)
3515 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3518 static void crypt_status(struct dm_target *ti, status_type_t type,
3519 unsigned int status_flags, char *result, unsigned int maxlen)
3521 struct crypt_config *cc = ti->private;
3522 unsigned int i, sz = 0;
3523 int num_feature_args = 0;
3526 case STATUSTYPE_INFO:
3530 case STATUSTYPE_TABLE:
3531 DMEMIT("%s ", cc->cipher_string);
3533 if (cc->key_size > 0) {
3535 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3537 for (i = 0; i < cc->key_size; i++) {
3538 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3539 hex2asc(cc->key[i] & 0xf));
3545 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3546 cc->dev->name, (unsigned long long)cc->start);
3548 num_feature_args += !!ti->num_discard_bios;
3549 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3550 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3551 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3552 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3553 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3554 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3555 if (cc->on_disk_tag_size)
3557 if (num_feature_args) {
3558 DMEMIT(" %d", num_feature_args);
3559 if (ti->num_discard_bios)
3560 DMEMIT(" allow_discards");
3561 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3562 DMEMIT(" same_cpu_crypt");
3563 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3564 DMEMIT(" submit_from_crypt_cpus");
3565 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3566 DMEMIT(" no_read_workqueue");
3567 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3568 DMEMIT(" no_write_workqueue");
3569 if (cc->on_disk_tag_size)
3570 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3571 if (cc->sector_size != (1 << SECTOR_SHIFT))
3572 DMEMIT(" sector_size:%d", cc->sector_size);
3573 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3574 DMEMIT(" iv_large_sectors");
3578 case STATUSTYPE_IMA:
3579 DMEMIT_TARGET_NAME_VERSION(ti->type);
3580 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3581 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3582 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3584 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3586 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3588 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3591 if (cc->on_disk_tag_size)
3592 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3593 cc->on_disk_tag_size, cc->cipher_auth);
3594 if (cc->sector_size != (1 << SECTOR_SHIFT))
3595 DMEMIT(",sector_size=%d", cc->sector_size);
3596 if (cc->cipher_string)
3597 DMEMIT(",cipher_string=%s", cc->cipher_string);
3599 DMEMIT(",key_size=%u", cc->key_size);
3600 DMEMIT(",key_parts=%u", cc->key_parts);
3601 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3602 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3608 static void crypt_postsuspend(struct dm_target *ti)
3610 struct crypt_config *cc = ti->private;
3612 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3615 static int crypt_preresume(struct dm_target *ti)
3617 struct crypt_config *cc = ti->private;
3619 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3620 DMERR("aborting resume - crypt key is not set.");
3627 static void crypt_resume(struct dm_target *ti)
3629 struct crypt_config *cc = ti->private;
3631 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3634 /* Message interface
3638 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3639 char *result, unsigned int maxlen)
3641 struct crypt_config *cc = ti->private;
3642 int key_size, ret = -EINVAL;
3647 if (!strcasecmp(argv[0], "key")) {
3648 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3649 DMWARN("not suspended during key manipulation.");
3652 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3653 /* The key size may not be changed. */
3654 key_size = get_key_size(&argv[2]);
3655 if (key_size < 0 || cc->key_size != key_size) {
3656 memset(argv[2], '0', strlen(argv[2]));
3660 ret = crypt_set_key(cc, argv[2]);
3663 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3664 ret = cc->iv_gen_ops->init(cc);
3665 /* wipe the kernel key payload copy */
3667 memset(cc->key, 0, cc->key_size * sizeof(u8));
3670 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3671 return crypt_wipe_key(cc);
3675 DMWARN("unrecognised message received.");
3679 static int crypt_iterate_devices(struct dm_target *ti,
3680 iterate_devices_callout_fn fn, void *data)
3682 struct crypt_config *cc = ti->private;
3684 return fn(ti, cc->dev, cc->start, ti->len, data);
3687 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3689 struct crypt_config *cc = ti->private;
3692 * Unfortunate constraint that is required to avoid the potential
3693 * for exceeding underlying device's max_segments limits -- due to
3694 * crypt_alloc_buffer() possibly allocating pages for the encryption
3695 * bio that are not as physically contiguous as the original bio.
3697 limits->max_segment_size = PAGE_SIZE;
3699 limits->logical_block_size =
3700 max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3701 limits->physical_block_size =
3702 max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3703 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3704 limits->dma_alignment = limits->logical_block_size - 1;
3707 static struct target_type crypt_target = {
3709 .version = {1, 25, 0},
3710 .module = THIS_MODULE,
3713 .features = DM_TARGET_ZONED_HM,
3714 .report_zones = crypt_report_zones,
3716 .status = crypt_status,
3717 .postsuspend = crypt_postsuspend,
3718 .preresume = crypt_preresume,
3719 .resume = crypt_resume,
3720 .message = crypt_message,
3721 .iterate_devices = crypt_iterate_devices,
3722 .io_hints = crypt_io_hints,
3726 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3727 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3728 MODULE_LICENSE("GPL");