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;
77 struct work_struct work;
79 struct convert_context ctx;
85 struct rb_node rb_node;
86 } CRYPTO_MINALIGN_ATTR;
88 struct dm_crypt_request {
89 struct convert_context *ctx;
90 struct scatterlist sg_in[4];
91 struct scatterlist sg_out[4];
97 struct crypt_iv_operations {
98 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
100 void (*dtr)(struct crypt_config *cc);
101 int (*init)(struct crypt_config *cc);
102 int (*wipe)(struct crypt_config *cc);
103 int (*generator)(struct crypt_config *cc, u8 *iv,
104 struct dm_crypt_request *dmreq);
105 int (*post)(struct crypt_config *cc, u8 *iv,
106 struct dm_crypt_request *dmreq);
109 struct iv_benbi_private {
113 #define LMK_SEED_SIZE 64 /* hash + 0 */
114 struct iv_lmk_private {
115 struct crypto_shash *hash_tfm;
119 #define TCW_WHITENING_SIZE 16
120 struct iv_tcw_private {
121 struct crypto_shash *crc32_tfm;
126 #define ELEPHANT_MAX_KEY_SIZE 32
127 struct iv_elephant_private {
128 struct crypto_skcipher *tfm;
132 * Crypt: maps a linear range of a block device
133 * and encrypts / decrypts at the same time.
135 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
136 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
137 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
138 DM_CRYPT_WRITE_INLINE };
141 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
142 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
143 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
147 * The fields in here must be read only after initialization.
149 struct crypt_config {
153 struct percpu_counter n_allocated_pages;
155 struct workqueue_struct *io_queue;
156 struct workqueue_struct *crypt_queue;
158 spinlock_t write_thread_lock;
159 struct task_struct *write_thread;
160 struct rb_root write_tree;
166 const struct crypt_iv_operations *iv_gen_ops;
168 struct iv_benbi_private benbi;
169 struct iv_lmk_private lmk;
170 struct iv_tcw_private tcw;
171 struct iv_elephant_private elephant;
174 unsigned int iv_size;
175 unsigned short sector_size;
176 unsigned char sector_shift;
179 struct crypto_skcipher **tfms;
180 struct crypto_aead **tfms_aead;
182 unsigned int tfms_count;
183 unsigned long cipher_flags;
186 * Layout of each crypto request:
188 * struct skcipher_request
191 * struct dm_crypt_request
195 * The padding is added so that dm_crypt_request and the IV are
198 unsigned int dmreq_start;
200 unsigned int per_bio_data_size;
203 unsigned int key_size;
204 unsigned int key_parts; /* independent parts in key buffer */
205 unsigned int key_extra_size; /* additional keys length */
206 unsigned int key_mac_size; /* MAC key size for authenc(...) */
208 unsigned int integrity_tag_size;
209 unsigned int integrity_iv_size;
210 unsigned int on_disk_tag_size;
213 * pool for per bio private data, crypto requests,
214 * encryption requeusts/buffer pages and integrity tags
216 unsigned int tag_pool_max_sectors;
222 struct mutex bio_alloc_lock;
224 u8 *authenc_key; /* space for keys in authenc() format (if used) */
225 u8 key[] __counted_by(key_size);
229 #define MAX_TAG_SIZE 480
230 #define POOL_ENTRY_SIZE 512
232 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
233 static unsigned int dm_crypt_clients_n;
234 static volatile unsigned long dm_crypt_pages_per_client;
235 #define DM_CRYPT_MEMORY_PERCENT 2
236 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
238 static void crypt_endio(struct bio *clone);
239 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
240 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
241 struct scatterlist *sg);
243 static bool crypt_integrity_aead(struct crypt_config *cc);
246 * Use this to access cipher attributes that are independent of the key.
248 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
250 return cc->cipher_tfm.tfms[0];
253 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
255 return cc->cipher_tfm.tfms_aead[0];
259 * Different IV generation algorithms:
261 * plain: the initial vector is the 32-bit little-endian version of the sector
262 * number, padded with zeros if necessary.
264 * plain64: the initial vector is the 64-bit little-endian version of the sector
265 * number, padded with zeros if necessary.
267 * plain64be: the initial vector is the 64-bit big-endian version of the sector
268 * number, padded with zeros if necessary.
270 * essiv: "encrypted sector|salt initial vector", the sector number is
271 * encrypted with the bulk cipher using a salt as key. The salt
272 * should be derived from the bulk cipher's key via hashing.
274 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
275 * (needed for LRW-32-AES and possible other narrow block modes)
277 * null: the initial vector is always zero. Provides compatibility with
278 * obsolete loop_fish2 devices. Do not use for new devices.
280 * lmk: Compatible implementation of the block chaining mode used
281 * by the Loop-AES block device encryption system
282 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
283 * It operates on full 512 byte sectors and uses CBC
284 * with an IV derived from the sector number, the data and
285 * optionally extra IV seed.
286 * This means that after decryption the first block
287 * of sector must be tweaked according to decrypted data.
288 * Loop-AES can use three encryption schemes:
289 * version 1: is plain aes-cbc mode
290 * version 2: uses 64 multikey scheme with lmk IV generator
291 * version 3: the same as version 2 with additional IV seed
292 * (it uses 65 keys, last key is used as IV seed)
294 * tcw: Compatible implementation of the block chaining mode used
295 * by the TrueCrypt device encryption system (prior to version 4.1).
296 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
297 * It operates on full 512 byte sectors and uses CBC
298 * with an IV derived from initial key and the sector number.
299 * In addition, whitening value is applied on every sector, whitening
300 * is calculated from initial key, sector number and mixed using CRC32.
301 * Note that this encryption scheme is vulnerable to watermarking attacks
302 * and should be used for old compatible containers access only.
304 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
305 * The IV is encrypted little-endian byte-offset (with the same key
306 * and cipher as the volume).
308 * elephant: The extended version of eboiv with additional Elephant diffuser
309 * used with Bitlocker CBC mode.
310 * This mode was used in older Windows systems
311 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
314 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
315 struct dm_crypt_request *dmreq)
317 memset(iv, 0, cc->iv_size);
318 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
323 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
324 struct dm_crypt_request *dmreq)
326 memset(iv, 0, cc->iv_size);
327 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
332 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
333 struct dm_crypt_request *dmreq)
335 memset(iv, 0, cc->iv_size);
336 /* iv_size is at least of size u64; usually it is 16 bytes */
337 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
342 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
343 struct dm_crypt_request *dmreq)
346 * ESSIV encryption of the IV is now handled by the crypto API,
347 * so just pass the plain sector number here.
349 memset(iv, 0, cc->iv_size);
350 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
355 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
361 if (crypt_integrity_aead(cc))
362 bs = crypto_aead_blocksize(any_tfm_aead(cc));
364 bs = crypto_skcipher_blocksize(any_tfm(cc));
368 * We need to calculate how far we must shift the sector count
369 * to get the cipher block count, we use this shift in _gen.
371 if (1 << log != bs) {
372 ti->error = "cypher blocksize is not a power of 2";
377 ti->error = "cypher blocksize is > 512";
381 cc->iv_gen_private.benbi.shift = 9 - log;
386 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
390 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
391 struct dm_crypt_request *dmreq)
395 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
397 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
398 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
403 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
404 struct dm_crypt_request *dmreq)
406 memset(iv, 0, cc->iv_size);
411 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
413 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
415 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
416 crypto_free_shash(lmk->hash_tfm);
417 lmk->hash_tfm = NULL;
419 kfree_sensitive(lmk->seed);
423 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
426 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
428 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
429 ti->error = "Unsupported sector size for LMK";
433 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
434 CRYPTO_ALG_ALLOCATES_MEMORY);
435 if (IS_ERR(lmk->hash_tfm)) {
436 ti->error = "Error initializing LMK hash";
437 return PTR_ERR(lmk->hash_tfm);
440 /* No seed in LMK version 2 */
441 if (cc->key_parts == cc->tfms_count) {
446 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
448 crypt_iv_lmk_dtr(cc);
449 ti->error = "Error kmallocing seed storage in LMK";
456 static int crypt_iv_lmk_init(struct crypt_config *cc)
458 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
459 int subkey_size = cc->key_size / cc->key_parts;
461 /* LMK seed is on the position of LMK_KEYS + 1 key */
463 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
464 crypto_shash_digestsize(lmk->hash_tfm));
469 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
471 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
474 memset(lmk->seed, 0, LMK_SEED_SIZE);
479 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
480 struct dm_crypt_request *dmreq,
483 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
484 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
485 struct md5_state md5state;
489 desc->tfm = lmk->hash_tfm;
491 r = crypto_shash_init(desc);
496 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
501 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
502 r = crypto_shash_update(desc, data + 16, 16 * 31);
506 /* Sector is cropped to 56 bits here */
507 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
508 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
509 buf[2] = cpu_to_le32(4024);
511 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
515 /* No MD5 padding here */
516 r = crypto_shash_export(desc, &md5state);
520 for (i = 0; i < MD5_HASH_WORDS; i++)
521 __cpu_to_le32s(&md5state.hash[i]);
522 memcpy(iv, &md5state.hash, cc->iv_size);
527 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
528 struct dm_crypt_request *dmreq)
530 struct scatterlist *sg;
534 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
535 sg = crypt_get_sg_data(cc, dmreq->sg_in);
536 src = kmap_local_page(sg_page(sg));
537 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
540 memset(iv, 0, cc->iv_size);
545 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
546 struct dm_crypt_request *dmreq)
548 struct scatterlist *sg;
552 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
555 sg = crypt_get_sg_data(cc, dmreq->sg_out);
556 dst = kmap_local_page(sg_page(sg));
557 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
559 /* Tweak the first block of plaintext sector */
561 crypto_xor(dst + sg->offset, iv, cc->iv_size);
567 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
569 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
571 kfree_sensitive(tcw->iv_seed);
573 kfree_sensitive(tcw->whitening);
574 tcw->whitening = NULL;
576 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
577 crypto_free_shash(tcw->crc32_tfm);
578 tcw->crc32_tfm = NULL;
581 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
584 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
586 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
587 ti->error = "Unsupported sector size for TCW";
591 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
592 ti->error = "Wrong key size for TCW";
596 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
597 CRYPTO_ALG_ALLOCATES_MEMORY);
598 if (IS_ERR(tcw->crc32_tfm)) {
599 ti->error = "Error initializing CRC32 in TCW";
600 return PTR_ERR(tcw->crc32_tfm);
603 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
604 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
605 if (!tcw->iv_seed || !tcw->whitening) {
606 crypt_iv_tcw_dtr(cc);
607 ti->error = "Error allocating seed storage in TCW";
614 static int crypt_iv_tcw_init(struct crypt_config *cc)
616 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
617 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
619 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
620 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
626 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
628 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
630 memset(tcw->iv_seed, 0, cc->iv_size);
631 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
636 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
637 struct dm_crypt_request *dmreq,
640 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
641 __le64 sector = cpu_to_le64(dmreq->iv_sector);
642 u8 buf[TCW_WHITENING_SIZE];
643 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
646 /* xor whitening with sector number */
647 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
648 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
650 /* calculate crc32 for every 32bit part and xor it */
651 desc->tfm = tcw->crc32_tfm;
652 for (i = 0; i < 4; i++) {
653 r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
657 crypto_xor(&buf[0], &buf[12], 4);
658 crypto_xor(&buf[4], &buf[8], 4);
660 /* apply whitening (8 bytes) to whole sector */
661 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
662 crypto_xor(data + i * 8, buf, 8);
664 memzero_explicit(buf, sizeof(buf));
668 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
669 struct dm_crypt_request *dmreq)
671 struct scatterlist *sg;
672 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
673 __le64 sector = cpu_to_le64(dmreq->iv_sector);
677 /* Remove whitening from ciphertext */
678 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
679 sg = crypt_get_sg_data(cc, dmreq->sg_in);
680 src = kmap_local_page(sg_page(sg));
681 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
686 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
688 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
694 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
695 struct dm_crypt_request *dmreq)
697 struct scatterlist *sg;
701 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
704 /* Apply whitening on ciphertext */
705 sg = crypt_get_sg_data(cc, dmreq->sg_out);
706 dst = kmap_local_page(sg_page(sg));
707 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
713 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
714 struct dm_crypt_request *dmreq)
716 /* Used only for writes, there must be an additional space to store IV */
717 get_random_bytes(iv, cc->iv_size);
721 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
724 if (crypt_integrity_aead(cc)) {
725 ti->error = "AEAD transforms not supported for EBOIV";
729 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
730 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
737 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
738 struct dm_crypt_request *dmreq)
740 struct crypto_skcipher *tfm = any_tfm(cc);
741 struct skcipher_request *req;
742 struct scatterlist src, dst;
743 DECLARE_CRYPTO_WAIT(wait);
744 unsigned int reqsize;
748 reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
749 reqsize = ALIGN(reqsize, __alignof__(__le64));
751 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
755 skcipher_request_set_tfm(req, tfm);
757 buf = (u8 *)req + reqsize;
758 memset(buf, 0, cc->iv_size);
759 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
761 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
762 sg_init_one(&dst, iv, cc->iv_size);
763 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
764 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
765 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
766 kfree_sensitive(req);
771 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
773 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
775 crypto_free_skcipher(elephant->tfm);
776 elephant->tfm = NULL;
779 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
782 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
785 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
786 CRYPTO_ALG_ALLOCATES_MEMORY);
787 if (IS_ERR(elephant->tfm)) {
788 r = PTR_ERR(elephant->tfm);
789 elephant->tfm = NULL;
793 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
795 crypt_iv_elephant_dtr(cc);
799 static void diffuser_disk_to_cpu(u32 *d, size_t n)
801 #ifndef __LITTLE_ENDIAN
804 for (i = 0; i < n; i++)
805 d[i] = le32_to_cpu((__le32)d[i]);
809 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
811 #ifndef __LITTLE_ENDIAN
814 for (i = 0; i < n; i++)
815 d[i] = cpu_to_le32((u32)d[i]);
819 static void diffuser_a_decrypt(u32 *d, size_t n)
823 for (i = 0; i < 5; i++) {
828 while (i1 < (n - 1)) {
829 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
835 d[i1] += d[i2] ^ d[i3];
841 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
844 d[i1] += d[i2] ^ d[i3];
850 static void diffuser_a_encrypt(u32 *d, size_t n)
854 for (i = 0; i < 5; i++) {
860 d[i1] -= d[i2] ^ d[i3];
863 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
869 d[i1] -= d[i2] ^ d[i3];
875 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
881 static void diffuser_b_decrypt(u32 *d, size_t n)
885 for (i = 0; i < 3; i++) {
890 while (i1 < (n - 1)) {
891 d[i1] += d[i2] ^ d[i3];
894 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
900 d[i1] += d[i2] ^ d[i3];
906 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
912 static void diffuser_b_encrypt(u32 *d, size_t n)
916 for (i = 0; i < 3; i++) {
922 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
928 d[i1] -= d[i2] ^ d[i3];
934 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
937 d[i1] -= d[i2] ^ d[i3];
943 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
945 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
946 u8 *es, *ks, *data, *data2, *data_offset;
947 struct skcipher_request *req;
948 struct scatterlist *sg, *sg2, src, dst;
949 DECLARE_CRYPTO_WAIT(wait);
952 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
953 es = kzalloc(16, GFP_NOIO); /* Key for AES */
954 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
956 if (!req || !es || !ks) {
961 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
964 sg_init_one(&src, es, 16);
965 sg_init_one(&dst, ks, 16);
966 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
967 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
968 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
974 sg_init_one(&dst, &ks[16], 16);
975 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
979 sg = crypt_get_sg_data(cc, dmreq->sg_out);
980 data = kmap_local_page(sg_page(sg));
981 data_offset = data + sg->offset;
983 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
984 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
985 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
986 data2 = kmap_local_page(sg_page(sg2));
987 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
991 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
992 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
993 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
994 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
995 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
998 for (i = 0; i < (cc->sector_size / 32); i++)
999 crypto_xor(data_offset + i * 32, ks, 32);
1001 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1002 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1003 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1004 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1005 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1010 kfree_sensitive(ks);
1011 kfree_sensitive(es);
1012 skcipher_request_free(req);
1016 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1017 struct dm_crypt_request *dmreq)
1021 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1022 r = crypt_iv_elephant(cc, dmreq);
1027 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1030 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1031 struct dm_crypt_request *dmreq)
1033 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1034 return crypt_iv_elephant(cc, dmreq);
1039 static int crypt_iv_elephant_init(struct crypt_config *cc)
1041 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1042 int key_offset = cc->key_size - cc->key_extra_size;
1044 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1047 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1049 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1050 u8 key[ELEPHANT_MAX_KEY_SIZE];
1052 memset(key, 0, cc->key_extra_size);
1053 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1056 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1057 .generator = crypt_iv_plain_gen
1060 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1061 .generator = crypt_iv_plain64_gen
1064 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1065 .generator = crypt_iv_plain64be_gen
1068 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1069 .generator = crypt_iv_essiv_gen
1072 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1073 .ctr = crypt_iv_benbi_ctr,
1074 .dtr = crypt_iv_benbi_dtr,
1075 .generator = crypt_iv_benbi_gen
1078 static const struct crypt_iv_operations crypt_iv_null_ops = {
1079 .generator = crypt_iv_null_gen
1082 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1083 .ctr = crypt_iv_lmk_ctr,
1084 .dtr = crypt_iv_lmk_dtr,
1085 .init = crypt_iv_lmk_init,
1086 .wipe = crypt_iv_lmk_wipe,
1087 .generator = crypt_iv_lmk_gen,
1088 .post = crypt_iv_lmk_post
1091 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1092 .ctr = crypt_iv_tcw_ctr,
1093 .dtr = crypt_iv_tcw_dtr,
1094 .init = crypt_iv_tcw_init,
1095 .wipe = crypt_iv_tcw_wipe,
1096 .generator = crypt_iv_tcw_gen,
1097 .post = crypt_iv_tcw_post
1100 static const struct crypt_iv_operations crypt_iv_random_ops = {
1101 .generator = crypt_iv_random_gen
1104 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1105 .ctr = crypt_iv_eboiv_ctr,
1106 .generator = crypt_iv_eboiv_gen
1109 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1110 .ctr = crypt_iv_elephant_ctr,
1111 .dtr = crypt_iv_elephant_dtr,
1112 .init = crypt_iv_elephant_init,
1113 .wipe = crypt_iv_elephant_wipe,
1114 .generator = crypt_iv_elephant_gen,
1115 .post = crypt_iv_elephant_post
1119 * Integrity extensions
1121 static bool crypt_integrity_aead(struct crypt_config *cc)
1123 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1126 static bool crypt_integrity_hmac(struct crypt_config *cc)
1128 return crypt_integrity_aead(cc) && cc->key_mac_size;
1131 /* Get sg containing data */
1132 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1133 struct scatterlist *sg)
1135 if (unlikely(crypt_integrity_aead(cc)))
1141 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1143 struct bio_integrity_payload *bip;
1144 unsigned int tag_len;
1147 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1150 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1152 return PTR_ERR(bip);
1154 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1156 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1158 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1159 tag_len, offset_in_page(io->integrity_metadata));
1160 if (unlikely(ret != tag_len))
1166 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1168 #ifdef CONFIG_BLK_DEV_INTEGRITY
1169 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1170 struct mapped_device *md = dm_table_get_md(ti->table);
1172 /* From now we require underlying device with our integrity profile */
1173 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1174 ti->error = "Integrity profile not supported.";
1178 if (bi->tag_size != cc->on_disk_tag_size ||
1179 bi->tuple_size != cc->on_disk_tag_size) {
1180 ti->error = "Integrity profile tag size mismatch.";
1183 if (1 << bi->interval_exp != cc->sector_size) {
1184 ti->error = "Integrity profile sector size mismatch.";
1188 if (crypt_integrity_aead(cc)) {
1189 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1190 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1191 cc->integrity_tag_size, cc->integrity_iv_size);
1193 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1194 ti->error = "Integrity AEAD auth tag size is not supported.";
1197 } else if (cc->integrity_iv_size)
1198 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1199 cc->integrity_iv_size);
1201 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1202 ti->error = "Not enough space for integrity tag in the profile.";
1208 ti->error = "Integrity profile not supported.";
1213 static void crypt_convert_init(struct crypt_config *cc,
1214 struct convert_context *ctx,
1215 struct bio *bio_out, struct bio *bio_in,
1218 ctx->bio_in = bio_in;
1219 ctx->bio_out = bio_out;
1221 ctx->iter_in = bio_in->bi_iter;
1223 ctx->iter_out = bio_out->bi_iter;
1224 ctx->cc_sector = sector + cc->iv_offset;
1225 init_completion(&ctx->restart);
1228 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1231 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1234 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1236 return (void *)((char *)dmreq - cc->dmreq_start);
1239 static u8 *iv_of_dmreq(struct crypt_config *cc,
1240 struct dm_crypt_request *dmreq)
1242 if (crypt_integrity_aead(cc))
1243 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1244 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1246 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1247 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1250 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1251 struct dm_crypt_request *dmreq)
1253 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1256 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1257 struct dm_crypt_request *dmreq)
1259 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1261 return (__le64 *) ptr;
1264 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1265 struct dm_crypt_request *dmreq)
1267 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1268 cc->iv_size + sizeof(uint64_t);
1270 return (unsigned int *)ptr;
1273 static void *tag_from_dmreq(struct crypt_config *cc,
1274 struct dm_crypt_request *dmreq)
1276 struct convert_context *ctx = dmreq->ctx;
1277 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1279 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1280 cc->on_disk_tag_size];
1283 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1284 struct dm_crypt_request *dmreq)
1286 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1289 static int crypt_convert_block_aead(struct crypt_config *cc,
1290 struct convert_context *ctx,
1291 struct aead_request *req,
1292 unsigned int tag_offset)
1294 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1295 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1296 struct dm_crypt_request *dmreq;
1297 u8 *iv, *org_iv, *tag_iv, *tag;
1301 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1303 /* Reject unexpected unaligned bio. */
1304 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1307 dmreq = dmreq_of_req(cc, req);
1308 dmreq->iv_sector = ctx->cc_sector;
1309 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1310 dmreq->iv_sector >>= cc->sector_shift;
1313 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1315 sector = org_sector_of_dmreq(cc, dmreq);
1316 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1318 iv = iv_of_dmreq(cc, dmreq);
1319 org_iv = org_iv_of_dmreq(cc, dmreq);
1320 tag = tag_from_dmreq(cc, dmreq);
1321 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1324 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1325 * | (authenticated) | (auth+encryption) | |
1326 * | sector_LE | IV | sector in/out | tag in/out |
1328 sg_init_table(dmreq->sg_in, 4);
1329 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1330 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1331 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1332 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1334 sg_init_table(dmreq->sg_out, 4);
1335 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1336 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1337 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1338 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1340 if (cc->iv_gen_ops) {
1341 /* For READs use IV stored in integrity metadata */
1342 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1343 memcpy(org_iv, tag_iv, cc->iv_size);
1345 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1348 /* Store generated IV in integrity metadata */
1349 if (cc->integrity_iv_size)
1350 memcpy(tag_iv, org_iv, cc->iv_size);
1352 /* Working copy of IV, to be modified in crypto API */
1353 memcpy(iv, org_iv, cc->iv_size);
1356 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1357 if (bio_data_dir(ctx->bio_in) == WRITE) {
1358 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1359 cc->sector_size, iv);
1360 r = crypto_aead_encrypt(req);
1361 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1362 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1363 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1365 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1366 cc->sector_size + cc->integrity_tag_size, iv);
1367 r = crypto_aead_decrypt(req);
1370 if (r == -EBADMSG) {
1371 sector_t s = le64_to_cpu(*sector);
1373 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1374 ctx->bio_in->bi_bdev, s);
1375 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1379 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1380 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1382 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1383 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1388 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1389 struct convert_context *ctx,
1390 struct skcipher_request *req,
1391 unsigned int tag_offset)
1393 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1394 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1395 struct scatterlist *sg_in, *sg_out;
1396 struct dm_crypt_request *dmreq;
1397 u8 *iv, *org_iv, *tag_iv;
1401 /* Reject unexpected unaligned bio. */
1402 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1405 dmreq = dmreq_of_req(cc, req);
1406 dmreq->iv_sector = ctx->cc_sector;
1407 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1408 dmreq->iv_sector >>= cc->sector_shift;
1411 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1413 iv = iv_of_dmreq(cc, dmreq);
1414 org_iv = org_iv_of_dmreq(cc, dmreq);
1415 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1417 sector = org_sector_of_dmreq(cc, dmreq);
1418 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1420 /* For skcipher we use only the first sg item */
1421 sg_in = &dmreq->sg_in[0];
1422 sg_out = &dmreq->sg_out[0];
1424 sg_init_table(sg_in, 1);
1425 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1427 sg_init_table(sg_out, 1);
1428 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1430 if (cc->iv_gen_ops) {
1431 /* For READs use IV stored in integrity metadata */
1432 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1433 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1435 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1438 /* Data can be already preprocessed in generator */
1439 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1441 /* Store generated IV in integrity metadata */
1442 if (cc->integrity_iv_size)
1443 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1445 /* Working copy of IV, to be modified in crypto API */
1446 memcpy(iv, org_iv, cc->iv_size);
1449 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1451 if (bio_data_dir(ctx->bio_in) == WRITE)
1452 r = crypto_skcipher_encrypt(req);
1454 r = crypto_skcipher_decrypt(req);
1456 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1457 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1459 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1460 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1465 static void kcryptd_async_done(void *async_req, int error);
1467 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1468 struct convert_context *ctx)
1470 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1473 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1478 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1481 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1482 * requests if driver request queue is full.
1484 skcipher_request_set_callback(ctx->r.req,
1485 CRYPTO_TFM_REQ_MAY_BACKLOG,
1486 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1491 static int crypt_alloc_req_aead(struct crypt_config *cc,
1492 struct convert_context *ctx)
1494 if (!ctx->r.req_aead) {
1495 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1496 if (!ctx->r.req_aead)
1500 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1503 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1504 * requests if driver request queue is full.
1506 aead_request_set_callback(ctx->r.req_aead,
1507 CRYPTO_TFM_REQ_MAY_BACKLOG,
1508 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1513 static int crypt_alloc_req(struct crypt_config *cc,
1514 struct convert_context *ctx)
1516 if (crypt_integrity_aead(cc))
1517 return crypt_alloc_req_aead(cc, ctx);
1519 return crypt_alloc_req_skcipher(cc, ctx);
1522 static void crypt_free_req_skcipher(struct crypt_config *cc,
1523 struct skcipher_request *req, struct bio *base_bio)
1525 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1527 if ((struct skcipher_request *)(io + 1) != req)
1528 mempool_free(req, &cc->req_pool);
1531 static void crypt_free_req_aead(struct crypt_config *cc,
1532 struct aead_request *req, struct bio *base_bio)
1534 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1536 if ((struct aead_request *)(io + 1) != req)
1537 mempool_free(req, &cc->req_pool);
1540 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1542 if (crypt_integrity_aead(cc))
1543 crypt_free_req_aead(cc, req, base_bio);
1545 crypt_free_req_skcipher(cc, req, base_bio);
1549 * Encrypt / decrypt data from one bio to another one (can be the same one)
1551 static blk_status_t crypt_convert(struct crypt_config *cc,
1552 struct convert_context *ctx, bool atomic, bool reset_pending)
1554 unsigned int tag_offset = 0;
1555 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1559 * if reset_pending is set we are dealing with the bio for the first time,
1560 * else we're continuing to work on the previous bio, so don't mess with
1561 * the cc_pending counter
1564 atomic_set(&ctx->cc_pending, 1);
1566 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1568 r = crypt_alloc_req(cc, ctx);
1570 complete(&ctx->restart);
1571 return BLK_STS_DEV_RESOURCE;
1574 atomic_inc(&ctx->cc_pending);
1576 if (crypt_integrity_aead(cc))
1577 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1579 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1583 * The request was queued by a crypto driver
1584 * but the driver request queue is full, let's wait.
1587 if (in_interrupt()) {
1588 if (try_wait_for_completion(&ctx->restart)) {
1590 * we don't have to block to wait for completion,
1595 * we can't wait for completion without blocking
1596 * exit and continue processing in a workqueue
1599 ctx->cc_sector += sector_step;
1601 return BLK_STS_DEV_RESOURCE;
1604 wait_for_completion(&ctx->restart);
1606 reinit_completion(&ctx->restart);
1609 * The request is queued and processed asynchronously,
1610 * completion function kcryptd_async_done() will be called.
1614 ctx->cc_sector += sector_step;
1618 * The request was already processed (synchronously).
1621 atomic_dec(&ctx->cc_pending);
1622 ctx->cc_sector += sector_step;
1628 * There was a data integrity error.
1631 atomic_dec(&ctx->cc_pending);
1632 return BLK_STS_PROTECTION;
1634 * There was an error while processing the request.
1637 atomic_dec(&ctx->cc_pending);
1638 return BLK_STS_IOERR;
1645 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1648 * Generate a new unfragmented bio with the given size
1649 * This should never violate the device limitations (but only because
1650 * max_segment_size is being constrained to PAGE_SIZE).
1652 * This function may be called concurrently. If we allocate from the mempool
1653 * concurrently, there is a possibility of deadlock. For example, if we have
1654 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1655 * the mempool concurrently, it may deadlock in a situation where both processes
1656 * have allocated 128 pages and the mempool is exhausted.
1658 * In order to avoid this scenario we allocate the pages under a mutex.
1660 * In order to not degrade performance with excessive locking, we try
1661 * non-blocking allocations without a mutex first but on failure we fallback
1662 * to blocking allocations with a mutex.
1664 * In order to reduce allocation overhead, we try to allocate compound pages in
1665 * the first pass. If they are not available, we fall back to the mempool.
1667 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1669 struct crypt_config *cc = io->cc;
1671 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1672 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1673 unsigned int remaining_size;
1674 unsigned int order = MAX_PAGE_ORDER;
1677 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1678 mutex_lock(&cc->bio_alloc_lock);
1680 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1682 clone->bi_private = io;
1683 clone->bi_end_io = crypt_endio;
1685 remaining_size = size;
1687 while (remaining_size) {
1689 unsigned size_to_add;
1690 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1691 order = min(order, remaining_order);
1694 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1695 (1 << order) > dm_crypt_pages_per_client))
1696 goto decrease_order;
1697 pages = alloc_pages(gfp_mask
1698 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1700 if (likely(pages != NULL)) {
1701 percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1708 pages = mempool_alloc(&cc->page_pool, gfp_mask);
1710 crypt_free_buffer_pages(cc, clone);
1712 gfp_mask |= __GFP_DIRECT_RECLAIM;
1718 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1719 __bio_add_page(clone, pages, size_to_add, 0);
1720 remaining_size -= size_to_add;
1723 /* Allocate space for integrity tags */
1724 if (dm_crypt_integrity_io_alloc(io, clone)) {
1725 crypt_free_buffer_pages(cc, clone);
1730 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1731 mutex_unlock(&cc->bio_alloc_lock);
1736 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1738 struct folio_iter fi;
1740 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1741 bio_for_each_folio_all(fi, clone) {
1742 if (folio_test_large(fi.folio)) {
1743 percpu_counter_sub(&cc->n_allocated_pages,
1744 1 << folio_order(fi.folio));
1745 folio_put(fi.folio);
1747 mempool_free(&fi.folio->page, &cc->page_pool);
1753 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1754 struct bio *bio, sector_t sector)
1758 io->sector = sector;
1760 io->ctx.r.req = NULL;
1761 io->integrity_metadata = NULL;
1762 io->integrity_metadata_from_pool = 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);
1772 * One of the bios was finished. Check for completion of
1773 * the whole request and correctly clean up the buffer.
1775 static void crypt_dec_pending(struct dm_crypt_io *io)
1777 struct crypt_config *cc = io->cc;
1778 struct bio *base_bio = io->base_bio;
1779 blk_status_t error = io->error;
1781 if (!atomic_dec_and_test(&io->io_pending))
1785 crypt_free_req(cc, io->ctx.r.req, base_bio);
1787 if (unlikely(io->integrity_metadata_from_pool))
1788 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1790 kfree(io->integrity_metadata);
1792 base_bio->bi_status = error;
1794 bio_endio(base_bio);
1798 * kcryptd/kcryptd_io:
1800 * Needed because it would be very unwise to do decryption in an
1801 * interrupt context.
1803 * kcryptd performs the actual encryption or decryption.
1805 * kcryptd_io performs the IO submission.
1807 * They must be separated as otherwise the final stages could be
1808 * starved by new requests which can block in the first stages due
1809 * to memory allocation.
1811 * The work is done per CPU global for all dm-crypt instances.
1812 * They should not depend on each other and do not block.
1814 static void crypt_endio(struct bio *clone)
1816 struct dm_crypt_io *io = clone->bi_private;
1817 struct crypt_config *cc = io->cc;
1818 unsigned int rw = bio_data_dir(clone);
1822 * free the processed pages
1825 crypt_free_buffer_pages(cc, clone);
1827 error = clone->bi_status;
1830 if (rw == READ && !error) {
1831 kcryptd_queue_crypt(io);
1835 if (unlikely(error))
1838 crypt_dec_pending(io);
1841 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1843 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1845 struct crypt_config *cc = io->cc;
1849 * We need the original biovec array in order to decrypt the whole bio
1850 * data *afterwards* -- thanks to immutable biovecs we don't need to
1851 * worry about the block layer modifying the biovec array; so leverage
1852 * bio_alloc_clone().
1854 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1857 clone->bi_private = io;
1858 clone->bi_end_io = crypt_endio;
1860 crypt_inc_pending(io);
1862 clone->bi_iter.bi_sector = cc->start + io->sector;
1864 if (dm_crypt_integrity_io_alloc(io, clone)) {
1865 crypt_dec_pending(io);
1870 dm_submit_bio_remap(io->base_bio, clone);
1874 static void kcryptd_io_read_work(struct work_struct *work)
1876 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1878 crypt_inc_pending(io);
1879 if (kcryptd_io_read(io, GFP_NOIO))
1880 io->error = BLK_STS_RESOURCE;
1881 crypt_dec_pending(io);
1884 static void kcryptd_queue_read(struct dm_crypt_io *io)
1886 struct crypt_config *cc = io->cc;
1888 INIT_WORK(&io->work, kcryptd_io_read_work);
1889 queue_work(cc->io_queue, &io->work);
1892 static void kcryptd_io_write(struct dm_crypt_io *io)
1894 struct bio *clone = io->ctx.bio_out;
1896 dm_submit_bio_remap(io->base_bio, clone);
1899 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1901 static int dmcrypt_write(void *data)
1903 struct crypt_config *cc = data;
1904 struct dm_crypt_io *io;
1907 struct rb_root write_tree;
1908 struct blk_plug plug;
1910 spin_lock_irq(&cc->write_thread_lock);
1913 if (!RB_EMPTY_ROOT(&cc->write_tree))
1916 set_current_state(TASK_INTERRUPTIBLE);
1918 spin_unlock_irq(&cc->write_thread_lock);
1920 if (unlikely(kthread_should_stop())) {
1921 set_current_state(TASK_RUNNING);
1927 set_current_state(TASK_RUNNING);
1928 spin_lock_irq(&cc->write_thread_lock);
1929 goto continue_locked;
1932 write_tree = cc->write_tree;
1933 cc->write_tree = RB_ROOT;
1934 spin_unlock_irq(&cc->write_thread_lock);
1936 BUG_ON(rb_parent(write_tree.rb_node));
1939 * Note: we cannot walk the tree here with rb_next because
1940 * the structures may be freed when kcryptd_io_write is called.
1942 blk_start_plug(&plug);
1944 io = crypt_io_from_node(rb_first(&write_tree));
1945 rb_erase(&io->rb_node, &write_tree);
1946 kcryptd_io_write(io);
1948 } while (!RB_EMPTY_ROOT(&write_tree));
1949 blk_finish_plug(&plug);
1954 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1956 struct bio *clone = io->ctx.bio_out;
1957 struct crypt_config *cc = io->cc;
1958 unsigned long flags;
1960 struct rb_node **rbp, *parent;
1962 if (unlikely(io->error)) {
1963 crypt_free_buffer_pages(cc, clone);
1965 crypt_dec_pending(io);
1969 /* crypt_convert should have filled the clone bio */
1970 BUG_ON(io->ctx.iter_out.bi_size);
1972 clone->bi_iter.bi_sector = cc->start + io->sector;
1974 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1975 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1976 dm_submit_bio_remap(io->base_bio, clone);
1980 spin_lock_irqsave(&cc->write_thread_lock, flags);
1981 if (RB_EMPTY_ROOT(&cc->write_tree))
1982 wake_up_process(cc->write_thread);
1983 rbp = &cc->write_tree.rb_node;
1985 sector = io->sector;
1988 if (sector < crypt_io_from_node(parent)->sector)
1989 rbp = &(*rbp)->rb_left;
1991 rbp = &(*rbp)->rb_right;
1993 rb_link_node(&io->rb_node, parent, rbp);
1994 rb_insert_color(&io->rb_node, &cc->write_tree);
1995 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1998 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1999 struct convert_context *ctx)
2002 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2006 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2007 * constraints so they do not need to be issued inline by
2008 * kcryptd_crypt_write_convert().
2010 switch (bio_op(ctx->bio_in)) {
2012 case REQ_OP_WRITE_ZEROES:
2019 static void kcryptd_crypt_write_continue(struct work_struct *work)
2021 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2022 struct crypt_config *cc = io->cc;
2023 struct convert_context *ctx = &io->ctx;
2025 sector_t sector = io->sector;
2028 wait_for_completion(&ctx->restart);
2029 reinit_completion(&ctx->restart);
2031 r = crypt_convert(cc, &io->ctx, true, false);
2034 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2035 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2036 /* Wait for completion signaled by kcryptd_async_done() */
2037 wait_for_completion(&ctx->restart);
2041 /* Encryption was already finished, submit io now */
2042 if (crypt_finished) {
2043 kcryptd_crypt_write_io_submit(io, 0);
2044 io->sector = sector;
2047 crypt_dec_pending(io);
2050 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2052 struct crypt_config *cc = io->cc;
2053 struct convert_context *ctx = &io->ctx;
2056 sector_t sector = io->sector;
2060 * Prevent io from disappearing until this function completes.
2062 crypt_inc_pending(io);
2063 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2065 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2066 if (unlikely(!clone)) {
2067 io->error = BLK_STS_IOERR;
2071 io->ctx.bio_out = clone;
2072 io->ctx.iter_out = clone->bi_iter;
2074 sector += bio_sectors(clone);
2076 crypt_inc_pending(io);
2077 r = crypt_convert(cc, ctx,
2078 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2080 * Crypto API backlogged the request, because its queue was full
2081 * and we're in softirq context, so continue from a workqueue
2082 * (TODO: is it actually possible to be in softirq in the write path?)
2084 if (r == BLK_STS_DEV_RESOURCE) {
2085 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2086 queue_work(cc->crypt_queue, &io->work);
2091 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2092 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2093 /* Wait for completion signaled by kcryptd_async_done() */
2094 wait_for_completion(&ctx->restart);
2098 /* Encryption was already finished, submit io now */
2099 if (crypt_finished) {
2100 kcryptd_crypt_write_io_submit(io, 0);
2101 io->sector = sector;
2105 crypt_dec_pending(io);
2108 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2110 crypt_dec_pending(io);
2113 static void kcryptd_crypt_read_continue(struct work_struct *work)
2115 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2116 struct crypt_config *cc = io->cc;
2119 wait_for_completion(&io->ctx.restart);
2120 reinit_completion(&io->ctx.restart);
2122 r = crypt_convert(cc, &io->ctx, true, false);
2126 if (atomic_dec_and_test(&io->ctx.cc_pending))
2127 kcryptd_crypt_read_done(io);
2129 crypt_dec_pending(io);
2132 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2134 struct crypt_config *cc = io->cc;
2137 crypt_inc_pending(io);
2139 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2142 r = crypt_convert(cc, &io->ctx,
2143 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2145 * Crypto API backlogged the request, because its queue was full
2146 * and we're in softirq context, so continue from a workqueue
2148 if (r == BLK_STS_DEV_RESOURCE) {
2149 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2150 queue_work(cc->crypt_queue, &io->work);
2156 if (atomic_dec_and_test(&io->ctx.cc_pending))
2157 kcryptd_crypt_read_done(io);
2159 crypt_dec_pending(io);
2162 static void kcryptd_async_done(void *data, int error)
2164 struct dm_crypt_request *dmreq = data;
2165 struct convert_context *ctx = dmreq->ctx;
2166 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2167 struct crypt_config *cc = io->cc;
2170 * A request from crypto driver backlog is going to be processed now,
2171 * finish the completion and continue in crypt_convert().
2172 * (Callback will be called for the second time for this request.)
2174 if (error == -EINPROGRESS) {
2175 complete(&ctx->restart);
2179 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2180 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2182 if (error == -EBADMSG) {
2183 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2185 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2186 ctx->bio_in->bi_bdev, s);
2187 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2189 io->error = BLK_STS_PROTECTION;
2190 } else if (error < 0)
2191 io->error = BLK_STS_IOERR;
2193 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2195 if (!atomic_dec_and_test(&ctx->cc_pending))
2199 * The request is fully completed: for inline writes, let
2200 * kcryptd_crypt_write_convert() do the IO submission.
2202 if (bio_data_dir(io->base_bio) == READ) {
2203 kcryptd_crypt_read_done(io);
2207 if (kcryptd_crypt_write_inline(cc, ctx)) {
2208 complete(&ctx->restart);
2212 kcryptd_crypt_write_io_submit(io, 1);
2215 static void kcryptd_crypt(struct work_struct *work)
2217 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2219 if (bio_data_dir(io->base_bio) == READ)
2220 kcryptd_crypt_read_convert(io);
2222 kcryptd_crypt_write_convert(io);
2225 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2227 struct crypt_config *cc = io->cc;
2229 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2230 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2232 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2233 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2234 * it is being executed with irqs disabled.
2236 if (!(in_hardirq() || irqs_disabled())) {
2237 kcryptd_crypt(&io->work);
2242 INIT_WORK(&io->work, kcryptd_crypt);
2243 queue_work(cc->crypt_queue, &io->work);
2246 static void crypt_free_tfms_aead(struct crypt_config *cc)
2248 if (!cc->cipher_tfm.tfms_aead)
2251 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2252 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2253 cc->cipher_tfm.tfms_aead[0] = NULL;
2256 kfree(cc->cipher_tfm.tfms_aead);
2257 cc->cipher_tfm.tfms_aead = NULL;
2260 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2264 if (!cc->cipher_tfm.tfms)
2267 for (i = 0; i < cc->tfms_count; i++)
2268 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2269 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2270 cc->cipher_tfm.tfms[i] = NULL;
2273 kfree(cc->cipher_tfm.tfms);
2274 cc->cipher_tfm.tfms = NULL;
2277 static void crypt_free_tfms(struct crypt_config *cc)
2279 if (crypt_integrity_aead(cc))
2280 crypt_free_tfms_aead(cc);
2282 crypt_free_tfms_skcipher(cc);
2285 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2290 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2291 sizeof(struct crypto_skcipher *),
2293 if (!cc->cipher_tfm.tfms)
2296 for (i = 0; i < cc->tfms_count; i++) {
2297 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2298 CRYPTO_ALG_ALLOCATES_MEMORY);
2299 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2300 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2301 crypt_free_tfms(cc);
2307 * dm-crypt performance can vary greatly depending on which crypto
2308 * algorithm implementation is used. Help people debug performance
2309 * problems by logging the ->cra_driver_name.
2311 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2312 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2316 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2320 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2321 if (!cc->cipher_tfm.tfms)
2324 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2325 CRYPTO_ALG_ALLOCATES_MEMORY);
2326 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2327 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2328 crypt_free_tfms(cc);
2332 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2333 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2337 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2339 if (crypt_integrity_aead(cc))
2340 return crypt_alloc_tfms_aead(cc, ciphermode);
2342 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2345 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2347 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2350 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2352 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2356 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2357 * the key must be for some reason in special format.
2358 * This funcion converts cc->key to this special format.
2360 static void crypt_copy_authenckey(char *p, const void *key,
2361 unsigned int enckeylen, unsigned int authkeylen)
2363 struct crypto_authenc_key_param *param;
2366 rta = (struct rtattr *)p;
2367 param = RTA_DATA(rta);
2368 param->enckeylen = cpu_to_be32(enckeylen);
2369 rta->rta_len = RTA_LENGTH(sizeof(*param));
2370 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2371 p += RTA_SPACE(sizeof(*param));
2372 memcpy(p, key + enckeylen, authkeylen);
2374 memcpy(p, key, enckeylen);
2377 static int crypt_setkey(struct crypt_config *cc)
2379 unsigned int subkey_size;
2382 /* Ignore extra keys (which are used for IV etc) */
2383 subkey_size = crypt_subkey_size(cc);
2385 if (crypt_integrity_hmac(cc)) {
2386 if (subkey_size < cc->key_mac_size)
2389 crypt_copy_authenckey(cc->authenc_key, cc->key,
2390 subkey_size - cc->key_mac_size,
2394 for (i = 0; i < cc->tfms_count; i++) {
2395 if (crypt_integrity_hmac(cc))
2396 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2397 cc->authenc_key, crypt_authenckey_size(cc));
2398 else if (crypt_integrity_aead(cc))
2399 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2400 cc->key + (i * subkey_size),
2403 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2404 cc->key + (i * subkey_size),
2410 if (crypt_integrity_hmac(cc))
2411 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2418 static bool contains_whitespace(const char *str)
2421 if (isspace(*str++))
2426 static int set_key_user(struct crypt_config *cc, struct key *key)
2428 const struct user_key_payload *ukp;
2430 ukp = user_key_payload_locked(key);
2432 return -EKEYREVOKED;
2434 if (cc->key_size != ukp->datalen)
2437 memcpy(cc->key, ukp->data, cc->key_size);
2442 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2444 const struct encrypted_key_payload *ekp;
2446 ekp = key->payload.data[0];
2448 return -EKEYREVOKED;
2450 if (cc->key_size != ekp->decrypted_datalen)
2453 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2458 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2460 const struct trusted_key_payload *tkp;
2462 tkp = key->payload.data[0];
2464 return -EKEYREVOKED;
2466 if (cc->key_size != tkp->key_len)
2469 memcpy(cc->key, tkp->key, cc->key_size);
2474 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2476 char *new_key_string, *key_desc;
2478 struct key_type *type;
2480 int (*set_key)(struct crypt_config *cc, struct key *key);
2483 * Reject key_string with whitespace. dm core currently lacks code for
2484 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2486 if (contains_whitespace(key_string)) {
2487 DMERR("whitespace chars not allowed in key string");
2491 /* look for next ':' separating key_type from key_description */
2492 key_desc = strchr(key_string, ':');
2493 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2496 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2497 type = &key_type_logon;
2498 set_key = set_key_user;
2499 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2500 type = &key_type_user;
2501 set_key = set_key_user;
2502 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2503 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2504 type = &key_type_encrypted;
2505 set_key = set_key_encrypted;
2506 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2507 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2508 type = &key_type_trusted;
2509 set_key = set_key_trusted;
2514 new_key_string = kstrdup(key_string, GFP_KERNEL);
2515 if (!new_key_string)
2518 key = request_key(type, key_desc + 1, NULL);
2520 kfree_sensitive(new_key_string);
2521 return PTR_ERR(key);
2524 down_read(&key->sem);
2526 ret = set_key(cc, key);
2530 kfree_sensitive(new_key_string);
2537 /* clear the flag since following operations may invalidate previously valid key */
2538 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2540 ret = crypt_setkey(cc);
2543 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2544 kfree_sensitive(cc->key_string);
2545 cc->key_string = new_key_string;
2547 kfree_sensitive(new_key_string);
2552 static int get_key_size(char **key_string)
2557 if (*key_string[0] != ':')
2558 return strlen(*key_string) >> 1;
2560 /* look for next ':' in key string */
2561 colon = strpbrk(*key_string + 1, ":");
2565 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2568 *key_string = colon;
2570 /* remaining key string should be :<logon|user>:<key_desc> */
2577 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2582 static int get_key_size(char **key_string)
2584 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2587 #endif /* CONFIG_KEYS */
2589 static int crypt_set_key(struct crypt_config *cc, char *key)
2592 int key_string_len = strlen(key);
2594 /* Hyphen (which gives a key_size of zero) means there is no key. */
2595 if (!cc->key_size && strcmp(key, "-"))
2598 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2599 if (key[0] == ':') {
2600 r = crypt_set_keyring_key(cc, key + 1);
2604 /* clear the flag since following operations may invalidate previously valid key */
2605 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2607 /* wipe references to any kernel keyring key */
2608 kfree_sensitive(cc->key_string);
2609 cc->key_string = NULL;
2611 /* Decode key from its hex representation. */
2612 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2615 r = crypt_setkey(cc);
2617 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2620 /* Hex key string not needed after here, so wipe it. */
2621 memset(key, '0', key_string_len);
2626 static int crypt_wipe_key(struct crypt_config *cc)
2630 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2631 get_random_bytes(&cc->key, cc->key_size);
2633 /* Wipe IV private keys */
2634 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2635 r = cc->iv_gen_ops->wipe(cc);
2640 kfree_sensitive(cc->key_string);
2641 cc->key_string = NULL;
2642 r = crypt_setkey(cc);
2643 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2648 static void crypt_calculate_pages_per_client(void)
2650 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2652 if (!dm_crypt_clients_n)
2655 pages /= dm_crypt_clients_n;
2656 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2657 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2658 dm_crypt_pages_per_client = pages;
2661 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2663 struct crypt_config *cc = pool_data;
2667 * Note, percpu_counter_read_positive() may over (and under) estimate
2668 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2669 * but avoids potential spinlock contention of an exact result.
2671 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2672 likely(gfp_mask & __GFP_NORETRY))
2675 page = alloc_page(gfp_mask);
2676 if (likely(page != NULL))
2677 percpu_counter_add(&cc->n_allocated_pages, 1);
2682 static void crypt_page_free(void *page, void *pool_data)
2684 struct crypt_config *cc = pool_data;
2687 percpu_counter_sub(&cc->n_allocated_pages, 1);
2690 static void crypt_dtr(struct dm_target *ti)
2692 struct crypt_config *cc = ti->private;
2699 if (cc->write_thread)
2700 kthread_stop(cc->write_thread);
2703 destroy_workqueue(cc->io_queue);
2704 if (cc->crypt_queue)
2705 destroy_workqueue(cc->crypt_queue);
2707 crypt_free_tfms(cc);
2709 bioset_exit(&cc->bs);
2711 mempool_exit(&cc->page_pool);
2712 mempool_exit(&cc->req_pool);
2713 mempool_exit(&cc->tag_pool);
2715 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2716 percpu_counter_destroy(&cc->n_allocated_pages);
2718 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2719 cc->iv_gen_ops->dtr(cc);
2722 dm_put_device(ti, cc->dev);
2724 kfree_sensitive(cc->cipher_string);
2725 kfree_sensitive(cc->key_string);
2726 kfree_sensitive(cc->cipher_auth);
2727 kfree_sensitive(cc->authenc_key);
2729 mutex_destroy(&cc->bio_alloc_lock);
2731 /* Must zero key material before freeing */
2732 kfree_sensitive(cc);
2734 spin_lock(&dm_crypt_clients_lock);
2735 WARN_ON(!dm_crypt_clients_n);
2736 dm_crypt_clients_n--;
2737 crypt_calculate_pages_per_client();
2738 spin_unlock(&dm_crypt_clients_lock);
2740 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2743 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2745 struct crypt_config *cc = ti->private;
2747 if (crypt_integrity_aead(cc))
2748 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2750 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2753 /* at least a 64 bit sector number should fit in our buffer */
2754 cc->iv_size = max(cc->iv_size,
2755 (unsigned int)(sizeof(u64) / sizeof(u8)));
2757 DMWARN("Selected cipher does not support IVs");
2761 /* Choose ivmode, see comments at iv code. */
2763 cc->iv_gen_ops = NULL;
2764 else if (strcmp(ivmode, "plain") == 0)
2765 cc->iv_gen_ops = &crypt_iv_plain_ops;
2766 else if (strcmp(ivmode, "plain64") == 0)
2767 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2768 else if (strcmp(ivmode, "plain64be") == 0)
2769 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2770 else if (strcmp(ivmode, "essiv") == 0)
2771 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2772 else if (strcmp(ivmode, "benbi") == 0)
2773 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2774 else if (strcmp(ivmode, "null") == 0)
2775 cc->iv_gen_ops = &crypt_iv_null_ops;
2776 else if (strcmp(ivmode, "eboiv") == 0)
2777 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2778 else if (strcmp(ivmode, "elephant") == 0) {
2779 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2781 cc->key_extra_size = cc->key_size / 2;
2782 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2784 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2785 } else if (strcmp(ivmode, "lmk") == 0) {
2786 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2788 * Version 2 and 3 is recognised according
2789 * to length of provided multi-key string.
2790 * If present (version 3), last key is used as IV seed.
2791 * All keys (including IV seed) are always the same size.
2793 if (cc->key_size % cc->key_parts) {
2795 cc->key_extra_size = cc->key_size / cc->key_parts;
2797 } else if (strcmp(ivmode, "tcw") == 0) {
2798 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2799 cc->key_parts += 2; /* IV + whitening */
2800 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2801 } else if (strcmp(ivmode, "random") == 0) {
2802 cc->iv_gen_ops = &crypt_iv_random_ops;
2803 /* Need storage space in integrity fields. */
2804 cc->integrity_iv_size = cc->iv_size;
2806 ti->error = "Invalid IV mode";
2814 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2815 * The HMAC is needed to calculate tag size (HMAC digest size).
2816 * This should be probably done by crypto-api calls (once available...)
2818 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2820 char *start, *end, *mac_alg = NULL;
2821 struct crypto_ahash *mac;
2823 if (!strstarts(cipher_api, "authenc("))
2826 start = strchr(cipher_api, '(');
2827 end = strchr(cipher_api, ',');
2828 if (!start || !end || ++start > end)
2831 mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2835 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2839 return PTR_ERR(mac);
2841 cc->key_mac_size = crypto_ahash_digestsize(mac);
2842 crypto_free_ahash(mac);
2844 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2845 if (!cc->authenc_key)
2851 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2852 char **ivmode, char **ivopts)
2854 struct crypt_config *cc = ti->private;
2855 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2861 * New format (capi: prefix)
2862 * capi:cipher_api_spec-iv:ivopts
2864 tmp = &cipher_in[strlen("capi:")];
2866 /* Separate IV options if present, it can contain another '-' in hash name */
2867 *ivopts = strrchr(tmp, ':');
2873 *ivmode = strrchr(tmp, '-');
2878 /* The rest is crypto API spec */
2881 /* Alloc AEAD, can be used only in new format. */
2882 if (crypt_integrity_aead(cc)) {
2883 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2885 ti->error = "Invalid AEAD cipher spec";
2890 if (*ivmode && !strcmp(*ivmode, "lmk"))
2891 cc->tfms_count = 64;
2893 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2895 ti->error = "Digest algorithm missing for ESSIV mode";
2898 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2899 cipher_api, *ivopts);
2900 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2901 ti->error = "Cannot allocate cipher string";
2907 cc->key_parts = cc->tfms_count;
2909 /* Allocate cipher */
2910 ret = crypt_alloc_tfms(cc, cipher_api);
2912 ti->error = "Error allocating crypto tfm";
2916 if (crypt_integrity_aead(cc))
2917 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2919 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2924 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2925 char **ivmode, char **ivopts)
2927 struct crypt_config *cc = ti->private;
2928 char *tmp, *cipher, *chainmode, *keycount;
2929 char *cipher_api = NULL;
2933 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2934 ti->error = "Bad cipher specification";
2939 * Legacy dm-crypt cipher specification
2940 * cipher[:keycount]-mode-iv:ivopts
2943 keycount = strsep(&tmp, "-");
2944 cipher = strsep(&keycount, ":");
2948 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2949 !is_power_of_2(cc->tfms_count)) {
2950 ti->error = "Bad cipher key count specification";
2953 cc->key_parts = cc->tfms_count;
2955 chainmode = strsep(&tmp, "-");
2956 *ivmode = strsep(&tmp, ":");
2960 * For compatibility with the original dm-crypt mapping format, if
2961 * only the cipher name is supplied, use cbc-plain.
2963 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2968 if (strcmp(chainmode, "ecb") && !*ivmode) {
2969 ti->error = "IV mechanism required";
2973 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2977 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2979 ti->error = "Digest algorithm missing for ESSIV mode";
2983 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2984 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2986 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2987 "%s(%s)", chainmode, cipher);
2989 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2994 /* Allocate cipher */
2995 ret = crypt_alloc_tfms(cc, cipher_api);
2997 ti->error = "Error allocating crypto tfm";
3005 ti->error = "Cannot allocate cipher strings";
3009 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3011 struct crypt_config *cc = ti->private;
3012 char *ivmode = NULL, *ivopts = NULL;
3015 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3016 if (!cc->cipher_string) {
3017 ti->error = "Cannot allocate cipher strings";
3021 if (strstarts(cipher_in, "capi:"))
3022 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3024 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3029 ret = crypt_ctr_ivmode(ti, ivmode);
3033 /* Initialize and set key */
3034 ret = crypt_set_key(cc, key);
3036 ti->error = "Error decoding and setting key";
3041 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3042 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3044 ti->error = "Error creating IV";
3049 /* Initialize IV (set keys for ESSIV etc) */
3050 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3051 ret = cc->iv_gen_ops->init(cc);
3053 ti->error = "Error initialising IV";
3058 /* wipe the kernel key payload copy */
3060 memset(cc->key, 0, cc->key_size * sizeof(u8));
3065 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3067 struct crypt_config *cc = ti->private;
3068 struct dm_arg_set as;
3069 static const struct dm_arg _args[] = {
3070 {0, 8, "Invalid number of feature args"},
3072 unsigned int opt_params, val;
3073 const char *opt_string, *sval;
3077 /* Optional parameters */
3081 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3085 while (opt_params--) {
3086 opt_string = dm_shift_arg(&as);
3088 ti->error = "Not enough feature arguments";
3092 if (!strcasecmp(opt_string, "allow_discards"))
3093 ti->num_discard_bios = 1;
3095 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3096 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3098 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3099 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3100 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3101 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3102 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3103 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3104 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3105 if (val == 0 || val > MAX_TAG_SIZE) {
3106 ti->error = "Invalid integrity arguments";
3109 cc->on_disk_tag_size = val;
3110 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3111 if (!strcasecmp(sval, "aead")) {
3112 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3113 } else if (strcasecmp(sval, "none")) {
3114 ti->error = "Unknown integrity profile";
3118 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3119 if (!cc->cipher_auth)
3121 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3122 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3123 cc->sector_size > 4096 ||
3124 (cc->sector_size & (cc->sector_size - 1))) {
3125 ti->error = "Invalid feature value for sector_size";
3128 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3129 ti->error = "Device size is not multiple of sector_size feature";
3132 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3133 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3134 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3136 ti->error = "Invalid feature arguments";
3144 #ifdef CONFIG_BLK_DEV_ZONED
3145 static int crypt_report_zones(struct dm_target *ti,
3146 struct dm_report_zones_args *args, unsigned int nr_zones)
3148 struct crypt_config *cc = ti->private;
3150 return dm_report_zones(cc->dev->bdev, cc->start,
3151 cc->start + dm_target_offset(ti, args->next_sector),
3155 #define crypt_report_zones NULL
3159 * Construct an encryption mapping:
3160 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3162 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3164 struct crypt_config *cc;
3165 const char *devname = dm_table_device_name(ti->table);
3167 unsigned int align_mask;
3168 unsigned long long tmpll;
3170 size_t iv_size_padding, additional_req_size;
3174 ti->error = "Not enough arguments";
3178 key_size = get_key_size(&argv[1]);
3180 ti->error = "Cannot parse key size";
3184 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3186 ti->error = "Cannot allocate encryption context";
3189 cc->key_size = key_size;
3190 cc->sector_size = (1 << SECTOR_SHIFT);
3191 cc->sector_shift = 0;
3195 spin_lock(&dm_crypt_clients_lock);
3196 dm_crypt_clients_n++;
3197 crypt_calculate_pages_per_client();
3198 spin_unlock(&dm_crypt_clients_lock);
3200 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3204 /* Optional parameters need to be read before cipher constructor */
3206 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3211 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3215 if (crypt_integrity_aead(cc)) {
3216 cc->dmreq_start = sizeof(struct aead_request);
3217 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3218 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3220 cc->dmreq_start = sizeof(struct skcipher_request);
3221 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3222 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3224 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3226 if (align_mask < CRYPTO_MINALIGN) {
3227 /* Allocate the padding exactly */
3228 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3232 * If the cipher requires greater alignment than kmalloc
3233 * alignment, we don't know the exact position of the
3234 * initialization vector. We must assume worst case.
3236 iv_size_padding = align_mask;
3239 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3240 additional_req_size = sizeof(struct dm_crypt_request) +
3241 iv_size_padding + cc->iv_size +
3244 sizeof(unsigned int);
3246 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3248 ti->error = "Cannot allocate crypt request mempool";
3252 cc->per_bio_data_size = ti->per_io_data_size =
3253 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3256 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3258 ti->error = "Cannot allocate page mempool";
3262 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3264 ti->error = "Cannot allocate crypt bioset";
3268 mutex_init(&cc->bio_alloc_lock);
3271 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3272 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3273 ti->error = "Invalid iv_offset sector";
3276 cc->iv_offset = tmpll;
3278 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3280 ti->error = "Device lookup failed";
3285 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3286 ti->error = "Invalid device sector";
3291 if (bdev_is_zoned(cc->dev->bdev)) {
3293 * For zoned block devices, we need to preserve the issuer write
3294 * ordering. To do so, disable write workqueues and force inline
3295 * encryption completion.
3297 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3298 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3301 * All zone append writes to a zone of a zoned block device will
3302 * have the same BIO sector, the start of the zone. When the
3303 * cypher IV mode uses sector values, all data targeting a
3304 * zone will be encrypted using the first sector numbers of the
3305 * zone. This will not result in write errors but will
3306 * cause most reads to fail as reads will use the sector values
3307 * for the actual data locations, resulting in IV mismatch.
3308 * To avoid this problem, ask DM core to emulate zone append
3309 * operations with regular writes.
3311 DMDEBUG("Zone append operations will be emulated");
3312 ti->emulate_zone_append = true;
3315 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3316 ret = crypt_integrity_ctr(cc, ti);
3320 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3321 if (!cc->tag_pool_max_sectors)
3322 cc->tag_pool_max_sectors = 1;
3324 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3325 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3327 ti->error = "Cannot allocate integrity tags mempool";
3331 cc->tag_pool_max_sectors <<= cc->sector_shift;
3335 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3336 if (!cc->io_queue) {
3337 ti->error = "Couldn't create kcryptd io queue";
3341 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3342 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3345 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3346 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3347 num_online_cpus(), devname);
3348 if (!cc->crypt_queue) {
3349 ti->error = "Couldn't create kcryptd queue";
3353 spin_lock_init(&cc->write_thread_lock);
3354 cc->write_tree = RB_ROOT;
3356 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3357 if (IS_ERR(cc->write_thread)) {
3358 ret = PTR_ERR(cc->write_thread);
3359 cc->write_thread = NULL;
3360 ti->error = "Couldn't spawn write thread";
3364 ti->num_flush_bios = 1;
3365 ti->limit_swap_bios = true;
3366 ti->accounts_remapped_io = true;
3368 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3372 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3377 static int crypt_map(struct dm_target *ti, struct bio *bio)
3379 struct dm_crypt_io *io;
3380 struct crypt_config *cc = ti->private;
3383 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3384 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3385 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3387 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3388 bio_op(bio) == REQ_OP_DISCARD)) {
3389 bio_set_dev(bio, cc->dev->bdev);
3390 if (bio_sectors(bio))
3391 bio->bi_iter.bi_sector = cc->start +
3392 dm_target_offset(ti, bio->bi_iter.bi_sector);
3393 return DM_MAPIO_REMAPPED;
3397 * Check if bio is too large, split as needed.
3399 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3400 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3401 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3404 * Ensure that bio is a multiple of internal sector encryption size
3405 * and is aligned to this size as defined in IO hints.
3407 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3408 return DM_MAPIO_KILL;
3410 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3411 return DM_MAPIO_KILL;
3413 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3414 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3416 if (cc->on_disk_tag_size) {
3417 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3419 if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3420 io->integrity_metadata = NULL;
3422 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3424 if (unlikely(!io->integrity_metadata)) {
3425 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3426 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3427 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3428 io->integrity_metadata_from_pool = true;
3432 if (crypt_integrity_aead(cc))
3433 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3435 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3437 if (bio_data_dir(io->base_bio) == READ) {
3438 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3439 kcryptd_queue_read(io);
3441 kcryptd_queue_crypt(io);
3443 return DM_MAPIO_SUBMITTED;
3446 static char hex2asc(unsigned char c)
3448 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3451 static void crypt_status(struct dm_target *ti, status_type_t type,
3452 unsigned int status_flags, char *result, unsigned int maxlen)
3454 struct crypt_config *cc = ti->private;
3455 unsigned int i, sz = 0;
3456 int num_feature_args = 0;
3459 case STATUSTYPE_INFO:
3463 case STATUSTYPE_TABLE:
3464 DMEMIT("%s ", cc->cipher_string);
3466 if (cc->key_size > 0) {
3468 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3470 for (i = 0; i < cc->key_size; i++) {
3471 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3472 hex2asc(cc->key[i] & 0xf));
3478 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3479 cc->dev->name, (unsigned long long)cc->start);
3481 num_feature_args += !!ti->num_discard_bios;
3482 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3483 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3484 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3485 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3486 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3487 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3488 if (cc->on_disk_tag_size)
3490 if (num_feature_args) {
3491 DMEMIT(" %d", num_feature_args);
3492 if (ti->num_discard_bios)
3493 DMEMIT(" allow_discards");
3494 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3495 DMEMIT(" same_cpu_crypt");
3496 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3497 DMEMIT(" submit_from_crypt_cpus");
3498 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3499 DMEMIT(" no_read_workqueue");
3500 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3501 DMEMIT(" no_write_workqueue");
3502 if (cc->on_disk_tag_size)
3503 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3504 if (cc->sector_size != (1 << SECTOR_SHIFT))
3505 DMEMIT(" sector_size:%d", cc->sector_size);
3506 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3507 DMEMIT(" iv_large_sectors");
3511 case STATUSTYPE_IMA:
3512 DMEMIT_TARGET_NAME_VERSION(ti->type);
3513 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3514 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3515 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3517 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3519 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3521 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3524 if (cc->on_disk_tag_size)
3525 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3526 cc->on_disk_tag_size, cc->cipher_auth);
3527 if (cc->sector_size != (1 << SECTOR_SHIFT))
3528 DMEMIT(",sector_size=%d", cc->sector_size);
3529 if (cc->cipher_string)
3530 DMEMIT(",cipher_string=%s", cc->cipher_string);
3532 DMEMIT(",key_size=%u", cc->key_size);
3533 DMEMIT(",key_parts=%u", cc->key_parts);
3534 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3535 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3541 static void crypt_postsuspend(struct dm_target *ti)
3543 struct crypt_config *cc = ti->private;
3545 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3548 static int crypt_preresume(struct dm_target *ti)
3550 struct crypt_config *cc = ti->private;
3552 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3553 DMERR("aborting resume - crypt key is not set.");
3560 static void crypt_resume(struct dm_target *ti)
3562 struct crypt_config *cc = ti->private;
3564 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3567 /* Message interface
3571 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3572 char *result, unsigned int maxlen)
3574 struct crypt_config *cc = ti->private;
3575 int key_size, ret = -EINVAL;
3580 if (!strcasecmp(argv[0], "key")) {
3581 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3582 DMWARN("not suspended during key manipulation.");
3585 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3586 /* The key size may not be changed. */
3587 key_size = get_key_size(&argv[2]);
3588 if (key_size < 0 || cc->key_size != key_size) {
3589 memset(argv[2], '0', strlen(argv[2]));
3593 ret = crypt_set_key(cc, argv[2]);
3596 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3597 ret = cc->iv_gen_ops->init(cc);
3598 /* wipe the kernel key payload copy */
3600 memset(cc->key, 0, cc->key_size * sizeof(u8));
3603 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3604 return crypt_wipe_key(cc);
3608 DMWARN("unrecognised message received.");
3612 static int crypt_iterate_devices(struct dm_target *ti,
3613 iterate_devices_callout_fn fn, void *data)
3615 struct crypt_config *cc = ti->private;
3617 return fn(ti, cc->dev, cc->start, ti->len, data);
3620 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3622 struct crypt_config *cc = ti->private;
3625 * Unfortunate constraint that is required to avoid the potential
3626 * for exceeding underlying device's max_segments limits -- due to
3627 * crypt_alloc_buffer() possibly allocating pages for the encryption
3628 * bio that are not as physically contiguous as the original bio.
3630 limits->max_segment_size = PAGE_SIZE;
3632 limits->logical_block_size =
3633 max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3634 limits->physical_block_size =
3635 max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3636 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3637 limits->dma_alignment = limits->logical_block_size - 1;
3640 static struct target_type crypt_target = {
3642 .version = {1, 24, 0},
3643 .module = THIS_MODULE,
3646 .features = DM_TARGET_ZONED_HM,
3647 .report_zones = crypt_report_zones,
3649 .status = crypt_status,
3650 .postsuspend = crypt_postsuspend,
3651 .preresume = crypt_preresume,
3652 .resume = crypt_resume,
3653 .message = crypt_message,
3654 .iterate_devices = crypt_iterate_devices,
3655 .io_hints = crypt_io_hints,
3659 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3660 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3661 MODULE_LICENSE("GPL");