2 * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4 * Copyright (C) 2006-2009 Red Hat, Inc. All rights reserved.
6 * This file is released under the GPL.
9 #include <linux/completion.h>
10 #include <linux/err.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/bio.h>
15 #include <linux/blkdev.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/crypto.h>
19 #include <linux/workqueue.h>
20 #include <linux/backing-dev.h>
21 #include <linux/percpu.h>
22 #include <linux/atomic.h>
23 #include <linux/scatterlist.h>
25 #include <asm/unaligned.h>
26 #include <crypto/hash.h>
27 #include <crypto/md5.h>
28 #include <crypto/algapi.h>
30 #include <linux/device-mapper.h>
32 #define DM_MSG_PREFIX "crypt"
35 * context holding the current state of a multi-part conversion
37 struct convert_context {
38 struct completion restart;
41 unsigned int offset_in;
42 unsigned int offset_out;
50 * per bio private data
53 struct crypt_config *cc;
55 struct work_struct work;
57 struct convert_context ctx;
62 struct dm_crypt_io *base_io;
65 struct dm_crypt_request {
66 struct convert_context *ctx;
67 struct scatterlist sg_in;
68 struct scatterlist sg_out;
74 struct crypt_iv_operations {
75 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
77 void (*dtr)(struct crypt_config *cc);
78 int (*init)(struct crypt_config *cc);
79 int (*wipe)(struct crypt_config *cc);
80 int (*generator)(struct crypt_config *cc, u8 *iv,
81 struct dm_crypt_request *dmreq);
82 int (*post)(struct crypt_config *cc, u8 *iv,
83 struct dm_crypt_request *dmreq);
86 struct iv_essiv_private {
87 struct crypto_hash *hash_tfm;
91 struct iv_benbi_private {
95 #define LMK_SEED_SIZE 64 /* hash + 0 */
96 struct iv_lmk_private {
97 struct crypto_shash *hash_tfm;
102 * Crypt: maps a linear range of a block device
103 * and encrypts / decrypts at the same time.
105 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID };
108 * Duplicated per-CPU state for cipher.
111 struct ablkcipher_request *req;
115 * The fields in here must be read only after initialization,
116 * changing state should be in crypt_cpu.
118 struct crypt_config {
123 * pool for per bio private data, crypto requests and
124 * encryption requeusts/buffer pages
128 mempool_t *page_pool;
131 struct workqueue_struct *io_queue;
132 struct workqueue_struct *crypt_queue;
137 struct crypt_iv_operations *iv_gen_ops;
139 struct iv_essiv_private essiv;
140 struct iv_benbi_private benbi;
141 struct iv_lmk_private lmk;
144 unsigned int iv_size;
147 * Duplicated per cpu state. Access through
148 * per_cpu_ptr() only.
150 struct crypt_cpu __percpu *cpu;
152 /* ESSIV: struct crypto_cipher *essiv_tfm */
154 struct crypto_ablkcipher **tfms;
158 * Layout of each crypto request:
160 * struct ablkcipher_request
163 * struct dm_crypt_request
167 * The padding is added so that dm_crypt_request and the IV are
170 unsigned int dmreq_start;
173 unsigned int key_size;
174 unsigned int key_parts;
179 #define MIN_POOL_PAGES 32
181 static struct kmem_cache *_crypt_io_pool;
183 static void clone_init(struct dm_crypt_io *, struct bio *);
184 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
185 static u8 *iv_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq);
187 static struct crypt_cpu *this_crypt_config(struct crypt_config *cc)
189 return this_cpu_ptr(cc->cpu);
193 * Use this to access cipher attributes that are the same for each CPU.
195 static struct crypto_ablkcipher *any_tfm(struct crypt_config *cc)
201 * Different IV generation algorithms:
203 * plain: the initial vector is the 32-bit little-endian version of the sector
204 * number, padded with zeros if necessary.
206 * plain64: the initial vector is the 64-bit little-endian version of the sector
207 * number, padded with zeros if necessary.
209 * essiv: "encrypted sector|salt initial vector", the sector number is
210 * encrypted with the bulk cipher using a salt as key. The salt
211 * should be derived from the bulk cipher's key via hashing.
213 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
214 * (needed for LRW-32-AES and possible other narrow block modes)
216 * null: the initial vector is always zero. Provides compatibility with
217 * obsolete loop_fish2 devices. Do not use for new devices.
219 * lmk: Compatible implementation of the block chaining mode used
220 * by the Loop-AES block device encryption system
221 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
222 * It operates on full 512 byte sectors and uses CBC
223 * with an IV derived from the sector number, the data and
224 * optionally extra IV seed.
225 * This means that after decryption the first block
226 * of sector must be tweaked according to decrypted data.
227 * Loop-AES can use three encryption schemes:
228 * version 1: is plain aes-cbc mode
229 * version 2: uses 64 multikey scheme with lmk IV generator
230 * version 3: the same as version 2 with additional IV seed
231 * (it uses 65 keys, last key is used as IV seed)
233 * plumb: unimplemented, see:
234 * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
237 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
238 struct dm_crypt_request *dmreq)
240 memset(iv, 0, cc->iv_size);
241 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
246 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
247 struct dm_crypt_request *dmreq)
249 memset(iv, 0, cc->iv_size);
250 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
255 /* Initialise ESSIV - compute salt but no local memory allocations */
256 static int crypt_iv_essiv_init(struct crypt_config *cc)
258 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
259 struct hash_desc desc;
260 struct scatterlist sg;
261 struct crypto_cipher *essiv_tfm;
264 sg_init_one(&sg, cc->key, cc->key_size);
265 desc.tfm = essiv->hash_tfm;
266 desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
268 err = crypto_hash_digest(&desc, &sg, cc->key_size, essiv->salt);
272 essiv_tfm = cc->iv_private;
274 err = crypto_cipher_setkey(essiv_tfm, essiv->salt,
275 crypto_hash_digestsize(essiv->hash_tfm));
282 /* Wipe salt and reset key derived from volume key */
283 static int crypt_iv_essiv_wipe(struct crypt_config *cc)
285 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
286 unsigned salt_size = crypto_hash_digestsize(essiv->hash_tfm);
287 struct crypto_cipher *essiv_tfm;
290 memset(essiv->salt, 0, salt_size);
292 essiv_tfm = cc->iv_private;
293 r = crypto_cipher_setkey(essiv_tfm, essiv->salt, salt_size);
300 /* Set up per cpu cipher state */
301 static struct crypto_cipher *setup_essiv_cpu(struct crypt_config *cc,
302 struct dm_target *ti,
303 u8 *salt, unsigned saltsize)
305 struct crypto_cipher *essiv_tfm;
308 /* Setup the essiv_tfm with the given salt */
309 essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
310 if (IS_ERR(essiv_tfm)) {
311 ti->error = "Error allocating crypto tfm for ESSIV";
315 if (crypto_cipher_blocksize(essiv_tfm) !=
316 crypto_ablkcipher_ivsize(any_tfm(cc))) {
317 ti->error = "Block size of ESSIV cipher does "
318 "not match IV size of block cipher";
319 crypto_free_cipher(essiv_tfm);
320 return ERR_PTR(-EINVAL);
323 err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
325 ti->error = "Failed to set key for ESSIV cipher";
326 crypto_free_cipher(essiv_tfm);
333 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
335 struct crypto_cipher *essiv_tfm;
336 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
338 crypto_free_hash(essiv->hash_tfm);
339 essiv->hash_tfm = NULL;
344 essiv_tfm = cc->iv_private;
347 crypto_free_cipher(essiv_tfm);
349 cc->iv_private = NULL;
352 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
355 struct crypto_cipher *essiv_tfm = NULL;
356 struct crypto_hash *hash_tfm = NULL;
361 ti->error = "Digest algorithm missing for ESSIV mode";
365 /* Allocate hash algorithm */
366 hash_tfm = crypto_alloc_hash(opts, 0, CRYPTO_ALG_ASYNC);
367 if (IS_ERR(hash_tfm)) {
368 ti->error = "Error initializing ESSIV hash";
369 err = PTR_ERR(hash_tfm);
373 salt = kzalloc(crypto_hash_digestsize(hash_tfm), GFP_KERNEL);
375 ti->error = "Error kmallocing salt storage in ESSIV";
380 cc->iv_gen_private.essiv.salt = salt;
381 cc->iv_gen_private.essiv.hash_tfm = hash_tfm;
383 essiv_tfm = setup_essiv_cpu(cc, ti, salt,
384 crypto_hash_digestsize(hash_tfm));
385 if (IS_ERR(essiv_tfm)) {
386 crypt_iv_essiv_dtr(cc);
387 return PTR_ERR(essiv_tfm);
389 cc->iv_private = essiv_tfm;
394 if (hash_tfm && !IS_ERR(hash_tfm))
395 crypto_free_hash(hash_tfm);
400 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
401 struct dm_crypt_request *dmreq)
403 struct crypto_cipher *essiv_tfm = cc->iv_private;
405 memset(iv, 0, cc->iv_size);
406 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
407 crypto_cipher_encrypt_one(essiv_tfm, iv, iv);
412 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
415 unsigned bs = crypto_ablkcipher_blocksize(any_tfm(cc));
418 /* we need to calculate how far we must shift the sector count
419 * to get the cipher block count, we use this shift in _gen */
421 if (1 << log != bs) {
422 ti->error = "cypher blocksize is not a power of 2";
427 ti->error = "cypher blocksize is > 512";
431 cc->iv_gen_private.benbi.shift = 9 - log;
436 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
440 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
441 struct dm_crypt_request *dmreq)
445 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
447 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
448 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
453 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
454 struct dm_crypt_request *dmreq)
456 memset(iv, 0, cc->iv_size);
461 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
463 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
465 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
466 crypto_free_shash(lmk->hash_tfm);
467 lmk->hash_tfm = NULL;
473 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
476 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
478 lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
479 if (IS_ERR(lmk->hash_tfm)) {
480 ti->error = "Error initializing LMK hash";
481 return PTR_ERR(lmk->hash_tfm);
484 /* No seed in LMK version 2 */
485 if (cc->key_parts == cc->tfms_count) {
490 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
492 crypt_iv_lmk_dtr(cc);
493 ti->error = "Error kmallocing seed storage in LMK";
500 static int crypt_iv_lmk_init(struct crypt_config *cc)
502 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
503 int subkey_size = cc->key_size / cc->key_parts;
505 /* LMK seed is on the position of LMK_KEYS + 1 key */
507 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
508 crypto_shash_digestsize(lmk->hash_tfm));
513 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
515 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
518 memset(lmk->seed, 0, LMK_SEED_SIZE);
523 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
524 struct dm_crypt_request *dmreq,
527 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
529 struct shash_desc desc;
530 char ctx[crypto_shash_descsize(lmk->hash_tfm)];
532 struct md5_state md5state;
536 sdesc.desc.tfm = lmk->hash_tfm;
537 sdesc.desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
539 r = crypto_shash_init(&sdesc.desc);
544 r = crypto_shash_update(&sdesc.desc, lmk->seed, LMK_SEED_SIZE);
549 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
550 r = crypto_shash_update(&sdesc.desc, data + 16, 16 * 31);
554 /* Sector is cropped to 56 bits here */
555 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
556 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
557 buf[2] = cpu_to_le32(4024);
559 r = crypto_shash_update(&sdesc.desc, (u8 *)buf, sizeof(buf));
563 /* No MD5 padding here */
564 r = crypto_shash_export(&sdesc.desc, &md5state);
568 for (i = 0; i < MD5_HASH_WORDS; i++)
569 __cpu_to_le32s(&md5state.hash[i]);
570 memcpy(iv, &md5state.hash, cc->iv_size);
575 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
576 struct dm_crypt_request *dmreq)
581 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
582 src = kmap_atomic(sg_page(&dmreq->sg_in));
583 r = crypt_iv_lmk_one(cc, iv, dmreq, src + dmreq->sg_in.offset);
586 memset(iv, 0, cc->iv_size);
591 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
592 struct dm_crypt_request *dmreq)
597 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
600 dst = kmap_atomic(sg_page(&dmreq->sg_out));
601 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + dmreq->sg_out.offset);
603 /* Tweak the first block of plaintext sector */
605 crypto_xor(dst + dmreq->sg_out.offset, iv, cc->iv_size);
611 static struct crypt_iv_operations crypt_iv_plain_ops = {
612 .generator = crypt_iv_plain_gen
615 static struct crypt_iv_operations crypt_iv_plain64_ops = {
616 .generator = crypt_iv_plain64_gen
619 static struct crypt_iv_operations crypt_iv_essiv_ops = {
620 .ctr = crypt_iv_essiv_ctr,
621 .dtr = crypt_iv_essiv_dtr,
622 .init = crypt_iv_essiv_init,
623 .wipe = crypt_iv_essiv_wipe,
624 .generator = crypt_iv_essiv_gen
627 static struct crypt_iv_operations crypt_iv_benbi_ops = {
628 .ctr = crypt_iv_benbi_ctr,
629 .dtr = crypt_iv_benbi_dtr,
630 .generator = crypt_iv_benbi_gen
633 static struct crypt_iv_operations crypt_iv_null_ops = {
634 .generator = crypt_iv_null_gen
637 static struct crypt_iv_operations crypt_iv_lmk_ops = {
638 .ctr = crypt_iv_lmk_ctr,
639 .dtr = crypt_iv_lmk_dtr,
640 .init = crypt_iv_lmk_init,
641 .wipe = crypt_iv_lmk_wipe,
642 .generator = crypt_iv_lmk_gen,
643 .post = crypt_iv_lmk_post
646 static void crypt_convert_init(struct crypt_config *cc,
647 struct convert_context *ctx,
648 struct bio *bio_out, struct bio *bio_in,
651 ctx->bio_in = bio_in;
652 ctx->bio_out = bio_out;
655 ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
656 ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
657 ctx->cc_sector = sector + cc->iv_offset;
658 init_completion(&ctx->restart);
661 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
662 struct ablkcipher_request *req)
664 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
667 static struct ablkcipher_request *req_of_dmreq(struct crypt_config *cc,
668 struct dm_crypt_request *dmreq)
670 return (struct ablkcipher_request *)((char *)dmreq - cc->dmreq_start);
673 static u8 *iv_of_dmreq(struct crypt_config *cc,
674 struct dm_crypt_request *dmreq)
676 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
677 crypto_ablkcipher_alignmask(any_tfm(cc)) + 1);
680 static int crypt_convert_block(struct crypt_config *cc,
681 struct convert_context *ctx,
682 struct ablkcipher_request *req)
684 struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
685 struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
686 struct dm_crypt_request *dmreq;
690 dmreq = dmreq_of_req(cc, req);
691 iv = iv_of_dmreq(cc, dmreq);
693 dmreq->iv_sector = ctx->cc_sector;
695 sg_init_table(&dmreq->sg_in, 1);
696 sg_set_page(&dmreq->sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT,
697 bv_in->bv_offset + ctx->offset_in);
699 sg_init_table(&dmreq->sg_out, 1);
700 sg_set_page(&dmreq->sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT,
701 bv_out->bv_offset + ctx->offset_out);
703 ctx->offset_in += 1 << SECTOR_SHIFT;
704 if (ctx->offset_in >= bv_in->bv_len) {
709 ctx->offset_out += 1 << SECTOR_SHIFT;
710 if (ctx->offset_out >= bv_out->bv_len) {
715 if (cc->iv_gen_ops) {
716 r = cc->iv_gen_ops->generator(cc, iv, dmreq);
721 ablkcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
722 1 << SECTOR_SHIFT, iv);
724 if (bio_data_dir(ctx->bio_in) == WRITE)
725 r = crypto_ablkcipher_encrypt(req);
727 r = crypto_ablkcipher_decrypt(req);
729 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
730 r = cc->iv_gen_ops->post(cc, iv, dmreq);
735 static void kcryptd_async_done(struct crypto_async_request *async_req,
738 static void crypt_alloc_req(struct crypt_config *cc,
739 struct convert_context *ctx)
741 struct crypt_cpu *this_cc = this_crypt_config(cc);
742 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
745 this_cc->req = mempool_alloc(cc->req_pool, GFP_NOIO);
747 ablkcipher_request_set_tfm(this_cc->req, cc->tfms[key_index]);
748 ablkcipher_request_set_callback(this_cc->req,
749 CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
750 kcryptd_async_done, dmreq_of_req(cc, this_cc->req));
754 * Encrypt / decrypt data from one bio to another one (can be the same one)
756 static int crypt_convert(struct crypt_config *cc,
757 struct convert_context *ctx)
759 struct crypt_cpu *this_cc = this_crypt_config(cc);
762 atomic_set(&ctx->cc_pending, 1);
764 while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
765 ctx->idx_out < ctx->bio_out->bi_vcnt) {
767 crypt_alloc_req(cc, ctx);
769 atomic_inc(&ctx->cc_pending);
771 r = crypt_convert_block(cc, ctx, this_cc->req);
776 wait_for_completion(&ctx->restart);
777 INIT_COMPLETION(ctx->restart);
786 atomic_dec(&ctx->cc_pending);
793 atomic_dec(&ctx->cc_pending);
802 * Generate a new unfragmented bio with the given size
803 * This should never violate the device limitations
804 * May return a smaller bio when running out of pages, indicated by
805 * *out_of_pages set to 1.
807 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size,
808 unsigned *out_of_pages)
810 struct crypt_config *cc = io->cc;
812 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
813 gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
817 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
821 clone_init(io, clone);
824 for (i = 0; i < nr_iovecs; i++) {
825 page = mempool_alloc(cc->page_pool, gfp_mask);
832 * If additional pages cannot be allocated without waiting,
833 * return a partially-allocated bio. The caller will then try
834 * to allocate more bios while submitting this partial bio.
836 gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
838 len = (size > PAGE_SIZE) ? PAGE_SIZE : size;
840 if (!bio_add_page(clone, page, len, 0)) {
841 mempool_free(page, cc->page_pool);
848 if (!clone->bi_size) {
856 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
861 for (i = 0; i < clone->bi_vcnt; i++) {
862 bv = bio_iovec_idx(clone, i);
863 BUG_ON(!bv->bv_page);
864 mempool_free(bv->bv_page, cc->page_pool);
869 static struct dm_crypt_io *crypt_io_alloc(struct crypt_config *cc,
870 struct bio *bio, sector_t sector)
872 struct dm_crypt_io *io;
874 io = mempool_alloc(cc->io_pool, GFP_NOIO);
880 atomic_set(&io->io_pending, 0);
885 static void crypt_inc_pending(struct dm_crypt_io *io)
887 atomic_inc(&io->io_pending);
891 * One of the bios was finished. Check for completion of
892 * the whole request and correctly clean up the buffer.
893 * If base_io is set, wait for the last fragment to complete.
895 static void crypt_dec_pending(struct dm_crypt_io *io)
897 struct crypt_config *cc = io->cc;
898 struct bio *base_bio = io->base_bio;
899 struct dm_crypt_io *base_io = io->base_io;
900 int error = io->error;
902 if (!atomic_dec_and_test(&io->io_pending))
905 mempool_free(io, cc->io_pool);
907 if (likely(!base_io))
908 bio_endio(base_bio, error);
910 if (error && !base_io->error)
911 base_io->error = error;
912 crypt_dec_pending(base_io);
917 * kcryptd/kcryptd_io:
919 * Needed because it would be very unwise to do decryption in an
922 * kcryptd performs the actual encryption or decryption.
924 * kcryptd_io performs the IO submission.
926 * They must be separated as otherwise the final stages could be
927 * starved by new requests which can block in the first stages due
928 * to memory allocation.
930 * The work is done per CPU global for all dm-crypt instances.
931 * They should not depend on each other and do not block.
933 static void crypt_endio(struct bio *clone, int error)
935 struct dm_crypt_io *io = clone->bi_private;
936 struct crypt_config *cc = io->cc;
937 unsigned rw = bio_data_dir(clone);
939 if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
943 * free the processed pages
946 crypt_free_buffer_pages(cc, clone);
950 if (rw == READ && !error) {
951 kcryptd_queue_crypt(io);
958 crypt_dec_pending(io);
961 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
963 struct crypt_config *cc = io->cc;
965 clone->bi_private = io;
966 clone->bi_end_io = crypt_endio;
967 clone->bi_bdev = cc->dev->bdev;
968 clone->bi_rw = io->base_bio->bi_rw;
971 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
973 struct crypt_config *cc = io->cc;
974 struct bio *base_bio = io->base_bio;
978 * The block layer might modify the bvec array, so always
979 * copy the required bvecs because we need the original
980 * one in order to decrypt the whole bio data *afterwards*.
982 clone = bio_clone_bioset(base_bio, gfp, cc->bs);
986 crypt_inc_pending(io);
988 clone_init(io, clone);
989 clone->bi_sector = cc->start + io->sector;
991 generic_make_request(clone);
995 static void kcryptd_io_write(struct dm_crypt_io *io)
997 struct bio *clone = io->ctx.bio_out;
998 generic_make_request(clone);
1001 static void kcryptd_io(struct work_struct *work)
1003 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1005 if (bio_data_dir(io->base_bio) == READ) {
1006 crypt_inc_pending(io);
1007 if (kcryptd_io_read(io, GFP_NOIO))
1008 io->error = -ENOMEM;
1009 crypt_dec_pending(io);
1011 kcryptd_io_write(io);
1014 static void kcryptd_queue_io(struct dm_crypt_io *io)
1016 struct crypt_config *cc = io->cc;
1018 INIT_WORK(&io->work, kcryptd_io);
1019 queue_work(cc->io_queue, &io->work);
1022 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1024 struct bio *clone = io->ctx.bio_out;
1025 struct crypt_config *cc = io->cc;
1027 if (unlikely(io->error < 0)) {
1028 crypt_free_buffer_pages(cc, clone);
1030 crypt_dec_pending(io);
1034 /* crypt_convert should have filled the clone bio */
1035 BUG_ON(io->ctx.idx_out < clone->bi_vcnt);
1037 clone->bi_sector = cc->start + io->sector;
1040 kcryptd_queue_io(io);
1042 generic_make_request(clone);
1045 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1047 struct crypt_config *cc = io->cc;
1049 struct dm_crypt_io *new_io;
1051 unsigned out_of_pages = 0;
1052 unsigned remaining = io->base_bio->bi_size;
1053 sector_t sector = io->sector;
1057 * Prevent io from disappearing until this function completes.
1059 crypt_inc_pending(io);
1060 crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1063 * The allocated buffers can be smaller than the whole bio,
1064 * so repeat the whole process until all the data can be handled.
1067 clone = crypt_alloc_buffer(io, remaining, &out_of_pages);
1068 if (unlikely(!clone)) {
1069 io->error = -ENOMEM;
1073 io->ctx.bio_out = clone;
1074 io->ctx.idx_out = 0;
1076 remaining -= clone->bi_size;
1077 sector += bio_sectors(clone);
1079 crypt_inc_pending(io);
1081 r = crypt_convert(cc, &io->ctx);
1085 crypt_finished = atomic_dec_and_test(&io->ctx.cc_pending);
1087 /* Encryption was already finished, submit io now */
1088 if (crypt_finished) {
1089 kcryptd_crypt_write_io_submit(io, 0);
1092 * If there was an error, do not try next fragments.
1093 * For async, error is processed in async handler.
1095 if (unlikely(r < 0))
1098 io->sector = sector;
1102 * Out of memory -> run queues
1103 * But don't wait if split was due to the io size restriction
1105 if (unlikely(out_of_pages))
1106 congestion_wait(BLK_RW_ASYNC, HZ/100);
1109 * With async crypto it is unsafe to share the crypto context
1110 * between fragments, so switch to a new dm_crypt_io structure.
1112 if (unlikely(!crypt_finished && remaining)) {
1113 new_io = crypt_io_alloc(io->cc, io->base_bio,
1115 crypt_inc_pending(new_io);
1116 crypt_convert_init(cc, &new_io->ctx, NULL,
1117 io->base_bio, sector);
1118 new_io->ctx.idx_in = io->ctx.idx_in;
1119 new_io->ctx.offset_in = io->ctx.offset_in;
1122 * Fragments after the first use the base_io
1126 new_io->base_io = io;
1128 new_io->base_io = io->base_io;
1129 crypt_inc_pending(io->base_io);
1130 crypt_dec_pending(io);
1137 crypt_dec_pending(io);
1140 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1142 crypt_dec_pending(io);
1145 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1147 struct crypt_config *cc = io->cc;
1150 crypt_inc_pending(io);
1152 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1155 r = crypt_convert(cc, &io->ctx);
1159 if (atomic_dec_and_test(&io->ctx.cc_pending))
1160 kcryptd_crypt_read_done(io);
1162 crypt_dec_pending(io);
1165 static void kcryptd_async_done(struct crypto_async_request *async_req,
1168 struct dm_crypt_request *dmreq = async_req->data;
1169 struct convert_context *ctx = dmreq->ctx;
1170 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1171 struct crypt_config *cc = io->cc;
1173 if (error == -EINPROGRESS) {
1174 complete(&ctx->restart);
1178 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1179 error = cc->iv_gen_ops->post(cc, iv_of_dmreq(cc, dmreq), dmreq);
1184 mempool_free(req_of_dmreq(cc, dmreq), cc->req_pool);
1186 if (!atomic_dec_and_test(&ctx->cc_pending))
1189 if (bio_data_dir(io->base_bio) == READ)
1190 kcryptd_crypt_read_done(io);
1192 kcryptd_crypt_write_io_submit(io, 1);
1195 static void kcryptd_crypt(struct work_struct *work)
1197 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1199 if (bio_data_dir(io->base_bio) == READ)
1200 kcryptd_crypt_read_convert(io);
1202 kcryptd_crypt_write_convert(io);
1205 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
1207 struct crypt_config *cc = io->cc;
1209 INIT_WORK(&io->work, kcryptd_crypt);
1210 queue_work(cc->crypt_queue, &io->work);
1214 * Decode key from its hex representation
1216 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
1223 for (i = 0; i < size; i++) {
1227 if (kstrtou8(buffer, 16, &key[i]))
1238 * Encode key into its hex representation
1240 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
1244 for (i = 0; i < size; i++) {
1245 sprintf(hex, "%02x", *key);
1251 static void crypt_free_tfms(struct crypt_config *cc)
1258 for (i = 0; i < cc->tfms_count; i++)
1259 if (cc->tfms[i] && !IS_ERR(cc->tfms[i])) {
1260 crypto_free_ablkcipher(cc->tfms[i]);
1268 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
1273 cc->tfms = kmalloc(cc->tfms_count * sizeof(struct crypto_ablkcipher *),
1278 for (i = 0; i < cc->tfms_count; i++) {
1279 cc->tfms[i] = crypto_alloc_ablkcipher(ciphermode, 0, 0);
1280 if (IS_ERR(cc->tfms[i])) {
1281 err = PTR_ERR(cc->tfms[i]);
1282 crypt_free_tfms(cc);
1290 static int crypt_setkey_allcpus(struct crypt_config *cc)
1292 unsigned subkey_size = cc->key_size >> ilog2(cc->tfms_count);
1295 for (i = 0; i < cc->tfms_count; i++) {
1296 r = crypto_ablkcipher_setkey(cc->tfms[i],
1297 cc->key + (i * subkey_size),
1306 static int crypt_set_key(struct crypt_config *cc, char *key)
1309 int key_string_len = strlen(key);
1311 /* The key size may not be changed. */
1312 if (cc->key_size != (key_string_len >> 1))
1315 /* Hyphen (which gives a key_size of zero) means there is no key. */
1316 if (!cc->key_size && strcmp(key, "-"))
1319 if (cc->key_size && crypt_decode_key(cc->key, key, cc->key_size) < 0)
1322 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1324 r = crypt_setkey_allcpus(cc);
1327 /* Hex key string not needed after here, so wipe it. */
1328 memset(key, '0', key_string_len);
1333 static int crypt_wipe_key(struct crypt_config *cc)
1335 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1336 memset(&cc->key, 0, cc->key_size * sizeof(u8));
1338 return crypt_setkey_allcpus(cc);
1341 static void crypt_dtr(struct dm_target *ti)
1343 struct crypt_config *cc = ti->private;
1344 struct crypt_cpu *cpu_cc;
1353 destroy_workqueue(cc->io_queue);
1354 if (cc->crypt_queue)
1355 destroy_workqueue(cc->crypt_queue);
1358 for_each_possible_cpu(cpu) {
1359 cpu_cc = per_cpu_ptr(cc->cpu, cpu);
1361 mempool_free(cpu_cc->req, cc->req_pool);
1364 crypt_free_tfms(cc);
1367 bioset_free(cc->bs);
1370 mempool_destroy(cc->page_pool);
1372 mempool_destroy(cc->req_pool);
1374 mempool_destroy(cc->io_pool);
1376 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1377 cc->iv_gen_ops->dtr(cc);
1380 dm_put_device(ti, cc->dev);
1383 free_percpu(cc->cpu);
1386 kzfree(cc->cipher_string);
1388 /* Must zero key material before freeing */
1392 static int crypt_ctr_cipher(struct dm_target *ti,
1393 char *cipher_in, char *key)
1395 struct crypt_config *cc = ti->private;
1396 char *tmp, *cipher, *chainmode, *ivmode, *ivopts, *keycount;
1397 char *cipher_api = NULL;
1401 /* Convert to crypto api definition? */
1402 if (strchr(cipher_in, '(')) {
1403 ti->error = "Bad cipher specification";
1407 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
1408 if (!cc->cipher_string)
1412 * Legacy dm-crypt cipher specification
1413 * cipher[:keycount]-mode-iv:ivopts
1416 keycount = strsep(&tmp, "-");
1417 cipher = strsep(&keycount, ":");
1421 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
1422 !is_power_of_2(cc->tfms_count)) {
1423 ti->error = "Bad cipher key count specification";
1426 cc->key_parts = cc->tfms_count;
1428 cc->cipher = kstrdup(cipher, GFP_KERNEL);
1432 chainmode = strsep(&tmp, "-");
1433 ivopts = strsep(&tmp, "-");
1434 ivmode = strsep(&ivopts, ":");
1437 DMWARN("Ignoring unexpected additional cipher options");
1439 cc->cpu = __alloc_percpu(sizeof(*(cc->cpu)),
1440 __alignof__(struct crypt_cpu));
1442 ti->error = "Cannot allocate per cpu state";
1447 * For compatibility with the original dm-crypt mapping format, if
1448 * only the cipher name is supplied, use cbc-plain.
1450 if (!chainmode || (!strcmp(chainmode, "plain") && !ivmode)) {
1455 if (strcmp(chainmode, "ecb") && !ivmode) {
1456 ti->error = "IV mechanism required";
1460 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
1464 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
1465 "%s(%s)", chainmode, cipher);
1471 /* Allocate cipher */
1472 ret = crypt_alloc_tfms(cc, cipher_api);
1474 ti->error = "Error allocating crypto tfm";
1478 /* Initialize and set key */
1479 ret = crypt_set_key(cc, key);
1481 ti->error = "Error decoding and setting key";
1486 cc->iv_size = crypto_ablkcipher_ivsize(any_tfm(cc));
1488 /* at least a 64 bit sector number should fit in our buffer */
1489 cc->iv_size = max(cc->iv_size,
1490 (unsigned int)(sizeof(u64) / sizeof(u8)));
1492 DMWARN("Selected cipher does not support IVs");
1496 /* Choose ivmode, see comments at iv code. */
1498 cc->iv_gen_ops = NULL;
1499 else if (strcmp(ivmode, "plain") == 0)
1500 cc->iv_gen_ops = &crypt_iv_plain_ops;
1501 else if (strcmp(ivmode, "plain64") == 0)
1502 cc->iv_gen_ops = &crypt_iv_plain64_ops;
1503 else if (strcmp(ivmode, "essiv") == 0)
1504 cc->iv_gen_ops = &crypt_iv_essiv_ops;
1505 else if (strcmp(ivmode, "benbi") == 0)
1506 cc->iv_gen_ops = &crypt_iv_benbi_ops;
1507 else if (strcmp(ivmode, "null") == 0)
1508 cc->iv_gen_ops = &crypt_iv_null_ops;
1509 else if (strcmp(ivmode, "lmk") == 0) {
1510 cc->iv_gen_ops = &crypt_iv_lmk_ops;
1511 /* Version 2 and 3 is recognised according
1512 * to length of provided multi-key string.
1513 * If present (version 3), last key is used as IV seed.
1515 if (cc->key_size % cc->key_parts)
1519 ti->error = "Invalid IV mode";
1524 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
1525 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
1527 ti->error = "Error creating IV";
1532 /* Initialize IV (set keys for ESSIV etc) */
1533 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
1534 ret = cc->iv_gen_ops->init(cc);
1536 ti->error = "Error initialising IV";
1547 ti->error = "Cannot allocate cipher strings";
1552 * Construct an encryption mapping:
1553 * <cipher> <key> <iv_offset> <dev_path> <start>
1555 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1557 struct crypt_config *cc;
1558 unsigned int key_size, opt_params;
1559 unsigned long long tmpll;
1561 struct dm_arg_set as;
1562 const char *opt_string;
1565 static struct dm_arg _args[] = {
1566 {0, 1, "Invalid number of feature args"},
1570 ti->error = "Not enough arguments";
1574 key_size = strlen(argv[1]) >> 1;
1576 cc = kzalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
1578 ti->error = "Cannot allocate encryption context";
1581 cc->key_size = key_size;
1584 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
1589 cc->io_pool = mempool_create_slab_pool(MIN_IOS, _crypt_io_pool);
1591 ti->error = "Cannot allocate crypt io mempool";
1595 cc->dmreq_start = sizeof(struct ablkcipher_request);
1596 cc->dmreq_start += crypto_ablkcipher_reqsize(any_tfm(cc));
1597 cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment());
1598 cc->dmreq_start += crypto_ablkcipher_alignmask(any_tfm(cc)) &
1599 ~(crypto_tfm_ctx_alignment() - 1);
1601 cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
1602 sizeof(struct dm_crypt_request) + cc->iv_size);
1603 if (!cc->req_pool) {
1604 ti->error = "Cannot allocate crypt request mempool";
1608 cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
1609 if (!cc->page_pool) {
1610 ti->error = "Cannot allocate page mempool";
1614 cc->bs = bioset_create(MIN_IOS, 0);
1616 ti->error = "Cannot allocate crypt bioset";
1621 if (sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) {
1622 ti->error = "Invalid iv_offset sector";
1625 cc->iv_offset = tmpll;
1627 if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev)) {
1628 ti->error = "Device lookup failed";
1632 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
1633 ti->error = "Invalid device sector";
1641 /* Optional parameters */
1646 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
1650 opt_string = dm_shift_arg(&as);
1652 if (opt_params == 1 && opt_string &&
1653 !strcasecmp(opt_string, "allow_discards"))
1654 ti->num_discard_requests = 1;
1655 else if (opt_params) {
1657 ti->error = "Invalid feature arguments";
1663 cc->io_queue = alloc_workqueue("kcryptd_io",
1667 if (!cc->io_queue) {
1668 ti->error = "Couldn't create kcryptd io queue";
1672 cc->crypt_queue = alloc_workqueue("kcryptd",
1677 if (!cc->crypt_queue) {
1678 ti->error = "Couldn't create kcryptd queue";
1682 ti->num_flush_requests = 1;
1683 ti->discard_zeroes_data_unsupported = true;
1692 static int crypt_map(struct dm_target *ti, struct bio *bio,
1693 union map_info *map_context)
1695 struct dm_crypt_io *io;
1696 struct crypt_config *cc = ti->private;
1699 * If bio is REQ_FLUSH or REQ_DISCARD, just bypass crypt queues.
1700 * - for REQ_FLUSH device-mapper core ensures that no IO is in-flight
1701 * - for REQ_DISCARD caller must use flush if IO ordering matters
1703 if (unlikely(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD))) {
1704 bio->bi_bdev = cc->dev->bdev;
1705 if (bio_sectors(bio))
1706 bio->bi_sector = cc->start + dm_target_offset(ti, bio->bi_sector);
1707 return DM_MAPIO_REMAPPED;
1710 io = crypt_io_alloc(cc, bio, dm_target_offset(ti, bio->bi_sector));
1712 if (bio_data_dir(io->base_bio) == READ) {
1713 if (kcryptd_io_read(io, GFP_NOWAIT))
1714 kcryptd_queue_io(io);
1716 kcryptd_queue_crypt(io);
1718 return DM_MAPIO_SUBMITTED;
1721 static int crypt_status(struct dm_target *ti, status_type_t type,
1722 unsigned status_flags, char *result, unsigned maxlen)
1724 struct crypt_config *cc = ti->private;
1725 unsigned int sz = 0;
1728 case STATUSTYPE_INFO:
1732 case STATUSTYPE_TABLE:
1733 DMEMIT("%s ", cc->cipher_string);
1735 if (cc->key_size > 0) {
1736 if ((maxlen - sz) < ((cc->key_size << 1) + 1))
1739 crypt_encode_key(result + sz, cc->key, cc->key_size);
1740 sz += cc->key_size << 1;
1747 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
1748 cc->dev->name, (unsigned long long)cc->start);
1750 if (ti->num_discard_requests)
1751 DMEMIT(" 1 allow_discards");
1758 static void crypt_postsuspend(struct dm_target *ti)
1760 struct crypt_config *cc = ti->private;
1762 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1765 static int crypt_preresume(struct dm_target *ti)
1767 struct crypt_config *cc = ti->private;
1769 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
1770 DMERR("aborting resume - crypt key is not set.");
1777 static void crypt_resume(struct dm_target *ti)
1779 struct crypt_config *cc = ti->private;
1781 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1784 /* Message interface
1788 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv)
1790 struct crypt_config *cc = ti->private;
1796 if (!strcasecmp(argv[0], "key")) {
1797 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
1798 DMWARN("not suspended during key manipulation.");
1801 if (argc == 3 && !strcasecmp(argv[1], "set")) {
1802 ret = crypt_set_key(cc, argv[2]);
1805 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
1806 ret = cc->iv_gen_ops->init(cc);
1809 if (argc == 2 && !strcasecmp(argv[1], "wipe")) {
1810 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
1811 ret = cc->iv_gen_ops->wipe(cc);
1815 return crypt_wipe_key(cc);
1820 DMWARN("unrecognised message received.");
1824 static int crypt_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
1825 struct bio_vec *biovec, int max_size)
1827 struct crypt_config *cc = ti->private;
1828 struct request_queue *q = bdev_get_queue(cc->dev->bdev);
1830 if (!q->merge_bvec_fn)
1833 bvm->bi_bdev = cc->dev->bdev;
1834 bvm->bi_sector = cc->start + dm_target_offset(ti, bvm->bi_sector);
1836 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
1839 static int crypt_iterate_devices(struct dm_target *ti,
1840 iterate_devices_callout_fn fn, void *data)
1842 struct crypt_config *cc = ti->private;
1844 return fn(ti, cc->dev, cc->start, ti->len, data);
1847 static struct target_type crypt_target = {
1849 .version = {1, 11, 0},
1850 .module = THIS_MODULE,
1854 .status = crypt_status,
1855 .postsuspend = crypt_postsuspend,
1856 .preresume = crypt_preresume,
1857 .resume = crypt_resume,
1858 .message = crypt_message,
1859 .merge = crypt_merge,
1860 .iterate_devices = crypt_iterate_devices,
1863 static int __init dm_crypt_init(void)
1867 _crypt_io_pool = KMEM_CACHE(dm_crypt_io, 0);
1868 if (!_crypt_io_pool)
1871 r = dm_register_target(&crypt_target);
1873 DMERR("register failed %d", r);
1874 kmem_cache_destroy(_crypt_io_pool);
1880 static void __exit dm_crypt_exit(void)
1882 dm_unregister_target(&crypt_target);
1883 kmem_cache_destroy(_crypt_io_pool);
1886 module_init(dm_crypt_init);
1887 module_exit(dm_crypt_exit);
1889 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
1890 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
1891 MODULE_LICENSE("GPL");