5 * This file is released under the GPL.
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/kernel.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/crypto.h>
17 #include <linux/workqueue.h>
18 #include <asm/atomic.h>
19 #include <linux/scatterlist.h>
24 #define DM_MSG_PREFIX "crypt"
27 * per bio private data
30 struct dm_target *target;
32 struct bio *first_clone;
33 struct work_struct work;
39 * context holding the current state of a multi-part conversion
41 struct convert_context {
44 unsigned int offset_in;
45 unsigned int offset_out;
54 struct crypt_iv_operations {
55 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
57 void (*dtr)(struct crypt_config *cc);
58 const char *(*status)(struct crypt_config *cc);
59 int (*generator)(struct crypt_config *cc, u8 *iv, sector_t sector);
63 * Crypt: maps a linear range of a block device
64 * and encrypts / decrypts at the same time.
71 * pool for per bio private data and
72 * for encryption buffer pages
80 struct crypt_iv_operations *iv_gen_ops;
82 struct crypto_cipher *iv_gen_private;
86 char cipher[CRYPTO_MAX_ALG_NAME];
87 char chainmode[CRYPTO_MAX_ALG_NAME];
88 struct crypto_blkcipher *tfm;
89 unsigned int key_size;
94 #define MIN_POOL_PAGES 32
95 #define MIN_BIO_PAGES 8
97 static kmem_cache_t *_crypt_io_pool;
100 * Different IV generation algorithms:
102 * plain: the initial vector is the 32-bit low-endian version of the sector
103 * number, padded with zeros if neccessary.
105 * ess_iv: "encrypted sector|salt initial vector", the sector number is
106 * encrypted with the bulk cipher using a salt as key. The salt
107 * should be derived from the bulk cipher's key via hashing.
109 * plumb: unimplemented, see:
110 * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
113 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
115 memset(iv, 0, cc->iv_size);
116 *(u32 *)iv = cpu_to_le32(sector & 0xffffffff);
121 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
124 struct crypto_cipher *essiv_tfm;
125 struct crypto_tfm *hash_tfm;
126 struct scatterlist sg;
127 unsigned int saltsize;
132 ti->error = "Digest algorithm missing for ESSIV mode";
136 /* Hash the cipher key with the given hash algorithm */
137 hash_tfm = crypto_alloc_tfm(opts, CRYPTO_TFM_REQ_MAY_SLEEP);
138 if (hash_tfm == NULL) {
139 ti->error = "Error initializing ESSIV hash";
143 if (crypto_tfm_alg_type(hash_tfm) != CRYPTO_ALG_TYPE_DIGEST) {
144 ti->error = "Expected digest algorithm for ESSIV hash";
145 crypto_free_tfm(hash_tfm);
149 saltsize = crypto_tfm_alg_digestsize(hash_tfm);
150 salt = kmalloc(saltsize, GFP_KERNEL);
152 ti->error = "Error kmallocing salt storage in ESSIV";
153 crypto_free_tfm(hash_tfm);
157 sg_set_buf(&sg, cc->key, cc->key_size);
158 crypto_digest_digest(hash_tfm, &sg, 1, salt);
159 crypto_free_tfm(hash_tfm);
161 /* Setup the essiv_tfm with the given salt */
162 essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
163 if (IS_ERR(essiv_tfm)) {
164 ti->error = "Error allocating crypto tfm for ESSIV";
166 return PTR_ERR(essiv_tfm);
168 if (crypto_cipher_blocksize(essiv_tfm) !=
169 crypto_blkcipher_ivsize(cc->tfm)) {
170 ti->error = "Block size of ESSIV cipher does "
171 "not match IV size of block cipher";
172 crypto_free_cipher(essiv_tfm);
176 err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
178 ti->error = "Failed to set key for ESSIV cipher";
179 crypto_free_cipher(essiv_tfm);
185 cc->iv_gen_private = essiv_tfm;
189 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
191 crypto_free_cipher(cc->iv_gen_private);
192 cc->iv_gen_private = NULL;
195 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
197 memset(iv, 0, cc->iv_size);
198 *(u64 *)iv = cpu_to_le64(sector);
199 crypto_cipher_encrypt_one(cc->iv_gen_private, iv, iv);
203 static struct crypt_iv_operations crypt_iv_plain_ops = {
204 .generator = crypt_iv_plain_gen
207 static struct crypt_iv_operations crypt_iv_essiv_ops = {
208 .ctr = crypt_iv_essiv_ctr,
209 .dtr = crypt_iv_essiv_dtr,
210 .generator = crypt_iv_essiv_gen
215 crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
216 struct scatterlist *in, unsigned int length,
217 int write, sector_t sector)
220 struct blkcipher_desc desc = {
223 .flags = CRYPTO_TFM_REQ_MAY_SLEEP,
227 if (cc->iv_gen_ops) {
228 r = cc->iv_gen_ops->generator(cc, iv, sector);
233 r = crypto_blkcipher_encrypt_iv(&desc, out, in, length);
235 r = crypto_blkcipher_decrypt_iv(&desc, out, in, length);
238 r = crypto_blkcipher_encrypt(&desc, out, in, length);
240 r = crypto_blkcipher_decrypt(&desc, out, in, length);
247 crypt_convert_init(struct crypt_config *cc, struct convert_context *ctx,
248 struct bio *bio_out, struct bio *bio_in,
249 sector_t sector, int write)
251 ctx->bio_in = bio_in;
252 ctx->bio_out = bio_out;
255 ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
256 ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
257 ctx->sector = sector + cc->iv_offset;
262 * Encrypt / decrypt data from one bio to another one (can be the same one)
264 static int crypt_convert(struct crypt_config *cc,
265 struct convert_context *ctx)
269 while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
270 ctx->idx_out < ctx->bio_out->bi_vcnt) {
271 struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
272 struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
273 struct scatterlist sg_in = {
274 .page = bv_in->bv_page,
275 .offset = bv_in->bv_offset + ctx->offset_in,
276 .length = 1 << SECTOR_SHIFT
278 struct scatterlist sg_out = {
279 .page = bv_out->bv_page,
280 .offset = bv_out->bv_offset + ctx->offset_out,
281 .length = 1 << SECTOR_SHIFT
284 ctx->offset_in += sg_in.length;
285 if (ctx->offset_in >= bv_in->bv_len) {
290 ctx->offset_out += sg_out.length;
291 if (ctx->offset_out >= bv_out->bv_len) {
296 r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length,
297 ctx->write, ctx->sector);
308 * Generate a new unfragmented bio with the given size
309 * This should never violate the device limitations
310 * May return a smaller bio when running out of pages
313 crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
314 struct bio *base_bio, unsigned int *bio_vec_idx)
317 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
318 gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
322 * Use __GFP_NOMEMALLOC to tell the VM to act less aggressively and
323 * to fail earlier. This is not necessary but increases throughput.
324 * FIXME: Is this really intelligent?
327 bio = bio_clone(base_bio, GFP_NOIO|__GFP_NOMEMALLOC);
329 bio = bio_alloc(GFP_NOIO|__GFP_NOMEMALLOC, nr_iovecs);
333 /* if the last bio was not complete, continue where that one ended */
334 bio->bi_idx = *bio_vec_idx;
335 bio->bi_vcnt = *bio_vec_idx;
337 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
339 /* bio->bi_idx pages have already been allocated */
340 size -= bio->bi_idx * PAGE_SIZE;
342 for(i = bio->bi_idx; i < nr_iovecs; i++) {
343 struct bio_vec *bv = bio_iovec_idx(bio, i);
345 bv->bv_page = mempool_alloc(cc->page_pool, gfp_mask);
350 * if additional pages cannot be allocated without waiting,
351 * return a partially allocated bio, the caller will then try
352 * to allocate additional bios while submitting this partial bio
354 if ((i - bio->bi_idx) == (MIN_BIO_PAGES - 1))
355 gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
358 if (size > PAGE_SIZE)
359 bv->bv_len = PAGE_SIZE;
363 bio->bi_size += bv->bv_len;
374 * Remember the last bio_vec allocated to be able
375 * to correctly continue after the splitting.
377 *bio_vec_idx = bio->bi_vcnt;
382 static void crypt_free_buffer_pages(struct crypt_config *cc,
383 struct bio *bio, unsigned int bytes)
385 unsigned int i, start, end;
389 * This is ugly, but Jens Axboe thinks that using bi_idx in the
390 * endio function is too dangerous at the moment, so I calculate the
391 * correct position using bi_vcnt and bi_size.
392 * The bv_offset and bv_len fields might already be modified but we
393 * know that we always allocated whole pages.
394 * A fix to the bi_idx issue in the kernel is in the works, so
395 * we will hopefully be able to revert to the cleaner solution soon.
397 i = bio->bi_vcnt - 1;
398 bv = bio_iovec_idx(bio, i);
399 end = (i << PAGE_SHIFT) + (bv->bv_offset + bv->bv_len) - bio->bi_size;
402 start >>= PAGE_SHIFT;
408 for(i = start; i < end; i++) {
409 bv = bio_iovec_idx(bio, i);
410 BUG_ON(!bv->bv_page);
411 mempool_free(bv->bv_page, cc->page_pool);
417 * One of the bios was finished. Check for completion of
418 * the whole request and correctly clean up the buffer.
420 static void dec_pending(struct crypt_io *io, int error)
422 struct crypt_config *cc = (struct crypt_config *) io->target->private;
427 if (!atomic_dec_and_test(&io->pending))
431 bio_put(io->first_clone);
433 bio_endio(io->bio, io->bio->bi_size, io->error);
435 mempool_free(io, cc->io_pool);
441 * Needed because it would be very unwise to do decryption in an
442 * interrupt context, so bios returning from read requests get
445 static struct workqueue_struct *_kcryptd_workqueue;
447 static void kcryptd_do_work(void *data)
449 struct crypt_io *io = (struct crypt_io *) data;
450 struct crypt_config *cc = (struct crypt_config *) io->target->private;
451 struct convert_context ctx;
454 crypt_convert_init(cc, &ctx, io->bio, io->bio,
455 io->bio->bi_sector - io->target->begin, 0);
456 r = crypt_convert(cc, &ctx);
461 static void kcryptd_queue_io(struct crypt_io *io)
463 INIT_WORK(&io->work, kcryptd_do_work, io);
464 queue_work(_kcryptd_workqueue, &io->work);
468 * Decode key from its hex representation
470 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
478 for(i = 0; i < size; i++) {
482 key[i] = (u8)simple_strtoul(buffer, &endp, 16);
484 if (endp != &buffer[2])
495 * Encode key into its hex representation
497 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
501 for(i = 0; i < size; i++) {
502 sprintf(hex, "%02x", *key);
509 * Construct an encryption mapping:
510 * <cipher> <key> <iv_offset> <dev_path> <start>
512 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
514 struct crypt_config *cc;
515 struct crypto_blkcipher *tfm;
521 unsigned int key_size;
522 unsigned long long tmpll;
525 ti->error = "Not enough arguments";
530 cipher = strsep(&tmp, "-");
531 chainmode = strsep(&tmp, "-");
532 ivopts = strsep(&tmp, "-");
533 ivmode = strsep(&ivopts, ":");
536 DMWARN("Unexpected additional cipher options");
538 key_size = strlen(argv[1]) >> 1;
540 cc = kmalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
543 "Cannot allocate transparent encryption context";
547 cc->key_size = key_size;
548 if ((!key_size && strcmp(argv[1], "-") != 0) ||
549 (key_size && crypt_decode_key(cc->key, argv[1], key_size) < 0)) {
550 ti->error = "Error decoding key";
554 /* Compatiblity mode for old dm-crypt cipher strings */
555 if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) {
560 if (strcmp(chainmode, "ecb") && !ivmode) {
561 ti->error = "This chaining mode requires an IV mechanism";
565 if (snprintf(cc->cipher, CRYPTO_MAX_ALG_NAME, "%s(%s)", chainmode,
566 cipher) >= CRYPTO_MAX_ALG_NAME) {
567 ti->error = "Chain mode + cipher name is too long";
571 tfm = crypto_alloc_blkcipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
573 ti->error = "Error allocating crypto tfm";
577 strcpy(cc->cipher, cipher);
578 strcpy(cc->chainmode, chainmode);
582 * Choose ivmode. Valid modes: "plain", "essiv:<esshash>".
583 * See comments at iv code
587 cc->iv_gen_ops = NULL;
588 else if (strcmp(ivmode, "plain") == 0)
589 cc->iv_gen_ops = &crypt_iv_plain_ops;
590 else if (strcmp(ivmode, "essiv") == 0)
591 cc->iv_gen_ops = &crypt_iv_essiv_ops;
593 ti->error = "Invalid IV mode";
597 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr &&
598 cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
601 cc->iv_size = crypto_blkcipher_ivsize(tfm);
603 /* at least a 64 bit sector number should fit in our buffer */
604 cc->iv_size = max(cc->iv_size,
605 (unsigned int)(sizeof(u64) / sizeof(u8)));
607 if (cc->iv_gen_ops) {
608 DMWARN("Selected cipher does not support IVs");
609 if (cc->iv_gen_ops->dtr)
610 cc->iv_gen_ops->dtr(cc);
611 cc->iv_gen_ops = NULL;
615 cc->io_pool = mempool_create_slab_pool(MIN_IOS, _crypt_io_pool);
617 ti->error = "Cannot allocate crypt io mempool";
621 cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
622 if (!cc->page_pool) {
623 ti->error = "Cannot allocate page mempool";
627 if (crypto_blkcipher_setkey(tfm, cc->key, key_size) < 0) {
628 ti->error = "Error setting key";
632 if (sscanf(argv[2], "%llu", &tmpll) != 1) {
633 ti->error = "Invalid iv_offset sector";
636 cc->iv_offset = tmpll;
638 if (sscanf(argv[4], "%llu", &tmpll) != 1) {
639 ti->error = "Invalid device sector";
644 if (dm_get_device(ti, argv[3], cc->start, ti->len,
645 dm_table_get_mode(ti->table), &cc->dev)) {
646 ti->error = "Device lookup failed";
650 if (ivmode && cc->iv_gen_ops) {
653 cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL);
655 ti->error = "Error kmallocing iv_mode string";
658 strcpy(cc->iv_mode, ivmode);
666 mempool_destroy(cc->page_pool);
668 mempool_destroy(cc->io_pool);
670 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
671 cc->iv_gen_ops->dtr(cc);
673 crypto_free_blkcipher(tfm);
675 /* Must zero key material before freeing */
676 memset(cc, 0, sizeof(*cc) + cc->key_size * sizeof(u8));
681 static void crypt_dtr(struct dm_target *ti)
683 struct crypt_config *cc = (struct crypt_config *) ti->private;
685 mempool_destroy(cc->page_pool);
686 mempool_destroy(cc->io_pool);
689 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
690 cc->iv_gen_ops->dtr(cc);
691 crypto_free_blkcipher(cc->tfm);
692 dm_put_device(ti, cc->dev);
694 /* Must zero key material before freeing */
695 memset(cc, 0, sizeof(*cc) + cc->key_size * sizeof(u8));
699 static int crypt_endio(struct bio *bio, unsigned int done, int error)
701 struct crypt_io *io = (struct crypt_io *) bio->bi_private;
702 struct crypt_config *cc = (struct crypt_config *) io->target->private;
704 if (bio_data_dir(bio) == WRITE) {
706 * free the processed pages, even if
707 * it's only a partially completed write
709 crypt_free_buffer_pages(cc, bio, done);
718 * successful reads are decrypted by the worker thread
720 if ((bio_data_dir(bio) == READ)
721 && bio_flagged(bio, BIO_UPTODATE)) {
722 kcryptd_queue_io(io);
726 dec_pending(io, error);
730 static inline struct bio *
731 crypt_clone(struct crypt_config *cc, struct crypt_io *io, struct bio *bio,
732 sector_t sector, unsigned int *bvec_idx,
733 struct convert_context *ctx)
737 if (bio_data_dir(bio) == WRITE) {
738 clone = crypt_alloc_buffer(cc, bio->bi_size,
739 io->first_clone, bvec_idx);
741 ctx->bio_out = clone;
742 if (crypt_convert(cc, ctx) < 0) {
743 crypt_free_buffer_pages(cc, clone,
751 * The block layer might modify the bvec array, so always
752 * copy the required bvecs because we need the original
753 * one in order to decrypt the whole bio data *afterwards*.
755 clone = bio_alloc(GFP_NOIO, bio_segments(bio));
758 clone->bi_vcnt = bio_segments(bio);
759 clone->bi_size = bio->bi_size;
760 memcpy(clone->bi_io_vec, bio_iovec(bio),
761 sizeof(struct bio_vec) * clone->bi_vcnt);
768 clone->bi_private = io;
769 clone->bi_end_io = crypt_endio;
770 clone->bi_bdev = cc->dev->bdev;
771 clone->bi_sector = cc->start + sector;
772 clone->bi_rw = bio->bi_rw;
777 static int crypt_map(struct dm_target *ti, struct bio *bio,
778 union map_info *map_context)
780 struct crypt_config *cc = (struct crypt_config *) ti->private;
781 struct crypt_io *io = mempool_alloc(cc->io_pool, GFP_NOIO);
782 struct convert_context ctx;
784 unsigned int remaining = bio->bi_size;
785 sector_t sector = bio->bi_sector - ti->begin;
786 unsigned int bvec_idx = 0;
790 io->first_clone = NULL;
792 atomic_set(&io->pending, 1); /* hold a reference */
794 if (bio_data_dir(bio) == WRITE)
795 crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
798 * The allocated buffers can be smaller than the whole bio,
799 * so repeat the whole process until all the data can be handled.
802 clone = crypt_clone(cc, io, bio, sector, &bvec_idx, &ctx);
806 if (!io->first_clone) {
808 * hold a reference to the first clone, because it
809 * holds the bio_vec array and that can't be freed
810 * before all other clones are released
813 io->first_clone = clone;
815 atomic_inc(&io->pending);
817 remaining -= clone->bi_size;
818 sector += bio_sectors(clone);
820 generic_make_request(clone);
822 /* out of memory -> run queues */
824 blk_congestion_wait(bio_data_dir(clone), HZ/100);
827 /* drop reference, clones could have returned before we reach this */
832 if (io->first_clone) {
833 dec_pending(io, -ENOMEM);
837 /* if no bio has been dispatched yet, we can directly return the error */
838 mempool_free(io, cc->io_pool);
842 static int crypt_status(struct dm_target *ti, status_type_t type,
843 char *result, unsigned int maxlen)
845 struct crypt_config *cc = (struct crypt_config *) ti->private;
847 const char *chainmode = NULL;
851 case STATUSTYPE_INFO:
855 case STATUSTYPE_TABLE:
856 cipher = crypto_blkcipher_name(cc->tfm);
858 chainmode = cc->chainmode;
861 DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
863 DMEMIT("%s-%s ", cipher, chainmode);
865 if (cc->key_size > 0) {
866 if ((maxlen - sz) < ((cc->key_size << 1) + 1))
869 crypt_encode_key(result + sz, cc->key, cc->key_size);
870 sz += cc->key_size << 1;
877 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
878 cc->dev->name, (unsigned long long)cc->start);
884 static struct target_type crypt_target = {
887 .module = THIS_MODULE,
891 .status = crypt_status,
894 static int __init dm_crypt_init(void)
898 _crypt_io_pool = kmem_cache_create("dm-crypt_io",
899 sizeof(struct crypt_io),
904 _kcryptd_workqueue = create_workqueue("kcryptd");
905 if (!_kcryptd_workqueue) {
907 DMERR("couldn't create kcryptd");
911 r = dm_register_target(&crypt_target);
913 DMERR("register failed %d", r);
920 destroy_workqueue(_kcryptd_workqueue);
922 kmem_cache_destroy(_crypt_io_pool);
926 static void __exit dm_crypt_exit(void)
928 int r = dm_unregister_target(&crypt_target);
931 DMERR("unregister failed %d", r);
933 destroy_workqueue(_kcryptd_workqueue);
934 kmem_cache_destroy(_crypt_io_pool);
937 module_init(dm_crypt_init);
938 module_exit(dm_crypt_exit);
941 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
942 MODULE_LICENSE("GPL");