2 * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
5 * This file is released under the GPL.
8 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/kernel.h>
11 #include <linux/bio.h>
12 #include <linux/blkdev.h>
13 #include <linux/mempool.h>
14 #include <linux/slab.h>
15 #include <linux/crypto.h>
16 #include <linux/workqueue.h>
17 #include <asm/atomic.h>
18 #include <asm/scatterlist.h>
26 * per bio private data
29 struct dm_target *target;
31 struct bio *first_clone;
32 struct work_struct work;
38 * context holding the current state of a multi-part conversion
40 struct convert_context {
43 unsigned int offset_in;
44 unsigned int offset_out;
53 struct crypt_iv_operations {
54 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
56 void (*dtr)(struct crypt_config *cc);
57 const char *(*status)(struct crypt_config *cc);
58 int (*generator)(struct crypt_config *cc, u8 *iv, sector_t sector);
62 * Crypt: maps a linear range of a block device
63 * and encrypts / decrypts at the same time.
70 * pool for per bio private data and
71 * for encryption buffer pages
79 struct crypt_iv_operations *iv_gen_ops;
85 struct crypto_tfm *tfm;
86 unsigned int key_size;
91 #define MIN_POOL_PAGES 32
92 #define MIN_BIO_PAGES 8
94 static kmem_cache_t *_crypt_io_pool;
97 * Mempool alloc and free functions for the page
99 static void *mempool_alloc_page(gfp_t gfp_mask, void *data)
101 return alloc_page(gfp_mask);
104 static void mempool_free_page(void *page, void *data)
111 * Different IV generation algorithms:
113 * plain: the initial vector is the 32-bit low-endian version of the sector
114 * number, padded with zeros if neccessary.
116 * ess_iv: "encrypted sector|salt initial vector", the sector number is
117 * encrypted with the bulk cipher using a salt as key. The salt
118 * should be derived from the bulk cipher's key via hashing.
120 * plumb: unimplemented, see:
121 * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
124 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
126 memset(iv, 0, cc->iv_size);
127 *(u32 *)iv = cpu_to_le32(sector & 0xffffffff);
132 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
135 struct crypto_tfm *essiv_tfm;
136 struct crypto_tfm *hash_tfm;
137 struct scatterlist sg;
138 unsigned int saltsize;
142 ti->error = PFX "Digest algorithm missing for ESSIV mode";
146 /* Hash the cipher key with the given hash algorithm */
147 hash_tfm = crypto_alloc_tfm(opts, CRYPTO_TFM_REQ_MAY_SLEEP);
148 if (hash_tfm == NULL) {
149 ti->error = PFX "Error initializing ESSIV hash";
153 if (crypto_tfm_alg_type(hash_tfm) != CRYPTO_ALG_TYPE_DIGEST) {
154 ti->error = PFX "Expected digest algorithm for ESSIV hash";
155 crypto_free_tfm(hash_tfm);
159 saltsize = crypto_tfm_alg_digestsize(hash_tfm);
160 salt = kmalloc(saltsize, GFP_KERNEL);
162 ti->error = PFX "Error kmallocing salt storage in ESSIV";
163 crypto_free_tfm(hash_tfm);
167 sg.page = virt_to_page(cc->key);
168 sg.offset = offset_in_page(cc->key);
169 sg.length = cc->key_size;
170 crypto_digest_digest(hash_tfm, &sg, 1, salt);
171 crypto_free_tfm(hash_tfm);
173 /* Setup the essiv_tfm with the given salt */
174 essiv_tfm = crypto_alloc_tfm(crypto_tfm_alg_name(cc->tfm),
175 CRYPTO_TFM_MODE_ECB |
176 CRYPTO_TFM_REQ_MAY_SLEEP);
177 if (essiv_tfm == NULL) {
178 ti->error = PFX "Error allocating crypto tfm for ESSIV";
182 if (crypto_tfm_alg_blocksize(essiv_tfm)
183 != crypto_tfm_alg_ivsize(cc->tfm)) {
184 ti->error = PFX "Block size of ESSIV cipher does "
185 "not match IV size of block cipher";
186 crypto_free_tfm(essiv_tfm);
190 if (crypto_cipher_setkey(essiv_tfm, salt, saltsize) < 0) {
191 ti->error = PFX "Failed to set key for ESSIV cipher";
192 crypto_free_tfm(essiv_tfm);
198 cc->iv_gen_private = (void *)essiv_tfm;
202 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
204 crypto_free_tfm((struct crypto_tfm *)cc->iv_gen_private);
205 cc->iv_gen_private = NULL;
208 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
210 struct scatterlist sg = { NULL, };
212 memset(iv, 0, cc->iv_size);
213 *(u64 *)iv = cpu_to_le64(sector);
215 sg.page = virt_to_page(iv);
216 sg.offset = offset_in_page(iv);
217 sg.length = cc->iv_size;
218 crypto_cipher_encrypt((struct crypto_tfm *)cc->iv_gen_private,
219 &sg, &sg, cc->iv_size);
224 static struct crypt_iv_operations crypt_iv_plain_ops = {
225 .generator = crypt_iv_plain_gen
228 static struct crypt_iv_operations crypt_iv_essiv_ops = {
229 .ctr = crypt_iv_essiv_ctr,
230 .dtr = crypt_iv_essiv_dtr,
231 .generator = crypt_iv_essiv_gen
236 crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
237 struct scatterlist *in, unsigned int length,
238 int write, sector_t sector)
243 if (cc->iv_gen_ops) {
244 r = cc->iv_gen_ops->generator(cc, iv, sector);
249 r = crypto_cipher_encrypt_iv(cc->tfm, out, in, length, iv);
251 r = crypto_cipher_decrypt_iv(cc->tfm, out, in, length, iv);
254 r = crypto_cipher_encrypt(cc->tfm, out, in, length);
256 r = crypto_cipher_decrypt(cc->tfm, out, in, length);
263 crypt_convert_init(struct crypt_config *cc, struct convert_context *ctx,
264 struct bio *bio_out, struct bio *bio_in,
265 sector_t sector, int write)
267 ctx->bio_in = bio_in;
268 ctx->bio_out = bio_out;
271 ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
272 ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
273 ctx->sector = sector + cc->iv_offset;
278 * Encrypt / decrypt data from one bio to another one (can be the same one)
280 static int crypt_convert(struct crypt_config *cc,
281 struct convert_context *ctx)
285 while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
286 ctx->idx_out < ctx->bio_out->bi_vcnt) {
287 struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
288 struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
289 struct scatterlist sg_in = {
290 .page = bv_in->bv_page,
291 .offset = bv_in->bv_offset + ctx->offset_in,
292 .length = 1 << SECTOR_SHIFT
294 struct scatterlist sg_out = {
295 .page = bv_out->bv_page,
296 .offset = bv_out->bv_offset + ctx->offset_out,
297 .length = 1 << SECTOR_SHIFT
300 ctx->offset_in += sg_in.length;
301 if (ctx->offset_in >= bv_in->bv_len) {
306 ctx->offset_out += sg_out.length;
307 if (ctx->offset_out >= bv_out->bv_len) {
312 r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length,
313 ctx->write, ctx->sector);
324 * Generate a new unfragmented bio with the given size
325 * This should never violate the device limitations
326 * May return a smaller bio when running out of pages
329 crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
330 struct bio *base_bio, unsigned int *bio_vec_idx)
333 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 int gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
338 * Use __GFP_NOMEMALLOC to tell the VM to act less aggressively and
339 * to fail earlier. This is not necessary but increases throughput.
340 * FIXME: Is this really intelligent?
343 bio = bio_clone(base_bio, GFP_NOIO|__GFP_NOMEMALLOC);
345 bio = bio_alloc(GFP_NOIO|__GFP_NOMEMALLOC, nr_iovecs);
349 /* if the last bio was not complete, continue where that one ended */
350 bio->bi_idx = *bio_vec_idx;
351 bio->bi_vcnt = *bio_vec_idx;
353 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
355 /* bio->bi_idx pages have already been allocated */
356 size -= bio->bi_idx * PAGE_SIZE;
358 for(i = bio->bi_idx; i < nr_iovecs; i++) {
359 struct bio_vec *bv = bio_iovec_idx(bio, i);
361 bv->bv_page = mempool_alloc(cc->page_pool, gfp_mask);
366 * if additional pages cannot be allocated without waiting,
367 * return a partially allocated bio, the caller will then try
368 * to allocate additional bios while submitting this partial bio
370 if ((i - bio->bi_idx) == (MIN_BIO_PAGES - 1))
371 gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
374 if (size > PAGE_SIZE)
375 bv->bv_len = PAGE_SIZE;
379 bio->bi_size += bv->bv_len;
390 * Remember the last bio_vec allocated to be able
391 * to correctly continue after the splitting.
393 *bio_vec_idx = bio->bi_vcnt;
398 static void crypt_free_buffer_pages(struct crypt_config *cc,
399 struct bio *bio, unsigned int bytes)
401 unsigned int i, start, end;
405 * This is ugly, but Jens Axboe thinks that using bi_idx in the
406 * endio function is too dangerous at the moment, so I calculate the
407 * correct position using bi_vcnt and bi_size.
408 * The bv_offset and bv_len fields might already be modified but we
409 * know that we always allocated whole pages.
410 * A fix to the bi_idx issue in the kernel is in the works, so
411 * we will hopefully be able to revert to the cleaner solution soon.
413 i = bio->bi_vcnt - 1;
414 bv = bio_iovec_idx(bio, i);
415 end = (i << PAGE_SHIFT) + (bv->bv_offset + bv->bv_len) - bio->bi_size;
418 start >>= PAGE_SHIFT;
424 for(i = start; i < end; i++) {
425 bv = bio_iovec_idx(bio, i);
426 BUG_ON(!bv->bv_page);
427 mempool_free(bv->bv_page, cc->page_pool);
433 * One of the bios was finished. Check for completion of
434 * the whole request and correctly clean up the buffer.
436 static void dec_pending(struct crypt_io *io, int error)
438 struct crypt_config *cc = (struct crypt_config *) io->target->private;
443 if (!atomic_dec_and_test(&io->pending))
447 bio_put(io->first_clone);
449 bio_endio(io->bio, io->bio->bi_size, io->error);
451 mempool_free(io, cc->io_pool);
457 * Needed because it would be very unwise to do decryption in an
458 * interrupt context, so bios returning from read requests get
461 static struct workqueue_struct *_kcryptd_workqueue;
463 static void kcryptd_do_work(void *data)
465 struct crypt_io *io = (struct crypt_io *) data;
466 struct crypt_config *cc = (struct crypt_config *) io->target->private;
467 struct convert_context ctx;
470 crypt_convert_init(cc, &ctx, io->bio, io->bio,
471 io->bio->bi_sector - io->target->begin, 0);
472 r = crypt_convert(cc, &ctx);
477 static void kcryptd_queue_io(struct crypt_io *io)
479 INIT_WORK(&io->work, kcryptd_do_work, io);
480 queue_work(_kcryptd_workqueue, &io->work);
484 * Decode key from its hex representation
486 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
494 for(i = 0; i < size; i++) {
498 key[i] = (u8)simple_strtoul(buffer, &endp, 16);
500 if (endp != &buffer[2])
511 * Encode key into its hex representation
513 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
517 for(i = 0; i < size; i++) {
518 sprintf(hex, "%02x", *key);
525 * Construct an encryption mapping:
526 * <cipher> <key> <iv_offset> <dev_path> <start>
528 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
530 struct crypt_config *cc;
531 struct crypto_tfm *tfm;
537 unsigned int crypto_flags;
538 unsigned int key_size;
541 ti->error = PFX "Not enough arguments";
546 cipher = strsep(&tmp, "-");
547 chainmode = strsep(&tmp, "-");
548 ivopts = strsep(&tmp, "-");
549 ivmode = strsep(&ivopts, ":");
552 DMWARN(PFX "Unexpected additional cipher options");
554 key_size = strlen(argv[1]) >> 1;
556 cc = kmalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
559 PFX "Cannot allocate transparent encryption context";
563 cc->key_size = key_size;
564 if ((!key_size && strcmp(argv[1], "-") != 0) ||
565 (key_size && crypt_decode_key(cc->key, argv[1], key_size) < 0)) {
566 ti->error = PFX "Error decoding key";
570 /* Compatiblity mode for old dm-crypt cipher strings */
571 if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) {
576 /* Choose crypto_flags according to chainmode */
577 if (strcmp(chainmode, "cbc") == 0)
578 crypto_flags = CRYPTO_TFM_MODE_CBC;
579 else if (strcmp(chainmode, "ecb") == 0)
580 crypto_flags = CRYPTO_TFM_MODE_ECB;
582 ti->error = PFX "Unknown chaining mode";
586 if (crypto_flags != CRYPTO_TFM_MODE_ECB && !ivmode) {
587 ti->error = PFX "This chaining mode requires an IV mechanism";
591 tfm = crypto_alloc_tfm(cipher, crypto_flags | CRYPTO_TFM_REQ_MAY_SLEEP);
593 ti->error = PFX "Error allocating crypto tfm";
596 if (crypto_tfm_alg_type(tfm) != CRYPTO_ALG_TYPE_CIPHER) {
597 ti->error = PFX "Expected cipher algorithm";
604 * Choose ivmode. Valid modes: "plain", "essiv:<esshash>".
605 * See comments at iv code
609 cc->iv_gen_ops = NULL;
610 else if (strcmp(ivmode, "plain") == 0)
611 cc->iv_gen_ops = &crypt_iv_plain_ops;
612 else if (strcmp(ivmode, "essiv") == 0)
613 cc->iv_gen_ops = &crypt_iv_essiv_ops;
615 ti->error = PFX "Invalid IV mode";
619 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr &&
620 cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
623 if (tfm->crt_cipher.cit_decrypt_iv && tfm->crt_cipher.cit_encrypt_iv)
624 /* at least a 64 bit sector number should fit in our buffer */
625 cc->iv_size = max(crypto_tfm_alg_ivsize(tfm),
626 (unsigned int)(sizeof(u64) / sizeof(u8)));
629 if (cc->iv_gen_ops) {
630 DMWARN(PFX "Selected cipher does not support IVs");
631 if (cc->iv_gen_ops->dtr)
632 cc->iv_gen_ops->dtr(cc);
633 cc->iv_gen_ops = NULL;
637 cc->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
638 mempool_free_slab, _crypt_io_pool);
640 ti->error = PFX "Cannot allocate crypt io mempool";
644 cc->page_pool = mempool_create(MIN_POOL_PAGES, mempool_alloc_page,
645 mempool_free_page, NULL);
646 if (!cc->page_pool) {
647 ti->error = PFX "Cannot allocate page mempool";
651 if (tfm->crt_cipher.cit_setkey(tfm, cc->key, key_size) < 0) {
652 ti->error = PFX "Error setting key";
656 if (sscanf(argv[2], SECTOR_FORMAT, &cc->iv_offset) != 1) {
657 ti->error = PFX "Invalid iv_offset sector";
661 if (sscanf(argv[4], SECTOR_FORMAT, &cc->start) != 1) {
662 ti->error = PFX "Invalid device sector";
666 if (dm_get_device(ti, argv[3], cc->start, ti->len,
667 dm_table_get_mode(ti->table), &cc->dev)) {
668 ti->error = PFX "Device lookup failed";
672 if (ivmode && cc->iv_gen_ops) {
675 cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL);
677 ti->error = PFX "Error kmallocing iv_mode string";
680 strcpy(cc->iv_mode, ivmode);
688 mempool_destroy(cc->page_pool);
690 mempool_destroy(cc->io_pool);
692 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
693 cc->iv_gen_ops->dtr(cc);
695 crypto_free_tfm(tfm);
701 static void crypt_dtr(struct dm_target *ti)
703 struct crypt_config *cc = (struct crypt_config *) ti->private;
705 mempool_destroy(cc->page_pool);
706 mempool_destroy(cc->io_pool);
709 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
710 cc->iv_gen_ops->dtr(cc);
711 crypto_free_tfm(cc->tfm);
712 dm_put_device(ti, cc->dev);
716 static int crypt_endio(struct bio *bio, unsigned int done, int error)
718 struct crypt_io *io = (struct crypt_io *) bio->bi_private;
719 struct crypt_config *cc = (struct crypt_config *) io->target->private;
721 if (bio_data_dir(bio) == WRITE) {
723 * free the processed pages, even if
724 * it's only a partially completed write
726 crypt_free_buffer_pages(cc, bio, done);
735 * successful reads are decrypted by the worker thread
737 if ((bio_data_dir(bio) == READ)
738 && bio_flagged(bio, BIO_UPTODATE)) {
739 kcryptd_queue_io(io);
743 dec_pending(io, error);
747 static inline struct bio *
748 crypt_clone(struct crypt_config *cc, struct crypt_io *io, struct bio *bio,
749 sector_t sector, unsigned int *bvec_idx,
750 struct convert_context *ctx)
754 if (bio_data_dir(bio) == WRITE) {
755 clone = crypt_alloc_buffer(cc, bio->bi_size,
756 io->first_clone, bvec_idx);
758 ctx->bio_out = clone;
759 if (crypt_convert(cc, ctx) < 0) {
760 crypt_free_buffer_pages(cc, clone,
768 * The block layer might modify the bvec array, so always
769 * copy the required bvecs because we need the original
770 * one in order to decrypt the whole bio data *afterwards*.
772 clone = bio_alloc(GFP_NOIO, bio_segments(bio));
775 clone->bi_vcnt = bio_segments(bio);
776 clone->bi_size = bio->bi_size;
777 memcpy(clone->bi_io_vec, bio_iovec(bio),
778 sizeof(struct bio_vec) * clone->bi_vcnt);
785 clone->bi_private = io;
786 clone->bi_end_io = crypt_endio;
787 clone->bi_bdev = cc->dev->bdev;
788 clone->bi_sector = cc->start + sector;
789 clone->bi_rw = bio->bi_rw;
794 static int crypt_map(struct dm_target *ti, struct bio *bio,
795 union map_info *map_context)
797 struct crypt_config *cc = (struct crypt_config *) ti->private;
798 struct crypt_io *io = mempool_alloc(cc->io_pool, GFP_NOIO);
799 struct convert_context ctx;
801 unsigned int remaining = bio->bi_size;
802 sector_t sector = bio->bi_sector - ti->begin;
803 unsigned int bvec_idx = 0;
807 io->first_clone = NULL;
809 atomic_set(&io->pending, 1); /* hold a reference */
811 if (bio_data_dir(bio) == WRITE)
812 crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
815 * The allocated buffers can be smaller than the whole bio,
816 * so repeat the whole process until all the data can be handled.
819 clone = crypt_clone(cc, io, bio, sector, &bvec_idx, &ctx);
823 if (!io->first_clone) {
825 * hold a reference to the first clone, because it
826 * holds the bio_vec array and that can't be freed
827 * before all other clones are released
830 io->first_clone = clone;
832 atomic_inc(&io->pending);
834 remaining -= clone->bi_size;
835 sector += bio_sectors(clone);
837 generic_make_request(clone);
839 /* out of memory -> run queues */
841 blk_congestion_wait(bio_data_dir(clone), HZ/100);
844 /* drop reference, clones could have returned before we reach this */
849 if (io->first_clone) {
850 dec_pending(io, -ENOMEM);
854 /* if no bio has been dispatched yet, we can directly return the error */
855 mempool_free(io, cc->io_pool);
859 static int crypt_status(struct dm_target *ti, status_type_t type,
860 char *result, unsigned int maxlen)
862 struct crypt_config *cc = (struct crypt_config *) ti->private;
864 const char *chainmode = NULL;
868 case STATUSTYPE_INFO:
872 case STATUSTYPE_TABLE:
873 cipher = crypto_tfm_alg_name(cc->tfm);
875 switch(cc->tfm->crt_cipher.cit_mode) {
876 case CRYPTO_TFM_MODE_CBC:
879 case CRYPTO_TFM_MODE_ECB:
887 DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
889 DMEMIT("%s-%s ", cipher, chainmode);
891 if (cc->key_size > 0) {
892 if ((maxlen - sz) < ((cc->key_size << 1) + 1))
895 crypt_encode_key(result + sz, cc->key, cc->key_size);
896 sz += cc->key_size << 1;
903 DMEMIT(" " SECTOR_FORMAT " %s " SECTOR_FORMAT,
904 cc->iv_offset, cc->dev->name, cc->start);
910 static struct target_type crypt_target = {
913 .module = THIS_MODULE,
917 .status = crypt_status,
920 static int __init dm_crypt_init(void)
924 _crypt_io_pool = kmem_cache_create("dm-crypt_io",
925 sizeof(struct crypt_io),
930 _kcryptd_workqueue = create_workqueue("kcryptd");
931 if (!_kcryptd_workqueue) {
933 DMERR(PFX "couldn't create kcryptd");
937 r = dm_register_target(&crypt_target);
939 DMERR(PFX "register failed %d", r);
946 destroy_workqueue(_kcryptd_workqueue);
948 kmem_cache_destroy(_crypt_io_pool);
952 static void __exit dm_crypt_exit(void)
954 int r = dm_unregister_target(&crypt_target);
957 DMERR(PFX "unregister failed %d", r);
959 destroy_workqueue(_kcryptd_workqueue);
960 kmem_cache_destroy(_crypt_io_pool);
963 module_init(dm_crypt_init);
964 module_exit(dm_crypt_exit);
966 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
967 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
968 MODULE_LICENSE("GPL");