2 * Copyright (C) 2012 Red Hat, Inc.
4 * Author: Mikulas Patocka <mpatocka@redhat.com>
6 * Based on Chromium dm-verity driver (C) 2011 The Chromium OS Authors
8 * This file is released under the GPLv2.
10 * In the file "/sys/module/dm_verity/parameters/prefetch_cluster" you can set
11 * default prefetch value. Data are read in "prefetch_cluster" chunks from the
12 * hash device. Setting this greatly improves performance when data and hash
13 * are on the same disk on different partitions on devices with poor random
19 #include <linux/module.h>
20 #include <linux/device-mapper.h>
21 #include <crypto/hash.h>
23 #define DM_MSG_PREFIX "verity"
25 #define DM_VERITY_IO_VEC_INLINE 16
26 #define DM_VERITY_MEMPOOL_SIZE 4
27 #define DM_VERITY_DEFAULT_PREFETCH_SIZE 262144
29 #define DM_VERITY_MAX_LEVELS 63
31 static unsigned dm_verity_prefetch_cluster = DM_VERITY_DEFAULT_PREFETCH_SIZE;
33 module_param_named(prefetch_cluster, dm_verity_prefetch_cluster, uint, S_IRUGO | S_IWUSR);
36 struct dm_dev *data_dev;
37 struct dm_dev *hash_dev;
39 struct dm_bufio_client *bufio;
41 struct crypto_shash *tfm;
42 u8 *root_digest; /* digest of the root block */
43 u8 *salt; /* salt: its size is salt_size */
45 sector_t data_start; /* data offset in 512-byte sectors */
46 sector_t hash_start; /* hash start in blocks */
47 sector_t data_blocks; /* the number of data blocks */
48 sector_t hash_blocks; /* the number of hash blocks */
49 unsigned char data_dev_block_bits; /* log2(data blocksize) */
50 unsigned char hash_dev_block_bits; /* log2(hash blocksize) */
51 unsigned char hash_per_block_bits; /* log2(hashes in hash block) */
52 unsigned char levels; /* the number of tree levels */
53 unsigned char version;
54 unsigned digest_size; /* digest size for the current hash algorithm */
55 unsigned shash_descsize;/* the size of temporary space for crypto */
56 int hash_failed; /* set to 1 if hash of any block failed */
58 mempool_t *vec_mempool; /* mempool of bio vector */
60 struct workqueue_struct *verify_wq;
62 /* starting blocks for each tree level. 0 is the lowest level. */
63 sector_t hash_level_block[DM_VERITY_MAX_LEVELS];
69 /* original values of bio->bi_end_io and bio->bi_private */
70 bio_end_io_t *orig_bi_end_io;
71 void *orig_bi_private;
76 /* saved bio vector */
77 struct bio_vec *io_vec;
80 struct work_struct work;
82 /* A space for short vectors; longer vectors are allocated separately. */
83 struct bio_vec io_vec_inline[DM_VERITY_IO_VEC_INLINE];
86 * Three variably-size fields follow this struct:
88 * u8 hash_desc[v->shash_descsize];
89 * u8 real_digest[v->digest_size];
90 * u8 want_digest[v->digest_size];
92 * To access them use: io_hash_desc(), io_real_digest() and io_want_digest().
96 static struct shash_desc *io_hash_desc(struct dm_verity *v, struct dm_verity_io *io)
98 return (struct shash_desc *)(io + 1);
101 static u8 *io_real_digest(struct dm_verity *v, struct dm_verity_io *io)
103 return (u8 *)(io + 1) + v->shash_descsize;
106 static u8 *io_want_digest(struct dm_verity *v, struct dm_verity_io *io)
108 return (u8 *)(io + 1) + v->shash_descsize + v->digest_size;
112 * Auxiliary structure appended to each dm-bufio buffer. If the value
113 * hash_verified is nonzero, hash of the block has been verified.
115 * The variable hash_verified is set to 0 when allocating the buffer, then
116 * it can be changed to 1 and it is never reset to 0 again.
118 * There is no lock around this value, a race condition can at worst cause
119 * that multiple processes verify the hash of the same buffer simultaneously
120 * and write 1 to hash_verified simultaneously.
121 * This condition is harmless, so we don't need locking.
128 * Initialize struct buffer_aux for a freshly created buffer.
130 static void dm_bufio_alloc_callback(struct dm_buffer *buf)
132 struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
134 aux->hash_verified = 0;
138 * Translate input sector number to the sector number on the target device.
140 static sector_t verity_map_sector(struct dm_verity *v, sector_t bi_sector)
142 return v->data_start + dm_target_offset(v->ti, bi_sector);
146 * Return hash position of a specified block at a specified tree level
147 * (0 is the lowest level).
148 * The lowest "hash_per_block_bits"-bits of the result denote hash position
149 * inside a hash block. The remaining bits denote location of the hash block.
151 static sector_t verity_position_at_level(struct dm_verity *v, sector_t block,
154 return block >> (level * v->hash_per_block_bits);
157 static void verity_hash_at_level(struct dm_verity *v, sector_t block, int level,
158 sector_t *hash_block, unsigned *offset)
160 sector_t position = verity_position_at_level(v, block, level);
163 *hash_block = v->hash_level_block[level] + (position >> v->hash_per_block_bits);
168 idx = position & ((1 << v->hash_per_block_bits) - 1);
170 *offset = idx * v->digest_size;
172 *offset = idx << (v->hash_dev_block_bits - v->hash_per_block_bits);
176 * Verify hash of a metadata block pertaining to the specified data block
177 * ("block" argument) at a specified level ("level" argument).
179 * On successful return, io_want_digest(v, io) contains the hash value for
180 * a lower tree level or for the data block (if we're at the lowest leve).
182 * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned.
183 * If "skip_unverified" is false, unverified buffer is hashed and verified
184 * against current value of io_want_digest(v, io).
186 static int verity_verify_level(struct dm_verity_io *io, sector_t block,
187 int level, bool skip_unverified)
189 struct dm_verity *v = io->v;
190 struct dm_buffer *buf;
191 struct buffer_aux *aux;
197 verity_hash_at_level(v, block, level, &hash_block, &offset);
199 data = dm_bufio_read(v->bufio, hash_block, &buf);
200 if (unlikely(IS_ERR(data)))
201 return PTR_ERR(data);
203 aux = dm_bufio_get_aux_data(buf);
205 if (!aux->hash_verified) {
206 struct shash_desc *desc;
209 if (skip_unverified) {
214 desc = io_hash_desc(v, io);
216 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
217 r = crypto_shash_init(desc);
219 DMERR("crypto_shash_init failed: %d", r);
223 if (likely(v->version >= 1)) {
224 r = crypto_shash_update(desc, v->salt, v->salt_size);
226 DMERR("crypto_shash_update failed: %d", r);
231 r = crypto_shash_update(desc, data, 1 << v->hash_dev_block_bits);
233 DMERR("crypto_shash_update failed: %d", r);
238 r = crypto_shash_update(desc, v->salt, v->salt_size);
240 DMERR("crypto_shash_update failed: %d", r);
245 result = io_real_digest(v, io);
246 r = crypto_shash_final(desc, result);
248 DMERR("crypto_shash_final failed: %d", r);
251 if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
252 DMERR_LIMIT("metadata block %llu is corrupted",
253 (unsigned long long)hash_block);
258 aux->hash_verified = 1;
263 memcpy(io_want_digest(v, io), data, v->digest_size);
265 dm_bufio_release(buf);
269 dm_bufio_release(buf);
275 * Verify one "dm_verity_io" structure.
277 static int verity_verify_io(struct dm_verity_io *io)
279 struct dm_verity *v = io->v;
282 unsigned vector = 0, offset = 0;
284 for (b = 0; b < io->n_blocks; b++) {
285 struct shash_desc *desc;
290 if (likely(v->levels)) {
292 * First, we try to get the requested hash for
293 * the current block. If the hash block itself is
294 * verified, zero is returned. If it isn't, this
295 * function returns 0 and we fall back to whole
296 * chain verification.
298 int r = verity_verify_level(io, io->block + b, 0, true);
300 goto test_block_hash;
305 memcpy(io_want_digest(v, io), v->root_digest, v->digest_size);
307 for (i = v->levels - 1; i >= 0; i--) {
308 int r = verity_verify_level(io, io->block + b, i, false);
314 desc = io_hash_desc(v, io);
316 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
317 r = crypto_shash_init(desc);
319 DMERR("crypto_shash_init failed: %d", r);
323 if (likely(v->version >= 1)) {
324 r = crypto_shash_update(desc, v->salt, v->salt_size);
326 DMERR("crypto_shash_update failed: %d", r);
331 todo = 1 << v->data_dev_block_bits;
337 BUG_ON(vector >= io->io_vec_size);
338 bv = &io->io_vec[vector];
339 page = kmap_atomic(bv->bv_page);
340 len = bv->bv_len - offset;
341 if (likely(len >= todo))
343 r = crypto_shash_update(desc,
344 page + bv->bv_offset + offset, len);
347 DMERR("crypto_shash_update failed: %d", r);
351 if (likely(offset == bv->bv_len)) {
359 r = crypto_shash_update(desc, v->salt, v->salt_size);
361 DMERR("crypto_shash_update failed: %d", r);
366 result = io_real_digest(v, io);
367 r = crypto_shash_final(desc, result);
369 DMERR("crypto_shash_final failed: %d", r);
372 if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
373 DMERR_LIMIT("data block %llu is corrupted",
374 (unsigned long long)(io->block + b));
379 BUG_ON(vector != io->io_vec_size);
386 * End one "io" structure with a given error.
388 static void verity_finish_io(struct dm_verity_io *io, int error)
390 struct dm_verity *v = io->v;
391 struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_bio_data_size);
393 bio->bi_end_io = io->orig_bi_end_io;
394 bio->bi_private = io->orig_bi_private;
396 if (io->io_vec != io->io_vec_inline)
397 mempool_free(io->io_vec, v->vec_mempool);
399 bio_endio(bio, error);
402 static void verity_work(struct work_struct *w)
404 struct dm_verity_io *io = container_of(w, struct dm_verity_io, work);
406 verity_finish_io(io, verity_verify_io(io));
409 static void verity_end_io(struct bio *bio, int error)
411 struct dm_verity_io *io = bio->bi_private;
414 verity_finish_io(io, error);
418 INIT_WORK(&io->work, verity_work);
419 queue_work(io->v->verify_wq, &io->work);
423 * Prefetch buffers for the specified io.
424 * The root buffer is not prefetched, it is assumed that it will be cached
427 static void verity_prefetch_io(struct dm_verity *v, struct dm_verity_io *io)
431 for (i = v->levels - 2; i >= 0; i--) {
432 sector_t hash_block_start;
433 sector_t hash_block_end;
434 verity_hash_at_level(v, io->block, i, &hash_block_start, NULL);
435 verity_hash_at_level(v, io->block + io->n_blocks - 1, i, &hash_block_end, NULL);
437 unsigned cluster = ACCESS_ONCE(dm_verity_prefetch_cluster);
439 cluster >>= v->data_dev_block_bits;
440 if (unlikely(!cluster))
441 goto no_prefetch_cluster;
443 if (unlikely(cluster & (cluster - 1)))
444 cluster = 1 << (fls(cluster) - 1);
446 hash_block_start &= ~(sector_t)(cluster - 1);
447 hash_block_end |= cluster - 1;
448 if (unlikely(hash_block_end >= v->hash_blocks))
449 hash_block_end = v->hash_blocks - 1;
452 dm_bufio_prefetch(v->bufio, hash_block_start,
453 hash_block_end - hash_block_start + 1);
458 * Bio map function. It allocates dm_verity_io structure and bio vector and
459 * fills them. Then it issues prefetches and the I/O.
461 static int verity_map(struct dm_target *ti, struct bio *bio)
463 struct dm_verity *v = ti->private;
464 struct dm_verity_io *io;
466 bio->bi_bdev = v->data_dev->bdev;
467 bio->bi_sector = verity_map_sector(v, bio->bi_sector);
469 if (((unsigned)bio->bi_sector | bio_sectors(bio)) &
470 ((1 << (v->data_dev_block_bits - SECTOR_SHIFT)) - 1)) {
471 DMERR_LIMIT("unaligned io");
475 if ((bio->bi_sector + bio_sectors(bio)) >>
476 (v->data_dev_block_bits - SECTOR_SHIFT) > v->data_blocks) {
477 DMERR_LIMIT("io out of range");
481 if (bio_data_dir(bio) == WRITE)
484 io = dm_per_bio_data(bio, ti->per_bio_data_size);
486 io->orig_bi_end_io = bio->bi_end_io;
487 io->orig_bi_private = bio->bi_private;
488 io->block = bio->bi_sector >> (v->data_dev_block_bits - SECTOR_SHIFT);
489 io->n_blocks = bio->bi_size >> v->data_dev_block_bits;
491 bio->bi_end_io = verity_end_io;
492 bio->bi_private = io;
493 io->io_vec_size = bio->bi_vcnt - bio->bi_idx;
494 if (io->io_vec_size < DM_VERITY_IO_VEC_INLINE)
495 io->io_vec = io->io_vec_inline;
497 io->io_vec = mempool_alloc(v->vec_mempool, GFP_NOIO);
498 memcpy(io->io_vec, bio_iovec(bio),
499 io->io_vec_size * sizeof(struct bio_vec));
501 verity_prefetch_io(v, io);
503 generic_make_request(bio);
505 return DM_MAPIO_SUBMITTED;
509 * Status: V (valid) or C (corruption found)
511 static int verity_status(struct dm_target *ti, status_type_t type,
512 unsigned status_flags, char *result, unsigned maxlen)
514 struct dm_verity *v = ti->private;
519 case STATUSTYPE_INFO:
520 DMEMIT("%c", v->hash_failed ? 'C' : 'V');
522 case STATUSTYPE_TABLE:
523 DMEMIT("%u %s %s %u %u %llu %llu %s ",
527 1 << v->data_dev_block_bits,
528 1 << v->hash_dev_block_bits,
529 (unsigned long long)v->data_blocks,
530 (unsigned long long)v->hash_start,
533 for (x = 0; x < v->digest_size; x++)
534 DMEMIT("%02x", v->root_digest[x]);
539 for (x = 0; x < v->salt_size; x++)
540 DMEMIT("%02x", v->salt[x]);
547 static int verity_ioctl(struct dm_target *ti, unsigned cmd,
550 struct dm_verity *v = ti->private;
554 ti->len != i_size_read(v->data_dev->bdev->bd_inode) >> SECTOR_SHIFT)
555 r = scsi_verify_blk_ioctl(NULL, cmd);
557 return r ? : __blkdev_driver_ioctl(v->data_dev->bdev, v->data_dev->mode,
561 static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
562 struct bio_vec *biovec, int max_size)
564 struct dm_verity *v = ti->private;
565 struct request_queue *q = bdev_get_queue(v->data_dev->bdev);
567 if (!q->merge_bvec_fn)
570 bvm->bi_bdev = v->data_dev->bdev;
571 bvm->bi_sector = verity_map_sector(v, bvm->bi_sector);
573 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
576 static int verity_iterate_devices(struct dm_target *ti,
577 iterate_devices_callout_fn fn, void *data)
579 struct dm_verity *v = ti->private;
581 return fn(ti, v->data_dev, v->data_start, ti->len, data);
584 static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)
586 struct dm_verity *v = ti->private;
588 if (limits->logical_block_size < 1 << v->data_dev_block_bits)
589 limits->logical_block_size = 1 << v->data_dev_block_bits;
591 if (limits->physical_block_size < 1 << v->data_dev_block_bits)
592 limits->physical_block_size = 1 << v->data_dev_block_bits;
594 blk_limits_io_min(limits, limits->logical_block_size);
597 static void verity_dtr(struct dm_target *ti)
599 struct dm_verity *v = ti->private;
602 destroy_workqueue(v->verify_wq);
605 mempool_destroy(v->vec_mempool);
608 dm_bufio_client_destroy(v->bufio);
611 kfree(v->root_digest);
614 crypto_free_shash(v->tfm);
619 dm_put_device(ti, v->hash_dev);
622 dm_put_device(ti, v->data_dev);
629 * <version> The current format is version 1.
630 * Vsn 0 is compatible with original Chromium OS releases.
635 * <the number of data blocks>
639 * <salt> Hex string or "-" if no salt.
641 static int verity_ctr(struct dm_target *ti, unsigned argc, char **argv)
645 unsigned long long num_ll;
648 sector_t hash_position;
651 v = kzalloc(sizeof(struct dm_verity), GFP_KERNEL);
653 ti->error = "Cannot allocate verity structure";
659 if ((dm_table_get_mode(ti->table) & ~FMODE_READ)) {
660 ti->error = "Device must be readonly";
666 ti->error = "Invalid argument count: exactly 10 arguments required";
671 if (sscanf(argv[0], "%d%c", &num, &dummy) != 1 ||
672 num < 0 || num > 1) {
673 ti->error = "Invalid version";
679 r = dm_get_device(ti, argv[1], FMODE_READ, &v->data_dev);
681 ti->error = "Data device lookup failed";
685 r = dm_get_device(ti, argv[2], FMODE_READ, &v->hash_dev);
687 ti->error = "Data device lookup failed";
691 if (sscanf(argv[3], "%u%c", &num, &dummy) != 1 ||
692 !num || (num & (num - 1)) ||
693 num < bdev_logical_block_size(v->data_dev->bdev) ||
695 ti->error = "Invalid data device block size";
699 v->data_dev_block_bits = ffs(num) - 1;
701 if (sscanf(argv[4], "%u%c", &num, &dummy) != 1 ||
702 !num || (num & (num - 1)) ||
703 num < bdev_logical_block_size(v->hash_dev->bdev) ||
705 ti->error = "Invalid hash device block size";
709 v->hash_dev_block_bits = ffs(num) - 1;
711 if (sscanf(argv[5], "%llu%c", &num_ll, &dummy) != 1 ||
712 (sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
713 >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll) {
714 ti->error = "Invalid data blocks";
718 v->data_blocks = num_ll;
720 if (ti->len > (v->data_blocks << (v->data_dev_block_bits - SECTOR_SHIFT))) {
721 ti->error = "Data device is too small";
726 if (sscanf(argv[6], "%llu%c", &num_ll, &dummy) != 1 ||
727 (sector_t)(num_ll << (v->hash_dev_block_bits - SECTOR_SHIFT))
728 >> (v->hash_dev_block_bits - SECTOR_SHIFT) != num_ll) {
729 ti->error = "Invalid hash start";
733 v->hash_start = num_ll;
735 v->alg_name = kstrdup(argv[7], GFP_KERNEL);
737 ti->error = "Cannot allocate algorithm name";
742 v->tfm = crypto_alloc_shash(v->alg_name, 0, 0);
743 if (IS_ERR(v->tfm)) {
744 ti->error = "Cannot initialize hash function";
749 v->digest_size = crypto_shash_digestsize(v->tfm);
750 if ((1 << v->hash_dev_block_bits) < v->digest_size * 2) {
751 ti->error = "Digest size too big";
756 sizeof(struct shash_desc) + crypto_shash_descsize(v->tfm);
758 v->root_digest = kmalloc(v->digest_size, GFP_KERNEL);
759 if (!v->root_digest) {
760 ti->error = "Cannot allocate root digest";
764 if (strlen(argv[8]) != v->digest_size * 2 ||
765 hex2bin(v->root_digest, argv[8], v->digest_size)) {
766 ti->error = "Invalid root digest";
771 if (strcmp(argv[9], "-")) {
772 v->salt_size = strlen(argv[9]) / 2;
773 v->salt = kmalloc(v->salt_size, GFP_KERNEL);
775 ti->error = "Cannot allocate salt";
779 if (strlen(argv[9]) != v->salt_size * 2 ||
780 hex2bin(v->salt, argv[9], v->salt_size)) {
781 ti->error = "Invalid salt";
787 v->hash_per_block_bits =
788 fls((1 << v->hash_dev_block_bits) / v->digest_size) - 1;
792 while (v->hash_per_block_bits * v->levels < 64 &&
793 (unsigned long long)(v->data_blocks - 1) >>
794 (v->hash_per_block_bits * v->levels))
797 if (v->levels > DM_VERITY_MAX_LEVELS) {
798 ti->error = "Too many tree levels";
803 hash_position = v->hash_start;
804 for (i = v->levels - 1; i >= 0; i--) {
806 v->hash_level_block[i] = hash_position;
807 s = verity_position_at_level(v, v->data_blocks, i);
808 s = (s >> v->hash_per_block_bits) +
809 !!(s & ((1 << v->hash_per_block_bits) - 1));
810 if (hash_position + s < hash_position) {
811 ti->error = "Hash device offset overflow";
817 v->hash_blocks = hash_position;
819 v->bufio = dm_bufio_client_create(v->hash_dev->bdev,
820 1 << v->hash_dev_block_bits, 1, sizeof(struct buffer_aux),
821 dm_bufio_alloc_callback, NULL);
822 if (IS_ERR(v->bufio)) {
823 ti->error = "Cannot initialize dm-bufio";
824 r = PTR_ERR(v->bufio);
829 if (dm_bufio_get_device_size(v->bufio) < v->hash_blocks) {
830 ti->error = "Hash device is too small";
835 ti->per_bio_data_size = roundup(sizeof(struct dm_verity_io) + v->shash_descsize + v->digest_size * 2, __alignof__(struct dm_verity_io));
837 v->vec_mempool = mempool_create_kmalloc_pool(DM_VERITY_MEMPOOL_SIZE,
838 BIO_MAX_PAGES * sizeof(struct bio_vec));
839 if (!v->vec_mempool) {
840 ti->error = "Cannot allocate vector mempool";
845 /* WQ_UNBOUND greatly improves performance when running on ramdisk */
846 v->verify_wq = alloc_workqueue("kverityd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, num_online_cpus());
848 ti->error = "Cannot allocate workqueue";
861 static struct target_type verity_target = {
863 .version = {1, 1, 0},
864 .module = THIS_MODULE,
868 .status = verity_status,
869 .ioctl = verity_ioctl,
870 .merge = verity_merge,
871 .iterate_devices = verity_iterate_devices,
872 .io_hints = verity_io_hints,
875 static int __init dm_verity_init(void)
879 r = dm_register_target(&verity_target);
881 DMERR("register failed %d", r);
886 static void __exit dm_verity_exit(void)
888 dm_unregister_target(&verity_target);
891 module_init(dm_verity_init);
892 module_exit(dm_verity_exit);
894 MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>");
895 MODULE_AUTHOR("Mandeep Baines <msb@chromium.org>");
896 MODULE_AUTHOR("Will Drewry <wad@chromium.org>");
897 MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
898 MODULE_LICENSE("GPL");