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 *io_mempool; /* mempool of struct dm_verity_io */
59 mempool_t *vec_mempool; /* mempool of bio vector */
61 struct workqueue_struct *verify_wq;
63 /* starting blocks for each tree level. 0 is the lowest level. */
64 sector_t hash_level_block[DM_VERITY_MAX_LEVELS];
71 /* original values of bio->bi_end_io and bio->bi_private */
72 bio_end_io_t *orig_bi_end_io;
73 void *orig_bi_private;
78 /* saved bio vector */
79 struct bio_vec *io_vec;
82 struct work_struct work;
84 /* A space for short vectors; longer vectors are allocated separately. */
85 struct bio_vec io_vec_inline[DM_VERITY_IO_VEC_INLINE];
88 * Three variably-size fields follow this struct:
90 * u8 hash_desc[v->shash_descsize];
91 * u8 real_digest[v->digest_size];
92 * u8 want_digest[v->digest_size];
94 * To access them use: io_hash_desc(), io_real_digest() and io_want_digest().
98 static struct shash_desc *io_hash_desc(struct dm_verity *v, struct dm_verity_io *io)
100 return (struct shash_desc *)(io + 1);
103 static u8 *io_real_digest(struct dm_verity *v, struct dm_verity_io *io)
105 return (u8 *)(io + 1) + v->shash_descsize;
108 static u8 *io_want_digest(struct dm_verity *v, struct dm_verity_io *io)
110 return (u8 *)(io + 1) + v->shash_descsize + v->digest_size;
114 * Auxiliary structure appended to each dm-bufio buffer. If the value
115 * hash_verified is nonzero, hash of the block has been verified.
117 * The variable hash_verified is set to 0 when allocating the buffer, then
118 * it can be changed to 1 and it is never reset to 0 again.
120 * There is no lock around this value, a race condition can at worst cause
121 * that multiple processes verify the hash of the same buffer simultaneously
122 * and write 1 to hash_verified simultaneously.
123 * This condition is harmless, so we don't need locking.
130 * Initialize struct buffer_aux for a freshly created buffer.
132 static void dm_bufio_alloc_callback(struct dm_buffer *buf)
134 struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
136 aux->hash_verified = 0;
140 * Translate input sector number to the sector number on the target device.
142 static sector_t verity_map_sector(struct dm_verity *v, sector_t bi_sector)
144 return v->data_start + dm_target_offset(v->ti, bi_sector);
148 * Return hash position of a specified block at a specified tree level
149 * (0 is the lowest level).
150 * The lowest "hash_per_block_bits"-bits of the result denote hash position
151 * inside a hash block. The remaining bits denote location of the hash block.
153 static sector_t verity_position_at_level(struct dm_verity *v, sector_t block,
156 return block >> (level * v->hash_per_block_bits);
159 static void verity_hash_at_level(struct dm_verity *v, sector_t block, int level,
160 sector_t *hash_block, unsigned *offset)
162 sector_t position = verity_position_at_level(v, block, level);
165 *hash_block = v->hash_level_block[level] + (position >> v->hash_per_block_bits);
170 idx = position & ((1 << v->hash_per_block_bits) - 1);
172 *offset = idx * v->digest_size;
174 *offset = idx << (v->hash_dev_block_bits - v->hash_per_block_bits);
178 * Verify hash of a metadata block pertaining to the specified data block
179 * ("block" argument) at a specified level ("level" argument).
181 * On successful return, io_want_digest(v, io) contains the hash value for
182 * a lower tree level or for the data block (if we're at the lowest leve).
184 * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned.
185 * If "skip_unverified" is false, unverified buffer is hashed and verified
186 * against current value of io_want_digest(v, io).
188 static int verity_verify_level(struct dm_verity_io *io, sector_t block,
189 int level, bool skip_unverified)
191 struct dm_verity *v = io->v;
192 struct dm_buffer *buf;
193 struct buffer_aux *aux;
199 verity_hash_at_level(v, block, level, &hash_block, &offset);
201 data = dm_bufio_read(v->bufio, hash_block, &buf);
202 if (unlikely(IS_ERR(data)))
203 return PTR_ERR(data);
205 aux = dm_bufio_get_aux_data(buf);
207 if (!aux->hash_verified) {
208 struct shash_desc *desc;
211 if (skip_unverified) {
216 desc = io_hash_desc(v, io);
218 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
219 r = crypto_shash_init(desc);
221 DMERR("crypto_shash_init failed: %d", r);
225 if (likely(v->version >= 1)) {
226 r = crypto_shash_update(desc, v->salt, v->salt_size);
228 DMERR("crypto_shash_update failed: %d", r);
233 r = crypto_shash_update(desc, data, 1 << v->hash_dev_block_bits);
235 DMERR("crypto_shash_update failed: %d", r);
240 r = crypto_shash_update(desc, v->salt, v->salt_size);
242 DMERR("crypto_shash_update failed: %d", r);
247 result = io_real_digest(v, io);
248 r = crypto_shash_final(desc, result);
250 DMERR("crypto_shash_final failed: %d", r);
253 if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
254 DMERR_LIMIT("metadata block %llu is corrupted",
255 (unsigned long long)hash_block);
260 aux->hash_verified = 1;
265 memcpy(io_want_digest(v, io), data, v->digest_size);
267 dm_bufio_release(buf);
271 dm_bufio_release(buf);
277 * Verify one "dm_verity_io" structure.
279 static int verity_verify_io(struct dm_verity_io *io)
281 struct dm_verity *v = io->v;
284 unsigned vector = 0, offset = 0;
286 for (b = 0; b < io->n_blocks; b++) {
287 struct shash_desc *desc;
292 if (likely(v->levels)) {
294 * First, we try to get the requested hash for
295 * the current block. If the hash block itself is
296 * verified, zero is returned. If it isn't, this
297 * function returns 0 and we fall back to whole
298 * chain verification.
300 int r = verity_verify_level(io, io->block + b, 0, true);
302 goto test_block_hash;
307 memcpy(io_want_digest(v, io), v->root_digest, v->digest_size);
309 for (i = v->levels - 1; i >= 0; i--) {
310 int r = verity_verify_level(io, io->block + b, i, false);
316 desc = io_hash_desc(v, io);
318 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
319 r = crypto_shash_init(desc);
321 DMERR("crypto_shash_init failed: %d", r);
325 if (likely(v->version >= 1)) {
326 r = crypto_shash_update(desc, v->salt, v->salt_size);
328 DMERR("crypto_shash_update failed: %d", r);
333 todo = 1 << v->data_dev_block_bits;
339 BUG_ON(vector >= io->io_vec_size);
340 bv = &io->io_vec[vector];
341 page = kmap_atomic(bv->bv_page);
342 len = bv->bv_len - offset;
343 if (likely(len >= todo))
345 r = crypto_shash_update(desc,
346 page + bv->bv_offset + offset, len);
349 DMERR("crypto_shash_update failed: %d", r);
353 if (likely(offset == bv->bv_len)) {
361 r = crypto_shash_update(desc, v->salt, v->salt_size);
363 DMERR("crypto_shash_update failed: %d", r);
368 result = io_real_digest(v, io);
369 r = crypto_shash_final(desc, result);
371 DMERR("crypto_shash_final failed: %d", r);
374 if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
375 DMERR_LIMIT("data block %llu is corrupted",
376 (unsigned long long)(io->block + b));
381 BUG_ON(vector != io->io_vec_size);
388 * End one "io" structure with a given error.
390 static void verity_finish_io(struct dm_verity_io *io, int error)
392 struct bio *bio = io->bio;
393 struct dm_verity *v = io->v;
395 bio->bi_end_io = io->orig_bi_end_io;
396 bio->bi_private = io->orig_bi_private;
398 if (io->io_vec != io->io_vec_inline)
399 mempool_free(io->io_vec, v->vec_mempool);
401 mempool_free(io, v->io_mempool);
403 bio_endio(bio, error);
406 static void verity_work(struct work_struct *w)
408 struct dm_verity_io *io = container_of(w, struct dm_verity_io, work);
410 verity_finish_io(io, verity_verify_io(io));
413 static void verity_end_io(struct bio *bio, int error)
415 struct dm_verity_io *io = bio->bi_private;
418 verity_finish_io(io, error);
422 INIT_WORK(&io->work, verity_work);
423 queue_work(io->v->verify_wq, &io->work);
427 * Prefetch buffers for the specified io.
428 * The root buffer is not prefetched, it is assumed that it will be cached
431 static void verity_prefetch_io(struct dm_verity *v, struct dm_verity_io *io)
435 for (i = v->levels - 2; i >= 0; i--) {
436 sector_t hash_block_start;
437 sector_t hash_block_end;
438 verity_hash_at_level(v, io->block, i, &hash_block_start, NULL);
439 verity_hash_at_level(v, io->block + io->n_blocks - 1, i, &hash_block_end, NULL);
441 unsigned cluster = ACCESS_ONCE(dm_verity_prefetch_cluster);
443 cluster >>= v->data_dev_block_bits;
444 if (unlikely(!cluster))
445 goto no_prefetch_cluster;
447 if (unlikely(cluster & (cluster - 1)))
448 cluster = 1 << (fls(cluster) - 1);
450 hash_block_start &= ~(sector_t)(cluster - 1);
451 hash_block_end |= cluster - 1;
452 if (unlikely(hash_block_end >= v->hash_blocks))
453 hash_block_end = v->hash_blocks - 1;
456 dm_bufio_prefetch(v->bufio, hash_block_start,
457 hash_block_end - hash_block_start + 1);
462 * Bio map function. It allocates dm_verity_io structure and bio vector and
463 * fills them. Then it issues prefetches and the I/O.
465 static int verity_map(struct dm_target *ti, struct bio *bio,
466 union map_info *map_context)
468 struct dm_verity *v = ti->private;
469 struct dm_verity_io *io;
471 bio->bi_bdev = v->data_dev->bdev;
472 bio->bi_sector = verity_map_sector(v, bio->bi_sector);
474 if (((unsigned)bio->bi_sector | bio_sectors(bio)) &
475 ((1 << (v->data_dev_block_bits - SECTOR_SHIFT)) - 1)) {
476 DMERR_LIMIT("unaligned io");
480 if ((bio->bi_sector + bio_sectors(bio)) >>
481 (v->data_dev_block_bits - SECTOR_SHIFT) > v->data_blocks) {
482 DMERR_LIMIT("io out of range");
486 if (bio_data_dir(bio) == WRITE)
489 io = mempool_alloc(v->io_mempool, GFP_NOIO);
492 io->orig_bi_end_io = bio->bi_end_io;
493 io->orig_bi_private = bio->bi_private;
494 io->block = bio->bi_sector >> (v->data_dev_block_bits - SECTOR_SHIFT);
495 io->n_blocks = bio->bi_size >> v->data_dev_block_bits;
497 bio->bi_end_io = verity_end_io;
498 bio->bi_private = io;
499 io->io_vec_size = bio->bi_vcnt - bio->bi_idx;
500 if (io->io_vec_size < DM_VERITY_IO_VEC_INLINE)
501 io->io_vec = io->io_vec_inline;
503 io->io_vec = mempool_alloc(v->vec_mempool, GFP_NOIO);
504 memcpy(io->io_vec, bio_iovec(bio),
505 io->io_vec_size * sizeof(struct bio_vec));
507 verity_prefetch_io(v, io);
509 generic_make_request(bio);
511 return DM_MAPIO_SUBMITTED;
515 * Status: V (valid) or C (corruption found)
517 static int verity_status(struct dm_target *ti, status_type_t type,
518 unsigned status_flags, char *result, unsigned maxlen)
520 struct dm_verity *v = ti->private;
525 case STATUSTYPE_INFO:
526 DMEMIT("%c", v->hash_failed ? 'C' : 'V');
528 case STATUSTYPE_TABLE:
529 DMEMIT("%u %s %s %u %u %llu %llu %s ",
533 1 << v->data_dev_block_bits,
534 1 << v->hash_dev_block_bits,
535 (unsigned long long)v->data_blocks,
536 (unsigned long long)v->hash_start,
539 for (x = 0; x < v->digest_size; x++)
540 DMEMIT("%02x", v->root_digest[x]);
545 for (x = 0; x < v->salt_size; x++)
546 DMEMIT("%02x", v->salt[x]);
553 static int verity_ioctl(struct dm_target *ti, unsigned cmd,
556 struct dm_verity *v = ti->private;
560 ti->len != i_size_read(v->data_dev->bdev->bd_inode) >> SECTOR_SHIFT)
561 r = scsi_verify_blk_ioctl(NULL, cmd);
563 return r ? : __blkdev_driver_ioctl(v->data_dev->bdev, v->data_dev->mode,
567 static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
568 struct bio_vec *biovec, int max_size)
570 struct dm_verity *v = ti->private;
571 struct request_queue *q = bdev_get_queue(v->data_dev->bdev);
573 if (!q->merge_bvec_fn)
576 bvm->bi_bdev = v->data_dev->bdev;
577 bvm->bi_sector = verity_map_sector(v, bvm->bi_sector);
579 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
582 static int verity_iterate_devices(struct dm_target *ti,
583 iterate_devices_callout_fn fn, void *data)
585 struct dm_verity *v = ti->private;
587 return fn(ti, v->data_dev, v->data_start, ti->len, data);
590 static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)
592 struct dm_verity *v = ti->private;
594 if (limits->logical_block_size < 1 << v->data_dev_block_bits)
595 limits->logical_block_size = 1 << v->data_dev_block_bits;
597 if (limits->physical_block_size < 1 << v->data_dev_block_bits)
598 limits->physical_block_size = 1 << v->data_dev_block_bits;
600 blk_limits_io_min(limits, limits->logical_block_size);
603 static void verity_dtr(struct dm_target *ti)
605 struct dm_verity *v = ti->private;
608 destroy_workqueue(v->verify_wq);
611 mempool_destroy(v->vec_mempool);
614 mempool_destroy(v->io_mempool);
617 dm_bufio_client_destroy(v->bufio);
620 kfree(v->root_digest);
623 crypto_free_shash(v->tfm);
628 dm_put_device(ti, v->hash_dev);
631 dm_put_device(ti, v->data_dev);
638 * <version> The current format is version 1.
639 * Vsn 0 is compatible with original Chromium OS releases.
644 * <the number of data blocks>
648 * <salt> Hex string or "-" if no salt.
650 static int verity_ctr(struct dm_target *ti, unsigned argc, char **argv)
654 unsigned long long num_ll;
657 sector_t hash_position;
660 v = kzalloc(sizeof(struct dm_verity), GFP_KERNEL);
662 ti->error = "Cannot allocate verity structure";
668 if ((dm_table_get_mode(ti->table) & ~FMODE_READ)) {
669 ti->error = "Device must be readonly";
675 ti->error = "Invalid argument count: exactly 10 arguments required";
680 if (sscanf(argv[0], "%d%c", &num, &dummy) != 1 ||
681 num < 0 || num > 1) {
682 ti->error = "Invalid version";
688 r = dm_get_device(ti, argv[1], FMODE_READ, &v->data_dev);
690 ti->error = "Data device lookup failed";
694 r = dm_get_device(ti, argv[2], FMODE_READ, &v->hash_dev);
696 ti->error = "Data device lookup failed";
700 if (sscanf(argv[3], "%u%c", &num, &dummy) != 1 ||
701 !num || (num & (num - 1)) ||
702 num < bdev_logical_block_size(v->data_dev->bdev) ||
704 ti->error = "Invalid data device block size";
708 v->data_dev_block_bits = ffs(num) - 1;
710 if (sscanf(argv[4], "%u%c", &num, &dummy) != 1 ||
711 !num || (num & (num - 1)) ||
712 num < bdev_logical_block_size(v->hash_dev->bdev) ||
714 ti->error = "Invalid hash device block size";
718 v->hash_dev_block_bits = ffs(num) - 1;
720 if (sscanf(argv[5], "%llu%c", &num_ll, &dummy) != 1 ||
721 (sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
722 >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll) {
723 ti->error = "Invalid data blocks";
727 v->data_blocks = num_ll;
729 if (ti->len > (v->data_blocks << (v->data_dev_block_bits - SECTOR_SHIFT))) {
730 ti->error = "Data device is too small";
735 if (sscanf(argv[6], "%llu%c", &num_ll, &dummy) != 1 ||
736 (sector_t)(num_ll << (v->hash_dev_block_bits - SECTOR_SHIFT))
737 >> (v->hash_dev_block_bits - SECTOR_SHIFT) != num_ll) {
738 ti->error = "Invalid hash start";
742 v->hash_start = num_ll;
744 v->alg_name = kstrdup(argv[7], GFP_KERNEL);
746 ti->error = "Cannot allocate algorithm name";
751 v->tfm = crypto_alloc_shash(v->alg_name, 0, 0);
752 if (IS_ERR(v->tfm)) {
753 ti->error = "Cannot initialize hash function";
758 v->digest_size = crypto_shash_digestsize(v->tfm);
759 if ((1 << v->hash_dev_block_bits) < v->digest_size * 2) {
760 ti->error = "Digest size too big";
765 sizeof(struct shash_desc) + crypto_shash_descsize(v->tfm);
767 v->root_digest = kmalloc(v->digest_size, GFP_KERNEL);
768 if (!v->root_digest) {
769 ti->error = "Cannot allocate root digest";
773 if (strlen(argv[8]) != v->digest_size * 2 ||
774 hex2bin(v->root_digest, argv[8], v->digest_size)) {
775 ti->error = "Invalid root digest";
780 if (strcmp(argv[9], "-")) {
781 v->salt_size = strlen(argv[9]) / 2;
782 v->salt = kmalloc(v->salt_size, GFP_KERNEL);
784 ti->error = "Cannot allocate salt";
788 if (strlen(argv[9]) != v->salt_size * 2 ||
789 hex2bin(v->salt, argv[9], v->salt_size)) {
790 ti->error = "Invalid salt";
796 v->hash_per_block_bits =
797 fls((1 << v->hash_dev_block_bits) / v->digest_size) - 1;
801 while (v->hash_per_block_bits * v->levels < 64 &&
802 (unsigned long long)(v->data_blocks - 1) >>
803 (v->hash_per_block_bits * v->levels))
806 if (v->levels > DM_VERITY_MAX_LEVELS) {
807 ti->error = "Too many tree levels";
812 hash_position = v->hash_start;
813 for (i = v->levels - 1; i >= 0; i--) {
815 v->hash_level_block[i] = hash_position;
816 s = verity_position_at_level(v, v->data_blocks, i);
817 s = (s >> v->hash_per_block_bits) +
818 !!(s & ((1 << v->hash_per_block_bits) - 1));
819 if (hash_position + s < hash_position) {
820 ti->error = "Hash device offset overflow";
826 v->hash_blocks = hash_position;
828 v->bufio = dm_bufio_client_create(v->hash_dev->bdev,
829 1 << v->hash_dev_block_bits, 1, sizeof(struct buffer_aux),
830 dm_bufio_alloc_callback, NULL);
831 if (IS_ERR(v->bufio)) {
832 ti->error = "Cannot initialize dm-bufio";
833 r = PTR_ERR(v->bufio);
838 if (dm_bufio_get_device_size(v->bufio) < v->hash_blocks) {
839 ti->error = "Hash device is too small";
844 v->io_mempool = mempool_create_kmalloc_pool(DM_VERITY_MEMPOOL_SIZE,
845 sizeof(struct dm_verity_io) + v->shash_descsize + v->digest_size * 2);
846 if (!v->io_mempool) {
847 ti->error = "Cannot allocate io mempool";
852 v->vec_mempool = mempool_create_kmalloc_pool(DM_VERITY_MEMPOOL_SIZE,
853 BIO_MAX_PAGES * sizeof(struct bio_vec));
854 if (!v->vec_mempool) {
855 ti->error = "Cannot allocate vector mempool";
860 /* WQ_UNBOUND greatly improves performance when running on ramdisk */
861 v->verify_wq = alloc_workqueue("kverityd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, num_online_cpus());
863 ti->error = "Cannot allocate workqueue";
876 static struct target_type verity_target = {
878 .version = {1, 0, 0},
879 .module = THIS_MODULE,
883 .status = verity_status,
884 .ioctl = verity_ioctl,
885 .merge = verity_merge,
886 .iterate_devices = verity_iterate_devices,
887 .io_hints = verity_io_hints,
890 static int __init dm_verity_init(void)
894 r = dm_register_target(&verity_target);
896 DMERR("register failed %d", r);
901 static void __exit dm_verity_exit(void)
903 dm_unregister_target(&verity_target);
906 module_init(dm_verity_init);
907 module_exit(dm_verity_exit);
909 MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>");
910 MODULE_AUTHOR("Mandeep Baines <msb@chromium.org>");
911 MODULE_AUTHOR("Will Drewry <wad@chromium.org>");
912 MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
913 MODULE_LICENSE("GPL");