1/*
2 * Copyright (C) 2012 Red Hat, Inc.
3 *
4 * Author: Mikulas Patocka <mpatocka@redhat.com>
5 *
6 * Based on Chromium dm-verity driver (C) 2011 The Chromium OS Authors
7 *
8 * This file is released under the GPLv2.
9 *
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
14 * access behavior.
15 */
16
17#include "dm-bufio.h"
18
19#include <linux/module.h>
20#include <linux/device-mapper.h>
21#include <linux/reboot.h>
22#include <crypto/hash.h>
23
24#define DM_MSG_PREFIX			"verity"
25
26#define DM_VERITY_ENV_LENGTH		42
27#define DM_VERITY_ENV_VAR_NAME		"DM_VERITY_ERR_BLOCK_NR"
28
29#define DM_VERITY_IO_VEC_INLINE		16
30#define DM_VERITY_MEMPOOL_SIZE		4
31#define DM_VERITY_DEFAULT_PREFETCH_SIZE	262144
32
33#define DM_VERITY_MAX_LEVELS		63
34#define DM_VERITY_MAX_CORRUPTED_ERRS	100
35
36#define DM_VERITY_OPT_LOGGING		"ignore_corruption"
37#define DM_VERITY_OPT_RESTART		"restart_on_corruption"
38
39static unsigned dm_verity_prefetch_cluster = DM_VERITY_DEFAULT_PREFETCH_SIZE;
40
41module_param_named(prefetch_cluster, dm_verity_prefetch_cluster, uint, S_IRUGO | S_IWUSR);
42
43enum verity_mode {
44	DM_VERITY_MODE_EIO,
45	DM_VERITY_MODE_LOGGING,
46	DM_VERITY_MODE_RESTART
47};
48
49enum verity_block_type {
50	DM_VERITY_BLOCK_TYPE_DATA,
51	DM_VERITY_BLOCK_TYPE_METADATA
52};
53
54struct dm_verity {
55	struct dm_dev *data_dev;
56	struct dm_dev *hash_dev;
57	struct dm_target *ti;
58	struct dm_bufio_client *bufio;
59	char *alg_name;
60	struct crypto_shash *tfm;
61	u8 *root_digest;	/* digest of the root block */
62	u8 *salt;		/* salt: its size is salt_size */
63	unsigned salt_size;
64	sector_t data_start;	/* data offset in 512-byte sectors */
65	sector_t hash_start;	/* hash start in blocks */
66	sector_t data_blocks;	/* the number of data blocks */
67	sector_t hash_blocks;	/* the number of hash blocks */
68	unsigned char data_dev_block_bits;	/* log2(data blocksize) */
69	unsigned char hash_dev_block_bits;	/* log2(hash blocksize) */
70	unsigned char hash_per_block_bits;	/* log2(hashes in hash block) */
71	unsigned char levels;	/* the number of tree levels */
72	unsigned char version;
73	unsigned digest_size;	/* digest size for the current hash algorithm */
74	unsigned shash_descsize;/* the size of temporary space for crypto */
75	int hash_failed;	/* set to 1 if hash of any block failed */
76	enum verity_mode mode;	/* mode for handling verification errors */
77	unsigned corrupted_errs;/* Number of errors for corrupted blocks */
78
79	mempool_t *vec_mempool;	/* mempool of bio vector */
80
81	struct workqueue_struct *verify_wq;
82
83	/* starting blocks for each tree level. 0 is the lowest level. */
84	sector_t hash_level_block[DM_VERITY_MAX_LEVELS];
85};
86
87struct dm_verity_io {
88	struct dm_verity *v;
89
90	/* original values of bio->bi_end_io and bio->bi_private */
91	bio_end_io_t *orig_bi_end_io;
92	void *orig_bi_private;
93
94	sector_t block;
95	unsigned n_blocks;
96
97	struct bvec_iter iter;
98
99	struct work_struct work;
100
101	/*
102	 * Three variably-size fields follow this struct:
103	 *
104	 * u8 hash_desc[v->shash_descsize];
105	 * u8 real_digest[v->digest_size];
106	 * u8 want_digest[v->digest_size];
107	 *
108	 * To access them use: io_hash_desc(), io_real_digest() and io_want_digest().
109	 */
110};
111
112struct dm_verity_prefetch_work {
113	struct work_struct work;
114	struct dm_verity *v;
115	sector_t block;
116	unsigned n_blocks;
117};
118
119static struct shash_desc *io_hash_desc(struct dm_verity *v, struct dm_verity_io *io)
120{
121	return (struct shash_desc *)(io + 1);
122}
123
124static u8 *io_real_digest(struct dm_verity *v, struct dm_verity_io *io)
125{
126	return (u8 *)(io + 1) + v->shash_descsize;
127}
128
129static u8 *io_want_digest(struct dm_verity *v, struct dm_verity_io *io)
130{
131	return (u8 *)(io + 1) + v->shash_descsize + v->digest_size;
132}
133
134/*
135 * Auxiliary structure appended to each dm-bufio buffer. If the value
136 * hash_verified is nonzero, hash of the block has been verified.
137 *
138 * The variable hash_verified is set to 0 when allocating the buffer, then
139 * it can be changed to 1 and it is never reset to 0 again.
140 *
141 * There is no lock around this value, a race condition can at worst cause
142 * that multiple processes verify the hash of the same buffer simultaneously
143 * and write 1 to hash_verified simultaneously.
144 * This condition is harmless, so we don't need locking.
145 */
146struct buffer_aux {
147	int hash_verified;
148};
149
150/*
151 * Initialize struct buffer_aux for a freshly created buffer.
152 */
153static void dm_bufio_alloc_callback(struct dm_buffer *buf)
154{
155	struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
156
157	aux->hash_verified = 0;
158}
159
160/*
161 * Translate input sector number to the sector number on the target device.
162 */
163static sector_t verity_map_sector(struct dm_verity *v, sector_t bi_sector)
164{
165	return v->data_start + dm_target_offset(v->ti, bi_sector);
166}
167
168/*
169 * Return hash position of a specified block at a specified tree level
170 * (0 is the lowest level).
171 * The lowest "hash_per_block_bits"-bits of the result denote hash position
172 * inside a hash block. The remaining bits denote location of the hash block.
173 */
174static sector_t verity_position_at_level(struct dm_verity *v, sector_t block,
175					 int level)
176{
177	return block >> (level * v->hash_per_block_bits);
178}
179
180static void verity_hash_at_level(struct dm_verity *v, sector_t block, int level,
181				 sector_t *hash_block, unsigned *offset)
182{
183	sector_t position = verity_position_at_level(v, block, level);
184	unsigned idx;
185
186	*hash_block = v->hash_level_block[level] + (position >> v->hash_per_block_bits);
187
188	if (!offset)
189		return;
190
191	idx = position & ((1 << v->hash_per_block_bits) - 1);
192	if (!v->version)
193		*offset = idx * v->digest_size;
194	else
195		*offset = idx << (v->hash_dev_block_bits - v->hash_per_block_bits);
196}
197
198/*
199 * Handle verification errors.
200 */
201static int verity_handle_err(struct dm_verity *v, enum verity_block_type type,
202			     unsigned long long block)
203{
204	char verity_env[DM_VERITY_ENV_LENGTH];
205	char *envp[] = { verity_env, NULL };
206	const char *type_str = "";
207	struct mapped_device *md = dm_table_get_md(v->ti->table);
208
209	/* Corruption should be visible in device status in all modes */
210	v->hash_failed = 1;
211
212	if (v->corrupted_errs >= DM_VERITY_MAX_CORRUPTED_ERRS)
213		goto out;
214
215	v->corrupted_errs++;
216
217	switch (type) {
218	case DM_VERITY_BLOCK_TYPE_DATA:
219		type_str = "data";
220		break;
221	case DM_VERITY_BLOCK_TYPE_METADATA:
222		type_str = "metadata";
223		break;
224	default:
225		BUG();
226	}
227
228	DMERR("%s: %s block %llu is corrupted", v->data_dev->name, type_str,
229		block);
230
231	if (v->corrupted_errs == DM_VERITY_MAX_CORRUPTED_ERRS)
232		DMERR("%s: reached maximum errors", v->data_dev->name);
233
234	snprintf(verity_env, DM_VERITY_ENV_LENGTH, "%s=%d,%llu",
235		DM_VERITY_ENV_VAR_NAME, type, block);
236
237	kobject_uevent_env(&disk_to_dev(dm_disk(md))->kobj, KOBJ_CHANGE, envp);
238
239out:
240	if (v->mode == DM_VERITY_MODE_LOGGING)
241		return 0;
242
243	if (v->mode == DM_VERITY_MODE_RESTART)
244		kernel_restart("dm-verity device corrupted");
245
246	return 1;
247}
248
249/*
250 * Verify hash of a metadata block pertaining to the specified data block
251 * ("block" argument) at a specified level ("level" argument).
252 *
253 * On successful return, io_want_digest(v, io) contains the hash value for
254 * a lower tree level or for the data block (if we're at the lowest leve).
255 *
256 * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned.
257 * If "skip_unverified" is false, unverified buffer is hashed and verified
258 * against current value of io_want_digest(v, io).
259 */
260static int verity_verify_level(struct dm_verity_io *io, sector_t block,
261			       int level, bool skip_unverified)
262{
263	struct dm_verity *v = io->v;
264	struct dm_buffer *buf;
265	struct buffer_aux *aux;
266	u8 *data;
267	int r;
268	sector_t hash_block;
269	unsigned offset;
270
271	verity_hash_at_level(v, block, level, &hash_block, &offset);
272
273	data = dm_bufio_read(v->bufio, hash_block, &buf);
274	if (unlikely(IS_ERR(data)))
275		return PTR_ERR(data);
276
277	aux = dm_bufio_get_aux_data(buf);
278
279	if (!aux->hash_verified) {
280		struct shash_desc *desc;
281		u8 *result;
282
283		if (skip_unverified) {
284			r = 1;
285			goto release_ret_r;
286		}
287
288		desc = io_hash_desc(v, io);
289		desc->tfm = v->tfm;
290		desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
291		r = crypto_shash_init(desc);
292		if (r < 0) {
293			DMERR("crypto_shash_init failed: %d", r);
294			goto release_ret_r;
295		}
296
297		if (likely(v->version >= 1)) {
298			r = crypto_shash_update(desc, v->salt, v->salt_size);
299			if (r < 0) {
300				DMERR("crypto_shash_update failed: %d", r);
301				goto release_ret_r;
302			}
303		}
304
305		r = crypto_shash_update(desc, data, 1 << v->hash_dev_block_bits);
306		if (r < 0) {
307			DMERR("crypto_shash_update failed: %d", r);
308			goto release_ret_r;
309		}
310
311		if (!v->version) {
312			r = crypto_shash_update(desc, v->salt, v->salt_size);
313			if (r < 0) {
314				DMERR("crypto_shash_update failed: %d", r);
315				goto release_ret_r;
316			}
317		}
318
319		result = io_real_digest(v, io);
320		r = crypto_shash_final(desc, result);
321		if (r < 0) {
322			DMERR("crypto_shash_final failed: %d", r);
323			goto release_ret_r;
324		}
325		if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
326			if (verity_handle_err(v, DM_VERITY_BLOCK_TYPE_METADATA,
327					      hash_block)) {
328				r = -EIO;
329				goto release_ret_r;
330			}
331		} else
332			aux->hash_verified = 1;
333	}
334
335	data += offset;
336
337	memcpy(io_want_digest(v, io), data, v->digest_size);
338
339	dm_bufio_release(buf);
340	return 0;
341
342release_ret_r:
343	dm_bufio_release(buf);
344
345	return r;
346}
347
348/*
349 * Verify one "dm_verity_io" structure.
350 */
351static int verity_verify_io(struct dm_verity_io *io)
352{
353	struct dm_verity *v = io->v;
354	struct bio *bio = dm_bio_from_per_bio_data(io,
355						   v->ti->per_bio_data_size);
356	unsigned b;
357	int i;
358
359	for (b = 0; b < io->n_blocks; b++) {
360		struct shash_desc *desc;
361		u8 *result;
362		int r;
363		unsigned todo;
364
365		if (likely(v->levels)) {
366			/*
367			 * First, we try to get the requested hash for
368			 * the current block. If the hash block itself is
369			 * verified, zero is returned. If it isn't, this
370			 * function returns 0 and we fall back to whole
371			 * chain verification.
372			 */
373			int r = verity_verify_level(io, io->block + b, 0, true);
374			if (likely(!r))
375				goto test_block_hash;
376			if (r < 0)
377				return r;
378		}
379
380		memcpy(io_want_digest(v, io), v->root_digest, v->digest_size);
381
382		for (i = v->levels - 1; i >= 0; i--) {
383			int r = verity_verify_level(io, io->block + b, i, false);
384			if (unlikely(r))
385				return r;
386		}
387
388test_block_hash:
389		desc = io_hash_desc(v, io);
390		desc->tfm = v->tfm;
391		desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
392		r = crypto_shash_init(desc);
393		if (r < 0) {
394			DMERR("crypto_shash_init failed: %d", r);
395			return r;
396		}
397
398		if (likely(v->version >= 1)) {
399			r = crypto_shash_update(desc, v->salt, v->salt_size);
400			if (r < 0) {
401				DMERR("crypto_shash_update failed: %d", r);
402				return r;
403			}
404		}
405		todo = 1 << v->data_dev_block_bits;
406		do {
407			u8 *page;
408			unsigned len;
409			struct bio_vec bv = bio_iter_iovec(bio, io->iter);
410
411			page = kmap_atomic(bv.bv_page);
412			len = bv.bv_len;
413			if (likely(len >= todo))
414				len = todo;
415			r = crypto_shash_update(desc, page + bv.bv_offset, len);
416			kunmap_atomic(page);
417
418			if (r < 0) {
419				DMERR("crypto_shash_update failed: %d", r);
420				return r;
421			}
422
423			bio_advance_iter(bio, &io->iter, len);
424			todo -= len;
425		} while (todo);
426
427		if (!v->version) {
428			r = crypto_shash_update(desc, v->salt, v->salt_size);
429			if (r < 0) {
430				DMERR("crypto_shash_update failed: %d", r);
431				return r;
432			}
433		}
434
435		result = io_real_digest(v, io);
436		r = crypto_shash_final(desc, result);
437		if (r < 0) {
438			DMERR("crypto_shash_final failed: %d", r);
439			return r;
440		}
441		if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) {
442			if (verity_handle_err(v, DM_VERITY_BLOCK_TYPE_DATA,
443					      io->block + b))
444				return -EIO;
445		}
446	}
447
448	return 0;
449}
450
451/*
452 * End one "io" structure with a given error.
453 */
454static void verity_finish_io(struct dm_verity_io *io, int error)
455{
456	struct dm_verity *v = io->v;
457	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_bio_data_size);
458
459	bio->bi_end_io = io->orig_bi_end_io;
460	bio->bi_private = io->orig_bi_private;
461
462	bio_endio_nodec(bio, error);
463}
464
465static void verity_work(struct work_struct *w)
466{
467	struct dm_verity_io *io = container_of(w, struct dm_verity_io, work);
468
469	verity_finish_io(io, verity_verify_io(io));
470}
471
472static void verity_end_io(struct bio *bio, int error)
473{
474	struct dm_verity_io *io = bio->bi_private;
475
476	if (error) {
477		verity_finish_io(io, error);
478		return;
479	}
480
481	INIT_WORK(&io->work, verity_work);
482	queue_work(io->v->verify_wq, &io->work);
483}
484
485/*
486 * Prefetch buffers for the specified io.
487 * The root buffer is not prefetched, it is assumed that it will be cached
488 * all the time.
489 */
490static void verity_prefetch_io(struct work_struct *work)
491{
492	struct dm_verity_prefetch_work *pw =
493		container_of(work, struct dm_verity_prefetch_work, work);
494	struct dm_verity *v = pw->v;
495	int i;
496
497	for (i = v->levels - 2; i >= 0; i--) {
498		sector_t hash_block_start;
499		sector_t hash_block_end;
500		verity_hash_at_level(v, pw->block, i, &hash_block_start, NULL);
501		verity_hash_at_level(v, pw->block + pw->n_blocks - 1, i, &hash_block_end, NULL);
502		if (!i) {
503			unsigned cluster = ACCESS_ONCE(dm_verity_prefetch_cluster);
504
505			cluster >>= v->data_dev_block_bits;
506			if (unlikely(!cluster))
507				goto no_prefetch_cluster;
508
509			if (unlikely(cluster & (cluster - 1)))
510				cluster = 1 << __fls(cluster);
511
512			hash_block_start &= ~(sector_t)(cluster - 1);
513			hash_block_end |= cluster - 1;
514			if (unlikely(hash_block_end >= v->hash_blocks))
515				hash_block_end = v->hash_blocks - 1;
516		}
517no_prefetch_cluster:
518		dm_bufio_prefetch(v->bufio, hash_block_start,
519				  hash_block_end - hash_block_start + 1);
520	}
521
522	kfree(pw);
523}
524
525static void verity_submit_prefetch(struct dm_verity *v, struct dm_verity_io *io)
526{
527	struct dm_verity_prefetch_work *pw;
528
529	pw = kmalloc(sizeof(struct dm_verity_prefetch_work),
530		GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
531
532	if (!pw)
533		return;
534
535	INIT_WORK(&pw->work, verity_prefetch_io);
536	pw->v = v;
537	pw->block = io->block;
538	pw->n_blocks = io->n_blocks;
539	queue_work(v->verify_wq, &pw->work);
540}
541
542/*
543 * Bio map function. It allocates dm_verity_io structure and bio vector and
544 * fills them. Then it issues prefetches and the I/O.
545 */
546static int verity_map(struct dm_target *ti, struct bio *bio)
547{
548	struct dm_verity *v = ti->private;
549	struct dm_verity_io *io;
550
551	bio->bi_bdev = v->data_dev->bdev;
552	bio->bi_iter.bi_sector = verity_map_sector(v, bio->bi_iter.bi_sector);
553
554	if (((unsigned)bio->bi_iter.bi_sector | bio_sectors(bio)) &
555	    ((1 << (v->data_dev_block_bits - SECTOR_SHIFT)) - 1)) {
556		DMERR_LIMIT("unaligned io");
557		return -EIO;
558	}
559
560	if (bio_end_sector(bio) >>
561	    (v->data_dev_block_bits - SECTOR_SHIFT) > v->data_blocks) {
562		DMERR_LIMIT("io out of range");
563		return -EIO;
564	}
565
566	if (bio_data_dir(bio) == WRITE)
567		return -EIO;
568
569	io = dm_per_bio_data(bio, ti->per_bio_data_size);
570	io->v = v;
571	io->orig_bi_end_io = bio->bi_end_io;
572	io->orig_bi_private = bio->bi_private;
573	io->block = bio->bi_iter.bi_sector >> (v->data_dev_block_bits - SECTOR_SHIFT);
574	io->n_blocks = bio->bi_iter.bi_size >> v->data_dev_block_bits;
575
576	bio->bi_end_io = verity_end_io;
577	bio->bi_private = io;
578	io->iter = bio->bi_iter;
579
580	verity_submit_prefetch(v, io);
581
582	generic_make_request(bio);
583
584	return DM_MAPIO_SUBMITTED;
585}
586
587/*
588 * Status: V (valid) or C (corruption found)
589 */
590static void verity_status(struct dm_target *ti, status_type_t type,
591			  unsigned status_flags, char *result, unsigned maxlen)
592{
593	struct dm_verity *v = ti->private;
594	unsigned sz = 0;
595	unsigned x;
596
597	switch (type) {
598	case STATUSTYPE_INFO:
599		DMEMIT("%c", v->hash_failed ? 'C' : 'V');
600		break;
601	case STATUSTYPE_TABLE:
602		DMEMIT("%u %s %s %u %u %llu %llu %s ",
603			v->version,
604			v->data_dev->name,
605			v->hash_dev->name,
606			1 << v->data_dev_block_bits,
607			1 << v->hash_dev_block_bits,
608			(unsigned long long)v->data_blocks,
609			(unsigned long long)v->hash_start,
610			v->alg_name
611			);
612		for (x = 0; x < v->digest_size; x++)
613			DMEMIT("%02x", v->root_digest[x]);
614		DMEMIT(" ");
615		if (!v->salt_size)
616			DMEMIT("-");
617		else
618			for (x = 0; x < v->salt_size; x++)
619				DMEMIT("%02x", v->salt[x]);
620		if (v->mode != DM_VERITY_MODE_EIO) {
621			DMEMIT(" 1 ");
622			switch (v->mode) {
623			case DM_VERITY_MODE_LOGGING:
624				DMEMIT(DM_VERITY_OPT_LOGGING);
625				break;
626			case DM_VERITY_MODE_RESTART:
627				DMEMIT(DM_VERITY_OPT_RESTART);
628				break;
629			default:
630				BUG();
631			}
632		}
633		break;
634	}
635}
636
637static int verity_ioctl(struct dm_target *ti, unsigned cmd,
638			unsigned long arg)
639{
640	struct dm_verity *v = ti->private;
641	int r = 0;
642
643	if (v->data_start ||
644	    ti->len != i_size_read(v->data_dev->bdev->bd_inode) >> SECTOR_SHIFT)
645		r = scsi_verify_blk_ioctl(NULL, cmd);
646
647	return r ? : __blkdev_driver_ioctl(v->data_dev->bdev, v->data_dev->mode,
648				     cmd, arg);
649}
650
651static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
652			struct bio_vec *biovec, int max_size)
653{
654	struct dm_verity *v = ti->private;
655	struct request_queue *q = bdev_get_queue(v->data_dev->bdev);
656
657	if (!q->merge_bvec_fn)
658		return max_size;
659
660	bvm->bi_bdev = v->data_dev->bdev;
661	bvm->bi_sector = verity_map_sector(v, bvm->bi_sector);
662
663	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
664}
665
666static int verity_iterate_devices(struct dm_target *ti,
667				  iterate_devices_callout_fn fn, void *data)
668{
669	struct dm_verity *v = ti->private;
670
671	return fn(ti, v->data_dev, v->data_start, ti->len, data);
672}
673
674static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)
675{
676	struct dm_verity *v = ti->private;
677
678	if (limits->logical_block_size < 1 << v->data_dev_block_bits)
679		limits->logical_block_size = 1 << v->data_dev_block_bits;
680
681	if (limits->physical_block_size < 1 << v->data_dev_block_bits)
682		limits->physical_block_size = 1 << v->data_dev_block_bits;
683
684	blk_limits_io_min(limits, limits->logical_block_size);
685}
686
687static void verity_dtr(struct dm_target *ti)
688{
689	struct dm_verity *v = ti->private;
690
691	if (v->verify_wq)
692		destroy_workqueue(v->verify_wq);
693
694	if (v->vec_mempool)
695		mempool_destroy(v->vec_mempool);
696
697	if (v->bufio)
698		dm_bufio_client_destroy(v->bufio);
699
700	kfree(v->salt);
701	kfree(v->root_digest);
702
703	if (v->tfm)
704		crypto_free_shash(v->tfm);
705
706	kfree(v->alg_name);
707
708	if (v->hash_dev)
709		dm_put_device(ti, v->hash_dev);
710
711	if (v->data_dev)
712		dm_put_device(ti, v->data_dev);
713
714	kfree(v);
715}
716
717/*
718 * Target parameters:
719 *	<version>	The current format is version 1.
720 *			Vsn 0 is compatible with original Chromium OS releases.
721 *	<data device>
722 *	<hash device>
723 *	<data block size>
724 *	<hash block size>
725 *	<the number of data blocks>
726 *	<hash start block>
727 *	<algorithm>
728 *	<digest>
729 *	<salt>		Hex string or "-" if no salt.
730 */
731static int verity_ctr(struct dm_target *ti, unsigned argc, char **argv)
732{
733	struct dm_verity *v;
734	struct dm_arg_set as;
735	const char *opt_string;
736	unsigned int num, opt_params;
737	unsigned long long num_ll;
738	int r;
739	int i;
740	sector_t hash_position;
741	char dummy;
742
743	static struct dm_arg _args[] = {
744		{0, 1, "Invalid number of feature args"},
745	};
746
747	v = kzalloc(sizeof(struct dm_verity), GFP_KERNEL);
748	if (!v) {
749		ti->error = "Cannot allocate verity structure";
750		return -ENOMEM;
751	}
752	ti->private = v;
753	v->ti = ti;
754
755	if ((dm_table_get_mode(ti->table) & ~FMODE_READ)) {
756		ti->error = "Device must be readonly";
757		r = -EINVAL;
758		goto bad;
759	}
760
761	if (argc < 10) {
762		ti->error = "Not enough arguments";
763		r = -EINVAL;
764		goto bad;
765	}
766
767	if (sscanf(argv[0], "%u%c", &num, &dummy) != 1 ||
768	    num > 1) {
769		ti->error = "Invalid version";
770		r = -EINVAL;
771		goto bad;
772	}
773	v->version = num;
774
775	r = dm_get_device(ti, argv[1], FMODE_READ, &v->data_dev);
776	if (r) {
777		ti->error = "Data device lookup failed";
778		goto bad;
779	}
780
781	r = dm_get_device(ti, argv[2], FMODE_READ, &v->hash_dev);
782	if (r) {
783		ti->error = "Data device lookup failed";
784		goto bad;
785	}
786
787	if (sscanf(argv[3], "%u%c", &num, &dummy) != 1 ||
788	    !num || (num & (num - 1)) ||
789	    num < bdev_logical_block_size(v->data_dev->bdev) ||
790	    num > PAGE_SIZE) {
791		ti->error = "Invalid data device block size";
792		r = -EINVAL;
793		goto bad;
794	}
795	v->data_dev_block_bits = __ffs(num);
796
797	if (sscanf(argv[4], "%u%c", &num, &dummy) != 1 ||
798	    !num || (num & (num - 1)) ||
799	    num < bdev_logical_block_size(v->hash_dev->bdev) ||
800	    num > INT_MAX) {
801		ti->error = "Invalid hash device block size";
802		r = -EINVAL;
803		goto bad;
804	}
805	v->hash_dev_block_bits = __ffs(num);
806
807	if (sscanf(argv[5], "%llu%c", &num_ll, &dummy) != 1 ||
808	    (sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
809	    >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll) {
810		ti->error = "Invalid data blocks";
811		r = -EINVAL;
812		goto bad;
813	}
814	v->data_blocks = num_ll;
815
816	if (ti->len > (v->data_blocks << (v->data_dev_block_bits - SECTOR_SHIFT))) {
817		ti->error = "Data device is too small";
818		r = -EINVAL;
819		goto bad;
820	}
821
822	if (sscanf(argv[6], "%llu%c", &num_ll, &dummy) != 1 ||
823	    (sector_t)(num_ll << (v->hash_dev_block_bits - SECTOR_SHIFT))
824	    >> (v->hash_dev_block_bits - SECTOR_SHIFT) != num_ll) {
825		ti->error = "Invalid hash start";
826		r = -EINVAL;
827		goto bad;
828	}
829	v->hash_start = num_ll;
830
831	v->alg_name = kstrdup(argv[7], GFP_KERNEL);
832	if (!v->alg_name) {
833		ti->error = "Cannot allocate algorithm name";
834		r = -ENOMEM;
835		goto bad;
836	}
837
838	v->tfm = crypto_alloc_shash(v->alg_name, 0, 0);
839	if (IS_ERR(v->tfm)) {
840		ti->error = "Cannot initialize hash function";
841		r = PTR_ERR(v->tfm);
842		v->tfm = NULL;
843		goto bad;
844	}
845	v->digest_size = crypto_shash_digestsize(v->tfm);
846	if ((1 << v->hash_dev_block_bits) < v->digest_size * 2) {
847		ti->error = "Digest size too big";
848		r = -EINVAL;
849		goto bad;
850	}
851	v->shash_descsize =
852		sizeof(struct shash_desc) + crypto_shash_descsize(v->tfm);
853
854	v->root_digest = kmalloc(v->digest_size, GFP_KERNEL);
855	if (!v->root_digest) {
856		ti->error = "Cannot allocate root digest";
857		r = -ENOMEM;
858		goto bad;
859	}
860	if (strlen(argv[8]) != v->digest_size * 2 ||
861	    hex2bin(v->root_digest, argv[8], v->digest_size)) {
862		ti->error = "Invalid root digest";
863		r = -EINVAL;
864		goto bad;
865	}
866
867	if (strcmp(argv[9], "-")) {
868		v->salt_size = strlen(argv[9]) / 2;
869		v->salt = kmalloc(v->salt_size, GFP_KERNEL);
870		if (!v->salt) {
871			ti->error = "Cannot allocate salt";
872			r = -ENOMEM;
873			goto bad;
874		}
875		if (strlen(argv[9]) != v->salt_size * 2 ||
876		    hex2bin(v->salt, argv[9], v->salt_size)) {
877			ti->error = "Invalid salt";
878			r = -EINVAL;
879			goto bad;
880		}
881	}
882
883	argv += 10;
884	argc -= 10;
885
886	/* Optional parameters */
887	if (argc) {
888		as.argc = argc;
889		as.argv = argv;
890
891		r = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
892		if (r)
893			goto bad;
894
895		while (opt_params) {
896			opt_params--;
897			opt_string = dm_shift_arg(&as);
898			if (!opt_string) {
899				ti->error = "Not enough feature arguments";
900				r = -EINVAL;
901				goto bad;
902			}
903
904			if (!strcasecmp(opt_string, DM_VERITY_OPT_LOGGING))
905				v->mode = DM_VERITY_MODE_LOGGING;
906			else if (!strcasecmp(opt_string, DM_VERITY_OPT_RESTART))
907				v->mode = DM_VERITY_MODE_RESTART;
908			else {
909				ti->error = "Invalid feature arguments";
910				r = -EINVAL;
911				goto bad;
912			}
913		}
914	}
915
916	v->hash_per_block_bits =
917		__fls((1 << v->hash_dev_block_bits) / v->digest_size);
918
919	v->levels = 0;
920	if (v->data_blocks)
921		while (v->hash_per_block_bits * v->levels < 64 &&
922		       (unsigned long long)(v->data_blocks - 1) >>
923		       (v->hash_per_block_bits * v->levels))
924			v->levels++;
925
926	if (v->levels > DM_VERITY_MAX_LEVELS) {
927		ti->error = "Too many tree levels";
928		r = -E2BIG;
929		goto bad;
930	}
931
932	hash_position = v->hash_start;
933	for (i = v->levels - 1; i >= 0; i--) {
934		sector_t s;
935		v->hash_level_block[i] = hash_position;
936		s = (v->data_blocks + ((sector_t)1 << ((i + 1) * v->hash_per_block_bits)) - 1)
937					>> ((i + 1) * v->hash_per_block_bits);
938		if (hash_position + s < hash_position) {
939			ti->error = "Hash device offset overflow";
940			r = -E2BIG;
941			goto bad;
942		}
943		hash_position += s;
944	}
945	v->hash_blocks = hash_position;
946
947	v->bufio = dm_bufio_client_create(v->hash_dev->bdev,
948		1 << v->hash_dev_block_bits, 1, sizeof(struct buffer_aux),
949		dm_bufio_alloc_callback, NULL);
950	if (IS_ERR(v->bufio)) {
951		ti->error = "Cannot initialize dm-bufio";
952		r = PTR_ERR(v->bufio);
953		v->bufio = NULL;
954		goto bad;
955	}
956
957	if (dm_bufio_get_device_size(v->bufio) < v->hash_blocks) {
958		ti->error = "Hash device is too small";
959		r = -E2BIG;
960		goto bad;
961	}
962
963	ti->per_bio_data_size = roundup(sizeof(struct dm_verity_io) + v->shash_descsize + v->digest_size * 2, __alignof__(struct dm_verity_io));
964
965	v->vec_mempool = mempool_create_kmalloc_pool(DM_VERITY_MEMPOOL_SIZE,
966					BIO_MAX_PAGES * sizeof(struct bio_vec));
967	if (!v->vec_mempool) {
968		ti->error = "Cannot allocate vector mempool";
969		r = -ENOMEM;
970		goto bad;
971	}
972
973	/* WQ_UNBOUND greatly improves performance when running on ramdisk */
974	v->verify_wq = alloc_workqueue("kverityd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, num_online_cpus());
975	if (!v->verify_wq) {
976		ti->error = "Cannot allocate workqueue";
977		r = -ENOMEM;
978		goto bad;
979	}
980
981	return 0;
982
983bad:
984	verity_dtr(ti);
985
986	return r;
987}
988
989static struct target_type verity_target = {
990	.name		= "verity",
991	.version	= {1, 2, 0},
992	.module		= THIS_MODULE,
993	.ctr		= verity_ctr,
994	.dtr		= verity_dtr,
995	.map		= verity_map,
996	.status		= verity_status,
997	.ioctl		= verity_ioctl,
998	.merge		= verity_merge,
999	.iterate_devices = verity_iterate_devices,
1000	.io_hints	= verity_io_hints,
1001};
1002
1003static int __init dm_verity_init(void)
1004{
1005	int r;
1006
1007	r = dm_register_target(&verity_target);
1008	if (r < 0)
1009		DMERR("register failed %d", r);
1010
1011	return r;
1012}
1013
1014static void __exit dm_verity_exit(void)
1015{
1016	dm_unregister_target(&verity_target);
1017}
1018
1019module_init(dm_verity_init);
1020module_exit(dm_verity_exit);
1021
1022MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>");
1023MODULE_AUTHOR("Mandeep Baines <msb@chromium.org>");
1024MODULE_AUTHOR("Will Drewry <wad@chromium.org>");
1025MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
1026MODULE_LICENSE("GPL");
1027