xref: /openbmc/linux/drivers/md/dm-verity-target.c (revision 929e2a61)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 Red Hat, Inc.
4  *
5  * Author: Mikulas Patocka <mpatocka@redhat.com>
6  *
7  * Based on Chromium dm-verity driver (C) 2011 The Chromium OS Authors
8  *
9  * In the file "/sys/module/dm_verity/parameters/prefetch_cluster" you can set
10  * default prefetch value. Data are read in "prefetch_cluster" chunks from the
11  * hash device. Setting this greatly improves performance when data and hash
12  * are on the same disk on different partitions on devices with poor random
13  * access behavior.
14  */
15 
16 #include "dm-verity.h"
17 #include "dm-verity-fec.h"
18 #include "dm-verity-verify-sig.h"
19 #include <linux/module.h>
20 #include <linux/reboot.h>
21 
22 #define DM_MSG_PREFIX			"verity"
23 
24 #define DM_VERITY_ENV_LENGTH		42
25 #define DM_VERITY_ENV_VAR_NAME		"DM_VERITY_ERR_BLOCK_NR"
26 
27 #define DM_VERITY_DEFAULT_PREFETCH_SIZE	262144
28 
29 #define DM_VERITY_MAX_CORRUPTED_ERRS	100
30 
31 #define DM_VERITY_OPT_LOGGING		"ignore_corruption"
32 #define DM_VERITY_OPT_RESTART		"restart_on_corruption"
33 #define DM_VERITY_OPT_PANIC		"panic_on_corruption"
34 #define DM_VERITY_OPT_IGN_ZEROES	"ignore_zero_blocks"
35 #define DM_VERITY_OPT_AT_MOST_ONCE	"check_at_most_once"
36 
37 #define DM_VERITY_OPTS_MAX		(2 + DM_VERITY_OPTS_FEC + \
38 					 DM_VERITY_ROOT_HASH_VERIFICATION_OPTS)
39 
40 static unsigned dm_verity_prefetch_cluster = DM_VERITY_DEFAULT_PREFETCH_SIZE;
41 
42 module_param_named(prefetch_cluster, dm_verity_prefetch_cluster, uint, S_IRUGO | S_IWUSR);
43 
44 struct dm_verity_prefetch_work {
45 	struct work_struct work;
46 	struct dm_verity *v;
47 	sector_t block;
48 	unsigned n_blocks;
49 };
50 
51 /*
52  * Auxiliary structure appended to each dm-bufio buffer. If the value
53  * hash_verified is nonzero, hash of the block has been verified.
54  *
55  * The variable hash_verified is set to 0 when allocating the buffer, then
56  * it can be changed to 1 and it is never reset to 0 again.
57  *
58  * There is no lock around this value, a race condition can at worst cause
59  * that multiple processes verify the hash of the same buffer simultaneously
60  * and write 1 to hash_verified simultaneously.
61  * This condition is harmless, so we don't need locking.
62  */
63 struct buffer_aux {
64 	int hash_verified;
65 };
66 
67 /*
68  * Initialize struct buffer_aux for a freshly created buffer.
69  */
70 static void dm_bufio_alloc_callback(struct dm_buffer *buf)
71 {
72 	struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
73 
74 	aux->hash_verified = 0;
75 }
76 
77 /*
78  * Translate input sector number to the sector number on the target device.
79  */
80 static sector_t verity_map_sector(struct dm_verity *v, sector_t bi_sector)
81 {
82 	return v->data_start + dm_target_offset(v->ti, bi_sector);
83 }
84 
85 /*
86  * Return hash position of a specified block at a specified tree level
87  * (0 is the lowest level).
88  * The lowest "hash_per_block_bits"-bits of the result denote hash position
89  * inside a hash block. The remaining bits denote location of the hash block.
90  */
91 static sector_t verity_position_at_level(struct dm_verity *v, sector_t block,
92 					 int level)
93 {
94 	return block >> (level * v->hash_per_block_bits);
95 }
96 
97 static int verity_hash_update(struct dm_verity *v, struct ahash_request *req,
98 				const u8 *data, size_t len,
99 				struct crypto_wait *wait)
100 {
101 	struct scatterlist sg;
102 
103 	if (likely(!is_vmalloc_addr(data))) {
104 		sg_init_one(&sg, data, len);
105 		ahash_request_set_crypt(req, &sg, NULL, len);
106 		return crypto_wait_req(crypto_ahash_update(req), wait);
107 	} else {
108 		do {
109 			int r;
110 			size_t this_step = min_t(size_t, len, PAGE_SIZE - offset_in_page(data));
111 			flush_kernel_vmap_range((void *)data, this_step);
112 			sg_init_table(&sg, 1);
113 			sg_set_page(&sg, vmalloc_to_page(data), this_step, offset_in_page(data));
114 			ahash_request_set_crypt(req, &sg, NULL, this_step);
115 			r = crypto_wait_req(crypto_ahash_update(req), wait);
116 			if (unlikely(r))
117 				return r;
118 			data += this_step;
119 			len -= this_step;
120 		} while (len);
121 		return 0;
122 	}
123 }
124 
125 /*
126  * Wrapper for crypto_ahash_init, which handles verity salting.
127  */
128 static int verity_hash_init(struct dm_verity *v, struct ahash_request *req,
129 				struct crypto_wait *wait)
130 {
131 	int r;
132 
133 	ahash_request_set_tfm(req, v->tfm);
134 	ahash_request_set_callback(req, CRYPTO_TFM_REQ_MAY_SLEEP |
135 					CRYPTO_TFM_REQ_MAY_BACKLOG,
136 					crypto_req_done, (void *)wait);
137 	crypto_init_wait(wait);
138 
139 	r = crypto_wait_req(crypto_ahash_init(req), wait);
140 
141 	if (unlikely(r < 0)) {
142 		DMERR("crypto_ahash_init failed: %d", r);
143 		return r;
144 	}
145 
146 	if (likely(v->salt_size && (v->version >= 1)))
147 		r = verity_hash_update(v, req, v->salt, v->salt_size, wait);
148 
149 	return r;
150 }
151 
152 static int verity_hash_final(struct dm_verity *v, struct ahash_request *req,
153 			     u8 *digest, struct crypto_wait *wait)
154 {
155 	int r;
156 
157 	if (unlikely(v->salt_size && (!v->version))) {
158 		r = verity_hash_update(v, req, v->salt, v->salt_size, wait);
159 
160 		if (r < 0) {
161 			DMERR("verity_hash_final failed updating salt: %d", r);
162 			goto out;
163 		}
164 	}
165 
166 	ahash_request_set_crypt(req, NULL, digest, 0);
167 	r = crypto_wait_req(crypto_ahash_final(req), wait);
168 out:
169 	return r;
170 }
171 
172 int verity_hash(struct dm_verity *v, struct ahash_request *req,
173 		const u8 *data, size_t len, u8 *digest)
174 {
175 	int r;
176 	struct crypto_wait wait;
177 
178 	r = verity_hash_init(v, req, &wait);
179 	if (unlikely(r < 0))
180 		goto out;
181 
182 	r = verity_hash_update(v, req, data, len, &wait);
183 	if (unlikely(r < 0))
184 		goto out;
185 
186 	r = verity_hash_final(v, req, digest, &wait);
187 
188 out:
189 	return r;
190 }
191 
192 static void verity_hash_at_level(struct dm_verity *v, sector_t block, int level,
193 				 sector_t *hash_block, unsigned *offset)
194 {
195 	sector_t position = verity_position_at_level(v, block, level);
196 	unsigned idx;
197 
198 	*hash_block = v->hash_level_block[level] + (position >> v->hash_per_block_bits);
199 
200 	if (!offset)
201 		return;
202 
203 	idx = position & ((1 << v->hash_per_block_bits) - 1);
204 	if (!v->version)
205 		*offset = idx * v->digest_size;
206 	else
207 		*offset = idx << (v->hash_dev_block_bits - v->hash_per_block_bits);
208 }
209 
210 /*
211  * Handle verification errors.
212  */
213 static int verity_handle_err(struct dm_verity *v, enum verity_block_type type,
214 			     unsigned long long block)
215 {
216 	char verity_env[DM_VERITY_ENV_LENGTH];
217 	char *envp[] = { verity_env, NULL };
218 	const char *type_str = "";
219 	struct mapped_device *md = dm_table_get_md(v->ti->table);
220 
221 	/* Corruption should be visible in device status in all modes */
222 	v->hash_failed = 1;
223 
224 	if (v->corrupted_errs >= DM_VERITY_MAX_CORRUPTED_ERRS)
225 		goto out;
226 
227 	v->corrupted_errs++;
228 
229 	switch (type) {
230 	case DM_VERITY_BLOCK_TYPE_DATA:
231 		type_str = "data";
232 		break;
233 	case DM_VERITY_BLOCK_TYPE_METADATA:
234 		type_str = "metadata";
235 		break;
236 	default:
237 		BUG();
238 	}
239 
240 	DMERR_LIMIT("%s: %s block %llu is corrupted", v->data_dev->name,
241 		    type_str, block);
242 
243 	if (v->corrupted_errs == DM_VERITY_MAX_CORRUPTED_ERRS)
244 		DMERR("%s: reached maximum errors", v->data_dev->name);
245 
246 	snprintf(verity_env, DM_VERITY_ENV_LENGTH, "%s=%d,%llu",
247 		DM_VERITY_ENV_VAR_NAME, type, block);
248 
249 	kobject_uevent_env(&disk_to_dev(dm_disk(md))->kobj, KOBJ_CHANGE, envp);
250 
251 out:
252 	if (v->mode == DM_VERITY_MODE_LOGGING)
253 		return 0;
254 
255 	if (v->mode == DM_VERITY_MODE_RESTART)
256 		kernel_restart("dm-verity device corrupted");
257 
258 	if (v->mode == DM_VERITY_MODE_PANIC)
259 		panic("dm-verity device corrupted");
260 
261 	return 1;
262 }
263 
264 /*
265  * Verify hash of a metadata block pertaining to the specified data block
266  * ("block" argument) at a specified level ("level" argument).
267  *
268  * On successful return, verity_io_want_digest(v, io) contains the hash value
269  * for a lower tree level or for the data block (if we're at the lowest level).
270  *
271  * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned.
272  * If "skip_unverified" is false, unverified buffer is hashed and verified
273  * against current value of verity_io_want_digest(v, io).
274  */
275 static int verity_verify_level(struct dm_verity *v, struct dm_verity_io *io,
276 			       sector_t block, int level, bool skip_unverified,
277 			       u8 *want_digest)
278 {
279 	struct dm_buffer *buf;
280 	struct buffer_aux *aux;
281 	u8 *data;
282 	int r;
283 	sector_t hash_block;
284 	unsigned offset;
285 
286 	verity_hash_at_level(v, block, level, &hash_block, &offset);
287 
288 	data = dm_bufio_read(v->bufio, hash_block, &buf);
289 	if (IS_ERR(data))
290 		return PTR_ERR(data);
291 
292 	aux = dm_bufio_get_aux_data(buf);
293 
294 	if (!aux->hash_verified) {
295 		if (skip_unverified) {
296 			r = 1;
297 			goto release_ret_r;
298 		}
299 
300 		r = verity_hash(v, verity_io_hash_req(v, io),
301 				data, 1 << v->hash_dev_block_bits,
302 				verity_io_real_digest(v, io));
303 		if (unlikely(r < 0))
304 			goto release_ret_r;
305 
306 		if (likely(memcmp(verity_io_real_digest(v, io), want_digest,
307 				  v->digest_size) == 0))
308 			aux->hash_verified = 1;
309 		else if (verity_fec_decode(v, io,
310 					   DM_VERITY_BLOCK_TYPE_METADATA,
311 					   hash_block, data, NULL) == 0)
312 			aux->hash_verified = 1;
313 		else if (verity_handle_err(v,
314 					   DM_VERITY_BLOCK_TYPE_METADATA,
315 					   hash_block)) {
316 			r = -EIO;
317 			goto release_ret_r;
318 		}
319 	}
320 
321 	data += offset;
322 	memcpy(want_digest, data, v->digest_size);
323 	r = 0;
324 
325 release_ret_r:
326 	dm_bufio_release(buf);
327 	return r;
328 }
329 
330 /*
331  * Find a hash for a given block, write it to digest and verify the integrity
332  * of the hash tree if necessary.
333  */
334 int verity_hash_for_block(struct dm_verity *v, struct dm_verity_io *io,
335 			  sector_t block, u8 *digest, bool *is_zero)
336 {
337 	int r = 0, i;
338 
339 	if (likely(v->levels)) {
340 		/*
341 		 * First, we try to get the requested hash for
342 		 * the current block. If the hash block itself is
343 		 * verified, zero is returned. If it isn't, this
344 		 * function returns 1 and we fall back to whole
345 		 * chain verification.
346 		 */
347 		r = verity_verify_level(v, io, block, 0, true, digest);
348 		if (likely(r <= 0))
349 			goto out;
350 	}
351 
352 	memcpy(digest, v->root_digest, v->digest_size);
353 
354 	for (i = v->levels - 1; i >= 0; i--) {
355 		r = verity_verify_level(v, io, block, i, false, digest);
356 		if (unlikely(r))
357 			goto out;
358 	}
359 out:
360 	if (!r && v->zero_digest)
361 		*is_zero = !memcmp(v->zero_digest, digest, v->digest_size);
362 	else
363 		*is_zero = false;
364 
365 	return r;
366 }
367 
368 /*
369  * Calculates the digest for the given bio
370  */
371 static int verity_for_io_block(struct dm_verity *v, struct dm_verity_io *io,
372 			       struct bvec_iter *iter, struct crypto_wait *wait)
373 {
374 	unsigned int todo = 1 << v->data_dev_block_bits;
375 	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
376 	struct scatterlist sg;
377 	struct ahash_request *req = verity_io_hash_req(v, io);
378 
379 	do {
380 		int r;
381 		unsigned int len;
382 		struct bio_vec bv = bio_iter_iovec(bio, *iter);
383 
384 		sg_init_table(&sg, 1);
385 
386 		len = bv.bv_len;
387 
388 		if (likely(len >= todo))
389 			len = todo;
390 		/*
391 		 * Operating on a single page at a time looks suboptimal
392 		 * until you consider the typical block size is 4,096B.
393 		 * Going through this loops twice should be very rare.
394 		 */
395 		sg_set_page(&sg, bv.bv_page, len, bv.bv_offset);
396 		ahash_request_set_crypt(req, &sg, NULL, len);
397 		r = crypto_wait_req(crypto_ahash_update(req), wait);
398 
399 		if (unlikely(r < 0)) {
400 			DMERR("verity_for_io_block crypto op failed: %d", r);
401 			return r;
402 		}
403 
404 		bio_advance_iter(bio, iter, len);
405 		todo -= len;
406 	} while (todo);
407 
408 	return 0;
409 }
410 
411 /*
412  * Calls function process for 1 << v->data_dev_block_bits bytes in the bio_vec
413  * starting from iter.
414  */
415 int verity_for_bv_block(struct dm_verity *v, struct dm_verity_io *io,
416 			struct bvec_iter *iter,
417 			int (*process)(struct dm_verity *v,
418 				       struct dm_verity_io *io, u8 *data,
419 				       size_t len))
420 {
421 	unsigned todo = 1 << v->data_dev_block_bits;
422 	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
423 
424 	do {
425 		int r;
426 		u8 *page;
427 		unsigned len;
428 		struct bio_vec bv = bio_iter_iovec(bio, *iter);
429 
430 		page = kmap_atomic(bv.bv_page);
431 		len = bv.bv_len;
432 
433 		if (likely(len >= todo))
434 			len = todo;
435 
436 		r = process(v, io, page + bv.bv_offset, len);
437 		kunmap_atomic(page);
438 
439 		if (r < 0)
440 			return r;
441 
442 		bio_advance_iter(bio, iter, len);
443 		todo -= len;
444 	} while (todo);
445 
446 	return 0;
447 }
448 
449 static int verity_bv_zero(struct dm_verity *v, struct dm_verity_io *io,
450 			  u8 *data, size_t len)
451 {
452 	memset(data, 0, len);
453 	return 0;
454 }
455 
456 /*
457  * Moves the bio iter one data block forward.
458  */
459 static inline void verity_bv_skip_block(struct dm_verity *v,
460 					struct dm_verity_io *io,
461 					struct bvec_iter *iter)
462 {
463 	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
464 
465 	bio_advance_iter(bio, iter, 1 << v->data_dev_block_bits);
466 }
467 
468 /*
469  * Verify one "dm_verity_io" structure.
470  */
471 static int verity_verify_io(struct dm_verity_io *io)
472 {
473 	bool is_zero;
474 	struct dm_verity *v = io->v;
475 	struct bvec_iter start;
476 	unsigned b;
477 	struct crypto_wait wait;
478 
479 	for (b = 0; b < io->n_blocks; b++) {
480 		int r;
481 		sector_t cur_block = io->block + b;
482 		struct ahash_request *req = verity_io_hash_req(v, io);
483 
484 		if (v->validated_blocks &&
485 		    likely(test_bit(cur_block, v->validated_blocks))) {
486 			verity_bv_skip_block(v, io, &io->iter);
487 			continue;
488 		}
489 
490 		r = verity_hash_for_block(v, io, cur_block,
491 					  verity_io_want_digest(v, io),
492 					  &is_zero);
493 		if (unlikely(r < 0))
494 			return r;
495 
496 		if (is_zero) {
497 			/*
498 			 * If we expect a zero block, don't validate, just
499 			 * return zeros.
500 			 */
501 			r = verity_for_bv_block(v, io, &io->iter,
502 						verity_bv_zero);
503 			if (unlikely(r < 0))
504 				return r;
505 
506 			continue;
507 		}
508 
509 		r = verity_hash_init(v, req, &wait);
510 		if (unlikely(r < 0))
511 			return r;
512 
513 		start = io->iter;
514 		r = verity_for_io_block(v, io, &io->iter, &wait);
515 		if (unlikely(r < 0))
516 			return r;
517 
518 		r = verity_hash_final(v, req, verity_io_real_digest(v, io),
519 					&wait);
520 		if (unlikely(r < 0))
521 			return r;
522 
523 		if (likely(memcmp(verity_io_real_digest(v, io),
524 				  verity_io_want_digest(v, io), v->digest_size) == 0)) {
525 			if (v->validated_blocks)
526 				set_bit(cur_block, v->validated_blocks);
527 			continue;
528 		}
529 		else if (verity_fec_decode(v, io, DM_VERITY_BLOCK_TYPE_DATA,
530 					   cur_block, NULL, &start) == 0)
531 			continue;
532 		else if (verity_handle_err(v, DM_VERITY_BLOCK_TYPE_DATA,
533 					   cur_block))
534 			return -EIO;
535 	}
536 
537 	return 0;
538 }
539 
540 /*
541  * End one "io" structure with a given error.
542  */
543 static void verity_finish_io(struct dm_verity_io *io, blk_status_t status)
544 {
545 	struct dm_verity *v = io->v;
546 	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
547 
548 	bio->bi_end_io = io->orig_bi_end_io;
549 	bio->bi_status = status;
550 
551 	verity_fec_finish_io(io);
552 
553 	bio_endio(bio);
554 }
555 
556 static void verity_work(struct work_struct *w)
557 {
558 	struct dm_verity_io *io = container_of(w, struct dm_verity_io, work);
559 
560 	verity_finish_io(io, errno_to_blk_status(verity_verify_io(io)));
561 }
562 
563 static void verity_end_io(struct bio *bio)
564 {
565 	struct dm_verity_io *io = bio->bi_private;
566 
567 	if (bio->bi_status && !verity_fec_is_enabled(io->v)) {
568 		verity_finish_io(io, bio->bi_status);
569 		return;
570 	}
571 
572 	INIT_WORK(&io->work, verity_work);
573 	queue_work(io->v->verify_wq, &io->work);
574 }
575 
576 /*
577  * Prefetch buffers for the specified io.
578  * The root buffer is not prefetched, it is assumed that it will be cached
579  * all the time.
580  */
581 static void verity_prefetch_io(struct work_struct *work)
582 {
583 	struct dm_verity_prefetch_work *pw =
584 		container_of(work, struct dm_verity_prefetch_work, work);
585 	struct dm_verity *v = pw->v;
586 	int i;
587 
588 	for (i = v->levels - 2; i >= 0; i--) {
589 		sector_t hash_block_start;
590 		sector_t hash_block_end;
591 		verity_hash_at_level(v, pw->block, i, &hash_block_start, NULL);
592 		verity_hash_at_level(v, pw->block + pw->n_blocks - 1, i, &hash_block_end, NULL);
593 		if (!i) {
594 			unsigned cluster = READ_ONCE(dm_verity_prefetch_cluster);
595 
596 			cluster >>= v->data_dev_block_bits;
597 			if (unlikely(!cluster))
598 				goto no_prefetch_cluster;
599 
600 			if (unlikely(cluster & (cluster - 1)))
601 				cluster = 1 << __fls(cluster);
602 
603 			hash_block_start &= ~(sector_t)(cluster - 1);
604 			hash_block_end |= cluster - 1;
605 			if (unlikely(hash_block_end >= v->hash_blocks))
606 				hash_block_end = v->hash_blocks - 1;
607 		}
608 no_prefetch_cluster:
609 		dm_bufio_prefetch(v->bufio, hash_block_start,
610 				  hash_block_end - hash_block_start + 1);
611 	}
612 
613 	kfree(pw);
614 }
615 
616 static void verity_submit_prefetch(struct dm_verity *v, struct dm_verity_io *io)
617 {
618 	sector_t block = io->block;
619 	unsigned int n_blocks = io->n_blocks;
620 	struct dm_verity_prefetch_work *pw;
621 
622 	if (v->validated_blocks) {
623 		while (n_blocks && test_bit(block, v->validated_blocks)) {
624 			block++;
625 			n_blocks--;
626 		}
627 		while (n_blocks && test_bit(block + n_blocks - 1,
628 					    v->validated_blocks))
629 			n_blocks--;
630 		if (!n_blocks)
631 			return;
632 	}
633 
634 	pw = kmalloc(sizeof(struct dm_verity_prefetch_work),
635 		GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
636 
637 	if (!pw)
638 		return;
639 
640 	INIT_WORK(&pw->work, verity_prefetch_io);
641 	pw->v = v;
642 	pw->block = block;
643 	pw->n_blocks = n_blocks;
644 	queue_work(v->verify_wq, &pw->work);
645 }
646 
647 /*
648  * Bio map function. It allocates dm_verity_io structure and bio vector and
649  * fills them. Then it issues prefetches and the I/O.
650  */
651 static int verity_map(struct dm_target *ti, struct bio *bio)
652 {
653 	struct dm_verity *v = ti->private;
654 	struct dm_verity_io *io;
655 
656 	bio_set_dev(bio, v->data_dev->bdev);
657 	bio->bi_iter.bi_sector = verity_map_sector(v, bio->bi_iter.bi_sector);
658 
659 	if (((unsigned)bio->bi_iter.bi_sector | bio_sectors(bio)) &
660 	    ((1 << (v->data_dev_block_bits - SECTOR_SHIFT)) - 1)) {
661 		DMERR_LIMIT("unaligned io");
662 		return DM_MAPIO_KILL;
663 	}
664 
665 	if (bio_end_sector(bio) >>
666 	    (v->data_dev_block_bits - SECTOR_SHIFT) > v->data_blocks) {
667 		DMERR_LIMIT("io out of range");
668 		return DM_MAPIO_KILL;
669 	}
670 
671 	if (bio_data_dir(bio) == WRITE)
672 		return DM_MAPIO_KILL;
673 
674 	io = dm_per_bio_data(bio, ti->per_io_data_size);
675 	io->v = v;
676 	io->orig_bi_end_io = bio->bi_end_io;
677 	io->block = bio->bi_iter.bi_sector >> (v->data_dev_block_bits - SECTOR_SHIFT);
678 	io->n_blocks = bio->bi_iter.bi_size >> v->data_dev_block_bits;
679 
680 	bio->bi_end_io = verity_end_io;
681 	bio->bi_private = io;
682 	io->iter = bio->bi_iter;
683 
684 	verity_fec_init_io(io);
685 
686 	verity_submit_prefetch(v, io);
687 
688 	submit_bio_noacct(bio);
689 
690 	return DM_MAPIO_SUBMITTED;
691 }
692 
693 /*
694  * Status: V (valid) or C (corruption found)
695  */
696 static void verity_status(struct dm_target *ti, status_type_t type,
697 			  unsigned status_flags, char *result, unsigned maxlen)
698 {
699 	struct dm_verity *v = ti->private;
700 	unsigned args = 0;
701 	unsigned sz = 0;
702 	unsigned x;
703 
704 	switch (type) {
705 	case STATUSTYPE_INFO:
706 		DMEMIT("%c", v->hash_failed ? 'C' : 'V');
707 		break;
708 	case STATUSTYPE_TABLE:
709 		DMEMIT("%u %s %s %u %u %llu %llu %s ",
710 			v->version,
711 			v->data_dev->name,
712 			v->hash_dev->name,
713 			1 << v->data_dev_block_bits,
714 			1 << v->hash_dev_block_bits,
715 			(unsigned long long)v->data_blocks,
716 			(unsigned long long)v->hash_start,
717 			v->alg_name
718 			);
719 		for (x = 0; x < v->digest_size; x++)
720 			DMEMIT("%02x", v->root_digest[x]);
721 		DMEMIT(" ");
722 		if (!v->salt_size)
723 			DMEMIT("-");
724 		else
725 			for (x = 0; x < v->salt_size; x++)
726 				DMEMIT("%02x", v->salt[x]);
727 		if (v->mode != DM_VERITY_MODE_EIO)
728 			args++;
729 		if (verity_fec_is_enabled(v))
730 			args += DM_VERITY_OPTS_FEC;
731 		if (v->zero_digest)
732 			args++;
733 		if (v->validated_blocks)
734 			args++;
735 		if (v->signature_key_desc)
736 			args += DM_VERITY_ROOT_HASH_VERIFICATION_OPTS;
737 		if (!args)
738 			return;
739 		DMEMIT(" %u", args);
740 		if (v->mode != DM_VERITY_MODE_EIO) {
741 			DMEMIT(" ");
742 			switch (v->mode) {
743 			case DM_VERITY_MODE_LOGGING:
744 				DMEMIT(DM_VERITY_OPT_LOGGING);
745 				break;
746 			case DM_VERITY_MODE_RESTART:
747 				DMEMIT(DM_VERITY_OPT_RESTART);
748 				break;
749 			case DM_VERITY_MODE_PANIC:
750 				DMEMIT(DM_VERITY_OPT_PANIC);
751 				break;
752 			default:
753 				BUG();
754 			}
755 		}
756 		if (v->zero_digest)
757 			DMEMIT(" " DM_VERITY_OPT_IGN_ZEROES);
758 		if (v->validated_blocks)
759 			DMEMIT(" " DM_VERITY_OPT_AT_MOST_ONCE);
760 		sz = verity_fec_status_table(v, sz, result, maxlen);
761 		if (v->signature_key_desc)
762 			DMEMIT(" " DM_VERITY_ROOT_HASH_VERIFICATION_OPT_SIG_KEY
763 				" %s", v->signature_key_desc);
764 		break;
765 	}
766 }
767 
768 static int verity_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
769 {
770 	struct dm_verity *v = ti->private;
771 
772 	*bdev = v->data_dev->bdev;
773 
774 	if (v->data_start ||
775 	    ti->len != i_size_read(v->data_dev->bdev->bd_inode) >> SECTOR_SHIFT)
776 		return 1;
777 	return 0;
778 }
779 
780 static int verity_iterate_devices(struct dm_target *ti,
781 				  iterate_devices_callout_fn fn, void *data)
782 {
783 	struct dm_verity *v = ti->private;
784 
785 	return fn(ti, v->data_dev, v->data_start, ti->len, data);
786 }
787 
788 static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)
789 {
790 	struct dm_verity *v = ti->private;
791 
792 	if (limits->logical_block_size < 1 << v->data_dev_block_bits)
793 		limits->logical_block_size = 1 << v->data_dev_block_bits;
794 
795 	if (limits->physical_block_size < 1 << v->data_dev_block_bits)
796 		limits->physical_block_size = 1 << v->data_dev_block_bits;
797 
798 	blk_limits_io_min(limits, limits->logical_block_size);
799 }
800 
801 static void verity_dtr(struct dm_target *ti)
802 {
803 	struct dm_verity *v = ti->private;
804 
805 	if (v->verify_wq)
806 		destroy_workqueue(v->verify_wq);
807 
808 	if (v->bufio)
809 		dm_bufio_client_destroy(v->bufio);
810 
811 	kvfree(v->validated_blocks);
812 	kfree(v->salt);
813 	kfree(v->root_digest);
814 	kfree(v->zero_digest);
815 
816 	if (v->tfm)
817 		crypto_free_ahash(v->tfm);
818 
819 	kfree(v->alg_name);
820 
821 	if (v->hash_dev)
822 		dm_put_device(ti, v->hash_dev);
823 
824 	if (v->data_dev)
825 		dm_put_device(ti, v->data_dev);
826 
827 	verity_fec_dtr(v);
828 
829 	kfree(v->signature_key_desc);
830 
831 	kfree(v);
832 }
833 
834 static int verity_alloc_most_once(struct dm_verity *v)
835 {
836 	struct dm_target *ti = v->ti;
837 
838 	/* the bitset can only handle INT_MAX blocks */
839 	if (v->data_blocks > INT_MAX) {
840 		ti->error = "device too large to use check_at_most_once";
841 		return -E2BIG;
842 	}
843 
844 	v->validated_blocks = kvcalloc(BITS_TO_LONGS(v->data_blocks),
845 				       sizeof(unsigned long),
846 				       GFP_KERNEL);
847 	if (!v->validated_blocks) {
848 		ti->error = "failed to allocate bitset for check_at_most_once";
849 		return -ENOMEM;
850 	}
851 
852 	return 0;
853 }
854 
855 static int verity_alloc_zero_digest(struct dm_verity *v)
856 {
857 	int r = -ENOMEM;
858 	struct ahash_request *req;
859 	u8 *zero_data;
860 
861 	v->zero_digest = kmalloc(v->digest_size, GFP_KERNEL);
862 
863 	if (!v->zero_digest)
864 		return r;
865 
866 	req = kmalloc(v->ahash_reqsize, GFP_KERNEL);
867 
868 	if (!req)
869 		return r; /* verity_dtr will free zero_digest */
870 
871 	zero_data = kzalloc(1 << v->data_dev_block_bits, GFP_KERNEL);
872 
873 	if (!zero_data)
874 		goto out;
875 
876 	r = verity_hash(v, req, zero_data, 1 << v->data_dev_block_bits,
877 			v->zero_digest);
878 
879 out:
880 	kfree(req);
881 	kfree(zero_data);
882 
883 	return r;
884 }
885 
886 static int verity_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
887 				 struct dm_verity_sig_opts *verify_args)
888 {
889 	int r;
890 	unsigned argc;
891 	struct dm_target *ti = v->ti;
892 	const char *arg_name;
893 
894 	static const struct dm_arg _args[] = {
895 		{0, DM_VERITY_OPTS_MAX, "Invalid number of feature args"},
896 	};
897 
898 	r = dm_read_arg_group(_args, as, &argc, &ti->error);
899 	if (r)
900 		return -EINVAL;
901 
902 	if (!argc)
903 		return 0;
904 
905 	do {
906 		arg_name = dm_shift_arg(as);
907 		argc--;
908 
909 		if (!strcasecmp(arg_name, DM_VERITY_OPT_LOGGING)) {
910 			v->mode = DM_VERITY_MODE_LOGGING;
911 			continue;
912 
913 		} else if (!strcasecmp(arg_name, DM_VERITY_OPT_RESTART)) {
914 			v->mode = DM_VERITY_MODE_RESTART;
915 			continue;
916 
917 		} else if (!strcasecmp(arg_name, DM_VERITY_OPT_PANIC)) {
918 			v->mode = DM_VERITY_MODE_PANIC;
919 			continue;
920 
921 		} else if (!strcasecmp(arg_name, DM_VERITY_OPT_IGN_ZEROES)) {
922 			r = verity_alloc_zero_digest(v);
923 			if (r) {
924 				ti->error = "Cannot allocate zero digest";
925 				return r;
926 			}
927 			continue;
928 
929 		} else if (!strcasecmp(arg_name, DM_VERITY_OPT_AT_MOST_ONCE)) {
930 			r = verity_alloc_most_once(v);
931 			if (r)
932 				return r;
933 			continue;
934 
935 		} else if (verity_is_fec_opt_arg(arg_name)) {
936 			r = verity_fec_parse_opt_args(as, v, &argc, arg_name);
937 			if (r)
938 				return r;
939 			continue;
940 		} else if (verity_verify_is_sig_opt_arg(arg_name)) {
941 			r = verity_verify_sig_parse_opt_args(as, v,
942 							     verify_args,
943 							     &argc, arg_name);
944 			if (r)
945 				return r;
946 			continue;
947 
948 		}
949 
950 		ti->error = "Unrecognized verity feature request";
951 		return -EINVAL;
952 	} while (argc && !r);
953 
954 	return r;
955 }
956 
957 /*
958  * Target parameters:
959  *	<version>	The current format is version 1.
960  *			Vsn 0 is compatible with original Chromium OS releases.
961  *	<data device>
962  *	<hash device>
963  *	<data block size>
964  *	<hash block size>
965  *	<the number of data blocks>
966  *	<hash start block>
967  *	<algorithm>
968  *	<digest>
969  *	<salt>		Hex string or "-" if no salt.
970  */
971 static int verity_ctr(struct dm_target *ti, unsigned argc, char **argv)
972 {
973 	struct dm_verity *v;
974 	struct dm_verity_sig_opts verify_args = {0};
975 	struct dm_arg_set as;
976 	unsigned int num;
977 	unsigned long long num_ll;
978 	int r;
979 	int i;
980 	sector_t hash_position;
981 	char dummy;
982 	char *root_hash_digest_to_validate;
983 
984 	v = kzalloc(sizeof(struct dm_verity), GFP_KERNEL);
985 	if (!v) {
986 		ti->error = "Cannot allocate verity structure";
987 		return -ENOMEM;
988 	}
989 	ti->private = v;
990 	v->ti = ti;
991 
992 	r = verity_fec_ctr_alloc(v);
993 	if (r)
994 		goto bad;
995 
996 	if ((dm_table_get_mode(ti->table) & ~FMODE_READ)) {
997 		ti->error = "Device must be readonly";
998 		r = -EINVAL;
999 		goto bad;
1000 	}
1001 
1002 	if (argc < 10) {
1003 		ti->error = "Not enough arguments";
1004 		r = -EINVAL;
1005 		goto bad;
1006 	}
1007 
1008 	if (sscanf(argv[0], "%u%c", &num, &dummy) != 1 ||
1009 	    num > 1) {
1010 		ti->error = "Invalid version";
1011 		r = -EINVAL;
1012 		goto bad;
1013 	}
1014 	v->version = num;
1015 
1016 	r = dm_get_device(ti, argv[1], FMODE_READ, &v->data_dev);
1017 	if (r) {
1018 		ti->error = "Data device lookup failed";
1019 		goto bad;
1020 	}
1021 
1022 	r = dm_get_device(ti, argv[2], FMODE_READ, &v->hash_dev);
1023 	if (r) {
1024 		ti->error = "Hash device lookup failed";
1025 		goto bad;
1026 	}
1027 
1028 	if (sscanf(argv[3], "%u%c", &num, &dummy) != 1 ||
1029 	    !num || (num & (num - 1)) ||
1030 	    num < bdev_logical_block_size(v->data_dev->bdev) ||
1031 	    num > PAGE_SIZE) {
1032 		ti->error = "Invalid data device block size";
1033 		r = -EINVAL;
1034 		goto bad;
1035 	}
1036 	v->data_dev_block_bits = __ffs(num);
1037 
1038 	if (sscanf(argv[4], "%u%c", &num, &dummy) != 1 ||
1039 	    !num || (num & (num - 1)) ||
1040 	    num < bdev_logical_block_size(v->hash_dev->bdev) ||
1041 	    num > INT_MAX) {
1042 		ti->error = "Invalid hash device block size";
1043 		r = -EINVAL;
1044 		goto bad;
1045 	}
1046 	v->hash_dev_block_bits = __ffs(num);
1047 
1048 	if (sscanf(argv[5], "%llu%c", &num_ll, &dummy) != 1 ||
1049 	    (sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
1050 	    >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll) {
1051 		ti->error = "Invalid data blocks";
1052 		r = -EINVAL;
1053 		goto bad;
1054 	}
1055 	v->data_blocks = num_ll;
1056 
1057 	if (ti->len > (v->data_blocks << (v->data_dev_block_bits - SECTOR_SHIFT))) {
1058 		ti->error = "Data device is too small";
1059 		r = -EINVAL;
1060 		goto bad;
1061 	}
1062 
1063 	if (sscanf(argv[6], "%llu%c", &num_ll, &dummy) != 1 ||
1064 	    (sector_t)(num_ll << (v->hash_dev_block_bits - SECTOR_SHIFT))
1065 	    >> (v->hash_dev_block_bits - SECTOR_SHIFT) != num_ll) {
1066 		ti->error = "Invalid hash start";
1067 		r = -EINVAL;
1068 		goto bad;
1069 	}
1070 	v->hash_start = num_ll;
1071 
1072 	v->alg_name = kstrdup(argv[7], GFP_KERNEL);
1073 	if (!v->alg_name) {
1074 		ti->error = "Cannot allocate algorithm name";
1075 		r = -ENOMEM;
1076 		goto bad;
1077 	}
1078 
1079 	v->tfm = crypto_alloc_ahash(v->alg_name, 0, 0);
1080 	if (IS_ERR(v->tfm)) {
1081 		ti->error = "Cannot initialize hash function";
1082 		r = PTR_ERR(v->tfm);
1083 		v->tfm = NULL;
1084 		goto bad;
1085 	}
1086 
1087 	/*
1088 	 * dm-verity performance can vary greatly depending on which hash
1089 	 * algorithm implementation is used.  Help people debug performance
1090 	 * problems by logging the ->cra_driver_name.
1091 	 */
1092 	DMINFO("%s using implementation \"%s\"", v->alg_name,
1093 	       crypto_hash_alg_common(v->tfm)->base.cra_driver_name);
1094 
1095 	v->digest_size = crypto_ahash_digestsize(v->tfm);
1096 	if ((1 << v->hash_dev_block_bits) < v->digest_size * 2) {
1097 		ti->error = "Digest size too big";
1098 		r = -EINVAL;
1099 		goto bad;
1100 	}
1101 	v->ahash_reqsize = sizeof(struct ahash_request) +
1102 		crypto_ahash_reqsize(v->tfm);
1103 
1104 	v->root_digest = kmalloc(v->digest_size, GFP_KERNEL);
1105 	if (!v->root_digest) {
1106 		ti->error = "Cannot allocate root digest";
1107 		r = -ENOMEM;
1108 		goto bad;
1109 	}
1110 	if (strlen(argv[8]) != v->digest_size * 2 ||
1111 	    hex2bin(v->root_digest, argv[8], v->digest_size)) {
1112 		ti->error = "Invalid root digest";
1113 		r = -EINVAL;
1114 		goto bad;
1115 	}
1116 	root_hash_digest_to_validate = argv[8];
1117 
1118 	if (strcmp(argv[9], "-")) {
1119 		v->salt_size = strlen(argv[9]) / 2;
1120 		v->salt = kmalloc(v->salt_size, GFP_KERNEL);
1121 		if (!v->salt) {
1122 			ti->error = "Cannot allocate salt";
1123 			r = -ENOMEM;
1124 			goto bad;
1125 		}
1126 		if (strlen(argv[9]) != v->salt_size * 2 ||
1127 		    hex2bin(v->salt, argv[9], v->salt_size)) {
1128 			ti->error = "Invalid salt";
1129 			r = -EINVAL;
1130 			goto bad;
1131 		}
1132 	}
1133 
1134 	argv += 10;
1135 	argc -= 10;
1136 
1137 	/* Optional parameters */
1138 	if (argc) {
1139 		as.argc = argc;
1140 		as.argv = argv;
1141 
1142 		r = verity_parse_opt_args(&as, v, &verify_args);
1143 		if (r < 0)
1144 			goto bad;
1145 	}
1146 
1147 	/* Root hash signature is  a optional parameter*/
1148 	r = verity_verify_root_hash(root_hash_digest_to_validate,
1149 				    strlen(root_hash_digest_to_validate),
1150 				    verify_args.sig,
1151 				    verify_args.sig_size);
1152 	if (r < 0) {
1153 		ti->error = "Root hash verification failed";
1154 		goto bad;
1155 	}
1156 	v->hash_per_block_bits =
1157 		__fls((1 << v->hash_dev_block_bits) / v->digest_size);
1158 
1159 	v->levels = 0;
1160 	if (v->data_blocks)
1161 		while (v->hash_per_block_bits * v->levels < 64 &&
1162 		       (unsigned long long)(v->data_blocks - 1) >>
1163 		       (v->hash_per_block_bits * v->levels))
1164 			v->levels++;
1165 
1166 	if (v->levels > DM_VERITY_MAX_LEVELS) {
1167 		ti->error = "Too many tree levels";
1168 		r = -E2BIG;
1169 		goto bad;
1170 	}
1171 
1172 	hash_position = v->hash_start;
1173 	for (i = v->levels - 1; i >= 0; i--) {
1174 		sector_t s;
1175 		v->hash_level_block[i] = hash_position;
1176 		s = (v->data_blocks + ((sector_t)1 << ((i + 1) * v->hash_per_block_bits)) - 1)
1177 					>> ((i + 1) * v->hash_per_block_bits);
1178 		if (hash_position + s < hash_position) {
1179 			ti->error = "Hash device offset overflow";
1180 			r = -E2BIG;
1181 			goto bad;
1182 		}
1183 		hash_position += s;
1184 	}
1185 	v->hash_blocks = hash_position;
1186 
1187 	v->bufio = dm_bufio_client_create(v->hash_dev->bdev,
1188 		1 << v->hash_dev_block_bits, 1, sizeof(struct buffer_aux),
1189 		dm_bufio_alloc_callback, NULL);
1190 	if (IS_ERR(v->bufio)) {
1191 		ti->error = "Cannot initialize dm-bufio";
1192 		r = PTR_ERR(v->bufio);
1193 		v->bufio = NULL;
1194 		goto bad;
1195 	}
1196 
1197 	if (dm_bufio_get_device_size(v->bufio) < v->hash_blocks) {
1198 		ti->error = "Hash device is too small";
1199 		r = -E2BIG;
1200 		goto bad;
1201 	}
1202 
1203 	/* WQ_UNBOUND greatly improves performance when running on ramdisk */
1204 	v->verify_wq = alloc_workqueue("kverityd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, num_online_cpus());
1205 	if (!v->verify_wq) {
1206 		ti->error = "Cannot allocate workqueue";
1207 		r = -ENOMEM;
1208 		goto bad;
1209 	}
1210 
1211 	ti->per_io_data_size = sizeof(struct dm_verity_io) +
1212 				v->ahash_reqsize + v->digest_size * 2;
1213 
1214 	r = verity_fec_ctr(v);
1215 	if (r)
1216 		goto bad;
1217 
1218 	ti->per_io_data_size = roundup(ti->per_io_data_size,
1219 				       __alignof__(struct dm_verity_io));
1220 
1221 	verity_verify_sig_opts_cleanup(&verify_args);
1222 
1223 	return 0;
1224 
1225 bad:
1226 
1227 	verity_verify_sig_opts_cleanup(&verify_args);
1228 	verity_dtr(ti);
1229 
1230 	return r;
1231 }
1232 
1233 static struct target_type verity_target = {
1234 	.name		= "verity",
1235 	.version	= {1, 7, 0},
1236 	.module		= THIS_MODULE,
1237 	.ctr		= verity_ctr,
1238 	.dtr		= verity_dtr,
1239 	.map		= verity_map,
1240 	.status		= verity_status,
1241 	.prepare_ioctl	= verity_prepare_ioctl,
1242 	.iterate_devices = verity_iterate_devices,
1243 	.io_hints	= verity_io_hints,
1244 };
1245 
1246 static int __init dm_verity_init(void)
1247 {
1248 	int r;
1249 
1250 	r = dm_register_target(&verity_target);
1251 	if (r < 0)
1252 		DMERR("register failed %d", r);
1253 
1254 	return r;
1255 }
1256 
1257 static void __exit dm_verity_exit(void)
1258 {
1259 	dm_unregister_target(&verity_target);
1260 }
1261 
1262 module_init(dm_verity_init);
1263 module_exit(dm_verity_exit);
1264 
1265 MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>");
1266 MODULE_AUTHOR("Mandeep Baines <msb@chromium.org>");
1267 MODULE_AUTHOR("Will Drewry <wad@chromium.org>");
1268 MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
1269 MODULE_LICENSE("GPL");
1270