Commit | Line | Data |
---|---|---|
a4ffc152 MP |
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 <crypto/hash.h> | |
22 | ||
23 | #define DM_MSG_PREFIX "verity" | |
24 | ||
25 | #define DM_VERITY_IO_VEC_INLINE 16 | |
26 | #define DM_VERITY_MEMPOOL_SIZE 4 | |
27 | #define DM_VERITY_DEFAULT_PREFETCH_SIZE 262144 | |
28 | ||
29 | #define DM_VERITY_MAX_LEVELS 63 | |
30 | ||
31 | static unsigned dm_verity_prefetch_cluster = DM_VERITY_DEFAULT_PREFETCH_SIZE; | |
32 | ||
33 | module_param_named(prefetch_cluster, dm_verity_prefetch_cluster, uint, S_IRUGO | S_IWUSR); | |
34 | ||
35 | struct dm_verity { | |
36 | struct dm_dev *data_dev; | |
37 | struct dm_dev *hash_dev; | |
38 | struct dm_target *ti; | |
39 | struct dm_bufio_client *bufio; | |
40 | char *alg_name; | |
41 | struct crypto_shash *tfm; | |
42 | u8 *root_digest; /* digest of the root block */ | |
43 | u8 *salt; /* salt: its size is salt_size */ | |
44 | unsigned 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 */ | |
57 | ||
a4ffc152 MP |
58 | mempool_t *vec_mempool; /* mempool of bio vector */ |
59 | ||
60 | struct workqueue_struct *verify_wq; | |
61 | ||
62 | /* starting blocks for each tree level. 0 is the lowest level. */ | |
63 | sector_t hash_level_block[DM_VERITY_MAX_LEVELS]; | |
64 | }; | |
65 | ||
66 | struct dm_verity_io { | |
67 | struct dm_verity *v; | |
a4ffc152 MP |
68 | |
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; | |
72 | ||
73 | sector_t block; | |
74 | unsigned n_blocks; | |
75 | ||
76 | /* saved bio vector */ | |
77 | struct bio_vec *io_vec; | |
78 | unsigned io_vec_size; | |
79 | ||
80 | struct work_struct work; | |
81 | ||
82 | /* A space for short vectors; longer vectors are allocated separately. */ | |
83 | struct bio_vec io_vec_inline[DM_VERITY_IO_VEC_INLINE]; | |
84 | ||
85 | /* | |
86 | * Three variably-size fields follow this struct: | |
87 | * | |
88 | * u8 hash_desc[v->shash_descsize]; | |
89 | * u8 real_digest[v->digest_size]; | |
90 | * u8 want_digest[v->digest_size]; | |
91 | * | |
92 | * To access them use: io_hash_desc(), io_real_digest() and io_want_digest(). | |
93 | */ | |
94 | }; | |
95 | ||
3b6b7813 MP |
96 | struct dm_verity_prefetch_work { |
97 | struct work_struct work; | |
98 | struct dm_verity *v; | |
99 | sector_t block; | |
100 | unsigned n_blocks; | |
101 | }; | |
102 | ||
a4ffc152 MP |
103 | static struct shash_desc *io_hash_desc(struct dm_verity *v, struct dm_verity_io *io) |
104 | { | |
105 | return (struct shash_desc *)(io + 1); | |
106 | } | |
107 | ||
108 | static u8 *io_real_digest(struct dm_verity *v, struct dm_verity_io *io) | |
109 | { | |
110 | return (u8 *)(io + 1) + v->shash_descsize; | |
111 | } | |
112 | ||
113 | static u8 *io_want_digest(struct dm_verity *v, struct dm_verity_io *io) | |
114 | { | |
115 | return (u8 *)(io + 1) + v->shash_descsize + v->digest_size; | |
116 | } | |
117 | ||
118 | /* | |
119 | * Auxiliary structure appended to each dm-bufio buffer. If the value | |
120 | * hash_verified is nonzero, hash of the block has been verified. | |
121 | * | |
122 | * The variable hash_verified is set to 0 when allocating the buffer, then | |
123 | * it can be changed to 1 and it is never reset to 0 again. | |
124 | * | |
125 | * There is no lock around this value, a race condition can at worst cause | |
126 | * that multiple processes verify the hash of the same buffer simultaneously | |
127 | * and write 1 to hash_verified simultaneously. | |
128 | * This condition is harmless, so we don't need locking. | |
129 | */ | |
130 | struct buffer_aux { | |
131 | int hash_verified; | |
132 | }; | |
133 | ||
134 | /* | |
135 | * Initialize struct buffer_aux for a freshly created buffer. | |
136 | */ | |
137 | static void dm_bufio_alloc_callback(struct dm_buffer *buf) | |
138 | { | |
139 | struct buffer_aux *aux = dm_bufio_get_aux_data(buf); | |
140 | ||
141 | aux->hash_verified = 0; | |
142 | } | |
143 | ||
144 | /* | |
145 | * Translate input sector number to the sector number on the target device. | |
146 | */ | |
147 | static sector_t verity_map_sector(struct dm_verity *v, sector_t bi_sector) | |
148 | { | |
149 | return v->data_start + dm_target_offset(v->ti, bi_sector); | |
150 | } | |
151 | ||
152 | /* | |
153 | * Return hash position of a specified block at a specified tree level | |
154 | * (0 is the lowest level). | |
155 | * The lowest "hash_per_block_bits"-bits of the result denote hash position | |
156 | * inside a hash block. The remaining bits denote location of the hash block. | |
157 | */ | |
158 | static sector_t verity_position_at_level(struct dm_verity *v, sector_t block, | |
159 | int level) | |
160 | { | |
161 | return block >> (level * v->hash_per_block_bits); | |
162 | } | |
163 | ||
164 | static void verity_hash_at_level(struct dm_verity *v, sector_t block, int level, | |
165 | sector_t *hash_block, unsigned *offset) | |
166 | { | |
167 | sector_t position = verity_position_at_level(v, block, level); | |
168 | unsigned idx; | |
169 | ||
170 | *hash_block = v->hash_level_block[level] + (position >> v->hash_per_block_bits); | |
171 | ||
172 | if (!offset) | |
173 | return; | |
174 | ||
175 | idx = position & ((1 << v->hash_per_block_bits) - 1); | |
176 | if (!v->version) | |
177 | *offset = idx * v->digest_size; | |
178 | else | |
179 | *offset = idx << (v->hash_dev_block_bits - v->hash_per_block_bits); | |
180 | } | |
181 | ||
182 | /* | |
183 | * Verify hash of a metadata block pertaining to the specified data block | |
184 | * ("block" argument) at a specified level ("level" argument). | |
185 | * | |
186 | * On successful return, io_want_digest(v, io) contains the hash value for | |
187 | * a lower tree level or for the data block (if we're at the lowest leve). | |
188 | * | |
189 | * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned. | |
190 | * If "skip_unverified" is false, unverified buffer is hashed and verified | |
191 | * against current value of io_want_digest(v, io). | |
192 | */ | |
193 | static int verity_verify_level(struct dm_verity_io *io, sector_t block, | |
194 | int level, bool skip_unverified) | |
195 | { | |
196 | struct dm_verity *v = io->v; | |
197 | struct dm_buffer *buf; | |
198 | struct buffer_aux *aux; | |
199 | u8 *data; | |
200 | int r; | |
201 | sector_t hash_block; | |
202 | unsigned offset; | |
203 | ||
204 | verity_hash_at_level(v, block, level, &hash_block, &offset); | |
205 | ||
206 | data = dm_bufio_read(v->bufio, hash_block, &buf); | |
207 | if (unlikely(IS_ERR(data))) | |
208 | return PTR_ERR(data); | |
209 | ||
210 | aux = dm_bufio_get_aux_data(buf); | |
211 | ||
212 | if (!aux->hash_verified) { | |
213 | struct shash_desc *desc; | |
214 | u8 *result; | |
215 | ||
216 | if (skip_unverified) { | |
217 | r = 1; | |
218 | goto release_ret_r; | |
219 | } | |
220 | ||
221 | desc = io_hash_desc(v, io); | |
222 | desc->tfm = v->tfm; | |
223 | desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP; | |
224 | r = crypto_shash_init(desc); | |
225 | if (r < 0) { | |
226 | DMERR("crypto_shash_init failed: %d", r); | |
227 | goto release_ret_r; | |
228 | } | |
229 | ||
230 | if (likely(v->version >= 1)) { | |
231 | r = crypto_shash_update(desc, v->salt, v->salt_size); | |
232 | if (r < 0) { | |
233 | DMERR("crypto_shash_update failed: %d", r); | |
234 | goto release_ret_r; | |
235 | } | |
236 | } | |
237 | ||
238 | r = crypto_shash_update(desc, data, 1 << v->hash_dev_block_bits); | |
239 | if (r < 0) { | |
240 | DMERR("crypto_shash_update failed: %d", r); | |
241 | goto release_ret_r; | |
242 | } | |
243 | ||
244 | if (!v->version) { | |
245 | r = crypto_shash_update(desc, v->salt, v->salt_size); | |
246 | if (r < 0) { | |
247 | DMERR("crypto_shash_update failed: %d", r); | |
248 | goto release_ret_r; | |
249 | } | |
250 | } | |
251 | ||
252 | result = io_real_digest(v, io); | |
253 | r = crypto_shash_final(desc, result); | |
254 | if (r < 0) { | |
255 | DMERR("crypto_shash_final failed: %d", r); | |
256 | goto release_ret_r; | |
257 | } | |
258 | if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) { | |
259 | DMERR_LIMIT("metadata block %llu is corrupted", | |
260 | (unsigned long long)hash_block); | |
261 | v->hash_failed = 1; | |
262 | r = -EIO; | |
263 | goto release_ret_r; | |
264 | } else | |
265 | aux->hash_verified = 1; | |
266 | } | |
267 | ||
268 | data += offset; | |
269 | ||
270 | memcpy(io_want_digest(v, io), data, v->digest_size); | |
271 | ||
272 | dm_bufio_release(buf); | |
273 | return 0; | |
274 | ||
275 | release_ret_r: | |
276 | dm_bufio_release(buf); | |
277 | ||
278 | return r; | |
279 | } | |
280 | ||
281 | /* | |
282 | * Verify one "dm_verity_io" structure. | |
283 | */ | |
284 | static int verity_verify_io(struct dm_verity_io *io) | |
285 | { | |
286 | struct dm_verity *v = io->v; | |
287 | unsigned b; | |
288 | int i; | |
289 | unsigned vector = 0, offset = 0; | |
290 | ||
291 | for (b = 0; b < io->n_blocks; b++) { | |
292 | struct shash_desc *desc; | |
293 | u8 *result; | |
294 | int r; | |
295 | unsigned todo; | |
296 | ||
297 | if (likely(v->levels)) { | |
298 | /* | |
299 | * First, we try to get the requested hash for | |
300 | * the current block. If the hash block itself is | |
301 | * verified, zero is returned. If it isn't, this | |
302 | * function returns 0 and we fall back to whole | |
303 | * chain verification. | |
304 | */ | |
305 | int r = verity_verify_level(io, io->block + b, 0, true); | |
306 | if (likely(!r)) | |
307 | goto test_block_hash; | |
308 | if (r < 0) | |
309 | return r; | |
310 | } | |
311 | ||
312 | memcpy(io_want_digest(v, io), v->root_digest, v->digest_size); | |
313 | ||
314 | for (i = v->levels - 1; i >= 0; i--) { | |
315 | int r = verity_verify_level(io, io->block + b, i, false); | |
316 | if (unlikely(r)) | |
317 | return r; | |
318 | } | |
319 | ||
320 | test_block_hash: | |
321 | desc = io_hash_desc(v, io); | |
322 | desc->tfm = v->tfm; | |
323 | desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP; | |
324 | r = crypto_shash_init(desc); | |
325 | if (r < 0) { | |
326 | DMERR("crypto_shash_init failed: %d", r); | |
327 | return r; | |
328 | } | |
329 | ||
330 | if (likely(v->version >= 1)) { | |
331 | r = crypto_shash_update(desc, v->salt, v->salt_size); | |
332 | if (r < 0) { | |
333 | DMERR("crypto_shash_update failed: %d", r); | |
334 | return r; | |
335 | } | |
336 | } | |
337 | ||
338 | todo = 1 << v->data_dev_block_bits; | |
339 | do { | |
340 | struct bio_vec *bv; | |
341 | u8 *page; | |
342 | unsigned len; | |
343 | ||
344 | BUG_ON(vector >= io->io_vec_size); | |
345 | bv = &io->io_vec[vector]; | |
346 | page = kmap_atomic(bv->bv_page); | |
347 | len = bv->bv_len - offset; | |
348 | if (likely(len >= todo)) | |
349 | len = todo; | |
350 | r = crypto_shash_update(desc, | |
351 | page + bv->bv_offset + offset, len); | |
352 | kunmap_atomic(page); | |
353 | if (r < 0) { | |
354 | DMERR("crypto_shash_update failed: %d", r); | |
355 | return r; | |
356 | } | |
357 | offset += len; | |
358 | if (likely(offset == bv->bv_len)) { | |
359 | offset = 0; | |
360 | vector++; | |
361 | } | |
362 | todo -= len; | |
363 | } while (todo); | |
364 | ||
365 | if (!v->version) { | |
366 | r = crypto_shash_update(desc, v->salt, v->salt_size); | |
367 | if (r < 0) { | |
368 | DMERR("crypto_shash_update failed: %d", r); | |
369 | return r; | |
370 | } | |
371 | } | |
372 | ||
373 | result = io_real_digest(v, io); | |
374 | r = crypto_shash_final(desc, result); | |
375 | if (r < 0) { | |
376 | DMERR("crypto_shash_final failed: %d", r); | |
377 | return r; | |
378 | } | |
379 | if (unlikely(memcmp(result, io_want_digest(v, io), v->digest_size))) { | |
380 | DMERR_LIMIT("data block %llu is corrupted", | |
381 | (unsigned long long)(io->block + b)); | |
382 | v->hash_failed = 1; | |
383 | return -EIO; | |
384 | } | |
385 | } | |
386 | BUG_ON(vector != io->io_vec_size); | |
387 | BUG_ON(offset); | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
392 | /* | |
393 | * End one "io" structure with a given error. | |
394 | */ | |
395 | static void verity_finish_io(struct dm_verity_io *io, int error) | |
396 | { | |
a4ffc152 | 397 | struct dm_verity *v = io->v; |
e42c3f91 | 398 | struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_bio_data_size); |
a4ffc152 MP |
399 | |
400 | bio->bi_end_io = io->orig_bi_end_io; | |
401 | bio->bi_private = io->orig_bi_private; | |
402 | ||
403 | if (io->io_vec != io->io_vec_inline) | |
404 | mempool_free(io->io_vec, v->vec_mempool); | |
405 | ||
a4ffc152 MP |
406 | bio_endio(bio, error); |
407 | } | |
408 | ||
409 | static void verity_work(struct work_struct *w) | |
410 | { | |
411 | struct dm_verity_io *io = container_of(w, struct dm_verity_io, work); | |
412 | ||
413 | verity_finish_io(io, verity_verify_io(io)); | |
414 | } | |
415 | ||
416 | static void verity_end_io(struct bio *bio, int error) | |
417 | { | |
418 | struct dm_verity_io *io = bio->bi_private; | |
419 | ||
420 | if (error) { | |
421 | verity_finish_io(io, error); | |
422 | return; | |
423 | } | |
424 | ||
425 | INIT_WORK(&io->work, verity_work); | |
426 | queue_work(io->v->verify_wq, &io->work); | |
427 | } | |
428 | ||
429 | /* | |
430 | * Prefetch buffers for the specified io. | |
431 | * The root buffer is not prefetched, it is assumed that it will be cached | |
432 | * all the time. | |
433 | */ | |
3b6b7813 | 434 | static void verity_prefetch_io(struct work_struct *work) |
a4ffc152 | 435 | { |
3b6b7813 MP |
436 | struct dm_verity_prefetch_work *pw = |
437 | container_of(work, struct dm_verity_prefetch_work, work); | |
438 | struct dm_verity *v = pw->v; | |
a4ffc152 MP |
439 | int i; |
440 | ||
441 | for (i = v->levels - 2; i >= 0; i--) { | |
442 | sector_t hash_block_start; | |
443 | sector_t hash_block_end; | |
3b6b7813 MP |
444 | verity_hash_at_level(v, pw->block, i, &hash_block_start, NULL); |
445 | verity_hash_at_level(v, pw->block + pw->n_blocks - 1, i, &hash_block_end, NULL); | |
a4ffc152 | 446 | if (!i) { |
fe5fe906 | 447 | unsigned cluster = ACCESS_ONCE(dm_verity_prefetch_cluster); |
a4ffc152 MP |
448 | |
449 | cluster >>= v->data_dev_block_bits; | |
450 | if (unlikely(!cluster)) | |
451 | goto no_prefetch_cluster; | |
452 | ||
453 | if (unlikely(cluster & (cluster - 1))) | |
553d8fe0 | 454 | cluster = 1 << __fls(cluster); |
a4ffc152 MP |
455 | |
456 | hash_block_start &= ~(sector_t)(cluster - 1); | |
457 | hash_block_end |= cluster - 1; | |
458 | if (unlikely(hash_block_end >= v->hash_blocks)) | |
459 | hash_block_end = v->hash_blocks - 1; | |
460 | } | |
461 | no_prefetch_cluster: | |
462 | dm_bufio_prefetch(v->bufio, hash_block_start, | |
463 | hash_block_end - hash_block_start + 1); | |
464 | } | |
3b6b7813 MP |
465 | |
466 | kfree(pw); | |
467 | } | |
468 | ||
469 | static void verity_submit_prefetch(struct dm_verity *v, struct dm_verity_io *io) | |
470 | { | |
471 | struct dm_verity_prefetch_work *pw; | |
472 | ||
473 | pw = kmalloc(sizeof(struct dm_verity_prefetch_work), | |
474 | GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
475 | ||
476 | if (!pw) | |
477 | return; | |
478 | ||
479 | INIT_WORK(&pw->work, verity_prefetch_io); | |
480 | pw->v = v; | |
481 | pw->block = io->block; | |
482 | pw->n_blocks = io->n_blocks; | |
483 | queue_work(v->verify_wq, &pw->work); | |
a4ffc152 MP |
484 | } |
485 | ||
486 | /* | |
487 | * Bio map function. It allocates dm_verity_io structure and bio vector and | |
488 | * fills them. Then it issues prefetches and the I/O. | |
489 | */ | |
7de3ee57 | 490 | static int verity_map(struct dm_target *ti, struct bio *bio) |
a4ffc152 MP |
491 | { |
492 | struct dm_verity *v = ti->private; | |
493 | struct dm_verity_io *io; | |
494 | ||
495 | bio->bi_bdev = v->data_dev->bdev; | |
496 | bio->bi_sector = verity_map_sector(v, bio->bi_sector); | |
497 | ||
498 | if (((unsigned)bio->bi_sector | bio_sectors(bio)) & | |
499 | ((1 << (v->data_dev_block_bits - SECTOR_SHIFT)) - 1)) { | |
500 | DMERR_LIMIT("unaligned io"); | |
501 | return -EIO; | |
502 | } | |
503 | ||
f73a1c7d | 504 | if (bio_end_sector(bio) >> |
a4ffc152 MP |
505 | (v->data_dev_block_bits - SECTOR_SHIFT) > v->data_blocks) { |
506 | DMERR_LIMIT("io out of range"); | |
507 | return -EIO; | |
508 | } | |
509 | ||
510 | if (bio_data_dir(bio) == WRITE) | |
511 | return -EIO; | |
512 | ||
e42c3f91 | 513 | io = dm_per_bio_data(bio, ti->per_bio_data_size); |
a4ffc152 | 514 | io->v = v; |
a4ffc152 MP |
515 | io->orig_bi_end_io = bio->bi_end_io; |
516 | io->orig_bi_private = bio->bi_private; | |
517 | io->block = bio->bi_sector >> (v->data_dev_block_bits - SECTOR_SHIFT); | |
518 | io->n_blocks = bio->bi_size >> v->data_dev_block_bits; | |
519 | ||
520 | bio->bi_end_io = verity_end_io; | |
521 | bio->bi_private = io; | |
4f2ac93c | 522 | io->io_vec_size = bio_segments(bio); |
a4ffc152 MP |
523 | if (io->io_vec_size < DM_VERITY_IO_VEC_INLINE) |
524 | io->io_vec = io->io_vec_inline; | |
525 | else | |
526 | io->io_vec = mempool_alloc(v->vec_mempool, GFP_NOIO); | |
527 | memcpy(io->io_vec, bio_iovec(bio), | |
528 | io->io_vec_size * sizeof(struct bio_vec)); | |
529 | ||
3b6b7813 | 530 | verity_submit_prefetch(v, io); |
a4ffc152 MP |
531 | |
532 | generic_make_request(bio); | |
533 | ||
534 | return DM_MAPIO_SUBMITTED; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Status: V (valid) or C (corruption found) | |
539 | */ | |
fd7c092e MP |
540 | static void verity_status(struct dm_target *ti, status_type_t type, |
541 | unsigned status_flags, char *result, unsigned maxlen) | |
a4ffc152 MP |
542 | { |
543 | struct dm_verity *v = ti->private; | |
544 | unsigned sz = 0; | |
545 | unsigned x; | |
546 | ||
547 | switch (type) { | |
548 | case STATUSTYPE_INFO: | |
549 | DMEMIT("%c", v->hash_failed ? 'C' : 'V'); | |
550 | break; | |
551 | case STATUSTYPE_TABLE: | |
552 | DMEMIT("%u %s %s %u %u %llu %llu %s ", | |
553 | v->version, | |
554 | v->data_dev->name, | |
555 | v->hash_dev->name, | |
556 | 1 << v->data_dev_block_bits, | |
557 | 1 << v->hash_dev_block_bits, | |
558 | (unsigned long long)v->data_blocks, | |
559 | (unsigned long long)v->hash_start, | |
560 | v->alg_name | |
561 | ); | |
562 | for (x = 0; x < v->digest_size; x++) | |
563 | DMEMIT("%02x", v->root_digest[x]); | |
564 | DMEMIT(" "); | |
565 | if (!v->salt_size) | |
566 | DMEMIT("-"); | |
567 | else | |
568 | for (x = 0; x < v->salt_size; x++) | |
569 | DMEMIT("%02x", v->salt[x]); | |
570 | break; | |
571 | } | |
a4ffc152 MP |
572 | } |
573 | ||
574 | static int verity_ioctl(struct dm_target *ti, unsigned cmd, | |
575 | unsigned long arg) | |
576 | { | |
577 | struct dm_verity *v = ti->private; | |
578 | int r = 0; | |
579 | ||
580 | if (v->data_start || | |
581 | ti->len != i_size_read(v->data_dev->bdev->bd_inode) >> SECTOR_SHIFT) | |
582 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
583 | ||
584 | return r ? : __blkdev_driver_ioctl(v->data_dev->bdev, v->data_dev->mode, | |
585 | cmd, arg); | |
586 | } | |
587 | ||
588 | static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
589 | struct bio_vec *biovec, int max_size) | |
590 | { | |
591 | struct dm_verity *v = ti->private; | |
592 | struct request_queue *q = bdev_get_queue(v->data_dev->bdev); | |
593 | ||
594 | if (!q->merge_bvec_fn) | |
595 | return max_size; | |
596 | ||
597 | bvm->bi_bdev = v->data_dev->bdev; | |
598 | bvm->bi_sector = verity_map_sector(v, bvm->bi_sector); | |
599 | ||
600 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
601 | } | |
602 | ||
603 | static int verity_iterate_devices(struct dm_target *ti, | |
604 | iterate_devices_callout_fn fn, void *data) | |
605 | { | |
606 | struct dm_verity *v = ti->private; | |
607 | ||
608 | return fn(ti, v->data_dev, v->data_start, ti->len, data); | |
609 | } | |
610 | ||
611 | static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
612 | { | |
613 | struct dm_verity *v = ti->private; | |
614 | ||
615 | if (limits->logical_block_size < 1 << v->data_dev_block_bits) | |
616 | limits->logical_block_size = 1 << v->data_dev_block_bits; | |
617 | ||
618 | if (limits->physical_block_size < 1 << v->data_dev_block_bits) | |
619 | limits->physical_block_size = 1 << v->data_dev_block_bits; | |
620 | ||
621 | blk_limits_io_min(limits, limits->logical_block_size); | |
622 | } | |
623 | ||
624 | static void verity_dtr(struct dm_target *ti) | |
625 | { | |
626 | struct dm_verity *v = ti->private; | |
627 | ||
628 | if (v->verify_wq) | |
629 | destroy_workqueue(v->verify_wq); | |
630 | ||
631 | if (v->vec_mempool) | |
632 | mempool_destroy(v->vec_mempool); | |
633 | ||
a4ffc152 MP |
634 | if (v->bufio) |
635 | dm_bufio_client_destroy(v->bufio); | |
636 | ||
637 | kfree(v->salt); | |
638 | kfree(v->root_digest); | |
639 | ||
640 | if (v->tfm) | |
641 | crypto_free_shash(v->tfm); | |
642 | ||
643 | kfree(v->alg_name); | |
644 | ||
645 | if (v->hash_dev) | |
646 | dm_put_device(ti, v->hash_dev); | |
647 | ||
648 | if (v->data_dev) | |
649 | dm_put_device(ti, v->data_dev); | |
650 | ||
651 | kfree(v); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Target parameters: | |
656 | * <version> The current format is version 1. | |
657 | * Vsn 0 is compatible with original Chromium OS releases. | |
658 | * <data device> | |
659 | * <hash device> | |
660 | * <data block size> | |
661 | * <hash block size> | |
662 | * <the number of data blocks> | |
663 | * <hash start block> | |
664 | * <algorithm> | |
665 | * <digest> | |
666 | * <salt> Hex string or "-" if no salt. | |
667 | */ | |
668 | static int verity_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
669 | { | |
670 | struct dm_verity *v; | |
671 | unsigned num; | |
672 | unsigned long long num_ll; | |
673 | int r; | |
674 | int i; | |
675 | sector_t hash_position; | |
676 | char dummy; | |
677 | ||
678 | v = kzalloc(sizeof(struct dm_verity), GFP_KERNEL); | |
679 | if (!v) { | |
680 | ti->error = "Cannot allocate verity structure"; | |
681 | return -ENOMEM; | |
682 | } | |
683 | ti->private = v; | |
684 | v->ti = ti; | |
685 | ||
686 | if ((dm_table_get_mode(ti->table) & ~FMODE_READ)) { | |
687 | ti->error = "Device must be readonly"; | |
688 | r = -EINVAL; | |
689 | goto bad; | |
690 | } | |
691 | ||
692 | if (argc != 10) { | |
693 | ti->error = "Invalid argument count: exactly 10 arguments required"; | |
694 | r = -EINVAL; | |
695 | goto bad; | |
696 | } | |
697 | ||
5d8be843 MP |
698 | if (sscanf(argv[0], "%u%c", &num, &dummy) != 1 || |
699 | num > 1) { | |
a4ffc152 MP |
700 | ti->error = "Invalid version"; |
701 | r = -EINVAL; | |
702 | goto bad; | |
703 | } | |
704 | v->version = num; | |
705 | ||
706 | r = dm_get_device(ti, argv[1], FMODE_READ, &v->data_dev); | |
707 | if (r) { | |
708 | ti->error = "Data device lookup failed"; | |
709 | goto bad; | |
710 | } | |
711 | ||
712 | r = dm_get_device(ti, argv[2], FMODE_READ, &v->hash_dev); | |
713 | if (r) { | |
714 | ti->error = "Data device lookup failed"; | |
715 | goto bad; | |
716 | } | |
717 | ||
718 | if (sscanf(argv[3], "%u%c", &num, &dummy) != 1 || | |
719 | !num || (num & (num - 1)) || | |
720 | num < bdev_logical_block_size(v->data_dev->bdev) || | |
721 | num > PAGE_SIZE) { | |
722 | ti->error = "Invalid data device block size"; | |
723 | r = -EINVAL; | |
724 | goto bad; | |
725 | } | |
553d8fe0 | 726 | v->data_dev_block_bits = __ffs(num); |
a4ffc152 MP |
727 | |
728 | if (sscanf(argv[4], "%u%c", &num, &dummy) != 1 || | |
729 | !num || (num & (num - 1)) || | |
730 | num < bdev_logical_block_size(v->hash_dev->bdev) || | |
731 | num > INT_MAX) { | |
732 | ti->error = "Invalid hash device block size"; | |
733 | r = -EINVAL; | |
734 | goto bad; | |
735 | } | |
553d8fe0 | 736 | v->hash_dev_block_bits = __ffs(num); |
a4ffc152 MP |
737 | |
738 | if (sscanf(argv[5], "%llu%c", &num_ll, &dummy) != 1 || | |
1d55f6bc MP |
739 | (sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) |
740 | >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll) { | |
a4ffc152 MP |
741 | ti->error = "Invalid data blocks"; |
742 | r = -EINVAL; | |
743 | goto bad; | |
744 | } | |
745 | v->data_blocks = num_ll; | |
746 | ||
747 | if (ti->len > (v->data_blocks << (v->data_dev_block_bits - SECTOR_SHIFT))) { | |
748 | ti->error = "Data device is too small"; | |
749 | r = -EINVAL; | |
750 | goto bad; | |
751 | } | |
752 | ||
753 | if (sscanf(argv[6], "%llu%c", &num_ll, &dummy) != 1 || | |
1d55f6bc MP |
754 | (sector_t)(num_ll << (v->hash_dev_block_bits - SECTOR_SHIFT)) |
755 | >> (v->hash_dev_block_bits - SECTOR_SHIFT) != num_ll) { | |
a4ffc152 MP |
756 | ti->error = "Invalid hash start"; |
757 | r = -EINVAL; | |
758 | goto bad; | |
759 | } | |
760 | v->hash_start = num_ll; | |
761 | ||
762 | v->alg_name = kstrdup(argv[7], GFP_KERNEL); | |
763 | if (!v->alg_name) { | |
764 | ti->error = "Cannot allocate algorithm name"; | |
765 | r = -ENOMEM; | |
766 | goto bad; | |
767 | } | |
768 | ||
769 | v->tfm = crypto_alloc_shash(v->alg_name, 0, 0); | |
770 | if (IS_ERR(v->tfm)) { | |
771 | ti->error = "Cannot initialize hash function"; | |
772 | r = PTR_ERR(v->tfm); | |
773 | v->tfm = NULL; | |
774 | goto bad; | |
775 | } | |
776 | v->digest_size = crypto_shash_digestsize(v->tfm); | |
777 | if ((1 << v->hash_dev_block_bits) < v->digest_size * 2) { | |
778 | ti->error = "Digest size too big"; | |
779 | r = -EINVAL; | |
780 | goto bad; | |
781 | } | |
782 | v->shash_descsize = | |
783 | sizeof(struct shash_desc) + crypto_shash_descsize(v->tfm); | |
784 | ||
785 | v->root_digest = kmalloc(v->digest_size, GFP_KERNEL); | |
786 | if (!v->root_digest) { | |
787 | ti->error = "Cannot allocate root digest"; | |
788 | r = -ENOMEM; | |
789 | goto bad; | |
790 | } | |
791 | if (strlen(argv[8]) != v->digest_size * 2 || | |
792 | hex2bin(v->root_digest, argv[8], v->digest_size)) { | |
793 | ti->error = "Invalid root digest"; | |
794 | r = -EINVAL; | |
795 | goto bad; | |
796 | } | |
797 | ||
798 | if (strcmp(argv[9], "-")) { | |
799 | v->salt_size = strlen(argv[9]) / 2; | |
800 | v->salt = kmalloc(v->salt_size, GFP_KERNEL); | |
801 | if (!v->salt) { | |
802 | ti->error = "Cannot allocate salt"; | |
803 | r = -ENOMEM; | |
804 | goto bad; | |
805 | } | |
806 | if (strlen(argv[9]) != v->salt_size * 2 || | |
807 | hex2bin(v->salt, argv[9], v->salt_size)) { | |
808 | ti->error = "Invalid salt"; | |
809 | r = -EINVAL; | |
810 | goto bad; | |
811 | } | |
812 | } | |
813 | ||
814 | v->hash_per_block_bits = | |
553d8fe0 | 815 | __fls((1 << v->hash_dev_block_bits) / v->digest_size); |
a4ffc152 MP |
816 | |
817 | v->levels = 0; | |
818 | if (v->data_blocks) | |
819 | while (v->hash_per_block_bits * v->levels < 64 && | |
820 | (unsigned long long)(v->data_blocks - 1) >> | |
821 | (v->hash_per_block_bits * v->levels)) | |
822 | v->levels++; | |
823 | ||
824 | if (v->levels > DM_VERITY_MAX_LEVELS) { | |
825 | ti->error = "Too many tree levels"; | |
826 | r = -E2BIG; | |
827 | goto bad; | |
828 | } | |
829 | ||
830 | hash_position = v->hash_start; | |
831 | for (i = v->levels - 1; i >= 0; i--) { | |
832 | sector_t s; | |
833 | v->hash_level_block[i] = hash_position; | |
b1bf2de0 MP |
834 | s = (v->data_blocks + ((sector_t)1 << ((i + 1) * v->hash_per_block_bits)) - 1) |
835 | >> ((i + 1) * v->hash_per_block_bits); | |
a4ffc152 MP |
836 | if (hash_position + s < hash_position) { |
837 | ti->error = "Hash device offset overflow"; | |
838 | r = -E2BIG; | |
839 | goto bad; | |
840 | } | |
841 | hash_position += s; | |
842 | } | |
843 | v->hash_blocks = hash_position; | |
844 | ||
845 | v->bufio = dm_bufio_client_create(v->hash_dev->bdev, | |
846 | 1 << v->hash_dev_block_bits, 1, sizeof(struct buffer_aux), | |
847 | dm_bufio_alloc_callback, NULL); | |
848 | if (IS_ERR(v->bufio)) { | |
849 | ti->error = "Cannot initialize dm-bufio"; | |
850 | r = PTR_ERR(v->bufio); | |
851 | v->bufio = NULL; | |
852 | goto bad; | |
853 | } | |
854 | ||
855 | if (dm_bufio_get_device_size(v->bufio) < v->hash_blocks) { | |
856 | ti->error = "Hash device is too small"; | |
857 | r = -E2BIG; | |
858 | goto bad; | |
859 | } | |
860 | ||
e42c3f91 | 861 | ti->per_bio_data_size = roundup(sizeof(struct dm_verity_io) + v->shash_descsize + v->digest_size * 2, __alignof__(struct dm_verity_io)); |
a4ffc152 MP |
862 | |
863 | v->vec_mempool = mempool_create_kmalloc_pool(DM_VERITY_MEMPOOL_SIZE, | |
864 | BIO_MAX_PAGES * sizeof(struct bio_vec)); | |
865 | if (!v->vec_mempool) { | |
866 | ti->error = "Cannot allocate vector mempool"; | |
867 | r = -ENOMEM; | |
868 | goto bad; | |
869 | } | |
870 | ||
871 | /* WQ_UNBOUND greatly improves performance when running on ramdisk */ | |
872 | v->verify_wq = alloc_workqueue("kverityd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, num_online_cpus()); | |
873 | if (!v->verify_wq) { | |
874 | ti->error = "Cannot allocate workqueue"; | |
875 | r = -ENOMEM; | |
876 | goto bad; | |
877 | } | |
878 | ||
879 | return 0; | |
880 | ||
881 | bad: | |
882 | verity_dtr(ti); | |
883 | ||
884 | return r; | |
885 | } | |
886 | ||
887 | static struct target_type verity_target = { | |
888 | .name = "verity", | |
3b6b7813 | 889 | .version = {1, 2, 0}, |
a4ffc152 MP |
890 | .module = THIS_MODULE, |
891 | .ctr = verity_ctr, | |
892 | .dtr = verity_dtr, | |
893 | .map = verity_map, | |
894 | .status = verity_status, | |
895 | .ioctl = verity_ioctl, | |
896 | .merge = verity_merge, | |
897 | .iterate_devices = verity_iterate_devices, | |
898 | .io_hints = verity_io_hints, | |
899 | }; | |
900 | ||
901 | static int __init dm_verity_init(void) | |
902 | { | |
903 | int r; | |
904 | ||
905 | r = dm_register_target(&verity_target); | |
906 | if (r < 0) | |
907 | DMERR("register failed %d", r); | |
908 | ||
909 | return r; | |
910 | } | |
911 | ||
912 | static void __exit dm_verity_exit(void) | |
913 | { | |
914 | dm_unregister_target(&verity_target); | |
915 | } | |
916 | ||
917 | module_init(dm_verity_init); | |
918 | module_exit(dm_verity_exit); | |
919 | ||
920 | MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>"); | |
921 | MODULE_AUTHOR("Mandeep Baines <msb@chromium.org>"); | |
922 | MODULE_AUTHOR("Will Drewry <wad@chromium.org>"); | |
923 | MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking"); | |
924 | MODULE_LICENSE("GPL"); |