drivers/md/dm-verity-target.c

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