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