fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/kthread.h>
  12 #include <linux/slab.h>
  13 #include <linux/migrate.h>
  14 #include <linux/ratelimit.h>
  15 #include <linux/uuid.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/crc32c.h>
  19 #include <linux/sched/mm.h>
  20 #include <asm/unaligned.h>
  21 #include <crypto/hash.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "bio.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "rcu-string.h"
  33 #include "dev-replace.h"
  34 #include "raid56.h"
  35 #include "sysfs.h"
  36 #include "qgroup.h"
  37 #include "compression.h"
  38 #include "tree-checker.h"
  39 #include "ref-verify.h"
  40 #include "block-group.h"
  41 #include "discard.h"
  42 #include "space-info.h"
  43 #include "zoned.h"
  44 #include "subpage.h"
  45 #include "fs.h"
  46 #include "accessors.h"
  47 #include "extent-tree.h"
  48 #include "root-tree.h"
  49 #include "defrag.h"
  50 #include "uuid-tree.h"
  51 #include "relocation.h"
  52 #include "scrub.h"
  53 #include "super.h"
  54
  55 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  56                                  BTRFS_HEADER_FLAG_RELOC |\
  57                                  BTRFS_SUPER_FLAG_ERROR |\
  58                                  BTRFS_SUPER_FLAG_SEEDING |\
  59                                  BTRFS_SUPER_FLAG_METADUMP |\
  60                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  61
  62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  64
  65 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  66 {
  67         if (fs_info->csum_shash)
  68                 crypto_free_shash(fs_info->csum_shash);
  69 }
  70
  71 /*
  72  * Compute the csum of a btree block and store the result to provided buffer.
  73  */
  74 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  75 {
  76         struct btrfs_fs_info *fs_info = buf->fs_info;
  77         const int num_pages = num_extent_pages(buf);
  78         const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  79         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  80         char *kaddr;
  81         int i;
  82
  83         shash->tfm = fs_info->csum_shash;
  84         crypto_shash_init(shash);
  85         kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
  86         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
  87                             first_page_part - BTRFS_CSUM_SIZE);
  88
  89         for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
  90                 kaddr = page_address(buf->pages[i]);
  91                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
  92         }
  93         memset(result, 0, BTRFS_CSUM_SIZE);
  94         crypto_shash_final(shash, result);
  95 }
  96
  97 /*
  98  * we can't consider a given block up to date unless the transid of the
  99  * block matches the transid in the parent node's pointer.  This is how we
 100  * detect blocks that either didn't get written at all or got written
 101  * in the wrong place.
 102  */
 103 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
 104 {
 105         if (!extent_buffer_uptodate(eb))
 106                 return 0;
 107
 108         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 109                 return 1;
 110
 111         if (atomic)
 112                 return -EAGAIN;
 113
 114         if (!extent_buffer_uptodate(eb) ||
 115             btrfs_header_generation(eb) != parent_transid) {
 116                 btrfs_err_rl(eb->fs_info,
 117 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 118                         eb->start, eb->read_mirror,
 119                         parent_transid, btrfs_header_generation(eb));
 120                 clear_extent_buffer_uptodate(eb);
 121                 return 0;
 122         }
 123         return 1;
 124 }
 125
 126 static bool btrfs_supported_super_csum(u16 csum_type)
 127 {
 128         switch (csum_type) {
 129         case BTRFS_CSUM_TYPE_CRC32:
 130         case BTRFS_CSUM_TYPE_XXHASH:
 131         case BTRFS_CSUM_TYPE_SHA256:
 132         case BTRFS_CSUM_TYPE_BLAKE2:
 133                 return true;
 134         default:
 135                 return false;
 136         }
 137 }
 138
 139 /*
 140  * Return 0 if the superblock checksum type matches the checksum value of that
 141  * algorithm. Pass the raw disk superblock data.
 142  */
 143 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 144                            const struct btrfs_super_block *disk_sb)
 145 {
 146         char result[BTRFS_CSUM_SIZE];
 147         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 148
 149         shash->tfm = fs_info->csum_shash;
 150
 151         /*
 152          * The super_block structure does not span the whole
 153          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 154          * filled with zeros and is included in the checksum.
 155          */
 156         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 157                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 158
 159         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 160                 return 1;
 161
 162         return 0;
 163 }
 164
 165 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 166                                       int mirror_num)
 167 {
 168         struct btrfs_fs_info *fs_info = eb->fs_info;
 169         int i, num_pages = num_extent_pages(eb);
 170         int ret = 0;
 171
 172         if (sb_rdonly(fs_info->sb))
 173                 return -EROFS;
 174
 175         for (i = 0; i < num_pages; i++) {
 176                 struct page *p = eb->pages[i];
 177                 u64 start = max_t(u64, eb->start, page_offset(p));
 178                 u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE);
 179                 u32 len = end - start;
 180
 181                 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
 182                                 start, p, offset_in_page(start), mirror_num);
 183                 if (ret)
 184                         break;
 185         }
 186
 187         return ret;
 188 }
 189
 190 /*
 191  * helper to read a given tree block, doing retries as required when
 192  * the checksums don't match and we have alternate mirrors to try.
 193  *
 194  * @check:              expected tree parentness check, see the comments of the
 195  *                      structure for details.
 196  */
 197 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 198                              struct btrfs_tree_parent_check *check)
 199 {
 200         struct btrfs_fs_info *fs_info = eb->fs_info;
 201         int failed = 0;
 202         int ret;
 203         int num_copies = 0;
 204         int mirror_num = 0;
 205         int failed_mirror = 0;
 206
 207         ASSERT(check);
 208
 209         while (1) {
 210                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 211                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
 212                 if (!ret)
 213                         break;
 214
 215                 num_copies = btrfs_num_copies(fs_info,
 216                                               eb->start, eb->len);
 217                 if (num_copies == 1)
 218                         break;
 219
 220                 if (!failed_mirror) {
 221                         failed = 1;
 222                         failed_mirror = eb->read_mirror;
 223                 }
 224
 225                 mirror_num++;
 226                 if (mirror_num == failed_mirror)
 227                         mirror_num++;
 228
 229                 if (mirror_num > num_copies)
 230                         break;
 231         }
 232
 233         if (failed && !ret && failed_mirror)
 234                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 235
 236         return ret;
 237 }
 238
 239 /*
 240  * Checksum a dirty tree block before IO.
 241  */
 242 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 243 {
 244         struct extent_buffer *eb = bbio->private;
 245         struct btrfs_fs_info *fs_info = eb->fs_info;
 246         u64 found_start = btrfs_header_bytenr(eb);
 247         u64 last_trans;
 248         u8 result[BTRFS_CSUM_SIZE];
 249         int ret;
 250
 251         /* Btree blocks are always contiguous on disk. */
 252         if (WARN_ON_ONCE(bbio->file_offset != eb->start))
 253                 return BLK_STS_IOERR;
 254         if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 255                 return BLK_STS_IOERR;
 256
 257         if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
 258                 WARN_ON_ONCE(found_start != 0);
 259                 return BLK_STS_OK;
 260         }
 261
 262         if (WARN_ON_ONCE(found_start != eb->start))
 263                 return BLK_STS_IOERR;
 264         if (WARN_ON(!btrfs_page_test_uptodate(fs_info, eb->pages[0], eb->start,
 265                                               eb->len)))
 266                 return BLK_STS_IOERR;
 267
 268         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 269                                     offsetof(struct btrfs_header, fsid),
 270                                     BTRFS_FSID_SIZE) == 0);
 271         csum_tree_block(eb, result);
 272
 273         if (btrfs_header_level(eb))
 274                 ret = btrfs_check_node(eb);
 275         else
 276                 ret = btrfs_check_leaf(eb);
 277
 278         if (ret < 0)
 279                 goto error;
 280
 281         /*
 282          * Also check the generation, the eb reached here must be newer than
 283          * last committed. Or something seriously wrong happened.
 284          */
 285         last_trans = btrfs_get_last_trans_committed(fs_info);
 286         if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
 287                 ret = -EUCLEAN;
 288                 btrfs_err(fs_info,
 289                         "block=%llu bad generation, have %llu expect > %llu",
 290                           eb->start, btrfs_header_generation(eb), last_trans);
 291                 goto error;
 292         }
 293         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 294         return BLK_STS_OK;
 295
 296 error:
 297         btrfs_print_tree(eb, 0);
 298         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 299                   eb->start);
 300         /*
 301          * Be noisy if this is an extent buffer from a log tree. We don't abort
 302          * a transaction in case there's a bad log tree extent buffer, we just
 303          * fallback to a transaction commit. Still we want to know when there is
 304          * a bad log tree extent buffer, as that may signal a bug somewhere.
 305          */
 306         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 307                 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 308         return errno_to_blk_status(ret);
 309 }
 310
 311 static bool check_tree_block_fsid(struct extent_buffer *eb)
 312 {
 313         struct btrfs_fs_info *fs_info = eb->fs_info;
 314         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 315         u8 fsid[BTRFS_FSID_SIZE];
 316
 317         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 318                            BTRFS_FSID_SIZE);
 319
 320         /*
 321          * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
 322          * This is then overwritten by metadata_uuid if it is present in the
 323          * device_list_add(). The same true for a seed device as well. So use of
 324          * fs_devices::metadata_uuid is appropriate here.
 325          */
 326         if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
 327                 return false;
 328
 329         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 330                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 331                         return false;
 332
 333         return true;
 334 }
 335
 336 /* Do basic extent buffer checks at read time */
 337 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
 338                                  struct btrfs_tree_parent_check *check)
 339 {
 340         struct btrfs_fs_info *fs_info = eb->fs_info;
 341         u64 found_start;
 342         const u32 csum_size = fs_info->csum_size;
 343         u8 found_level;
 344         u8 result[BTRFS_CSUM_SIZE];
 345         const u8 *header_csum;
 346         int ret = 0;
 347
 348         ASSERT(check);
 349
 350         found_start = btrfs_header_bytenr(eb);
 351         if (found_start != eb->start) {
 352                 btrfs_err_rl(fs_info,
 353                         "bad tree block start, mirror %u want %llu have %llu",
 354                              eb->read_mirror, eb->start, found_start);
 355                 ret = -EIO;
 356                 goto out;
 357         }
 358         if (check_tree_block_fsid(eb)) {
 359                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 360                              eb->start, eb->read_mirror);
 361                 ret = -EIO;
 362                 goto out;
 363         }
 364         found_level = btrfs_header_level(eb);
 365         if (found_level >= BTRFS_MAX_LEVEL) {
 366                 btrfs_err(fs_info,
 367                         "bad tree block level, mirror %u level %d on logical %llu",
 368                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 369                 ret = -EIO;
 370                 goto out;
 371         }
 372
 373         csum_tree_block(eb, result);
 374         header_csum = page_address(eb->pages[0]) +
 375                 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
 376
 377         if (memcmp(result, header_csum, csum_size) != 0) {
 378                 btrfs_warn_rl(fs_info,
 379 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 380                               eb->start, eb->read_mirror,
 381                               CSUM_FMT_VALUE(csum_size, header_csum),
 382                               CSUM_FMT_VALUE(csum_size, result),
 383                               btrfs_header_level(eb));
 384                 ret = -EUCLEAN;
 385                 goto out;
 386         }
 387
 388         if (found_level != check->level) {
 389                 btrfs_err(fs_info,
 390                 "level verify failed on logical %llu mirror %u wanted %u found %u",
 391                           eb->start, eb->read_mirror, check->level, found_level);
 392                 ret = -EIO;
 393                 goto out;
 394         }
 395         if (unlikely(check->transid &&
 396                      btrfs_header_generation(eb) != check->transid)) {
 397                 btrfs_err_rl(eb->fs_info,
 398 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 399                                 eb->start, eb->read_mirror, check->transid,
 400                                 btrfs_header_generation(eb));
 401                 ret = -EIO;
 402                 goto out;
 403         }
 404         if (check->has_first_key) {
 405                 struct btrfs_key *expect_key = &check->first_key;
 406                 struct btrfs_key found_key;
 407
 408                 if (found_level)
 409                         btrfs_node_key_to_cpu(eb, &found_key, 0);
 410                 else
 411                         btrfs_item_key_to_cpu(eb, &found_key, 0);
 412                 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 413                         btrfs_err(fs_info,
 414 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 415                                   eb->start, check->transid,
 416                                   expect_key->objectid,
 417                                   expect_key->type, expect_key->offset,
 418                                   found_key.objectid, found_key.type,
 419                                   found_key.offset);
 420                         ret = -EUCLEAN;
 421                         goto out;
 422                 }
 423         }
 424         if (check->owner_root) {
 425                 ret = btrfs_check_eb_owner(eb, check->owner_root);
 426                 if (ret < 0)
 427                         goto out;
 428         }
 429
 430         /*
 431          * If this is a leaf block and it is corrupt, set the corrupt bit so
 432          * that we don't try and read the other copies of this block, just
 433          * return -EIO.
 434          */
 435         if (found_level == 0 && btrfs_check_leaf(eb)) {
 436                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 437                 ret = -EIO;
 438         }
 439
 440         if (found_level > 0 && btrfs_check_node(eb))
 441                 ret = -EIO;
 442
 443         if (ret)
 444                 btrfs_err(fs_info,
 445                 "read time tree block corruption detected on logical %llu mirror %u",
 446                           eb->start, eb->read_mirror);
 447 out:
 448         return ret;
 449 }
 450
 451 #ifdef CONFIG_MIGRATION
 452 static int btree_migrate_folio(struct address_space *mapping,
 453                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 454 {
 455         /*
 456          * we can't safely write a btree page from here,
 457          * we haven't done the locking hook
 458          */
 459         if (folio_test_dirty(src))
 460                 return -EAGAIN;
 461         /*
 462          * Buffers may be managed in a filesystem specific way.
 463          * We must have no buffers or drop them.
 464          */
 465         if (folio_get_private(src) &&
 466             !filemap_release_folio(src, GFP_KERNEL))
 467                 return -EAGAIN;
 468         return migrate_folio(mapping, dst, src, mode);
 469 }
 470 #else
 471 #define btree_migrate_folio NULL
 472 #endif
 473
 474 static int btree_writepages(struct address_space *mapping,
 475                             struct writeback_control *wbc)
 476 {
 477         struct btrfs_fs_info *fs_info;
 478         int ret;
 479
 480         if (wbc->sync_mode == WB_SYNC_NONE) {
 481
 482                 if (wbc->for_kupdate)
 483                         return 0;
 484
 485                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 486                 /* this is a bit racy, but that's ok */
 487                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 488                                              BTRFS_DIRTY_METADATA_THRESH,
 489                                              fs_info->dirty_metadata_batch);
 490                 if (ret < 0)
 491                         return 0;
 492         }
 493         return btree_write_cache_pages(mapping, wbc);
 494 }
 495
 496 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 497 {
 498         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 499                 return false;
 500
 501         return try_release_extent_buffer(&folio->page);
 502 }
 503
 504 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 505                                  size_t length)
 506 {
 507         struct extent_io_tree *tree;
 508         tree = &BTRFS_I(folio->mapping->host)->io_tree;
 509         extent_invalidate_folio(tree, folio, offset);
 510         btree_release_folio(folio, GFP_NOFS);
 511         if (folio_get_private(folio)) {
 512                 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 513                            "folio private not zero on folio %llu",
 514                            (unsigned long long)folio_pos(folio));
 515                 folio_detach_private(folio);
 516         }
 517 }
 518
 519 #ifdef DEBUG
 520 static bool btree_dirty_folio(struct address_space *mapping,
 521                 struct folio *folio)
 522 {
 523         struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 524         struct btrfs_subpage_info *spi = fs_info->subpage_info;
 525         struct btrfs_subpage *subpage;
 526         struct extent_buffer *eb;
 527         int cur_bit = 0;
 528         u64 page_start = folio_pos(folio);
 529
 530         if (fs_info->sectorsize == PAGE_SIZE) {
 531                 eb = folio_get_private(folio);
 532                 BUG_ON(!eb);
 533                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 534                 BUG_ON(!atomic_read(&eb->refs));
 535                 btrfs_assert_tree_write_locked(eb);
 536                 return filemap_dirty_folio(mapping, folio);
 537         }
 538
 539         ASSERT(spi);
 540         subpage = folio_get_private(folio);
 541
 542         for (cur_bit = spi->dirty_offset;
 543              cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
 544              cur_bit++) {
 545                 unsigned long flags;
 546                 u64 cur;
 547
 548                 spin_lock_irqsave(&subpage->lock, flags);
 549                 if (!test_bit(cur_bit, subpage->bitmaps)) {
 550                         spin_unlock_irqrestore(&subpage->lock, flags);
 551                         continue;
 552                 }
 553                 spin_unlock_irqrestore(&subpage->lock, flags);
 554                 cur = page_start + cur_bit * fs_info->sectorsize;
 555
 556                 eb = find_extent_buffer(fs_info, cur);
 557                 ASSERT(eb);
 558                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 559                 ASSERT(atomic_read(&eb->refs));
 560                 btrfs_assert_tree_write_locked(eb);
 561                 free_extent_buffer(eb);
 562
 563                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
 564         }
 565         return filemap_dirty_folio(mapping, folio);
 566 }
 567 #else
 568 #define btree_dirty_folio filemap_dirty_folio
 569 #endif
 570
 571 static const struct address_space_operations btree_aops = {
 572         .writepages     = btree_writepages,
 573         .release_folio  = btree_release_folio,
 574         .invalidate_folio = btree_invalidate_folio,
 575         .migrate_folio  = btree_migrate_folio,
 576         .dirty_folio    = btree_dirty_folio,
 577 };
 578
 579 struct extent_buffer *btrfs_find_create_tree_block(
 580                                                 struct btrfs_fs_info *fs_info,
 581                                                 u64 bytenr, u64 owner_root,
 582                                                 int level)
 583 {
 584         if (btrfs_is_testing(fs_info))
 585                 return alloc_test_extent_buffer(fs_info, bytenr);
 586         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 587 }
 588
 589 /*
 590  * Read tree block at logical address @bytenr and do variant basic but critical
 591  * verification.
 592  *
 593  * @check:              expected tree parentness check, see comments of the
 594  *                      structure for details.
 595  */
 596 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 597                                       struct btrfs_tree_parent_check *check)
 598 {
 599         struct extent_buffer *buf = NULL;
 600         int ret;
 601
 602         ASSERT(check);
 603
 604         buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 605                                            check->level);
 606         if (IS_ERR(buf))
 607                 return buf;
 608
 609         ret = btrfs_read_extent_buffer(buf, check);
 610         if (ret) {
 611                 free_extent_buffer_stale(buf);
 612                 return ERR_PTR(ret);
 613         }
 614         if (btrfs_check_eb_owner(buf, check->owner_root)) {
 615                 free_extent_buffer_stale(buf);
 616                 return ERR_PTR(-EUCLEAN);
 617         }
 618         return buf;
 619
 620 }
 621
 622 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 623                          u64 objectid)
 624 {
 625         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 626
 627         memset(&root->root_key, 0, sizeof(root->root_key));
 628         memset(&root->root_item, 0, sizeof(root->root_item));
 629         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 630         root->fs_info = fs_info;
 631         root->root_key.objectid = objectid;
 632         root->node = NULL;
 633         root->commit_root = NULL;
 634         root->state = 0;
 635         RB_CLEAR_NODE(&root->rb_node);
 636
 637         root->last_trans = 0;
 638         root->free_objectid = 0;
 639         root->nr_delalloc_inodes = 0;
 640         root->nr_ordered_extents = 0;
 641         root->inode_tree = RB_ROOT;
 642         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
 643
 644         btrfs_init_root_block_rsv(root);
 645
 646         INIT_LIST_HEAD(&root->dirty_list);
 647         INIT_LIST_HEAD(&root->root_list);
 648         INIT_LIST_HEAD(&root->delalloc_inodes);
 649         INIT_LIST_HEAD(&root->delalloc_root);
 650         INIT_LIST_HEAD(&root->ordered_extents);
 651         INIT_LIST_HEAD(&root->ordered_root);
 652         INIT_LIST_HEAD(&root->reloc_dirty_list);
 653         INIT_LIST_HEAD(&root->logged_list[0]);
 654         INIT_LIST_HEAD(&root->logged_list[1]);
 655         spin_lock_init(&root->inode_lock);
 656         spin_lock_init(&root->delalloc_lock);
 657         spin_lock_init(&root->ordered_extent_lock);
 658         spin_lock_init(&root->accounting_lock);
 659         spin_lock_init(&root->log_extents_lock[0]);
 660         spin_lock_init(&root->log_extents_lock[1]);
 661         spin_lock_init(&root->qgroup_meta_rsv_lock);
 662         mutex_init(&root->objectid_mutex);
 663         mutex_init(&root->log_mutex);
 664         mutex_init(&root->ordered_extent_mutex);
 665         mutex_init(&root->delalloc_mutex);
 666         init_waitqueue_head(&root->qgroup_flush_wait);
 667         init_waitqueue_head(&root->log_writer_wait);
 668         init_waitqueue_head(&root->log_commit_wait[0]);
 669         init_waitqueue_head(&root->log_commit_wait[1]);
 670         INIT_LIST_HEAD(&root->log_ctxs[0]);
 671         INIT_LIST_HEAD(&root->log_ctxs[1]);
 672         atomic_set(&root->log_commit[0], 0);
 673         atomic_set(&root->log_commit[1], 0);
 674         atomic_set(&root->log_writers, 0);
 675         atomic_set(&root->log_batch, 0);
 676         refcount_set(&root->refs, 1);
 677         atomic_set(&root->snapshot_force_cow, 0);
 678         atomic_set(&root->nr_swapfiles, 0);
 679         btrfs_set_root_log_transid(root, 0);
 680         root->log_transid_committed = -1;
 681         btrfs_set_root_last_log_commit(root, 0);
 682         root->anon_dev = 0;
 683         if (!dummy) {
 684                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
 685                                     IO_TREE_ROOT_DIRTY_LOG_PAGES);
 686                 extent_io_tree_init(fs_info, &root->log_csum_range,
 687                                     IO_TREE_LOG_CSUM_RANGE);
 688         }
 689
 690         spin_lock_init(&root->root_item_lock);
 691         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 692 #ifdef CONFIG_BTRFS_DEBUG
 693         INIT_LIST_HEAD(&root->leak_list);
 694         spin_lock(&fs_info->fs_roots_radix_lock);
 695         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 696         spin_unlock(&fs_info->fs_roots_radix_lock);
 697 #endif
 698 }
 699
 700 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 701                                            u64 objectid, gfp_t flags)
 702 {
 703         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
 704         if (root)
 705                 __setup_root(root, fs_info, objectid);
 706         return root;
 707 }
 708
 709 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 710 /* Should only be used by the testing infrastructure */
 711 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 712 {
 713         struct btrfs_root *root;
 714
 715         if (!fs_info)
 716                 return ERR_PTR(-EINVAL);
 717
 718         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 719         if (!root)
 720                 return ERR_PTR(-ENOMEM);
 721
 722         /* We don't use the stripesize in selftest, set it as sectorsize */
 723         root->alloc_bytenr = 0;
 724
 725         return root;
 726 }
 727 #endif
 728
 729 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 730 {
 731         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 732         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 733
 734         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 735 }
 736
 737 static int global_root_key_cmp(const void *k, const struct rb_node *node)
 738 {
 739         const struct btrfs_key *key = k;
 740         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 741
 742         return btrfs_comp_cpu_keys(key, &root->root_key);
 743 }
 744
 745 int btrfs_global_root_insert(struct btrfs_root *root)
 746 {
 747         struct btrfs_fs_info *fs_info = root->fs_info;
 748         struct rb_node *tmp;
 749         int ret = 0;
 750
 751         write_lock(&fs_info->global_root_lock);
 752         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
 753         write_unlock(&fs_info->global_root_lock);
 754
 755         if (tmp) {
 756                 ret = -EEXIST;
 757                 btrfs_warn(fs_info, "global root %llu %llu already exists",
 758                                 root->root_key.objectid, root->root_key.offset);
 759         }
 760         return ret;
 761 }
 762
 763 void btrfs_global_root_delete(struct btrfs_root *root)
 764 {
 765         struct btrfs_fs_info *fs_info = root->fs_info;
 766
 767         write_lock(&fs_info->global_root_lock);
 768         rb_erase(&root->rb_node, &fs_info->global_root_tree);
 769         write_unlock(&fs_info->global_root_lock);
 770 }
 771
 772 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 773                                      struct btrfs_key *key)
 774 {
 775         struct rb_node *node;
 776         struct btrfs_root *root = NULL;
 777
 778         read_lock(&fs_info->global_root_lock);
 779         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
 780         if (node)
 781                 root = container_of(node, struct btrfs_root, rb_node);
 782         read_unlock(&fs_info->global_root_lock);
 783
 784         return root;
 785 }
 786
 787 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
 788 {
 789         struct btrfs_block_group *block_group;
 790         u64 ret;
 791
 792         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
 793                 return 0;
 794
 795         if (bytenr)
 796                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
 797         else
 798                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
 799         ASSERT(block_group);
 800         if (!block_group)
 801                 return 0;
 802         ret = block_group->global_root_id;
 803         btrfs_put_block_group(block_group);
 804
 805         return ret;
 806 }
 807
 808 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 809 {
 810         struct btrfs_key key = {
 811                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
 812                 .type = BTRFS_ROOT_ITEM_KEY,
 813                 .offset = btrfs_global_root_id(fs_info, bytenr),
 814         };
 815
 816         return btrfs_global_root(fs_info, &key);
 817 }
 818
 819 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 820 {
 821         struct btrfs_key key = {
 822                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
 823                 .type = BTRFS_ROOT_ITEM_KEY,
 824                 .offset = btrfs_global_root_id(fs_info, bytenr),
 825         };
 826
 827         return btrfs_global_root(fs_info, &key);
 828 }
 829
 830 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
 831 {
 832         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
 833                 return fs_info->block_group_root;
 834         return btrfs_extent_root(fs_info, 0);
 835 }
 836
 837 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 838                                      u64 objectid)
 839 {
 840         struct btrfs_fs_info *fs_info = trans->fs_info;
 841         struct extent_buffer *leaf;
 842         struct btrfs_root *tree_root = fs_info->tree_root;
 843         struct btrfs_root *root;
 844         struct btrfs_key key;
 845         unsigned int nofs_flag;
 846         int ret = 0;
 847
 848         /*
 849          * We're holding a transaction handle, so use a NOFS memory allocation
 850          * context to avoid deadlock if reclaim happens.
 851          */
 852         nofs_flag = memalloc_nofs_save();
 853         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
 854         memalloc_nofs_restore(nofs_flag);
 855         if (!root)
 856                 return ERR_PTR(-ENOMEM);
 857
 858         root->root_key.objectid = objectid;
 859         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 860         root->root_key.offset = 0;
 861
 862         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 863                                       0, BTRFS_NESTING_NORMAL);
 864         if (IS_ERR(leaf)) {
 865                 ret = PTR_ERR(leaf);
 866                 leaf = NULL;
 867                 goto fail;
 868         }
 869
 870         root->node = leaf;
 871         btrfs_mark_buffer_dirty(trans, leaf);
 872
 873         root->commit_root = btrfs_root_node(root);
 874         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 875
 876         btrfs_set_root_flags(&root->root_item, 0);
 877         btrfs_set_root_limit(&root->root_item, 0);
 878         btrfs_set_root_bytenr(&root->root_item, leaf->start);
 879         btrfs_set_root_generation(&root->root_item, trans->transid);
 880         btrfs_set_root_level(&root->root_item, 0);
 881         btrfs_set_root_refs(&root->root_item, 1);
 882         btrfs_set_root_used(&root->root_item, leaf->len);
 883         btrfs_set_root_last_snapshot(&root->root_item, 0);
 884         btrfs_set_root_dirid(&root->root_item, 0);
 885         if (is_fstree(objectid))
 886                 generate_random_guid(root->root_item.uuid);
 887         else
 888                 export_guid(root->root_item.uuid, &guid_null);
 889         btrfs_set_root_drop_level(&root->root_item, 0);
 890
 891         btrfs_tree_unlock(leaf);
 892
 893         key.objectid = objectid;
 894         key.type = BTRFS_ROOT_ITEM_KEY;
 895         key.offset = 0;
 896         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
 897         if (ret)
 898                 goto fail;
 899
 900         return root;
 901
 902 fail:
 903         btrfs_put_root(root);
 904
 905         return ERR_PTR(ret);
 906 }
 907
 908 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 909                                          struct btrfs_fs_info *fs_info)
 910 {
 911         struct btrfs_root *root;
 912
 913         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
 914         if (!root)
 915                 return ERR_PTR(-ENOMEM);
 916
 917         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 918         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 919         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 920
 921         return root;
 922 }
 923
 924 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 925                               struct btrfs_root *root)
 926 {
 927         struct extent_buffer *leaf;
 928
 929         /*
 930          * DON'T set SHAREABLE bit for log trees.
 931          *
 932          * Log trees are not exposed to user space thus can't be snapshotted,
 933          * and they go away before a real commit is actually done.
 934          *
 935          * They do store pointers to file data extents, and those reference
 936          * counts still get updated (along with back refs to the log tree).
 937          */
 938
 939         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
 940                         NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
 941         if (IS_ERR(leaf))
 942                 return PTR_ERR(leaf);
 943
 944         root->node = leaf;
 945
 946         btrfs_mark_buffer_dirty(trans, root->node);
 947         btrfs_tree_unlock(root->node);
 948
 949         return 0;
 950 }
 951
 952 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 953                              struct btrfs_fs_info *fs_info)
 954 {
 955         struct btrfs_root *log_root;
 956
 957         log_root = alloc_log_tree(trans, fs_info);
 958         if (IS_ERR(log_root))
 959                 return PTR_ERR(log_root);
 960
 961         if (!btrfs_is_zoned(fs_info)) {
 962                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
 963
 964                 if (ret) {
 965                         btrfs_put_root(log_root);
 966                         return ret;
 967                 }
 968         }
 969
 970         WARN_ON(fs_info->log_root_tree);
 971         fs_info->log_root_tree = log_root;
 972         return 0;
 973 }
 974
 975 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 976                        struct btrfs_root *root)
 977 {
 978         struct btrfs_fs_info *fs_info = root->fs_info;
 979         struct btrfs_root *log_root;
 980         struct btrfs_inode_item *inode_item;
 981         int ret;
 982
 983         log_root = alloc_log_tree(trans, fs_info);
 984         if (IS_ERR(log_root))
 985                 return PTR_ERR(log_root);
 986
 987         ret = btrfs_alloc_log_tree_node(trans, log_root);
 988         if (ret) {
 989                 btrfs_put_root(log_root);
 990                 return ret;
 991         }
 992
 993         log_root->last_trans = trans->transid;
 994         log_root->root_key.offset = root->root_key.objectid;
 995
 996         inode_item = &log_root->root_item.inode;
 997         btrfs_set_stack_inode_generation(inode_item, 1);
 998         btrfs_set_stack_inode_size(inode_item, 3);
 999         btrfs_set_stack_inode_nlink(inode_item, 1);
1000         btrfs_set_stack_inode_nbytes(inode_item,
1001                                      fs_info->nodesize);
1002         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1003
1004         btrfs_set_root_node(&log_root->root_item, log_root->node);
1005
1006         WARN_ON(root->log_root);
1007         root->log_root = log_root;
1008         btrfs_set_root_log_transid(root, 0);
1009         root->log_transid_committed = -1;
1010         btrfs_set_root_last_log_commit(root, 0);
1011         return 0;
1012 }
1013
1014 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1015                                               struct btrfs_path *path,
1016                                               struct btrfs_key *key)
1017 {
1018         struct btrfs_root *root;
1019         struct btrfs_tree_parent_check check = { 0 };
1020         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1021         u64 generation;
1022         int ret;
1023         int level;
1024
1025         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1026         if (!root)
1027                 return ERR_PTR(-ENOMEM);
1028
1029         ret = btrfs_find_root(tree_root, key, path,
1030                               &root->root_item, &root->root_key);
1031         if (ret) {
1032                 if (ret > 0)
1033                         ret = -ENOENT;
1034                 goto fail;
1035         }
1036
1037         generation = btrfs_root_generation(&root->root_item);
1038         level = btrfs_root_level(&root->root_item);
1039         check.level = level;
1040         check.transid = generation;
1041         check.owner_root = key->objectid;
1042         root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1043                                      &check);
1044         if (IS_ERR(root->node)) {
1045                 ret = PTR_ERR(root->node);
1046                 root->node = NULL;
1047                 goto fail;
1048         }
1049         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1050                 ret = -EIO;
1051                 goto fail;
1052         }
1053
1054         /*
1055          * For real fs, and not log/reloc trees, root owner must
1056          * match its root node owner
1057          */
1058         if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1059             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1060             root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1061             root->root_key.objectid != btrfs_header_owner(root->node)) {
1062                 btrfs_crit(fs_info,
1063 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1064                            root->root_key.objectid, root->node->start,
1065                            btrfs_header_owner(root->node),
1066                            root->root_key.objectid);
1067                 ret = -EUCLEAN;
1068                 goto fail;
1069         }
1070         root->commit_root = btrfs_root_node(root);
1071         return root;
1072 fail:
1073         btrfs_put_root(root);
1074         return ERR_PTR(ret);
1075 }
1076
1077 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1078                                         struct btrfs_key *key)
1079 {
1080         struct btrfs_root *root;
1081         struct btrfs_path *path;
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return ERR_PTR(-ENOMEM);
1086         root = read_tree_root_path(tree_root, path, key);
1087         btrfs_free_path(path);
1088
1089         return root;
1090 }
1091
1092 /*
1093  * Initialize subvolume root in-memory structure
1094  *
1095  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1096  */
1097 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1098 {
1099         int ret;
1100
1101         btrfs_drew_lock_init(&root->snapshot_lock);
1102
1103         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1104             !btrfs_is_data_reloc_root(root) &&
1105             is_fstree(root->root_key.objectid)) {
1106                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1107                 btrfs_check_and_init_root_item(&root->root_item);
1108         }
1109
1110         /*
1111          * Don't assign anonymous block device to roots that are not exposed to
1112          * userspace, the id pool is limited to 1M
1113          */
1114         if (is_fstree(root->root_key.objectid) &&
1115             btrfs_root_refs(&root->root_item) > 0) {
1116                 if (!anon_dev) {
1117                         ret = get_anon_bdev(&root->anon_dev);
1118                         if (ret)
1119                                 goto fail;
1120                 } else {
1121                         root->anon_dev = anon_dev;
1122                 }
1123         }
1124
1125         mutex_lock(&root->objectid_mutex);
1126         ret = btrfs_init_root_free_objectid(root);
1127         if (ret) {
1128                 mutex_unlock(&root->objectid_mutex);
1129                 goto fail;
1130         }
1131
1132         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1133
1134         mutex_unlock(&root->objectid_mutex);
1135
1136         return 0;
1137 fail:
1138         /* The caller is responsible to call btrfs_free_fs_root */
1139         return ret;
1140 }
1141
1142 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1143                                                u64 root_id)
1144 {
1145         struct btrfs_root *root;
1146
1147         spin_lock(&fs_info->fs_roots_radix_lock);
1148         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1149                                  (unsigned long)root_id);
1150         root = btrfs_grab_root(root);
1151         spin_unlock(&fs_info->fs_roots_radix_lock);
1152         return root;
1153 }
1154
1155 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1156                                                 u64 objectid)
1157 {
1158         struct btrfs_key key = {
1159                 .objectid = objectid,
1160                 .type = BTRFS_ROOT_ITEM_KEY,
1161                 .offset = 0,
1162         };
1163
1164         switch (objectid) {
1165         case BTRFS_ROOT_TREE_OBJECTID:
1166                 return btrfs_grab_root(fs_info->tree_root);
1167         case BTRFS_EXTENT_TREE_OBJECTID:
1168                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1169         case BTRFS_CHUNK_TREE_OBJECTID:
1170                 return btrfs_grab_root(fs_info->chunk_root);
1171         case BTRFS_DEV_TREE_OBJECTID:
1172                 return btrfs_grab_root(fs_info->dev_root);
1173         case BTRFS_CSUM_TREE_OBJECTID:
1174                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1175         case BTRFS_QUOTA_TREE_OBJECTID:
1176                 return btrfs_grab_root(fs_info->quota_root);
1177         case BTRFS_UUID_TREE_OBJECTID:
1178                 return btrfs_grab_root(fs_info->uuid_root);
1179         case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1180                 return btrfs_grab_root(fs_info->block_group_root);
1181         case BTRFS_FREE_SPACE_TREE_OBJECTID:
1182                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1183         case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1184                 return btrfs_grab_root(fs_info->stripe_root);
1185         default:
1186                 return NULL;
1187         }
1188 }
1189
1190 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1191                          struct btrfs_root *root)
1192 {
1193         int ret;
1194
1195         ret = radix_tree_preload(GFP_NOFS);
1196         if (ret)
1197                 return ret;
1198
1199         spin_lock(&fs_info->fs_roots_radix_lock);
1200         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1201                                 (unsigned long)root->root_key.objectid,
1202                                 root);
1203         if (ret == 0) {
1204                 btrfs_grab_root(root);
1205                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1206         }
1207         spin_unlock(&fs_info->fs_roots_radix_lock);
1208         radix_tree_preload_end();
1209
1210         return ret;
1211 }
1212
1213 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1214 {
1215 #ifdef CONFIG_BTRFS_DEBUG
1216         struct btrfs_root *root;
1217
1218         while (!list_empty(&fs_info->allocated_roots)) {
1219                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1220
1221                 root = list_first_entry(&fs_info->allocated_roots,
1222                                         struct btrfs_root, leak_list);
1223                 btrfs_err(fs_info, "leaked root %s refcount %d",
1224                           btrfs_root_name(&root->root_key, buf),
1225                           refcount_read(&root->refs));
1226                 while (refcount_read(&root->refs) > 1)
1227                         btrfs_put_root(root);
1228                 btrfs_put_root(root);
1229         }
1230 #endif
1231 }
1232
1233 static void free_global_roots(struct btrfs_fs_info *fs_info)
1234 {
1235         struct btrfs_root *root;
1236         struct rb_node *node;
1237
1238         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1239                 root = rb_entry(node, struct btrfs_root, rb_node);
1240                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1241                 btrfs_put_root(root);
1242         }
1243 }
1244
1245 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1246 {
1247         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1248         percpu_counter_destroy(&fs_info->delalloc_bytes);
1249         percpu_counter_destroy(&fs_info->ordered_bytes);
1250         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1251         btrfs_free_csum_hash(fs_info);
1252         btrfs_free_stripe_hash_table(fs_info);
1253         btrfs_free_ref_cache(fs_info);
1254         kfree(fs_info->balance_ctl);
1255         kfree(fs_info->delayed_root);
1256         free_global_roots(fs_info);
1257         btrfs_put_root(fs_info->tree_root);
1258         btrfs_put_root(fs_info->chunk_root);
1259         btrfs_put_root(fs_info->dev_root);
1260         btrfs_put_root(fs_info->quota_root);
1261         btrfs_put_root(fs_info->uuid_root);
1262         btrfs_put_root(fs_info->fs_root);
1263         btrfs_put_root(fs_info->data_reloc_root);
1264         btrfs_put_root(fs_info->block_group_root);
1265         btrfs_put_root(fs_info->stripe_root);
1266         btrfs_check_leaked_roots(fs_info);
1267         btrfs_extent_buffer_leak_debug_check(fs_info);
1268         kfree(fs_info->super_copy);
1269         kfree(fs_info->super_for_commit);
1270         kfree(fs_info->subpage_info);
1271         kvfree(fs_info);
1272 }
1273
1274
1275 /*
1276  * Get an in-memory reference of a root structure.
1277  *
1278  * For essential trees like root/extent tree, we grab it from fs_info directly.
1279  * For subvolume trees, we check the cached filesystem roots first. If not
1280  * found, then read it from disk and add it to cached fs roots.
1281  *
1282  * Caller should release the root by calling btrfs_put_root() after the usage.
1283  *
1284  * NOTE: Reloc and log trees can't be read by this function as they share the
1285  *       same root objectid.
1286  *
1287  * @objectid:   root id
1288  * @anon_dev:   preallocated anonymous block device number for new roots,
1289  *              pass 0 for new allocation.
1290  * @check_ref:  whether to check root item references, If true, return -ENOENT
1291  *              for orphan roots
1292  */
1293 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1294                                              u64 objectid, dev_t anon_dev,
1295                                              bool check_ref)
1296 {
1297         struct btrfs_root *root;
1298         struct btrfs_path *path;
1299         struct btrfs_key key;
1300         int ret;
1301
1302         root = btrfs_get_global_root(fs_info, objectid);
1303         if (root)
1304                 return root;
1305
1306         /*
1307          * If we're called for non-subvolume trees, and above function didn't
1308          * find one, do not try to read it from disk.
1309          *
1310          * This is namely for free-space-tree and quota tree, which can change
1311          * at runtime and should only be grabbed from fs_info.
1312          */
1313         if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1314                 return ERR_PTR(-ENOENT);
1315 again:
1316         root = btrfs_lookup_fs_root(fs_info, objectid);
1317         if (root) {
1318                 /* Shouldn't get preallocated anon_dev for cached roots */
1319                 ASSERT(!anon_dev);
1320                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1321                         btrfs_put_root(root);
1322                         return ERR_PTR(-ENOENT);
1323                 }
1324                 return root;
1325         }
1326
1327         key.objectid = objectid;
1328         key.type = BTRFS_ROOT_ITEM_KEY;
1329         key.offset = (u64)-1;
1330         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1331         if (IS_ERR(root))
1332                 return root;
1333
1334         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1335                 ret = -ENOENT;
1336                 goto fail;
1337         }
1338
1339         ret = btrfs_init_fs_root(root, anon_dev);
1340         if (ret)
1341                 goto fail;
1342
1343         path = btrfs_alloc_path();
1344         if (!path) {
1345                 ret = -ENOMEM;
1346                 goto fail;
1347         }
1348         key.objectid = BTRFS_ORPHAN_OBJECTID;
1349         key.type = BTRFS_ORPHAN_ITEM_KEY;
1350         key.offset = objectid;
1351
1352         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1353         btrfs_free_path(path);
1354         if (ret < 0)
1355                 goto fail;
1356         if (ret == 0)
1357                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1358
1359         ret = btrfs_insert_fs_root(fs_info, root);
1360         if (ret) {
1361                 if (ret == -EEXIST) {
1362                         btrfs_put_root(root);
1363                         goto again;
1364                 }
1365                 goto fail;
1366         }
1367         return root;
1368 fail:
1369         /*
1370          * If our caller provided us an anonymous device, then it's his
1371          * responsibility to free it in case we fail. So we have to set our
1372          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1373          * and once again by our caller.
1374          */
1375         if (anon_dev)
1376                 root->anon_dev = 0;
1377         btrfs_put_root(root);
1378         return ERR_PTR(ret);
1379 }
1380
1381 /*
1382  * Get in-memory reference of a root structure
1383  *
1384  * @objectid:   tree objectid
1385  * @check_ref:  if set, verify that the tree exists and the item has at least
1386  *              one reference
1387  */
1388 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1389                                      u64 objectid, bool check_ref)
1390 {
1391         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1392 }
1393
1394 /*
1395  * Get in-memory reference of a root structure, created as new, optionally pass
1396  * the anonymous block device id
1397  *
1398  * @objectid:   tree objectid
1399  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1400  *              parameter value
1401  */
1402 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1403                                          u64 objectid, dev_t anon_dev)
1404 {
1405         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1406 }
1407
1408 /*
1409  * Return a root for the given objectid.
1410  *
1411  * @fs_info:    the fs_info
1412  * @objectid:   the objectid we need to lookup
1413  *
1414  * This is exclusively used for backref walking, and exists specifically because
1415  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1416  * creation time, which means we may have to read the tree_root in order to look
1417  * up a fs root that is not in memory.  If the root is not in memory we will
1418  * read the tree root commit root and look up the fs root from there.  This is a
1419  * temporary root, it will not be inserted into the radix tree as it doesn't
1420  * have the most uptodate information, it'll simply be discarded once the
1421  * backref code is finished using the root.
1422  */
1423 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1424                                                  struct btrfs_path *path,
1425                                                  u64 objectid)
1426 {
1427         struct btrfs_root *root;
1428         struct btrfs_key key;
1429
1430         ASSERT(path->search_commit_root && path->skip_locking);
1431
1432         /*
1433          * This can return -ENOENT if we ask for a root that doesn't exist, but
1434          * since this is called via the backref walking code we won't be looking
1435          * up a root that doesn't exist, unless there's corruption.  So if root
1436          * != NULL just return it.
1437          */
1438         root = btrfs_get_global_root(fs_info, objectid);
1439         if (root)
1440                 return root;
1441
1442         root = btrfs_lookup_fs_root(fs_info, objectid);
1443         if (root)
1444                 return root;
1445
1446         key.objectid = objectid;
1447         key.type = BTRFS_ROOT_ITEM_KEY;
1448         key.offset = (u64)-1;
1449         root = read_tree_root_path(fs_info->tree_root, path, &key);
1450         btrfs_release_path(path);
1451
1452         return root;
1453 }
1454
1455 static int cleaner_kthread(void *arg)
1456 {
1457         struct btrfs_fs_info *fs_info = arg;
1458         int again;
1459
1460         while (1) {
1461                 again = 0;
1462
1463                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1464
1465                 /* Make the cleaner go to sleep early. */
1466                 if (btrfs_need_cleaner_sleep(fs_info))
1467                         goto sleep;
1468
1469                 /*
1470                  * Do not do anything if we might cause open_ctree() to block
1471                  * before we have finished mounting the filesystem.
1472                  */
1473                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1474                         goto sleep;
1475
1476                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1477                         goto sleep;
1478
1479                 /*
1480                  * Avoid the problem that we change the status of the fs
1481                  * during the above check and trylock.
1482                  */
1483                 if (btrfs_need_cleaner_sleep(fs_info)) {
1484                         mutex_unlock(&fs_info->cleaner_mutex);
1485                         goto sleep;
1486                 }
1487
1488                 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1489                         btrfs_sysfs_feature_update(fs_info);
1490
1491                 btrfs_run_delayed_iputs(fs_info);
1492
1493                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1494                 mutex_unlock(&fs_info->cleaner_mutex);
1495
1496                 /*
1497                  * The defragger has dealt with the R/O remount and umount,
1498                  * needn't do anything special here.
1499                  */
1500                 btrfs_run_defrag_inodes(fs_info);
1501
1502                 /*
1503                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1504                  * with relocation (btrfs_relocate_chunk) and relocation
1505                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1506                  * after acquiring fs_info->reclaim_bgs_lock. So we
1507                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1508                  * unused block groups.
1509                  */
1510                 btrfs_delete_unused_bgs(fs_info);
1511
1512                 /*
1513                  * Reclaim block groups in the reclaim_bgs list after we deleted
1514                  * all unused block_groups. This possibly gives us some more free
1515                  * space.
1516                  */
1517                 btrfs_reclaim_bgs(fs_info);
1518 sleep:
1519                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1520                 if (kthread_should_park())
1521                         kthread_parkme();
1522                 if (kthread_should_stop())
1523                         return 0;
1524                 if (!again) {
1525                         set_current_state(TASK_INTERRUPTIBLE);
1526                         schedule();
1527                         __set_current_state(TASK_RUNNING);
1528                 }
1529         }
1530 }
1531
1532 static int transaction_kthread(void *arg)
1533 {
1534         struct btrfs_root *root = arg;
1535         struct btrfs_fs_info *fs_info = root->fs_info;
1536         struct btrfs_trans_handle *trans;
1537         struct btrfs_transaction *cur;
1538         u64 transid;
1539         time64_t delta;
1540         unsigned long delay;
1541         bool cannot_commit;
1542
1543         do {
1544                 cannot_commit = false;
1545                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1546                 mutex_lock(&fs_info->transaction_kthread_mutex);
1547
1548                 spin_lock(&fs_info->trans_lock);
1549                 cur = fs_info->running_transaction;
1550                 if (!cur) {
1551                         spin_unlock(&fs_info->trans_lock);
1552                         goto sleep;
1553                 }
1554
1555                 delta = ktime_get_seconds() - cur->start_time;
1556                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1557                     cur->state < TRANS_STATE_COMMIT_PREP &&
1558                     delta < fs_info->commit_interval) {
1559                         spin_unlock(&fs_info->trans_lock);
1560                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1561                         delay = min(delay,
1562                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1563                         goto sleep;
1564                 }
1565                 transid = cur->transid;
1566                 spin_unlock(&fs_info->trans_lock);
1567
1568                 /* If the file system is aborted, this will always fail. */
1569                 trans = btrfs_attach_transaction(root);
1570                 if (IS_ERR(trans)) {
1571                         if (PTR_ERR(trans) != -ENOENT)
1572                                 cannot_commit = true;
1573                         goto sleep;
1574                 }
1575                 if (transid == trans->transid) {
1576                         btrfs_commit_transaction(trans);
1577                 } else {
1578                         btrfs_end_transaction(trans);
1579                 }
1580 sleep:
1581                 wake_up_process(fs_info->cleaner_kthread);
1582                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1583
1584                 if (BTRFS_FS_ERROR(fs_info))
1585                         btrfs_cleanup_transaction(fs_info);
1586                 if (!kthread_should_stop() &&
1587                                 (!btrfs_transaction_blocked(fs_info) ||
1588                                  cannot_commit))
1589                         schedule_timeout_interruptible(delay);
1590         } while (!kthread_should_stop());
1591         return 0;
1592 }
1593
1594 /*
1595  * This will find the highest generation in the array of root backups.  The
1596  * index of the highest array is returned, or -EINVAL if we can't find
1597  * anything.
1598  *
1599  * We check to make sure the array is valid by comparing the
1600  * generation of the latest  root in the array with the generation
1601  * in the super block.  If they don't match we pitch it.
1602  */
1603 static int find_newest_super_backup(struct btrfs_fs_info *info)
1604 {
1605         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1606         u64 cur;
1607         struct btrfs_root_backup *root_backup;
1608         int i;
1609
1610         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1611                 root_backup = info->super_copy->super_roots + i;
1612                 cur = btrfs_backup_tree_root_gen(root_backup);
1613                 if (cur == newest_gen)
1614                         return i;
1615         }
1616
1617         return -EINVAL;
1618 }
1619
1620 /*
1621  * copy all the root pointers into the super backup array.
1622  * this will bump the backup pointer by one when it is
1623  * done
1624  */
1625 static void backup_super_roots(struct btrfs_fs_info *info)
1626 {
1627         const int next_backup = info->backup_root_index;
1628         struct btrfs_root_backup *root_backup;
1629
1630         root_backup = info->super_for_commit->super_roots + next_backup;
1631
1632         /*
1633          * make sure all of our padding and empty slots get zero filled
1634          * regardless of which ones we use today
1635          */
1636         memset(root_backup, 0, sizeof(*root_backup));
1637
1638         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1639
1640         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1641         btrfs_set_backup_tree_root_gen(root_backup,
1642                                btrfs_header_generation(info->tree_root->node));
1643
1644         btrfs_set_backup_tree_root_level(root_backup,
1645                                btrfs_header_level(info->tree_root->node));
1646
1647         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1648         btrfs_set_backup_chunk_root_gen(root_backup,
1649                                btrfs_header_generation(info->chunk_root->node));
1650         btrfs_set_backup_chunk_root_level(root_backup,
1651                                btrfs_header_level(info->chunk_root->node));
1652
1653         if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1654                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1655                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1656
1657                 btrfs_set_backup_extent_root(root_backup,
1658                                              extent_root->node->start);
1659                 btrfs_set_backup_extent_root_gen(root_backup,
1660                                 btrfs_header_generation(extent_root->node));
1661                 btrfs_set_backup_extent_root_level(root_backup,
1662                                         btrfs_header_level(extent_root->node));
1663
1664                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1665                 btrfs_set_backup_csum_root_gen(root_backup,
1666                                                btrfs_header_generation(csum_root->node));
1667                 btrfs_set_backup_csum_root_level(root_backup,
1668                                                  btrfs_header_level(csum_root->node));
1669         }
1670
1671         /*
1672          * we might commit during log recovery, which happens before we set
1673          * the fs_root.  Make sure it is valid before we fill it in.
1674          */
1675         if (info->fs_root && info->fs_root->node) {
1676                 btrfs_set_backup_fs_root(root_backup,
1677                                          info->fs_root->node->start);
1678                 btrfs_set_backup_fs_root_gen(root_backup,
1679                                btrfs_header_generation(info->fs_root->node));
1680                 btrfs_set_backup_fs_root_level(root_backup,
1681                                btrfs_header_level(info->fs_root->node));
1682         }
1683
1684         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1685         btrfs_set_backup_dev_root_gen(root_backup,
1686                                btrfs_header_generation(info->dev_root->node));
1687         btrfs_set_backup_dev_root_level(root_backup,
1688                                        btrfs_header_level(info->dev_root->node));
1689
1690         btrfs_set_backup_total_bytes(root_backup,
1691                              btrfs_super_total_bytes(info->super_copy));
1692         btrfs_set_backup_bytes_used(root_backup,
1693                              btrfs_super_bytes_used(info->super_copy));
1694         btrfs_set_backup_num_devices(root_backup,
1695                              btrfs_super_num_devices(info->super_copy));
1696
1697         /*
1698          * if we don't copy this out to the super_copy, it won't get remembered
1699          * for the next commit
1700          */
1701         memcpy(&info->super_copy->super_roots,
1702                &info->super_for_commit->super_roots,
1703                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1704 }
1705
1706 /*
1707  * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1708  * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1709  *
1710  * @fs_info:  filesystem whose backup roots need to be read
1711  * @priority: priority of backup root required
1712  *
1713  * Returns backup root index on success and -EINVAL otherwise.
1714  */
1715 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1716 {
1717         int backup_index = find_newest_super_backup(fs_info);
1718         struct btrfs_super_block *super = fs_info->super_copy;
1719         struct btrfs_root_backup *root_backup;
1720
1721         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1722                 if (priority == 0)
1723                         return backup_index;
1724
1725                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1726                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1727         } else {
1728                 return -EINVAL;
1729         }
1730
1731         root_backup = super->super_roots + backup_index;
1732
1733         btrfs_set_super_generation(super,
1734                                    btrfs_backup_tree_root_gen(root_backup));
1735         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1736         btrfs_set_super_root_level(super,
1737                                    btrfs_backup_tree_root_level(root_backup));
1738         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1739
1740         /*
1741          * Fixme: the total bytes and num_devices need to match or we should
1742          * need a fsck
1743          */
1744         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1745         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1746
1747         return backup_index;
1748 }
1749
1750 /* helper to cleanup workers */
1751 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1752 {
1753         btrfs_destroy_workqueue(fs_info->fixup_workers);
1754         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1755         btrfs_destroy_workqueue(fs_info->workers);
1756         if (fs_info->endio_workers)
1757                 destroy_workqueue(fs_info->endio_workers);
1758         if (fs_info->rmw_workers)
1759                 destroy_workqueue(fs_info->rmw_workers);
1760         if (fs_info->compressed_write_workers)
1761                 destroy_workqueue(fs_info->compressed_write_workers);
1762         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1763         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1764         btrfs_destroy_workqueue(fs_info->delayed_workers);
1765         btrfs_destroy_workqueue(fs_info->caching_workers);
1766         btrfs_destroy_workqueue(fs_info->flush_workers);
1767         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1768         if (fs_info->discard_ctl.discard_workers)
1769                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1770         /*
1771          * Now that all other work queues are destroyed, we can safely destroy
1772          * the queues used for metadata I/O, since tasks from those other work
1773          * queues can do metadata I/O operations.
1774          */
1775         if (fs_info->endio_meta_workers)
1776                 destroy_workqueue(fs_info->endio_meta_workers);
1777 }
1778
1779 static void free_root_extent_buffers(struct btrfs_root *root)
1780 {
1781         if (root) {
1782                 free_extent_buffer(root->node);
1783                 free_extent_buffer(root->commit_root);
1784                 root->node = NULL;
1785                 root->commit_root = NULL;
1786         }
1787 }
1788
1789 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1790 {
1791         struct btrfs_root *root, *tmp;
1792
1793         rbtree_postorder_for_each_entry_safe(root, tmp,
1794                                              &fs_info->global_root_tree,
1795                                              rb_node)
1796                 free_root_extent_buffers(root);
1797 }
1798
1799 /* helper to cleanup tree roots */
1800 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1801 {
1802         free_root_extent_buffers(info->tree_root);
1803
1804         free_global_root_pointers(info);
1805         free_root_extent_buffers(info->dev_root);
1806         free_root_extent_buffers(info->quota_root);
1807         free_root_extent_buffers(info->uuid_root);
1808         free_root_extent_buffers(info->fs_root);
1809         free_root_extent_buffers(info->data_reloc_root);
1810         free_root_extent_buffers(info->block_group_root);
1811         free_root_extent_buffers(info->stripe_root);
1812         if (free_chunk_root)
1813                 free_root_extent_buffers(info->chunk_root);
1814 }
1815
1816 void btrfs_put_root(struct btrfs_root *root)
1817 {
1818         if (!root)
1819                 return;
1820
1821         if (refcount_dec_and_test(&root->refs)) {
1822                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1823                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1824                 if (root->anon_dev)
1825                         free_anon_bdev(root->anon_dev);
1826                 free_root_extent_buffers(root);
1827 #ifdef CONFIG_BTRFS_DEBUG
1828                 spin_lock(&root->fs_info->fs_roots_radix_lock);
1829                 list_del_init(&root->leak_list);
1830                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1831 #endif
1832                 kfree(root);
1833         }
1834 }
1835
1836 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1837 {
1838         int ret;
1839         struct btrfs_root *gang[8];
1840         int i;
1841
1842         while (!list_empty(&fs_info->dead_roots)) {
1843                 gang[0] = list_entry(fs_info->dead_roots.next,
1844                                      struct btrfs_root, root_list);
1845                 list_del(&gang[0]->root_list);
1846
1847                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1848                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1849                 btrfs_put_root(gang[0]);
1850         }
1851
1852         while (1) {
1853                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1854                                              (void **)gang, 0,
1855                                              ARRAY_SIZE(gang));
1856                 if (!ret)
1857                         break;
1858                 for (i = 0; i < ret; i++)
1859                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1860         }
1861 }
1862
1863 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1864 {
1865         mutex_init(&fs_info->scrub_lock);
1866         atomic_set(&fs_info->scrubs_running, 0);
1867         atomic_set(&fs_info->scrub_pause_req, 0);
1868         atomic_set(&fs_info->scrubs_paused, 0);
1869         atomic_set(&fs_info->scrub_cancel_req, 0);
1870         init_waitqueue_head(&fs_info->scrub_pause_wait);
1871         refcount_set(&fs_info->scrub_workers_refcnt, 0);
1872 }
1873
1874 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1875 {
1876         spin_lock_init(&fs_info->balance_lock);
1877         mutex_init(&fs_info->balance_mutex);
1878         atomic_set(&fs_info->balance_pause_req, 0);
1879         atomic_set(&fs_info->balance_cancel_req, 0);
1880         fs_info->balance_ctl = NULL;
1881         init_waitqueue_head(&fs_info->balance_wait_q);
1882         atomic_set(&fs_info->reloc_cancel_req, 0);
1883 }
1884
1885 static int btrfs_init_btree_inode(struct super_block *sb)
1886 {
1887         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1888         unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1889                                               fs_info->tree_root);
1890         struct inode *inode;
1891
1892         inode = new_inode(sb);
1893         if (!inode)
1894                 return -ENOMEM;
1895
1896         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1897         set_nlink(inode, 1);
1898         /*
1899          * we set the i_size on the btree inode to the max possible int.
1900          * the real end of the address space is determined by all of
1901          * the devices in the system
1902          */
1903         inode->i_size = OFFSET_MAX;
1904         inode->i_mapping->a_ops = &btree_aops;
1905         mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1906
1907         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1908         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1909                             IO_TREE_BTREE_INODE_IO);
1910         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1911
1912         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1913         BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1914         BTRFS_I(inode)->location.type = 0;
1915         BTRFS_I(inode)->location.offset = 0;
1916         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1917         __insert_inode_hash(inode, hash);
1918         fs_info->btree_inode = inode;
1919
1920         return 0;
1921 }
1922
1923 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1924 {
1925         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1926         init_rwsem(&fs_info->dev_replace.rwsem);
1927         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1928 }
1929
1930 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1931 {
1932         spin_lock_init(&fs_info->qgroup_lock);
1933         mutex_init(&fs_info->qgroup_ioctl_lock);
1934         fs_info->qgroup_tree = RB_ROOT;
1935         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1936         fs_info->qgroup_seq = 1;
1937         fs_info->qgroup_ulist = NULL;
1938         fs_info->qgroup_rescan_running = false;
1939         fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1940         mutex_init(&fs_info->qgroup_rescan_lock);
1941 }
1942
1943 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1944 {
1945         u32 max_active = fs_info->thread_pool_size;
1946         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1947         unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1948
1949         fs_info->workers =
1950                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1951
1952         fs_info->delalloc_workers =
1953                 btrfs_alloc_workqueue(fs_info, "delalloc",
1954                                       flags, max_active, 2);
1955
1956         fs_info->flush_workers =
1957                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1958                                       flags, max_active, 0);
1959
1960         fs_info->caching_workers =
1961                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1962
1963         fs_info->fixup_workers =
1964                 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1965
1966         fs_info->endio_workers =
1967                 alloc_workqueue("btrfs-endio", flags, max_active);
1968         fs_info->endio_meta_workers =
1969                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1970         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1971         fs_info->endio_write_workers =
1972                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1973                                       max_active, 2);
1974         fs_info->compressed_write_workers =
1975                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1976         fs_info->endio_freespace_worker =
1977                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1978                                       max_active, 0);
1979         fs_info->delayed_workers =
1980                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1981                                       max_active, 0);
1982         fs_info->qgroup_rescan_workers =
1983                 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1984                                               ordered_flags);
1985         fs_info->discard_ctl.discard_workers =
1986                 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
1987
1988         if (!(fs_info->workers &&
1989               fs_info->delalloc_workers && fs_info->flush_workers &&
1990               fs_info->endio_workers && fs_info->endio_meta_workers &&
1991               fs_info->compressed_write_workers &&
1992               fs_info->endio_write_workers &&
1993               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1994               fs_info->caching_workers && fs_info->fixup_workers &&
1995               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1996               fs_info->discard_ctl.discard_workers)) {
1997                 return -ENOMEM;
1998         }
1999
2000         return 0;
2001 }
2002
2003 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2004 {
2005         struct crypto_shash *csum_shash;
2006         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2007
2008         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2009
2010         if (IS_ERR(csum_shash)) {
2011                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2012                           csum_driver);
2013                 return PTR_ERR(csum_shash);
2014         }
2015
2016         fs_info->csum_shash = csum_shash;
2017
2018         /*
2019          * Check if the checksum implementation is a fast accelerated one.
2020          * As-is this is a bit of a hack and should be replaced once the csum
2021          * implementations provide that information themselves.
2022          */
2023         switch (csum_type) {
2024         case BTRFS_CSUM_TYPE_CRC32:
2025                 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2026                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2027                 break;
2028         case BTRFS_CSUM_TYPE_XXHASH:
2029                 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2030                 break;
2031         default:
2032                 break;
2033         }
2034
2035         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2036                         btrfs_super_csum_name(csum_type),
2037                         crypto_shash_driver_name(csum_shash));
2038         return 0;
2039 }
2040
2041 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2042                             struct btrfs_fs_devices *fs_devices)
2043 {
2044         int ret;
2045         struct btrfs_tree_parent_check check = { 0 };
2046         struct btrfs_root *log_tree_root;
2047         struct btrfs_super_block *disk_super = fs_info->super_copy;
2048         u64 bytenr = btrfs_super_log_root(disk_super);
2049         int level = btrfs_super_log_root_level(disk_super);
2050
2051         if (fs_devices->rw_devices == 0) {
2052                 btrfs_warn(fs_info, "log replay required on RO media");
2053                 return -EIO;
2054         }
2055
2056         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2057                                          GFP_KERNEL);
2058         if (!log_tree_root)
2059                 return -ENOMEM;
2060
2061         check.level = level;
2062         check.transid = fs_info->generation + 1;
2063         check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2064         log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2065         if (IS_ERR(log_tree_root->node)) {
2066                 btrfs_warn(fs_info, "failed to read log tree");
2067                 ret = PTR_ERR(log_tree_root->node);
2068                 log_tree_root->node = NULL;
2069                 btrfs_put_root(log_tree_root);
2070                 return ret;
2071         }
2072         if (!extent_buffer_uptodate(log_tree_root->node)) {
2073                 btrfs_err(fs_info, "failed to read log tree");
2074                 btrfs_put_root(log_tree_root);
2075                 return -EIO;
2076         }
2077
2078         /* returns with log_tree_root freed on success */
2079         ret = btrfs_recover_log_trees(log_tree_root);
2080         if (ret) {
2081                 btrfs_handle_fs_error(fs_info, ret,
2082                                       "Failed to recover log tree");
2083                 btrfs_put_root(log_tree_root);
2084                 return ret;
2085         }
2086
2087         if (sb_rdonly(fs_info->sb)) {
2088                 ret = btrfs_commit_super(fs_info);
2089                 if (ret)
2090                         return ret;
2091         }
2092
2093         return 0;
2094 }
2095
2096 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2097                                       struct btrfs_path *path, u64 objectid,
2098                                       const char *name)
2099 {
2100         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2101         struct btrfs_root *root;
2102         u64 max_global_id = 0;
2103         int ret;
2104         struct btrfs_key key = {
2105                 .objectid = objectid,
2106                 .type = BTRFS_ROOT_ITEM_KEY,
2107                 .offset = 0,
2108         };
2109         bool found = false;
2110
2111         /* If we have IGNOREDATACSUMS skip loading these roots. */
2112         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2113             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2114                 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2115                 return 0;
2116         }
2117
2118         while (1) {
2119                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2120                 if (ret < 0)
2121                         break;
2122
2123                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2124                         ret = btrfs_next_leaf(tree_root, path);
2125                         if (ret) {
2126                                 if (ret > 0)
2127                                         ret = 0;
2128                                 break;
2129                         }
2130                 }
2131                 ret = 0;
2132
2133                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2134                 if (key.objectid != objectid)
2135                         break;
2136                 btrfs_release_path(path);
2137
2138                 /*
2139                  * Just worry about this for extent tree, it'll be the same for
2140                  * everybody.
2141                  */
2142                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2143                         max_global_id = max(max_global_id, key.offset);
2144
2145                 found = true;
2146                 root = read_tree_root_path(tree_root, path, &key);
2147                 if (IS_ERR(root)) {
2148                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2149                                 ret = PTR_ERR(root);
2150                         break;
2151                 }
2152                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2153                 ret = btrfs_global_root_insert(root);
2154                 if (ret) {
2155                         btrfs_put_root(root);
2156                         break;
2157                 }
2158                 key.offset++;
2159         }
2160         btrfs_release_path(path);
2161
2162         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2163                 fs_info->nr_global_roots = max_global_id + 1;
2164
2165         if (!found || ret) {
2166                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2167                         set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2168
2169                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2170                         ret = ret ? ret : -ENOENT;
2171                 else
2172                         ret = 0;
2173                 btrfs_err(fs_info, "failed to load root %s", name);
2174         }
2175         return ret;
2176 }
2177
2178 static int load_global_roots(struct btrfs_root *tree_root)
2179 {
2180         struct btrfs_path *path;
2181         int ret = 0;
2182
2183         path = btrfs_alloc_path();
2184         if (!path)
2185                 return -ENOMEM;
2186
2187         ret = load_global_roots_objectid(tree_root, path,
2188                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2189         if (ret)
2190                 goto out;
2191         ret = load_global_roots_objectid(tree_root, path,
2192                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2193         if (ret)
2194                 goto out;
2195         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2196                 goto out;
2197         ret = load_global_roots_objectid(tree_root, path,
2198                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2199                                          "free space");
2200 out:
2201         btrfs_free_path(path);
2202         return ret;
2203 }
2204
2205 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2206 {
2207         struct btrfs_root *tree_root = fs_info->tree_root;
2208         struct btrfs_root *root;
2209         struct btrfs_key location;
2210         int ret;
2211
2212         BUG_ON(!fs_info->tree_root);
2213
2214         ret = load_global_roots(tree_root);
2215         if (ret)
2216                 return ret;
2217
2218         location.type = BTRFS_ROOT_ITEM_KEY;
2219         location.offset = 0;
2220
2221         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2222                 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2223                 root = btrfs_read_tree_root(tree_root, &location);
2224                 if (IS_ERR(root)) {
2225                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2226                                 ret = PTR_ERR(root);
2227                                 goto out;
2228                         }
2229                 } else {
2230                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2231                         fs_info->block_group_root = root;
2232                 }
2233         }
2234
2235         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2236         root = btrfs_read_tree_root(tree_root, &location);
2237         if (IS_ERR(root)) {
2238                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2239                         ret = PTR_ERR(root);
2240                         goto out;
2241                 }
2242         } else {
2243                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2244                 fs_info->dev_root = root;
2245         }
2246         /* Initialize fs_info for all devices in any case */
2247         ret = btrfs_init_devices_late(fs_info);
2248         if (ret)
2249                 goto out;
2250
2251         /*
2252          * This tree can share blocks with some other fs tree during relocation
2253          * and we need a proper setup by btrfs_get_fs_root
2254          */
2255         root = btrfs_get_fs_root(tree_root->fs_info,
2256                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2257         if (IS_ERR(root)) {
2258                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2259                         ret = PTR_ERR(root);
2260                         goto out;
2261                 }
2262         } else {
2263                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2264                 fs_info->data_reloc_root = root;
2265         }
2266
2267         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2268         root = btrfs_read_tree_root(tree_root, &location);
2269         if (!IS_ERR(root)) {
2270                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2271                 fs_info->quota_root = root;
2272         }
2273
2274         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2275         root = btrfs_read_tree_root(tree_root, &location);
2276         if (IS_ERR(root)) {
2277                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2278                         ret = PTR_ERR(root);
2279                         if (ret != -ENOENT)
2280                                 goto out;
2281                 }
2282         } else {
2283                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2284                 fs_info->uuid_root = root;
2285         }
2286
2287         if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2288                 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2289                 root = btrfs_read_tree_root(tree_root, &location);
2290                 if (IS_ERR(root)) {
2291                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2292                                 ret = PTR_ERR(root);
2293                                 goto out;
2294                         }
2295                 } else {
2296                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2297                         fs_info->stripe_root = root;
2298                 }
2299         }
2300
2301         return 0;
2302 out:
2303         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2304                    location.objectid, ret);
2305         return ret;
2306 }
2307
2308 /*
2309  * Real super block validation
2310  * NOTE: super csum type and incompat features will not be checked here.
2311  *
2312  * @sb:         super block to check
2313  * @mirror_num: the super block number to check its bytenr:
2314  *              0       the primary (1st) sb
2315  *              1, 2    2nd and 3rd backup copy
2316  *             -1       skip bytenr check
2317  */
2318 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2319                          struct btrfs_super_block *sb, int mirror_num)
2320 {
2321         u64 nodesize = btrfs_super_nodesize(sb);
2322         u64 sectorsize = btrfs_super_sectorsize(sb);
2323         int ret = 0;
2324
2325         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2326                 btrfs_err(fs_info, "no valid FS found");
2327                 ret = -EINVAL;
2328         }
2329         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2330                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2331                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2332                 ret = -EINVAL;
2333         }
2334         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2335                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2336                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2337                 ret = -EINVAL;
2338         }
2339         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2340                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2341                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2342                 ret = -EINVAL;
2343         }
2344         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2345                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2346                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2347                 ret = -EINVAL;
2348         }
2349
2350         /*
2351          * Check sectorsize and nodesize first, other check will need it.
2352          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2353          */
2354         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2355             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2356                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2357                 ret = -EINVAL;
2358         }
2359
2360         /*
2361          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2362          *
2363          * We can support 16K sectorsize with 64K page size without problem,
2364          * but such sectorsize/pagesize combination doesn't make much sense.
2365          * 4K will be our future standard, PAGE_SIZE is supported from the very
2366          * beginning.
2367          */
2368         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2369                 btrfs_err(fs_info,
2370                         "sectorsize %llu not yet supported for page size %lu",
2371                         sectorsize, PAGE_SIZE);
2372                 ret = -EINVAL;
2373         }
2374
2375         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2376             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2377                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2378                 ret = -EINVAL;
2379         }
2380         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2381                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2382                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2383                 ret = -EINVAL;
2384         }
2385
2386         /* Root alignment check */
2387         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2388                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2389                            btrfs_super_root(sb));
2390                 ret = -EINVAL;
2391         }
2392         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2393                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2394                            btrfs_super_chunk_root(sb));
2395                 ret = -EINVAL;
2396         }
2397         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2398                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2399                            btrfs_super_log_root(sb));
2400                 ret = -EINVAL;
2401         }
2402
2403         if (!fs_info->fs_devices->temp_fsid &&
2404             memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2405                 btrfs_err(fs_info,
2406                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2407                           sb->fsid, fs_info->fs_devices->fsid);
2408                 ret = -EINVAL;
2409         }
2410
2411         if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2412                    BTRFS_FSID_SIZE) != 0) {
2413                 btrfs_err(fs_info,
2414 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2415                           btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2416                 ret = -EINVAL;
2417         }
2418
2419         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2420                    BTRFS_FSID_SIZE) != 0) {
2421                 btrfs_err(fs_info,
2422                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2423                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2424                 ret = -EINVAL;
2425         }
2426
2427         /*
2428          * Artificial requirement for block-group-tree to force newer features
2429          * (free-space-tree, no-holes) so the test matrix is smaller.
2430          */
2431         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2432             (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2433              !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2434                 btrfs_err(fs_info,
2435                 "block-group-tree feature requires fres-space-tree and no-holes");
2436                 ret = -EINVAL;
2437         }
2438
2439         /*
2440          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2441          * done later
2442          */
2443         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2444                 btrfs_err(fs_info, "bytes_used is too small %llu",
2445                           btrfs_super_bytes_used(sb));
2446                 ret = -EINVAL;
2447         }
2448         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2449                 btrfs_err(fs_info, "invalid stripesize %u",
2450                           btrfs_super_stripesize(sb));
2451                 ret = -EINVAL;
2452         }
2453         if (btrfs_super_num_devices(sb) > (1UL << 31))
2454                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2455                            btrfs_super_num_devices(sb));
2456         if (btrfs_super_num_devices(sb) == 0) {
2457                 btrfs_err(fs_info, "number of devices is 0");
2458                 ret = -EINVAL;
2459         }
2460
2461         if (mirror_num >= 0 &&
2462             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2463                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2464                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2465                 ret = -EINVAL;
2466         }
2467
2468         /*
2469          * Obvious sys_chunk_array corruptions, it must hold at least one key
2470          * and one chunk
2471          */
2472         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2473                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2474                           btrfs_super_sys_array_size(sb),
2475                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2476                 ret = -EINVAL;
2477         }
2478         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2479                         + sizeof(struct btrfs_chunk)) {
2480                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2481                           btrfs_super_sys_array_size(sb),
2482                           sizeof(struct btrfs_disk_key)
2483                           + sizeof(struct btrfs_chunk));
2484                 ret = -EINVAL;
2485         }
2486
2487         /*
2488          * The generation is a global counter, we'll trust it more than the others
2489          * but it's still possible that it's the one that's wrong.
2490          */
2491         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2492                 btrfs_warn(fs_info,
2493                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2494                         btrfs_super_generation(sb),
2495                         btrfs_super_chunk_root_generation(sb));
2496         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2497             && btrfs_super_cache_generation(sb) != (u64)-1)
2498                 btrfs_warn(fs_info,
2499                         "suspicious: generation < cache_generation: %llu < %llu",
2500                         btrfs_super_generation(sb),
2501                         btrfs_super_cache_generation(sb));
2502
2503         return ret;
2504 }
2505
2506 /*
2507  * Validation of super block at mount time.
2508  * Some checks already done early at mount time, like csum type and incompat
2509  * flags will be skipped.
2510  */
2511 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2512 {
2513         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2514 }
2515
2516 /*
2517  * Validation of super block at write time.
2518  * Some checks like bytenr check will be skipped as their values will be
2519  * overwritten soon.
2520  * Extra checks like csum type and incompat flags will be done here.
2521  */
2522 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2523                                       struct btrfs_super_block *sb)
2524 {
2525         int ret;
2526
2527         ret = btrfs_validate_super(fs_info, sb, -1);
2528         if (ret < 0)
2529                 goto out;
2530         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2531                 ret = -EUCLEAN;
2532                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2533                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2534                 goto out;
2535         }
2536         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2537                 ret = -EUCLEAN;
2538                 btrfs_err(fs_info,
2539                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2540                           btrfs_super_incompat_flags(sb),
2541                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2542                 goto out;
2543         }
2544 out:
2545         if (ret < 0)
2546                 btrfs_err(fs_info,
2547                 "super block corruption detected before writing it to disk");
2548         return ret;
2549 }
2550
2551 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2552 {
2553         struct btrfs_tree_parent_check check = {
2554                 .level = level,
2555                 .transid = gen,
2556                 .owner_root = root->root_key.objectid
2557         };
2558         int ret = 0;
2559
2560         root->node = read_tree_block(root->fs_info, bytenr, &check);
2561         if (IS_ERR(root->node)) {
2562                 ret = PTR_ERR(root->node);
2563                 root->node = NULL;
2564                 return ret;
2565         }
2566         if (!extent_buffer_uptodate(root->node)) {
2567                 free_extent_buffer(root->node);
2568                 root->node = NULL;
2569                 return -EIO;
2570         }
2571
2572         btrfs_set_root_node(&root->root_item, root->node);
2573         root->commit_root = btrfs_root_node(root);
2574         btrfs_set_root_refs(&root->root_item, 1);
2575         return ret;
2576 }
2577
2578 static int load_important_roots(struct btrfs_fs_info *fs_info)
2579 {
2580         struct btrfs_super_block *sb = fs_info->super_copy;
2581         u64 gen, bytenr;
2582         int level, ret;
2583
2584         bytenr = btrfs_super_root(sb);
2585         gen = btrfs_super_generation(sb);
2586         level = btrfs_super_root_level(sb);
2587         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2588         if (ret) {
2589                 btrfs_warn(fs_info, "couldn't read tree root");
2590                 return ret;
2591         }
2592         return 0;
2593 }
2594
2595 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2596 {
2597         int backup_index = find_newest_super_backup(fs_info);
2598         struct btrfs_super_block *sb = fs_info->super_copy;
2599         struct btrfs_root *tree_root = fs_info->tree_root;
2600         bool handle_error = false;
2601         int ret = 0;
2602         int i;
2603
2604         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2605                 if (handle_error) {
2606                         if (!IS_ERR(tree_root->node))
2607                                 free_extent_buffer(tree_root->node);
2608                         tree_root->node = NULL;
2609
2610                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2611                                 break;
2612
2613                         free_root_pointers(fs_info, 0);
2614
2615                         /*
2616                          * Don't use the log in recovery mode, it won't be
2617                          * valid
2618                          */
2619                         btrfs_set_super_log_root(sb, 0);
2620
2621                         /* We can't trust the free space cache either */
2622                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2623
2624                         btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2625                         ret = read_backup_root(fs_info, i);
2626                         backup_index = ret;
2627                         if (ret < 0)
2628                                 return ret;
2629                 }
2630
2631                 ret = load_important_roots(fs_info);
2632                 if (ret) {
2633                         handle_error = true;
2634                         continue;
2635                 }
2636
2637                 /*
2638                  * No need to hold btrfs_root::objectid_mutex since the fs
2639                  * hasn't been fully initialised and we are the only user
2640                  */
2641                 ret = btrfs_init_root_free_objectid(tree_root);
2642                 if (ret < 0) {
2643                         handle_error = true;
2644                         continue;
2645                 }
2646
2647                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2648
2649                 ret = btrfs_read_roots(fs_info);
2650                 if (ret < 0) {
2651                         handle_error = true;
2652                         continue;
2653                 }
2654
2655                 /* All successful */
2656                 fs_info->generation = btrfs_header_generation(tree_root->node);
2657                 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2658                 fs_info->last_reloc_trans = 0;
2659
2660                 /* Always begin writing backup roots after the one being used */
2661                 if (backup_index < 0) {
2662                         fs_info->backup_root_index = 0;
2663                 } else {
2664                         fs_info->backup_root_index = backup_index + 1;
2665                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2666                 }
2667                 break;
2668         }
2669
2670         return ret;
2671 }
2672
2673 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2674 {
2675         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2676         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2677         INIT_LIST_HEAD(&fs_info->trans_list);
2678         INIT_LIST_HEAD(&fs_info->dead_roots);
2679         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2680         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2681         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2682         spin_lock_init(&fs_info->delalloc_root_lock);
2683         spin_lock_init(&fs_info->trans_lock);
2684         spin_lock_init(&fs_info->fs_roots_radix_lock);
2685         spin_lock_init(&fs_info->delayed_iput_lock);
2686         spin_lock_init(&fs_info->defrag_inodes_lock);
2687         spin_lock_init(&fs_info->super_lock);
2688         spin_lock_init(&fs_info->buffer_lock);
2689         spin_lock_init(&fs_info->unused_bgs_lock);
2690         spin_lock_init(&fs_info->treelog_bg_lock);
2691         spin_lock_init(&fs_info->zone_active_bgs_lock);
2692         spin_lock_init(&fs_info->relocation_bg_lock);
2693         rwlock_init(&fs_info->tree_mod_log_lock);
2694         rwlock_init(&fs_info->global_root_lock);
2695         mutex_init(&fs_info->unused_bg_unpin_mutex);
2696         mutex_init(&fs_info->reclaim_bgs_lock);
2697         mutex_init(&fs_info->reloc_mutex);
2698         mutex_init(&fs_info->delalloc_root_mutex);
2699         mutex_init(&fs_info->zoned_meta_io_lock);
2700         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2701         seqlock_init(&fs_info->profiles_lock);
2702
2703         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2704         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2705         btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2706         btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2707         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2708                                      BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2709         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2710                                      BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2711         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2712                                      BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2713         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2714                                      BTRFS_LOCKDEP_TRANS_COMPLETED);
2715
2716         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2717         INIT_LIST_HEAD(&fs_info->space_info);
2718         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2719         INIT_LIST_HEAD(&fs_info->unused_bgs);
2720         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2721         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2722 #ifdef CONFIG_BTRFS_DEBUG
2723         INIT_LIST_HEAD(&fs_info->allocated_roots);
2724         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2725         spin_lock_init(&fs_info->eb_leak_lock);
2726 #endif
2727         extent_map_tree_init(&fs_info->mapping_tree);
2728         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2729                              BTRFS_BLOCK_RSV_GLOBAL);
2730         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2731         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2732         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2733         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2734                              BTRFS_BLOCK_RSV_DELOPS);
2735         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2736                              BTRFS_BLOCK_RSV_DELREFS);
2737
2738         atomic_set(&fs_info->async_delalloc_pages, 0);
2739         atomic_set(&fs_info->defrag_running, 0);
2740         atomic_set(&fs_info->nr_delayed_iputs, 0);
2741         atomic64_set(&fs_info->tree_mod_seq, 0);
2742         fs_info->global_root_tree = RB_ROOT;
2743         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2744         fs_info->metadata_ratio = 0;
2745         fs_info->defrag_inodes = RB_ROOT;
2746         atomic64_set(&fs_info->free_chunk_space, 0);
2747         fs_info->tree_mod_log = RB_ROOT;
2748         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2749         btrfs_init_ref_verify(fs_info);
2750
2751         fs_info->thread_pool_size = min_t(unsigned long,
2752                                           num_online_cpus() + 2, 8);
2753
2754         INIT_LIST_HEAD(&fs_info->ordered_roots);
2755         spin_lock_init(&fs_info->ordered_root_lock);
2756
2757         btrfs_init_scrub(fs_info);
2758         btrfs_init_balance(fs_info);
2759         btrfs_init_async_reclaim_work(fs_info);
2760
2761         rwlock_init(&fs_info->block_group_cache_lock);
2762         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2763
2764         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2765                             IO_TREE_FS_EXCLUDED_EXTENTS);
2766
2767         mutex_init(&fs_info->ordered_operations_mutex);
2768         mutex_init(&fs_info->tree_log_mutex);
2769         mutex_init(&fs_info->chunk_mutex);
2770         mutex_init(&fs_info->transaction_kthread_mutex);
2771         mutex_init(&fs_info->cleaner_mutex);
2772         mutex_init(&fs_info->ro_block_group_mutex);
2773         init_rwsem(&fs_info->commit_root_sem);
2774         init_rwsem(&fs_info->cleanup_work_sem);
2775         init_rwsem(&fs_info->subvol_sem);
2776         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2777
2778         btrfs_init_dev_replace_locks(fs_info);
2779         btrfs_init_qgroup(fs_info);
2780         btrfs_discard_init(fs_info);
2781
2782         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2783         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2784
2785         init_waitqueue_head(&fs_info->transaction_throttle);
2786         init_waitqueue_head(&fs_info->transaction_wait);
2787         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2788         init_waitqueue_head(&fs_info->async_submit_wait);
2789         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2790
2791         /* Usable values until the real ones are cached from the superblock */
2792         fs_info->nodesize = 4096;
2793         fs_info->sectorsize = 4096;
2794         fs_info->sectorsize_bits = ilog2(4096);
2795         fs_info->stripesize = 4096;
2796
2797         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2798
2799         spin_lock_init(&fs_info->swapfile_pins_lock);
2800         fs_info->swapfile_pins = RB_ROOT;
2801
2802         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2803         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2804 }
2805
2806 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2807 {
2808         int ret;
2809
2810         fs_info->sb = sb;
2811         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2812         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2813
2814         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2815         if (ret)
2816                 return ret;
2817
2818         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2819         if (ret)
2820                 return ret;
2821
2822         fs_info->dirty_metadata_batch = PAGE_SIZE *
2823                                         (1 + ilog2(nr_cpu_ids));
2824
2825         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2826         if (ret)
2827                 return ret;
2828
2829         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2830                         GFP_KERNEL);
2831         if (ret)
2832                 return ret;
2833
2834         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2835                                         GFP_KERNEL);
2836         if (!fs_info->delayed_root)
2837                 return -ENOMEM;
2838         btrfs_init_delayed_root(fs_info->delayed_root);
2839
2840         if (sb_rdonly(sb))
2841                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2842
2843         return btrfs_alloc_stripe_hash_table(fs_info);
2844 }
2845
2846 static int btrfs_uuid_rescan_kthread(void *data)
2847 {
2848         struct btrfs_fs_info *fs_info = data;
2849         int ret;
2850
2851         /*
2852          * 1st step is to iterate through the existing UUID tree and
2853          * to delete all entries that contain outdated data.
2854          * 2nd step is to add all missing entries to the UUID tree.
2855          */
2856         ret = btrfs_uuid_tree_iterate(fs_info);
2857         if (ret < 0) {
2858                 if (ret != -EINTR)
2859                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2860                                    ret);
2861                 up(&fs_info->uuid_tree_rescan_sem);
2862                 return ret;
2863         }
2864         return btrfs_uuid_scan_kthread(data);
2865 }
2866
2867 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2868 {
2869         struct task_struct *task;
2870
2871         down(&fs_info->uuid_tree_rescan_sem);
2872         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2873         if (IS_ERR(task)) {
2874                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2875                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2876                 up(&fs_info->uuid_tree_rescan_sem);
2877                 return PTR_ERR(task);
2878         }
2879
2880         return 0;
2881 }
2882
2883 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2884 {
2885         u64 root_objectid = 0;
2886         struct btrfs_root *gang[8];
2887         int i = 0;
2888         int err = 0;
2889         unsigned int ret = 0;
2890
2891         while (1) {
2892                 spin_lock(&fs_info->fs_roots_radix_lock);
2893                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2894                                              (void **)gang, root_objectid,
2895                                              ARRAY_SIZE(gang));
2896                 if (!ret) {
2897                         spin_unlock(&fs_info->fs_roots_radix_lock);
2898                         break;
2899                 }
2900                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2901
2902                 for (i = 0; i < ret; i++) {
2903                         /* Avoid to grab roots in dead_roots. */
2904                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2905                                 gang[i] = NULL;
2906                                 continue;
2907                         }
2908                         /* Grab all the search result for later use. */
2909                         gang[i] = btrfs_grab_root(gang[i]);
2910                 }
2911                 spin_unlock(&fs_info->fs_roots_radix_lock);
2912
2913                 for (i = 0; i < ret; i++) {
2914                         if (!gang[i])
2915                                 continue;
2916                         root_objectid = gang[i]->root_key.objectid;
2917                         err = btrfs_orphan_cleanup(gang[i]);
2918                         if (err)
2919                                 goto out;
2920                         btrfs_put_root(gang[i]);
2921                 }
2922                 root_objectid++;
2923         }
2924 out:
2925         /* Release the uncleaned roots due to error. */
2926         for (; i < ret; i++) {
2927                 if (gang[i])
2928                         btrfs_put_root(gang[i]);
2929         }
2930         return err;
2931 }
2932
2933 /*
2934  * Some options only have meaning at mount time and shouldn't persist across
2935  * remounts, or be displayed. Clear these at the end of mount and remount
2936  * code paths.
2937  */
2938 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2939 {
2940         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2941         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2942 }
2943
2944 /*
2945  * Mounting logic specific to read-write file systems. Shared by open_ctree
2946  * and btrfs_remount when remounting from read-only to read-write.
2947  */
2948 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2949 {
2950         int ret;
2951         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2952         bool rebuild_free_space_tree = false;
2953
2954         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2955             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2956                 rebuild_free_space_tree = true;
2957         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2958                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2959                 btrfs_warn(fs_info, "free space tree is invalid");
2960                 rebuild_free_space_tree = true;
2961         }
2962
2963         if (rebuild_free_space_tree) {
2964                 btrfs_info(fs_info, "rebuilding free space tree");
2965                 ret = btrfs_rebuild_free_space_tree(fs_info);
2966                 if (ret) {
2967                         btrfs_warn(fs_info,
2968                                    "failed to rebuild free space tree: %d", ret);
2969                         goto out;
2970                 }
2971         }
2972
2973         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2974             !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2975                 btrfs_info(fs_info, "disabling free space tree");
2976                 ret = btrfs_delete_free_space_tree(fs_info);
2977                 if (ret) {
2978                         btrfs_warn(fs_info,
2979                                    "failed to disable free space tree: %d", ret);
2980                         goto out;
2981                 }
2982         }
2983
2984         /*
2985          * btrfs_find_orphan_roots() is responsible for finding all the dead
2986          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
2987          * them into the fs_info->fs_roots_radix tree. This must be done before
2988          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
2989          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
2990          * item before the root's tree is deleted - this means that if we unmount
2991          * or crash before the deletion completes, on the next mount we will not
2992          * delete what remains of the tree because the orphan item does not
2993          * exists anymore, which is what tells us we have a pending deletion.
2994          */
2995         ret = btrfs_find_orphan_roots(fs_info);
2996         if (ret)
2997                 goto out;
2998
2999         ret = btrfs_cleanup_fs_roots(fs_info);
3000         if (ret)
3001                 goto out;
3002
3003         down_read(&fs_info->cleanup_work_sem);
3004         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3005             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3006                 up_read(&fs_info->cleanup_work_sem);
3007                 goto out;
3008         }
3009         up_read(&fs_info->cleanup_work_sem);
3010
3011         mutex_lock(&fs_info->cleaner_mutex);
3012         ret = btrfs_recover_relocation(fs_info);
3013         mutex_unlock(&fs_info->cleaner_mutex);
3014         if (ret < 0) {
3015                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3016                 goto out;
3017         }
3018
3019         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3020             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3021                 btrfs_info(fs_info, "creating free space tree");
3022                 ret = btrfs_create_free_space_tree(fs_info);
3023                 if (ret) {
3024                         btrfs_warn(fs_info,
3025                                 "failed to create free space tree: %d", ret);
3026                         goto out;
3027                 }
3028         }
3029
3030         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3031                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3032                 if (ret)
3033                         goto out;
3034         }
3035
3036         ret = btrfs_resume_balance_async(fs_info);
3037         if (ret)
3038                 goto out;
3039
3040         ret = btrfs_resume_dev_replace_async(fs_info);
3041         if (ret) {
3042                 btrfs_warn(fs_info, "failed to resume dev_replace");
3043                 goto out;
3044         }
3045
3046         btrfs_qgroup_rescan_resume(fs_info);
3047
3048         if (!fs_info->uuid_root) {
3049                 btrfs_info(fs_info, "creating UUID tree");
3050                 ret = btrfs_create_uuid_tree(fs_info);
3051                 if (ret) {
3052                         btrfs_warn(fs_info,
3053                                    "failed to create the UUID tree %d", ret);
3054                         goto out;
3055                 }
3056         }
3057
3058 out:
3059         return ret;
3060 }
3061
3062 /*
3063  * Do various sanity and dependency checks of different features.
3064  *
3065  * @is_rw_mount:        If the mount is read-write.
3066  *
3067  * This is the place for less strict checks (like for subpage or artificial
3068  * feature dependencies).
3069  *
3070  * For strict checks or possible corruption detection, see
3071  * btrfs_validate_super().
3072  *
3073  * This should be called after btrfs_parse_options(), as some mount options
3074  * (space cache related) can modify on-disk format like free space tree and
3075  * screw up certain feature dependencies.
3076  */
3077 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3078 {
3079         struct btrfs_super_block *disk_super = fs_info->super_copy;
3080         u64 incompat = btrfs_super_incompat_flags(disk_super);
3081         const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3082         const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3083
3084         if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3085                 btrfs_err(fs_info,
3086                 "cannot mount because of unknown incompat features (0x%llx)",
3087                     incompat);
3088                 return -EINVAL;
3089         }
3090
3091         /* Runtime limitation for mixed block groups. */
3092         if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3093             (fs_info->sectorsize != fs_info->nodesize)) {
3094                 btrfs_err(fs_info,
3095 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3096                         fs_info->nodesize, fs_info->sectorsize);
3097                 return -EINVAL;
3098         }
3099
3100         /* Mixed backref is an always-enabled feature. */
3101         incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3102
3103         /* Set compression related flags just in case. */
3104         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3105                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3106         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3107                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3108
3109         /*
3110          * An ancient flag, which should really be marked deprecated.
3111          * Such runtime limitation doesn't really need a incompat flag.
3112          */
3113         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3114                 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3115
3116         if (compat_ro_unsupp && is_rw_mount) {
3117                 btrfs_err(fs_info,
3118         "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3119                        compat_ro);
3120                 return -EINVAL;
3121         }
3122
3123         /*
3124          * We have unsupported RO compat features, although RO mounted, we
3125          * should not cause any metadata writes, including log replay.
3126          * Or we could screw up whatever the new feature requires.
3127          */
3128         if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3129             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3130                 btrfs_err(fs_info,
3131 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3132                           compat_ro);
3133                 return -EINVAL;
3134         }
3135
3136         /*
3137          * Artificial limitations for block group tree, to force
3138          * block-group-tree to rely on no-holes and free-space-tree.
3139          */
3140         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3141             (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3142              !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3143                 btrfs_err(fs_info,
3144 "block-group-tree feature requires no-holes and free-space-tree features");
3145                 return -EINVAL;
3146         }
3147
3148         /*
3149          * Subpage runtime limitation on v1 cache.
3150          *
3151          * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3152          * we're already defaulting to v2 cache, no need to bother v1 as it's
3153          * going to be deprecated anyway.
3154          */
3155         if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3156                 btrfs_warn(fs_info,
3157         "v1 space cache is not supported for page size %lu with sectorsize %u",
3158                            PAGE_SIZE, fs_info->sectorsize);
3159                 return -EINVAL;
3160         }
3161
3162         /* This can be called by remount, we need to protect the super block. */
3163         spin_lock(&fs_info->super_lock);
3164         btrfs_set_super_incompat_flags(disk_super, incompat);
3165         spin_unlock(&fs_info->super_lock);
3166
3167         return 0;
3168 }
3169
3170 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3171                       char *options)
3172 {
3173         u32 sectorsize;
3174         u32 nodesize;
3175         u32 stripesize;
3176         u64 generation;
3177         u16 csum_type;
3178         struct btrfs_super_block *disk_super;
3179         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3180         struct btrfs_root *tree_root;
3181         struct btrfs_root *chunk_root;
3182         int ret;
3183         int level;
3184
3185         ret = init_mount_fs_info(fs_info, sb);
3186         if (ret)
3187                 goto fail;
3188
3189         /* These need to be init'ed before we start creating inodes and such. */
3190         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3191                                      GFP_KERNEL);
3192         fs_info->tree_root = tree_root;
3193         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3194                                       GFP_KERNEL);
3195         fs_info->chunk_root = chunk_root;
3196         if (!tree_root || !chunk_root) {
3197                 ret = -ENOMEM;
3198                 goto fail;
3199         }
3200
3201         ret = btrfs_init_btree_inode(sb);
3202         if (ret)
3203                 goto fail;
3204
3205         invalidate_bdev(fs_devices->latest_dev->bdev);
3206
3207         /*
3208          * Read super block and check the signature bytes only
3209          */
3210         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3211         if (IS_ERR(disk_super)) {
3212                 ret = PTR_ERR(disk_super);
3213                 goto fail_alloc;
3214         }
3215
3216         /*
3217          * Verify the type first, if that or the checksum value are
3218          * corrupted, we'll find out
3219          */
3220         csum_type = btrfs_super_csum_type(disk_super);
3221         if (!btrfs_supported_super_csum(csum_type)) {
3222                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3223                           csum_type);
3224                 ret = -EINVAL;
3225                 btrfs_release_disk_super(disk_super);
3226                 goto fail_alloc;
3227         }
3228
3229         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3230
3231         ret = btrfs_init_csum_hash(fs_info, csum_type);
3232         if (ret) {
3233                 btrfs_release_disk_super(disk_super);
3234                 goto fail_alloc;
3235         }
3236
3237         /*
3238          * We want to check superblock checksum, the type is stored inside.
3239          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3240          */
3241         if (btrfs_check_super_csum(fs_info, disk_super)) {
3242                 btrfs_err(fs_info, "superblock checksum mismatch");
3243                 ret = -EINVAL;
3244                 btrfs_release_disk_super(disk_super);
3245                 goto fail_alloc;
3246         }
3247
3248         /*
3249          * super_copy is zeroed at allocation time and we never touch the
3250          * following bytes up to INFO_SIZE, the checksum is calculated from
3251          * the whole block of INFO_SIZE
3252          */
3253         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3254         btrfs_release_disk_super(disk_super);
3255
3256         disk_super = fs_info->super_copy;
3257
3258         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3259                sizeof(*fs_info->super_for_commit));
3260
3261         ret = btrfs_validate_mount_super(fs_info);
3262         if (ret) {
3263                 btrfs_err(fs_info, "superblock contains fatal errors");
3264                 ret = -EINVAL;
3265                 goto fail_alloc;
3266         }
3267
3268         if (!btrfs_super_root(disk_super)) {
3269                 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3270                 ret = -EINVAL;
3271                 goto fail_alloc;
3272         }
3273
3274         /* check FS state, whether FS is broken. */
3275         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3276                 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3277
3278         /*
3279          * In the long term, we'll store the compression type in the super
3280          * block, and it'll be used for per file compression control.
3281          */
3282         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3283
3284
3285         /* Set up fs_info before parsing mount options */
3286         nodesize = btrfs_super_nodesize(disk_super);
3287         sectorsize = btrfs_super_sectorsize(disk_super);
3288         stripesize = sectorsize;
3289         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3290         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3291
3292         fs_info->nodesize = nodesize;
3293         fs_info->sectorsize = sectorsize;
3294         fs_info->sectorsize_bits = ilog2(sectorsize);
3295         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3296         fs_info->stripesize = stripesize;
3297
3298         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3299         if (ret)
3300                 goto fail_alloc;
3301
3302         ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3303         if (ret < 0)
3304                 goto fail_alloc;
3305
3306         if (sectorsize < PAGE_SIZE) {
3307                 struct btrfs_subpage_info *subpage_info;
3308
3309                 /*
3310                  * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3311                  * going to be deprecated.
3312                  *
3313                  * Force to use v2 cache for subpage case.
3314                  */
3315                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3316                 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3317                         "forcing free space tree for sector size %u with page size %lu",
3318                         sectorsize, PAGE_SIZE);
3319
3320                 btrfs_warn(fs_info,
3321                 "read-write for sector size %u with page size %lu is experimental",
3322                            sectorsize, PAGE_SIZE);
3323                 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3324                 if (!subpage_info) {
3325                         ret = -ENOMEM;
3326                         goto fail_alloc;
3327                 }
3328                 btrfs_init_subpage_info(subpage_info, sectorsize);
3329                 fs_info->subpage_info = subpage_info;
3330         }
3331
3332         ret = btrfs_init_workqueues(fs_info);
3333         if (ret)
3334                 goto fail_sb_buffer;
3335
3336         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3337         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3338
3339         sb->s_blocksize = sectorsize;
3340         sb->s_blocksize_bits = blksize_bits(sectorsize);
3341         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3342
3343         mutex_lock(&fs_info->chunk_mutex);
3344         ret = btrfs_read_sys_array(fs_info);
3345         mutex_unlock(&fs_info->chunk_mutex);
3346         if (ret) {
3347                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3348                 goto fail_sb_buffer;
3349         }
3350
3351         generation = btrfs_super_chunk_root_generation(disk_super);
3352         level = btrfs_super_chunk_root_level(disk_super);
3353         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3354                               generation, level);
3355         if (ret) {
3356                 btrfs_err(fs_info, "failed to read chunk root");
3357                 goto fail_tree_roots;
3358         }
3359
3360         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3361                            offsetof(struct btrfs_header, chunk_tree_uuid),
3362                            BTRFS_UUID_SIZE);
3363
3364         ret = btrfs_read_chunk_tree(fs_info);
3365         if (ret) {
3366                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3367                 goto fail_tree_roots;
3368         }
3369
3370         /*
3371          * At this point we know all the devices that make this filesystem,
3372          * including the seed devices but we don't know yet if the replace
3373          * target is required. So free devices that are not part of this
3374          * filesystem but skip the replace target device which is checked
3375          * below in btrfs_init_dev_replace().
3376          */
3377         btrfs_free_extra_devids(fs_devices);
3378         if (!fs_devices->latest_dev->bdev) {
3379                 btrfs_err(fs_info, "failed to read devices");
3380                 ret = -EIO;
3381                 goto fail_tree_roots;
3382         }
3383
3384         ret = init_tree_roots(fs_info);
3385         if (ret)
3386                 goto fail_tree_roots;
3387
3388         /*
3389          * Get zone type information of zoned block devices. This will also
3390          * handle emulation of a zoned filesystem if a regular device has the
3391          * zoned incompat feature flag set.
3392          */
3393         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3394         if (ret) {
3395                 btrfs_err(fs_info,
3396                           "zoned: failed to read device zone info: %d", ret);
3397                 goto fail_block_groups;
3398         }
3399
3400         /*
3401          * If we have a uuid root and we're not being told to rescan we need to
3402          * check the generation here so we can set the
3403          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3404          * transaction during a balance or the log replay without updating the
3405          * uuid generation, and then if we crash we would rescan the uuid tree,
3406          * even though it was perfectly fine.
3407          */
3408         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3409             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3410                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3411
3412         ret = btrfs_verify_dev_extents(fs_info);
3413         if (ret) {
3414                 btrfs_err(fs_info,
3415                           "failed to verify dev extents against chunks: %d",
3416                           ret);
3417                 goto fail_block_groups;
3418         }
3419         ret = btrfs_recover_balance(fs_info);
3420         if (ret) {
3421                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3422                 goto fail_block_groups;
3423         }
3424
3425         ret = btrfs_init_dev_stats(fs_info);
3426         if (ret) {
3427                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3428                 goto fail_block_groups;
3429         }
3430
3431         ret = btrfs_init_dev_replace(fs_info);
3432         if (ret) {
3433                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3434                 goto fail_block_groups;
3435         }
3436
3437         ret = btrfs_check_zoned_mode(fs_info);
3438         if (ret) {
3439                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3440                           ret);
3441                 goto fail_block_groups;
3442         }
3443
3444         ret = btrfs_sysfs_add_fsid(fs_devices);
3445         if (ret) {
3446                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3447                                 ret);
3448                 goto fail_block_groups;
3449         }
3450
3451         ret = btrfs_sysfs_add_mounted(fs_info);
3452         if (ret) {
3453                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3454                 goto fail_fsdev_sysfs;
3455         }
3456
3457         ret = btrfs_init_space_info(fs_info);
3458         if (ret) {
3459                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3460                 goto fail_sysfs;
3461         }
3462
3463         ret = btrfs_read_block_groups(fs_info);
3464         if (ret) {
3465                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3466                 goto fail_sysfs;
3467         }
3468
3469         btrfs_free_zone_cache(fs_info);
3470
3471         btrfs_check_active_zone_reservation(fs_info);
3472
3473         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3474             !btrfs_check_rw_degradable(fs_info, NULL)) {
3475                 btrfs_warn(fs_info,
3476                 "writable mount is not allowed due to too many missing devices");
3477                 ret = -EINVAL;
3478                 goto fail_sysfs;
3479         }
3480
3481         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3482                                                "btrfs-cleaner");
3483         if (IS_ERR(fs_info->cleaner_kthread)) {
3484                 ret = PTR_ERR(fs_info->cleaner_kthread);
3485                 goto fail_sysfs;
3486         }
3487
3488         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3489                                                    tree_root,
3490                                                    "btrfs-transaction");
3491         if (IS_ERR(fs_info->transaction_kthread)) {
3492                 ret = PTR_ERR(fs_info->transaction_kthread);
3493                 goto fail_cleaner;
3494         }
3495
3496         if (!btrfs_test_opt(fs_info, NOSSD) &&
3497             !fs_info->fs_devices->rotating) {
3498                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3499         }
3500
3501         /*
3502          * For devices supporting discard turn on discard=async automatically,
3503          * unless it's already set or disabled. This could be turned off by
3504          * nodiscard for the same mount.
3505          *
3506          * The zoned mode piggy backs on the discard functionality for
3507          * resetting a zone. There is no reason to delay the zone reset as it is
3508          * fast enough. So, do not enable async discard for zoned mode.
3509          */
3510         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3511               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3512               btrfs_test_opt(fs_info, NODISCARD)) &&
3513             fs_info->fs_devices->discardable &&
3514             !btrfs_is_zoned(fs_info)) {
3515                 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3516                                    "auto enabling async discard");
3517         }
3518
3519         ret = btrfs_read_qgroup_config(fs_info);
3520         if (ret)
3521                 goto fail_trans_kthread;
3522
3523         if (btrfs_build_ref_tree(fs_info))
3524                 btrfs_err(fs_info, "couldn't build ref tree");
3525
3526         /* do not make disk changes in broken FS or nologreplay is given */
3527         if (btrfs_super_log_root(disk_super) != 0 &&
3528             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529                 btrfs_info(fs_info, "start tree-log replay");
3530                 ret = btrfs_replay_log(fs_info, fs_devices);
3531                 if (ret)
3532                         goto fail_qgroup;
3533         }
3534
3535         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3536         if (IS_ERR(fs_info->fs_root)) {
3537                 ret = PTR_ERR(fs_info->fs_root);
3538                 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539                 fs_info->fs_root = NULL;
3540                 goto fail_qgroup;
3541         }
3542
3543         if (sb_rdonly(sb))
3544                 goto clear_oneshot;
3545
3546         ret = btrfs_start_pre_rw_mount(fs_info);
3547         if (ret) {
3548                 close_ctree(fs_info);
3549                 return ret;
3550         }
3551         btrfs_discard_resume(fs_info);
3552
3553         if (fs_info->uuid_root &&
3554             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3556                 btrfs_info(fs_info, "checking UUID tree");
3557                 ret = btrfs_check_uuid_tree(fs_info);
3558                 if (ret) {
3559                         btrfs_warn(fs_info,
3560                                 "failed to check the UUID tree: %d", ret);
3561                         close_ctree(fs_info);
3562                         return ret;
3563                 }
3564         }
3565
3566         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3567
3568         /* Kick the cleaner thread so it'll start deleting snapshots. */
3569         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570                 wake_up_process(fs_info->cleaner_kthread);
3571
3572 clear_oneshot:
3573         btrfs_clear_oneshot_options(fs_info);
3574         return 0;
3575
3576 fail_qgroup:
3577         btrfs_free_qgroup_config(fs_info);
3578 fail_trans_kthread:
3579         kthread_stop(fs_info->transaction_kthread);
3580         btrfs_cleanup_transaction(fs_info);
3581         btrfs_free_fs_roots(fs_info);
3582 fail_cleaner:
3583         kthread_stop(fs_info->cleaner_kthread);
3584
3585         /*
3586          * make sure we're done with the btree inode before we stop our
3587          * kthreads
3588          */
3589         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3590
3591 fail_sysfs:
3592         btrfs_sysfs_remove_mounted(fs_info);
3593
3594 fail_fsdev_sysfs:
3595         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3596
3597 fail_block_groups:
3598         btrfs_put_block_group_cache(fs_info);
3599
3600 fail_tree_roots:
3601         if (fs_info->data_reloc_root)
3602                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3603         free_root_pointers(fs_info, true);
3604         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3605
3606 fail_sb_buffer:
3607         btrfs_stop_all_workers(fs_info);
3608         btrfs_free_block_groups(fs_info);
3609 fail_alloc:
3610         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3611
3612         iput(fs_info->btree_inode);
3613 fail:
3614         btrfs_close_devices(fs_info->fs_devices);
3615         ASSERT(ret < 0);
3616         return ret;
3617 }
3618 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3619
3620 static void btrfs_end_super_write(struct bio *bio)
3621 {
3622         struct btrfs_device *device = bio->bi_private;
3623         struct bio_vec *bvec;
3624         struct bvec_iter_all iter_all;
3625         struct page *page;
3626
3627         bio_for_each_segment_all(bvec, bio, iter_all) {
3628                 page = bvec->bv_page;
3629
3630                 if (bio->bi_status) {
3631                         btrfs_warn_rl_in_rcu(device->fs_info,
3632                                 "lost page write due to IO error on %s (%d)",
3633                                 btrfs_dev_name(device),
3634                                 blk_status_to_errno(bio->bi_status));
3635                         ClearPageUptodate(page);
3636                         SetPageError(page);
3637                         btrfs_dev_stat_inc_and_print(device,
3638                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3639                 } else {
3640                         SetPageUptodate(page);
3641                 }
3642
3643                 put_page(page);
3644                 unlock_page(page);
3645         }
3646
3647         bio_put(bio);
3648 }
3649
3650 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3651                                                    int copy_num, bool drop_cache)
3652 {
3653         struct btrfs_super_block *super;
3654         struct page *page;
3655         u64 bytenr, bytenr_orig;
3656         struct address_space *mapping = bdev->bd_inode->i_mapping;
3657         int ret;
3658
3659         bytenr_orig = btrfs_sb_offset(copy_num);
3660         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3661         if (ret == -ENOENT)
3662                 return ERR_PTR(-EINVAL);
3663         else if (ret)
3664                 return ERR_PTR(ret);
3665
3666         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3667                 return ERR_PTR(-EINVAL);
3668
3669         if (drop_cache) {
3670                 /* This should only be called with the primary sb. */
3671                 ASSERT(copy_num == 0);
3672
3673                 /*
3674                  * Drop the page of the primary superblock, so later read will
3675                  * always read from the device.
3676                  */
3677                 invalidate_inode_pages2_range(mapping,
3678                                 bytenr >> PAGE_SHIFT,
3679                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3680         }
3681
3682         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3683         if (IS_ERR(page))
3684                 return ERR_CAST(page);
3685
3686         super = page_address(page);
3687         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3688                 btrfs_release_disk_super(super);
3689                 return ERR_PTR(-ENODATA);
3690         }
3691
3692         if (btrfs_super_bytenr(super) != bytenr_orig) {
3693                 btrfs_release_disk_super(super);
3694                 return ERR_PTR(-EINVAL);
3695         }
3696
3697         return super;
3698 }
3699
3700
3701 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3702 {
3703         struct btrfs_super_block *super, *latest = NULL;
3704         int i;
3705         u64 transid = 0;
3706
3707         /* we would like to check all the supers, but that would make
3708          * a btrfs mount succeed after a mkfs from a different FS.
3709          * So, we need to add a special mount option to scan for
3710          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3711          */
3712         for (i = 0; i < 1; i++) {
3713                 super = btrfs_read_dev_one_super(bdev, i, false);
3714                 if (IS_ERR(super))
3715                         continue;
3716
3717                 if (!latest || btrfs_super_generation(super) > transid) {
3718                         if (latest)
3719                                 btrfs_release_disk_super(super);
3720
3721                         latest = super;
3722                         transid = btrfs_super_generation(super);
3723                 }
3724         }
3725
3726         return super;
3727 }
3728
3729 /*
3730  * Write superblock @sb to the @device. Do not wait for completion, all the
3731  * pages we use for writing are locked.
3732  *
3733  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3734  * the expected device size at commit time. Note that max_mirrors must be
3735  * same for write and wait phases.
3736  *
3737  * Return number of errors when page is not found or submission fails.
3738  */
3739 static int write_dev_supers(struct btrfs_device *device,
3740                             struct btrfs_super_block *sb, int max_mirrors)
3741 {
3742         struct btrfs_fs_info *fs_info = device->fs_info;
3743         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3744         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3745         int i;
3746         int errors = 0;
3747         int ret;
3748         u64 bytenr, bytenr_orig;
3749
3750         if (max_mirrors == 0)
3751                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3752
3753         shash->tfm = fs_info->csum_shash;
3754
3755         for (i = 0; i < max_mirrors; i++) {
3756                 struct page *page;
3757                 struct bio *bio;
3758                 struct btrfs_super_block *disk_super;
3759
3760                 bytenr_orig = btrfs_sb_offset(i);
3761                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3762                 if (ret == -ENOENT) {
3763                         continue;
3764                 } else if (ret < 0) {
3765                         btrfs_err(device->fs_info,
3766                                 "couldn't get super block location for mirror %d",
3767                                 i);
3768                         errors++;
3769                         continue;
3770                 }
3771                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3772                     device->commit_total_bytes)
3773                         break;
3774
3775                 btrfs_set_super_bytenr(sb, bytenr_orig);
3776
3777                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3778                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3779                                     sb->csum);
3780
3781                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3782                                            GFP_NOFS);
3783                 if (!page) {
3784                         btrfs_err(device->fs_info,
3785                             "couldn't get super block page for bytenr %llu",
3786                             bytenr);
3787                         errors++;
3788                         continue;
3789                 }
3790
3791                 /* Bump the refcount for wait_dev_supers() */
3792                 get_page(page);
3793
3794                 disk_super = page_address(page);
3795                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3796
3797                 /*
3798                  * Directly use bios here instead of relying on the page cache
3799                  * to do I/O, so we don't lose the ability to do integrity
3800                  * checking.
3801                  */
3802                 bio = bio_alloc(device->bdev, 1,
3803                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3804                                 GFP_NOFS);
3805                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3806                 bio->bi_private = device;
3807                 bio->bi_end_io = btrfs_end_super_write;
3808                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3809                                offset_in_page(bytenr));
3810
3811                 /*
3812                  * We FUA only the first super block.  The others we allow to
3813                  * go down lazy and there's a short window where the on-disk
3814                  * copies might still contain the older version.
3815                  */
3816                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3817                         bio->bi_opf |= REQ_FUA;
3818                 submit_bio(bio);
3819
3820                 if (btrfs_advance_sb_log(device, i))
3821                         errors++;
3822         }
3823         return errors < i ? 0 : -1;
3824 }
3825
3826 /*
3827  * Wait for write completion of superblocks done by write_dev_supers,
3828  * @max_mirrors same for write and wait phases.
3829  *
3830  * Return number of errors when page is not found or not marked up to
3831  * date.
3832  */
3833 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3834 {
3835         int i;
3836         int errors = 0;
3837         bool primary_failed = false;
3838         int ret;
3839         u64 bytenr;
3840
3841         if (max_mirrors == 0)
3842                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3843
3844         for (i = 0; i < max_mirrors; i++) {
3845                 struct page *page;
3846
3847                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3848                 if (ret == -ENOENT) {
3849                         break;
3850                 } else if (ret < 0) {
3851                         errors++;
3852                         if (i == 0)
3853                                 primary_failed = true;
3854                         continue;
3855                 }
3856                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3857                     device->commit_total_bytes)
3858                         break;
3859
3860                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3861                                      bytenr >> PAGE_SHIFT);
3862                 if (!page) {
3863                         errors++;
3864                         if (i == 0)
3865                                 primary_failed = true;
3866                         continue;
3867                 }
3868                 /* Page is submitted locked and unlocked once the IO completes */
3869                 wait_on_page_locked(page);
3870                 if (PageError(page)) {
3871                         errors++;
3872                         if (i == 0)
3873                                 primary_failed = true;
3874                 }
3875
3876                 /* Drop our reference */
3877                 put_page(page);
3878
3879                 /* Drop the reference from the writing run */
3880                 put_page(page);
3881         }
3882
3883         /* log error, force error return */
3884         if (primary_failed) {
3885                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3886                           device->devid);
3887                 return -1;
3888         }
3889
3890         return errors < i ? 0 : -1;
3891 }
3892
3893 /*
3894  * endio for the write_dev_flush, this will wake anyone waiting
3895  * for the barrier when it is done
3896  */
3897 static void btrfs_end_empty_barrier(struct bio *bio)
3898 {
3899         bio_uninit(bio);
3900         complete(bio->bi_private);
3901 }
3902
3903 /*
3904  * Submit a flush request to the device if it supports it. Error handling is
3905  * done in the waiting counterpart.
3906  */
3907 static void write_dev_flush(struct btrfs_device *device)
3908 {
3909         struct bio *bio = &device->flush_bio;
3910
3911         device->last_flush_error = BLK_STS_OK;
3912
3913         bio_init(bio, device->bdev, NULL, 0,
3914                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3915         bio->bi_end_io = btrfs_end_empty_barrier;
3916         init_completion(&device->flush_wait);
3917         bio->bi_private = &device->flush_wait;
3918         submit_bio(bio);
3919         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3920 }
3921
3922 /*
3923  * If the flush bio has been submitted by write_dev_flush, wait for it.
3924  * Return true for any error, and false otherwise.
3925  */
3926 static bool wait_dev_flush(struct btrfs_device *device)
3927 {
3928         struct bio *bio = &device->flush_bio;
3929
3930         if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3931                 return false;
3932
3933         wait_for_completion_io(&device->flush_wait);
3934
3935         if (bio->bi_status) {
3936                 device->last_flush_error = bio->bi_status;
3937                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3938                 return true;
3939         }
3940
3941         return false;
3942 }
3943
3944 /*
3945  * send an empty flush down to each device in parallel,
3946  * then wait for them
3947  */
3948 static int barrier_all_devices(struct btrfs_fs_info *info)
3949 {
3950         struct list_head *head;
3951         struct btrfs_device *dev;
3952         int errors_wait = 0;
3953
3954         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3955         /* send down all the barriers */
3956         head = &info->fs_devices->devices;
3957         list_for_each_entry(dev, head, dev_list) {
3958                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3959                         continue;
3960                 if (!dev->bdev)
3961                         continue;
3962                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3963                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3964                         continue;
3965
3966                 write_dev_flush(dev);
3967         }
3968
3969         /* wait for all the barriers */
3970         list_for_each_entry(dev, head, dev_list) {
3971                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3972                         continue;
3973                 if (!dev->bdev) {
3974                         errors_wait++;
3975                         continue;
3976                 }
3977                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3978                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3979                         continue;
3980
3981                 if (wait_dev_flush(dev))
3982                         errors_wait++;
3983         }
3984
3985         /*
3986          * Checks last_flush_error of disks in order to determine the device
3987          * state.
3988          */
3989         if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3990                 return -EIO;
3991
3992         return 0;
3993 }
3994
3995 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3996 {
3997         int raid_type;
3998         int min_tolerated = INT_MAX;
3999
4000         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4001             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4002                 min_tolerated = min_t(int, min_tolerated,
4003                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4004                                     tolerated_failures);
4005
4006         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4007                 if (raid_type == BTRFS_RAID_SINGLE)
4008                         continue;
4009                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4010                         continue;
4011                 min_tolerated = min_t(int, min_tolerated,
4012                                     btrfs_raid_array[raid_type].
4013                                     tolerated_failures);
4014         }
4015
4016         if (min_tolerated == INT_MAX) {
4017                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4018                 min_tolerated = 0;
4019         }
4020
4021         return min_tolerated;
4022 }
4023
4024 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4025 {
4026         struct list_head *head;
4027         struct btrfs_device *dev;
4028         struct btrfs_super_block *sb;
4029         struct btrfs_dev_item *dev_item;
4030         int ret;
4031         int do_barriers;
4032         int max_errors;
4033         int total_errors = 0;
4034         u64 flags;
4035
4036         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4037
4038         /*
4039          * max_mirrors == 0 indicates we're from commit_transaction,
4040          * not from fsync where the tree roots in fs_info have not
4041          * been consistent on disk.
4042          */
4043         if (max_mirrors == 0)
4044                 backup_super_roots(fs_info);
4045
4046         sb = fs_info->super_for_commit;
4047         dev_item = &sb->dev_item;
4048
4049         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4050         head = &fs_info->fs_devices->devices;
4051         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4052
4053         if (do_barriers) {
4054                 ret = barrier_all_devices(fs_info);
4055                 if (ret) {
4056                         mutex_unlock(
4057                                 &fs_info->fs_devices->device_list_mutex);
4058                         btrfs_handle_fs_error(fs_info, ret,
4059                                               "errors while submitting device barriers.");
4060                         return ret;
4061                 }
4062         }
4063
4064         list_for_each_entry(dev, head, dev_list) {
4065                 if (!dev->bdev) {
4066                         total_errors++;
4067                         continue;
4068                 }
4069                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4070                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4071                         continue;
4072
4073                 btrfs_set_stack_device_generation(dev_item, 0);
4074                 btrfs_set_stack_device_type(dev_item, dev->type);
4075                 btrfs_set_stack_device_id(dev_item, dev->devid);
4076                 btrfs_set_stack_device_total_bytes(dev_item,
4077                                                    dev->commit_total_bytes);
4078                 btrfs_set_stack_device_bytes_used(dev_item,
4079                                                   dev->commit_bytes_used);
4080                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4081                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4082                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4083                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4084                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4085                        BTRFS_FSID_SIZE);
4086
4087                 flags = btrfs_super_flags(sb);
4088                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4089
4090                 ret = btrfs_validate_write_super(fs_info, sb);
4091                 if (ret < 0) {
4092                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4093                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4094                                 "unexpected superblock corruption detected");
4095                         return -EUCLEAN;
4096                 }
4097
4098                 ret = write_dev_supers(dev, sb, max_mirrors);
4099                 if (ret)
4100                         total_errors++;
4101         }
4102         if (total_errors > max_errors) {
4103                 btrfs_err(fs_info, "%d errors while writing supers",
4104                           total_errors);
4105                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4106
4107                 /* FUA is masked off if unsupported and can't be the reason */
4108                 btrfs_handle_fs_error(fs_info, -EIO,
4109                                       "%d errors while writing supers",
4110                                       total_errors);
4111                 return -EIO;
4112         }
4113
4114         total_errors = 0;
4115         list_for_each_entry(dev, head, dev_list) {
4116                 if (!dev->bdev)
4117                         continue;
4118                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4119                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4120                         continue;
4121
4122                 ret = wait_dev_supers(dev, max_mirrors);
4123                 if (ret)
4124                         total_errors++;
4125         }
4126         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4127         if (total_errors > max_errors) {
4128                 btrfs_handle_fs_error(fs_info, -EIO,
4129                                       "%d errors while writing supers",
4130                                       total_errors);
4131                 return -EIO;
4132         }
4133         return 0;
4134 }
4135
4136 /* Drop a fs root from the radix tree and free it. */
4137 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4138                                   struct btrfs_root *root)
4139 {
4140         bool drop_ref = false;
4141
4142         spin_lock(&fs_info->fs_roots_radix_lock);
4143         radix_tree_delete(&fs_info->fs_roots_radix,
4144                           (unsigned long)root->root_key.objectid);
4145         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4146                 drop_ref = true;
4147         spin_unlock(&fs_info->fs_roots_radix_lock);
4148
4149         if (BTRFS_FS_ERROR(fs_info)) {
4150                 ASSERT(root->log_root == NULL);
4151                 if (root->reloc_root) {
4152                         btrfs_put_root(root->reloc_root);
4153                         root->reloc_root = NULL;
4154                 }
4155         }
4156
4157         if (drop_ref)
4158                 btrfs_put_root(root);
4159 }
4160
4161 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4162 {
4163         struct btrfs_root *root = fs_info->tree_root;
4164         struct btrfs_trans_handle *trans;
4165
4166         mutex_lock(&fs_info->cleaner_mutex);
4167         btrfs_run_delayed_iputs(fs_info);
4168         mutex_unlock(&fs_info->cleaner_mutex);
4169         wake_up_process(fs_info->cleaner_kthread);
4170
4171         /* wait until ongoing cleanup work done */
4172         down_write(&fs_info->cleanup_work_sem);
4173         up_write(&fs_info->cleanup_work_sem);
4174
4175         trans = btrfs_join_transaction(root);
4176         if (IS_ERR(trans))
4177                 return PTR_ERR(trans);
4178         return btrfs_commit_transaction(trans);
4179 }
4180
4181 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4182 {
4183         struct btrfs_transaction *trans;
4184         struct btrfs_transaction *tmp;
4185         bool found = false;
4186
4187         if (list_empty(&fs_info->trans_list))
4188                 return;
4189
4190         /*
4191          * This function is only called at the very end of close_ctree(),
4192          * thus no other running transaction, no need to take trans_lock.
4193          */
4194         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4195         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4196                 struct extent_state *cached = NULL;
4197                 u64 dirty_bytes = 0;
4198                 u64 cur = 0;
4199                 u64 found_start;
4200                 u64 found_end;
4201
4202                 found = true;
4203                 while (find_first_extent_bit(&trans->dirty_pages, cur,
4204                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4205                         dirty_bytes += found_end + 1 - found_start;
4206                         cur = found_end + 1;
4207                 }
4208                 btrfs_warn(fs_info,
4209         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4210                            trans->transid, dirty_bytes);
4211                 btrfs_cleanup_one_transaction(trans, fs_info);
4212
4213                 if (trans == fs_info->running_transaction)
4214                         fs_info->running_transaction = NULL;
4215                 list_del_init(&trans->list);
4216
4217                 btrfs_put_transaction(trans);
4218                 trace_btrfs_transaction_commit(fs_info);
4219         }
4220         ASSERT(!found);
4221 }
4222
4223 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4224 {
4225         int ret;
4226
4227         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4228
4229         /*
4230          * If we had UNFINISHED_DROPS we could still be processing them, so
4231          * clear that bit and wake up relocation so it can stop.
4232          * We must do this before stopping the block group reclaim task, because
4233          * at btrfs_relocate_block_group() we wait for this bit, and after the
4234          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4235          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4236          * return 1.
4237          */
4238         btrfs_wake_unfinished_drop(fs_info);
4239
4240         /*
4241          * We may have the reclaim task running and relocating a data block group,
4242          * in which case it may create delayed iputs. So stop it before we park
4243          * the cleaner kthread otherwise we can get new delayed iputs after
4244          * parking the cleaner, and that can make the async reclaim task to hang
4245          * if it's waiting for delayed iputs to complete, since the cleaner is
4246          * parked and can not run delayed iputs - this will make us hang when
4247          * trying to stop the async reclaim task.
4248          */
4249         cancel_work_sync(&fs_info->reclaim_bgs_work);
4250         /*
4251          * We don't want the cleaner to start new transactions, add more delayed
4252          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4253          * because that frees the task_struct, and the transaction kthread might
4254          * still try to wake up the cleaner.
4255          */
4256         kthread_park(fs_info->cleaner_kthread);
4257
4258         /* wait for the qgroup rescan worker to stop */
4259         btrfs_qgroup_wait_for_completion(fs_info, false);
4260
4261         /* wait for the uuid_scan task to finish */
4262         down(&fs_info->uuid_tree_rescan_sem);
4263         /* avoid complains from lockdep et al., set sem back to initial state */
4264         up(&fs_info->uuid_tree_rescan_sem);
4265
4266         /* pause restriper - we want to resume on mount */
4267         btrfs_pause_balance(fs_info);
4268
4269         btrfs_dev_replace_suspend_for_unmount(fs_info);
4270
4271         btrfs_scrub_cancel(fs_info);
4272
4273         /* wait for any defraggers to finish */
4274         wait_event(fs_info->transaction_wait,
4275                    (atomic_read(&fs_info->defrag_running) == 0));
4276
4277         /* clear out the rbtree of defraggable inodes */
4278         btrfs_cleanup_defrag_inodes(fs_info);
4279
4280         /*
4281          * After we parked the cleaner kthread, ordered extents may have
4282          * completed and created new delayed iputs. If one of the async reclaim
4283          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4284          * can hang forever trying to stop it, because if a delayed iput is
4285          * added after it ran btrfs_run_delayed_iputs() and before it called
4286          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4287          * no one else to run iputs.
4288          *
4289          * So wait for all ongoing ordered extents to complete and then run
4290          * delayed iputs. This works because once we reach this point no one
4291          * can either create new ordered extents nor create delayed iputs
4292          * through some other means.
4293          *
4294          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4295          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4296          * but the delayed iput for the respective inode is made only when doing
4297          * the final btrfs_put_ordered_extent() (which must happen at
4298          * btrfs_finish_ordered_io() when we are unmounting).
4299          */
4300         btrfs_flush_workqueue(fs_info->endio_write_workers);
4301         /* Ordered extents for free space inodes. */
4302         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4303         btrfs_run_delayed_iputs(fs_info);
4304
4305         cancel_work_sync(&fs_info->async_reclaim_work);
4306         cancel_work_sync(&fs_info->async_data_reclaim_work);
4307         cancel_work_sync(&fs_info->preempt_reclaim_work);
4308
4309         /* Cancel or finish ongoing discard work */
4310         btrfs_discard_cleanup(fs_info);
4311
4312         if (!sb_rdonly(fs_info->sb)) {
4313                 /*
4314                  * The cleaner kthread is stopped, so do one final pass over
4315                  * unused block groups.
4316                  */
4317                 btrfs_delete_unused_bgs(fs_info);
4318
4319                 /*
4320                  * There might be existing delayed inode workers still running
4321                  * and holding an empty delayed inode item. We must wait for
4322                  * them to complete first because they can create a transaction.
4323                  * This happens when someone calls btrfs_balance_delayed_items()
4324                  * and then a transaction commit runs the same delayed nodes
4325                  * before any delayed worker has done something with the nodes.
4326                  * We must wait for any worker here and not at transaction
4327                  * commit time since that could cause a deadlock.
4328                  * This is a very rare case.
4329                  */
4330                 btrfs_flush_workqueue(fs_info->delayed_workers);
4331
4332                 ret = btrfs_commit_super(fs_info);
4333                 if (ret)
4334                         btrfs_err(fs_info, "commit super ret %d", ret);
4335         }
4336
4337         if (BTRFS_FS_ERROR(fs_info))
4338                 btrfs_error_commit_super(fs_info);
4339
4340         kthread_stop(fs_info->transaction_kthread);
4341         kthread_stop(fs_info->cleaner_kthread);
4342
4343         ASSERT(list_empty(&fs_info->delayed_iputs));
4344         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4345
4346         if (btrfs_check_quota_leak(fs_info)) {
4347                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4348                 btrfs_err(fs_info, "qgroup reserved space leaked");
4349         }
4350
4351         btrfs_free_qgroup_config(fs_info);
4352         ASSERT(list_empty(&fs_info->delalloc_roots));
4353
4354         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4355                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4356                        percpu_counter_sum(&fs_info->delalloc_bytes));
4357         }
4358
4359         if (percpu_counter_sum(&fs_info->ordered_bytes))
4360                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4361                            percpu_counter_sum(&fs_info->ordered_bytes));
4362
4363         btrfs_sysfs_remove_mounted(fs_info);
4364         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4365
4366         btrfs_put_block_group_cache(fs_info);
4367
4368         /*
4369          * we must make sure there is not any read request to
4370          * submit after we stopping all workers.
4371          */
4372         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4373         btrfs_stop_all_workers(fs_info);
4374
4375         /* We shouldn't have any transaction open at this point */
4376         warn_about_uncommitted_trans(fs_info);
4377
4378         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4379         free_root_pointers(fs_info, true);
4380         btrfs_free_fs_roots(fs_info);
4381
4382         /*
4383          * We must free the block groups after dropping the fs_roots as we could
4384          * have had an IO error and have left over tree log blocks that aren't
4385          * cleaned up until the fs roots are freed.  This makes the block group
4386          * accounting appear to be wrong because there's pending reserved bytes,
4387          * so make sure we do the block group cleanup afterwards.
4388          */
4389         btrfs_free_block_groups(fs_info);
4390
4391         iput(fs_info->btree_inode);
4392
4393         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4394         btrfs_close_devices(fs_info->fs_devices);
4395 }
4396
4397 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4398                              struct extent_buffer *buf)
4399 {
4400         struct btrfs_fs_info *fs_info = buf->fs_info;
4401         u64 transid = btrfs_header_generation(buf);
4402
4403 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4404         /*
4405          * This is a fast path so only do this check if we have sanity tests
4406          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4407          * outside of the sanity tests.
4408          */
4409         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4410                 return;
4411 #endif
4412         /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4413         ASSERT(trans->transid == fs_info->generation);
4414         btrfs_assert_tree_write_locked(buf);
4415         if (unlikely(transid != fs_info->generation)) {
4416                 btrfs_abort_transaction(trans, -EUCLEAN);
4417                 btrfs_crit(fs_info,
4418 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4419                            buf->start, transid, fs_info->generation);
4420         }
4421         set_extent_buffer_dirty(buf);
4422 }
4423
4424 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4425                                         int flush_delayed)
4426 {
4427         /*
4428          * looks as though older kernels can get into trouble with
4429          * this code, they end up stuck in balance_dirty_pages forever
4430          */
4431         int ret;
4432
4433         if (current->flags & PF_MEMALLOC)
4434                 return;
4435
4436         if (flush_delayed)
4437                 btrfs_balance_delayed_items(fs_info);
4438
4439         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4440                                      BTRFS_DIRTY_METADATA_THRESH,
4441                                      fs_info->dirty_metadata_batch);
4442         if (ret > 0) {
4443                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4444         }
4445 }
4446
4447 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4448 {
4449         __btrfs_btree_balance_dirty(fs_info, 1);
4450 }
4451
4452 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4453 {
4454         __btrfs_btree_balance_dirty(fs_info, 0);
4455 }
4456
4457 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4458 {
4459         /* cleanup FS via transaction */
4460         btrfs_cleanup_transaction(fs_info);
4461
4462         mutex_lock(&fs_info->cleaner_mutex);
4463         btrfs_run_delayed_iputs(fs_info);
4464         mutex_unlock(&fs_info->cleaner_mutex);
4465
4466         down_write(&fs_info->cleanup_work_sem);
4467         up_write(&fs_info->cleanup_work_sem);
4468 }
4469
4470 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4471 {
4472         struct btrfs_root *gang[8];
4473         u64 root_objectid = 0;
4474         int ret;
4475
4476         spin_lock(&fs_info->fs_roots_radix_lock);
4477         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4478                                              (void **)gang, root_objectid,
4479                                              ARRAY_SIZE(gang))) != 0) {
4480                 int i;
4481
4482                 for (i = 0; i < ret; i++)
4483                         gang[i] = btrfs_grab_root(gang[i]);
4484                 spin_unlock(&fs_info->fs_roots_radix_lock);
4485
4486                 for (i = 0; i < ret; i++) {
4487                         if (!gang[i])
4488                                 continue;
4489                         root_objectid = gang[i]->root_key.objectid;
4490                         btrfs_free_log(NULL, gang[i]);
4491                         btrfs_put_root(gang[i]);
4492                 }
4493                 root_objectid++;
4494                 spin_lock(&fs_info->fs_roots_radix_lock);
4495         }
4496         spin_unlock(&fs_info->fs_roots_radix_lock);
4497         btrfs_free_log_root_tree(NULL, fs_info);
4498 }
4499
4500 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4501 {
4502         struct btrfs_ordered_extent *ordered;
4503
4504         spin_lock(&root->ordered_extent_lock);
4505         /*
4506          * This will just short circuit the ordered completion stuff which will
4507          * make sure the ordered extent gets properly cleaned up.
4508          */
4509         list_for_each_entry(ordered, &root->ordered_extents,
4510                             root_extent_list)
4511                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4512         spin_unlock(&root->ordered_extent_lock);
4513 }
4514
4515 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4516 {
4517         struct btrfs_root *root;
4518         LIST_HEAD(splice);
4519
4520         spin_lock(&fs_info->ordered_root_lock);
4521         list_splice_init(&fs_info->ordered_roots, &splice);
4522         while (!list_empty(&splice)) {
4523                 root = list_first_entry(&splice, struct btrfs_root,
4524                                         ordered_root);
4525                 list_move_tail(&root->ordered_root,
4526                                &fs_info->ordered_roots);
4527
4528                 spin_unlock(&fs_info->ordered_root_lock);
4529                 btrfs_destroy_ordered_extents(root);
4530
4531                 cond_resched();
4532                 spin_lock(&fs_info->ordered_root_lock);
4533         }
4534         spin_unlock(&fs_info->ordered_root_lock);
4535
4536         /*
4537          * We need this here because if we've been flipped read-only we won't
4538          * get sync() from the umount, so we need to make sure any ordered
4539          * extents that haven't had their dirty pages IO start writeout yet
4540          * actually get run and error out properly.
4541          */
4542         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4543 }
4544
4545 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4546                                        struct btrfs_fs_info *fs_info)
4547 {
4548         struct rb_node *node;
4549         struct btrfs_delayed_ref_root *delayed_refs;
4550         struct btrfs_delayed_ref_node *ref;
4551
4552         delayed_refs = &trans->delayed_refs;
4553
4554         spin_lock(&delayed_refs->lock);
4555         if (atomic_read(&delayed_refs->num_entries) == 0) {
4556                 spin_unlock(&delayed_refs->lock);
4557                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4558                 return;
4559         }
4560
4561         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4562                 struct btrfs_delayed_ref_head *head;
4563                 struct rb_node *n;
4564                 bool pin_bytes = false;
4565
4566                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4567                                 href_node);
4568                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4569                         continue;
4570
4571                 spin_lock(&head->lock);
4572                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4573                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4574                                        ref_node);
4575                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4576                         RB_CLEAR_NODE(&ref->ref_node);
4577                         if (!list_empty(&ref->add_list))
4578                                 list_del(&ref->add_list);
4579                         atomic_dec(&delayed_refs->num_entries);
4580                         btrfs_put_delayed_ref(ref);
4581                         btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4582                 }
4583                 if (head->must_insert_reserved)
4584                         pin_bytes = true;
4585                 btrfs_free_delayed_extent_op(head->extent_op);
4586                 btrfs_delete_ref_head(delayed_refs, head);
4587                 spin_unlock(&head->lock);
4588                 spin_unlock(&delayed_refs->lock);
4589                 mutex_unlock(&head->mutex);
4590
4591                 if (pin_bytes) {
4592                         struct btrfs_block_group *cache;
4593
4594                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4595                         BUG_ON(!cache);
4596
4597                         spin_lock(&cache->space_info->lock);
4598                         spin_lock(&cache->lock);
4599                         cache->pinned += head->num_bytes;
4600                         btrfs_space_info_update_bytes_pinned(fs_info,
4601                                 cache->space_info, head->num_bytes);
4602                         cache->reserved -= head->num_bytes;
4603                         cache->space_info->bytes_reserved -= head->num_bytes;
4604                         spin_unlock(&cache->lock);
4605                         spin_unlock(&cache->space_info->lock);
4606
4607                         btrfs_put_block_group(cache);
4608
4609                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4610                                 head->bytenr + head->num_bytes - 1);
4611                 }
4612                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4613                 btrfs_put_delayed_ref_head(head);
4614                 cond_resched();
4615                 spin_lock(&delayed_refs->lock);
4616         }
4617         btrfs_qgroup_destroy_extent_records(trans);
4618
4619         spin_unlock(&delayed_refs->lock);
4620 }
4621
4622 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4623 {
4624         struct btrfs_inode *btrfs_inode;
4625         LIST_HEAD(splice);
4626
4627         spin_lock(&root->delalloc_lock);
4628         list_splice_init(&root->delalloc_inodes, &splice);
4629
4630         while (!list_empty(&splice)) {
4631                 struct inode *inode = NULL;
4632                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4633                                                delalloc_inodes);
4634                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4635                 spin_unlock(&root->delalloc_lock);
4636
4637                 /*
4638                  * Make sure we get a live inode and that it'll not disappear
4639                  * meanwhile.
4640                  */
4641                 inode = igrab(&btrfs_inode->vfs_inode);
4642                 if (inode) {
4643                         unsigned int nofs_flag;
4644
4645                         nofs_flag = memalloc_nofs_save();
4646                         invalidate_inode_pages2(inode->i_mapping);
4647                         memalloc_nofs_restore(nofs_flag);
4648                         iput(inode);
4649                 }
4650                 spin_lock(&root->delalloc_lock);
4651         }
4652         spin_unlock(&root->delalloc_lock);
4653 }
4654
4655 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4656 {
4657         struct btrfs_root *root;
4658         LIST_HEAD(splice);
4659
4660         spin_lock(&fs_info->delalloc_root_lock);
4661         list_splice_init(&fs_info->delalloc_roots, &splice);
4662         while (!list_empty(&splice)) {
4663                 root = list_first_entry(&splice, struct btrfs_root,
4664                                          delalloc_root);
4665                 root = btrfs_grab_root(root);
4666                 BUG_ON(!root);
4667                 spin_unlock(&fs_info->delalloc_root_lock);
4668
4669                 btrfs_destroy_delalloc_inodes(root);
4670                 btrfs_put_root(root);
4671
4672                 spin_lock(&fs_info->delalloc_root_lock);
4673         }
4674         spin_unlock(&fs_info->delalloc_root_lock);
4675 }
4676
4677 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4678                                          struct extent_io_tree *dirty_pages,
4679                                          int mark)
4680 {
4681         struct extent_buffer *eb;
4682         u64 start = 0;
4683         u64 end;
4684
4685         while (find_first_extent_bit(dirty_pages, start, &start, &end,
4686                                      mark, NULL)) {
4687                 clear_extent_bits(dirty_pages, start, end, mark);
4688                 while (start <= end) {
4689                         eb = find_extent_buffer(fs_info, start);
4690                         start += fs_info->nodesize;
4691                         if (!eb)
4692                                 continue;
4693
4694                         btrfs_tree_lock(eb);
4695                         wait_on_extent_buffer_writeback(eb);
4696                         btrfs_clear_buffer_dirty(NULL, eb);
4697                         btrfs_tree_unlock(eb);
4698
4699                         free_extent_buffer_stale(eb);
4700                 }
4701         }
4702 }
4703
4704 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4705                                         struct extent_io_tree *unpin)
4706 {
4707         u64 start;
4708         u64 end;
4709
4710         while (1) {
4711                 struct extent_state *cached_state = NULL;
4712
4713                 /*
4714                  * The btrfs_finish_extent_commit() may get the same range as
4715                  * ours between find_first_extent_bit and clear_extent_dirty.
4716                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4717                  * the same extent range.
4718                  */
4719                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4720                 if (!find_first_extent_bit(unpin, 0, &start, &end,
4721                                            EXTENT_DIRTY, &cached_state)) {
4722                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4723                         break;
4724                 }
4725
4726                 clear_extent_dirty(unpin, start, end, &cached_state);
4727                 free_extent_state(cached_state);
4728                 btrfs_error_unpin_extent_range(fs_info, start, end);
4729                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4730                 cond_resched();
4731         }
4732 }
4733
4734 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4735 {
4736         struct inode *inode;
4737
4738         inode = cache->io_ctl.inode;
4739         if (inode) {
4740                 unsigned int nofs_flag;
4741
4742                 nofs_flag = memalloc_nofs_save();
4743                 invalidate_inode_pages2(inode->i_mapping);
4744                 memalloc_nofs_restore(nofs_flag);
4745
4746                 BTRFS_I(inode)->generation = 0;
4747                 cache->io_ctl.inode = NULL;
4748                 iput(inode);
4749         }
4750         ASSERT(cache->io_ctl.pages == NULL);
4751         btrfs_put_block_group(cache);
4752 }
4753
4754 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4755                              struct btrfs_fs_info *fs_info)
4756 {
4757         struct btrfs_block_group *cache;
4758
4759         spin_lock(&cur_trans->dirty_bgs_lock);
4760         while (!list_empty(&cur_trans->dirty_bgs)) {
4761                 cache = list_first_entry(&cur_trans->dirty_bgs,
4762                                          struct btrfs_block_group,
4763                                          dirty_list);
4764
4765                 if (!list_empty(&cache->io_list)) {
4766                         spin_unlock(&cur_trans->dirty_bgs_lock);
4767                         list_del_init(&cache->io_list);
4768                         btrfs_cleanup_bg_io(cache);
4769                         spin_lock(&cur_trans->dirty_bgs_lock);
4770                 }
4771
4772                 list_del_init(&cache->dirty_list);
4773                 spin_lock(&cache->lock);
4774                 cache->disk_cache_state = BTRFS_DC_ERROR;
4775                 spin_unlock(&cache->lock);
4776
4777                 spin_unlock(&cur_trans->dirty_bgs_lock);
4778                 btrfs_put_block_group(cache);
4779                 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4780                 spin_lock(&cur_trans->dirty_bgs_lock);
4781         }
4782         spin_unlock(&cur_trans->dirty_bgs_lock);
4783
4784         /*
4785          * Refer to the definition of io_bgs member for details why it's safe
4786          * to use it without any locking
4787          */
4788         while (!list_empty(&cur_trans->io_bgs)) {
4789                 cache = list_first_entry(&cur_trans->io_bgs,
4790                                          struct btrfs_block_group,
4791                                          io_list);
4792
4793                 list_del_init(&cache->io_list);
4794                 spin_lock(&cache->lock);
4795                 cache->disk_cache_state = BTRFS_DC_ERROR;
4796                 spin_unlock(&cache->lock);
4797                 btrfs_cleanup_bg_io(cache);
4798         }
4799 }
4800
4801 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4802                                    struct btrfs_fs_info *fs_info)
4803 {
4804         struct btrfs_device *dev, *tmp;
4805
4806         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4807         ASSERT(list_empty(&cur_trans->dirty_bgs));
4808         ASSERT(list_empty(&cur_trans->io_bgs));
4809
4810         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4811                                  post_commit_list) {
4812                 list_del_init(&dev->post_commit_list);
4813         }
4814
4815         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4816
4817         cur_trans->state = TRANS_STATE_COMMIT_START;
4818         wake_up(&fs_info->transaction_blocked_wait);
4819
4820         cur_trans->state = TRANS_STATE_UNBLOCKED;
4821         wake_up(&fs_info->transaction_wait);
4822
4823         btrfs_destroy_delayed_inodes(fs_info);
4824
4825         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4826                                      EXTENT_DIRTY);
4827         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4828
4829         cur_trans->state =TRANS_STATE_COMPLETED;
4830         wake_up(&cur_trans->commit_wait);
4831 }
4832
4833 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4834 {
4835         struct btrfs_transaction *t;
4836
4837         mutex_lock(&fs_info->transaction_kthread_mutex);
4838
4839         spin_lock(&fs_info->trans_lock);
4840         while (!list_empty(&fs_info->trans_list)) {
4841                 t = list_first_entry(&fs_info->trans_list,
4842                                      struct btrfs_transaction, list);
4843                 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4844                         refcount_inc(&t->use_count);
4845                         spin_unlock(&fs_info->trans_lock);
4846                         btrfs_wait_for_commit(fs_info, t->transid);
4847                         btrfs_put_transaction(t);
4848                         spin_lock(&fs_info->trans_lock);
4849                         continue;
4850                 }
4851                 if (t == fs_info->running_transaction) {
4852                         t->state = TRANS_STATE_COMMIT_DOING;
4853                         spin_unlock(&fs_info->trans_lock);
4854                         /*
4855                          * We wait for 0 num_writers since we don't hold a trans
4856                          * handle open currently for this transaction.
4857                          */
4858                         wait_event(t->writer_wait,
4859                                    atomic_read(&t->num_writers) == 0);
4860                 } else {
4861                         spin_unlock(&fs_info->trans_lock);
4862                 }
4863                 btrfs_cleanup_one_transaction(t, fs_info);
4864
4865                 spin_lock(&fs_info->trans_lock);
4866                 if (t == fs_info->running_transaction)
4867                         fs_info->running_transaction = NULL;
4868                 list_del_init(&t->list);
4869                 spin_unlock(&fs_info->trans_lock);
4870
4871                 btrfs_put_transaction(t);
4872                 trace_btrfs_transaction_commit(fs_info);
4873                 spin_lock(&fs_info->trans_lock);
4874         }
4875         spin_unlock(&fs_info->trans_lock);
4876         btrfs_destroy_all_ordered_extents(fs_info);
4877         btrfs_destroy_delayed_inodes(fs_info);
4878         btrfs_assert_delayed_root_empty(fs_info);
4879         btrfs_destroy_all_delalloc_inodes(fs_info);
4880         btrfs_drop_all_logs(fs_info);
4881         mutex_unlock(&fs_info->transaction_kthread_mutex);
4882
4883         return 0;
4884 }
4885
4886 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4887 {
4888         struct btrfs_path *path;
4889         int ret;
4890         struct extent_buffer *l;
4891         struct btrfs_key search_key;
4892         struct btrfs_key found_key;
4893         int slot;
4894
4895         path = btrfs_alloc_path();
4896         if (!path)
4897                 return -ENOMEM;
4898
4899         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4900         search_key.type = -1;
4901         search_key.offset = (u64)-1;
4902         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4903         if (ret < 0)
4904                 goto error;
4905         BUG_ON(ret == 0); /* Corruption */
4906         if (path->slots[0] > 0) {
4907                 slot = path->slots[0] - 1;
4908                 l = path->nodes[0];
4909                 btrfs_item_key_to_cpu(l, &found_key, slot);
4910                 root->free_objectid = max_t(u64, found_key.objectid + 1,
4911                                             BTRFS_FIRST_FREE_OBJECTID);
4912         } else {
4913                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4914         }
4915         ret = 0;
4916 error:
4917         btrfs_free_path(path);
4918         return ret;
4919 }
4920
4921 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4922 {
4923         int ret;
4924         mutex_lock(&root->objectid_mutex);
4925
4926         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4927                 btrfs_warn(root->fs_info,
4928                            "the objectid of root %llu reaches its highest value",
4929                            root->root_key.objectid);
4930                 ret = -ENOSPC;
4931                 goto out;
4932         }
4933
4934         *objectid = root->free_objectid++;
4935         ret = 0;
4936 out:
4937         mutex_unlock(&root->objectid_mutex);
4938         return ret;
4939 }