root/fs/btrfs/disk-io.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. btrfs_end_io_wq_init
  2. btrfs_end_io_wq_exit
  3. btrfs_init_lockdep
  4. btrfs_set_buffer_lockdep_class
  5. btree_get_extent
  6. csum_tree_block
  7. verify_parent_transid
  8. btrfs_supported_super_csum
  9. btrfs_check_super_csum
  10. btrfs_verify_level_key
  11. btree_read_extent_buffer_pages
  12. csum_dirty_buffer
  13. check_tree_block_fsid
  14. btree_readpage_end_io_hook
  15. end_workqueue_bio
  16. btrfs_bio_wq_end_io
  17. run_one_async_start
  18. run_one_async_done
  19. run_one_async_free
  20. btrfs_wq_submit_bio
  21. btree_csum_one_bio
  22. btree_submit_bio_start
  23. check_async_write
  24. btree_submit_bio_hook
  25. btree_migratepage
  26. btree_writepages
  27. btree_readpage
  28. btree_releasepage
  29. btree_invalidatepage
  30. btree_set_page_dirty
  31. readahead_tree_block
  32. btrfs_find_create_tree_block
  33. read_tree_block
  34. btrfs_clean_tree_block
  35. btrfs_alloc_subvolume_writers
  36. btrfs_free_subvolume_writers
  37. __setup_root
  38. btrfs_alloc_root
  39. btrfs_alloc_dummy_root
  40. btrfs_create_tree
  41. alloc_log_tree
  42. btrfs_init_log_root_tree
  43. btrfs_add_log_tree
  44. btrfs_read_tree_root
  45. btrfs_read_fs_root
  46. btrfs_init_fs_root
  47. btrfs_lookup_fs_root
  48. btrfs_insert_fs_root
  49. btrfs_get_fs_root
  50. btrfs_congested_fn
  51. end_workqueue_fn
  52. cleaner_kthread
  53. transaction_kthread
  54. find_newest_super_backup
  55. find_oldest_super_backup
  56. backup_super_roots
  57. next_root_backup
  58. btrfs_stop_all_workers
  59. free_root_extent_buffers
  60. free_root_pointers
  61. btrfs_free_fs_roots
  62. btrfs_init_scrub
  63. btrfs_init_balance
  64. btrfs_init_btree_inode
  65. btrfs_init_dev_replace_locks
  66. btrfs_init_qgroup
  67. btrfs_init_workqueues
  68. btrfs_init_csum_hash
  69. btrfs_free_csum_hash
  70. btrfs_replay_log
  71. btrfs_read_roots
  72. validate_super
  73. btrfs_validate_mount_super
  74. btrfs_validate_write_super
  75. open_ctree
  76. btrfs_end_buffer_write_sync
  77. btrfs_read_dev_one_super
  78. btrfs_read_dev_super
  79. write_dev_supers
  80. wait_dev_supers
  81. btrfs_end_empty_barrier
  82. write_dev_flush
  83. wait_dev_flush
  84. check_barrier_error
  85. barrier_all_devices
  86. btrfs_get_num_tolerated_disk_barrier_failures
  87. write_all_supers
  88. btrfs_drop_and_free_fs_root
  89. btrfs_free_fs_root
  90. btrfs_cleanup_fs_roots
  91. btrfs_commit_super
  92. close_ctree
  93. btrfs_buffer_uptodate
  94. btrfs_mark_buffer_dirty
  95. __btrfs_btree_balance_dirty
  96. btrfs_btree_balance_dirty
  97. btrfs_btree_balance_dirty_nodelay
  98. btrfs_read_buffer
  99. btrfs_error_commit_super
  100. btrfs_destroy_ordered_extents
  101. btrfs_destroy_all_ordered_extents
  102. btrfs_destroy_delayed_refs
  103. btrfs_destroy_delalloc_inodes
  104. btrfs_destroy_all_delalloc_inodes
  105. btrfs_destroy_marked_extents
  106. btrfs_destroy_pinned_extent
  107. btrfs_cleanup_bg_io
  108. btrfs_cleanup_dirty_bgs
  109. btrfs_cleanup_one_transaction
  110. btrfs_cleanup_transaction

   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/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <linux/sched/mm.h>
  21 #include <asm/unaligned.h>
  22 #include <crypto/hash.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "btrfs_inode.h"
  27 #include "volumes.h"
  28 #include "print-tree.h"
  29 #include "locking.h"
  30 #include "tree-log.h"
  31 #include "free-space-cache.h"
  32 #include "free-space-tree.h"
  33 #include "inode-map.h"
  34 #include "check-integrity.h"
  35 #include "rcu-string.h"
  36 #include "dev-replace.h"
  37 #include "raid56.h"
  38 #include "sysfs.h"
  39 #include "qgroup.h"
  40 #include "compression.h"
  41 #include "tree-checker.h"
  42 #include "ref-verify.h"
  43 #include "block-group.h"
  44 
  45 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  46                                  BTRFS_HEADER_FLAG_RELOC |\
  47                                  BTRFS_SUPER_FLAG_ERROR |\
  48                                  BTRFS_SUPER_FLAG_SEEDING |\
  49                                  BTRFS_SUPER_FLAG_METADUMP |\
  50                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  51 
  52 static const struct extent_io_ops btree_extent_io_ops;
  53 static void end_workqueue_fn(struct btrfs_work *work);
  54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  56                                       struct btrfs_fs_info *fs_info);
  57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  65 
  66 /*
  67  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  68  * is complete.  This is used during reads to verify checksums, and it is used
  69  * by writes to insert metadata for new file extents after IO is complete.
  70  */
  71 struct btrfs_end_io_wq {
  72         struct bio *bio;
  73         bio_end_io_t *end_io;
  74         void *private;
  75         struct btrfs_fs_info *info;
  76         blk_status_t status;
  77         enum btrfs_wq_endio_type metadata;
  78         struct btrfs_work work;
  79 };
  80 
  81 static struct kmem_cache *btrfs_end_io_wq_cache;
  82 
  83 int __init btrfs_end_io_wq_init(void)
  84 {
  85         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  86                                         sizeof(struct btrfs_end_io_wq),
  87                                         0,
  88                                         SLAB_MEM_SPREAD,
  89                                         NULL);
  90         if (!btrfs_end_io_wq_cache)
  91                 return -ENOMEM;
  92         return 0;
  93 }
  94 
  95 void __cold btrfs_end_io_wq_exit(void)
  96 {
  97         kmem_cache_destroy(btrfs_end_io_wq_cache);
  98 }
  99 
 100 /*
 101  * async submit bios are used to offload expensive checksumming
 102  * onto the worker threads.  They checksum file and metadata bios
 103  * just before they are sent down the IO stack.
 104  */
 105 struct async_submit_bio {
 106         void *private_data;
 107         struct bio *bio;
 108         extent_submit_bio_start_t *submit_bio_start;
 109         int mirror_num;
 110         /*
 111          * bio_offset is optional, can be used if the pages in the bio
 112          * can't tell us where in the file the bio should go
 113          */
 114         u64 bio_offset;
 115         struct btrfs_work work;
 116         blk_status_t status;
 117 };
 118 
 119 /*
 120  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 121  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 122  * the level the eb occupies in the tree.
 123  *
 124  * Different roots are used for different purposes and may nest inside each
 125  * other and they require separate keysets.  As lockdep keys should be
 126  * static, assign keysets according to the purpose of the root as indicated
 127  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 128  * roots have separate keysets.
 129  *
 130  * Lock-nesting across peer nodes is always done with the immediate parent
 131  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 132  * subclass to avoid triggering lockdep warning in such cases.
 133  *
 134  * The key is set by the readpage_end_io_hook after the buffer has passed
 135  * csum validation but before the pages are unlocked.  It is also set by
 136  * btrfs_init_new_buffer on freshly allocated blocks.
 137  *
 138  * We also add a check to make sure the highest level of the tree is the
 139  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 140  * needs update as well.
 141  */
 142 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 143 # if BTRFS_MAX_LEVEL != 8
 144 #  error
 145 # endif
 146 
 147 static struct btrfs_lockdep_keyset {
 148         u64                     id;             /* root objectid */
 149         const char              *name_stem;     /* lock name stem */
 150         char                    names[BTRFS_MAX_LEVEL + 1][20];
 151         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 152 } btrfs_lockdep_keysets[] = {
 153         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 154         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 155         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 156         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 157         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 158         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 159         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 160         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 161         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 162         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 163         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 164         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 165         { .id = 0,                              .name_stem = "tree"     },
 166 };
 167 
 168 void __init btrfs_init_lockdep(void)
 169 {
 170         int i, j;
 171 
 172         /* initialize lockdep class names */
 173         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 174                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 175 
 176                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 177                         snprintf(ks->names[j], sizeof(ks->names[j]),
 178                                  "btrfs-%s-%02d", ks->name_stem, j);
 179         }
 180 }
 181 
 182 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 183                                     int level)
 184 {
 185         struct btrfs_lockdep_keyset *ks;
 186 
 187         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 188 
 189         /* find the matching keyset, id 0 is the default entry */
 190         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 191                 if (ks->id == objectid)
 192                         break;
 193 
 194         lockdep_set_class_and_name(&eb->lock,
 195                                    &ks->keys[level], ks->names[level]);
 196 }
 197 
 198 #endif
 199 
 200 /*
 201  * extents on the btree inode are pretty simple, there's one extent
 202  * that covers the entire device
 203  */
 204 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 205                 struct page *page, size_t pg_offset, u64 start, u64 len,
 206                 int create)
 207 {
 208         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 209         struct extent_map_tree *em_tree = &inode->extent_tree;
 210         struct extent_map *em;
 211         int ret;
 212 
 213         read_lock(&em_tree->lock);
 214         em = lookup_extent_mapping(em_tree, start, len);
 215         if (em) {
 216                 em->bdev = fs_info->fs_devices->latest_bdev;
 217                 read_unlock(&em_tree->lock);
 218                 goto out;
 219         }
 220         read_unlock(&em_tree->lock);
 221 
 222         em = alloc_extent_map();
 223         if (!em) {
 224                 em = ERR_PTR(-ENOMEM);
 225                 goto out;
 226         }
 227         em->start = 0;
 228         em->len = (u64)-1;
 229         em->block_len = (u64)-1;
 230         em->block_start = 0;
 231         em->bdev = fs_info->fs_devices->latest_bdev;
 232 
 233         write_lock(&em_tree->lock);
 234         ret = add_extent_mapping(em_tree, em, 0);
 235         if (ret == -EEXIST) {
 236                 free_extent_map(em);
 237                 em = lookup_extent_mapping(em_tree, start, len);
 238                 if (!em)
 239                         em = ERR_PTR(-EIO);
 240         } else if (ret) {
 241                 free_extent_map(em);
 242                 em = ERR_PTR(ret);
 243         }
 244         write_unlock(&em_tree->lock);
 245 
 246 out:
 247         return em;
 248 }
 249 
 250 /*
 251  * Compute the csum of a btree block and store the result to provided buffer.
 252  *
 253  * Returns error if the extent buffer cannot be mapped.
 254  */
 255 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
 256 {
 257         struct btrfs_fs_info *fs_info = buf->fs_info;
 258         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 259         unsigned long len;
 260         unsigned long cur_len;
 261         unsigned long offset = BTRFS_CSUM_SIZE;
 262         char *kaddr;
 263         unsigned long map_start;
 264         unsigned long map_len;
 265         int err;
 266 
 267         shash->tfm = fs_info->csum_shash;
 268         crypto_shash_init(shash);
 269 
 270         len = buf->len - offset;
 271 
 272         while (len > 0) {
 273                 /*
 274                  * Note: we don't need to check for the err == 1 case here, as
 275                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 276                  * and 'min_len = 32' and the currently implemented mapping
 277                  * algorithm we cannot cross a page boundary.
 278                  */
 279                 err = map_private_extent_buffer(buf, offset, 32,
 280                                         &kaddr, &map_start, &map_len);
 281                 if (WARN_ON(err))
 282                         return err;
 283                 cur_len = min(len, map_len - (offset - map_start));
 284                 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
 285                 len -= cur_len;
 286                 offset += cur_len;
 287         }
 288         memset(result, 0, BTRFS_CSUM_SIZE);
 289 
 290         crypto_shash_final(shash, result);
 291 
 292         return 0;
 293 }
 294 
 295 /*
 296  * we can't consider a given block up to date unless the transid of the
 297  * block matches the transid in the parent node's pointer.  This is how we
 298  * detect blocks that either didn't get written at all or got written
 299  * in the wrong place.
 300  */
 301 static int verify_parent_transid(struct extent_io_tree *io_tree,
 302                                  struct extent_buffer *eb, u64 parent_transid,
 303                                  int atomic)
 304 {
 305         struct extent_state *cached_state = NULL;
 306         int ret;
 307         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 308 
 309         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 310                 return 0;
 311 
 312         if (atomic)
 313                 return -EAGAIN;
 314 
 315         if (need_lock) {
 316                 btrfs_tree_read_lock(eb);
 317                 btrfs_set_lock_blocking_read(eb);
 318         }
 319 
 320         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 321                          &cached_state);
 322         if (extent_buffer_uptodate(eb) &&
 323             btrfs_header_generation(eb) == parent_transid) {
 324                 ret = 0;
 325                 goto out;
 326         }
 327         btrfs_err_rl(eb->fs_info,
 328                 "parent transid verify failed on %llu wanted %llu found %llu",
 329                         eb->start,
 330                         parent_transid, btrfs_header_generation(eb));
 331         ret = 1;
 332 
 333         /*
 334          * Things reading via commit roots that don't have normal protection,
 335          * like send, can have a really old block in cache that may point at a
 336          * block that has been freed and re-allocated.  So don't clear uptodate
 337          * if we find an eb that is under IO (dirty/writeback) because we could
 338          * end up reading in the stale data and then writing it back out and
 339          * making everybody very sad.
 340          */
 341         if (!extent_buffer_under_io(eb))
 342                 clear_extent_buffer_uptodate(eb);
 343 out:
 344         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 345                              &cached_state);
 346         if (need_lock)
 347                 btrfs_tree_read_unlock_blocking(eb);
 348         return ret;
 349 }
 350 
 351 static bool btrfs_supported_super_csum(u16 csum_type)
 352 {
 353         switch (csum_type) {
 354         case BTRFS_CSUM_TYPE_CRC32:
 355                 return true;
 356         default:
 357                 return false;
 358         }
 359 }
 360 
 361 /*
 362  * Return 0 if the superblock checksum type matches the checksum value of that
 363  * algorithm. Pass the raw disk superblock data.
 364  */
 365 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 366                                   char *raw_disk_sb)
 367 {
 368         struct btrfs_super_block *disk_sb =
 369                 (struct btrfs_super_block *)raw_disk_sb;
 370         char result[BTRFS_CSUM_SIZE];
 371         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 372 
 373         shash->tfm = fs_info->csum_shash;
 374         crypto_shash_init(shash);
 375 
 376         /*
 377          * The super_block structure does not span the whole
 378          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 379          * filled with zeros and is included in the checksum.
 380          */
 381         crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
 382                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 383         crypto_shash_final(shash, result);
 384 
 385         if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
 386                 return 1;
 387 
 388         return 0;
 389 }
 390 
 391 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 392                            struct btrfs_key *first_key, u64 parent_transid)
 393 {
 394         struct btrfs_fs_info *fs_info = eb->fs_info;
 395         int found_level;
 396         struct btrfs_key found_key;
 397         int ret;
 398 
 399         found_level = btrfs_header_level(eb);
 400         if (found_level != level) {
 401                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 402                      KERN_ERR "BTRFS: tree level check failed\n");
 403                 btrfs_err(fs_info,
 404 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 405                           eb->start, level, found_level);
 406                 return -EIO;
 407         }
 408 
 409         if (!first_key)
 410                 return 0;
 411 
 412         /*
 413          * For live tree block (new tree blocks in current transaction),
 414          * we need proper lock context to avoid race, which is impossible here.
 415          * So we only checks tree blocks which is read from disk, whose
 416          * generation <= fs_info->last_trans_committed.
 417          */
 418         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 419                 return 0;
 420 
 421         /* We have @first_key, so this @eb must have at least one item */
 422         if (btrfs_header_nritems(eb) == 0) {
 423                 btrfs_err(fs_info,
 424                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 425                           eb->start);
 426                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 427                 return -EUCLEAN;
 428         }
 429 
 430         if (found_level)
 431                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 432         else
 433                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 434         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 435 
 436         if (ret) {
 437                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 438                      KERN_ERR "BTRFS: tree first key check failed\n");
 439                 btrfs_err(fs_info,
 440 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 441                           eb->start, parent_transid, first_key->objectid,
 442                           first_key->type, first_key->offset,
 443                           found_key.objectid, found_key.type,
 444                           found_key.offset);
 445         }
 446         return ret;
 447 }
 448 
 449 /*
 450  * helper to read a given tree block, doing retries as required when
 451  * the checksums don't match and we have alternate mirrors to try.
 452  *
 453  * @parent_transid:     expected transid, skip check if 0
 454  * @level:              expected level, mandatory check
 455  * @first_key:          expected key of first slot, skip check if NULL
 456  */
 457 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 458                                           u64 parent_transid, int level,
 459                                           struct btrfs_key *first_key)
 460 {
 461         struct btrfs_fs_info *fs_info = eb->fs_info;
 462         struct extent_io_tree *io_tree;
 463         int failed = 0;
 464         int ret;
 465         int num_copies = 0;
 466         int mirror_num = 0;
 467         int failed_mirror = 0;
 468 
 469         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 470         while (1) {
 471                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 472                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 473                 if (!ret) {
 474                         if (verify_parent_transid(io_tree, eb,
 475                                                    parent_transid, 0))
 476                                 ret = -EIO;
 477                         else if (btrfs_verify_level_key(eb, level,
 478                                                 first_key, parent_transid))
 479                                 ret = -EUCLEAN;
 480                         else
 481                                 break;
 482                 }
 483 
 484                 num_copies = btrfs_num_copies(fs_info,
 485                                               eb->start, eb->len);
 486                 if (num_copies == 1)
 487                         break;
 488 
 489                 if (!failed_mirror) {
 490                         failed = 1;
 491                         failed_mirror = eb->read_mirror;
 492                 }
 493 
 494                 mirror_num++;
 495                 if (mirror_num == failed_mirror)
 496                         mirror_num++;
 497 
 498                 if (mirror_num > num_copies)
 499                         break;
 500         }
 501 
 502         if (failed && !ret && failed_mirror)
 503                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 504 
 505         return ret;
 506 }
 507 
 508 /*
 509  * checksum a dirty tree block before IO.  This has extra checks to make sure
 510  * we only fill in the checksum field in the first page of a multi-page block
 511  */
 512 
 513 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 514 {
 515         u64 start = page_offset(page);
 516         u64 found_start;
 517         u8 result[BTRFS_CSUM_SIZE];
 518         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 519         struct extent_buffer *eb;
 520         int ret;
 521 
 522         eb = (struct extent_buffer *)page->private;
 523         if (page != eb->pages[0])
 524                 return 0;
 525 
 526         found_start = btrfs_header_bytenr(eb);
 527         /*
 528          * Please do not consolidate these warnings into a single if.
 529          * It is useful to know what went wrong.
 530          */
 531         if (WARN_ON(found_start != start))
 532                 return -EUCLEAN;
 533         if (WARN_ON(!PageUptodate(page)))
 534                 return -EUCLEAN;
 535 
 536         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 537                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 538 
 539         if (csum_tree_block(eb, result))
 540                 return -EINVAL;
 541 
 542         if (btrfs_header_level(eb))
 543                 ret = btrfs_check_node(eb);
 544         else
 545                 ret = btrfs_check_leaf_full(eb);
 546 
 547         if (ret < 0) {
 548                 btrfs_err(fs_info,
 549                 "block=%llu write time tree block corruption detected",
 550                           eb->start);
 551                 return ret;
 552         }
 553         write_extent_buffer(eb, result, 0, csum_size);
 554 
 555         return 0;
 556 }
 557 
 558 static int check_tree_block_fsid(struct extent_buffer *eb)
 559 {
 560         struct btrfs_fs_info *fs_info = eb->fs_info;
 561         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 562         u8 fsid[BTRFS_FSID_SIZE];
 563         int ret = 1;
 564 
 565         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 566         while (fs_devices) {
 567                 u8 *metadata_uuid;
 568 
 569                 /*
 570                  * Checking the incompat flag is only valid for the current
 571                  * fs. For seed devices it's forbidden to have their uuid
 572                  * changed so reading ->fsid in this case is fine
 573                  */
 574                 if (fs_devices == fs_info->fs_devices &&
 575                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 576                         metadata_uuid = fs_devices->metadata_uuid;
 577                 else
 578                         metadata_uuid = fs_devices->fsid;
 579 
 580                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 581                         ret = 0;
 582                         break;
 583                 }
 584                 fs_devices = fs_devices->seed;
 585         }
 586         return ret;
 587 }
 588 
 589 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 590                                       u64 phy_offset, struct page *page,
 591                                       u64 start, u64 end, int mirror)
 592 {
 593         u64 found_start;
 594         int found_level;
 595         struct extent_buffer *eb;
 596         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 597         struct btrfs_fs_info *fs_info = root->fs_info;
 598         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 599         int ret = 0;
 600         u8 result[BTRFS_CSUM_SIZE];
 601         int reads_done;
 602 
 603         if (!page->private)
 604                 goto out;
 605 
 606         eb = (struct extent_buffer *)page->private;
 607 
 608         /* the pending IO might have been the only thing that kept this buffer
 609          * in memory.  Make sure we have a ref for all this other checks
 610          */
 611         extent_buffer_get(eb);
 612 
 613         reads_done = atomic_dec_and_test(&eb->io_pages);
 614         if (!reads_done)
 615                 goto err;
 616 
 617         eb->read_mirror = mirror;
 618         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 619                 ret = -EIO;
 620                 goto err;
 621         }
 622 
 623         found_start = btrfs_header_bytenr(eb);
 624         if (found_start != eb->start) {
 625                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 626                              eb->start, found_start);
 627                 ret = -EIO;
 628                 goto err;
 629         }
 630         if (check_tree_block_fsid(eb)) {
 631                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 632                              eb->start);
 633                 ret = -EIO;
 634                 goto err;
 635         }
 636         found_level = btrfs_header_level(eb);
 637         if (found_level >= BTRFS_MAX_LEVEL) {
 638                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 639                           (int)btrfs_header_level(eb), eb->start);
 640                 ret = -EIO;
 641                 goto err;
 642         }
 643 
 644         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 645                                        eb, found_level);
 646 
 647         ret = csum_tree_block(eb, result);
 648         if (ret)
 649                 goto err;
 650 
 651         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 652                 u32 val;
 653                 u32 found = 0;
 654 
 655                 memcpy(&found, result, csum_size);
 656 
 657                 read_extent_buffer(eb, &val, 0, csum_size);
 658                 btrfs_warn_rl(fs_info,
 659                 "%s checksum verify failed on %llu wanted %x found %x level %d",
 660                               fs_info->sb->s_id, eb->start,
 661                               val, found, btrfs_header_level(eb));
 662                 ret = -EUCLEAN;
 663                 goto err;
 664         }
 665 
 666         /*
 667          * If this is a leaf block and it is corrupt, set the corrupt bit so
 668          * that we don't try and read the other copies of this block, just
 669          * return -EIO.
 670          */
 671         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 672                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 673                 ret = -EIO;
 674         }
 675 
 676         if (found_level > 0 && btrfs_check_node(eb))
 677                 ret = -EIO;
 678 
 679         if (!ret)
 680                 set_extent_buffer_uptodate(eb);
 681         else
 682                 btrfs_err(fs_info,
 683                           "block=%llu read time tree block corruption detected",
 684                           eb->start);
 685 err:
 686         if (reads_done &&
 687             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 688                 btree_readahead_hook(eb, ret);
 689 
 690         if (ret) {
 691                 /*
 692                  * our io error hook is going to dec the io pages
 693                  * again, we have to make sure it has something
 694                  * to decrement
 695                  */
 696                 atomic_inc(&eb->io_pages);
 697                 clear_extent_buffer_uptodate(eb);
 698         }
 699         free_extent_buffer(eb);
 700 out:
 701         return ret;
 702 }
 703 
 704 static void end_workqueue_bio(struct bio *bio)
 705 {
 706         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 707         struct btrfs_fs_info *fs_info;
 708         struct btrfs_workqueue *wq;
 709 
 710         fs_info = end_io_wq->info;
 711         end_io_wq->status = bio->bi_status;
 712 
 713         if (bio_op(bio) == REQ_OP_WRITE) {
 714                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 715                         wq = fs_info->endio_meta_write_workers;
 716                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 717                         wq = fs_info->endio_freespace_worker;
 718                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 719                         wq = fs_info->endio_raid56_workers;
 720                 else
 721                         wq = fs_info->endio_write_workers;
 722         } else {
 723                 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
 724                         wq = fs_info->endio_repair_workers;
 725                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 726                         wq = fs_info->endio_raid56_workers;
 727                 else if (end_io_wq->metadata)
 728                         wq = fs_info->endio_meta_workers;
 729                 else
 730                         wq = fs_info->endio_workers;
 731         }
 732 
 733         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 734         btrfs_queue_work(wq, &end_io_wq->work);
 735 }
 736 
 737 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 738                         enum btrfs_wq_endio_type metadata)
 739 {
 740         struct btrfs_end_io_wq *end_io_wq;
 741 
 742         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 743         if (!end_io_wq)
 744                 return BLK_STS_RESOURCE;
 745 
 746         end_io_wq->private = bio->bi_private;
 747         end_io_wq->end_io = bio->bi_end_io;
 748         end_io_wq->info = info;
 749         end_io_wq->status = 0;
 750         end_io_wq->bio = bio;
 751         end_io_wq->metadata = metadata;
 752 
 753         bio->bi_private = end_io_wq;
 754         bio->bi_end_io = end_workqueue_bio;
 755         return 0;
 756 }
 757 
 758 static void run_one_async_start(struct btrfs_work *work)
 759 {
 760         struct async_submit_bio *async;
 761         blk_status_t ret;
 762 
 763         async = container_of(work, struct  async_submit_bio, work);
 764         ret = async->submit_bio_start(async->private_data, async->bio,
 765                                       async->bio_offset);
 766         if (ret)
 767                 async->status = ret;
 768 }
 769 
 770 /*
 771  * In order to insert checksums into the metadata in large chunks, we wait
 772  * until bio submission time.   All the pages in the bio are checksummed and
 773  * sums are attached onto the ordered extent record.
 774  *
 775  * At IO completion time the csums attached on the ordered extent record are
 776  * inserted into the tree.
 777  */
 778 static void run_one_async_done(struct btrfs_work *work)
 779 {
 780         struct async_submit_bio *async;
 781         struct inode *inode;
 782         blk_status_t ret;
 783 
 784         async = container_of(work, struct  async_submit_bio, work);
 785         inode = async->private_data;
 786 
 787         /* If an error occurred we just want to clean up the bio and move on */
 788         if (async->status) {
 789                 async->bio->bi_status = async->status;
 790                 bio_endio(async->bio);
 791                 return;
 792         }
 793 
 794         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 795                         async->mirror_num, 1);
 796         if (ret) {
 797                 async->bio->bi_status = ret;
 798                 bio_endio(async->bio);
 799         }
 800 }
 801 
 802 static void run_one_async_free(struct btrfs_work *work)
 803 {
 804         struct async_submit_bio *async;
 805 
 806         async = container_of(work, struct  async_submit_bio, work);
 807         kfree(async);
 808 }
 809 
 810 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 811                                  int mirror_num, unsigned long bio_flags,
 812                                  u64 bio_offset, void *private_data,
 813                                  extent_submit_bio_start_t *submit_bio_start)
 814 {
 815         struct async_submit_bio *async;
 816 
 817         async = kmalloc(sizeof(*async), GFP_NOFS);
 818         if (!async)
 819                 return BLK_STS_RESOURCE;
 820 
 821         async->private_data = private_data;
 822         async->bio = bio;
 823         async->mirror_num = mirror_num;
 824         async->submit_bio_start = submit_bio_start;
 825 
 826         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 827                         run_one_async_free);
 828 
 829         async->bio_offset = bio_offset;
 830 
 831         async->status = 0;
 832 
 833         if (op_is_sync(bio->bi_opf))
 834                 btrfs_set_work_high_priority(&async->work);
 835 
 836         btrfs_queue_work(fs_info->workers, &async->work);
 837         return 0;
 838 }
 839 
 840 static blk_status_t btree_csum_one_bio(struct bio *bio)
 841 {
 842         struct bio_vec *bvec;
 843         struct btrfs_root *root;
 844         int ret = 0;
 845         struct bvec_iter_all iter_all;
 846 
 847         ASSERT(!bio_flagged(bio, BIO_CLONED));
 848         bio_for_each_segment_all(bvec, bio, iter_all) {
 849                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 850                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 851                 if (ret)
 852                         break;
 853         }
 854 
 855         return errno_to_blk_status(ret);
 856 }
 857 
 858 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 859                                              u64 bio_offset)
 860 {
 861         /*
 862          * when we're called for a write, we're already in the async
 863          * submission context.  Just jump into btrfs_map_bio
 864          */
 865         return btree_csum_one_bio(bio);
 866 }
 867 
 868 static int check_async_write(struct btrfs_fs_info *fs_info,
 869                              struct btrfs_inode *bi)
 870 {
 871         if (atomic_read(&bi->sync_writers))
 872                 return 0;
 873         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 874                 return 0;
 875         return 1;
 876 }
 877 
 878 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 879                                           int mirror_num,
 880                                           unsigned long bio_flags)
 881 {
 882         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 883         int async = check_async_write(fs_info, BTRFS_I(inode));
 884         blk_status_t ret;
 885 
 886         if (bio_op(bio) != REQ_OP_WRITE) {
 887                 /*
 888                  * called for a read, do the setup so that checksum validation
 889                  * can happen in the async kernel threads
 890                  */
 891                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 892                                           BTRFS_WQ_ENDIO_METADATA);
 893                 if (ret)
 894                         goto out_w_error;
 895                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 896         } else if (!async) {
 897                 ret = btree_csum_one_bio(bio);
 898                 if (ret)
 899                         goto out_w_error;
 900                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 901         } else {
 902                 /*
 903                  * kthread helpers are used to submit writes so that
 904                  * checksumming can happen in parallel across all CPUs
 905                  */
 906                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 907                                           0, inode, btree_submit_bio_start);
 908         }
 909 
 910         if (ret)
 911                 goto out_w_error;
 912         return 0;
 913 
 914 out_w_error:
 915         bio->bi_status = ret;
 916         bio_endio(bio);
 917         return ret;
 918 }
 919 
 920 #ifdef CONFIG_MIGRATION
 921 static int btree_migratepage(struct address_space *mapping,
 922                         struct page *newpage, struct page *page,
 923                         enum migrate_mode mode)
 924 {
 925         /*
 926          * we can't safely write a btree page from here,
 927          * we haven't done the locking hook
 928          */
 929         if (PageDirty(page))
 930                 return -EAGAIN;
 931         /*
 932          * Buffers may be managed in a filesystem specific way.
 933          * We must have no buffers or drop them.
 934          */
 935         if (page_has_private(page) &&
 936             !try_to_release_page(page, GFP_KERNEL))
 937                 return -EAGAIN;
 938         return migrate_page(mapping, newpage, page, mode);
 939 }
 940 #endif
 941 
 942 
 943 static int btree_writepages(struct address_space *mapping,
 944                             struct writeback_control *wbc)
 945 {
 946         struct btrfs_fs_info *fs_info;
 947         int ret;
 948 
 949         if (wbc->sync_mode == WB_SYNC_NONE) {
 950 
 951                 if (wbc->for_kupdate)
 952                         return 0;
 953 
 954                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 955                 /* this is a bit racy, but that's ok */
 956                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 957                                              BTRFS_DIRTY_METADATA_THRESH,
 958                                              fs_info->dirty_metadata_batch);
 959                 if (ret < 0)
 960                         return 0;
 961         }
 962         return btree_write_cache_pages(mapping, wbc);
 963 }
 964 
 965 static int btree_readpage(struct file *file, struct page *page)
 966 {
 967         struct extent_io_tree *tree;
 968         tree = &BTRFS_I(page->mapping->host)->io_tree;
 969         return extent_read_full_page(tree, page, btree_get_extent, 0);
 970 }
 971 
 972 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 973 {
 974         if (PageWriteback(page) || PageDirty(page))
 975                 return 0;
 976 
 977         return try_release_extent_buffer(page);
 978 }
 979 
 980 static void btree_invalidatepage(struct page *page, unsigned int offset,
 981                                  unsigned int length)
 982 {
 983         struct extent_io_tree *tree;
 984         tree = &BTRFS_I(page->mapping->host)->io_tree;
 985         extent_invalidatepage(tree, page, offset);
 986         btree_releasepage(page, GFP_NOFS);
 987         if (PagePrivate(page)) {
 988                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 989                            "page private not zero on page %llu",
 990                            (unsigned long long)page_offset(page));
 991                 ClearPagePrivate(page);
 992                 set_page_private(page, 0);
 993                 put_page(page);
 994         }
 995 }
 996 
 997 static int btree_set_page_dirty(struct page *page)
 998 {
 999 #ifdef DEBUG
1000         struct extent_buffer *eb;
1001 
1002         BUG_ON(!PagePrivate(page));
1003         eb = (struct extent_buffer *)page->private;
1004         BUG_ON(!eb);
1005         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1006         BUG_ON(!atomic_read(&eb->refs));
1007         btrfs_assert_tree_locked(eb);
1008 #endif
1009         return __set_page_dirty_nobuffers(page);
1010 }
1011 
1012 static const struct address_space_operations btree_aops = {
1013         .readpage       = btree_readpage,
1014         .writepages     = btree_writepages,
1015         .releasepage    = btree_releasepage,
1016         .invalidatepage = btree_invalidatepage,
1017 #ifdef CONFIG_MIGRATION
1018         .migratepage    = btree_migratepage,
1019 #endif
1020         .set_page_dirty = btree_set_page_dirty,
1021 };
1022 
1023 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1024 {
1025         struct extent_buffer *buf = NULL;
1026         int ret;
1027 
1028         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1029         if (IS_ERR(buf))
1030                 return;
1031 
1032         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1033         if (ret < 0)
1034                 free_extent_buffer_stale(buf);
1035         else
1036                 free_extent_buffer(buf);
1037 }
1038 
1039 struct extent_buffer *btrfs_find_create_tree_block(
1040                                                 struct btrfs_fs_info *fs_info,
1041                                                 u64 bytenr)
1042 {
1043         if (btrfs_is_testing(fs_info))
1044                 return alloc_test_extent_buffer(fs_info, bytenr);
1045         return alloc_extent_buffer(fs_info, bytenr);
1046 }
1047 
1048 /*
1049  * Read tree block at logical address @bytenr and do variant basic but critical
1050  * verification.
1051  *
1052  * @parent_transid:     expected transid of this tree block, skip check if 0
1053  * @level:              expected level, mandatory check
1054  * @first_key:          expected key in slot 0, skip check if NULL
1055  */
1056 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1057                                       u64 parent_transid, int level,
1058                                       struct btrfs_key *first_key)
1059 {
1060         struct extent_buffer *buf = NULL;
1061         int ret;
1062 
1063         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1064         if (IS_ERR(buf))
1065                 return buf;
1066 
1067         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1068                                              level, first_key);
1069         if (ret) {
1070                 free_extent_buffer_stale(buf);
1071                 return ERR_PTR(ret);
1072         }
1073         return buf;
1074 
1075 }
1076 
1077 void btrfs_clean_tree_block(struct extent_buffer *buf)
1078 {
1079         struct btrfs_fs_info *fs_info = buf->fs_info;
1080         if (btrfs_header_generation(buf) ==
1081             fs_info->running_transaction->transid) {
1082                 btrfs_assert_tree_locked(buf);
1083 
1084                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1085                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1086                                                  -buf->len,
1087                                                  fs_info->dirty_metadata_batch);
1088                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1089                         btrfs_set_lock_blocking_write(buf);
1090                         clear_extent_buffer_dirty(buf);
1091                 }
1092         }
1093 }
1094 
1095 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1096 {
1097         struct btrfs_subvolume_writers *writers;
1098         int ret;
1099 
1100         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1101         if (!writers)
1102                 return ERR_PTR(-ENOMEM);
1103 
1104         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1105         if (ret < 0) {
1106                 kfree(writers);
1107                 return ERR_PTR(ret);
1108         }
1109 
1110         init_waitqueue_head(&writers->wait);
1111         return writers;
1112 }
1113 
1114 static void
1115 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1116 {
1117         percpu_counter_destroy(&writers->counter);
1118         kfree(writers);
1119 }
1120 
1121 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1122                          u64 objectid)
1123 {
1124         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1125         root->node = NULL;
1126         root->commit_root = NULL;
1127         root->state = 0;
1128         root->orphan_cleanup_state = 0;
1129 
1130         root->last_trans = 0;
1131         root->highest_objectid = 0;
1132         root->nr_delalloc_inodes = 0;
1133         root->nr_ordered_extents = 0;
1134         root->inode_tree = RB_ROOT;
1135         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1136         root->block_rsv = NULL;
1137 
1138         INIT_LIST_HEAD(&root->dirty_list);
1139         INIT_LIST_HEAD(&root->root_list);
1140         INIT_LIST_HEAD(&root->delalloc_inodes);
1141         INIT_LIST_HEAD(&root->delalloc_root);
1142         INIT_LIST_HEAD(&root->ordered_extents);
1143         INIT_LIST_HEAD(&root->ordered_root);
1144         INIT_LIST_HEAD(&root->reloc_dirty_list);
1145         INIT_LIST_HEAD(&root->logged_list[0]);
1146         INIT_LIST_HEAD(&root->logged_list[1]);
1147         spin_lock_init(&root->inode_lock);
1148         spin_lock_init(&root->delalloc_lock);
1149         spin_lock_init(&root->ordered_extent_lock);
1150         spin_lock_init(&root->accounting_lock);
1151         spin_lock_init(&root->log_extents_lock[0]);
1152         spin_lock_init(&root->log_extents_lock[1]);
1153         spin_lock_init(&root->qgroup_meta_rsv_lock);
1154         mutex_init(&root->objectid_mutex);
1155         mutex_init(&root->log_mutex);
1156         mutex_init(&root->ordered_extent_mutex);
1157         mutex_init(&root->delalloc_mutex);
1158         init_waitqueue_head(&root->log_writer_wait);
1159         init_waitqueue_head(&root->log_commit_wait[0]);
1160         init_waitqueue_head(&root->log_commit_wait[1]);
1161         INIT_LIST_HEAD(&root->log_ctxs[0]);
1162         INIT_LIST_HEAD(&root->log_ctxs[1]);
1163         atomic_set(&root->log_commit[0], 0);
1164         atomic_set(&root->log_commit[1], 0);
1165         atomic_set(&root->log_writers, 0);
1166         atomic_set(&root->log_batch, 0);
1167         refcount_set(&root->refs, 1);
1168         atomic_set(&root->will_be_snapshotted, 0);
1169         atomic_set(&root->snapshot_force_cow, 0);
1170         atomic_set(&root->nr_swapfiles, 0);
1171         root->log_transid = 0;
1172         root->log_transid_committed = -1;
1173         root->last_log_commit = 0;
1174         if (!dummy)
1175                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1176                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1177 
1178         memset(&root->root_key, 0, sizeof(root->root_key));
1179         memset(&root->root_item, 0, sizeof(root->root_item));
1180         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1181         if (!dummy)
1182                 root->defrag_trans_start = fs_info->generation;
1183         else
1184                 root->defrag_trans_start = 0;
1185         root->root_key.objectid = objectid;
1186         root->anon_dev = 0;
1187 
1188         spin_lock_init(&root->root_item_lock);
1189         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1190 }
1191 
1192 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1193                 gfp_t flags)
1194 {
1195         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1196         if (root)
1197                 root->fs_info = fs_info;
1198         return root;
1199 }
1200 
1201 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1202 /* Should only be used by the testing infrastructure */
1203 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1204 {
1205         struct btrfs_root *root;
1206 
1207         if (!fs_info)
1208                 return ERR_PTR(-EINVAL);
1209 
1210         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1211         if (!root)
1212                 return ERR_PTR(-ENOMEM);
1213 
1214         /* We don't use the stripesize in selftest, set it as sectorsize */
1215         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1216         root->alloc_bytenr = 0;
1217 
1218         return root;
1219 }
1220 #endif
1221 
1222 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1223                                      u64 objectid)
1224 {
1225         struct btrfs_fs_info *fs_info = trans->fs_info;
1226         struct extent_buffer *leaf;
1227         struct btrfs_root *tree_root = fs_info->tree_root;
1228         struct btrfs_root *root;
1229         struct btrfs_key key;
1230         unsigned int nofs_flag;
1231         int ret = 0;
1232         uuid_le uuid = NULL_UUID_LE;
1233 
1234         /*
1235          * We're holding a transaction handle, so use a NOFS memory allocation
1236          * context to avoid deadlock if reclaim happens.
1237          */
1238         nofs_flag = memalloc_nofs_save();
1239         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1240         memalloc_nofs_restore(nofs_flag);
1241         if (!root)
1242                 return ERR_PTR(-ENOMEM);
1243 
1244         __setup_root(root, fs_info, objectid);
1245         root->root_key.objectid = objectid;
1246         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1247         root->root_key.offset = 0;
1248 
1249         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1250         if (IS_ERR(leaf)) {
1251                 ret = PTR_ERR(leaf);
1252                 leaf = NULL;
1253                 goto fail;
1254         }
1255 
1256         root->node = leaf;
1257         btrfs_mark_buffer_dirty(leaf);
1258 
1259         root->commit_root = btrfs_root_node(root);
1260         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1261 
1262         root->root_item.flags = 0;
1263         root->root_item.byte_limit = 0;
1264         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1265         btrfs_set_root_generation(&root->root_item, trans->transid);
1266         btrfs_set_root_level(&root->root_item, 0);
1267         btrfs_set_root_refs(&root->root_item, 1);
1268         btrfs_set_root_used(&root->root_item, leaf->len);
1269         btrfs_set_root_last_snapshot(&root->root_item, 0);
1270         btrfs_set_root_dirid(&root->root_item, 0);
1271         if (is_fstree(objectid))
1272                 uuid_le_gen(&uuid);
1273         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1274         root->root_item.drop_level = 0;
1275 
1276         key.objectid = objectid;
1277         key.type = BTRFS_ROOT_ITEM_KEY;
1278         key.offset = 0;
1279         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1280         if (ret)
1281                 goto fail;
1282 
1283         btrfs_tree_unlock(leaf);
1284 
1285         return root;
1286 
1287 fail:
1288         if (leaf) {
1289                 btrfs_tree_unlock(leaf);
1290                 free_extent_buffer(root->commit_root);
1291                 free_extent_buffer(leaf);
1292         }
1293         kfree(root);
1294 
1295         return ERR_PTR(ret);
1296 }
1297 
1298 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1299                                          struct btrfs_fs_info *fs_info)
1300 {
1301         struct btrfs_root *root;
1302         struct extent_buffer *leaf;
1303 
1304         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1305         if (!root)
1306                 return ERR_PTR(-ENOMEM);
1307 
1308         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1309 
1310         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1311         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1312         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1313 
1314         /*
1315          * DON'T set REF_COWS for log trees
1316          *
1317          * log trees do not get reference counted because they go away
1318          * before a real commit is actually done.  They do store pointers
1319          * to file data extents, and those reference counts still get
1320          * updated (along with back refs to the log tree).
1321          */
1322 
1323         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1324                         NULL, 0, 0, 0);
1325         if (IS_ERR(leaf)) {
1326                 kfree(root);
1327                 return ERR_CAST(leaf);
1328         }
1329 
1330         root->node = leaf;
1331 
1332         btrfs_mark_buffer_dirty(root->node);
1333         btrfs_tree_unlock(root->node);
1334         return root;
1335 }
1336 
1337 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1338                              struct btrfs_fs_info *fs_info)
1339 {
1340         struct btrfs_root *log_root;
1341 
1342         log_root = alloc_log_tree(trans, fs_info);
1343         if (IS_ERR(log_root))
1344                 return PTR_ERR(log_root);
1345         WARN_ON(fs_info->log_root_tree);
1346         fs_info->log_root_tree = log_root;
1347         return 0;
1348 }
1349 
1350 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1351                        struct btrfs_root *root)
1352 {
1353         struct btrfs_fs_info *fs_info = root->fs_info;
1354         struct btrfs_root *log_root;
1355         struct btrfs_inode_item *inode_item;
1356 
1357         log_root = alloc_log_tree(trans, fs_info);
1358         if (IS_ERR(log_root))
1359                 return PTR_ERR(log_root);
1360 
1361         log_root->last_trans = trans->transid;
1362         log_root->root_key.offset = root->root_key.objectid;
1363 
1364         inode_item = &log_root->root_item.inode;
1365         btrfs_set_stack_inode_generation(inode_item, 1);
1366         btrfs_set_stack_inode_size(inode_item, 3);
1367         btrfs_set_stack_inode_nlink(inode_item, 1);
1368         btrfs_set_stack_inode_nbytes(inode_item,
1369                                      fs_info->nodesize);
1370         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1371 
1372         btrfs_set_root_node(&log_root->root_item, log_root->node);
1373 
1374         WARN_ON(root->log_root);
1375         root->log_root = log_root;
1376         root->log_transid = 0;
1377         root->log_transid_committed = -1;
1378         root->last_log_commit = 0;
1379         return 0;
1380 }
1381 
1382 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1383                                                struct btrfs_key *key)
1384 {
1385         struct btrfs_root *root;
1386         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1387         struct btrfs_path *path;
1388         u64 generation;
1389         int ret;
1390         int level;
1391 
1392         path = btrfs_alloc_path();
1393         if (!path)
1394                 return ERR_PTR(-ENOMEM);
1395 
1396         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1397         if (!root) {
1398                 ret = -ENOMEM;
1399                 goto alloc_fail;
1400         }
1401 
1402         __setup_root(root, fs_info, key->objectid);
1403 
1404         ret = btrfs_find_root(tree_root, key, path,
1405                               &root->root_item, &root->root_key);
1406         if (ret) {
1407                 if (ret > 0)
1408                         ret = -ENOENT;
1409                 goto find_fail;
1410         }
1411 
1412         generation = btrfs_root_generation(&root->root_item);
1413         level = btrfs_root_level(&root->root_item);
1414         root->node = read_tree_block(fs_info,
1415                                      btrfs_root_bytenr(&root->root_item),
1416                                      generation, level, NULL);
1417         if (IS_ERR(root->node)) {
1418                 ret = PTR_ERR(root->node);
1419                 goto find_fail;
1420         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1421                 ret = -EIO;
1422                 free_extent_buffer(root->node);
1423                 goto find_fail;
1424         }
1425         root->commit_root = btrfs_root_node(root);
1426 out:
1427         btrfs_free_path(path);
1428         return root;
1429 
1430 find_fail:
1431         kfree(root);
1432 alloc_fail:
1433         root = ERR_PTR(ret);
1434         goto out;
1435 }
1436 
1437 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1438                                       struct btrfs_key *location)
1439 {
1440         struct btrfs_root *root;
1441 
1442         root = btrfs_read_tree_root(tree_root, location);
1443         if (IS_ERR(root))
1444                 return root;
1445 
1446         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1447                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1448                 btrfs_check_and_init_root_item(&root->root_item);
1449         }
1450 
1451         return root;
1452 }
1453 
1454 int btrfs_init_fs_root(struct btrfs_root *root)
1455 {
1456         int ret;
1457         struct btrfs_subvolume_writers *writers;
1458 
1459         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1460         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1461                                         GFP_NOFS);
1462         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1463                 ret = -ENOMEM;
1464                 goto fail;
1465         }
1466 
1467         writers = btrfs_alloc_subvolume_writers();
1468         if (IS_ERR(writers)) {
1469                 ret = PTR_ERR(writers);
1470                 goto fail;
1471         }
1472         root->subv_writers = writers;
1473 
1474         btrfs_init_free_ino_ctl(root);
1475         spin_lock_init(&root->ino_cache_lock);
1476         init_waitqueue_head(&root->ino_cache_wait);
1477 
1478         ret = get_anon_bdev(&root->anon_dev);
1479         if (ret)
1480                 goto fail;
1481 
1482         mutex_lock(&root->objectid_mutex);
1483         ret = btrfs_find_highest_objectid(root,
1484                                         &root->highest_objectid);
1485         if (ret) {
1486                 mutex_unlock(&root->objectid_mutex);
1487                 goto fail;
1488         }
1489 
1490         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1491 
1492         mutex_unlock(&root->objectid_mutex);
1493 
1494         return 0;
1495 fail:
1496         /* The caller is responsible to call btrfs_free_fs_root */
1497         return ret;
1498 }
1499 
1500 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1501                                         u64 root_id)
1502 {
1503         struct btrfs_root *root;
1504 
1505         spin_lock(&fs_info->fs_roots_radix_lock);
1506         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1507                                  (unsigned long)root_id);
1508         spin_unlock(&fs_info->fs_roots_radix_lock);
1509         return root;
1510 }
1511 
1512 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1513                          struct btrfs_root *root)
1514 {
1515         int ret;
1516 
1517         ret = radix_tree_preload(GFP_NOFS);
1518         if (ret)
1519                 return ret;
1520 
1521         spin_lock(&fs_info->fs_roots_radix_lock);
1522         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1523                                 (unsigned long)root->root_key.objectid,
1524                                 root);
1525         if (ret == 0)
1526                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1527         spin_unlock(&fs_info->fs_roots_radix_lock);
1528         radix_tree_preload_end();
1529 
1530         return ret;
1531 }
1532 
1533 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1534                                      struct btrfs_key *location,
1535                                      bool check_ref)
1536 {
1537         struct btrfs_root *root;
1538         struct btrfs_path *path;
1539         struct btrfs_key key;
1540         int ret;
1541 
1542         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1543                 return fs_info->tree_root;
1544         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1545                 return fs_info->extent_root;
1546         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1547                 return fs_info->chunk_root;
1548         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1549                 return fs_info->dev_root;
1550         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1551                 return fs_info->csum_root;
1552         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1553                 return fs_info->quota_root ? fs_info->quota_root :
1554                                              ERR_PTR(-ENOENT);
1555         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1556                 return fs_info->uuid_root ? fs_info->uuid_root :
1557                                             ERR_PTR(-ENOENT);
1558         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1559                 return fs_info->free_space_root ? fs_info->free_space_root :
1560                                                   ERR_PTR(-ENOENT);
1561 again:
1562         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1563         if (root) {
1564                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1565                         return ERR_PTR(-ENOENT);
1566                 return root;
1567         }
1568 
1569         root = btrfs_read_fs_root(fs_info->tree_root, location);
1570         if (IS_ERR(root))
1571                 return root;
1572 
1573         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1574                 ret = -ENOENT;
1575                 goto fail;
1576         }
1577 
1578         ret = btrfs_init_fs_root(root);
1579         if (ret)
1580                 goto fail;
1581 
1582         path = btrfs_alloc_path();
1583         if (!path) {
1584                 ret = -ENOMEM;
1585                 goto fail;
1586         }
1587         key.objectid = BTRFS_ORPHAN_OBJECTID;
1588         key.type = BTRFS_ORPHAN_ITEM_KEY;
1589         key.offset = location->objectid;
1590 
1591         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1592         btrfs_free_path(path);
1593         if (ret < 0)
1594                 goto fail;
1595         if (ret == 0)
1596                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1597 
1598         ret = btrfs_insert_fs_root(fs_info, root);
1599         if (ret) {
1600                 if (ret == -EEXIST) {
1601                         btrfs_free_fs_root(root);
1602                         goto again;
1603                 }
1604                 goto fail;
1605         }
1606         return root;
1607 fail:
1608         btrfs_free_fs_root(root);
1609         return ERR_PTR(ret);
1610 }
1611 
1612 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1613 {
1614         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1615         int ret = 0;
1616         struct btrfs_device *device;
1617         struct backing_dev_info *bdi;
1618 
1619         rcu_read_lock();
1620         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1621                 if (!device->bdev)
1622                         continue;
1623                 bdi = device->bdev->bd_bdi;
1624                 if (bdi_congested(bdi, bdi_bits)) {
1625                         ret = 1;
1626                         break;
1627                 }
1628         }
1629         rcu_read_unlock();
1630         return ret;
1631 }
1632 
1633 /*
1634  * called by the kthread helper functions to finally call the bio end_io
1635  * functions.  This is where read checksum verification actually happens
1636  */
1637 static void end_workqueue_fn(struct btrfs_work *work)
1638 {
1639         struct bio *bio;
1640         struct btrfs_end_io_wq *end_io_wq;
1641 
1642         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1643         bio = end_io_wq->bio;
1644 
1645         bio->bi_status = end_io_wq->status;
1646         bio->bi_private = end_io_wq->private;
1647         bio->bi_end_io = end_io_wq->end_io;
1648         bio_endio(bio);
1649         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1650 }
1651 
1652 static int cleaner_kthread(void *arg)
1653 {
1654         struct btrfs_root *root = arg;
1655         struct btrfs_fs_info *fs_info = root->fs_info;
1656         int again;
1657 
1658         while (1) {
1659                 again = 0;
1660 
1661                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1662 
1663                 /* Make the cleaner go to sleep early. */
1664                 if (btrfs_need_cleaner_sleep(fs_info))
1665                         goto sleep;
1666 
1667                 /*
1668                  * Do not do anything if we might cause open_ctree() to block
1669                  * before we have finished mounting the filesystem.
1670                  */
1671                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1672                         goto sleep;
1673 
1674                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1675                         goto sleep;
1676 
1677                 /*
1678                  * Avoid the problem that we change the status of the fs
1679                  * during the above check and trylock.
1680                  */
1681                 if (btrfs_need_cleaner_sleep(fs_info)) {
1682                         mutex_unlock(&fs_info->cleaner_mutex);
1683                         goto sleep;
1684                 }
1685 
1686                 btrfs_run_delayed_iputs(fs_info);
1687 
1688                 again = btrfs_clean_one_deleted_snapshot(root);
1689                 mutex_unlock(&fs_info->cleaner_mutex);
1690 
1691                 /*
1692                  * The defragger has dealt with the R/O remount and umount,
1693                  * needn't do anything special here.
1694                  */
1695                 btrfs_run_defrag_inodes(fs_info);
1696 
1697                 /*
1698                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1699                  * with relocation (btrfs_relocate_chunk) and relocation
1700                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1701                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1702                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1703                  * unused block groups.
1704                  */
1705                 btrfs_delete_unused_bgs(fs_info);
1706 sleep:
1707                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1708                 if (kthread_should_park())
1709                         kthread_parkme();
1710                 if (kthread_should_stop())
1711                         return 0;
1712                 if (!again) {
1713                         set_current_state(TASK_INTERRUPTIBLE);
1714                         schedule();
1715                         __set_current_state(TASK_RUNNING);
1716                 }
1717         }
1718 }
1719 
1720 static int transaction_kthread(void *arg)
1721 {
1722         struct btrfs_root *root = arg;
1723         struct btrfs_fs_info *fs_info = root->fs_info;
1724         struct btrfs_trans_handle *trans;
1725         struct btrfs_transaction *cur;
1726         u64 transid;
1727         time64_t now;
1728         unsigned long delay;
1729         bool cannot_commit;
1730 
1731         do {
1732                 cannot_commit = false;
1733                 delay = HZ * fs_info->commit_interval;
1734                 mutex_lock(&fs_info->transaction_kthread_mutex);
1735 
1736                 spin_lock(&fs_info->trans_lock);
1737                 cur = fs_info->running_transaction;
1738                 if (!cur) {
1739                         spin_unlock(&fs_info->trans_lock);
1740                         goto sleep;
1741                 }
1742 
1743                 now = ktime_get_seconds();
1744                 if (cur->state < TRANS_STATE_BLOCKED &&
1745                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1746                     (now < cur->start_time ||
1747                      now - cur->start_time < fs_info->commit_interval)) {
1748                         spin_unlock(&fs_info->trans_lock);
1749                         delay = HZ * 5;
1750                         goto sleep;
1751                 }
1752                 transid = cur->transid;
1753                 spin_unlock(&fs_info->trans_lock);
1754 
1755                 /* If the file system is aborted, this will always fail. */
1756                 trans = btrfs_attach_transaction(root);
1757                 if (IS_ERR(trans)) {
1758                         if (PTR_ERR(trans) != -ENOENT)
1759                                 cannot_commit = true;
1760                         goto sleep;
1761                 }
1762                 if (transid == trans->transid) {
1763                         btrfs_commit_transaction(trans);
1764                 } else {
1765                         btrfs_end_transaction(trans);
1766                 }
1767 sleep:
1768                 wake_up_process(fs_info->cleaner_kthread);
1769                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1770 
1771                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1772                                       &fs_info->fs_state)))
1773                         btrfs_cleanup_transaction(fs_info);
1774                 if (!kthread_should_stop() &&
1775                                 (!btrfs_transaction_blocked(fs_info) ||
1776                                  cannot_commit))
1777                         schedule_timeout_interruptible(delay);
1778         } while (!kthread_should_stop());
1779         return 0;
1780 }
1781 
1782 /*
1783  * this will find the highest generation in the array of
1784  * root backups.  The index of the highest array is returned,
1785  * or -1 if we can't find anything.
1786  *
1787  * We check to make sure the array is valid by comparing the
1788  * generation of the latest  root in the array with the generation
1789  * in the super block.  If they don't match we pitch it.
1790  */
1791 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1792 {
1793         u64 cur;
1794         int newest_index = -1;
1795         struct btrfs_root_backup *root_backup;
1796         int i;
1797 
1798         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1799                 root_backup = info->super_copy->super_roots + i;
1800                 cur = btrfs_backup_tree_root_gen(root_backup);
1801                 if (cur == newest_gen)
1802                         newest_index = i;
1803         }
1804 
1805         /* check to see if we actually wrapped around */
1806         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1807                 root_backup = info->super_copy->super_roots;
1808                 cur = btrfs_backup_tree_root_gen(root_backup);
1809                 if (cur == newest_gen)
1810                         newest_index = 0;
1811         }
1812         return newest_index;
1813 }
1814 
1815 
1816 /*
1817  * find the oldest backup so we know where to store new entries
1818  * in the backup array.  This will set the backup_root_index
1819  * field in the fs_info struct
1820  */
1821 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1822                                      u64 newest_gen)
1823 {
1824         int newest_index = -1;
1825 
1826         newest_index = find_newest_super_backup(info, newest_gen);
1827         /* if there was garbage in there, just move along */
1828         if (newest_index == -1) {
1829                 info->backup_root_index = 0;
1830         } else {
1831                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1832         }
1833 }
1834 
1835 /*
1836  * copy all the root pointers into the super backup array.
1837  * this will bump the backup pointer by one when it is
1838  * done
1839  */
1840 static void backup_super_roots(struct btrfs_fs_info *info)
1841 {
1842         int next_backup;
1843         struct btrfs_root_backup *root_backup;
1844         int last_backup;
1845 
1846         next_backup = info->backup_root_index;
1847         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1848                 BTRFS_NUM_BACKUP_ROOTS;
1849 
1850         /*
1851          * just overwrite the last backup if we're at the same generation
1852          * this happens only at umount
1853          */
1854         root_backup = info->super_for_commit->super_roots + last_backup;
1855         if (btrfs_backup_tree_root_gen(root_backup) ==
1856             btrfs_header_generation(info->tree_root->node))
1857                 next_backup = last_backup;
1858 
1859         root_backup = info->super_for_commit->super_roots + next_backup;
1860 
1861         /*
1862          * make sure all of our padding and empty slots get zero filled
1863          * regardless of which ones we use today
1864          */
1865         memset(root_backup, 0, sizeof(*root_backup));
1866 
1867         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1868 
1869         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1870         btrfs_set_backup_tree_root_gen(root_backup,
1871                                btrfs_header_generation(info->tree_root->node));
1872 
1873         btrfs_set_backup_tree_root_level(root_backup,
1874                                btrfs_header_level(info->tree_root->node));
1875 
1876         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1877         btrfs_set_backup_chunk_root_gen(root_backup,
1878                                btrfs_header_generation(info->chunk_root->node));
1879         btrfs_set_backup_chunk_root_level(root_backup,
1880                                btrfs_header_level(info->chunk_root->node));
1881 
1882         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1883         btrfs_set_backup_extent_root_gen(root_backup,
1884                                btrfs_header_generation(info->extent_root->node));
1885         btrfs_set_backup_extent_root_level(root_backup,
1886                                btrfs_header_level(info->extent_root->node));
1887 
1888         /*
1889          * we might commit during log recovery, which happens before we set
1890          * the fs_root.  Make sure it is valid before we fill it in.
1891          */
1892         if (info->fs_root && info->fs_root->node) {
1893                 btrfs_set_backup_fs_root(root_backup,
1894                                          info->fs_root->node->start);
1895                 btrfs_set_backup_fs_root_gen(root_backup,
1896                                btrfs_header_generation(info->fs_root->node));
1897                 btrfs_set_backup_fs_root_level(root_backup,
1898                                btrfs_header_level(info->fs_root->node));
1899         }
1900 
1901         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1902         btrfs_set_backup_dev_root_gen(root_backup,
1903                                btrfs_header_generation(info->dev_root->node));
1904         btrfs_set_backup_dev_root_level(root_backup,
1905                                        btrfs_header_level(info->dev_root->node));
1906 
1907         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1908         btrfs_set_backup_csum_root_gen(root_backup,
1909                                btrfs_header_generation(info->csum_root->node));
1910         btrfs_set_backup_csum_root_level(root_backup,
1911                                btrfs_header_level(info->csum_root->node));
1912 
1913         btrfs_set_backup_total_bytes(root_backup,
1914                              btrfs_super_total_bytes(info->super_copy));
1915         btrfs_set_backup_bytes_used(root_backup,
1916                              btrfs_super_bytes_used(info->super_copy));
1917         btrfs_set_backup_num_devices(root_backup,
1918                              btrfs_super_num_devices(info->super_copy));
1919 
1920         /*
1921          * if we don't copy this out to the super_copy, it won't get remembered
1922          * for the next commit
1923          */
1924         memcpy(&info->super_copy->super_roots,
1925                &info->super_for_commit->super_roots,
1926                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1927 }
1928 
1929 /*
1930  * this copies info out of the root backup array and back into
1931  * the in-memory super block.  It is meant to help iterate through
1932  * the array, so you send it the number of backups you've already
1933  * tried and the last backup index you used.
1934  *
1935  * this returns -1 when it has tried all the backups
1936  */
1937 static noinline int next_root_backup(struct btrfs_fs_info *info,
1938                                      struct btrfs_super_block *super,
1939                                      int *num_backups_tried, int *backup_index)
1940 {
1941         struct btrfs_root_backup *root_backup;
1942         int newest = *backup_index;
1943 
1944         if (*num_backups_tried == 0) {
1945                 u64 gen = btrfs_super_generation(super);
1946 
1947                 newest = find_newest_super_backup(info, gen);
1948                 if (newest == -1)
1949                         return -1;
1950 
1951                 *backup_index = newest;
1952                 *num_backups_tried = 1;
1953         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1954                 /* we've tried all the backups, all done */
1955                 return -1;
1956         } else {
1957                 /* jump to the next oldest backup */
1958                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1959                         BTRFS_NUM_BACKUP_ROOTS;
1960                 *backup_index = newest;
1961                 *num_backups_tried += 1;
1962         }
1963         root_backup = super->super_roots + newest;
1964 
1965         btrfs_set_super_generation(super,
1966                                    btrfs_backup_tree_root_gen(root_backup));
1967         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1968         btrfs_set_super_root_level(super,
1969                                    btrfs_backup_tree_root_level(root_backup));
1970         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1971 
1972         /*
1973          * fixme: the total bytes and num_devices need to match or we should
1974          * need a fsck
1975          */
1976         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1977         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1978         return 0;
1979 }
1980 
1981 /* helper to cleanup workers */
1982 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1983 {
1984         btrfs_destroy_workqueue(fs_info->fixup_workers);
1985         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1986         btrfs_destroy_workqueue(fs_info->workers);
1987         btrfs_destroy_workqueue(fs_info->endio_workers);
1988         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1989         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1990         btrfs_destroy_workqueue(fs_info->rmw_workers);
1991         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1992         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1993         btrfs_destroy_workqueue(fs_info->submit_workers);
1994         btrfs_destroy_workqueue(fs_info->delayed_workers);
1995         btrfs_destroy_workqueue(fs_info->caching_workers);
1996         btrfs_destroy_workqueue(fs_info->readahead_workers);
1997         btrfs_destroy_workqueue(fs_info->flush_workers);
1998         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1999         /*
2000          * Now that all other work queues are destroyed, we can safely destroy
2001          * the queues used for metadata I/O, since tasks from those other work
2002          * queues can do metadata I/O operations.
2003          */
2004         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2005         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2006 }
2007 
2008 static void free_root_extent_buffers(struct btrfs_root *root)
2009 {
2010         if (root) {
2011                 free_extent_buffer(root->node);
2012                 free_extent_buffer(root->commit_root);
2013                 root->node = NULL;
2014                 root->commit_root = NULL;
2015         }
2016 }
2017 
2018 /* helper to cleanup tree roots */
2019 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2020 {
2021         free_root_extent_buffers(info->tree_root);
2022 
2023         free_root_extent_buffers(info->dev_root);
2024         free_root_extent_buffers(info->extent_root);
2025         free_root_extent_buffers(info->csum_root);
2026         free_root_extent_buffers(info->quota_root);
2027         free_root_extent_buffers(info->uuid_root);
2028         if (free_chunk_root)
2029                 free_root_extent_buffers(info->chunk_root);
2030         free_root_extent_buffers(info->free_space_root);
2031 }
2032 
2033 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2034 {
2035         int ret;
2036         struct btrfs_root *gang[8];
2037         int i;
2038 
2039         while (!list_empty(&fs_info->dead_roots)) {
2040                 gang[0] = list_entry(fs_info->dead_roots.next,
2041                                      struct btrfs_root, root_list);
2042                 list_del(&gang[0]->root_list);
2043 
2044                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2045                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2046                 } else {
2047                         free_extent_buffer(gang[0]->node);
2048                         free_extent_buffer(gang[0]->commit_root);
2049                         btrfs_put_fs_root(gang[0]);
2050                 }
2051         }
2052 
2053         while (1) {
2054                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2055                                              (void **)gang, 0,
2056                                              ARRAY_SIZE(gang));
2057                 if (!ret)
2058                         break;
2059                 for (i = 0; i < ret; i++)
2060                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2061         }
2062 
2063         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2064                 btrfs_free_log_root_tree(NULL, fs_info);
2065                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2066         }
2067 }
2068 
2069 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2070 {
2071         mutex_init(&fs_info->scrub_lock);
2072         atomic_set(&fs_info->scrubs_running, 0);
2073         atomic_set(&fs_info->scrub_pause_req, 0);
2074         atomic_set(&fs_info->scrubs_paused, 0);
2075         atomic_set(&fs_info->scrub_cancel_req, 0);
2076         init_waitqueue_head(&fs_info->scrub_pause_wait);
2077         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2078 }
2079 
2080 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2081 {
2082         spin_lock_init(&fs_info->balance_lock);
2083         mutex_init(&fs_info->balance_mutex);
2084         atomic_set(&fs_info->balance_pause_req, 0);
2085         atomic_set(&fs_info->balance_cancel_req, 0);
2086         fs_info->balance_ctl = NULL;
2087         init_waitqueue_head(&fs_info->balance_wait_q);
2088 }
2089 
2090 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2091 {
2092         struct inode *inode = fs_info->btree_inode;
2093 
2094         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2095         set_nlink(inode, 1);
2096         /*
2097          * we set the i_size on the btree inode to the max possible int.
2098          * the real end of the address space is determined by all of
2099          * the devices in the system
2100          */
2101         inode->i_size = OFFSET_MAX;
2102         inode->i_mapping->a_ops = &btree_aops;
2103 
2104         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2105         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2106                             IO_TREE_INODE_IO, inode);
2107         BTRFS_I(inode)->io_tree.track_uptodate = false;
2108         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2109 
2110         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2111 
2112         BTRFS_I(inode)->root = fs_info->tree_root;
2113         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2114         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2115         btrfs_insert_inode_hash(inode);
2116 }
2117 
2118 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2119 {
2120         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2121         init_rwsem(&fs_info->dev_replace.rwsem);
2122         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2123 }
2124 
2125 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2126 {
2127         spin_lock_init(&fs_info->qgroup_lock);
2128         mutex_init(&fs_info->qgroup_ioctl_lock);
2129         fs_info->qgroup_tree = RB_ROOT;
2130         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2131         fs_info->qgroup_seq = 1;
2132         fs_info->qgroup_ulist = NULL;
2133         fs_info->qgroup_rescan_running = false;
2134         mutex_init(&fs_info->qgroup_rescan_lock);
2135 }
2136 
2137 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2138                 struct btrfs_fs_devices *fs_devices)
2139 {
2140         u32 max_active = fs_info->thread_pool_size;
2141         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2142 
2143         fs_info->workers =
2144                 btrfs_alloc_workqueue(fs_info, "worker",
2145                                       flags | WQ_HIGHPRI, max_active, 16);
2146 
2147         fs_info->delalloc_workers =
2148                 btrfs_alloc_workqueue(fs_info, "delalloc",
2149                                       flags, max_active, 2);
2150 
2151         fs_info->flush_workers =
2152                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2153                                       flags, max_active, 0);
2154 
2155         fs_info->caching_workers =
2156                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2157 
2158         /*
2159          * a higher idle thresh on the submit workers makes it much more
2160          * likely that bios will be send down in a sane order to the
2161          * devices
2162          */
2163         fs_info->submit_workers =
2164                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2165                                       min_t(u64, fs_devices->num_devices,
2166                                             max_active), 64);
2167 
2168         fs_info->fixup_workers =
2169                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2170 
2171         /*
2172          * endios are largely parallel and should have a very
2173          * low idle thresh
2174          */
2175         fs_info->endio_workers =
2176                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2177         fs_info->endio_meta_workers =
2178                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2179                                       max_active, 4);
2180         fs_info->endio_meta_write_workers =
2181                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2182                                       max_active, 2);
2183         fs_info->endio_raid56_workers =
2184                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2185                                       max_active, 4);
2186         fs_info->endio_repair_workers =
2187                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2188         fs_info->rmw_workers =
2189                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2190         fs_info->endio_write_workers =
2191                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2192                                       max_active, 2);
2193         fs_info->endio_freespace_worker =
2194                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2195                                       max_active, 0);
2196         fs_info->delayed_workers =
2197                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2198                                       max_active, 0);
2199         fs_info->readahead_workers =
2200                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2201                                       max_active, 2);
2202         fs_info->qgroup_rescan_workers =
2203                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2204 
2205         if (!(fs_info->workers && fs_info->delalloc_workers &&
2206               fs_info->submit_workers && fs_info->flush_workers &&
2207               fs_info->endio_workers && fs_info->endio_meta_workers &&
2208               fs_info->endio_meta_write_workers &&
2209               fs_info->endio_repair_workers &&
2210               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2211               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2212               fs_info->caching_workers && fs_info->readahead_workers &&
2213               fs_info->fixup_workers && fs_info->delayed_workers &&
2214               fs_info->qgroup_rescan_workers)) {
2215                 return -ENOMEM;
2216         }
2217 
2218         return 0;
2219 }
2220 
2221 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2222 {
2223         struct crypto_shash *csum_shash;
2224         const char *csum_name = btrfs_super_csum_name(csum_type);
2225 
2226         csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2227 
2228         if (IS_ERR(csum_shash)) {
2229                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2230                           csum_name);
2231                 return PTR_ERR(csum_shash);
2232         }
2233 
2234         fs_info->csum_shash = csum_shash;
2235 
2236         return 0;
2237 }
2238 
2239 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2240 {
2241         crypto_free_shash(fs_info->csum_shash);
2242 }
2243 
2244 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2245                             struct btrfs_fs_devices *fs_devices)
2246 {
2247         int ret;
2248         struct btrfs_root *log_tree_root;
2249         struct btrfs_super_block *disk_super = fs_info->super_copy;
2250         u64 bytenr = btrfs_super_log_root(disk_super);
2251         int level = btrfs_super_log_root_level(disk_super);
2252 
2253         if (fs_devices->rw_devices == 0) {
2254                 btrfs_warn(fs_info, "log replay required on RO media");
2255                 return -EIO;
2256         }
2257 
2258         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2259         if (!log_tree_root)
2260                 return -ENOMEM;
2261 
2262         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2263 
2264         log_tree_root->node = read_tree_block(fs_info, bytenr,
2265                                               fs_info->generation + 1,
2266                                               level, NULL);
2267         if (IS_ERR(log_tree_root->node)) {
2268                 btrfs_warn(fs_info, "failed to read log tree");
2269                 ret = PTR_ERR(log_tree_root->node);
2270                 kfree(log_tree_root);
2271                 return ret;
2272         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2273                 btrfs_err(fs_info, "failed to read log tree");
2274                 free_extent_buffer(log_tree_root->node);
2275                 kfree(log_tree_root);
2276                 return -EIO;
2277         }
2278         /* returns with log_tree_root freed on success */
2279         ret = btrfs_recover_log_trees(log_tree_root);
2280         if (ret) {
2281                 btrfs_handle_fs_error(fs_info, ret,
2282                                       "Failed to recover log tree");
2283                 free_extent_buffer(log_tree_root->node);
2284                 kfree(log_tree_root);
2285                 return ret;
2286         }
2287 
2288         if (sb_rdonly(fs_info->sb)) {
2289                 ret = btrfs_commit_super(fs_info);
2290                 if (ret)
2291                         return ret;
2292         }
2293 
2294         return 0;
2295 }
2296 
2297 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2298 {
2299         struct btrfs_root *tree_root = fs_info->tree_root;
2300         struct btrfs_root *root;
2301         struct btrfs_key location;
2302         int ret;
2303 
2304         BUG_ON(!fs_info->tree_root);
2305 
2306         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2307         location.type = BTRFS_ROOT_ITEM_KEY;
2308         location.offset = 0;
2309 
2310         root = btrfs_read_tree_root(tree_root, &location);
2311         if (IS_ERR(root)) {
2312                 ret = PTR_ERR(root);
2313                 goto out;
2314         }
2315         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2316         fs_info->extent_root = root;
2317 
2318         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2319         root = btrfs_read_tree_root(tree_root, &location);
2320         if (IS_ERR(root)) {
2321                 ret = PTR_ERR(root);
2322                 goto out;
2323         }
2324         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2325         fs_info->dev_root = root;
2326         btrfs_init_devices_late(fs_info);
2327 
2328         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2329         root = btrfs_read_tree_root(tree_root, &location);
2330         if (IS_ERR(root)) {
2331                 ret = PTR_ERR(root);
2332                 goto out;
2333         }
2334         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2335         fs_info->csum_root = root;
2336 
2337         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2338         root = btrfs_read_tree_root(tree_root, &location);
2339         if (!IS_ERR(root)) {
2340                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2342                 fs_info->quota_root = root;
2343         }
2344 
2345         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2346         root = btrfs_read_tree_root(tree_root, &location);
2347         if (IS_ERR(root)) {
2348                 ret = PTR_ERR(root);
2349                 if (ret != -ENOENT)
2350                         goto out;
2351         } else {
2352                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353                 fs_info->uuid_root = root;
2354         }
2355 
2356         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2357                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2358                 root = btrfs_read_tree_root(tree_root, &location);
2359                 if (IS_ERR(root)) {
2360                         ret = PTR_ERR(root);
2361                         goto out;
2362                 }
2363                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2364                 fs_info->free_space_root = root;
2365         }
2366 
2367         return 0;
2368 out:
2369         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2370                    location.objectid, ret);
2371         return ret;
2372 }
2373 
2374 /*
2375  * Real super block validation
2376  * NOTE: super csum type and incompat features will not be checked here.
2377  *
2378  * @sb:         super block to check
2379  * @mirror_num: the super block number to check its bytenr:
2380  *              0       the primary (1st) sb
2381  *              1, 2    2nd and 3rd backup copy
2382  *             -1       skip bytenr check
2383  */
2384 static int validate_super(struct btrfs_fs_info *fs_info,
2385                             struct btrfs_super_block *sb, int mirror_num)
2386 {
2387         u64 nodesize = btrfs_super_nodesize(sb);
2388         u64 sectorsize = btrfs_super_sectorsize(sb);
2389         int ret = 0;
2390 
2391         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2392                 btrfs_err(fs_info, "no valid FS found");
2393                 ret = -EINVAL;
2394         }
2395         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2396                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2397                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2398                 ret = -EINVAL;
2399         }
2400         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2401                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2402                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2403                 ret = -EINVAL;
2404         }
2405         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2406                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2407                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2408                 ret = -EINVAL;
2409         }
2410         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2411                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2412                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2413                 ret = -EINVAL;
2414         }
2415 
2416         /*
2417          * Check sectorsize and nodesize first, other check will need it.
2418          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2419          */
2420         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2421             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2422                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2423                 ret = -EINVAL;
2424         }
2425         /* Only PAGE SIZE is supported yet */
2426         if (sectorsize != PAGE_SIZE) {
2427                 btrfs_err(fs_info,
2428                         "sectorsize %llu not supported yet, only support %lu",
2429                         sectorsize, PAGE_SIZE);
2430                 ret = -EINVAL;
2431         }
2432         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2433             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2434                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2435                 ret = -EINVAL;
2436         }
2437         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2438                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2439                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2440                 ret = -EINVAL;
2441         }
2442 
2443         /* Root alignment check */
2444         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2445                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2446                            btrfs_super_root(sb));
2447                 ret = -EINVAL;
2448         }
2449         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2450                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2451                            btrfs_super_chunk_root(sb));
2452                 ret = -EINVAL;
2453         }
2454         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2455                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2456                            btrfs_super_log_root(sb));
2457                 ret = -EINVAL;
2458         }
2459 
2460         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2461                    BTRFS_FSID_SIZE) != 0) {
2462                 btrfs_err(fs_info,
2463                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2464                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2465                 ret = -EINVAL;
2466         }
2467 
2468         /*
2469          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2470          * done later
2471          */
2472         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2473                 btrfs_err(fs_info, "bytes_used is too small %llu",
2474                           btrfs_super_bytes_used(sb));
2475                 ret = -EINVAL;
2476         }
2477         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2478                 btrfs_err(fs_info, "invalid stripesize %u",
2479                           btrfs_super_stripesize(sb));
2480                 ret = -EINVAL;
2481         }
2482         if (btrfs_super_num_devices(sb) > (1UL << 31))
2483                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2484                            btrfs_super_num_devices(sb));
2485         if (btrfs_super_num_devices(sb) == 0) {
2486                 btrfs_err(fs_info, "number of devices is 0");
2487                 ret = -EINVAL;
2488         }
2489 
2490         if (mirror_num >= 0 &&
2491             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2492                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2493                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2494                 ret = -EINVAL;
2495         }
2496 
2497         /*
2498          * Obvious sys_chunk_array corruptions, it must hold at least one key
2499          * and one chunk
2500          */
2501         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2502                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2503                           btrfs_super_sys_array_size(sb),
2504                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2505                 ret = -EINVAL;
2506         }
2507         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2508                         + sizeof(struct btrfs_chunk)) {
2509                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2510                           btrfs_super_sys_array_size(sb),
2511                           sizeof(struct btrfs_disk_key)
2512                           + sizeof(struct btrfs_chunk));
2513                 ret = -EINVAL;
2514         }
2515 
2516         /*
2517          * The generation is a global counter, we'll trust it more than the others
2518          * but it's still possible that it's the one that's wrong.
2519          */
2520         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2521                 btrfs_warn(fs_info,
2522                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2523                         btrfs_super_generation(sb),
2524                         btrfs_super_chunk_root_generation(sb));
2525         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2526             && btrfs_super_cache_generation(sb) != (u64)-1)
2527                 btrfs_warn(fs_info,
2528                         "suspicious: generation < cache_generation: %llu < %llu",
2529                         btrfs_super_generation(sb),
2530                         btrfs_super_cache_generation(sb));
2531 
2532         return ret;
2533 }
2534 
2535 /*
2536  * Validation of super block at mount time.
2537  * Some checks already done early at mount time, like csum type and incompat
2538  * flags will be skipped.
2539  */
2540 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2541 {
2542         return validate_super(fs_info, fs_info->super_copy, 0);
2543 }
2544 
2545 /*
2546  * Validation of super block at write time.
2547  * Some checks like bytenr check will be skipped as their values will be
2548  * overwritten soon.
2549  * Extra checks like csum type and incompat flags will be done here.
2550  */
2551 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2552                                       struct btrfs_super_block *sb)
2553 {
2554         int ret;
2555 
2556         ret = validate_super(fs_info, sb, -1);
2557         if (ret < 0)
2558                 goto out;
2559         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2560                 ret = -EUCLEAN;
2561                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2562                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2563                 goto out;
2564         }
2565         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2566                 ret = -EUCLEAN;
2567                 btrfs_err(fs_info,
2568                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2569                           btrfs_super_incompat_flags(sb),
2570                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2571                 goto out;
2572         }
2573 out:
2574         if (ret < 0)
2575                 btrfs_err(fs_info,
2576                 "super block corruption detected before writing it to disk");
2577         return ret;
2578 }
2579 
2580 int open_ctree(struct super_block *sb,
2581                struct btrfs_fs_devices *fs_devices,
2582                char *options)
2583 {
2584         u32 sectorsize;
2585         u32 nodesize;
2586         u32 stripesize;
2587         u64 generation;
2588         u64 features;
2589         u16 csum_type;
2590         struct btrfs_key location;
2591         struct buffer_head *bh;
2592         struct btrfs_super_block *disk_super;
2593         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2594         struct btrfs_root *tree_root;
2595         struct btrfs_root *chunk_root;
2596         int ret;
2597         int err = -EINVAL;
2598         int num_backups_tried = 0;
2599         int backup_index = 0;
2600         int clear_free_space_tree = 0;
2601         int level;
2602 
2603         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2604         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2605         if (!tree_root || !chunk_root) {
2606                 err = -ENOMEM;
2607                 goto fail;
2608         }
2609 
2610         ret = init_srcu_struct(&fs_info->subvol_srcu);
2611         if (ret) {
2612                 err = ret;
2613                 goto fail;
2614         }
2615 
2616         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2617         if (ret) {
2618                 err = ret;
2619                 goto fail_srcu;
2620         }
2621 
2622         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2623         if (ret) {
2624                 err = ret;
2625                 goto fail_dio_bytes;
2626         }
2627         fs_info->dirty_metadata_batch = PAGE_SIZE *
2628                                         (1 + ilog2(nr_cpu_ids));
2629 
2630         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2631         if (ret) {
2632                 err = ret;
2633                 goto fail_dirty_metadata_bytes;
2634         }
2635 
2636         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2637                         GFP_KERNEL);
2638         if (ret) {
2639                 err = ret;
2640                 goto fail_delalloc_bytes;
2641         }
2642 
2643         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2644         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2645         INIT_LIST_HEAD(&fs_info->trans_list);
2646         INIT_LIST_HEAD(&fs_info->dead_roots);
2647         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2648         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2649         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2650         spin_lock_init(&fs_info->delalloc_root_lock);
2651         spin_lock_init(&fs_info->trans_lock);
2652         spin_lock_init(&fs_info->fs_roots_radix_lock);
2653         spin_lock_init(&fs_info->delayed_iput_lock);
2654         spin_lock_init(&fs_info->defrag_inodes_lock);
2655         spin_lock_init(&fs_info->super_lock);
2656         spin_lock_init(&fs_info->buffer_lock);
2657         spin_lock_init(&fs_info->unused_bgs_lock);
2658         rwlock_init(&fs_info->tree_mod_log_lock);
2659         mutex_init(&fs_info->unused_bg_unpin_mutex);
2660         mutex_init(&fs_info->delete_unused_bgs_mutex);
2661         mutex_init(&fs_info->reloc_mutex);
2662         mutex_init(&fs_info->delalloc_root_mutex);
2663         seqlock_init(&fs_info->profiles_lock);
2664 
2665         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2666         INIT_LIST_HEAD(&fs_info->space_info);
2667         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2668         INIT_LIST_HEAD(&fs_info->unused_bgs);
2669         extent_map_tree_init(&fs_info->mapping_tree);
2670         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2671                              BTRFS_BLOCK_RSV_GLOBAL);
2672         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2673         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2674         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2675         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2676                              BTRFS_BLOCK_RSV_DELOPS);
2677         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2678                              BTRFS_BLOCK_RSV_DELREFS);
2679 
2680         atomic_set(&fs_info->async_delalloc_pages, 0);
2681         atomic_set(&fs_info->defrag_running, 0);
2682         atomic_set(&fs_info->reada_works_cnt, 0);
2683         atomic_set(&fs_info->nr_delayed_iputs, 0);
2684         atomic64_set(&fs_info->tree_mod_seq, 0);
2685         fs_info->sb = sb;
2686         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2687         fs_info->metadata_ratio = 0;
2688         fs_info->defrag_inodes = RB_ROOT;
2689         atomic64_set(&fs_info->free_chunk_space, 0);
2690         fs_info->tree_mod_log = RB_ROOT;
2691         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2692         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2693         /* readahead state */
2694         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2695         spin_lock_init(&fs_info->reada_lock);
2696         btrfs_init_ref_verify(fs_info);
2697 
2698         fs_info->thread_pool_size = min_t(unsigned long,
2699                                           num_online_cpus() + 2, 8);
2700 
2701         INIT_LIST_HEAD(&fs_info->ordered_roots);
2702         spin_lock_init(&fs_info->ordered_root_lock);
2703 
2704         fs_info->btree_inode = new_inode(sb);
2705         if (!fs_info->btree_inode) {
2706                 err = -ENOMEM;
2707                 goto fail_bio_counter;
2708         }
2709         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2710 
2711         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2712                                         GFP_KERNEL);
2713         if (!fs_info->delayed_root) {
2714                 err = -ENOMEM;
2715                 goto fail_iput;
2716         }
2717         btrfs_init_delayed_root(fs_info->delayed_root);
2718 
2719         btrfs_init_scrub(fs_info);
2720 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2721         fs_info->check_integrity_print_mask = 0;
2722 #endif
2723         btrfs_init_balance(fs_info);
2724         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2725 
2726         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2727         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2728 
2729         btrfs_init_btree_inode(fs_info);
2730 
2731         spin_lock_init(&fs_info->block_group_cache_lock);
2732         fs_info->block_group_cache_tree = RB_ROOT;
2733         fs_info->first_logical_byte = (u64)-1;
2734 
2735         extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2736                             IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2737         extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2738                             IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2739         fs_info->pinned_extents = &fs_info->freed_extents[0];
2740         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2741 
2742         mutex_init(&fs_info->ordered_operations_mutex);
2743         mutex_init(&fs_info->tree_log_mutex);
2744         mutex_init(&fs_info->chunk_mutex);
2745         mutex_init(&fs_info->transaction_kthread_mutex);
2746         mutex_init(&fs_info->cleaner_mutex);
2747         mutex_init(&fs_info->ro_block_group_mutex);
2748         init_rwsem(&fs_info->commit_root_sem);
2749         init_rwsem(&fs_info->cleanup_work_sem);
2750         init_rwsem(&fs_info->subvol_sem);
2751         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2752 
2753         btrfs_init_dev_replace_locks(fs_info);
2754         btrfs_init_qgroup(fs_info);
2755 
2756         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2757         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2758 
2759         init_waitqueue_head(&fs_info->transaction_throttle);
2760         init_waitqueue_head(&fs_info->transaction_wait);
2761         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2762         init_waitqueue_head(&fs_info->async_submit_wait);
2763         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2764 
2765         /* Usable values until the real ones are cached from the superblock */
2766         fs_info->nodesize = 4096;
2767         fs_info->sectorsize = 4096;
2768         fs_info->stripesize = 4096;
2769 
2770         spin_lock_init(&fs_info->swapfile_pins_lock);
2771         fs_info->swapfile_pins = RB_ROOT;
2772 
2773         fs_info->send_in_progress = 0;
2774 
2775         ret = btrfs_alloc_stripe_hash_table(fs_info);
2776         if (ret) {
2777                 err = ret;
2778                 goto fail_alloc;
2779         }
2780 
2781         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2782 
2783         invalidate_bdev(fs_devices->latest_bdev);
2784 
2785         /*
2786          * Read super block and check the signature bytes only
2787          */
2788         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2789         if (IS_ERR(bh)) {
2790                 err = PTR_ERR(bh);
2791                 goto fail_alloc;
2792         }
2793 
2794         /*
2795          * Verify the type first, if that or the the checksum value are
2796          * corrupted, we'll find out
2797          */
2798         csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2799         if (!btrfs_supported_super_csum(csum_type)) {
2800                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2801                           csum_type);
2802                 err = -EINVAL;
2803                 brelse(bh);
2804                 goto fail_alloc;
2805         }
2806 
2807         ret = btrfs_init_csum_hash(fs_info, csum_type);
2808         if (ret) {
2809                 err = ret;
2810                 goto fail_alloc;
2811         }
2812 
2813         /*
2814          * We want to check superblock checksum, the type is stored inside.
2815          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2816          */
2817         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2818                 btrfs_err(fs_info, "superblock checksum mismatch");
2819                 err = -EINVAL;
2820                 brelse(bh);
2821                 goto fail_csum;
2822         }
2823 
2824         /*
2825          * super_copy is zeroed at allocation time and we never touch the
2826          * following bytes up to INFO_SIZE, the checksum is calculated from
2827          * the whole block of INFO_SIZE
2828          */
2829         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2830         brelse(bh);
2831 
2832         disk_super = fs_info->super_copy;
2833 
2834         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2835                        BTRFS_FSID_SIZE));
2836 
2837         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2838                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2839                                 fs_info->super_copy->metadata_uuid,
2840                                 BTRFS_FSID_SIZE));
2841         }
2842 
2843         features = btrfs_super_flags(disk_super);
2844         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2845                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2846                 btrfs_set_super_flags(disk_super, features);
2847                 btrfs_info(fs_info,
2848                         "found metadata UUID change in progress flag, clearing");
2849         }
2850 
2851         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2852                sizeof(*fs_info->super_for_commit));
2853 
2854         ret = btrfs_validate_mount_super(fs_info);
2855         if (ret) {
2856                 btrfs_err(fs_info, "superblock contains fatal errors");
2857                 err = -EINVAL;
2858                 goto fail_csum;
2859         }
2860 
2861         if (!btrfs_super_root(disk_super))
2862                 goto fail_csum;
2863 
2864         /* check FS state, whether FS is broken. */
2865         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2866                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2867 
2868         /*
2869          * run through our array of backup supers and setup
2870          * our ring pointer to the oldest one
2871          */
2872         generation = btrfs_super_generation(disk_super);
2873         find_oldest_super_backup(fs_info, generation);
2874 
2875         /*
2876          * In the long term, we'll store the compression type in the super
2877          * block, and it'll be used for per file compression control.
2878          */
2879         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2880 
2881         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2882         if (ret) {
2883                 err = ret;
2884                 goto fail_csum;
2885         }
2886 
2887         features = btrfs_super_incompat_flags(disk_super) &
2888                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2889         if (features) {
2890                 btrfs_err(fs_info,
2891                     "cannot mount because of unsupported optional features (%llx)",
2892                     features);
2893                 err = -EINVAL;
2894                 goto fail_csum;
2895         }
2896 
2897         features = btrfs_super_incompat_flags(disk_super);
2898         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2899         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2900                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2901         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2902                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2903 
2904         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2905                 btrfs_info(fs_info, "has skinny extents");
2906 
2907         /*
2908          * flag our filesystem as having big metadata blocks if
2909          * they are bigger than the page size
2910          */
2911         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2912                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2913                         btrfs_info(fs_info,
2914                                 "flagging fs with big metadata feature");
2915                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2916         }
2917 
2918         nodesize = btrfs_super_nodesize(disk_super);
2919         sectorsize = btrfs_super_sectorsize(disk_super);
2920         stripesize = sectorsize;
2921         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2922         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2923 
2924         /* Cache block sizes */
2925         fs_info->nodesize = nodesize;
2926         fs_info->sectorsize = sectorsize;
2927         fs_info->stripesize = stripesize;
2928 
2929         /*
2930          * mixed block groups end up with duplicate but slightly offset
2931          * extent buffers for the same range.  It leads to corruptions
2932          */
2933         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2934             (sectorsize != nodesize)) {
2935                 btrfs_err(fs_info,
2936 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2937                         nodesize, sectorsize);
2938                 goto fail_csum;
2939         }
2940 
2941         /*
2942          * Needn't use the lock because there is no other task which will
2943          * update the flag.
2944          */
2945         btrfs_set_super_incompat_flags(disk_super, features);
2946 
2947         features = btrfs_super_compat_ro_flags(disk_super) &
2948                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2949         if (!sb_rdonly(sb) && features) {
2950                 btrfs_err(fs_info,
2951         "cannot mount read-write because of unsupported optional features (%llx)",
2952                        features);
2953                 err = -EINVAL;
2954                 goto fail_csum;
2955         }
2956 
2957         ret = btrfs_init_workqueues(fs_info, fs_devices);
2958         if (ret) {
2959                 err = ret;
2960                 goto fail_sb_buffer;
2961         }
2962 
2963         sb->s_bdi->congested_fn = btrfs_congested_fn;
2964         sb->s_bdi->congested_data = fs_info;
2965         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2966         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2967         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2968         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2969 
2970         sb->s_blocksize = sectorsize;
2971         sb->s_blocksize_bits = blksize_bits(sectorsize);
2972         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2973 
2974         mutex_lock(&fs_info->chunk_mutex);
2975         ret = btrfs_read_sys_array(fs_info);
2976         mutex_unlock(&fs_info->chunk_mutex);
2977         if (ret) {
2978                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2979                 goto fail_sb_buffer;
2980         }
2981 
2982         generation = btrfs_super_chunk_root_generation(disk_super);
2983         level = btrfs_super_chunk_root_level(disk_super);
2984 
2985         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2986 
2987         chunk_root->node = read_tree_block(fs_info,
2988                                            btrfs_super_chunk_root(disk_super),
2989                                            generation, level, NULL);
2990         if (IS_ERR(chunk_root->node) ||
2991             !extent_buffer_uptodate(chunk_root->node)) {
2992                 btrfs_err(fs_info, "failed to read chunk root");
2993                 if (!IS_ERR(chunk_root->node))
2994                         free_extent_buffer(chunk_root->node);
2995                 chunk_root->node = NULL;
2996                 goto fail_tree_roots;
2997         }
2998         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2999         chunk_root->commit_root = btrfs_root_node(chunk_root);
3000 
3001         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3002            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3003 
3004         ret = btrfs_read_chunk_tree(fs_info);
3005         if (ret) {
3006                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3007                 goto fail_tree_roots;
3008         }
3009 
3010         /*
3011          * Keep the devid that is marked to be the target device for the
3012          * device replace procedure
3013          */
3014         btrfs_free_extra_devids(fs_devices, 0);
3015 
3016         if (!fs_devices->latest_bdev) {
3017                 btrfs_err(fs_info, "failed to read devices");
3018                 goto fail_tree_roots;
3019         }
3020 
3021 retry_root_backup:
3022         generation = btrfs_super_generation(disk_super);
3023         level = btrfs_super_root_level(disk_super);
3024 
3025         tree_root->node = read_tree_block(fs_info,
3026                                           btrfs_super_root(disk_super),
3027                                           generation, level, NULL);
3028         if (IS_ERR(tree_root->node) ||
3029             !extent_buffer_uptodate(tree_root->node)) {
3030                 btrfs_warn(fs_info, "failed to read tree root");
3031                 if (!IS_ERR(tree_root->node))
3032                         free_extent_buffer(tree_root->node);
3033                 tree_root->node = NULL;
3034                 goto recovery_tree_root;
3035         }
3036 
3037         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3038         tree_root->commit_root = btrfs_root_node(tree_root);
3039         btrfs_set_root_refs(&tree_root->root_item, 1);
3040 
3041         mutex_lock(&tree_root->objectid_mutex);
3042         ret = btrfs_find_highest_objectid(tree_root,
3043                                         &tree_root->highest_objectid);
3044         if (ret) {
3045                 mutex_unlock(&tree_root->objectid_mutex);
3046                 goto recovery_tree_root;
3047         }
3048 
3049         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3050 
3051         mutex_unlock(&tree_root->objectid_mutex);
3052 
3053         ret = btrfs_read_roots(fs_info);
3054         if (ret)
3055                 goto recovery_tree_root;
3056 
3057         fs_info->generation = generation;
3058         fs_info->last_trans_committed = generation;
3059 
3060         /*
3061          * If we have a uuid root and we're not being told to rescan we need to
3062          * check the generation here so we can set the
3063          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3064          * transaction during a balance or the log replay without updating the
3065          * uuid generation, and then if we crash we would rescan the uuid tree,
3066          * even though it was perfectly fine.
3067          */
3068         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3069             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3070                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3071 
3072         ret = btrfs_verify_dev_extents(fs_info);
3073         if (ret) {
3074                 btrfs_err(fs_info,
3075                           "failed to verify dev extents against chunks: %d",
3076                           ret);
3077                 goto fail_block_groups;
3078         }
3079         ret = btrfs_recover_balance(fs_info);
3080         if (ret) {
3081                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3082                 goto fail_block_groups;
3083         }
3084 
3085         ret = btrfs_init_dev_stats(fs_info);
3086         if (ret) {
3087                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3088                 goto fail_block_groups;
3089         }
3090 
3091         ret = btrfs_init_dev_replace(fs_info);
3092         if (ret) {
3093                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3094                 goto fail_block_groups;
3095         }
3096 
3097         btrfs_free_extra_devids(fs_devices, 1);
3098 
3099         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3100         if (ret) {
3101                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3102                                 ret);
3103                 goto fail_block_groups;
3104         }
3105 
3106         ret = btrfs_sysfs_add_device(fs_devices);
3107         if (ret) {
3108                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3109                                 ret);
3110                 goto fail_fsdev_sysfs;
3111         }
3112 
3113         ret = btrfs_sysfs_add_mounted(fs_info);
3114         if (ret) {
3115                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3116                 goto fail_fsdev_sysfs;
3117         }
3118 
3119         ret = btrfs_init_space_info(fs_info);
3120         if (ret) {
3121                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3122                 goto fail_sysfs;
3123         }
3124 
3125         ret = btrfs_read_block_groups(fs_info);
3126         if (ret) {
3127                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3128                 goto fail_sysfs;
3129         }
3130 
3131         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3132                 btrfs_warn(fs_info,
3133                 "writable mount is not allowed due to too many missing devices");
3134                 goto fail_sysfs;
3135         }
3136 
3137         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3138                                                "btrfs-cleaner");
3139         if (IS_ERR(fs_info->cleaner_kthread))
3140                 goto fail_sysfs;
3141 
3142         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3143                                                    tree_root,
3144                                                    "btrfs-transaction");
3145         if (IS_ERR(fs_info->transaction_kthread))
3146                 goto fail_cleaner;
3147 
3148         if (!btrfs_test_opt(fs_info, NOSSD) &&
3149             !fs_info->fs_devices->rotating) {
3150                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3151         }
3152 
3153         /*
3154          * Mount does not set all options immediately, we can do it now and do
3155          * not have to wait for transaction commit
3156          */
3157         btrfs_apply_pending_changes(fs_info);
3158 
3159 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3160         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3161                 ret = btrfsic_mount(fs_info, fs_devices,
3162                                     btrfs_test_opt(fs_info,
3163                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3164                                     1 : 0,
3165                                     fs_info->check_integrity_print_mask);
3166                 if (ret)
3167                         btrfs_warn(fs_info,
3168                                 "failed to initialize integrity check module: %d",
3169                                 ret);
3170         }
3171 #endif
3172         ret = btrfs_read_qgroup_config(fs_info);
3173         if (ret)
3174                 goto fail_trans_kthread;
3175 
3176         if (btrfs_build_ref_tree(fs_info))
3177                 btrfs_err(fs_info, "couldn't build ref tree");
3178 
3179         /* do not make disk changes in broken FS or nologreplay is given */
3180         if (btrfs_super_log_root(disk_super) != 0 &&
3181             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3182                 btrfs_info(fs_info, "start tree-log replay");
3183                 ret = btrfs_replay_log(fs_info, fs_devices);
3184                 if (ret) {
3185                         err = ret;
3186                         goto fail_qgroup;
3187                 }
3188         }
3189 
3190         ret = btrfs_find_orphan_roots(fs_info);
3191         if (ret)
3192                 goto fail_qgroup;
3193 
3194         if (!sb_rdonly(sb)) {
3195                 ret = btrfs_cleanup_fs_roots(fs_info);
3196                 if (ret)
3197                         goto fail_qgroup;
3198 
3199                 mutex_lock(&fs_info->cleaner_mutex);
3200                 ret = btrfs_recover_relocation(tree_root);
3201                 mutex_unlock(&fs_info->cleaner_mutex);
3202                 if (ret < 0) {
3203                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3204                                         ret);
3205                         err = -EINVAL;
3206                         goto fail_qgroup;
3207                 }
3208         }
3209 
3210         location.objectid = BTRFS_FS_TREE_OBJECTID;
3211         location.type = BTRFS_ROOT_ITEM_KEY;
3212         location.offset = 0;
3213 
3214         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3215         if (IS_ERR(fs_info->fs_root)) {
3216                 err = PTR_ERR(fs_info->fs_root);
3217                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3218                 fs_info->fs_root = NULL;
3219                 goto fail_qgroup;
3220         }
3221 
3222         if (sb_rdonly(sb))
3223                 return 0;
3224 
3225         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3226             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3227                 clear_free_space_tree = 1;
3228         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3229                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3230                 btrfs_warn(fs_info, "free space tree is invalid");
3231                 clear_free_space_tree = 1;
3232         }
3233 
3234         if (clear_free_space_tree) {
3235                 btrfs_info(fs_info, "clearing free space tree");
3236                 ret = btrfs_clear_free_space_tree(fs_info);
3237                 if (ret) {
3238                         btrfs_warn(fs_info,
3239                                    "failed to clear free space tree: %d", ret);
3240                         close_ctree(fs_info);
3241                         return ret;
3242                 }
3243         }
3244 
3245         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3246             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3247                 btrfs_info(fs_info, "creating free space tree");
3248                 ret = btrfs_create_free_space_tree(fs_info);
3249                 if (ret) {
3250                         btrfs_warn(fs_info,
3251                                 "failed to create free space tree: %d", ret);
3252                         close_ctree(fs_info);
3253                         return ret;
3254                 }
3255         }
3256 
3257         down_read(&fs_info->cleanup_work_sem);
3258         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3259             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3260                 up_read(&fs_info->cleanup_work_sem);
3261                 close_ctree(fs_info);
3262                 return ret;
3263         }
3264         up_read(&fs_info->cleanup_work_sem);
3265 
3266         ret = btrfs_resume_balance_async(fs_info);
3267         if (ret) {
3268                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3269                 close_ctree(fs_info);
3270                 return ret;
3271         }
3272 
3273         ret = btrfs_resume_dev_replace_async(fs_info);
3274         if (ret) {
3275                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3276                 close_ctree(fs_info);
3277                 return ret;
3278         }
3279 
3280         btrfs_qgroup_rescan_resume(fs_info);
3281 
3282         if (!fs_info->uuid_root) {
3283                 btrfs_info(fs_info, "creating UUID tree");
3284                 ret = btrfs_create_uuid_tree(fs_info);
3285                 if (ret) {
3286                         btrfs_warn(fs_info,
3287                                 "failed to create the UUID tree: %d", ret);
3288                         close_ctree(fs_info);
3289                         return ret;
3290                 }
3291         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3292                    fs_info->generation !=
3293                                 btrfs_super_uuid_tree_generation(disk_super)) {
3294                 btrfs_info(fs_info, "checking UUID tree");
3295                 ret = btrfs_check_uuid_tree(fs_info);
3296                 if (ret) {
3297                         btrfs_warn(fs_info,
3298                                 "failed to check the UUID tree: %d", ret);
3299                         close_ctree(fs_info);
3300                         return ret;
3301                 }
3302         }
3303         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3304 
3305         /*
3306          * backuproot only affect mount behavior, and if open_ctree succeeded,
3307          * no need to keep the flag
3308          */
3309         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3310 
3311         return 0;
3312 
3313 fail_qgroup:
3314         btrfs_free_qgroup_config(fs_info);
3315 fail_trans_kthread:
3316         kthread_stop(fs_info->transaction_kthread);
3317         btrfs_cleanup_transaction(fs_info);
3318         btrfs_free_fs_roots(fs_info);
3319 fail_cleaner:
3320         kthread_stop(fs_info->cleaner_kthread);
3321 
3322         /*
3323          * make sure we're done with the btree inode before we stop our
3324          * kthreads
3325          */
3326         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3327 
3328 fail_sysfs:
3329         btrfs_sysfs_remove_mounted(fs_info);
3330 
3331 fail_fsdev_sysfs:
3332         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3333 
3334 fail_block_groups:
3335         btrfs_put_block_group_cache(fs_info);
3336 
3337 fail_tree_roots:
3338         free_root_pointers(fs_info, true);
3339         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3340 
3341 fail_sb_buffer:
3342         btrfs_stop_all_workers(fs_info);
3343         btrfs_free_block_groups(fs_info);
3344 fail_csum:
3345         btrfs_free_csum_hash(fs_info);
3346 fail_alloc:
3347 fail_iput:
3348         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3349 
3350         iput(fs_info->btree_inode);
3351 fail_bio_counter:
3352         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3353 fail_delalloc_bytes:
3354         percpu_counter_destroy(&fs_info->delalloc_bytes);
3355 fail_dirty_metadata_bytes:
3356         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3357 fail_dio_bytes:
3358         percpu_counter_destroy(&fs_info->dio_bytes);
3359 fail_srcu:
3360         cleanup_srcu_struct(&fs_info->subvol_srcu);
3361 fail:
3362         btrfs_free_stripe_hash_table(fs_info);
3363         btrfs_close_devices(fs_info->fs_devices);
3364         return err;
3365 
3366 recovery_tree_root:
3367         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3368                 goto fail_tree_roots;
3369 
3370         free_root_pointers(fs_info, false);
3371 
3372         /* don't use the log in recovery mode, it won't be valid */
3373         btrfs_set_super_log_root(disk_super, 0);
3374 
3375         /* we can't trust the free space cache either */
3376         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3377 
3378         ret = next_root_backup(fs_info, fs_info->super_copy,
3379                                &num_backups_tried, &backup_index);
3380         if (ret == -1)
3381                 goto fail_block_groups;
3382         goto retry_root_backup;
3383 }
3384 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3385 
3386 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3387 {
3388         if (uptodate) {
3389                 set_buffer_uptodate(bh);
3390         } else {
3391                 struct btrfs_device *device = (struct btrfs_device *)
3392                         bh->b_private;
3393 
3394                 btrfs_warn_rl_in_rcu(device->fs_info,
3395                                 "lost page write due to IO error on %s",
3396                                           rcu_str_deref(device->name));
3397                 /* note, we don't set_buffer_write_io_error because we have
3398                  * our own ways of dealing with the IO errors
3399                  */
3400                 clear_buffer_uptodate(bh);
3401                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3402         }
3403         unlock_buffer(bh);
3404         put_bh(bh);
3405 }
3406 
3407 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3408                         struct buffer_head **bh_ret)
3409 {
3410         struct buffer_head *bh;
3411         struct btrfs_super_block *super;
3412         u64 bytenr;
3413 
3414         bytenr = btrfs_sb_offset(copy_num);
3415         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3416                 return -EINVAL;
3417 
3418         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3419         /*
3420          * If we fail to read from the underlying devices, as of now
3421          * the best option we have is to mark it EIO.
3422          */
3423         if (!bh)
3424                 return -EIO;
3425 
3426         super = (struct btrfs_super_block *)bh->b_data;
3427         if (btrfs_super_bytenr(super) != bytenr ||
3428                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3429                 brelse(bh);
3430                 return -EINVAL;
3431         }
3432 
3433         *bh_ret = bh;
3434         return 0;
3435 }
3436 
3437 
3438 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3439 {
3440         struct buffer_head *bh;
3441         struct buffer_head *latest = NULL;
3442         struct btrfs_super_block *super;
3443         int i;
3444         u64 transid = 0;
3445         int ret = -EINVAL;
3446 
3447         /* we would like to check all the supers, but that would make
3448          * a btrfs mount succeed after a mkfs from a different FS.
3449          * So, we need to add a special mount option to scan for
3450          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3451          */
3452         for (i = 0; i < 1; i++) {
3453                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3454                 if (ret)
3455                         continue;
3456 
3457                 super = (struct btrfs_super_block *)bh->b_data;
3458 
3459                 if (!latest || btrfs_super_generation(super) > transid) {
3460                         brelse(latest);
3461                         latest = bh;
3462                         transid = btrfs_super_generation(super);
3463                 } else {
3464                         brelse(bh);
3465                 }
3466         }
3467 
3468         if (!latest)
3469                 return ERR_PTR(ret);
3470 
3471         return latest;
3472 }
3473 
3474 /*
3475  * Write superblock @sb to the @device. Do not wait for completion, all the
3476  * buffer heads we write are pinned.
3477  *
3478  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3479  * the expected device size at commit time. Note that max_mirrors must be
3480  * same for write and wait phases.
3481  *
3482  * Return number of errors when buffer head is not found or submission fails.
3483  */
3484 static int write_dev_supers(struct btrfs_device *device,
3485                             struct btrfs_super_block *sb, int max_mirrors)
3486 {
3487         struct btrfs_fs_info *fs_info = device->fs_info;
3488         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3489         struct buffer_head *bh;
3490         int i;
3491         int ret;
3492         int errors = 0;
3493         u64 bytenr;
3494         int op_flags;
3495 
3496         if (max_mirrors == 0)
3497                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3498 
3499         shash->tfm = fs_info->csum_shash;
3500 
3501         for (i = 0; i < max_mirrors; i++) {
3502                 bytenr = btrfs_sb_offset(i);
3503                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3504                     device->commit_total_bytes)
3505                         break;
3506 
3507                 btrfs_set_super_bytenr(sb, bytenr);
3508 
3509                 crypto_shash_init(shash);
3510                 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3511                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3512                 crypto_shash_final(shash, sb->csum);
3513 
3514                 /* One reference for us, and we leave it for the caller */
3515                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3516                               BTRFS_SUPER_INFO_SIZE);
3517                 if (!bh) {
3518                         btrfs_err(device->fs_info,
3519                             "couldn't get super buffer head for bytenr %llu",
3520                             bytenr);
3521                         errors++;
3522                         continue;
3523                 }
3524 
3525                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3526 
3527                 /* one reference for submit_bh */
3528                 get_bh(bh);
3529 
3530                 set_buffer_uptodate(bh);
3531                 lock_buffer(bh);
3532                 bh->b_end_io = btrfs_end_buffer_write_sync;
3533                 bh->b_private = device;
3534 
3535                 /*
3536                  * we fua the first super.  The others we allow
3537                  * to go down lazy.
3538                  */
3539                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3540                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3541                         op_flags |= REQ_FUA;
3542                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3543                 if (ret)
3544                         errors++;
3545         }
3546         return errors < i ? 0 : -1;
3547 }
3548 
3549 /*
3550  * Wait for write completion of superblocks done by write_dev_supers,
3551  * @max_mirrors same for write and wait phases.
3552  *
3553  * Return number of errors when buffer head is not found or not marked up to
3554  * date.
3555  */
3556 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3557 {
3558         struct buffer_head *bh;
3559         int i;
3560         int errors = 0;
3561         bool primary_failed = false;
3562         u64 bytenr;
3563 
3564         if (max_mirrors == 0)
3565                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3566 
3567         for (i = 0; i < max_mirrors; i++) {
3568                 bytenr = btrfs_sb_offset(i);
3569                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3570                     device->commit_total_bytes)
3571                         break;
3572 
3573                 bh = __find_get_block(device->bdev,
3574                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3575                                       BTRFS_SUPER_INFO_SIZE);
3576                 if (!bh) {
3577                         errors++;
3578                         if (i == 0)
3579                                 primary_failed = true;
3580                         continue;
3581                 }
3582                 wait_on_buffer(bh);
3583                 if (!buffer_uptodate(bh)) {
3584                         errors++;
3585                         if (i == 0)
3586                                 primary_failed = true;
3587                 }
3588 
3589                 /* drop our reference */
3590                 brelse(bh);
3591 
3592                 /* drop the reference from the writing run */
3593                 brelse(bh);
3594         }
3595 
3596         /* log error, force error return */
3597         if (primary_failed) {
3598                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3599                           device->devid);
3600                 return -1;
3601         }
3602 
3603         return errors < i ? 0 : -1;
3604 }
3605 
3606 /*
3607  * endio for the write_dev_flush, this will wake anyone waiting
3608  * for the barrier when it is done
3609  */
3610 static void btrfs_end_empty_barrier(struct bio *bio)
3611 {
3612         complete(bio->bi_private);
3613 }
3614 
3615 /*
3616  * Submit a flush request to the device if it supports it. Error handling is
3617  * done in the waiting counterpart.
3618  */
3619 static void write_dev_flush(struct btrfs_device *device)
3620 {
3621         struct request_queue *q = bdev_get_queue(device->bdev);
3622         struct bio *bio = device->flush_bio;
3623 
3624         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3625                 return;
3626 
3627         bio_reset(bio);
3628         bio->bi_end_io = btrfs_end_empty_barrier;
3629         bio_set_dev(bio, device->bdev);
3630         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3631         init_completion(&device->flush_wait);
3632         bio->bi_private = &device->flush_wait;
3633 
3634         btrfsic_submit_bio(bio);
3635         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3636 }
3637 
3638 /*
3639  * If the flush bio has been submitted by write_dev_flush, wait for it.
3640  */
3641 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3642 {
3643         struct bio *bio = device->flush_bio;
3644 
3645         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3646                 return BLK_STS_OK;
3647 
3648         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3649         wait_for_completion_io(&device->flush_wait);
3650 
3651         return bio->bi_status;
3652 }
3653 
3654 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3655 {
3656         if (!btrfs_check_rw_degradable(fs_info, NULL))
3657                 return -EIO;
3658         return 0;
3659 }
3660 
3661 /*
3662  * send an empty flush down to each device in parallel,
3663  * then wait for them
3664  */
3665 static int barrier_all_devices(struct btrfs_fs_info *info)
3666 {
3667         struct list_head *head;
3668         struct btrfs_device *dev;
3669         int errors_wait = 0;
3670         blk_status_t ret;
3671 
3672         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3673         /* send down all the barriers */
3674         head = &info->fs_devices->devices;
3675         list_for_each_entry(dev, head, dev_list) {
3676                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3677                         continue;
3678                 if (!dev->bdev)
3679                         continue;
3680                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3681                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3682                         continue;
3683 
3684                 write_dev_flush(dev);
3685                 dev->last_flush_error = BLK_STS_OK;
3686         }
3687 
3688         /* wait for all the barriers */
3689         list_for_each_entry(dev, head, dev_list) {
3690                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3691                         continue;
3692                 if (!dev->bdev) {
3693                         errors_wait++;
3694                         continue;
3695                 }
3696                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3697                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3698                         continue;
3699 
3700                 ret = wait_dev_flush(dev);
3701                 if (ret) {
3702                         dev->last_flush_error = ret;
3703                         btrfs_dev_stat_inc_and_print(dev,
3704                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3705                         errors_wait++;
3706                 }
3707         }
3708 
3709         if (errors_wait) {
3710                 /*
3711                  * At some point we need the status of all disks
3712                  * to arrive at the volume status. So error checking
3713                  * is being pushed to a separate loop.
3714                  */
3715                 return check_barrier_error(info);
3716         }
3717         return 0;
3718 }
3719 
3720 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3721 {
3722         int raid_type;
3723         int min_tolerated = INT_MAX;
3724 
3725         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3726             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3727                 min_tolerated = min_t(int, min_tolerated,
3728                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3729                                     tolerated_failures);
3730 
3731         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3732                 if (raid_type == BTRFS_RAID_SINGLE)
3733                         continue;
3734                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3735                         continue;
3736                 min_tolerated = min_t(int, min_tolerated,
3737                                     btrfs_raid_array[raid_type].
3738                                     tolerated_failures);
3739         }
3740 
3741         if (min_tolerated == INT_MAX) {
3742                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3743                 min_tolerated = 0;
3744         }
3745 
3746         return min_tolerated;
3747 }
3748 
3749 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3750 {
3751         struct list_head *head;
3752         struct btrfs_device *dev;
3753         struct btrfs_super_block *sb;
3754         struct btrfs_dev_item *dev_item;
3755         int ret;
3756         int do_barriers;
3757         int max_errors;
3758         int total_errors = 0;
3759         u64 flags;
3760 
3761         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3762 
3763         /*
3764          * max_mirrors == 0 indicates we're from commit_transaction,
3765          * not from fsync where the tree roots in fs_info have not
3766          * been consistent on disk.
3767          */
3768         if (max_mirrors == 0)
3769                 backup_super_roots(fs_info);
3770 
3771         sb = fs_info->super_for_commit;
3772         dev_item = &sb->dev_item;
3773 
3774         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3775         head = &fs_info->fs_devices->devices;
3776         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3777 
3778         if (do_barriers) {
3779                 ret = barrier_all_devices(fs_info);
3780                 if (ret) {
3781                         mutex_unlock(
3782                                 &fs_info->fs_devices->device_list_mutex);
3783                         btrfs_handle_fs_error(fs_info, ret,
3784                                               "errors while submitting device barriers.");
3785                         return ret;
3786                 }
3787         }
3788 
3789         list_for_each_entry(dev, head, dev_list) {
3790                 if (!dev->bdev) {
3791                         total_errors++;
3792                         continue;
3793                 }
3794                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3795                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3796                         continue;
3797 
3798                 btrfs_set_stack_device_generation(dev_item, 0);
3799                 btrfs_set_stack_device_type(dev_item, dev->type);
3800                 btrfs_set_stack_device_id(dev_item, dev->devid);
3801                 btrfs_set_stack_device_total_bytes(dev_item,
3802                                                    dev->commit_total_bytes);
3803                 btrfs_set_stack_device_bytes_used(dev_item,
3804                                                   dev->commit_bytes_used);
3805                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3806                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3807                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3808                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3809                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3810                        BTRFS_FSID_SIZE);
3811 
3812                 flags = btrfs_super_flags(sb);
3813                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3814 
3815                 ret = btrfs_validate_write_super(fs_info, sb);
3816                 if (ret < 0) {
3817                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3818                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3819                                 "unexpected superblock corruption detected");
3820                         return -EUCLEAN;
3821                 }
3822 
3823                 ret = write_dev_supers(dev, sb, max_mirrors);
3824                 if (ret)
3825                         total_errors++;
3826         }
3827         if (total_errors > max_errors) {
3828                 btrfs_err(fs_info, "%d errors while writing supers",
3829                           total_errors);
3830                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3831 
3832                 /* FUA is masked off if unsupported and can't be the reason */
3833                 btrfs_handle_fs_error(fs_info, -EIO,
3834                                       "%d errors while writing supers",
3835                                       total_errors);
3836                 return -EIO;
3837         }
3838 
3839         total_errors = 0;
3840         list_for_each_entry(dev, head, dev_list) {
3841                 if (!dev->bdev)
3842                         continue;
3843                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3844                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3845                         continue;
3846 
3847                 ret = wait_dev_supers(dev, max_mirrors);
3848                 if (ret)
3849                         total_errors++;
3850         }
3851         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3852         if (total_errors > max_errors) {
3853                 btrfs_handle_fs_error(fs_info, -EIO,
3854                                       "%d errors while writing supers",
3855                                       total_errors);
3856                 return -EIO;
3857         }
3858         return 0;
3859 }
3860 
3861 /* Drop a fs root from the radix tree and free it. */
3862 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3863                                   struct btrfs_root *root)
3864 {
3865         spin_lock(&fs_info->fs_roots_radix_lock);
3866         radix_tree_delete(&fs_info->fs_roots_radix,
3867                           (unsigned long)root->root_key.objectid);
3868         spin_unlock(&fs_info->fs_roots_radix_lock);
3869 
3870         if (btrfs_root_refs(&root->root_item) == 0)
3871                 synchronize_srcu(&fs_info->subvol_srcu);
3872 
3873         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3874                 btrfs_free_log(NULL, root);
3875                 if (root->reloc_root) {
3876                         free_extent_buffer(root->reloc_root->node);
3877                         free_extent_buffer(root->reloc_root->commit_root);
3878                         btrfs_put_fs_root(root->reloc_root);
3879                         root->reloc_root = NULL;
3880                 }
3881         }
3882 
3883         if (root->free_ino_pinned)
3884                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3885         if (root->free_ino_ctl)
3886                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3887         btrfs_free_fs_root(root);
3888 }
3889 
3890 void btrfs_free_fs_root(struct btrfs_root *root)
3891 {
3892         iput(root->ino_cache_inode);
3893         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3894         if (root->anon_dev)
3895                 free_anon_bdev(root->anon_dev);
3896         if (root->subv_writers)
3897                 btrfs_free_subvolume_writers(root->subv_writers);
3898         free_extent_buffer(root->node);
3899         free_extent_buffer(root->commit_root);
3900         kfree(root->free_ino_ctl);
3901         kfree(root->free_ino_pinned);
3902         btrfs_put_fs_root(root);
3903 }
3904 
3905 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3906 {
3907         u64 root_objectid = 0;
3908         struct btrfs_root *gang[8];
3909         int i = 0;
3910         int err = 0;
3911         unsigned int ret = 0;
3912         int index;
3913 
3914         while (1) {
3915                 index = srcu_read_lock(&fs_info->subvol_srcu);
3916                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3917                                              (void **)gang, root_objectid,
3918                                              ARRAY_SIZE(gang));
3919                 if (!ret) {
3920                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3921                         break;
3922                 }
3923                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3924 
3925                 for (i = 0; i < ret; i++) {
3926                         /* Avoid to grab roots in dead_roots */
3927                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3928                                 gang[i] = NULL;
3929                                 continue;
3930                         }
3931                         /* grab all the search result for later use */
3932                         gang[i] = btrfs_grab_fs_root(gang[i]);
3933                 }
3934                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3935 
3936                 for (i = 0; i < ret; i++) {
3937                         if (!gang[i])
3938                                 continue;
3939                         root_objectid = gang[i]->root_key.objectid;
3940                         err = btrfs_orphan_cleanup(gang[i]);
3941                         if (err)
3942                                 break;
3943                         btrfs_put_fs_root(gang[i]);
3944                 }
3945                 root_objectid++;
3946         }
3947 
3948         /* release the uncleaned roots due to error */
3949         for (; i < ret; i++) {
3950                 if (gang[i])
3951                         btrfs_put_fs_root(gang[i]);
3952         }
3953         return err;
3954 }
3955 
3956 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3957 {
3958         struct btrfs_root *root = fs_info->tree_root;
3959         struct btrfs_trans_handle *trans;
3960 
3961         mutex_lock(&fs_info->cleaner_mutex);
3962         btrfs_run_delayed_iputs(fs_info);
3963         mutex_unlock(&fs_info->cleaner_mutex);
3964         wake_up_process(fs_info->cleaner_kthread);
3965 
3966         /* wait until ongoing cleanup work done */
3967         down_write(&fs_info->cleanup_work_sem);
3968         up_write(&fs_info->cleanup_work_sem);
3969 
3970         trans = btrfs_join_transaction(root);
3971         if (IS_ERR(trans))
3972                 return PTR_ERR(trans);
3973         return btrfs_commit_transaction(trans);
3974 }
3975 
3976 void close_ctree(struct btrfs_fs_info *fs_info)
3977 {
3978         int ret;
3979 
3980         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3981         /*
3982          * We don't want the cleaner to start new transactions, add more delayed
3983          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3984          * because that frees the task_struct, and the transaction kthread might
3985          * still try to wake up the cleaner.
3986          */
3987         kthread_park(fs_info->cleaner_kthread);
3988 
3989         /* wait for the qgroup rescan worker to stop */
3990         btrfs_qgroup_wait_for_completion(fs_info, false);
3991 
3992         /* wait for the uuid_scan task to finish */
3993         down(&fs_info->uuid_tree_rescan_sem);
3994         /* avoid complains from lockdep et al., set sem back to initial state */
3995         up(&fs_info->uuid_tree_rescan_sem);
3996 
3997         /* pause restriper - we want to resume on mount */
3998         btrfs_pause_balance(fs_info);
3999 
4000         btrfs_dev_replace_suspend_for_unmount(fs_info);
4001 
4002         btrfs_scrub_cancel(fs_info);
4003 
4004         /* wait for any defraggers to finish */
4005         wait_event(fs_info->transaction_wait,
4006                    (atomic_read(&fs_info->defrag_running) == 0));
4007 
4008         /* clear out the rbtree of defraggable inodes */
4009         btrfs_cleanup_defrag_inodes(fs_info);
4010 
4011         cancel_work_sync(&fs_info->async_reclaim_work);
4012 
4013         if (!sb_rdonly(fs_info->sb)) {
4014                 /*
4015                  * The cleaner kthread is stopped, so do one final pass over
4016                  * unused block groups.
4017                  */
4018                 btrfs_delete_unused_bgs(fs_info);
4019 
4020                 /*
4021                  * There might be existing delayed inode workers still running
4022                  * and holding an empty delayed inode item. We must wait for
4023                  * them to complete first because they can create a transaction.
4024                  * This happens when someone calls btrfs_balance_delayed_items()
4025                  * and then a transaction commit runs the same delayed nodes
4026                  * before any delayed worker has done something with the nodes.
4027                  * We must wait for any worker here and not at transaction
4028                  * commit time since that could cause a deadlock.
4029                  * This is a very rare case.
4030                  */
4031                 btrfs_flush_workqueue(fs_info->delayed_workers);
4032 
4033                 ret = btrfs_commit_super(fs_info);
4034                 if (ret)
4035                         btrfs_err(fs_info, "commit super ret %d", ret);
4036         }
4037 
4038         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4039             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4040                 btrfs_error_commit_super(fs_info);
4041 
4042         kthread_stop(fs_info->transaction_kthread);
4043         kthread_stop(fs_info->cleaner_kthread);
4044 
4045         ASSERT(list_empty(&fs_info->delayed_iputs));
4046         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4047 
4048         btrfs_free_qgroup_config(fs_info);
4049         ASSERT(list_empty(&fs_info->delalloc_roots));
4050 
4051         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4052                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4053                        percpu_counter_sum(&fs_info->delalloc_bytes));
4054         }
4055 
4056         if (percpu_counter_sum(&fs_info->dio_bytes))
4057                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4058                            percpu_counter_sum(&fs_info->dio_bytes));
4059 
4060         btrfs_sysfs_remove_mounted(fs_info);
4061         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4062 
4063         btrfs_free_fs_roots(fs_info);
4064 
4065         btrfs_put_block_group_cache(fs_info);
4066 
4067         /*
4068          * we must make sure there is not any read request to
4069          * submit after we stopping all workers.
4070          */
4071         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4072         btrfs_stop_all_workers(fs_info);
4073 
4074         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4075         free_root_pointers(fs_info, true);
4076 
4077         /*
4078          * We must free the block groups after dropping the fs_roots as we could
4079          * have had an IO error and have left over tree log blocks that aren't
4080          * cleaned up until the fs roots are freed.  This makes the block group
4081          * accounting appear to be wrong because there's pending reserved bytes,
4082          * so make sure we do the block group cleanup afterwards.
4083          */
4084         btrfs_free_block_groups(fs_info);
4085 
4086         iput(fs_info->btree_inode);
4087 
4088 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4089         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4090                 btrfsic_unmount(fs_info->fs_devices);
4091 #endif
4092 
4093         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4094         btrfs_close_devices(fs_info->fs_devices);
4095 
4096         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4097         percpu_counter_destroy(&fs_info->delalloc_bytes);
4098         percpu_counter_destroy(&fs_info->dio_bytes);
4099         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4100         cleanup_srcu_struct(&fs_info->subvol_srcu);
4101 
4102         btrfs_free_csum_hash(fs_info);
4103         btrfs_free_stripe_hash_table(fs_info);
4104         btrfs_free_ref_cache(fs_info);
4105 }
4106 
4107 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4108                           int atomic)
4109 {
4110         int ret;
4111         struct inode *btree_inode = buf->pages[0]->mapping->host;
4112 
4113         ret = extent_buffer_uptodate(buf);
4114         if (!ret)
4115                 return ret;
4116 
4117         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4118                                     parent_transid, atomic);
4119         if (ret == -EAGAIN)
4120                 return ret;
4121         return !ret;
4122 }
4123 
4124 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4125 {
4126         struct btrfs_fs_info *fs_info;
4127         struct btrfs_root *root;
4128         u64 transid = btrfs_header_generation(buf);
4129         int was_dirty;
4130 
4131 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4132         /*
4133          * This is a fast path so only do this check if we have sanity tests
4134          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4135          * outside of the sanity tests.
4136          */
4137         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4138                 return;
4139 #endif
4140         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4141         fs_info = root->fs_info;
4142         btrfs_assert_tree_locked(buf);
4143         if (transid != fs_info->generation)
4144                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4145                         buf->start, transid, fs_info->generation);
4146         was_dirty = set_extent_buffer_dirty(buf);
4147         if (!was_dirty)
4148                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4149                                          buf->len,
4150                                          fs_info->dirty_metadata_batch);
4151 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4152         /*
4153          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4154          * but item data not updated.
4155          * So here we should only check item pointers, not item data.
4156          */
4157         if (btrfs_header_level(buf) == 0 &&
4158             btrfs_check_leaf_relaxed(buf)) {
4159                 btrfs_print_leaf(buf);
4160                 ASSERT(0);
4161         }
4162 #endif
4163 }
4164 
4165 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4166                                         int flush_delayed)
4167 {
4168         /*
4169          * looks as though older kernels can get into trouble with
4170          * this code, they end up stuck in balance_dirty_pages forever
4171          */
4172         int ret;
4173 
4174         if (current->flags & PF_MEMALLOC)
4175                 return;
4176 
4177         if (flush_delayed)
4178                 btrfs_balance_delayed_items(fs_info);
4179 
4180         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4181                                      BTRFS_DIRTY_METADATA_THRESH,
4182                                      fs_info->dirty_metadata_batch);
4183         if (ret > 0) {
4184                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4185         }
4186 }
4187 
4188 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4189 {
4190         __btrfs_btree_balance_dirty(fs_info, 1);
4191 }
4192 
4193 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4194 {
4195         __btrfs_btree_balance_dirty(fs_info, 0);
4196 }
4197 
4198 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4199                       struct btrfs_key *first_key)
4200 {
4201         return btree_read_extent_buffer_pages(buf, parent_transid,
4202                                               level, first_key);
4203 }
4204 
4205 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4206 {
4207         /* cleanup FS via transaction */
4208         btrfs_cleanup_transaction(fs_info);
4209 
4210         mutex_lock(&fs_info->cleaner_mutex);
4211         btrfs_run_delayed_iputs(fs_info);
4212         mutex_unlock(&fs_info->cleaner_mutex);
4213 
4214         down_write(&fs_info->cleanup_work_sem);
4215         up_write(&fs_info->cleanup_work_sem);
4216 }
4217 
4218 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4219 {
4220         struct btrfs_ordered_extent *ordered;
4221 
4222         spin_lock(&root->ordered_extent_lock);
4223         /*
4224          * This will just short circuit the ordered completion stuff which will
4225          * make sure the ordered extent gets properly cleaned up.
4226          */
4227         list_for_each_entry(ordered, &root->ordered_extents,
4228                             root_extent_list)
4229                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4230         spin_unlock(&root->ordered_extent_lock);
4231 }
4232 
4233 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4234 {
4235         struct btrfs_root *root;
4236         struct list_head splice;
4237 
4238         INIT_LIST_HEAD(&splice);
4239 
4240         spin_lock(&fs_info->ordered_root_lock);
4241         list_splice_init(&fs_info->ordered_roots, &splice);
4242         while (!list_empty(&splice)) {
4243                 root = list_first_entry(&splice, struct btrfs_root,
4244                                         ordered_root);
4245                 list_move_tail(&root->ordered_root,
4246                                &fs_info->ordered_roots);
4247 
4248                 spin_unlock(&fs_info->ordered_root_lock);
4249                 btrfs_destroy_ordered_extents(root);
4250 
4251                 cond_resched();
4252                 spin_lock(&fs_info->ordered_root_lock);
4253         }
4254         spin_unlock(&fs_info->ordered_root_lock);
4255 
4256         /*
4257          * We need this here because if we've been flipped read-only we won't
4258          * get sync() from the umount, so we need to make sure any ordered
4259          * extents that haven't had their dirty pages IO start writeout yet
4260          * actually get run and error out properly.
4261          */
4262         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4263 }
4264 
4265 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4266                                       struct btrfs_fs_info *fs_info)
4267 {
4268         struct rb_node *node;
4269         struct btrfs_delayed_ref_root *delayed_refs;
4270         struct btrfs_delayed_ref_node *ref;
4271         int ret = 0;
4272 
4273         delayed_refs = &trans->delayed_refs;
4274 
4275         spin_lock(&delayed_refs->lock);
4276         if (atomic_read(&delayed_refs->num_entries) == 0) {
4277                 spin_unlock(&delayed_refs->lock);
4278                 btrfs_info(fs_info, "delayed_refs has NO entry");
4279                 return ret;
4280         }
4281 
4282         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4283                 struct btrfs_delayed_ref_head *head;
4284                 struct rb_node *n;
4285                 bool pin_bytes = false;
4286 
4287                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4288                                 href_node);
4289                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4290                         continue;
4291 
4292                 spin_lock(&head->lock);
4293                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4294                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4295                                        ref_node);
4296                         ref->in_tree = 0;
4297                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4298                         RB_CLEAR_NODE(&ref->ref_node);
4299                         if (!list_empty(&ref->add_list))
4300                                 list_del(&ref->add_list);
4301                         atomic_dec(&delayed_refs->num_entries);
4302                         btrfs_put_delayed_ref(ref);
4303                 }
4304                 if (head->must_insert_reserved)
4305                         pin_bytes = true;
4306                 btrfs_free_delayed_extent_op(head->extent_op);
4307                 btrfs_delete_ref_head(delayed_refs, head);
4308                 spin_unlock(&head->lock);
4309                 spin_unlock(&delayed_refs->lock);
4310                 mutex_unlock(&head->mutex);
4311 
4312                 if (pin_bytes)
4313                         btrfs_pin_extent(fs_info, head->bytenr,
4314                                          head->num_bytes, 1);
4315                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4316                 btrfs_put_delayed_ref_head(head);
4317                 cond_resched();
4318                 spin_lock(&delayed_refs->lock);
4319         }
4320         btrfs_qgroup_destroy_extent_records(trans);
4321 
4322         spin_unlock(&delayed_refs->lock);
4323 
4324         return ret;
4325 }
4326 
4327 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4328 {
4329         struct btrfs_inode *btrfs_inode;
4330         struct list_head splice;
4331 
4332         INIT_LIST_HEAD(&splice);
4333 
4334         spin_lock(&root->delalloc_lock);
4335         list_splice_init(&root->delalloc_inodes, &splice);
4336 
4337         while (!list_empty(&splice)) {
4338                 struct inode *inode = NULL;
4339                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4340                                                delalloc_inodes);
4341                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4342                 spin_unlock(&root->delalloc_lock);
4343 
4344                 /*
4345                  * Make sure we get a live inode and that it'll not disappear
4346                  * meanwhile.
4347                  */
4348                 inode = igrab(&btrfs_inode->vfs_inode);
4349                 if (inode) {
4350                         invalidate_inode_pages2(inode->i_mapping);
4351                         iput(inode);
4352                 }
4353                 spin_lock(&root->delalloc_lock);
4354         }
4355         spin_unlock(&root->delalloc_lock);
4356 }
4357 
4358 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4359 {
4360         struct btrfs_root *root;
4361         struct list_head splice;
4362 
4363         INIT_LIST_HEAD(&splice);
4364 
4365         spin_lock(&fs_info->delalloc_root_lock);
4366         list_splice_init(&fs_info->delalloc_roots, &splice);
4367         while (!list_empty(&splice)) {
4368                 root = list_first_entry(&splice, struct btrfs_root,
4369                                          delalloc_root);
4370                 root = btrfs_grab_fs_root(root);
4371                 BUG_ON(!root);
4372                 spin_unlock(&fs_info->delalloc_root_lock);
4373 
4374                 btrfs_destroy_delalloc_inodes(root);
4375                 btrfs_put_fs_root(root);
4376 
4377                 spin_lock(&fs_info->delalloc_root_lock);
4378         }
4379         spin_unlock(&fs_info->delalloc_root_lock);
4380 }
4381 
4382 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4383                                         struct extent_io_tree *dirty_pages,
4384                                         int mark)
4385 {
4386         int ret;
4387         struct extent_buffer *eb;
4388         u64 start = 0;
4389         u64 end;
4390 
4391         while (1) {
4392                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4393                                             mark, NULL);
4394                 if (ret)
4395                         break;
4396 
4397                 clear_extent_bits(dirty_pages, start, end, mark);
4398                 while (start <= end) {
4399                         eb = find_extent_buffer(fs_info, start);
4400                         start += fs_info->nodesize;
4401                         if (!eb)
4402                                 continue;
4403                         wait_on_extent_buffer_writeback(eb);
4404 
4405                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4406                                                &eb->bflags))
4407                                 clear_extent_buffer_dirty(eb);
4408                         free_extent_buffer_stale(eb);
4409                 }
4410         }
4411 
4412         return ret;
4413 }
4414 
4415 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4416                                        struct extent_io_tree *pinned_extents)
4417 {
4418         struct extent_io_tree *unpin;
4419         u64 start;
4420         u64 end;
4421         int ret;
4422         bool loop = true;
4423 
4424         unpin = pinned_extents;
4425 again:
4426         while (1) {
4427                 struct extent_state *cached_state = NULL;
4428 
4429                 /*
4430                  * The btrfs_finish_extent_commit() may get the same range as
4431                  * ours between find_first_extent_bit and clear_extent_dirty.
4432                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4433                  * the same extent range.
4434                  */
4435                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4436                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4437                                             EXTENT_DIRTY, &cached_state);
4438                 if (ret) {
4439                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4440                         break;
4441                 }
4442 
4443                 clear_extent_dirty(unpin, start, end, &cached_state);
4444                 free_extent_state(cached_state);
4445                 btrfs_error_unpin_extent_range(fs_info, start, end);
4446                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4447                 cond_resched();
4448         }
4449 
4450         if (loop) {
4451                 if (unpin == &fs_info->freed_extents[0])
4452                         unpin = &fs_info->freed_extents[1];
4453                 else
4454                         unpin = &fs_info->freed_extents[0];
4455                 loop = false;
4456                 goto again;
4457         }
4458 
4459         return 0;
4460 }
4461 
4462 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4463 {
4464         struct inode *inode;
4465 
4466         inode = cache->io_ctl.inode;
4467         if (inode) {
4468                 invalidate_inode_pages2(inode->i_mapping);
4469                 BTRFS_I(inode)->generation = 0;
4470                 cache->io_ctl.inode = NULL;
4471                 iput(inode);
4472         }
4473         btrfs_put_block_group(cache);
4474 }
4475 
4476 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4477                              struct btrfs_fs_info *fs_info)
4478 {
4479         struct btrfs_block_group_cache *cache;
4480 
4481         spin_lock(&cur_trans->dirty_bgs_lock);
4482         while (!list_empty(&cur_trans->dirty_bgs)) {
4483                 cache = list_first_entry(&cur_trans->dirty_bgs,
4484                                          struct btrfs_block_group_cache,
4485                                          dirty_list);
4486 
4487                 if (!list_empty(&cache->io_list)) {
4488                         spin_unlock(&cur_trans->dirty_bgs_lock);
4489                         list_del_init(&cache->io_list);
4490                         btrfs_cleanup_bg_io(cache);
4491                         spin_lock(&cur_trans->dirty_bgs_lock);
4492                 }
4493 
4494                 list_del_init(&cache->dirty_list);
4495                 spin_lock(&cache->lock);
4496                 cache->disk_cache_state = BTRFS_DC_ERROR;
4497                 spin_unlock(&cache->lock);
4498 
4499                 spin_unlock(&cur_trans->dirty_bgs_lock);
4500                 btrfs_put_block_group(cache);
4501                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4502                 spin_lock(&cur_trans->dirty_bgs_lock);
4503         }
4504         spin_unlock(&cur_trans->dirty_bgs_lock);
4505 
4506         /*
4507          * Refer to the definition of io_bgs member for details why it's safe
4508          * to use it without any locking
4509          */
4510         while (!list_empty(&cur_trans->io_bgs)) {
4511                 cache = list_first_entry(&cur_trans->io_bgs,
4512                                          struct btrfs_block_group_cache,
4513                                          io_list);
4514 
4515                 list_del_init(&cache->io_list);
4516                 spin_lock(&cache->lock);
4517                 cache->disk_cache_state = BTRFS_DC_ERROR;
4518                 spin_unlock(&cache->lock);
4519                 btrfs_cleanup_bg_io(cache);
4520         }
4521 }
4522 
4523 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4524                                    struct btrfs_fs_info *fs_info)
4525 {
4526         struct btrfs_device *dev, *tmp;
4527 
4528         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4529         ASSERT(list_empty(&cur_trans->dirty_bgs));
4530         ASSERT(list_empty(&cur_trans->io_bgs));
4531 
4532         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4533                                  post_commit_list) {
4534                 list_del_init(&dev->post_commit_list);
4535         }
4536 
4537         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4538 
4539         cur_trans->state = TRANS_STATE_COMMIT_START;
4540         wake_up(&fs_info->transaction_blocked_wait);
4541 
4542         cur_trans->state = TRANS_STATE_UNBLOCKED;
4543         wake_up(&fs_info->transaction_wait);
4544 
4545         btrfs_destroy_delayed_inodes(fs_info);
4546 
4547         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4548                                      EXTENT_DIRTY);
4549         btrfs_destroy_pinned_extent(fs_info,
4550                                     fs_info->pinned_extents);
4551 
4552         cur_trans->state =TRANS_STATE_COMPLETED;
4553         wake_up(&cur_trans->commit_wait);
4554 }
4555 
4556 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4557 {
4558         struct btrfs_transaction *t;
4559 
4560         mutex_lock(&fs_info->transaction_kthread_mutex);
4561 
4562         spin_lock(&fs_info->trans_lock);
4563         while (!list_empty(&fs_info->trans_list)) {
4564                 t = list_first_entry(&fs_info->trans_list,
4565                                      struct btrfs_transaction, list);
4566                 if (t->state >= TRANS_STATE_COMMIT_START) {
4567                         refcount_inc(&t->use_count);
4568                         spin_unlock(&fs_info->trans_lock);
4569                         btrfs_wait_for_commit(fs_info, t->transid);
4570                         btrfs_put_transaction(t);
4571                         spin_lock(&fs_info->trans_lock);
4572                         continue;
4573                 }
4574                 if (t == fs_info->running_transaction) {
4575                         t->state = TRANS_STATE_COMMIT_DOING;
4576                         spin_unlock(&fs_info->trans_lock);
4577                         /*
4578                          * We wait for 0 num_writers since we don't hold a trans
4579                          * handle open currently for this transaction.
4580                          */
4581                         wait_event(t->writer_wait,
4582                                    atomic_read(&t->num_writers) == 0);
4583                 } else {
4584                         spin_unlock(&fs_info->trans_lock);
4585                 }
4586                 btrfs_cleanup_one_transaction(t, fs_info);
4587 
4588                 spin_lock(&fs_info->trans_lock);
4589                 if (t == fs_info->running_transaction)
4590                         fs_info->running_transaction = NULL;
4591                 list_del_init(&t->list);
4592                 spin_unlock(&fs_info->trans_lock);
4593 
4594                 btrfs_put_transaction(t);
4595                 trace_btrfs_transaction_commit(fs_info->tree_root);
4596                 spin_lock(&fs_info->trans_lock);
4597         }
4598         spin_unlock(&fs_info->trans_lock);
4599         btrfs_destroy_all_ordered_extents(fs_info);
4600         btrfs_destroy_delayed_inodes(fs_info);
4601         btrfs_assert_delayed_root_empty(fs_info);
4602         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4603         btrfs_destroy_all_delalloc_inodes(fs_info);
4604         mutex_unlock(&fs_info->transaction_kthread_mutex);
4605 
4606         return 0;
4607 }
4608 
4609 static const struct extent_io_ops btree_extent_io_ops = {
4610         /* mandatory callbacks */
4611         .submit_bio_hook = btree_submit_bio_hook,
4612         .readpage_end_io_hook = btree_readpage_end_io_hook,
4613 };

/* [<][>][^][v][top][bottom][index][help] */