fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/kthread.h>
  12 #include <linux/slab.h>
  13 #include <linux/migrate.h>
  14 #include <linux/ratelimit.h>
  15 #include <linux/uuid.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/crc32c.h>
  19 #include <linux/sched/mm.h>
  20 #include <asm/unaligned.h>
  21 #include <crypto/hash.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42 #include "block-group.h"
  43 #include "discard.h"
  44 #include "space-info.h"
  45
  46 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  47                                  BTRFS_HEADER_FLAG_RELOC |\
  48                                  BTRFS_SUPER_FLAG_ERROR |\
  49                                  BTRFS_SUPER_FLAG_SEEDING |\
  50                                  BTRFS_SUPER_FLAG_METADUMP |\
  51                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  52
  53 static const struct extent_io_ops btree_extent_io_ops;
  54 static void end_workqueue_fn(struct btrfs_work *work);
  55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  57                                       struct btrfs_fs_info *fs_info);
  58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  60                                         struct extent_io_tree *dirty_pages,
  61                                         int mark);
  62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  63                                        struct extent_io_tree *pinned_extents);
  64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  66
  67 /*
  68  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  69  * is complete.  This is used during reads to verify checksums, and it is used
  70  * by writes to insert metadata for new file extents after IO is complete.
  71  */
  72 struct btrfs_end_io_wq {
  73         struct bio *bio;
  74         bio_end_io_t *end_io;
  75         void *private;
  76         struct btrfs_fs_info *info;
  77         blk_status_t status;
  78         enum btrfs_wq_endio_type metadata;
  79         struct btrfs_work work;
  80 };
  81
  82 static struct kmem_cache *btrfs_end_io_wq_cache;
  83
  84 int __init btrfs_end_io_wq_init(void)
  85 {
  86         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  87                                         sizeof(struct btrfs_end_io_wq),
  88                                         0,
  89                                         SLAB_MEM_SPREAD,
  90                                         NULL);
  91         if (!btrfs_end_io_wq_cache)
  92                 return -ENOMEM;
  93         return 0;
  94 }
  95
  96 void __cold btrfs_end_io_wq_exit(void)
  97 {
  98         kmem_cache_destroy(btrfs_end_io_wq_cache);
  99 }
 100
 101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
 102 {
 103         if (fs_info->csum_shash)
 104                 crypto_free_shash(fs_info->csum_shash);
 105 }
 106
 107 /*
 108  * async submit bios are used to offload expensive checksumming
 109  * onto the worker threads.  They checksum file and metadata bios
 110  * just before they are sent down the IO stack.
 111  */
 112 struct async_submit_bio {
 113         void *private_data;
 114         struct bio *bio;
 115         extent_submit_bio_start_t *submit_bio_start;
 116         int mirror_num;
 117         /*
 118          * bio_offset is optional, can be used if the pages in the bio
 119          * can't tell us where in the file the bio should go
 120          */
 121         u64 bio_offset;
 122         struct btrfs_work work;
 123         blk_status_t status;
 124 };
 125
 126 /*
 127  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 128  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 129  * the level the eb occupies in the tree.
 130  *
 131  * Different roots are used for different purposes and may nest inside each
 132  * other and they require separate keysets.  As lockdep keys should be
 133  * static, assign keysets according to the purpose of the root as indicated
 134  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 135  * roots have separate keysets.
 136  *
 137  * Lock-nesting across peer nodes is always done with the immediate parent
 138  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 139  * subclass to avoid triggering lockdep warning in such cases.
 140  *
 141  * The key is set by the readpage_end_io_hook after the buffer has passed
 142  * csum validation but before the pages are unlocked.  It is also set by
 143  * btrfs_init_new_buffer on freshly allocated blocks.
 144  *
 145  * We also add a check to make sure the highest level of the tree is the
 146  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 147  * needs update as well.
 148  */
 149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 150 # if BTRFS_MAX_LEVEL != 8
 151 #  error
 152 # endif
 153
 154 static struct btrfs_lockdep_keyset {
 155         u64                     id;             /* root objectid */
 156         const char              *name_stem;     /* lock name stem */
 157         char                    names[BTRFS_MAX_LEVEL + 1][20];
 158         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 159 } btrfs_lockdep_keysets[] = {
 160         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 161         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 162         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 163         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 164         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 165         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 166         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 167         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 168         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 169         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 170         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 171         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 172         { .id = 0,                              .name_stem = "tree"     },
 173 };
 174
 175 void __init btrfs_init_lockdep(void)
 176 {
 177         int i, j;
 178
 179         /* initialize lockdep class names */
 180         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 181                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 182
 183                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 184                         snprintf(ks->names[j], sizeof(ks->names[j]),
 185                                  "btrfs-%s-%02d", ks->name_stem, j);
 186         }
 187 }
 188
 189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 190                                     int level)
 191 {
 192         struct btrfs_lockdep_keyset *ks;
 193
 194         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 195
 196         /* find the matching keyset, id 0 is the default entry */
 197         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 198                 if (ks->id == objectid)
 199                         break;
 200
 201         lockdep_set_class_and_name(&eb->lock,
 202                                    &ks->keys[level], ks->names[level]);
 203 }
 204
 205 #endif
 206
 207 /*
 208  * extents on the btree inode are pretty simple, there's one extent
 209  * that covers the entire device
 210  */
 211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 212                                     struct page *page, size_t pg_offset,
 213                                     u64 start, u64 len)
 214 {
 215         struct extent_map_tree *em_tree = &inode->extent_tree;
 216         struct extent_map *em;
 217         int ret;
 218
 219         read_lock(&em_tree->lock);
 220         em = lookup_extent_mapping(em_tree, start, len);
 221         if (em) {
 222                 read_unlock(&em_tree->lock);
 223                 goto out;
 224         }
 225         read_unlock(&em_tree->lock);
 226
 227         em = alloc_extent_map();
 228         if (!em) {
 229                 em = ERR_PTR(-ENOMEM);
 230                 goto out;
 231         }
 232         em->start = 0;
 233         em->len = (u64)-1;
 234         em->block_len = (u64)-1;
 235         em->block_start = 0;
 236
 237         write_lock(&em_tree->lock);
 238         ret = add_extent_mapping(em_tree, em, 0);
 239         if (ret == -EEXIST) {
 240                 free_extent_map(em);
 241                 em = lookup_extent_mapping(em_tree, start, len);
 242                 if (!em)
 243                         em = ERR_PTR(-EIO);
 244         } else if (ret) {
 245                 free_extent_map(em);
 246                 em = ERR_PTR(ret);
 247         }
 248         write_unlock(&em_tree->lock);
 249
 250 out:
 251         return em;
 252 }
 253
 254 /*
 255  * Compute the csum of a btree block and store the result to provided buffer.
 256  */
 257 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
 258 {
 259         struct btrfs_fs_info *fs_info = buf->fs_info;
 260         const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
 261         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 262         char *kaddr;
 263         int i;
 264
 265         shash->tfm = fs_info->csum_shash;
 266         crypto_shash_init(shash);
 267         kaddr = page_address(buf->pages[0]);
 268         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
 269                             PAGE_SIZE - BTRFS_CSUM_SIZE);
 270
 271         for (i = 1; i < num_pages; i++) {
 272                 kaddr = page_address(buf->pages[i]);
 273                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 274         }
 275         memset(result, 0, BTRFS_CSUM_SIZE);
 276         crypto_shash_final(shash, result);
 277 }
 278
 279 /*
 280  * we can't consider a given block up to date unless the transid of the
 281  * block matches the transid in the parent node's pointer.  This is how we
 282  * detect blocks that either didn't get written at all or got written
 283  * in the wrong place.
 284  */
 285 static int verify_parent_transid(struct extent_io_tree *io_tree,
 286                                  struct extent_buffer *eb, u64 parent_transid,
 287                                  int atomic)
 288 {
 289         struct extent_state *cached_state = NULL;
 290         int ret;
 291         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 292
 293         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 294                 return 0;
 295
 296         if (atomic)
 297                 return -EAGAIN;
 298
 299         if (need_lock) {
 300                 btrfs_tree_read_lock(eb);
 301                 btrfs_set_lock_blocking_read(eb);
 302         }
 303
 304         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 305                          &cached_state);
 306         if (extent_buffer_uptodate(eb) &&
 307             btrfs_header_generation(eb) == parent_transid) {
 308                 ret = 0;
 309                 goto out;
 310         }
 311         btrfs_err_rl(eb->fs_info,
 312                 "parent transid verify failed on %llu wanted %llu found %llu",
 313                         eb->start,
 314                         parent_transid, btrfs_header_generation(eb));
 315         ret = 1;
 316
 317         /*
 318          * Things reading via commit roots that don't have normal protection,
 319          * like send, can have a really old block in cache that may point at a
 320          * block that has been freed and re-allocated.  So don't clear uptodate
 321          * if we find an eb that is under IO (dirty/writeback) because we could
 322          * end up reading in the stale data and then writing it back out and
 323          * making everybody very sad.
 324          */
 325         if (!extent_buffer_under_io(eb))
 326                 clear_extent_buffer_uptodate(eb);
 327 out:
 328         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 329                              &cached_state);
 330         if (need_lock)
 331                 btrfs_tree_read_unlock_blocking(eb);
 332         return ret;
 333 }
 334
 335 static bool btrfs_supported_super_csum(u16 csum_type)
 336 {
 337         switch (csum_type) {
 338         case BTRFS_CSUM_TYPE_CRC32:
 339         case BTRFS_CSUM_TYPE_XXHASH:
 340         case BTRFS_CSUM_TYPE_SHA256:
 341         case BTRFS_CSUM_TYPE_BLAKE2:
 342                 return true;
 343         default:
 344                 return false;
 345         }
 346 }
 347
 348 /*
 349  * Return 0 if the superblock checksum type matches the checksum value of that
 350  * algorithm. Pass the raw disk superblock data.
 351  */
 352 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 353                                   char *raw_disk_sb)
 354 {
 355         struct btrfs_super_block *disk_sb =
 356                 (struct btrfs_super_block *)raw_disk_sb;
 357         char result[BTRFS_CSUM_SIZE];
 358         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 359
 360         shash->tfm = fs_info->csum_shash;
 361         crypto_shash_init(shash);
 362
 363         /*
 364          * The super_block structure does not span the whole
 365          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 366          * filled with zeros and is included in the checksum.
 367          */
 368         crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
 369                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 370         crypto_shash_final(shash, result);
 371
 372         if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
 373                 return 1;
 374
 375         return 0;
 376 }
 377
 378 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 379                            struct btrfs_key *first_key, u64 parent_transid)
 380 {
 381         struct btrfs_fs_info *fs_info = eb->fs_info;
 382         int found_level;
 383         struct btrfs_key found_key;
 384         int ret;
 385
 386         found_level = btrfs_header_level(eb);
 387         if (found_level != level) {
 388                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 389                      KERN_ERR "BTRFS: tree level check failed\n");
 390                 btrfs_err(fs_info,
 391 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 392                           eb->start, level, found_level);
 393                 return -EIO;
 394         }
 395
 396         if (!first_key)
 397                 return 0;
 398
 399         /*
 400          * For live tree block (new tree blocks in current transaction),
 401          * we need proper lock context to avoid race, which is impossible here.
 402          * So we only checks tree blocks which is read from disk, whose
 403          * generation <= fs_info->last_trans_committed.
 404          */
 405         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 406                 return 0;
 407
 408         /* We have @first_key, so this @eb must have at least one item */
 409         if (btrfs_header_nritems(eb) == 0) {
 410                 btrfs_err(fs_info,
 411                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 412                           eb->start);
 413                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 414                 return -EUCLEAN;
 415         }
 416
 417         if (found_level)
 418                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 419         else
 420                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 421         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 422
 423         if (ret) {
 424                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 425                      KERN_ERR "BTRFS: tree first key check failed\n");
 426                 btrfs_err(fs_info,
 427 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 428                           eb->start, parent_transid, first_key->objectid,
 429                           first_key->type, first_key->offset,
 430                           found_key.objectid, found_key.type,
 431                           found_key.offset);
 432         }
 433         return ret;
 434 }
 435
 436 /*
 437  * helper to read a given tree block, doing retries as required when
 438  * the checksums don't match and we have alternate mirrors to try.
 439  *
 440  * @parent_transid:     expected transid, skip check if 0
 441  * @level:              expected level, mandatory check
 442  * @first_key:          expected key of first slot, skip check if NULL
 443  */
 444 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 445                                           u64 parent_transid, int level,
 446                                           struct btrfs_key *first_key)
 447 {
 448         struct btrfs_fs_info *fs_info = eb->fs_info;
 449         struct extent_io_tree *io_tree;
 450         int failed = 0;
 451         int ret;
 452         int num_copies = 0;
 453         int mirror_num = 0;
 454         int failed_mirror = 0;
 455
 456         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 457         while (1) {
 458                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 459                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 460                 if (!ret) {
 461                         if (verify_parent_transid(io_tree, eb,
 462                                                    parent_transid, 0))
 463                                 ret = -EIO;
 464                         else if (btrfs_verify_level_key(eb, level,
 465                                                 first_key, parent_transid))
 466                                 ret = -EUCLEAN;
 467                         else
 468                                 break;
 469                 }
 470
 471                 num_copies = btrfs_num_copies(fs_info,
 472                                               eb->start, eb->len);
 473                 if (num_copies == 1)
 474                         break;
 475
 476                 if (!failed_mirror) {
 477                         failed = 1;
 478                         failed_mirror = eb->read_mirror;
 479                 }
 480
 481                 mirror_num++;
 482                 if (mirror_num == failed_mirror)
 483                         mirror_num++;
 484
 485                 if (mirror_num > num_copies)
 486                         break;
 487         }
 488
 489         if (failed && !ret && failed_mirror)
 490                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 491
 492         return ret;
 493 }
 494
 495 /*
 496  * checksum a dirty tree block before IO.  This has extra checks to make sure
 497  * we only fill in the checksum field in the first page of a multi-page block
 498  */
 499
 500 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 501 {
 502         u64 start = page_offset(page);
 503         u64 found_start;
 504         u8 result[BTRFS_CSUM_SIZE];
 505         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 506         struct extent_buffer *eb;
 507         int ret;
 508
 509         eb = (struct extent_buffer *)page->private;
 510         if (page != eb->pages[0])
 511                 return 0;
 512
 513         found_start = btrfs_header_bytenr(eb);
 514         /*
 515          * Please do not consolidate these warnings into a single if.
 516          * It is useful to know what went wrong.
 517          */
 518         if (WARN_ON(found_start != start))
 519                 return -EUCLEAN;
 520         if (WARN_ON(!PageUptodate(page)))
 521                 return -EUCLEAN;
 522
 523         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 524                                     offsetof(struct btrfs_header, fsid),
 525                                     BTRFS_FSID_SIZE) == 0);
 526
 527         csum_tree_block(eb, result);
 528
 529         if (btrfs_header_level(eb))
 530                 ret = btrfs_check_node(eb);
 531         else
 532                 ret = btrfs_check_leaf_full(eb);
 533
 534         if (ret < 0) {
 535                 btrfs_print_tree(eb, 0);
 536                 btrfs_err(fs_info,
 537                 "block=%llu write time tree block corruption detected",
 538                           eb->start);
 539                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 540                 return ret;
 541         }
 542         write_extent_buffer(eb, result, 0, csum_size);
 543
 544         return 0;
 545 }
 546
 547 static int check_tree_block_fsid(struct extent_buffer *eb)
 548 {
 549         struct btrfs_fs_info *fs_info = eb->fs_info;
 550         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 551         u8 fsid[BTRFS_FSID_SIZE];
 552         int ret = 1;
 553
 554         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 555                            BTRFS_FSID_SIZE);
 556         while (fs_devices) {
 557                 u8 *metadata_uuid;
 558
 559                 /*
 560                  * Checking the incompat flag is only valid for the current
 561                  * fs. For seed devices it's forbidden to have their uuid
 562                  * changed so reading ->fsid in this case is fine
 563                  */
 564                 if (fs_devices == fs_info->fs_devices &&
 565                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 566                         metadata_uuid = fs_devices->metadata_uuid;
 567                 else
 568                         metadata_uuid = fs_devices->fsid;
 569
 570                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 571                         ret = 0;
 572                         break;
 573                 }
 574                 fs_devices = fs_devices->seed;
 575         }
 576         return ret;
 577 }
 578
 579 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 580                                       u64 phy_offset, struct page *page,
 581                                       u64 start, u64 end, int mirror)
 582 {
 583         u64 found_start;
 584         int found_level;
 585         struct extent_buffer *eb;
 586         struct btrfs_fs_info *fs_info;
 587         u16 csum_size;
 588         int ret = 0;
 589         u8 result[BTRFS_CSUM_SIZE];
 590         int reads_done;
 591
 592         if (!page->private)
 593                 goto out;
 594
 595         eb = (struct extent_buffer *)page->private;
 596         fs_info = eb->fs_info;
 597         csum_size = btrfs_super_csum_size(fs_info->super_copy);
 598
 599         /* the pending IO might have been the only thing that kept this buffer
 600          * in memory.  Make sure we have a ref for all this other checks
 601          */
 602         atomic_inc(&eb->refs);
 603
 604         reads_done = atomic_dec_and_test(&eb->io_pages);
 605         if (!reads_done)
 606                 goto err;
 607
 608         eb->read_mirror = mirror;
 609         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 610                 ret = -EIO;
 611                 goto err;
 612         }
 613
 614         found_start = btrfs_header_bytenr(eb);
 615         if (found_start != eb->start) {
 616                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 617                              eb->start, found_start);
 618                 ret = -EIO;
 619                 goto err;
 620         }
 621         if (check_tree_block_fsid(eb)) {
 622                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 623                              eb->start);
 624                 ret = -EIO;
 625                 goto err;
 626         }
 627         found_level = btrfs_header_level(eb);
 628         if (found_level >= BTRFS_MAX_LEVEL) {
 629                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 630                           (int)btrfs_header_level(eb), eb->start);
 631                 ret = -EIO;
 632                 goto err;
 633         }
 634
 635         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 636                                        eb, found_level);
 637
 638         csum_tree_block(eb, result);
 639
 640         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 641                 u32 val;
 642                 u32 found = 0;
 643
 644                 memcpy(&found, result, csum_size);
 645
 646                 read_extent_buffer(eb, &val, 0, csum_size);
 647                 btrfs_warn_rl(fs_info,
 648                 "%s checksum verify failed on %llu wanted %x found %x level %d",
 649                               fs_info->sb->s_id, eb->start,
 650                               val, found, btrfs_header_level(eb));
 651                 ret = -EUCLEAN;
 652                 goto err;
 653         }
 654
 655         /*
 656          * If this is a leaf block and it is corrupt, set the corrupt bit so
 657          * that we don't try and read the other copies of this block, just
 658          * return -EIO.
 659          */
 660         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 661                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 662                 ret = -EIO;
 663         }
 664
 665         if (found_level > 0 && btrfs_check_node(eb))
 666                 ret = -EIO;
 667
 668         if (!ret)
 669                 set_extent_buffer_uptodate(eb);
 670         else
 671                 btrfs_err(fs_info,
 672                           "block=%llu read time tree block corruption detected",
 673                           eb->start);
 674 err:
 675         if (reads_done &&
 676             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 677                 btree_readahead_hook(eb, ret);
 678
 679         if (ret) {
 680                 /*
 681                  * our io error hook is going to dec the io pages
 682                  * again, we have to make sure it has something
 683                  * to decrement
 684                  */
 685                 atomic_inc(&eb->io_pages);
 686                 clear_extent_buffer_uptodate(eb);
 687         }
 688         free_extent_buffer(eb);
 689 out:
 690         return ret;
 691 }
 692
 693 static void end_workqueue_bio(struct bio *bio)
 694 {
 695         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 696         struct btrfs_fs_info *fs_info;
 697         struct btrfs_workqueue *wq;
 698
 699         fs_info = end_io_wq->info;
 700         end_io_wq->status = bio->bi_status;
 701
 702         if (bio_op(bio) == REQ_OP_WRITE) {
 703                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 704                         wq = fs_info->endio_meta_write_workers;
 705                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 706                         wq = fs_info->endio_freespace_worker;
 707                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 708                         wq = fs_info->endio_raid56_workers;
 709                 else
 710                         wq = fs_info->endio_write_workers;
 711         } else {
 712                 if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
 713                         wq = fs_info->endio_repair_workers;
 714                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 715                         wq = fs_info->endio_raid56_workers;
 716                 else if (end_io_wq->metadata)
 717                         wq = fs_info->endio_meta_workers;
 718                 else
 719                         wq = fs_info->endio_workers;
 720         }
 721
 722         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 723         btrfs_queue_work(wq, &end_io_wq->work);
 724 }
 725
 726 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 727                         enum btrfs_wq_endio_type metadata)
 728 {
 729         struct btrfs_end_io_wq *end_io_wq;
 730
 731         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 732         if (!end_io_wq)
 733                 return BLK_STS_RESOURCE;
 734
 735         end_io_wq->private = bio->bi_private;
 736         end_io_wq->end_io = bio->bi_end_io;
 737         end_io_wq->info = info;
 738         end_io_wq->status = 0;
 739         end_io_wq->bio = bio;
 740         end_io_wq->metadata = metadata;
 741
 742         bio->bi_private = end_io_wq;
 743         bio->bi_end_io = end_workqueue_bio;
 744         return 0;
 745 }
 746
 747 static void run_one_async_start(struct btrfs_work *work)
 748 {
 749         struct async_submit_bio *async;
 750         blk_status_t ret;
 751
 752         async = container_of(work, struct  async_submit_bio, work);
 753         ret = async->submit_bio_start(async->private_data, async->bio,
 754                                       async->bio_offset);
 755         if (ret)
 756                 async->status = ret;
 757 }
 758
 759 /*
 760  * In order to insert checksums into the metadata in large chunks, we wait
 761  * until bio submission time.   All the pages in the bio are checksummed and
 762  * sums are attached onto the ordered extent record.
 763  *
 764  * At IO completion time the csums attached on the ordered extent record are
 765  * inserted into the tree.
 766  */
 767 static void run_one_async_done(struct btrfs_work *work)
 768 {
 769         struct async_submit_bio *async;
 770         struct inode *inode;
 771         blk_status_t ret;
 772
 773         async = container_of(work, struct  async_submit_bio, work);
 774         inode = async->private_data;
 775
 776         /* If an error occurred we just want to clean up the bio and move on */
 777         if (async->status) {
 778                 async->bio->bi_status = async->status;
 779                 bio_endio(async->bio);
 780                 return;
 781         }
 782
 783         /*
 784          * All of the bios that pass through here are from async helpers.
 785          * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
 786          * This changes nothing when cgroups aren't in use.
 787          */
 788         async->bio->bi_opf |= REQ_CGROUP_PUNT;
 789         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
 790         if (ret) {
 791                 async->bio->bi_status = ret;
 792                 bio_endio(async->bio);
 793         }
 794 }
 795
 796 static void run_one_async_free(struct btrfs_work *work)
 797 {
 798         struct async_submit_bio *async;
 799
 800         async = container_of(work, struct  async_submit_bio, work);
 801         kfree(async);
 802 }
 803
 804 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 805                                  int mirror_num, unsigned long bio_flags,
 806                                  u64 bio_offset, void *private_data,
 807                                  extent_submit_bio_start_t *submit_bio_start)
 808 {
 809         struct async_submit_bio *async;
 810
 811         async = kmalloc(sizeof(*async), GFP_NOFS);
 812         if (!async)
 813                 return BLK_STS_RESOURCE;
 814
 815         async->private_data = private_data;
 816         async->bio = bio;
 817         async->mirror_num = mirror_num;
 818         async->submit_bio_start = submit_bio_start;
 819
 820         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 821                         run_one_async_free);
 822
 823         async->bio_offset = bio_offset;
 824
 825         async->status = 0;
 826
 827         if (op_is_sync(bio->bi_opf))
 828                 btrfs_set_work_high_priority(&async->work);
 829
 830         btrfs_queue_work(fs_info->workers, &async->work);
 831         return 0;
 832 }
 833
 834 static blk_status_t btree_csum_one_bio(struct bio *bio)
 835 {
 836         struct bio_vec *bvec;
 837         struct btrfs_root *root;
 838         int ret = 0;
 839         struct bvec_iter_all iter_all;
 840
 841         ASSERT(!bio_flagged(bio, BIO_CLONED));
 842         bio_for_each_segment_all(bvec, bio, iter_all) {
 843                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 844                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 845                 if (ret)
 846                         break;
 847         }
 848
 849         return errno_to_blk_status(ret);
 850 }
 851
 852 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 853                                              u64 bio_offset)
 854 {
 855         /*
 856          * when we're called for a write, we're already in the async
 857          * submission context.  Just jump into btrfs_map_bio
 858          */
 859         return btree_csum_one_bio(bio);
 860 }
 861
 862 static int check_async_write(struct btrfs_fs_info *fs_info,
 863                              struct btrfs_inode *bi)
 864 {
 865         if (atomic_read(&bi->sync_writers))
 866                 return 0;
 867         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 868                 return 0;
 869         return 1;
 870 }
 871
 872 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 873                                           int mirror_num,
 874                                           unsigned long bio_flags)
 875 {
 876         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 877         int async = check_async_write(fs_info, BTRFS_I(inode));
 878         blk_status_t ret;
 879
 880         if (bio_op(bio) != REQ_OP_WRITE) {
 881                 /*
 882                  * called for a read, do the setup so that checksum validation
 883                  * can happen in the async kernel threads
 884                  */
 885                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 886                                           BTRFS_WQ_ENDIO_METADATA);
 887                 if (ret)
 888                         goto out_w_error;
 889                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 890         } else if (!async) {
 891                 ret = btree_csum_one_bio(bio);
 892                 if (ret)
 893                         goto out_w_error;
 894                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 895         } else {
 896                 /*
 897                  * kthread helpers are used to submit writes so that
 898                  * checksumming can happen in parallel across all CPUs
 899                  */
 900                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 901                                           0, inode, btree_submit_bio_start);
 902         }
 903
 904         if (ret)
 905                 goto out_w_error;
 906         return 0;
 907
 908 out_w_error:
 909         bio->bi_status = ret;
 910         bio_endio(bio);
 911         return ret;
 912 }
 913
 914 #ifdef CONFIG_MIGRATION
 915 static int btree_migratepage(struct address_space *mapping,
 916                         struct page *newpage, struct page *page,
 917                         enum migrate_mode mode)
 918 {
 919         /*
 920          * we can't safely write a btree page from here,
 921          * we haven't done the locking hook
 922          */
 923         if (PageDirty(page))
 924                 return -EAGAIN;
 925         /*
 926          * Buffers may be managed in a filesystem specific way.
 927          * We must have no buffers or drop them.
 928          */
 929         if (page_has_private(page) &&
 930             !try_to_release_page(page, GFP_KERNEL))
 931                 return -EAGAIN;
 932         return migrate_page(mapping, newpage, page, mode);
 933 }
 934 #endif
 935
 936
 937 static int btree_writepages(struct address_space *mapping,
 938                             struct writeback_control *wbc)
 939 {
 940         struct btrfs_fs_info *fs_info;
 941         int ret;
 942
 943         if (wbc->sync_mode == WB_SYNC_NONE) {
 944
 945                 if (wbc->for_kupdate)
 946                         return 0;
 947
 948                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 949                 /* this is a bit racy, but that's ok */
 950                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 951                                              BTRFS_DIRTY_METADATA_THRESH,
 952                                              fs_info->dirty_metadata_batch);
 953                 if (ret < 0)
 954                         return 0;
 955         }
 956         return btree_write_cache_pages(mapping, wbc);
 957 }
 958
 959 static int btree_readpage(struct file *file, struct page *page)
 960 {
 961         return extent_read_full_page(page, btree_get_extent, 0);
 962 }
 963
 964 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 965 {
 966         if (PageWriteback(page) || PageDirty(page))
 967                 return 0;
 968
 969         return try_release_extent_buffer(page);
 970 }
 971
 972 static void btree_invalidatepage(struct page *page, unsigned int offset,
 973                                  unsigned int length)
 974 {
 975         struct extent_io_tree *tree;
 976         tree = &BTRFS_I(page->mapping->host)->io_tree;
 977         extent_invalidatepage(tree, page, offset);
 978         btree_releasepage(page, GFP_NOFS);
 979         if (PagePrivate(page)) {
 980                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 981                            "page private not zero on page %llu",
 982                            (unsigned long long)page_offset(page));
 983                 ClearPagePrivate(page);
 984                 set_page_private(page, 0);
 985                 put_page(page);
 986         }
 987 }
 988
 989 static int btree_set_page_dirty(struct page *page)
 990 {
 991 #ifdef DEBUG
 992         struct extent_buffer *eb;
 993
 994         BUG_ON(!PagePrivate(page));
 995         eb = (struct extent_buffer *)page->private;
 996         BUG_ON(!eb);
 997         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 998         BUG_ON(!atomic_read(&eb->refs));
 999         btrfs_assert_tree_locked(eb);
1000 #endif
1001         return __set_page_dirty_nobuffers(page);
1002 }
1003
1004 static const struct address_space_operations btree_aops = {
1005         .readpage       = btree_readpage,
1006         .writepages     = btree_writepages,
1007         .releasepage    = btree_releasepage,
1008         .invalidatepage = btree_invalidatepage,
1009 #ifdef CONFIG_MIGRATION
1010         .migratepage    = btree_migratepage,
1011 #endif
1012         .set_page_dirty = btree_set_page_dirty,
1013 };
1014
1015 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1016 {
1017         struct extent_buffer *buf = NULL;
1018         int ret;
1019
1020         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1021         if (IS_ERR(buf))
1022                 return;
1023
1024         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1025         if (ret < 0)
1026                 free_extent_buffer_stale(buf);
1027         else
1028                 free_extent_buffer(buf);
1029 }
1030
1031 struct extent_buffer *btrfs_find_create_tree_block(
1032                                                 struct btrfs_fs_info *fs_info,
1033                                                 u64 bytenr)
1034 {
1035         if (btrfs_is_testing(fs_info))
1036                 return alloc_test_extent_buffer(fs_info, bytenr);
1037         return alloc_extent_buffer(fs_info, bytenr);
1038 }
1039
1040 /*
1041  * Read tree block at logical address @bytenr and do variant basic but critical
1042  * verification.
1043  *
1044  * @parent_transid:     expected transid of this tree block, skip check if 0
1045  * @level:              expected level, mandatory check
1046  * @first_key:          expected key in slot 0, skip check if NULL
1047  */
1048 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1049                                       u64 parent_transid, int level,
1050                                       struct btrfs_key *first_key)
1051 {
1052         struct extent_buffer *buf = NULL;
1053         int ret;
1054
1055         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1056         if (IS_ERR(buf))
1057                 return buf;
1058
1059         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1060                                              level, first_key);
1061         if (ret) {
1062                 free_extent_buffer_stale(buf);
1063                 return ERR_PTR(ret);
1064         }
1065         return buf;
1066
1067 }
1068
1069 void btrfs_clean_tree_block(struct extent_buffer *buf)
1070 {
1071         struct btrfs_fs_info *fs_info = buf->fs_info;
1072         if (btrfs_header_generation(buf) ==
1073             fs_info->running_transaction->transid) {
1074                 btrfs_assert_tree_locked(buf);
1075
1076                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1077                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1078                                                  -buf->len,
1079                                                  fs_info->dirty_metadata_batch);
1080                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1081                         btrfs_set_lock_blocking_write(buf);
1082                         clear_extent_buffer_dirty(buf);
1083                 }
1084         }
1085 }
1086
1087 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1088                          u64 objectid)
1089 {
1090         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1091         root->fs_info = fs_info;
1092         root->node = NULL;
1093         root->commit_root = NULL;
1094         root->state = 0;
1095         root->orphan_cleanup_state = 0;
1096
1097         root->last_trans = 0;
1098         root->highest_objectid = 0;
1099         root->nr_delalloc_inodes = 0;
1100         root->nr_ordered_extents = 0;
1101         root->inode_tree = RB_ROOT;
1102         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1103         root->block_rsv = NULL;
1104
1105         INIT_LIST_HEAD(&root->dirty_list);
1106         INIT_LIST_HEAD(&root->root_list);
1107         INIT_LIST_HEAD(&root->delalloc_inodes);
1108         INIT_LIST_HEAD(&root->delalloc_root);
1109         INIT_LIST_HEAD(&root->ordered_extents);
1110         INIT_LIST_HEAD(&root->ordered_root);
1111         INIT_LIST_HEAD(&root->reloc_dirty_list);
1112         INIT_LIST_HEAD(&root->logged_list[0]);
1113         INIT_LIST_HEAD(&root->logged_list[1]);
1114         spin_lock_init(&root->inode_lock);
1115         spin_lock_init(&root->delalloc_lock);
1116         spin_lock_init(&root->ordered_extent_lock);
1117         spin_lock_init(&root->accounting_lock);
1118         spin_lock_init(&root->log_extents_lock[0]);
1119         spin_lock_init(&root->log_extents_lock[1]);
1120         spin_lock_init(&root->qgroup_meta_rsv_lock);
1121         mutex_init(&root->objectid_mutex);
1122         mutex_init(&root->log_mutex);
1123         mutex_init(&root->ordered_extent_mutex);
1124         mutex_init(&root->delalloc_mutex);
1125         init_waitqueue_head(&root->log_writer_wait);
1126         init_waitqueue_head(&root->log_commit_wait[0]);
1127         init_waitqueue_head(&root->log_commit_wait[1]);
1128         INIT_LIST_HEAD(&root->log_ctxs[0]);
1129         INIT_LIST_HEAD(&root->log_ctxs[1]);
1130         atomic_set(&root->log_commit[0], 0);
1131         atomic_set(&root->log_commit[1], 0);
1132         atomic_set(&root->log_writers, 0);
1133         atomic_set(&root->log_batch, 0);
1134         refcount_set(&root->refs, 1);
1135         atomic_set(&root->snapshot_force_cow, 0);
1136         atomic_set(&root->nr_swapfiles, 0);
1137         root->log_transid = 0;
1138         root->log_transid_committed = -1;
1139         root->last_log_commit = 0;
1140         if (!dummy)
1141                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1142                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1143
1144         memset(&root->root_key, 0, sizeof(root->root_key));
1145         memset(&root->root_item, 0, sizeof(root->root_item));
1146         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1147         if (!dummy)
1148                 root->defrag_trans_start = fs_info->generation;
1149         else
1150                 root->defrag_trans_start = 0;
1151         root->root_key.objectid = objectid;
1152         root->anon_dev = 0;
1153
1154         spin_lock_init(&root->root_item_lock);
1155         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1156 #ifdef CONFIG_BTRFS_DEBUG
1157         INIT_LIST_HEAD(&root->leak_list);
1158         spin_lock(&fs_info->fs_roots_radix_lock);
1159         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1160         spin_unlock(&fs_info->fs_roots_radix_lock);
1161 #endif
1162 }
1163
1164 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1165                                            u64 objectid, gfp_t flags)
1166 {
1167         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1168         if (root)
1169                 __setup_root(root, fs_info, objectid);
1170         return root;
1171 }
1172
1173 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1174 /* Should only be used by the testing infrastructure */
1175 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1176 {
1177         struct btrfs_root *root;
1178
1179         if (!fs_info)
1180                 return ERR_PTR(-EINVAL);
1181
1182         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1183         if (!root)
1184                 return ERR_PTR(-ENOMEM);
1185
1186         /* We don't use the stripesize in selftest, set it as sectorsize */
1187         root->alloc_bytenr = 0;
1188
1189         return root;
1190 }
1191 #endif
1192
1193 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1194                                      u64 objectid)
1195 {
1196         struct btrfs_fs_info *fs_info = trans->fs_info;
1197         struct extent_buffer *leaf;
1198         struct btrfs_root *tree_root = fs_info->tree_root;
1199         struct btrfs_root *root;
1200         struct btrfs_key key;
1201         unsigned int nofs_flag;
1202         int ret = 0;
1203
1204         /*
1205          * We're holding a transaction handle, so use a NOFS memory allocation
1206          * context to avoid deadlock if reclaim happens.
1207          */
1208         nofs_flag = memalloc_nofs_save();
1209         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1210         memalloc_nofs_restore(nofs_flag);
1211         if (!root)
1212                 return ERR_PTR(-ENOMEM);
1213
1214         root->root_key.objectid = objectid;
1215         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1216         root->root_key.offset = 0;
1217
1218         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1219         if (IS_ERR(leaf)) {
1220                 ret = PTR_ERR(leaf);
1221                 leaf = NULL;
1222                 goto fail;
1223         }
1224
1225         root->node = leaf;
1226         btrfs_mark_buffer_dirty(leaf);
1227
1228         root->commit_root = btrfs_root_node(root);
1229         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1230
1231         root->root_item.flags = 0;
1232         root->root_item.byte_limit = 0;
1233         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1234         btrfs_set_root_generation(&root->root_item, trans->transid);
1235         btrfs_set_root_level(&root->root_item, 0);
1236         btrfs_set_root_refs(&root->root_item, 1);
1237         btrfs_set_root_used(&root->root_item, leaf->len);
1238         btrfs_set_root_last_snapshot(&root->root_item, 0);
1239         btrfs_set_root_dirid(&root->root_item, 0);
1240         if (is_fstree(objectid))
1241                 generate_random_guid(root->root_item.uuid);
1242         else
1243                 export_guid(root->root_item.uuid, &guid_null);
1244         root->root_item.drop_level = 0;
1245
1246         key.objectid = objectid;
1247         key.type = BTRFS_ROOT_ITEM_KEY;
1248         key.offset = 0;
1249         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1250         if (ret)
1251                 goto fail;
1252
1253         btrfs_tree_unlock(leaf);
1254
1255         return root;
1256
1257 fail:
1258         if (leaf)
1259                 btrfs_tree_unlock(leaf);
1260         btrfs_put_root(root);
1261
1262         return ERR_PTR(ret);
1263 }
1264
1265 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1266                                          struct btrfs_fs_info *fs_info)
1267 {
1268         struct btrfs_root *root;
1269         struct extent_buffer *leaf;
1270
1271         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1272         if (!root)
1273                 return ERR_PTR(-ENOMEM);
1274
1275         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1276         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1278
1279         /*
1280          * DON'T set REF_COWS for log trees
1281          *
1282          * log trees do not get reference counted because they go away
1283          * before a real commit is actually done.  They do store pointers
1284          * to file data extents, and those reference counts still get
1285          * updated (along with back refs to the log tree).
1286          */
1287
1288         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1289                         NULL, 0, 0, 0);
1290         if (IS_ERR(leaf)) {
1291                 btrfs_put_root(root);
1292                 return ERR_CAST(leaf);
1293         }
1294
1295         root->node = leaf;
1296
1297         btrfs_mark_buffer_dirty(root->node);
1298         btrfs_tree_unlock(root->node);
1299         return root;
1300 }
1301
1302 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1303                              struct btrfs_fs_info *fs_info)
1304 {
1305         struct btrfs_root *log_root;
1306
1307         log_root = alloc_log_tree(trans, fs_info);
1308         if (IS_ERR(log_root))
1309                 return PTR_ERR(log_root);
1310         WARN_ON(fs_info->log_root_tree);
1311         fs_info->log_root_tree = log_root;
1312         return 0;
1313 }
1314
1315 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1316                        struct btrfs_root *root)
1317 {
1318         struct btrfs_fs_info *fs_info = root->fs_info;
1319         struct btrfs_root *log_root;
1320         struct btrfs_inode_item *inode_item;
1321
1322         log_root = alloc_log_tree(trans, fs_info);
1323         if (IS_ERR(log_root))
1324                 return PTR_ERR(log_root);
1325
1326         log_root->last_trans = trans->transid;
1327         log_root->root_key.offset = root->root_key.objectid;
1328
1329         inode_item = &log_root->root_item.inode;
1330         btrfs_set_stack_inode_generation(inode_item, 1);
1331         btrfs_set_stack_inode_size(inode_item, 3);
1332         btrfs_set_stack_inode_nlink(inode_item, 1);
1333         btrfs_set_stack_inode_nbytes(inode_item,
1334                                      fs_info->nodesize);
1335         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1336
1337         btrfs_set_root_node(&log_root->root_item, log_root->node);
1338
1339         WARN_ON(root->log_root);
1340         root->log_root = log_root;
1341         root->log_transid = 0;
1342         root->log_transid_committed = -1;
1343         root->last_log_commit = 0;
1344         return 0;
1345 }
1346
1347 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1348                                         struct btrfs_key *key)
1349 {
1350         struct btrfs_root *root;
1351         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1352         struct btrfs_path *path;
1353         u64 generation;
1354         int ret;
1355         int level;
1356
1357         path = btrfs_alloc_path();
1358         if (!path)
1359                 return ERR_PTR(-ENOMEM);
1360
1361         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1362         if (!root) {
1363                 ret = -ENOMEM;
1364                 goto alloc_fail;
1365         }
1366
1367         ret = btrfs_find_root(tree_root, key, path,
1368                               &root->root_item, &root->root_key);
1369         if (ret) {
1370                 if (ret > 0)
1371                         ret = -ENOENT;
1372                 goto find_fail;
1373         }
1374
1375         generation = btrfs_root_generation(&root->root_item);
1376         level = btrfs_root_level(&root->root_item);
1377         root->node = read_tree_block(fs_info,
1378                                      btrfs_root_bytenr(&root->root_item),
1379                                      generation, level, NULL);
1380         if (IS_ERR(root->node)) {
1381                 ret = PTR_ERR(root->node);
1382                 root->node = NULL;
1383                 goto find_fail;
1384         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1385                 ret = -EIO;
1386                 goto find_fail;
1387         }
1388         root->commit_root = btrfs_root_node(root);
1389 out:
1390         btrfs_free_path(path);
1391         return root;
1392
1393 find_fail:
1394         btrfs_put_root(root);
1395 alloc_fail:
1396         root = ERR_PTR(ret);
1397         goto out;
1398 }
1399
1400 static int btrfs_init_fs_root(struct btrfs_root *root)
1401 {
1402         int ret;
1403         unsigned int nofs_flag;
1404
1405         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1406         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1407                                         GFP_NOFS);
1408         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1409                 ret = -ENOMEM;
1410                 goto fail;
1411         }
1412
1413         /*
1414          * We might be called under a transaction (e.g. indirect backref
1415          * resolution) which could deadlock if it triggers memory reclaim
1416          */
1417         nofs_flag = memalloc_nofs_save();
1418         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1419         memalloc_nofs_restore(nofs_flag);
1420         if (ret)
1421                 goto fail;
1422
1423         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1424                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1425                 btrfs_check_and_init_root_item(&root->root_item);
1426         }
1427
1428         btrfs_init_free_ino_ctl(root);
1429         spin_lock_init(&root->ino_cache_lock);
1430         init_waitqueue_head(&root->ino_cache_wait);
1431
1432         ret = get_anon_bdev(&root->anon_dev);
1433         if (ret)
1434                 goto fail;
1435
1436         mutex_lock(&root->objectid_mutex);
1437         ret = btrfs_find_highest_objectid(root,
1438                                         &root->highest_objectid);
1439         if (ret) {
1440                 mutex_unlock(&root->objectid_mutex);
1441                 goto fail;
1442         }
1443
1444         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1445
1446         mutex_unlock(&root->objectid_mutex);
1447
1448         return 0;
1449 fail:
1450         /* The caller is responsible to call btrfs_free_fs_root */
1451         return ret;
1452 }
1453
1454 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1455                                                u64 root_id)
1456 {
1457         struct btrfs_root *root;
1458
1459         spin_lock(&fs_info->fs_roots_radix_lock);
1460         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1461                                  (unsigned long)root_id);
1462         if (root)
1463                 root = btrfs_grab_root(root);
1464         spin_unlock(&fs_info->fs_roots_radix_lock);
1465         return root;
1466 }
1467
1468 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1469                          struct btrfs_root *root)
1470 {
1471         int ret;
1472
1473         ret = radix_tree_preload(GFP_NOFS);
1474         if (ret)
1475                 return ret;
1476
1477         spin_lock(&fs_info->fs_roots_radix_lock);
1478         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1479                                 (unsigned long)root->root_key.objectid,
1480                                 root);
1481         if (ret == 0) {
1482                 btrfs_grab_root(root);
1483                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1484         }
1485         spin_unlock(&fs_info->fs_roots_radix_lock);
1486         radix_tree_preload_end();
1487
1488         return ret;
1489 }
1490
1491 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1492 {
1493 #ifdef CONFIG_BTRFS_DEBUG
1494         struct btrfs_root *root;
1495
1496         while (!list_empty(&fs_info->allocated_roots)) {
1497                 root = list_first_entry(&fs_info->allocated_roots,
1498                                         struct btrfs_root, leak_list);
1499                 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1500                           root->root_key.objectid, root->root_key.offset,
1501                           refcount_read(&root->refs));
1502                 while (refcount_read(&root->refs) > 1)
1503                         btrfs_put_root(root);
1504                 btrfs_put_root(root);
1505         }
1506 #endif
1507 }
1508
1509 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1510 {
1511         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1512         percpu_counter_destroy(&fs_info->delalloc_bytes);
1513         percpu_counter_destroy(&fs_info->dio_bytes);
1514         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1515         btrfs_free_csum_hash(fs_info);
1516         btrfs_free_stripe_hash_table(fs_info);
1517         btrfs_free_ref_cache(fs_info);
1518         kfree(fs_info->balance_ctl);
1519         kfree(fs_info->delayed_root);
1520         btrfs_put_root(fs_info->extent_root);
1521         btrfs_put_root(fs_info->tree_root);
1522         btrfs_put_root(fs_info->chunk_root);
1523         btrfs_put_root(fs_info->dev_root);
1524         btrfs_put_root(fs_info->csum_root);
1525         btrfs_put_root(fs_info->quota_root);
1526         btrfs_put_root(fs_info->uuid_root);
1527         btrfs_put_root(fs_info->free_space_root);
1528         btrfs_put_root(fs_info->fs_root);
1529         btrfs_check_leaked_roots(fs_info);
1530         btrfs_extent_buffer_leak_debug_check(fs_info);
1531         kfree(fs_info->super_copy);
1532         kfree(fs_info->super_for_commit);
1533         kvfree(fs_info);
1534 }
1535
1536
1537 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1538                                      struct btrfs_key *location,
1539                                      bool check_ref)
1540 {
1541         struct btrfs_root *root;
1542         struct btrfs_path *path;
1543         struct btrfs_key key;
1544         int ret;
1545
1546         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1547                 return btrfs_grab_root(fs_info->tree_root);
1548         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1549                 return btrfs_grab_root(fs_info->extent_root);
1550         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1551                 return btrfs_grab_root(fs_info->chunk_root);
1552         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1553                 return btrfs_grab_root(fs_info->dev_root);
1554         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1555                 return btrfs_grab_root(fs_info->csum_root);
1556         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1557                 return btrfs_grab_root(fs_info->quota_root) ?
1558                         fs_info->quota_root : ERR_PTR(-ENOENT);
1559         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1560                 return btrfs_grab_root(fs_info->uuid_root) ?
1561                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1562         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1563                 return btrfs_grab_root(fs_info->free_space_root) ?
1564                         fs_info->free_space_root : ERR_PTR(-ENOENT);
1565 again:
1566         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1567         if (root) {
1568                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1569                         btrfs_put_root(root);
1570                         return ERR_PTR(-ENOENT);
1571                 }
1572                 return root;
1573         }
1574
1575         root = btrfs_read_tree_root(fs_info->tree_root, location);
1576         if (IS_ERR(root))
1577                 return root;
1578
1579         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1580                 ret = -ENOENT;
1581                 goto fail;
1582         }
1583
1584         ret = btrfs_init_fs_root(root);
1585         if (ret)
1586                 goto fail;
1587
1588         path = btrfs_alloc_path();
1589         if (!path) {
1590                 ret = -ENOMEM;
1591                 goto fail;
1592         }
1593         key.objectid = BTRFS_ORPHAN_OBJECTID;
1594         key.type = BTRFS_ORPHAN_ITEM_KEY;
1595         key.offset = location->objectid;
1596
1597         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1598         btrfs_free_path(path);
1599         if (ret < 0)
1600                 goto fail;
1601         if (ret == 0)
1602                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1603
1604         ret = btrfs_insert_fs_root(fs_info, root);
1605         if (ret) {
1606                 btrfs_put_root(root);
1607                 if (ret == -EEXIST)
1608                         goto again;
1609                 goto fail;
1610         }
1611         return root;
1612 fail:
1613         btrfs_put_root(root);
1614         return ERR_PTR(ret);
1615 }
1616
1617 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1618 {
1619         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1620         int ret = 0;
1621         struct btrfs_device *device;
1622         struct backing_dev_info *bdi;
1623
1624         rcu_read_lock();
1625         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1626                 if (!device->bdev)
1627                         continue;
1628                 bdi = device->bdev->bd_bdi;
1629                 if (bdi_congested(bdi, bdi_bits)) {
1630                         ret = 1;
1631                         break;
1632                 }
1633         }
1634         rcu_read_unlock();
1635         return ret;
1636 }
1637
1638 /*
1639  * called by the kthread helper functions to finally call the bio end_io
1640  * functions.  This is where read checksum verification actually happens
1641  */
1642 static void end_workqueue_fn(struct btrfs_work *work)
1643 {
1644         struct bio *bio;
1645         struct btrfs_end_io_wq *end_io_wq;
1646
1647         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1648         bio = end_io_wq->bio;
1649
1650         bio->bi_status = end_io_wq->status;
1651         bio->bi_private = end_io_wq->private;
1652         bio->bi_end_io = end_io_wq->end_io;
1653         bio_endio(bio);
1654         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1655 }
1656
1657 static int cleaner_kthread(void *arg)
1658 {
1659         struct btrfs_root *root = arg;
1660         struct btrfs_fs_info *fs_info = root->fs_info;
1661         int again;
1662
1663         while (1) {
1664                 again = 0;
1665
1666                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1667
1668                 /* Make the cleaner go to sleep early. */
1669                 if (btrfs_need_cleaner_sleep(fs_info))
1670                         goto sleep;
1671
1672                 /*
1673                  * Do not do anything if we might cause open_ctree() to block
1674                  * before we have finished mounting the filesystem.
1675                  */
1676                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1677                         goto sleep;
1678
1679                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1680                         goto sleep;
1681
1682                 /*
1683                  * Avoid the problem that we change the status of the fs
1684                  * during the above check and trylock.
1685                  */
1686                 if (btrfs_need_cleaner_sleep(fs_info)) {
1687                         mutex_unlock(&fs_info->cleaner_mutex);
1688                         goto sleep;
1689                 }
1690
1691                 btrfs_run_delayed_iputs(fs_info);
1692
1693                 again = btrfs_clean_one_deleted_snapshot(root);
1694                 mutex_unlock(&fs_info->cleaner_mutex);
1695
1696                 /*
1697                  * The defragger has dealt with the R/O remount and umount,
1698                  * needn't do anything special here.
1699                  */
1700                 btrfs_run_defrag_inodes(fs_info);
1701
1702                 /*
1703                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1704                  * with relocation (btrfs_relocate_chunk) and relocation
1705                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1706                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1707                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1708                  * unused block groups.
1709                  */
1710                 btrfs_delete_unused_bgs(fs_info);
1711 sleep:
1712                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1713                 if (kthread_should_park())
1714                         kthread_parkme();
1715                 if (kthread_should_stop())
1716                         return 0;
1717                 if (!again) {
1718                         set_current_state(TASK_INTERRUPTIBLE);
1719                         schedule();
1720                         __set_current_state(TASK_RUNNING);
1721                 }
1722         }
1723 }
1724
1725 static int transaction_kthread(void *arg)
1726 {
1727         struct btrfs_root *root = arg;
1728         struct btrfs_fs_info *fs_info = root->fs_info;
1729         struct btrfs_trans_handle *trans;
1730         struct btrfs_transaction *cur;
1731         u64 transid;
1732         time64_t now;
1733         unsigned long delay;
1734         bool cannot_commit;
1735
1736         do {
1737                 cannot_commit = false;
1738                 delay = HZ * fs_info->commit_interval;
1739                 mutex_lock(&fs_info->transaction_kthread_mutex);
1740
1741                 spin_lock(&fs_info->trans_lock);
1742                 cur = fs_info->running_transaction;
1743                 if (!cur) {
1744                         spin_unlock(&fs_info->trans_lock);
1745                         goto sleep;
1746                 }
1747
1748                 now = ktime_get_seconds();
1749                 if (cur->state < TRANS_STATE_COMMIT_START &&
1750                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1751                     (now < cur->start_time ||
1752                      now - cur->start_time < fs_info->commit_interval)) {
1753                         spin_unlock(&fs_info->trans_lock);
1754                         delay = HZ * 5;
1755                         goto sleep;
1756                 }
1757                 transid = cur->transid;
1758                 spin_unlock(&fs_info->trans_lock);
1759
1760                 /* If the file system is aborted, this will always fail. */
1761                 trans = btrfs_attach_transaction(root);
1762                 if (IS_ERR(trans)) {
1763                         if (PTR_ERR(trans) != -ENOENT)
1764                                 cannot_commit = true;
1765                         goto sleep;
1766                 }
1767                 if (transid == trans->transid) {
1768                         btrfs_commit_transaction(trans);
1769                 } else {
1770                         btrfs_end_transaction(trans);
1771                 }
1772 sleep:
1773                 wake_up_process(fs_info->cleaner_kthread);
1774                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1775
1776                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1777                                       &fs_info->fs_state)))
1778                         btrfs_cleanup_transaction(fs_info);
1779                 if (!kthread_should_stop() &&
1780                                 (!btrfs_transaction_blocked(fs_info) ||
1781                                  cannot_commit))
1782                         schedule_timeout_interruptible(delay);
1783         } while (!kthread_should_stop());
1784         return 0;
1785 }
1786
1787 /*
1788  * This will find the highest generation in the array of root backups.  The
1789  * index of the highest array is returned, or -EINVAL if we can't find
1790  * anything.
1791  *
1792  * We check to make sure the array is valid by comparing the
1793  * generation of the latest  root in the array with the generation
1794  * in the super block.  If they don't match we pitch it.
1795  */
1796 static int find_newest_super_backup(struct btrfs_fs_info *info)
1797 {
1798         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1799         u64 cur;
1800         struct btrfs_root_backup *root_backup;
1801         int i;
1802
1803         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1804                 root_backup = info->super_copy->super_roots + i;
1805                 cur = btrfs_backup_tree_root_gen(root_backup);
1806                 if (cur == newest_gen)
1807                         return i;
1808         }
1809
1810         return -EINVAL;
1811 }
1812
1813 /*
1814  * copy all the root pointers into the super backup array.
1815  * this will bump the backup pointer by one when it is
1816  * done
1817  */
1818 static void backup_super_roots(struct btrfs_fs_info *info)
1819 {
1820         const int next_backup = info->backup_root_index;
1821         struct btrfs_root_backup *root_backup;
1822
1823         root_backup = info->super_for_commit->super_roots + next_backup;
1824
1825         /*
1826          * make sure all of our padding and empty slots get zero filled
1827          * regardless of which ones we use today
1828          */
1829         memset(root_backup, 0, sizeof(*root_backup));
1830
1831         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1832
1833         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1834         btrfs_set_backup_tree_root_gen(root_backup,
1835                                btrfs_header_generation(info->tree_root->node));
1836
1837         btrfs_set_backup_tree_root_level(root_backup,
1838                                btrfs_header_level(info->tree_root->node));
1839
1840         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1841         btrfs_set_backup_chunk_root_gen(root_backup,
1842                                btrfs_header_generation(info->chunk_root->node));
1843         btrfs_set_backup_chunk_root_level(root_backup,
1844                                btrfs_header_level(info->chunk_root->node));
1845
1846         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1847         btrfs_set_backup_extent_root_gen(root_backup,
1848                                btrfs_header_generation(info->extent_root->node));
1849         btrfs_set_backup_extent_root_level(root_backup,
1850                                btrfs_header_level(info->extent_root->node));
1851
1852         /*
1853          * we might commit during log recovery, which happens before we set
1854          * the fs_root.  Make sure it is valid before we fill it in.
1855          */
1856         if (info->fs_root && info->fs_root->node) {
1857                 btrfs_set_backup_fs_root(root_backup,
1858                                          info->fs_root->node->start);
1859                 btrfs_set_backup_fs_root_gen(root_backup,
1860                                btrfs_header_generation(info->fs_root->node));
1861                 btrfs_set_backup_fs_root_level(root_backup,
1862                                btrfs_header_level(info->fs_root->node));
1863         }
1864
1865         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1866         btrfs_set_backup_dev_root_gen(root_backup,
1867                                btrfs_header_generation(info->dev_root->node));
1868         btrfs_set_backup_dev_root_level(root_backup,
1869                                        btrfs_header_level(info->dev_root->node));
1870
1871         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1872         btrfs_set_backup_csum_root_gen(root_backup,
1873                                btrfs_header_generation(info->csum_root->node));
1874         btrfs_set_backup_csum_root_level(root_backup,
1875                                btrfs_header_level(info->csum_root->node));
1876
1877         btrfs_set_backup_total_bytes(root_backup,
1878                              btrfs_super_total_bytes(info->super_copy));
1879         btrfs_set_backup_bytes_used(root_backup,
1880                              btrfs_super_bytes_used(info->super_copy));
1881         btrfs_set_backup_num_devices(root_backup,
1882                              btrfs_super_num_devices(info->super_copy));
1883
1884         /*
1885          * if we don't copy this out to the super_copy, it won't get remembered
1886          * for the next commit
1887          */
1888         memcpy(&info->super_copy->super_roots,
1889                &info->super_for_commit->super_roots,
1890                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1891 }
1892
1893 /*
1894  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1895  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1896  *
1897  * fs_info - filesystem whose backup roots need to be read
1898  * priority - priority of backup root required
1899  *
1900  * Returns backup root index on success and -EINVAL otherwise.
1901  */
1902 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1903 {
1904         int backup_index = find_newest_super_backup(fs_info);
1905         struct btrfs_super_block *super = fs_info->super_copy;
1906         struct btrfs_root_backup *root_backup;
1907
1908         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1909                 if (priority == 0)
1910                         return backup_index;
1911
1912                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1913                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1914         } else {
1915                 return -EINVAL;
1916         }
1917
1918         root_backup = super->super_roots + backup_index;
1919
1920         btrfs_set_super_generation(super,
1921                                    btrfs_backup_tree_root_gen(root_backup));
1922         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1923         btrfs_set_super_root_level(super,
1924                                    btrfs_backup_tree_root_level(root_backup));
1925         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1926
1927         /*
1928          * Fixme: the total bytes and num_devices need to match or we should
1929          * need a fsck
1930          */
1931         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1932         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1933
1934         return backup_index;
1935 }
1936
1937 /* helper to cleanup workers */
1938 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1939 {
1940         btrfs_destroy_workqueue(fs_info->fixup_workers);
1941         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1942         btrfs_destroy_workqueue(fs_info->workers);
1943         btrfs_destroy_workqueue(fs_info->endio_workers);
1944         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1945         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1946         btrfs_destroy_workqueue(fs_info->rmw_workers);
1947         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1948         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1949         btrfs_destroy_workqueue(fs_info->delayed_workers);
1950         btrfs_destroy_workqueue(fs_info->caching_workers);
1951         btrfs_destroy_workqueue(fs_info->readahead_workers);
1952         btrfs_destroy_workqueue(fs_info->flush_workers);
1953         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1954         if (fs_info->discard_ctl.discard_workers)
1955                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1956         /*
1957          * Now that all other work queues are destroyed, we can safely destroy
1958          * the queues used for metadata I/O, since tasks from those other work
1959          * queues can do metadata I/O operations.
1960          */
1961         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1962         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1963 }
1964
1965 static void free_root_extent_buffers(struct btrfs_root *root)
1966 {
1967         if (root) {
1968                 free_extent_buffer(root->node);
1969                 free_extent_buffer(root->commit_root);
1970                 root->node = NULL;
1971                 root->commit_root = NULL;
1972         }
1973 }
1974
1975 /* helper to cleanup tree roots */
1976 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1977 {
1978         free_root_extent_buffers(info->tree_root);
1979
1980         free_root_extent_buffers(info->dev_root);
1981         free_root_extent_buffers(info->extent_root);
1982         free_root_extent_buffers(info->csum_root);
1983         free_root_extent_buffers(info->quota_root);
1984         free_root_extent_buffers(info->uuid_root);
1985         free_root_extent_buffers(info->fs_root);
1986         if (free_chunk_root)
1987                 free_root_extent_buffers(info->chunk_root);
1988         free_root_extent_buffers(info->free_space_root);
1989 }
1990
1991 void btrfs_put_root(struct btrfs_root *root)
1992 {
1993         if (!root)
1994                 return;
1995
1996         if (refcount_dec_and_test(&root->refs)) {
1997                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1998                 if (root->anon_dev)
1999                         free_anon_bdev(root->anon_dev);
2000                 btrfs_drew_lock_destroy(&root->snapshot_lock);
2001                 free_extent_buffer(root->node);
2002                 free_extent_buffer(root->commit_root);
2003                 kfree(root->free_ino_ctl);
2004                 kfree(root->free_ino_pinned);
2005 #ifdef CONFIG_BTRFS_DEBUG
2006                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2007                 list_del_init(&root->leak_list);
2008                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2009 #endif
2010                 kfree(root);
2011         }
2012 }
2013
2014 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2015 {
2016         int ret;
2017         struct btrfs_root *gang[8];
2018         int i;
2019
2020         while (!list_empty(&fs_info->dead_roots)) {
2021                 gang[0] = list_entry(fs_info->dead_roots.next,
2022                                      struct btrfs_root, root_list);
2023                 list_del(&gang[0]->root_list);
2024
2025                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2026                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2027                 btrfs_put_root(gang[0]);
2028         }
2029
2030         while (1) {
2031                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2032                                              (void **)gang, 0,
2033                                              ARRAY_SIZE(gang));
2034                 if (!ret)
2035                         break;
2036                 for (i = 0; i < ret; i++)
2037                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2038         }
2039
2040         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
2041                 btrfs_free_log_root_tree(NULL, fs_info);
2042 }
2043
2044 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2045 {
2046         mutex_init(&fs_info->scrub_lock);
2047         atomic_set(&fs_info->scrubs_running, 0);
2048         atomic_set(&fs_info->scrub_pause_req, 0);
2049         atomic_set(&fs_info->scrubs_paused, 0);
2050         atomic_set(&fs_info->scrub_cancel_req, 0);
2051         init_waitqueue_head(&fs_info->scrub_pause_wait);
2052         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2053 }
2054
2055 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2056 {
2057         spin_lock_init(&fs_info->balance_lock);
2058         mutex_init(&fs_info->balance_mutex);
2059         atomic_set(&fs_info->balance_pause_req, 0);
2060         atomic_set(&fs_info->balance_cancel_req, 0);
2061         fs_info->balance_ctl = NULL;
2062         init_waitqueue_head(&fs_info->balance_wait_q);
2063 }
2064
2065 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2066 {
2067         struct inode *inode = fs_info->btree_inode;
2068
2069         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2070         set_nlink(inode, 1);
2071         /*
2072          * we set the i_size on the btree inode to the max possible int.
2073          * the real end of the address space is determined by all of
2074          * the devices in the system
2075          */
2076         inode->i_size = OFFSET_MAX;
2077         inode->i_mapping->a_ops = &btree_aops;
2078
2079         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2080         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2081                             IO_TREE_INODE_IO, inode);
2082         BTRFS_I(inode)->io_tree.track_uptodate = false;
2083         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2084
2085         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2086
2087         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2088         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2089         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2090         btrfs_insert_inode_hash(inode);
2091 }
2092
2093 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2094 {
2095         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2096         init_rwsem(&fs_info->dev_replace.rwsem);
2097         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2098 }
2099
2100 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2101 {
2102         spin_lock_init(&fs_info->qgroup_lock);
2103         mutex_init(&fs_info->qgroup_ioctl_lock);
2104         fs_info->qgroup_tree = RB_ROOT;
2105         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2106         fs_info->qgroup_seq = 1;
2107         fs_info->qgroup_ulist = NULL;
2108         fs_info->qgroup_rescan_running = false;
2109         mutex_init(&fs_info->qgroup_rescan_lock);
2110 }
2111
2112 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2113                 struct btrfs_fs_devices *fs_devices)
2114 {
2115         u32 max_active = fs_info->thread_pool_size;
2116         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2117
2118         fs_info->workers =
2119                 btrfs_alloc_workqueue(fs_info, "worker",
2120                                       flags | WQ_HIGHPRI, max_active, 16);
2121
2122         fs_info->delalloc_workers =
2123                 btrfs_alloc_workqueue(fs_info, "delalloc",
2124                                       flags, max_active, 2);
2125
2126         fs_info->flush_workers =
2127                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2128                                       flags, max_active, 0);
2129
2130         fs_info->caching_workers =
2131                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2132
2133         fs_info->fixup_workers =
2134                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2135
2136         /*
2137          * endios are largely parallel and should have a very
2138          * low idle thresh
2139          */
2140         fs_info->endio_workers =
2141                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2142         fs_info->endio_meta_workers =
2143                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2144                                       max_active, 4);
2145         fs_info->endio_meta_write_workers =
2146                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2147                                       max_active, 2);
2148         fs_info->endio_raid56_workers =
2149                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2150                                       max_active, 4);
2151         fs_info->endio_repair_workers =
2152                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2153         fs_info->rmw_workers =
2154                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2155         fs_info->endio_write_workers =
2156                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2157                                       max_active, 2);
2158         fs_info->endio_freespace_worker =
2159                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2160                                       max_active, 0);
2161         fs_info->delayed_workers =
2162                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2163                                       max_active, 0);
2164         fs_info->readahead_workers =
2165                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2166                                       max_active, 2);
2167         fs_info->qgroup_rescan_workers =
2168                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2169         fs_info->discard_ctl.discard_workers =
2170                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2171
2172         if (!(fs_info->workers && fs_info->delalloc_workers &&
2173               fs_info->flush_workers &&
2174               fs_info->endio_workers && fs_info->endio_meta_workers &&
2175               fs_info->endio_meta_write_workers &&
2176               fs_info->endio_repair_workers &&
2177               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2178               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2179               fs_info->caching_workers && fs_info->readahead_workers &&
2180               fs_info->fixup_workers && fs_info->delayed_workers &&
2181               fs_info->qgroup_rescan_workers &&
2182               fs_info->discard_ctl.discard_workers)) {
2183                 return -ENOMEM;
2184         }
2185
2186         return 0;
2187 }
2188
2189 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2190 {
2191         struct crypto_shash *csum_shash;
2192         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2193
2194         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2195
2196         if (IS_ERR(csum_shash)) {
2197                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2198                           csum_driver);
2199                 return PTR_ERR(csum_shash);
2200         }
2201
2202         fs_info->csum_shash = csum_shash;
2203
2204         return 0;
2205 }
2206
2207 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2208                             struct btrfs_fs_devices *fs_devices)
2209 {
2210         int ret;
2211         struct btrfs_root *log_tree_root;
2212         struct btrfs_super_block *disk_super = fs_info->super_copy;
2213         u64 bytenr = btrfs_super_log_root(disk_super);
2214         int level = btrfs_super_log_root_level(disk_super);
2215
2216         if (fs_devices->rw_devices == 0) {
2217                 btrfs_warn(fs_info, "log replay required on RO media");
2218                 return -EIO;
2219         }
2220
2221         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2222                                          GFP_KERNEL);
2223         if (!log_tree_root)
2224                 return -ENOMEM;
2225
2226         log_tree_root->node = read_tree_block(fs_info, bytenr,
2227                                               fs_info->generation + 1,
2228                                               level, NULL);
2229         if (IS_ERR(log_tree_root->node)) {
2230                 btrfs_warn(fs_info, "failed to read log tree");
2231                 ret = PTR_ERR(log_tree_root->node);
2232                 log_tree_root->node = NULL;
2233                 btrfs_put_root(log_tree_root);
2234                 return ret;
2235         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2236                 btrfs_err(fs_info, "failed to read log tree");
2237                 btrfs_put_root(log_tree_root);
2238                 return -EIO;
2239         }
2240         /* returns with log_tree_root freed on success */
2241         ret = btrfs_recover_log_trees(log_tree_root);
2242         if (ret) {
2243                 btrfs_handle_fs_error(fs_info, ret,
2244                                       "Failed to recover log tree");
2245                 btrfs_put_root(log_tree_root);
2246                 return ret;
2247         }
2248
2249         if (sb_rdonly(fs_info->sb)) {
2250                 ret = btrfs_commit_super(fs_info);
2251                 if (ret)
2252                         return ret;
2253         }
2254
2255         return 0;
2256 }
2257
2258 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2259 {
2260         struct btrfs_root *tree_root = fs_info->tree_root;
2261         struct btrfs_root *root;
2262         struct btrfs_key location;
2263         int ret;
2264
2265         BUG_ON(!fs_info->tree_root);
2266
2267         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2268         location.type = BTRFS_ROOT_ITEM_KEY;
2269         location.offset = 0;
2270
2271         root = btrfs_read_tree_root(tree_root, &location);
2272         if (IS_ERR(root)) {
2273                 ret = PTR_ERR(root);
2274                 goto out;
2275         }
2276         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2277         fs_info->extent_root = root;
2278
2279         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2280         root = btrfs_read_tree_root(tree_root, &location);
2281         if (IS_ERR(root)) {
2282                 ret = PTR_ERR(root);
2283                 goto out;
2284         }
2285         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2286         fs_info->dev_root = root;
2287         btrfs_init_devices_late(fs_info);
2288
2289         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2290         root = btrfs_read_tree_root(tree_root, &location);
2291         if (IS_ERR(root)) {
2292                 ret = PTR_ERR(root);
2293                 goto out;
2294         }
2295         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2296         fs_info->csum_root = root;
2297
2298         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2299         root = btrfs_read_tree_root(tree_root, &location);
2300         if (!IS_ERR(root)) {
2301                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2303                 fs_info->quota_root = root;
2304         }
2305
2306         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2307         root = btrfs_read_tree_root(tree_root, &location);
2308         if (IS_ERR(root)) {
2309                 ret = PTR_ERR(root);
2310                 if (ret != -ENOENT)
2311                         goto out;
2312         } else {
2313                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2314                 fs_info->uuid_root = root;
2315         }
2316
2317         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2318                 location.objectid = BTRFS_FREE_SPACE_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->free_space_root = root;
2326         }
2327
2328         return 0;
2329 out:
2330         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2331                    location.objectid, ret);
2332         return ret;
2333 }
2334
2335 /*
2336  * Real super block validation
2337  * NOTE: super csum type and incompat features will not be checked here.
2338  *
2339  * @sb:         super block to check
2340  * @mirror_num: the super block number to check its bytenr:
2341  *              0       the primary (1st) sb
2342  *              1, 2    2nd and 3rd backup copy
2343  *             -1       skip bytenr check
2344  */
2345 static int validate_super(struct btrfs_fs_info *fs_info,
2346                             struct btrfs_super_block *sb, int mirror_num)
2347 {
2348         u64 nodesize = btrfs_super_nodesize(sb);
2349         u64 sectorsize = btrfs_super_sectorsize(sb);
2350         int ret = 0;
2351
2352         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2353                 btrfs_err(fs_info, "no valid FS found");
2354                 ret = -EINVAL;
2355         }
2356         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2357                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2358                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2359                 ret = -EINVAL;
2360         }
2361         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2362                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2363                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2364                 ret = -EINVAL;
2365         }
2366         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2367                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2368                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2369                 ret = -EINVAL;
2370         }
2371         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2372                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2373                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2374                 ret = -EINVAL;
2375         }
2376
2377         /*
2378          * Check sectorsize and nodesize first, other check will need it.
2379          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2380          */
2381         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2382             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2383                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2384                 ret = -EINVAL;
2385         }
2386         /* Only PAGE SIZE is supported yet */
2387         if (sectorsize != PAGE_SIZE) {
2388                 btrfs_err(fs_info,
2389                         "sectorsize %llu not supported yet, only support %lu",
2390                         sectorsize, PAGE_SIZE);
2391                 ret = -EINVAL;
2392         }
2393         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2394             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2395                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2396                 ret = -EINVAL;
2397         }
2398         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2399                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2400                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2401                 ret = -EINVAL;
2402         }
2403
2404         /* Root alignment check */
2405         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2406                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2407                            btrfs_super_root(sb));
2408                 ret = -EINVAL;
2409         }
2410         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2411                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2412                            btrfs_super_chunk_root(sb));
2413                 ret = -EINVAL;
2414         }
2415         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2416                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2417                            btrfs_super_log_root(sb));
2418                 ret = -EINVAL;
2419         }
2420
2421         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2422                    BTRFS_FSID_SIZE) != 0) {
2423                 btrfs_err(fs_info,
2424                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2425                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2426                 ret = -EINVAL;
2427         }
2428
2429         /*
2430          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2431          * done later
2432          */
2433         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2434                 btrfs_err(fs_info, "bytes_used is too small %llu",
2435                           btrfs_super_bytes_used(sb));
2436                 ret = -EINVAL;
2437         }
2438         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2439                 btrfs_err(fs_info, "invalid stripesize %u",
2440                           btrfs_super_stripesize(sb));
2441                 ret = -EINVAL;
2442         }
2443         if (btrfs_super_num_devices(sb) > (1UL << 31))
2444                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2445                            btrfs_super_num_devices(sb));
2446         if (btrfs_super_num_devices(sb) == 0) {
2447                 btrfs_err(fs_info, "number of devices is 0");
2448                 ret = -EINVAL;
2449         }
2450
2451         if (mirror_num >= 0 &&
2452             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2453                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2454                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2455                 ret = -EINVAL;
2456         }
2457
2458         /*
2459          * Obvious sys_chunk_array corruptions, it must hold at least one key
2460          * and one chunk
2461          */
2462         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2463                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2464                           btrfs_super_sys_array_size(sb),
2465                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2466                 ret = -EINVAL;
2467         }
2468         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2469                         + sizeof(struct btrfs_chunk)) {
2470                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2471                           btrfs_super_sys_array_size(sb),
2472                           sizeof(struct btrfs_disk_key)
2473                           + sizeof(struct btrfs_chunk));
2474                 ret = -EINVAL;
2475         }
2476
2477         /*
2478          * The generation is a global counter, we'll trust it more than the others
2479          * but it's still possible that it's the one that's wrong.
2480          */
2481         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2482                 btrfs_warn(fs_info,
2483                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2484                         btrfs_super_generation(sb),
2485                         btrfs_super_chunk_root_generation(sb));
2486         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2487             && btrfs_super_cache_generation(sb) != (u64)-1)
2488                 btrfs_warn(fs_info,
2489                         "suspicious: generation < cache_generation: %llu < %llu",
2490                         btrfs_super_generation(sb),
2491                         btrfs_super_cache_generation(sb));
2492
2493         return ret;
2494 }
2495
2496 /*
2497  * Validation of super block at mount time.
2498  * Some checks already done early at mount time, like csum type and incompat
2499  * flags will be skipped.
2500  */
2501 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2502 {
2503         return validate_super(fs_info, fs_info->super_copy, 0);
2504 }
2505
2506 /*
2507  * Validation of super block at write time.
2508  * Some checks like bytenr check will be skipped as their values will be
2509  * overwritten soon.
2510  * Extra checks like csum type and incompat flags will be done here.
2511  */
2512 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2513                                       struct btrfs_super_block *sb)
2514 {
2515         int ret;
2516
2517         ret = validate_super(fs_info, sb, -1);
2518         if (ret < 0)
2519                 goto out;
2520         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2521                 ret = -EUCLEAN;
2522                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2523                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2524                 goto out;
2525         }
2526         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2527                 ret = -EUCLEAN;
2528                 btrfs_err(fs_info,
2529                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2530                           btrfs_super_incompat_flags(sb),
2531                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2532                 goto out;
2533         }
2534 out:
2535         if (ret < 0)
2536                 btrfs_err(fs_info,
2537                 "super block corruption detected before writing it to disk");
2538         return ret;
2539 }
2540
2541 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2542 {
2543         int backup_index = find_newest_super_backup(fs_info);
2544         struct btrfs_super_block *sb = fs_info->super_copy;
2545         struct btrfs_root *tree_root = fs_info->tree_root;
2546         bool handle_error = false;
2547         int ret = 0;
2548         int i;
2549
2550         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2551                 u64 generation;
2552                 int level;
2553
2554                 if (handle_error) {
2555                         if (!IS_ERR(tree_root->node))
2556                                 free_extent_buffer(tree_root->node);
2557                         tree_root->node = NULL;
2558
2559                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2560                                 break;
2561
2562                         free_root_pointers(fs_info, 0);
2563
2564                         /*
2565                          * Don't use the log in recovery mode, it won't be
2566                          * valid
2567                          */
2568                         btrfs_set_super_log_root(sb, 0);
2569
2570                         /* We can't trust the free space cache either */
2571                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2572
2573                         ret = read_backup_root(fs_info, i);
2574                         backup_index = ret;
2575                         if (ret < 0)
2576                                 return ret;
2577                 }
2578                 generation = btrfs_super_generation(sb);
2579                 level = btrfs_super_root_level(sb);
2580                 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2581                                                   generation, level, NULL);
2582                 if (IS_ERR(tree_root->node) ||
2583                     !extent_buffer_uptodate(tree_root->node)) {
2584                         handle_error = true;
2585
2586                         if (IS_ERR(tree_root->node))
2587                                 ret = PTR_ERR(tree_root->node);
2588                         else if (!extent_buffer_uptodate(tree_root->node))
2589                                 ret = -EUCLEAN;
2590
2591                         btrfs_warn(fs_info, "failed to read tree root");
2592                         continue;
2593                 }
2594
2595                 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2596                 tree_root->commit_root = btrfs_root_node(tree_root);
2597                 btrfs_set_root_refs(&tree_root->root_item, 1);
2598
2599                 /*
2600                  * No need to hold btrfs_root::objectid_mutex since the fs
2601                  * hasn't been fully initialised and we are the only user
2602                  */
2603                 ret = btrfs_find_highest_objectid(tree_root,
2604                                                 &tree_root->highest_objectid);
2605                 if (ret < 0) {
2606                         handle_error = true;
2607                         continue;
2608                 }
2609
2610                 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2611
2612                 ret = btrfs_read_roots(fs_info);
2613                 if (ret < 0) {
2614                         handle_error = true;
2615                         continue;
2616                 }
2617
2618                 /* All successful */
2619                 fs_info->generation = generation;
2620                 fs_info->last_trans_committed = generation;
2621
2622                 /* Always begin writing backup roots after the one being used */
2623                 if (backup_index < 0) {
2624                         fs_info->backup_root_index = 0;
2625                 } else {
2626                         fs_info->backup_root_index = backup_index + 1;
2627                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2628                 }
2629                 break;
2630         }
2631
2632         return ret;
2633 }
2634
2635 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2636 {
2637         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2638         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2639         INIT_LIST_HEAD(&fs_info->trans_list);
2640         INIT_LIST_HEAD(&fs_info->dead_roots);
2641         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2642         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2643         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2644         spin_lock_init(&fs_info->delalloc_root_lock);
2645         spin_lock_init(&fs_info->trans_lock);
2646         spin_lock_init(&fs_info->fs_roots_radix_lock);
2647         spin_lock_init(&fs_info->delayed_iput_lock);
2648         spin_lock_init(&fs_info->defrag_inodes_lock);
2649         spin_lock_init(&fs_info->super_lock);
2650         spin_lock_init(&fs_info->buffer_lock);
2651         spin_lock_init(&fs_info->unused_bgs_lock);
2652         rwlock_init(&fs_info->tree_mod_log_lock);
2653         mutex_init(&fs_info->unused_bg_unpin_mutex);
2654         mutex_init(&fs_info->delete_unused_bgs_mutex);
2655         mutex_init(&fs_info->reloc_mutex);
2656         mutex_init(&fs_info->delalloc_root_mutex);
2657         seqlock_init(&fs_info->profiles_lock);
2658
2659         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2660         INIT_LIST_HEAD(&fs_info->space_info);
2661         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2662         INIT_LIST_HEAD(&fs_info->unused_bgs);
2663 #ifdef CONFIG_BTRFS_DEBUG
2664         INIT_LIST_HEAD(&fs_info->allocated_roots);
2665         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2666         spin_lock_init(&fs_info->eb_leak_lock);
2667 #endif
2668         extent_map_tree_init(&fs_info->mapping_tree);
2669         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2670                              BTRFS_BLOCK_RSV_GLOBAL);
2671         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2672         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2673         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2674         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2675                              BTRFS_BLOCK_RSV_DELOPS);
2676         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2677                              BTRFS_BLOCK_RSV_DELREFS);
2678
2679         atomic_set(&fs_info->async_delalloc_pages, 0);
2680         atomic_set(&fs_info->defrag_running, 0);
2681         atomic_set(&fs_info->reada_works_cnt, 0);
2682         atomic_set(&fs_info->nr_delayed_iputs, 0);
2683         atomic64_set(&fs_info->tree_mod_seq, 0);
2684         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2685         fs_info->metadata_ratio = 0;
2686         fs_info->defrag_inodes = RB_ROOT;
2687         atomic64_set(&fs_info->free_chunk_space, 0);
2688         fs_info->tree_mod_log = RB_ROOT;
2689         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2690         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2691         /* readahead state */
2692         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2693         spin_lock_init(&fs_info->reada_lock);
2694         btrfs_init_ref_verify(fs_info);
2695
2696         fs_info->thread_pool_size = min_t(unsigned long,
2697                                           num_online_cpus() + 2, 8);
2698
2699         INIT_LIST_HEAD(&fs_info->ordered_roots);
2700         spin_lock_init(&fs_info->ordered_root_lock);
2701
2702         btrfs_init_scrub(fs_info);
2703 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2704         fs_info->check_integrity_print_mask = 0;
2705 #endif
2706         btrfs_init_balance(fs_info);
2707         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2708
2709         spin_lock_init(&fs_info->block_group_cache_lock);
2710         fs_info->block_group_cache_tree = RB_ROOT;
2711         fs_info->first_logical_byte = (u64)-1;
2712
2713         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2714                             IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2715         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2716
2717         mutex_init(&fs_info->ordered_operations_mutex);
2718         mutex_init(&fs_info->tree_log_mutex);
2719         mutex_init(&fs_info->chunk_mutex);
2720         mutex_init(&fs_info->transaction_kthread_mutex);
2721         mutex_init(&fs_info->cleaner_mutex);
2722         mutex_init(&fs_info->ro_block_group_mutex);
2723         init_rwsem(&fs_info->commit_root_sem);
2724         init_rwsem(&fs_info->cleanup_work_sem);
2725         init_rwsem(&fs_info->subvol_sem);
2726         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2727
2728         btrfs_init_dev_replace_locks(fs_info);
2729         btrfs_init_qgroup(fs_info);
2730         btrfs_discard_init(fs_info);
2731
2732         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2733         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2734
2735         init_waitqueue_head(&fs_info->transaction_throttle);
2736         init_waitqueue_head(&fs_info->transaction_wait);
2737         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2738         init_waitqueue_head(&fs_info->async_submit_wait);
2739         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2740
2741         /* Usable values until the real ones are cached from the superblock */
2742         fs_info->nodesize = 4096;
2743         fs_info->sectorsize = 4096;
2744         fs_info->stripesize = 4096;
2745
2746         spin_lock_init(&fs_info->swapfile_pins_lock);
2747         fs_info->swapfile_pins = RB_ROOT;
2748
2749         fs_info->send_in_progress = 0;
2750 }
2751
2752 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2753 {
2754         int ret;
2755
2756         fs_info->sb = sb;
2757         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2758         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2759
2760         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2761         if (ret)
2762                 return ret;
2763
2764         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2765         if (ret)
2766                 return ret;
2767
2768         fs_info->dirty_metadata_batch = PAGE_SIZE *
2769                                         (1 + ilog2(nr_cpu_ids));
2770
2771         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2772         if (ret)
2773                 return ret;
2774
2775         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2776                         GFP_KERNEL);
2777         if (ret)
2778                 return ret;
2779
2780         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2781                                         GFP_KERNEL);
2782         if (!fs_info->delayed_root)
2783                 return -ENOMEM;
2784         btrfs_init_delayed_root(fs_info->delayed_root);
2785
2786         return btrfs_alloc_stripe_hash_table(fs_info);
2787 }
2788
2789 static int btrfs_uuid_rescan_kthread(void *data)
2790 {
2791         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2792         int ret;
2793
2794         /*
2795          * 1st step is to iterate through the existing UUID tree and
2796          * to delete all entries that contain outdated data.
2797          * 2nd step is to add all missing entries to the UUID tree.
2798          */
2799         ret = btrfs_uuid_tree_iterate(fs_info);
2800         if (ret < 0) {
2801                 if (ret != -EINTR)
2802                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2803                                    ret);
2804                 up(&fs_info->uuid_tree_rescan_sem);
2805                 return ret;
2806         }
2807         return btrfs_uuid_scan_kthread(data);
2808 }
2809
2810 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2811 {
2812         struct task_struct *task;
2813
2814         down(&fs_info->uuid_tree_rescan_sem);
2815         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2816         if (IS_ERR(task)) {
2817                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2818                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2819                 up(&fs_info->uuid_tree_rescan_sem);
2820                 return PTR_ERR(task);
2821         }
2822
2823         return 0;
2824 }
2825
2826 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2827                       char *options)
2828 {
2829         u32 sectorsize;
2830         u32 nodesize;
2831         u32 stripesize;
2832         u64 generation;
2833         u64 features;
2834         u16 csum_type;
2835         struct btrfs_key location;
2836         struct btrfs_super_block *disk_super;
2837         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2838         struct btrfs_root *tree_root;
2839         struct btrfs_root *chunk_root;
2840         int ret;
2841         int err = -EINVAL;
2842         int clear_free_space_tree = 0;
2843         int level;
2844
2845         ret = init_mount_fs_info(fs_info, sb);
2846         if (ret) {
2847                 err = ret;
2848                 goto fail;
2849         }
2850
2851         /* These need to be init'ed before we start creating inodes and such. */
2852         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2853                                      GFP_KERNEL);
2854         fs_info->tree_root = tree_root;
2855         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2856                                       GFP_KERNEL);
2857         fs_info->chunk_root = chunk_root;
2858         if (!tree_root || !chunk_root) {
2859                 err = -ENOMEM;
2860                 goto fail;
2861         }
2862
2863         fs_info->btree_inode = new_inode(sb);
2864         if (!fs_info->btree_inode) {
2865                 err = -ENOMEM;
2866                 goto fail;
2867         }
2868         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2869         btrfs_init_btree_inode(fs_info);
2870
2871         invalidate_bdev(fs_devices->latest_bdev);
2872
2873         /*
2874          * Read super block and check the signature bytes only
2875          */
2876         disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2877         if (IS_ERR(disk_super)) {
2878                 err = PTR_ERR(disk_super);
2879                 goto fail_alloc;
2880         }
2881
2882         /*
2883          * Verify the type first, if that or the the checksum value are
2884          * corrupted, we'll find out
2885          */
2886         csum_type = btrfs_super_csum_type(disk_super);
2887         if (!btrfs_supported_super_csum(csum_type)) {
2888                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2889                           csum_type);
2890                 err = -EINVAL;
2891                 btrfs_release_disk_super(disk_super);
2892                 goto fail_alloc;
2893         }
2894
2895         ret = btrfs_init_csum_hash(fs_info, csum_type);
2896         if (ret) {
2897                 err = ret;
2898                 btrfs_release_disk_super(disk_super);
2899                 goto fail_alloc;
2900         }
2901
2902         /*
2903          * We want to check superblock checksum, the type is stored inside.
2904          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2905          */
2906         if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2907                 btrfs_err(fs_info, "superblock checksum mismatch");
2908                 err = -EINVAL;
2909                 btrfs_release_disk_super(disk_super);
2910                 goto fail_alloc;
2911         }
2912
2913         /*
2914          * super_copy is zeroed at allocation time and we never touch the
2915          * following bytes up to INFO_SIZE, the checksum is calculated from
2916          * the whole block of INFO_SIZE
2917          */
2918         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2919         btrfs_release_disk_super(disk_super);
2920
2921         disk_super = fs_info->super_copy;
2922
2923         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2924                        BTRFS_FSID_SIZE));
2925
2926         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2927                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2928                                 fs_info->super_copy->metadata_uuid,
2929                                 BTRFS_FSID_SIZE));
2930         }
2931
2932         features = btrfs_super_flags(disk_super);
2933         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2934                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2935                 btrfs_set_super_flags(disk_super, features);
2936                 btrfs_info(fs_info,
2937                         "found metadata UUID change in progress flag, clearing");
2938         }
2939
2940         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2941                sizeof(*fs_info->super_for_commit));
2942
2943         ret = btrfs_validate_mount_super(fs_info);
2944         if (ret) {
2945                 btrfs_err(fs_info, "superblock contains fatal errors");
2946                 err = -EINVAL;
2947                 goto fail_alloc;
2948         }
2949
2950         if (!btrfs_super_root(disk_super))
2951                 goto fail_alloc;
2952
2953         /* check FS state, whether FS is broken. */
2954         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2955                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2956
2957         /*
2958          * In the long term, we'll store the compression type in the super
2959          * block, and it'll be used for per file compression control.
2960          */
2961         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2962
2963         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2964         if (ret) {
2965                 err = ret;
2966                 goto fail_alloc;
2967         }
2968
2969         features = btrfs_super_incompat_flags(disk_super) &
2970                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2971         if (features) {
2972                 btrfs_err(fs_info,
2973                     "cannot mount because of unsupported optional features (%llx)",
2974                     features);
2975                 err = -EINVAL;
2976                 goto fail_alloc;
2977         }
2978
2979         features = btrfs_super_incompat_flags(disk_super);
2980         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2981         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2982                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2983         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2984                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2985
2986         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2987                 btrfs_info(fs_info, "has skinny extents");
2988
2989         /*
2990          * flag our filesystem as having big metadata blocks if
2991          * they are bigger than the page size
2992          */
2993         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2994                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2995                         btrfs_info(fs_info,
2996                                 "flagging fs with big metadata feature");
2997                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2998         }
2999
3000         nodesize = btrfs_super_nodesize(disk_super);
3001         sectorsize = btrfs_super_sectorsize(disk_super);
3002         stripesize = sectorsize;
3003         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3004         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3005
3006         /* Cache block sizes */
3007         fs_info->nodesize = nodesize;
3008         fs_info->sectorsize = sectorsize;
3009         fs_info->stripesize = stripesize;
3010
3011         /*
3012          * mixed block groups end up with duplicate but slightly offset
3013          * extent buffers for the same range.  It leads to corruptions
3014          */
3015         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3016             (sectorsize != nodesize)) {
3017                 btrfs_err(fs_info,
3018 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3019                         nodesize, sectorsize);
3020                 goto fail_alloc;
3021         }
3022
3023         /*
3024          * Needn't use the lock because there is no other task which will
3025          * update the flag.
3026          */
3027         btrfs_set_super_incompat_flags(disk_super, features);
3028
3029         features = btrfs_super_compat_ro_flags(disk_super) &
3030                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3031         if (!sb_rdonly(sb) && features) {
3032                 btrfs_err(fs_info,
3033         "cannot mount read-write because of unsupported optional features (%llx)",
3034                        features);
3035                 err = -EINVAL;
3036                 goto fail_alloc;
3037         }
3038
3039         ret = btrfs_init_workqueues(fs_info, fs_devices);
3040         if (ret) {
3041                 err = ret;
3042                 goto fail_sb_buffer;
3043         }
3044
3045         sb->s_bdi->congested_fn = btrfs_congested_fn;
3046         sb->s_bdi->congested_data = fs_info;
3047         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3048         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3049         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3050         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3051
3052         sb->s_blocksize = sectorsize;
3053         sb->s_blocksize_bits = blksize_bits(sectorsize);
3054         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3055
3056         mutex_lock(&fs_info->chunk_mutex);
3057         ret = btrfs_read_sys_array(fs_info);
3058         mutex_unlock(&fs_info->chunk_mutex);
3059         if (ret) {
3060                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3061                 goto fail_sb_buffer;
3062         }
3063
3064         generation = btrfs_super_chunk_root_generation(disk_super);
3065         level = btrfs_super_chunk_root_level(disk_super);
3066
3067         chunk_root->node = read_tree_block(fs_info,
3068                                            btrfs_super_chunk_root(disk_super),
3069                                            generation, level, NULL);
3070         if (IS_ERR(chunk_root->node) ||
3071             !extent_buffer_uptodate(chunk_root->node)) {
3072                 btrfs_err(fs_info, "failed to read chunk root");
3073                 if (!IS_ERR(chunk_root->node))
3074                         free_extent_buffer(chunk_root->node);
3075                 chunk_root->node = NULL;
3076                 goto fail_tree_roots;
3077         }
3078         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3079         chunk_root->commit_root = btrfs_root_node(chunk_root);
3080
3081         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3082                            offsetof(struct btrfs_header, chunk_tree_uuid),
3083                            BTRFS_UUID_SIZE);
3084
3085         ret = btrfs_read_chunk_tree(fs_info);
3086         if (ret) {
3087                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3088                 goto fail_tree_roots;
3089         }
3090
3091         /*
3092          * Keep the devid that is marked to be the target device for the
3093          * device replace procedure
3094          */
3095         btrfs_free_extra_devids(fs_devices, 0);
3096
3097         if (!fs_devices->latest_bdev) {
3098                 btrfs_err(fs_info, "failed to read devices");
3099                 goto fail_tree_roots;
3100         }
3101
3102         ret = init_tree_roots(fs_info);
3103         if (ret)
3104                 goto fail_tree_roots;
3105
3106         /*
3107          * If we have a uuid root and we're not being told to rescan we need to
3108          * check the generation here so we can set the
3109          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3110          * transaction during a balance or the log replay without updating the
3111          * uuid generation, and then if we crash we would rescan the uuid tree,
3112          * even though it was perfectly fine.
3113          */
3114         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3115             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3116                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3117
3118         ret = btrfs_verify_dev_extents(fs_info);
3119         if (ret) {
3120                 btrfs_err(fs_info,
3121                           "failed to verify dev extents against chunks: %d",
3122                           ret);
3123                 goto fail_block_groups;
3124         }
3125         ret = btrfs_recover_balance(fs_info);
3126         if (ret) {
3127                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3128                 goto fail_block_groups;
3129         }
3130
3131         ret = btrfs_init_dev_stats(fs_info);
3132         if (ret) {
3133                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3134                 goto fail_block_groups;
3135         }
3136
3137         ret = btrfs_init_dev_replace(fs_info);
3138         if (ret) {
3139                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3140                 goto fail_block_groups;
3141         }
3142
3143         btrfs_free_extra_devids(fs_devices, 1);
3144
3145         ret = btrfs_sysfs_add_fsid(fs_devices);
3146         if (ret) {
3147                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3148                                 ret);
3149                 goto fail_block_groups;
3150         }
3151
3152         ret = btrfs_sysfs_add_mounted(fs_info);
3153         if (ret) {
3154                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3155                 goto fail_fsdev_sysfs;
3156         }
3157
3158         ret = btrfs_init_space_info(fs_info);
3159         if (ret) {
3160                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3161                 goto fail_sysfs;
3162         }
3163
3164         ret = btrfs_read_block_groups(fs_info);
3165         if (ret) {
3166                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3167                 goto fail_sysfs;
3168         }
3169
3170         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3171                 btrfs_warn(fs_info,
3172                 "writable mount is not allowed due to too many missing devices");
3173                 goto fail_sysfs;
3174         }
3175
3176         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3177                                                "btrfs-cleaner");
3178         if (IS_ERR(fs_info->cleaner_kthread))
3179                 goto fail_sysfs;
3180
3181         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3182                                                    tree_root,
3183                                                    "btrfs-transaction");
3184         if (IS_ERR(fs_info->transaction_kthread))
3185                 goto fail_cleaner;
3186
3187         if (!btrfs_test_opt(fs_info, NOSSD) &&
3188             !fs_info->fs_devices->rotating) {
3189                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3190         }
3191
3192         /*
3193          * Mount does not set all options immediately, we can do it now and do
3194          * not have to wait for transaction commit
3195          */
3196         btrfs_apply_pending_changes(fs_info);
3197
3198 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3199         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3200                 ret = btrfsic_mount(fs_info, fs_devices,
3201                                     btrfs_test_opt(fs_info,
3202                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3203                                     1 : 0,
3204                                     fs_info->check_integrity_print_mask);
3205                 if (ret)
3206                         btrfs_warn(fs_info,
3207                                 "failed to initialize integrity check module: %d",
3208                                 ret);
3209         }
3210 #endif
3211         ret = btrfs_read_qgroup_config(fs_info);
3212         if (ret)
3213                 goto fail_trans_kthread;
3214
3215         if (btrfs_build_ref_tree(fs_info))
3216                 btrfs_err(fs_info, "couldn't build ref tree");
3217
3218         /* do not make disk changes in broken FS or nologreplay is given */
3219         if (btrfs_super_log_root(disk_super) != 0 &&
3220             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3221                 btrfs_info(fs_info, "start tree-log replay");
3222                 ret = btrfs_replay_log(fs_info, fs_devices);
3223                 if (ret) {
3224                         err = ret;
3225                         goto fail_qgroup;
3226                 }
3227         }
3228
3229         ret = btrfs_find_orphan_roots(fs_info);
3230         if (ret)
3231                 goto fail_qgroup;
3232
3233         if (!sb_rdonly(sb)) {
3234                 ret = btrfs_cleanup_fs_roots(fs_info);
3235                 if (ret)
3236                         goto fail_qgroup;
3237
3238                 mutex_lock(&fs_info->cleaner_mutex);
3239                 ret = btrfs_recover_relocation(tree_root);
3240                 mutex_unlock(&fs_info->cleaner_mutex);
3241                 if (ret < 0) {
3242                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3243                                         ret);
3244                         err = -EINVAL;
3245                         goto fail_qgroup;
3246                 }
3247         }
3248
3249         location.objectid = BTRFS_FS_TREE_OBJECTID;
3250         location.type = BTRFS_ROOT_ITEM_KEY;
3251         location.offset = 0;
3252
3253         fs_info->fs_root = btrfs_get_fs_root(fs_info, &location, true);
3254         if (IS_ERR(fs_info->fs_root)) {
3255                 err = PTR_ERR(fs_info->fs_root);
3256                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3257                 fs_info->fs_root = NULL;
3258                 goto fail_qgroup;
3259         }
3260
3261         if (sb_rdonly(sb))
3262                 return 0;
3263
3264         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3265             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3266                 clear_free_space_tree = 1;
3267         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3268                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3269                 btrfs_warn(fs_info, "free space tree is invalid");
3270                 clear_free_space_tree = 1;
3271         }
3272
3273         if (clear_free_space_tree) {
3274                 btrfs_info(fs_info, "clearing free space tree");
3275                 ret = btrfs_clear_free_space_tree(fs_info);
3276                 if (ret) {
3277                         btrfs_warn(fs_info,
3278                                    "failed to clear free space tree: %d", ret);
3279                         close_ctree(fs_info);
3280                         return ret;
3281                 }
3282         }
3283
3284         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3285             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3286                 btrfs_info(fs_info, "creating free space tree");
3287                 ret = btrfs_create_free_space_tree(fs_info);
3288                 if (ret) {
3289                         btrfs_warn(fs_info,
3290                                 "failed to create free space tree: %d", ret);
3291                         close_ctree(fs_info);
3292                         return ret;
3293                 }
3294         }
3295
3296         down_read(&fs_info->cleanup_work_sem);
3297         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3298             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3299                 up_read(&fs_info->cleanup_work_sem);
3300                 close_ctree(fs_info);
3301                 return ret;
3302         }
3303         up_read(&fs_info->cleanup_work_sem);
3304
3305         ret = btrfs_resume_balance_async(fs_info);
3306         if (ret) {
3307                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3308                 close_ctree(fs_info);
3309                 return ret;
3310         }
3311
3312         ret = btrfs_resume_dev_replace_async(fs_info);
3313         if (ret) {
3314                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3315                 close_ctree(fs_info);
3316                 return ret;
3317         }
3318
3319         btrfs_qgroup_rescan_resume(fs_info);
3320         btrfs_discard_resume(fs_info);
3321
3322         if (!fs_info->uuid_root) {
3323                 btrfs_info(fs_info, "creating UUID tree");
3324                 ret = btrfs_create_uuid_tree(fs_info);
3325                 if (ret) {
3326                         btrfs_warn(fs_info,
3327                                 "failed to create the UUID tree: %d", ret);
3328                         close_ctree(fs_info);
3329                         return ret;
3330                 }
3331         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3332                    fs_info->generation !=
3333                                 btrfs_super_uuid_tree_generation(disk_super)) {
3334                 btrfs_info(fs_info, "checking UUID tree");
3335                 ret = btrfs_check_uuid_tree(fs_info);
3336                 if (ret) {
3337                         btrfs_warn(fs_info,
3338                                 "failed to check the UUID tree: %d", ret);
3339                         close_ctree(fs_info);
3340                         return ret;
3341                 }
3342         }
3343         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3344
3345         /*
3346          * backuproot only affect mount behavior, and if open_ctree succeeded,
3347          * no need to keep the flag
3348          */
3349         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3350
3351         return 0;
3352
3353 fail_qgroup:
3354         btrfs_free_qgroup_config(fs_info);
3355 fail_trans_kthread:
3356         kthread_stop(fs_info->transaction_kthread);
3357         btrfs_cleanup_transaction(fs_info);
3358         btrfs_free_fs_roots(fs_info);
3359 fail_cleaner:
3360         kthread_stop(fs_info->cleaner_kthread);
3361
3362         /*
3363          * make sure we're done with the btree inode before we stop our
3364          * kthreads
3365          */
3366         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3367
3368 fail_sysfs:
3369         btrfs_sysfs_remove_mounted(fs_info);
3370
3371 fail_fsdev_sysfs:
3372         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3373
3374 fail_block_groups:
3375         btrfs_put_block_group_cache(fs_info);
3376
3377 fail_tree_roots:
3378         free_root_pointers(fs_info, true);
3379         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3380
3381 fail_sb_buffer:
3382         btrfs_stop_all_workers(fs_info);
3383         btrfs_free_block_groups(fs_info);
3384 fail_alloc:
3385         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3386
3387         iput(fs_info->btree_inode);
3388 fail:
3389         btrfs_close_devices(fs_info->fs_devices);
3390         return err;
3391 }
3392 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3393
3394 static void btrfs_end_super_write(struct bio *bio)
3395 {
3396         struct btrfs_device *device = bio->bi_private;
3397         struct bio_vec *bvec;
3398         struct bvec_iter_all iter_all;
3399         struct page *page;
3400
3401         bio_for_each_segment_all(bvec, bio, iter_all) {
3402                 page = bvec->bv_page;
3403
3404                 if (bio->bi_status) {
3405                         btrfs_warn_rl_in_rcu(device->fs_info,
3406                                 "lost page write due to IO error on %s (%d)",
3407                                 rcu_str_deref(device->name),
3408                                 blk_status_to_errno(bio->bi_status));
3409                         ClearPageUptodate(page);
3410                         SetPageError(page);
3411                         btrfs_dev_stat_inc_and_print(device,
3412                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3413                 } else {
3414                         SetPageUptodate(page);
3415                 }
3416
3417                 put_page(page);
3418                 unlock_page(page);
3419         }
3420
3421         bio_put(bio);
3422 }
3423
3424 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3425                                                    int copy_num)
3426 {
3427         struct btrfs_super_block *super;
3428         struct page *page;
3429         u64 bytenr;
3430         struct address_space *mapping = bdev->bd_inode->i_mapping;
3431
3432         bytenr = btrfs_sb_offset(copy_num);
3433         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3434                 return ERR_PTR(-EINVAL);
3435
3436         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3437         if (IS_ERR(page))
3438                 return ERR_CAST(page);
3439
3440         super = page_address(page);
3441         if (btrfs_super_bytenr(super) != bytenr ||
3442                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3443                 btrfs_release_disk_super(super);
3444                 return ERR_PTR(-EINVAL);
3445         }
3446
3447         return super;
3448 }
3449
3450
3451 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3452 {
3453         struct btrfs_super_block *super, *latest = NULL;
3454         int i;
3455         u64 transid = 0;
3456
3457         /* we would like to check all the supers, but that would make
3458          * a btrfs mount succeed after a mkfs from a different FS.
3459          * So, we need to add a special mount option to scan for
3460          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3461          */
3462         for (i = 0; i < 1; i++) {
3463                 super = btrfs_read_dev_one_super(bdev, i);
3464                 if (IS_ERR(super))
3465                         continue;
3466
3467                 if (!latest || btrfs_super_generation(super) > transid) {
3468                         if (latest)
3469                                 btrfs_release_disk_super(super);
3470
3471                         latest = super;
3472                         transid = btrfs_super_generation(super);
3473                 }
3474         }
3475
3476         return super;
3477 }
3478
3479 /*
3480  * Write superblock @sb to the @device. Do not wait for completion, all the
3481  * pages we use for writing are locked.
3482  *
3483  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3484  * the expected device size at commit time. Note that max_mirrors must be
3485  * same for write and wait phases.
3486  *
3487  * Return number of errors when page is not found or submission fails.
3488  */
3489 static int write_dev_supers(struct btrfs_device *device,
3490                             struct btrfs_super_block *sb, int max_mirrors)
3491 {
3492         struct btrfs_fs_info *fs_info = device->fs_info;
3493         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3494         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3495         int i;
3496         int errors = 0;
3497         u64 bytenr;
3498
3499         if (max_mirrors == 0)
3500                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3501
3502         shash->tfm = fs_info->csum_shash;
3503
3504         for (i = 0; i < max_mirrors; i++) {
3505                 struct page *page;
3506                 struct bio *bio;
3507                 struct btrfs_super_block *disk_super;
3508
3509                 bytenr = btrfs_sb_offset(i);
3510                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3511                     device->commit_total_bytes)
3512                         break;
3513
3514                 btrfs_set_super_bytenr(sb, bytenr);
3515
3516                 crypto_shash_init(shash);
3517                 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3518                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3519                 crypto_shash_final(shash, sb->csum);
3520
3521                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3522                                            GFP_NOFS);
3523                 if (!page) {
3524                         btrfs_err(device->fs_info,
3525                             "couldn't get super block page for bytenr %llu",
3526                             bytenr);
3527                         errors++;
3528                         continue;
3529                 }
3530
3531                 /* Bump the refcount for wait_dev_supers() */
3532                 get_page(page);
3533
3534                 disk_super = page_address(page);
3535                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3536
3537                 /*
3538                  * Directly use bios here instead of relying on the page cache
3539                  * to do I/O, so we don't lose the ability to do integrity
3540                  * checking.
3541                  */
3542                 bio = bio_alloc(GFP_NOFS, 1);
3543                 bio_set_dev(bio, device->bdev);
3544                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3545                 bio->bi_private = device;
3546                 bio->bi_end_io = btrfs_end_super_write;
3547                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3548                                offset_in_page(bytenr));
3549
3550                 /*
3551                  * We FUA only the first super block.  The others we allow to
3552                  * go down lazy and there's a short window where the on-disk
3553                  * copies might still contain the older version.
3554                  */
3555                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3556                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3557                         bio->bi_opf |= REQ_FUA;
3558
3559                 btrfsic_submit_bio(bio);
3560         }
3561         return errors < i ? 0 : -1;
3562 }
3563
3564 /*
3565  * Wait for write completion of superblocks done by write_dev_supers,
3566  * @max_mirrors same for write and wait phases.
3567  *
3568  * Return number of errors when page is not found or not marked up to
3569  * date.
3570  */
3571 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3572 {
3573         int i;
3574         int errors = 0;
3575         bool primary_failed = false;
3576         u64 bytenr;
3577
3578         if (max_mirrors == 0)
3579                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3580
3581         for (i = 0; i < max_mirrors; i++) {
3582                 struct page *page;
3583
3584                 bytenr = btrfs_sb_offset(i);
3585                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3586                     device->commit_total_bytes)
3587                         break;
3588
3589                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3590                                      bytenr >> PAGE_SHIFT);
3591                 if (!page) {
3592                         errors++;
3593                         if (i == 0)
3594                                 primary_failed = true;
3595                         continue;
3596                 }
3597                 /* Page is submitted locked and unlocked once the IO completes */
3598                 wait_on_page_locked(page);
3599                 if (PageError(page)) {
3600                         errors++;
3601                         if (i == 0)
3602                                 primary_failed = true;
3603                 }
3604
3605                 /* Drop our reference */
3606                 put_page(page);
3607
3608                 /* Drop the reference from the writing run */
3609                 put_page(page);
3610         }
3611
3612         /* log error, force error return */
3613         if (primary_failed) {
3614                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3615                           device->devid);
3616                 return -1;
3617         }
3618
3619         return errors < i ? 0 : -1;
3620 }
3621
3622 /*
3623  * endio for the write_dev_flush, this will wake anyone waiting
3624  * for the barrier when it is done
3625  */
3626 static void btrfs_end_empty_barrier(struct bio *bio)
3627 {
3628         complete(bio->bi_private);
3629 }
3630
3631 /*
3632  * Submit a flush request to the device if it supports it. Error handling is
3633  * done in the waiting counterpart.
3634  */
3635 static void write_dev_flush(struct btrfs_device *device)
3636 {
3637         struct request_queue *q = bdev_get_queue(device->bdev);
3638         struct bio *bio = device->flush_bio;
3639
3640         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3641                 return;
3642
3643         bio_reset(bio);
3644         bio->bi_end_io = btrfs_end_empty_barrier;
3645         bio_set_dev(bio, device->bdev);
3646         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3647         init_completion(&device->flush_wait);
3648         bio->bi_private = &device->flush_wait;
3649
3650         btrfsic_submit_bio(bio);
3651         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3652 }
3653
3654 /*
3655  * If the flush bio has been submitted by write_dev_flush, wait for it.
3656  */
3657 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3658 {
3659         struct bio *bio = device->flush_bio;
3660
3661         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3662                 return BLK_STS_OK;
3663
3664         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3665         wait_for_completion_io(&device->flush_wait);
3666
3667         return bio->bi_status;
3668 }
3669
3670 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3671 {
3672         if (!btrfs_check_rw_degradable(fs_info, NULL))
3673                 return -EIO;
3674         return 0;
3675 }
3676
3677 /*
3678  * send an empty flush down to each device in parallel,
3679  * then wait for them
3680  */
3681 static int barrier_all_devices(struct btrfs_fs_info *info)
3682 {
3683         struct list_head *head;
3684         struct btrfs_device *dev;
3685         int errors_wait = 0;
3686         blk_status_t ret;
3687
3688         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3689         /* send down all the barriers */
3690         head = &info->fs_devices->devices;
3691         list_for_each_entry(dev, head, dev_list) {
3692                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3693                         continue;
3694                 if (!dev->bdev)
3695                         continue;
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                 write_dev_flush(dev);
3701                 dev->last_flush_error = BLK_STS_OK;
3702         }
3703
3704         /* wait for all the barriers */
3705         list_for_each_entry(dev, head, dev_list) {
3706                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3707                         continue;
3708                 if (!dev->bdev) {
3709                         errors_wait++;
3710                         continue;
3711                 }
3712                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3713                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3714                         continue;
3715
3716                 ret = wait_dev_flush(dev);
3717                 if (ret) {
3718                         dev->last_flush_error = ret;
3719                         btrfs_dev_stat_inc_and_print(dev,
3720                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3721                         errors_wait++;
3722                 }
3723         }
3724
3725         if (errors_wait) {
3726                 /*
3727                  * At some point we need the status of all disks
3728                  * to arrive at the volume status. So error checking
3729                  * is being pushed to a separate loop.
3730                  */
3731                 return check_barrier_error(info);
3732         }
3733         return 0;
3734 }
3735
3736 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3737 {
3738         int raid_type;
3739         int min_tolerated = INT_MAX;
3740
3741         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3742             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3743                 min_tolerated = min_t(int, min_tolerated,
3744                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3745                                     tolerated_failures);
3746
3747         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3748                 if (raid_type == BTRFS_RAID_SINGLE)
3749                         continue;
3750                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3751                         continue;
3752                 min_tolerated = min_t(int, min_tolerated,
3753                                     btrfs_raid_array[raid_type].
3754                                     tolerated_failures);
3755         }
3756
3757         if (min_tolerated == INT_MAX) {
3758                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3759                 min_tolerated = 0;
3760         }
3761
3762         return min_tolerated;
3763 }
3764
3765 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3766 {
3767         struct list_head *head;
3768         struct btrfs_device *dev;
3769         struct btrfs_super_block *sb;
3770         struct btrfs_dev_item *dev_item;
3771         int ret;
3772         int do_barriers;
3773         int max_errors;
3774         int total_errors = 0;
3775         u64 flags;
3776
3777         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3778
3779         /*
3780          * max_mirrors == 0 indicates we're from commit_transaction,
3781          * not from fsync where the tree roots in fs_info have not
3782          * been consistent on disk.
3783          */
3784         if (max_mirrors == 0)
3785                 backup_super_roots(fs_info);
3786
3787         sb = fs_info->super_for_commit;
3788         dev_item = &sb->dev_item;
3789
3790         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3791         head = &fs_info->fs_devices->devices;
3792         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3793
3794         if (do_barriers) {
3795                 ret = barrier_all_devices(fs_info);
3796                 if (ret) {
3797                         mutex_unlock(
3798                                 &fs_info->fs_devices->device_list_mutex);
3799                         btrfs_handle_fs_error(fs_info, ret,
3800                                               "errors while submitting device barriers.");
3801                         return ret;
3802                 }
3803         }
3804
3805         list_for_each_entry(dev, head, dev_list) {
3806                 if (!dev->bdev) {
3807                         total_errors++;
3808                         continue;
3809                 }
3810                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3811                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3812                         continue;
3813
3814                 btrfs_set_stack_device_generation(dev_item, 0);
3815                 btrfs_set_stack_device_type(dev_item, dev->type);
3816                 btrfs_set_stack_device_id(dev_item, dev->devid);
3817                 btrfs_set_stack_device_total_bytes(dev_item,
3818                                                    dev->commit_total_bytes);
3819                 btrfs_set_stack_device_bytes_used(dev_item,
3820                                                   dev->commit_bytes_used);
3821                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3822                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3823                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3824                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3825                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3826                        BTRFS_FSID_SIZE);
3827
3828                 flags = btrfs_super_flags(sb);
3829                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3830
3831                 ret = btrfs_validate_write_super(fs_info, sb);
3832                 if (ret < 0) {
3833                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3834                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3835                                 "unexpected superblock corruption detected");
3836                         return -EUCLEAN;
3837                 }
3838
3839                 ret = write_dev_supers(dev, sb, max_mirrors);
3840                 if (ret)
3841                         total_errors++;
3842         }
3843         if (total_errors > max_errors) {
3844                 btrfs_err(fs_info, "%d errors while writing supers",
3845                           total_errors);
3846                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3847
3848                 /* FUA is masked off if unsupported and can't be the reason */
3849                 btrfs_handle_fs_error(fs_info, -EIO,
3850                                       "%d errors while writing supers",
3851                                       total_errors);
3852                 return -EIO;
3853         }
3854
3855         total_errors = 0;
3856         list_for_each_entry(dev, head, dev_list) {
3857                 if (!dev->bdev)
3858                         continue;
3859                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3860                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3861                         continue;
3862
3863                 ret = wait_dev_supers(dev, max_mirrors);
3864                 if (ret)
3865                         total_errors++;
3866         }
3867         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3868         if (total_errors > max_errors) {
3869                 btrfs_handle_fs_error(fs_info, -EIO,
3870                                       "%d errors while writing supers",
3871                                       total_errors);
3872                 return -EIO;
3873         }
3874         return 0;
3875 }
3876
3877 /* Drop a fs root from the radix tree and free it. */
3878 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3879                                   struct btrfs_root *root)
3880 {
3881         bool drop_ref = false;
3882
3883         spin_lock(&fs_info->fs_roots_radix_lock);
3884         radix_tree_delete(&fs_info->fs_roots_radix,
3885                           (unsigned long)root->root_key.objectid);
3886         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3887                 drop_ref = true;
3888         spin_unlock(&fs_info->fs_roots_radix_lock);
3889
3890         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3891                 btrfs_free_log(NULL, root);
3892                 if (root->reloc_root) {
3893                         btrfs_put_root(root->reloc_root);
3894                         root->reloc_root = NULL;
3895                 }
3896         }
3897
3898         if (root->free_ino_pinned)
3899                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3900         if (root->free_ino_ctl)
3901                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3902         if (root->ino_cache_inode) {
3903                 iput(root->ino_cache_inode);
3904                 root->ino_cache_inode = NULL;
3905         }
3906         if (drop_ref)
3907                 btrfs_put_root(root);
3908 }
3909
3910 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3911 {
3912         u64 root_objectid = 0;
3913         struct btrfs_root *gang[8];
3914         int i = 0;
3915         int err = 0;
3916         unsigned int ret = 0;
3917
3918         while (1) {
3919                 spin_lock(&fs_info->fs_roots_radix_lock);
3920                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3921                                              (void **)gang, root_objectid,
3922                                              ARRAY_SIZE(gang));
3923                 if (!ret) {
3924                         spin_unlock(&fs_info->fs_roots_radix_lock);
3925                         break;
3926                 }
3927                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3928
3929                 for (i = 0; i < ret; i++) {
3930                         /* Avoid to grab roots in dead_roots */
3931                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3932                                 gang[i] = NULL;
3933                                 continue;
3934                         }
3935                         /* grab all the search result for later use */
3936                         gang[i] = btrfs_grab_root(gang[i]);
3937                 }
3938                 spin_unlock(&fs_info->fs_roots_radix_lock);
3939
3940                 for (i = 0; i < ret; i++) {
3941                         if (!gang[i])
3942                                 continue;
3943                         root_objectid = gang[i]->root_key.objectid;
3944                         err = btrfs_orphan_cleanup(gang[i]);
3945                         if (err)
3946                                 break;
3947                         btrfs_put_root(gang[i]);
3948                 }
3949                 root_objectid++;
3950         }
3951
3952         /* release the uncleaned roots due to error */
3953         for (; i < ret; i++) {
3954                 if (gang[i])
3955                         btrfs_put_root(gang[i]);
3956         }
3957         return err;
3958 }
3959
3960 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3961 {
3962         struct btrfs_root *root = fs_info->tree_root;
3963         struct btrfs_trans_handle *trans;
3964
3965         mutex_lock(&fs_info->cleaner_mutex);
3966         btrfs_run_delayed_iputs(fs_info);
3967         mutex_unlock(&fs_info->cleaner_mutex);
3968         wake_up_process(fs_info->cleaner_kthread);
3969
3970         /* wait until ongoing cleanup work done */
3971         down_write(&fs_info->cleanup_work_sem);
3972         up_write(&fs_info->cleanup_work_sem);
3973
3974         trans = btrfs_join_transaction(root);
3975         if (IS_ERR(trans))
3976                 return PTR_ERR(trans);
3977         return btrfs_commit_transaction(trans);
3978 }
3979
3980 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3981 {
3982         int ret;
3983
3984         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3985         /*
3986          * We don't want the cleaner to start new transactions, add more delayed
3987          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3988          * because that frees the task_struct, and the transaction kthread might
3989          * still try to wake up the cleaner.
3990          */
3991         kthread_park(fs_info->cleaner_kthread);
3992
3993         /* wait for the qgroup rescan worker to stop */
3994         btrfs_qgroup_wait_for_completion(fs_info, false);
3995
3996         /* wait for the uuid_scan task to finish */
3997         down(&fs_info->uuid_tree_rescan_sem);
3998         /* avoid complains from lockdep et al., set sem back to initial state */
3999         up(&fs_info->uuid_tree_rescan_sem);
4000
4001         /* pause restriper - we want to resume on mount */
4002         btrfs_pause_balance(fs_info);
4003
4004         btrfs_dev_replace_suspend_for_unmount(fs_info);
4005
4006         btrfs_scrub_cancel(fs_info);
4007
4008         /* wait for any defraggers to finish */
4009         wait_event(fs_info->transaction_wait,
4010                    (atomic_read(&fs_info->defrag_running) == 0));
4011
4012         /* clear out the rbtree of defraggable inodes */
4013         btrfs_cleanup_defrag_inodes(fs_info);
4014
4015         cancel_work_sync(&fs_info->async_reclaim_work);
4016
4017         /* Cancel or finish ongoing discard work */
4018         btrfs_discard_cleanup(fs_info);
4019
4020         if (!sb_rdonly(fs_info->sb)) {
4021                 /*
4022                  * The cleaner kthread is stopped, so do one final pass over
4023                  * unused block groups.
4024                  */
4025                 btrfs_delete_unused_bgs(fs_info);
4026
4027                 /*
4028                  * There might be existing delayed inode workers still running
4029                  * and holding an empty delayed inode item. We must wait for
4030                  * them to complete first because they can create a transaction.
4031                  * This happens when someone calls btrfs_balance_delayed_items()
4032                  * and then a transaction commit runs the same delayed nodes
4033                  * before any delayed worker has done something with the nodes.
4034                  * We must wait for any worker here and not at transaction
4035                  * commit time since that could cause a deadlock.
4036                  * This is a very rare case.
4037                  */
4038                 btrfs_flush_workqueue(fs_info->delayed_workers);
4039
4040                 ret = btrfs_commit_super(fs_info);
4041                 if (ret)
4042                         btrfs_err(fs_info, "commit super ret %d", ret);
4043         }
4044
4045         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4046             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4047                 btrfs_error_commit_super(fs_info);
4048
4049         kthread_stop(fs_info->transaction_kthread);
4050         kthread_stop(fs_info->cleaner_kthread);
4051
4052         ASSERT(list_empty(&fs_info->delayed_iputs));
4053         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4054
4055         btrfs_free_qgroup_config(fs_info);
4056         ASSERT(list_empty(&fs_info->delalloc_roots));
4057
4058         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4059                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4060                        percpu_counter_sum(&fs_info->delalloc_bytes));
4061         }
4062
4063         if (percpu_counter_sum(&fs_info->dio_bytes))
4064                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4065                            percpu_counter_sum(&fs_info->dio_bytes));
4066
4067         btrfs_sysfs_remove_mounted(fs_info);
4068         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4069
4070         btrfs_put_block_group_cache(fs_info);
4071
4072         /*
4073          * we must make sure there is not any read request to
4074          * submit after we stopping all workers.
4075          */
4076         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4077         btrfs_stop_all_workers(fs_info);
4078
4079         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4080         free_root_pointers(fs_info, true);
4081         btrfs_free_fs_roots(fs_info);
4082
4083         /*
4084          * We must free the block groups after dropping the fs_roots as we could
4085          * have had an IO error and have left over tree log blocks that aren't
4086          * cleaned up until the fs roots are freed.  This makes the block group
4087          * accounting appear to be wrong because there's pending reserved bytes,
4088          * so make sure we do the block group cleanup afterwards.
4089          */
4090         btrfs_free_block_groups(fs_info);
4091
4092         iput(fs_info->btree_inode);
4093
4094 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4095         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4096                 btrfsic_unmount(fs_info->fs_devices);
4097 #endif
4098
4099         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4100         btrfs_close_devices(fs_info->fs_devices);
4101 }
4102
4103 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4104                           int atomic)
4105 {
4106         int ret;
4107         struct inode *btree_inode = buf->pages[0]->mapping->host;
4108
4109         ret = extent_buffer_uptodate(buf);
4110         if (!ret)
4111                 return ret;
4112
4113         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4114                                     parent_transid, atomic);
4115         if (ret == -EAGAIN)
4116                 return ret;
4117         return !ret;
4118 }
4119
4120 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4121 {
4122         struct btrfs_fs_info *fs_info;
4123         struct btrfs_root *root;
4124         u64 transid = btrfs_header_generation(buf);
4125         int was_dirty;
4126
4127 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4128         /*
4129          * This is a fast path so only do this check if we have sanity tests
4130          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4131          * outside of the sanity tests.
4132          */
4133         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4134                 return;
4135 #endif
4136         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4137         fs_info = root->fs_info;
4138         btrfs_assert_tree_locked(buf);
4139         if (transid != fs_info->generation)
4140                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4141                         buf->start, transid, fs_info->generation);
4142         was_dirty = set_extent_buffer_dirty(buf);
4143         if (!was_dirty)
4144                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4145                                          buf->len,
4146                                          fs_info->dirty_metadata_batch);
4147 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4148         /*
4149          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4150          * but item data not updated.
4151          * So here we should only check item pointers, not item data.
4152          */
4153         if (btrfs_header_level(buf) == 0 &&
4154             btrfs_check_leaf_relaxed(buf)) {
4155                 btrfs_print_leaf(buf);
4156                 ASSERT(0);
4157         }
4158 #endif
4159 }
4160
4161 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4162                                         int flush_delayed)
4163 {
4164         /*
4165          * looks as though older kernels can get into trouble with
4166          * this code, they end up stuck in balance_dirty_pages forever
4167          */
4168         int ret;
4169
4170         if (current->flags & PF_MEMALLOC)
4171                 return;
4172
4173         if (flush_delayed)
4174                 btrfs_balance_delayed_items(fs_info);
4175
4176         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4177                                      BTRFS_DIRTY_METADATA_THRESH,
4178                                      fs_info->dirty_metadata_batch);
4179         if (ret > 0) {
4180                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4181         }
4182 }
4183
4184 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4185 {
4186         __btrfs_btree_balance_dirty(fs_info, 1);
4187 }
4188
4189 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4190 {
4191         __btrfs_btree_balance_dirty(fs_info, 0);
4192 }
4193
4194 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4195                       struct btrfs_key *first_key)
4196 {
4197         return btree_read_extent_buffer_pages(buf, parent_transid,
4198                                               level, first_key);
4199 }
4200
4201 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4202 {
4203         /* cleanup FS via transaction */
4204         btrfs_cleanup_transaction(fs_info);
4205
4206         mutex_lock(&fs_info->cleaner_mutex);
4207         btrfs_run_delayed_iputs(fs_info);
4208         mutex_unlock(&fs_info->cleaner_mutex);
4209
4210         down_write(&fs_info->cleanup_work_sem);
4211         up_write(&fs_info->cleanup_work_sem);
4212 }
4213
4214 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4215 {
4216         struct btrfs_ordered_extent *ordered;
4217
4218         spin_lock(&root->ordered_extent_lock);
4219         /*
4220          * This will just short circuit the ordered completion stuff which will
4221          * make sure the ordered extent gets properly cleaned up.
4222          */
4223         list_for_each_entry(ordered, &root->ordered_extents,
4224                             root_extent_list)
4225                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4226         spin_unlock(&root->ordered_extent_lock);
4227 }
4228
4229 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4230 {
4231         struct btrfs_root *root;
4232         struct list_head splice;
4233
4234         INIT_LIST_HEAD(&splice);
4235
4236         spin_lock(&fs_info->ordered_root_lock);
4237         list_splice_init(&fs_info->ordered_roots, &splice);
4238         while (!list_empty(&splice)) {
4239                 root = list_first_entry(&splice, struct btrfs_root,
4240                                         ordered_root);
4241                 list_move_tail(&root->ordered_root,
4242                                &fs_info->ordered_roots);
4243
4244                 spin_unlock(&fs_info->ordered_root_lock);
4245                 btrfs_destroy_ordered_extents(root);
4246
4247                 cond_resched();
4248                 spin_lock(&fs_info->ordered_root_lock);
4249         }
4250         spin_unlock(&fs_info->ordered_root_lock);
4251
4252         /*
4253          * We need this here because if we've been flipped read-only we won't
4254          * get sync() from the umount, so we need to make sure any ordered
4255          * extents that haven't had their dirty pages IO start writeout yet
4256          * actually get run and error out properly.
4257          */
4258         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4259 }
4260
4261 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4262                                       struct btrfs_fs_info *fs_info)
4263 {
4264         struct rb_node *node;
4265         struct btrfs_delayed_ref_root *delayed_refs;
4266         struct btrfs_delayed_ref_node *ref;
4267         int ret = 0;
4268
4269         delayed_refs = &trans->delayed_refs;
4270
4271         spin_lock(&delayed_refs->lock);
4272         if (atomic_read(&delayed_refs->num_entries) == 0) {
4273                 spin_unlock(&delayed_refs->lock);
4274                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4275                 return ret;
4276         }
4277
4278         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4279                 struct btrfs_delayed_ref_head *head;
4280                 struct rb_node *n;
4281                 bool pin_bytes = false;
4282
4283                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4284                                 href_node);
4285                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4286                         continue;
4287
4288                 spin_lock(&head->lock);
4289                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4290                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4291                                        ref_node);
4292                         ref->in_tree = 0;
4293                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4294                         RB_CLEAR_NODE(&ref->ref_node);
4295                         if (!list_empty(&ref->add_list))
4296                                 list_del(&ref->add_list);
4297                         atomic_dec(&delayed_refs->num_entries);
4298                         btrfs_put_delayed_ref(ref);
4299                 }
4300                 if (head->must_insert_reserved)
4301                         pin_bytes = true;
4302                 btrfs_free_delayed_extent_op(head->extent_op);
4303                 btrfs_delete_ref_head(delayed_refs, head);
4304                 spin_unlock(&head->lock);
4305                 spin_unlock(&delayed_refs->lock);
4306                 mutex_unlock(&head->mutex);
4307
4308                 if (pin_bytes) {
4309                         struct btrfs_block_group *cache;
4310
4311                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4312                         BUG_ON(!cache);
4313
4314                         spin_lock(&cache->space_info->lock);
4315                         spin_lock(&cache->lock);
4316                         cache->pinned += head->num_bytes;
4317                         btrfs_space_info_update_bytes_pinned(fs_info,
4318                                 cache->space_info, head->num_bytes);
4319                         cache->reserved -= head->num_bytes;
4320                         cache->space_info->bytes_reserved -= head->num_bytes;
4321                         spin_unlock(&cache->lock);
4322                         spin_unlock(&cache->space_info->lock);
4323                         percpu_counter_add_batch(
4324                                 &cache->space_info->total_bytes_pinned,
4325                                 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4326
4327                         btrfs_put_block_group(cache);
4328
4329                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4330                                 head->bytenr + head->num_bytes - 1);
4331                 }
4332                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4333                 btrfs_put_delayed_ref_head(head);
4334                 cond_resched();
4335                 spin_lock(&delayed_refs->lock);
4336         }
4337         btrfs_qgroup_destroy_extent_records(trans);
4338
4339         spin_unlock(&delayed_refs->lock);
4340
4341         return ret;
4342 }
4343
4344 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4345 {
4346         struct btrfs_inode *btrfs_inode;
4347         struct list_head splice;
4348
4349         INIT_LIST_HEAD(&splice);
4350
4351         spin_lock(&root->delalloc_lock);
4352         list_splice_init(&root->delalloc_inodes, &splice);
4353
4354         while (!list_empty(&splice)) {
4355                 struct inode *inode = NULL;
4356                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4357                                                delalloc_inodes);
4358                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4359                 spin_unlock(&root->delalloc_lock);
4360
4361                 /*
4362                  * Make sure we get a live inode and that it'll not disappear
4363                  * meanwhile.
4364                  */
4365                 inode = igrab(&btrfs_inode->vfs_inode);
4366                 if (inode) {
4367                         invalidate_inode_pages2(inode->i_mapping);
4368                         iput(inode);
4369                 }
4370                 spin_lock(&root->delalloc_lock);
4371         }
4372         spin_unlock(&root->delalloc_lock);
4373 }
4374
4375 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4376 {
4377         struct btrfs_root *root;
4378         struct list_head splice;
4379
4380         INIT_LIST_HEAD(&splice);
4381
4382         spin_lock(&fs_info->delalloc_root_lock);
4383         list_splice_init(&fs_info->delalloc_roots, &splice);
4384         while (!list_empty(&splice)) {
4385                 root = list_first_entry(&splice, struct btrfs_root,
4386                                          delalloc_root);
4387                 root = btrfs_grab_root(root);
4388                 BUG_ON(!root);
4389                 spin_unlock(&fs_info->delalloc_root_lock);
4390
4391                 btrfs_destroy_delalloc_inodes(root);
4392                 btrfs_put_root(root);
4393
4394                 spin_lock(&fs_info->delalloc_root_lock);
4395         }
4396         spin_unlock(&fs_info->delalloc_root_lock);
4397 }
4398
4399 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4400                                         struct extent_io_tree *dirty_pages,
4401                                         int mark)
4402 {
4403         int ret;
4404         struct extent_buffer *eb;
4405         u64 start = 0;
4406         u64 end;
4407
4408         while (1) {
4409                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4410                                             mark, NULL);
4411                 if (ret)
4412                         break;
4413
4414                 clear_extent_bits(dirty_pages, start, end, mark);
4415                 while (start <= end) {
4416                         eb = find_extent_buffer(fs_info, start);
4417                         start += fs_info->nodesize;
4418                         if (!eb)
4419                                 continue;
4420                         wait_on_extent_buffer_writeback(eb);
4421
4422                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4423                                                &eb->bflags))
4424                                 clear_extent_buffer_dirty(eb);
4425                         free_extent_buffer_stale(eb);
4426                 }
4427         }
4428
4429         return ret;
4430 }
4431
4432 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4433                                        struct extent_io_tree *unpin)
4434 {
4435         u64 start;
4436         u64 end;
4437         int ret;
4438
4439         while (1) {
4440                 struct extent_state *cached_state = NULL;
4441
4442                 /*
4443                  * The btrfs_finish_extent_commit() may get the same range as
4444                  * ours between find_first_extent_bit and clear_extent_dirty.
4445                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4446                  * the same extent range.
4447                  */
4448                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4449                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4450                                             EXTENT_DIRTY, &cached_state);
4451                 if (ret) {
4452                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4453                         break;
4454                 }
4455
4456                 clear_extent_dirty(unpin, start, end, &cached_state);
4457                 free_extent_state(cached_state);
4458                 btrfs_error_unpin_extent_range(fs_info, start, end);
4459                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4460                 cond_resched();
4461         }
4462
4463         return 0;
4464 }
4465
4466 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4467 {
4468         struct inode *inode;
4469
4470         inode = cache->io_ctl.inode;
4471         if (inode) {
4472                 invalidate_inode_pages2(inode->i_mapping);
4473                 BTRFS_I(inode)->generation = 0;
4474                 cache->io_ctl.inode = NULL;
4475                 iput(inode);
4476         }
4477         btrfs_put_block_group(cache);
4478 }
4479
4480 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4481                              struct btrfs_fs_info *fs_info)
4482 {
4483         struct btrfs_block_group *cache;
4484
4485         spin_lock(&cur_trans->dirty_bgs_lock);
4486         while (!list_empty(&cur_trans->dirty_bgs)) {
4487                 cache = list_first_entry(&cur_trans->dirty_bgs,
4488                                          struct btrfs_block_group,
4489                                          dirty_list);
4490
4491                 if (!list_empty(&cache->io_list)) {
4492                         spin_unlock(&cur_trans->dirty_bgs_lock);
4493                         list_del_init(&cache->io_list);
4494                         btrfs_cleanup_bg_io(cache);
4495                         spin_lock(&cur_trans->dirty_bgs_lock);
4496                 }
4497
4498                 list_del_init(&cache->dirty_list);
4499                 spin_lock(&cache->lock);
4500                 cache->disk_cache_state = BTRFS_DC_ERROR;
4501                 spin_unlock(&cache->lock);
4502
4503                 spin_unlock(&cur_trans->dirty_bgs_lock);
4504                 btrfs_put_block_group(cache);
4505                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4506                 spin_lock(&cur_trans->dirty_bgs_lock);
4507         }
4508         spin_unlock(&cur_trans->dirty_bgs_lock);
4509
4510         /*
4511          * Refer to the definition of io_bgs member for details why it's safe
4512          * to use it without any locking
4513          */
4514         while (!list_empty(&cur_trans->io_bgs)) {
4515                 cache = list_first_entry(&cur_trans->io_bgs,
4516                                          struct btrfs_block_group,
4517                                          io_list);
4518
4519                 list_del_init(&cache->io_list);
4520                 spin_lock(&cache->lock);
4521                 cache->disk_cache_state = BTRFS_DC_ERROR;
4522                 spin_unlock(&cache->lock);
4523                 btrfs_cleanup_bg_io(cache);
4524         }
4525 }
4526
4527 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4528                                    struct btrfs_fs_info *fs_info)
4529 {
4530         struct btrfs_device *dev, *tmp;
4531
4532         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4533         ASSERT(list_empty(&cur_trans->dirty_bgs));
4534         ASSERT(list_empty(&cur_trans->io_bgs));
4535
4536         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4537                                  post_commit_list) {
4538                 list_del_init(&dev->post_commit_list);
4539         }
4540
4541         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4542
4543         cur_trans->state = TRANS_STATE_COMMIT_START;
4544         wake_up(&fs_info->transaction_blocked_wait);
4545
4546         cur_trans->state = TRANS_STATE_UNBLOCKED;
4547         wake_up(&fs_info->transaction_wait);
4548
4549         btrfs_destroy_delayed_inodes(fs_info);
4550
4551         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4552                                      EXTENT_DIRTY);
4553         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4554
4555         cur_trans->state =TRANS_STATE_COMPLETED;
4556         wake_up(&cur_trans->commit_wait);
4557 }
4558
4559 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4560 {
4561         struct btrfs_transaction *t;
4562
4563         mutex_lock(&fs_info->transaction_kthread_mutex);
4564
4565         spin_lock(&fs_info->trans_lock);
4566         while (!list_empty(&fs_info->trans_list)) {
4567                 t = list_first_entry(&fs_info->trans_list,
4568                                      struct btrfs_transaction, list);
4569                 if (t->state >= TRANS_STATE_COMMIT_START) {
4570                         refcount_inc(&t->use_count);
4571                         spin_unlock(&fs_info->trans_lock);
4572                         btrfs_wait_for_commit(fs_info, t->transid);
4573                         btrfs_put_transaction(t);
4574                         spin_lock(&fs_info->trans_lock);
4575                         continue;
4576                 }
4577                 if (t == fs_info->running_transaction) {
4578                         t->state = TRANS_STATE_COMMIT_DOING;
4579                         spin_unlock(&fs_info->trans_lock);
4580                         /*
4581                          * We wait for 0 num_writers since we don't hold a trans
4582                          * handle open currently for this transaction.
4583                          */
4584                         wait_event(t->writer_wait,
4585                                    atomic_read(&t->num_writers) == 0);
4586                 } else {
4587                         spin_unlock(&fs_info->trans_lock);
4588                 }
4589                 btrfs_cleanup_one_transaction(t, fs_info);
4590
4591                 spin_lock(&fs_info->trans_lock);
4592                 if (t == fs_info->running_transaction)
4593                         fs_info->running_transaction = NULL;
4594                 list_del_init(&t->list);
4595                 spin_unlock(&fs_info->trans_lock);
4596
4597                 btrfs_put_transaction(t);
4598                 trace_btrfs_transaction_commit(fs_info->tree_root);
4599                 spin_lock(&fs_info->trans_lock);
4600         }
4601         spin_unlock(&fs_info->trans_lock);
4602         btrfs_destroy_all_ordered_extents(fs_info);
4603         btrfs_destroy_delayed_inodes(fs_info);
4604         btrfs_assert_delayed_root_empty(fs_info);
4605         btrfs_destroy_all_delalloc_inodes(fs_info);
4606         mutex_unlock(&fs_info->transaction_kthread_mutex);
4607
4608         return 0;
4609 }
4610
4611 static const struct extent_io_ops btree_extent_io_ops = {
4612         /* mandatory callbacks */
4613         .submit_bio_hook = btree_submit_bio_hook,
4614         .readpage_end_io_hook = btree_readpage_end_io_hook,
4615 };