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                 detach_page_private(page);
 984         }
 985 }
 986
 987 static int btree_set_page_dirty(struct page *page)
 988 {
 989 #ifdef DEBUG
 990         struct extent_buffer *eb;
 991
 992         BUG_ON(!PagePrivate(page));
 993         eb = (struct extent_buffer *)page->private;
 994         BUG_ON(!eb);
 995         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 996         BUG_ON(!atomic_read(&eb->refs));
 997         btrfs_assert_tree_locked(eb);
 998 #endif
 999         return __set_page_dirty_nobuffers(page);
1000 }
1001
1002 static const struct address_space_operations btree_aops = {
1003         .readpage       = btree_readpage,
1004         .writepages     = btree_writepages,
1005         .releasepage    = btree_releasepage,
1006         .invalidatepage = btree_invalidatepage,
1007 #ifdef CONFIG_MIGRATION
1008         .migratepage    = btree_migratepage,
1009 #endif
1010         .set_page_dirty = btree_set_page_dirty,
1011 };
1012
1013 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1014 {
1015         struct extent_buffer *buf = NULL;
1016         int ret;
1017
1018         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1019         if (IS_ERR(buf))
1020                 return;
1021
1022         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1023         if (ret < 0)
1024                 free_extent_buffer_stale(buf);
1025         else
1026                 free_extent_buffer(buf);
1027 }
1028
1029 struct extent_buffer *btrfs_find_create_tree_block(
1030                                                 struct btrfs_fs_info *fs_info,
1031                                                 u64 bytenr)
1032 {
1033         if (btrfs_is_testing(fs_info))
1034                 return alloc_test_extent_buffer(fs_info, bytenr);
1035         return alloc_extent_buffer(fs_info, bytenr);
1036 }
1037
1038 /*
1039  * Read tree block at logical address @bytenr and do variant basic but critical
1040  * verification.
1041  *
1042  * @parent_transid:     expected transid of this tree block, skip check if 0
1043  * @level:              expected level, mandatory check
1044  * @first_key:          expected key in slot 0, skip check if NULL
1045  */
1046 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1047                                       u64 parent_transid, int level,
1048                                       struct btrfs_key *first_key)
1049 {
1050         struct extent_buffer *buf = NULL;
1051         int ret;
1052
1053         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1054         if (IS_ERR(buf))
1055                 return buf;
1056
1057         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1058                                              level, first_key);
1059         if (ret) {
1060                 free_extent_buffer_stale(buf);
1061                 return ERR_PTR(ret);
1062         }
1063         return buf;
1064
1065 }
1066
1067 void btrfs_clean_tree_block(struct extent_buffer *buf)
1068 {
1069         struct btrfs_fs_info *fs_info = buf->fs_info;
1070         if (btrfs_header_generation(buf) ==
1071             fs_info->running_transaction->transid) {
1072                 btrfs_assert_tree_locked(buf);
1073
1074                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1075                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1076                                                  -buf->len,
1077                                                  fs_info->dirty_metadata_batch);
1078                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1079                         btrfs_set_lock_blocking_write(buf);
1080                         clear_extent_buffer_dirty(buf);
1081                 }
1082         }
1083 }
1084
1085 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1086                          u64 objectid)
1087 {
1088         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1089         root->fs_info = fs_info;
1090         root->node = NULL;
1091         root->commit_root = NULL;
1092         root->state = 0;
1093         root->orphan_cleanup_state = 0;
1094
1095         root->last_trans = 0;
1096         root->highest_objectid = 0;
1097         root->nr_delalloc_inodes = 0;
1098         root->nr_ordered_extents = 0;
1099         root->inode_tree = RB_ROOT;
1100         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1101         root->block_rsv = NULL;
1102
1103         INIT_LIST_HEAD(&root->dirty_list);
1104         INIT_LIST_HEAD(&root->root_list);
1105         INIT_LIST_HEAD(&root->delalloc_inodes);
1106         INIT_LIST_HEAD(&root->delalloc_root);
1107         INIT_LIST_HEAD(&root->ordered_extents);
1108         INIT_LIST_HEAD(&root->ordered_root);
1109         INIT_LIST_HEAD(&root->reloc_dirty_list);
1110         INIT_LIST_HEAD(&root->logged_list[0]);
1111         INIT_LIST_HEAD(&root->logged_list[1]);
1112         spin_lock_init(&root->inode_lock);
1113         spin_lock_init(&root->delalloc_lock);
1114         spin_lock_init(&root->ordered_extent_lock);
1115         spin_lock_init(&root->accounting_lock);
1116         spin_lock_init(&root->log_extents_lock[0]);
1117         spin_lock_init(&root->log_extents_lock[1]);
1118         spin_lock_init(&root->qgroup_meta_rsv_lock);
1119         mutex_init(&root->objectid_mutex);
1120         mutex_init(&root->log_mutex);
1121         mutex_init(&root->ordered_extent_mutex);
1122         mutex_init(&root->delalloc_mutex);
1123         init_waitqueue_head(&root->log_writer_wait);
1124         init_waitqueue_head(&root->log_commit_wait[0]);
1125         init_waitqueue_head(&root->log_commit_wait[1]);
1126         INIT_LIST_HEAD(&root->log_ctxs[0]);
1127         INIT_LIST_HEAD(&root->log_ctxs[1]);
1128         atomic_set(&root->log_commit[0], 0);
1129         atomic_set(&root->log_commit[1], 0);
1130         atomic_set(&root->log_writers, 0);
1131         atomic_set(&root->log_batch, 0);
1132         refcount_set(&root->refs, 1);
1133         atomic_set(&root->snapshot_force_cow, 0);
1134         atomic_set(&root->nr_swapfiles, 0);
1135         root->log_transid = 0;
1136         root->log_transid_committed = -1;
1137         root->last_log_commit = 0;
1138         if (!dummy)
1139                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1140                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1141
1142         memset(&root->root_key, 0, sizeof(root->root_key));
1143         memset(&root->root_item, 0, sizeof(root->root_item));
1144         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1145         if (!dummy)
1146                 root->defrag_trans_start = fs_info->generation;
1147         else
1148                 root->defrag_trans_start = 0;
1149         root->root_key.objectid = objectid;
1150         root->anon_dev = 0;
1151
1152         spin_lock_init(&root->root_item_lock);
1153         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1154 #ifdef CONFIG_BTRFS_DEBUG
1155         INIT_LIST_HEAD(&root->leak_list);
1156         spin_lock(&fs_info->fs_roots_radix_lock);
1157         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1158         spin_unlock(&fs_info->fs_roots_radix_lock);
1159 #endif
1160 }
1161
1162 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1163                                            u64 objectid, gfp_t flags)
1164 {
1165         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1166         if (root)
1167                 __setup_root(root, fs_info, objectid);
1168         return root;
1169 }
1170
1171 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1172 /* Should only be used by the testing infrastructure */
1173 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1174 {
1175         struct btrfs_root *root;
1176
1177         if (!fs_info)
1178                 return ERR_PTR(-EINVAL);
1179
1180         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1181         if (!root)
1182                 return ERR_PTR(-ENOMEM);
1183
1184         /* We don't use the stripesize in selftest, set it as sectorsize */
1185         root->alloc_bytenr = 0;
1186
1187         return root;
1188 }
1189 #endif
1190
1191 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1192                                      u64 objectid)
1193 {
1194         struct btrfs_fs_info *fs_info = trans->fs_info;
1195         struct extent_buffer *leaf;
1196         struct btrfs_root *tree_root = fs_info->tree_root;
1197         struct btrfs_root *root;
1198         struct btrfs_key key;
1199         unsigned int nofs_flag;
1200         int ret = 0;
1201
1202         /*
1203          * We're holding a transaction handle, so use a NOFS memory allocation
1204          * context to avoid deadlock if reclaim happens.
1205          */
1206         nofs_flag = memalloc_nofs_save();
1207         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1208         memalloc_nofs_restore(nofs_flag);
1209         if (!root)
1210                 return ERR_PTR(-ENOMEM);
1211
1212         root->root_key.objectid = objectid;
1213         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1214         root->root_key.offset = 0;
1215
1216         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1217         if (IS_ERR(leaf)) {
1218                 ret = PTR_ERR(leaf);
1219                 leaf = NULL;
1220                 goto fail;
1221         }
1222
1223         root->node = leaf;
1224         btrfs_mark_buffer_dirty(leaf);
1225
1226         root->commit_root = btrfs_root_node(root);
1227         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1228
1229         root->root_item.flags = 0;
1230         root->root_item.byte_limit = 0;
1231         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1232         btrfs_set_root_generation(&root->root_item, trans->transid);
1233         btrfs_set_root_level(&root->root_item, 0);
1234         btrfs_set_root_refs(&root->root_item, 1);
1235         btrfs_set_root_used(&root->root_item, leaf->len);
1236         btrfs_set_root_last_snapshot(&root->root_item, 0);
1237         btrfs_set_root_dirid(&root->root_item, 0);
1238         if (is_fstree(objectid))
1239                 generate_random_guid(root->root_item.uuid);
1240         else
1241                 export_guid(root->root_item.uuid, &guid_null);
1242         root->root_item.drop_level = 0;
1243
1244         key.objectid = objectid;
1245         key.type = BTRFS_ROOT_ITEM_KEY;
1246         key.offset = 0;
1247         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1248         if (ret)
1249                 goto fail;
1250
1251         btrfs_tree_unlock(leaf);
1252
1253         return root;
1254
1255 fail:
1256         if (leaf)
1257                 btrfs_tree_unlock(leaf);
1258         btrfs_put_root(root);
1259
1260         return ERR_PTR(ret);
1261 }
1262
1263 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1264                                          struct btrfs_fs_info *fs_info)
1265 {
1266         struct btrfs_root *root;
1267         struct extent_buffer *leaf;
1268
1269         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1270         if (!root)
1271                 return ERR_PTR(-ENOMEM);
1272
1273         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1274         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1275         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1276
1277         /*
1278          * DON'T set REF_COWS for log trees
1279          *
1280          * log trees do not get reference counted because they go away
1281          * before a real commit is actually done.  They do store pointers
1282          * to file data extents, and those reference counts still get
1283          * updated (along with back refs to the log tree).
1284          */
1285
1286         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1287                         NULL, 0, 0, 0);
1288         if (IS_ERR(leaf)) {
1289                 btrfs_put_root(root);
1290                 return ERR_CAST(leaf);
1291         }
1292
1293         root->node = leaf;
1294
1295         btrfs_mark_buffer_dirty(root->node);
1296         btrfs_tree_unlock(root->node);
1297         return root;
1298 }
1299
1300 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1301                              struct btrfs_fs_info *fs_info)
1302 {
1303         struct btrfs_root *log_root;
1304
1305         log_root = alloc_log_tree(trans, fs_info);
1306         if (IS_ERR(log_root))
1307                 return PTR_ERR(log_root);
1308         WARN_ON(fs_info->log_root_tree);
1309         fs_info->log_root_tree = log_root;
1310         return 0;
1311 }
1312
1313 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1314                        struct btrfs_root *root)
1315 {
1316         struct btrfs_fs_info *fs_info = root->fs_info;
1317         struct btrfs_root *log_root;
1318         struct btrfs_inode_item *inode_item;
1319
1320         log_root = alloc_log_tree(trans, fs_info);
1321         if (IS_ERR(log_root))
1322                 return PTR_ERR(log_root);
1323
1324         log_root->last_trans = trans->transid;
1325         log_root->root_key.offset = root->root_key.objectid;
1326
1327         inode_item = &log_root->root_item.inode;
1328         btrfs_set_stack_inode_generation(inode_item, 1);
1329         btrfs_set_stack_inode_size(inode_item, 3);
1330         btrfs_set_stack_inode_nlink(inode_item, 1);
1331         btrfs_set_stack_inode_nbytes(inode_item,
1332                                      fs_info->nodesize);
1333         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1334
1335         btrfs_set_root_node(&log_root->root_item, log_root->node);
1336
1337         WARN_ON(root->log_root);
1338         root->log_root = log_root;
1339         root->log_transid = 0;
1340         root->log_transid_committed = -1;
1341         root->last_log_commit = 0;
1342         return 0;
1343 }
1344
1345 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1346                                         struct btrfs_key *key)
1347 {
1348         struct btrfs_root *root;
1349         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1350         struct btrfs_path *path;
1351         u64 generation;
1352         int ret;
1353         int level;
1354
1355         path = btrfs_alloc_path();
1356         if (!path)
1357                 return ERR_PTR(-ENOMEM);
1358
1359         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1360         if (!root) {
1361                 ret = -ENOMEM;
1362                 goto alloc_fail;
1363         }
1364
1365         ret = btrfs_find_root(tree_root, key, path,
1366                               &root->root_item, &root->root_key);
1367         if (ret) {
1368                 if (ret > 0)
1369                         ret = -ENOENT;
1370                 goto find_fail;
1371         }
1372
1373         generation = btrfs_root_generation(&root->root_item);
1374         level = btrfs_root_level(&root->root_item);
1375         root->node = read_tree_block(fs_info,
1376                                      btrfs_root_bytenr(&root->root_item),
1377                                      generation, level, NULL);
1378         if (IS_ERR(root->node)) {
1379                 ret = PTR_ERR(root->node);
1380                 root->node = NULL;
1381                 goto find_fail;
1382         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1383                 ret = -EIO;
1384                 goto find_fail;
1385         }
1386         root->commit_root = btrfs_root_node(root);
1387 out:
1388         btrfs_free_path(path);
1389         return root;
1390
1391 find_fail:
1392         btrfs_put_root(root);
1393 alloc_fail:
1394         root = ERR_PTR(ret);
1395         goto out;
1396 }
1397
1398 static int btrfs_init_fs_root(struct btrfs_root *root)
1399 {
1400         int ret;
1401         unsigned int nofs_flag;
1402
1403         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1404         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1405                                         GFP_NOFS);
1406         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1407                 ret = -ENOMEM;
1408                 goto fail;
1409         }
1410
1411         /*
1412          * We might be called under a transaction (e.g. indirect backref
1413          * resolution) which could deadlock if it triggers memory reclaim
1414          */
1415         nofs_flag = memalloc_nofs_save();
1416         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1417         memalloc_nofs_restore(nofs_flag);
1418         if (ret)
1419                 goto fail;
1420
1421         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1422                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1423                 btrfs_check_and_init_root_item(&root->root_item);
1424         }
1425
1426         btrfs_init_free_ino_ctl(root);
1427         spin_lock_init(&root->ino_cache_lock);
1428         init_waitqueue_head(&root->ino_cache_wait);
1429
1430         ret = get_anon_bdev(&root->anon_dev);
1431         if (ret)
1432                 goto fail;
1433
1434         mutex_lock(&root->objectid_mutex);
1435         ret = btrfs_find_highest_objectid(root,
1436                                         &root->highest_objectid);
1437         if (ret) {
1438                 mutex_unlock(&root->objectid_mutex);
1439                 goto fail;
1440         }
1441
1442         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1443
1444         mutex_unlock(&root->objectid_mutex);
1445
1446         return 0;
1447 fail:
1448         /* The caller is responsible to call btrfs_free_fs_root */
1449         return ret;
1450 }
1451
1452 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1453                                                u64 root_id)
1454 {
1455         struct btrfs_root *root;
1456
1457         spin_lock(&fs_info->fs_roots_radix_lock);
1458         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1459                                  (unsigned long)root_id);
1460         if (root)
1461                 root = btrfs_grab_root(root);
1462         spin_unlock(&fs_info->fs_roots_radix_lock);
1463         return root;
1464 }
1465
1466 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1467                          struct btrfs_root *root)
1468 {
1469         int ret;
1470
1471         ret = radix_tree_preload(GFP_NOFS);
1472         if (ret)
1473                 return ret;
1474
1475         spin_lock(&fs_info->fs_roots_radix_lock);
1476         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1477                                 (unsigned long)root->root_key.objectid,
1478                                 root);
1479         if (ret == 0) {
1480                 btrfs_grab_root(root);
1481                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1482         }
1483         spin_unlock(&fs_info->fs_roots_radix_lock);
1484         radix_tree_preload_end();
1485
1486         return ret;
1487 }
1488
1489 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1490 {
1491 #ifdef CONFIG_BTRFS_DEBUG
1492         struct btrfs_root *root;
1493
1494         while (!list_empty(&fs_info->allocated_roots)) {
1495                 root = list_first_entry(&fs_info->allocated_roots,
1496                                         struct btrfs_root, leak_list);
1497                 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1498                           root->root_key.objectid, root->root_key.offset,
1499                           refcount_read(&root->refs));
1500                 while (refcount_read(&root->refs) > 1)
1501                         btrfs_put_root(root);
1502                 btrfs_put_root(root);
1503         }
1504 #endif
1505 }
1506
1507 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1508 {
1509         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1510         percpu_counter_destroy(&fs_info->delalloc_bytes);
1511         percpu_counter_destroy(&fs_info->dio_bytes);
1512         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1513         btrfs_free_csum_hash(fs_info);
1514         btrfs_free_stripe_hash_table(fs_info);
1515         btrfs_free_ref_cache(fs_info);
1516         kfree(fs_info->balance_ctl);
1517         kfree(fs_info->delayed_root);
1518         btrfs_put_root(fs_info->extent_root);
1519         btrfs_put_root(fs_info->tree_root);
1520         btrfs_put_root(fs_info->chunk_root);
1521         btrfs_put_root(fs_info->dev_root);
1522         btrfs_put_root(fs_info->csum_root);
1523         btrfs_put_root(fs_info->quota_root);
1524         btrfs_put_root(fs_info->uuid_root);
1525         btrfs_put_root(fs_info->free_space_root);
1526         btrfs_put_root(fs_info->fs_root);
1527         btrfs_check_leaked_roots(fs_info);
1528         btrfs_extent_buffer_leak_debug_check(fs_info);
1529         kfree(fs_info->super_copy);
1530         kfree(fs_info->super_for_commit);
1531         kvfree(fs_info);
1532 }
1533
1534
1535 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1536                                      struct btrfs_key *location,
1537                                      bool check_ref)
1538 {
1539         struct btrfs_root *root;
1540         struct btrfs_path *path;
1541         struct btrfs_key key;
1542         int ret;
1543
1544         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1545                 return btrfs_grab_root(fs_info->tree_root);
1546         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1547                 return btrfs_grab_root(fs_info->extent_root);
1548         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1549                 return btrfs_grab_root(fs_info->chunk_root);
1550         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1551                 return btrfs_grab_root(fs_info->dev_root);
1552         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1553                 return btrfs_grab_root(fs_info->csum_root);
1554         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1555                 return btrfs_grab_root(fs_info->quota_root) ?
1556                         fs_info->quota_root : ERR_PTR(-ENOENT);
1557         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1558                 return btrfs_grab_root(fs_info->uuid_root) ?
1559                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1560         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1561                 return btrfs_grab_root(fs_info->free_space_root) ?
1562                         fs_info->free_space_root : ERR_PTR(-ENOENT);
1563 again:
1564         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1565         if (root) {
1566                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1567                         btrfs_put_root(root);
1568                         return ERR_PTR(-ENOENT);
1569                 }
1570                 return root;
1571         }
1572
1573         root = btrfs_read_tree_root(fs_info->tree_root, location);
1574         if (IS_ERR(root))
1575                 return root;
1576
1577         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1578                 ret = -ENOENT;
1579                 goto fail;
1580         }
1581
1582         ret = btrfs_init_fs_root(root);
1583         if (ret)
1584                 goto fail;
1585
1586         path = btrfs_alloc_path();
1587         if (!path) {
1588                 ret = -ENOMEM;
1589                 goto fail;
1590         }
1591         key.objectid = BTRFS_ORPHAN_OBJECTID;
1592         key.type = BTRFS_ORPHAN_ITEM_KEY;
1593         key.offset = location->objectid;
1594
1595         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1596         btrfs_free_path(path);
1597         if (ret < 0)
1598                 goto fail;
1599         if (ret == 0)
1600                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1601
1602         ret = btrfs_insert_fs_root(fs_info, root);
1603         if (ret) {
1604                 btrfs_put_root(root);
1605                 if (ret == -EEXIST)
1606                         goto again;
1607                 goto fail;
1608         }
1609         return root;
1610 fail:
1611         btrfs_put_root(root);
1612         return ERR_PTR(ret);
1613 }
1614
1615 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1616 {
1617         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1618         int ret = 0;
1619         struct btrfs_device *device;
1620         struct backing_dev_info *bdi;
1621
1622         rcu_read_lock();
1623         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1624                 if (!device->bdev)
1625                         continue;
1626                 bdi = device->bdev->bd_bdi;
1627                 if (bdi_congested(bdi, bdi_bits)) {
1628                         ret = 1;
1629                         break;
1630                 }
1631         }
1632         rcu_read_unlock();
1633         return ret;
1634 }
1635
1636 /*
1637  * called by the kthread helper functions to finally call the bio end_io
1638  * functions.  This is where read checksum verification actually happens
1639  */
1640 static void end_workqueue_fn(struct btrfs_work *work)
1641 {
1642         struct bio *bio;
1643         struct btrfs_end_io_wq *end_io_wq;
1644
1645         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1646         bio = end_io_wq->bio;
1647
1648         bio->bi_status = end_io_wq->status;
1649         bio->bi_private = end_io_wq->private;
1650         bio->bi_end_io = end_io_wq->end_io;
1651         bio_endio(bio);
1652         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1653 }
1654
1655 static int cleaner_kthread(void *arg)
1656 {
1657         struct btrfs_root *root = arg;
1658         struct btrfs_fs_info *fs_info = root->fs_info;
1659         int again;
1660
1661         while (1) {
1662                 again = 0;
1663
1664                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1665
1666                 /* Make the cleaner go to sleep early. */
1667                 if (btrfs_need_cleaner_sleep(fs_info))
1668                         goto sleep;
1669
1670                 /*
1671                  * Do not do anything if we might cause open_ctree() to block
1672                  * before we have finished mounting the filesystem.
1673                  */
1674                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1675                         goto sleep;
1676
1677                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1678                         goto sleep;
1679
1680                 /*
1681                  * Avoid the problem that we change the status of the fs
1682                  * during the above check and trylock.
1683                  */
1684                 if (btrfs_need_cleaner_sleep(fs_info)) {
1685                         mutex_unlock(&fs_info->cleaner_mutex);
1686                         goto sleep;
1687                 }
1688
1689                 btrfs_run_delayed_iputs(fs_info);
1690
1691                 again = btrfs_clean_one_deleted_snapshot(root);
1692                 mutex_unlock(&fs_info->cleaner_mutex);
1693
1694                 /*
1695                  * The defragger has dealt with the R/O remount and umount,
1696                  * needn't do anything special here.
1697                  */
1698                 btrfs_run_defrag_inodes(fs_info);
1699
1700                 /*
1701                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1702                  * with relocation (btrfs_relocate_chunk) and relocation
1703                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1704                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1705                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1706                  * unused block groups.
1707                  */
1708                 btrfs_delete_unused_bgs(fs_info);
1709 sleep:
1710                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1711                 if (kthread_should_park())
1712                         kthread_parkme();
1713                 if (kthread_should_stop())
1714                         return 0;
1715                 if (!again) {
1716                         set_current_state(TASK_INTERRUPTIBLE);
1717                         schedule();
1718                         __set_current_state(TASK_RUNNING);
1719                 }
1720         }
1721 }
1722
1723 static int transaction_kthread(void *arg)
1724 {
1725         struct btrfs_root *root = arg;
1726         struct btrfs_fs_info *fs_info = root->fs_info;
1727         struct btrfs_trans_handle *trans;
1728         struct btrfs_transaction *cur;
1729         u64 transid;
1730         time64_t now;
1731         unsigned long delay;
1732         bool cannot_commit;
1733
1734         do {
1735                 cannot_commit = false;
1736                 delay = HZ * fs_info->commit_interval;
1737                 mutex_lock(&fs_info->transaction_kthread_mutex);
1738
1739                 spin_lock(&fs_info->trans_lock);
1740                 cur = fs_info->running_transaction;
1741                 if (!cur) {
1742                         spin_unlock(&fs_info->trans_lock);
1743                         goto sleep;
1744                 }
1745
1746                 now = ktime_get_seconds();
1747                 if (cur->state < TRANS_STATE_COMMIT_START &&
1748                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1749                     (now < cur->start_time ||
1750                      now - cur->start_time < fs_info->commit_interval)) {
1751                         spin_unlock(&fs_info->trans_lock);
1752                         delay = HZ * 5;
1753                         goto sleep;
1754                 }
1755                 transid = cur->transid;
1756                 spin_unlock(&fs_info->trans_lock);
1757
1758                 /* If the file system is aborted, this will always fail. */
1759                 trans = btrfs_attach_transaction(root);
1760                 if (IS_ERR(trans)) {
1761                         if (PTR_ERR(trans) != -ENOENT)
1762                                 cannot_commit = true;
1763                         goto sleep;
1764                 }
1765                 if (transid == trans->transid) {
1766                         btrfs_commit_transaction(trans);
1767                 } else {
1768                         btrfs_end_transaction(trans);
1769                 }
1770 sleep:
1771                 wake_up_process(fs_info->cleaner_kthread);
1772                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1773
1774                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1775                                       &fs_info->fs_state)))
1776                         btrfs_cleanup_transaction(fs_info);
1777                 if (!kthread_should_stop() &&
1778                                 (!btrfs_transaction_blocked(fs_info) ||
1779                                  cannot_commit))
1780                         schedule_timeout_interruptible(delay);
1781         } while (!kthread_should_stop());
1782         return 0;
1783 }
1784
1785 /*
1786  * This will find the highest generation in the array of root backups.  The
1787  * index of the highest array is returned, or -EINVAL if we can't find
1788  * anything.
1789  *
1790  * We check to make sure the array is valid by comparing the
1791  * generation of the latest  root in the array with the generation
1792  * in the super block.  If they don't match we pitch it.
1793  */
1794 static int find_newest_super_backup(struct btrfs_fs_info *info)
1795 {
1796         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1797         u64 cur;
1798         struct btrfs_root_backup *root_backup;
1799         int i;
1800
1801         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1802                 root_backup = info->super_copy->super_roots + i;
1803                 cur = btrfs_backup_tree_root_gen(root_backup);
1804                 if (cur == newest_gen)
1805                         return i;
1806         }
1807
1808         return -EINVAL;
1809 }
1810
1811 /*
1812  * copy all the root pointers into the super backup array.
1813  * this will bump the backup pointer by one when it is
1814  * done
1815  */
1816 static void backup_super_roots(struct btrfs_fs_info *info)
1817 {
1818         const int next_backup = info->backup_root_index;
1819         struct btrfs_root_backup *root_backup;
1820
1821         root_backup = info->super_for_commit->super_roots + next_backup;
1822
1823         /*
1824          * make sure all of our padding and empty slots get zero filled
1825          * regardless of which ones we use today
1826          */
1827         memset(root_backup, 0, sizeof(*root_backup));
1828
1829         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1830
1831         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1832         btrfs_set_backup_tree_root_gen(root_backup,
1833                                btrfs_header_generation(info->tree_root->node));
1834
1835         btrfs_set_backup_tree_root_level(root_backup,
1836                                btrfs_header_level(info->tree_root->node));
1837
1838         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1839         btrfs_set_backup_chunk_root_gen(root_backup,
1840                                btrfs_header_generation(info->chunk_root->node));
1841         btrfs_set_backup_chunk_root_level(root_backup,
1842                                btrfs_header_level(info->chunk_root->node));
1843
1844         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1845         btrfs_set_backup_extent_root_gen(root_backup,
1846                                btrfs_header_generation(info->extent_root->node));
1847         btrfs_set_backup_extent_root_level(root_backup,
1848                                btrfs_header_level(info->extent_root->node));
1849
1850         /*
1851          * we might commit during log recovery, which happens before we set
1852          * the fs_root.  Make sure it is valid before we fill it in.
1853          */
1854         if (info->fs_root && info->fs_root->node) {
1855                 btrfs_set_backup_fs_root(root_backup,
1856                                          info->fs_root->node->start);
1857                 btrfs_set_backup_fs_root_gen(root_backup,
1858                                btrfs_header_generation(info->fs_root->node));
1859                 btrfs_set_backup_fs_root_level(root_backup,
1860                                btrfs_header_level(info->fs_root->node));
1861         }
1862
1863         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1864         btrfs_set_backup_dev_root_gen(root_backup,
1865                                btrfs_header_generation(info->dev_root->node));
1866         btrfs_set_backup_dev_root_level(root_backup,
1867                                        btrfs_header_level(info->dev_root->node));
1868
1869         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1870         btrfs_set_backup_csum_root_gen(root_backup,
1871                                btrfs_header_generation(info->csum_root->node));
1872         btrfs_set_backup_csum_root_level(root_backup,
1873                                btrfs_header_level(info->csum_root->node));
1874
1875         btrfs_set_backup_total_bytes(root_backup,
1876                              btrfs_super_total_bytes(info->super_copy));
1877         btrfs_set_backup_bytes_used(root_backup,
1878                              btrfs_super_bytes_used(info->super_copy));
1879         btrfs_set_backup_num_devices(root_backup,
1880                              btrfs_super_num_devices(info->super_copy));
1881
1882         /*
1883          * if we don't copy this out to the super_copy, it won't get remembered
1884          * for the next commit
1885          */
1886         memcpy(&info->super_copy->super_roots,
1887                &info->super_for_commit->super_roots,
1888                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1889 }
1890
1891 /*
1892  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1893  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1894  *
1895  * fs_info - filesystem whose backup roots need to be read
1896  * priority - priority of backup root required
1897  *
1898  * Returns backup root index on success and -EINVAL otherwise.
1899  */
1900 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1901 {
1902         int backup_index = find_newest_super_backup(fs_info);
1903         struct btrfs_super_block *super = fs_info->super_copy;
1904         struct btrfs_root_backup *root_backup;
1905
1906         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1907                 if (priority == 0)
1908                         return backup_index;
1909
1910                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1911                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1912         } else {
1913                 return -EINVAL;
1914         }
1915
1916         root_backup = super->super_roots + backup_index;
1917
1918         btrfs_set_super_generation(super,
1919                                    btrfs_backup_tree_root_gen(root_backup));
1920         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1921         btrfs_set_super_root_level(super,
1922                                    btrfs_backup_tree_root_level(root_backup));
1923         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1924
1925         /*
1926          * Fixme: the total bytes and num_devices need to match or we should
1927          * need a fsck
1928          */
1929         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1930         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1931
1932         return backup_index;
1933 }
1934
1935 /* helper to cleanup workers */
1936 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1937 {
1938         btrfs_destroy_workqueue(fs_info->fixup_workers);
1939         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1940         btrfs_destroy_workqueue(fs_info->workers);
1941         btrfs_destroy_workqueue(fs_info->endio_workers);
1942         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1943         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1944         btrfs_destroy_workqueue(fs_info->rmw_workers);
1945         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1946         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1947         btrfs_destroy_workqueue(fs_info->delayed_workers);
1948         btrfs_destroy_workqueue(fs_info->caching_workers);
1949         btrfs_destroy_workqueue(fs_info->readahead_workers);
1950         btrfs_destroy_workqueue(fs_info->flush_workers);
1951         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1952         if (fs_info->discard_ctl.discard_workers)
1953                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1954         /*
1955          * Now that all other work queues are destroyed, we can safely destroy
1956          * the queues used for metadata I/O, since tasks from those other work
1957          * queues can do metadata I/O operations.
1958          */
1959         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1960         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1961 }
1962
1963 static void free_root_extent_buffers(struct btrfs_root *root)
1964 {
1965         if (root) {
1966                 free_extent_buffer(root->node);
1967                 free_extent_buffer(root->commit_root);
1968                 root->node = NULL;
1969                 root->commit_root = NULL;
1970         }
1971 }
1972
1973 /* helper to cleanup tree roots */
1974 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1975 {
1976         free_root_extent_buffers(info->tree_root);
1977
1978         free_root_extent_buffers(info->dev_root);
1979         free_root_extent_buffers(info->extent_root);
1980         free_root_extent_buffers(info->csum_root);
1981         free_root_extent_buffers(info->quota_root);
1982         free_root_extent_buffers(info->uuid_root);
1983         free_root_extent_buffers(info->fs_root);
1984         if (free_chunk_root)
1985                 free_root_extent_buffers(info->chunk_root);
1986         free_root_extent_buffers(info->free_space_root);
1987 }
1988
1989 void btrfs_put_root(struct btrfs_root *root)
1990 {
1991         if (!root)
1992                 return;
1993
1994         if (refcount_dec_and_test(&root->refs)) {
1995                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1996                 if (root->anon_dev)
1997                         free_anon_bdev(root->anon_dev);
1998                 btrfs_drew_lock_destroy(&root->snapshot_lock);
1999                 free_extent_buffer(root->node);
2000                 free_extent_buffer(root->commit_root);
2001                 kfree(root->free_ino_ctl);
2002                 kfree(root->free_ino_pinned);
2003 #ifdef CONFIG_BTRFS_DEBUG
2004                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2005                 list_del_init(&root->leak_list);
2006                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2007 #endif
2008                 kfree(root);
2009         }
2010 }
2011
2012 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2013 {
2014         int ret;
2015         struct btrfs_root *gang[8];
2016         int i;
2017
2018         while (!list_empty(&fs_info->dead_roots)) {
2019                 gang[0] = list_entry(fs_info->dead_roots.next,
2020                                      struct btrfs_root, root_list);
2021                 list_del(&gang[0]->root_list);
2022
2023                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2024                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2025                 btrfs_put_root(gang[0]);
2026         }
2027
2028         while (1) {
2029                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2030                                              (void **)gang, 0,
2031                                              ARRAY_SIZE(gang));
2032                 if (!ret)
2033                         break;
2034                 for (i = 0; i < ret; i++)
2035                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2036         }
2037 }
2038
2039 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2040 {
2041         mutex_init(&fs_info->scrub_lock);
2042         atomic_set(&fs_info->scrubs_running, 0);
2043         atomic_set(&fs_info->scrub_pause_req, 0);
2044         atomic_set(&fs_info->scrubs_paused, 0);
2045         atomic_set(&fs_info->scrub_cancel_req, 0);
2046         init_waitqueue_head(&fs_info->scrub_pause_wait);
2047         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2048 }
2049
2050 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2051 {
2052         spin_lock_init(&fs_info->balance_lock);
2053         mutex_init(&fs_info->balance_mutex);
2054         atomic_set(&fs_info->balance_pause_req, 0);
2055         atomic_set(&fs_info->balance_cancel_req, 0);
2056         fs_info->balance_ctl = NULL;
2057         init_waitqueue_head(&fs_info->balance_wait_q);
2058 }
2059
2060 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2061 {
2062         struct inode *inode = fs_info->btree_inode;
2063
2064         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2065         set_nlink(inode, 1);
2066         /*
2067          * we set the i_size on the btree inode to the max possible int.
2068          * the real end of the address space is determined by all of
2069          * the devices in the system
2070          */
2071         inode->i_size = OFFSET_MAX;
2072         inode->i_mapping->a_ops = &btree_aops;
2073
2074         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2075         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2076                             IO_TREE_INODE_IO, inode);
2077         BTRFS_I(inode)->io_tree.track_uptodate = false;
2078         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2079
2080         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2081
2082         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2083         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2084         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2085         btrfs_insert_inode_hash(inode);
2086 }
2087
2088 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2089 {
2090         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2091         init_rwsem(&fs_info->dev_replace.rwsem);
2092         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2093 }
2094
2095 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2096 {
2097         spin_lock_init(&fs_info->qgroup_lock);
2098         mutex_init(&fs_info->qgroup_ioctl_lock);
2099         fs_info->qgroup_tree = RB_ROOT;
2100         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2101         fs_info->qgroup_seq = 1;
2102         fs_info->qgroup_ulist = NULL;
2103         fs_info->qgroup_rescan_running = false;
2104         mutex_init(&fs_info->qgroup_rescan_lock);
2105 }
2106
2107 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2108                 struct btrfs_fs_devices *fs_devices)
2109 {
2110         u32 max_active = fs_info->thread_pool_size;
2111         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2112
2113         fs_info->workers =
2114                 btrfs_alloc_workqueue(fs_info, "worker",
2115                                       flags | WQ_HIGHPRI, max_active, 16);
2116
2117         fs_info->delalloc_workers =
2118                 btrfs_alloc_workqueue(fs_info, "delalloc",
2119                                       flags, max_active, 2);
2120
2121         fs_info->flush_workers =
2122                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2123                                       flags, max_active, 0);
2124
2125         fs_info->caching_workers =
2126                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2127
2128         fs_info->fixup_workers =
2129                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2130
2131         /*
2132          * endios are largely parallel and should have a very
2133          * low idle thresh
2134          */
2135         fs_info->endio_workers =
2136                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2137         fs_info->endio_meta_workers =
2138                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2139                                       max_active, 4);
2140         fs_info->endio_meta_write_workers =
2141                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2142                                       max_active, 2);
2143         fs_info->endio_raid56_workers =
2144                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2145                                       max_active, 4);
2146         fs_info->endio_repair_workers =
2147                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2148         fs_info->rmw_workers =
2149                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2150         fs_info->endio_write_workers =
2151                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2152                                       max_active, 2);
2153         fs_info->endio_freespace_worker =
2154                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2155                                       max_active, 0);
2156         fs_info->delayed_workers =
2157                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2158                                       max_active, 0);
2159         fs_info->readahead_workers =
2160                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2161                                       max_active, 2);
2162         fs_info->qgroup_rescan_workers =
2163                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2164         fs_info->discard_ctl.discard_workers =
2165                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2166
2167         if (!(fs_info->workers && fs_info->delalloc_workers &&
2168               fs_info->flush_workers &&
2169               fs_info->endio_workers && fs_info->endio_meta_workers &&
2170               fs_info->endio_meta_write_workers &&
2171               fs_info->endio_repair_workers &&
2172               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2173               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2174               fs_info->caching_workers && fs_info->readahead_workers &&
2175               fs_info->fixup_workers && fs_info->delayed_workers &&
2176               fs_info->qgroup_rescan_workers &&
2177               fs_info->discard_ctl.discard_workers)) {
2178                 return -ENOMEM;
2179         }
2180
2181         return 0;
2182 }
2183
2184 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2185 {
2186         struct crypto_shash *csum_shash;
2187         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2188
2189         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2190
2191         if (IS_ERR(csum_shash)) {
2192                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2193                           csum_driver);
2194                 return PTR_ERR(csum_shash);
2195         }
2196
2197         fs_info->csum_shash = csum_shash;
2198
2199         return 0;
2200 }
2201
2202 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2203                             struct btrfs_fs_devices *fs_devices)
2204 {
2205         int ret;
2206         struct btrfs_root *log_tree_root;
2207         struct btrfs_super_block *disk_super = fs_info->super_copy;
2208         u64 bytenr = btrfs_super_log_root(disk_super);
2209         int level = btrfs_super_log_root_level(disk_super);
2210
2211         if (fs_devices->rw_devices == 0) {
2212                 btrfs_warn(fs_info, "log replay required on RO media");
2213                 return -EIO;
2214         }
2215
2216         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2217                                          GFP_KERNEL);
2218         if (!log_tree_root)
2219                 return -ENOMEM;
2220
2221         log_tree_root->node = read_tree_block(fs_info, bytenr,
2222                                               fs_info->generation + 1,
2223                                               level, NULL);
2224         if (IS_ERR(log_tree_root->node)) {
2225                 btrfs_warn(fs_info, "failed to read log tree");
2226                 ret = PTR_ERR(log_tree_root->node);
2227                 log_tree_root->node = NULL;
2228                 btrfs_put_root(log_tree_root);
2229                 return ret;
2230         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2231                 btrfs_err(fs_info, "failed to read log tree");
2232                 btrfs_put_root(log_tree_root);
2233                 return -EIO;
2234         }
2235         /* returns with log_tree_root freed on success */
2236         ret = btrfs_recover_log_trees(log_tree_root);
2237         if (ret) {
2238                 btrfs_handle_fs_error(fs_info, ret,
2239                                       "Failed to recover log tree");
2240                 btrfs_put_root(log_tree_root);
2241                 return ret;
2242         }
2243
2244         if (sb_rdonly(fs_info->sb)) {
2245                 ret = btrfs_commit_super(fs_info);
2246                 if (ret)
2247                         return ret;
2248         }
2249
2250         return 0;
2251 }
2252
2253 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2254 {
2255         struct btrfs_root *tree_root = fs_info->tree_root;
2256         struct btrfs_root *root;
2257         struct btrfs_key location;
2258         int ret;
2259
2260         BUG_ON(!fs_info->tree_root);
2261
2262         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2263         location.type = BTRFS_ROOT_ITEM_KEY;
2264         location.offset = 0;
2265
2266         root = btrfs_read_tree_root(tree_root, &location);
2267         if (IS_ERR(root)) {
2268                 ret = PTR_ERR(root);
2269                 goto out;
2270         }
2271         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2272         fs_info->extent_root = root;
2273
2274         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2275         root = btrfs_read_tree_root(tree_root, &location);
2276         if (IS_ERR(root)) {
2277                 ret = PTR_ERR(root);
2278                 goto out;
2279         }
2280         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2281         fs_info->dev_root = root;
2282         btrfs_init_devices_late(fs_info);
2283
2284         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2285         root = btrfs_read_tree_root(tree_root, &location);
2286         if (IS_ERR(root)) {
2287                 ret = PTR_ERR(root);
2288                 goto out;
2289         }
2290         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2291         fs_info->csum_root = root;
2292
2293         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2294         root = btrfs_read_tree_root(tree_root, &location);
2295         if (!IS_ERR(root)) {
2296                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2297                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2298                 fs_info->quota_root = root;
2299         }
2300
2301         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2302         root = btrfs_read_tree_root(tree_root, &location);
2303         if (IS_ERR(root)) {
2304                 ret = PTR_ERR(root);
2305                 if (ret != -ENOENT)
2306                         goto out;
2307         } else {
2308                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2309                 fs_info->uuid_root = root;
2310         }
2311
2312         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2313                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2314                 root = btrfs_read_tree_root(tree_root, &location);
2315                 if (IS_ERR(root)) {
2316                         ret = PTR_ERR(root);
2317                         goto out;
2318                 }
2319                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320                 fs_info->free_space_root = root;
2321         }
2322
2323         return 0;
2324 out:
2325         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2326                    location.objectid, ret);
2327         return ret;
2328 }
2329
2330 /*
2331  * Real super block validation
2332  * NOTE: super csum type and incompat features will not be checked here.
2333  *
2334  * @sb:         super block to check
2335  * @mirror_num: the super block number to check its bytenr:
2336  *              0       the primary (1st) sb
2337  *              1, 2    2nd and 3rd backup copy
2338  *             -1       skip bytenr check
2339  */
2340 static int validate_super(struct btrfs_fs_info *fs_info,
2341                             struct btrfs_super_block *sb, int mirror_num)
2342 {
2343         u64 nodesize = btrfs_super_nodesize(sb);
2344         u64 sectorsize = btrfs_super_sectorsize(sb);
2345         int ret = 0;
2346
2347         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2348                 btrfs_err(fs_info, "no valid FS found");
2349                 ret = -EINVAL;
2350         }
2351         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2352                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2353                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2354                 ret = -EINVAL;
2355         }
2356         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2357                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2358                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2359                 ret = -EINVAL;
2360         }
2361         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2362                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2363                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2364                 ret = -EINVAL;
2365         }
2366         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2367                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2368                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2369                 ret = -EINVAL;
2370         }
2371
2372         /*
2373          * Check sectorsize and nodesize first, other check will need it.
2374          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2375          */
2376         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2377             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2378                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2379                 ret = -EINVAL;
2380         }
2381         /* Only PAGE SIZE is supported yet */
2382         if (sectorsize != PAGE_SIZE) {
2383                 btrfs_err(fs_info,
2384                         "sectorsize %llu not supported yet, only support %lu",
2385                         sectorsize, PAGE_SIZE);
2386                 ret = -EINVAL;
2387         }
2388         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2389             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2390                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2391                 ret = -EINVAL;
2392         }
2393         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2394                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2395                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2396                 ret = -EINVAL;
2397         }
2398
2399         /* Root alignment check */
2400         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2401                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2402                            btrfs_super_root(sb));
2403                 ret = -EINVAL;
2404         }
2405         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2406                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2407                            btrfs_super_chunk_root(sb));
2408                 ret = -EINVAL;
2409         }
2410         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2411                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2412                            btrfs_super_log_root(sb));
2413                 ret = -EINVAL;
2414         }
2415
2416         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2417                    BTRFS_FSID_SIZE) != 0) {
2418                 btrfs_err(fs_info,
2419                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2420                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2421                 ret = -EINVAL;
2422         }
2423
2424         /*
2425          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2426          * done later
2427          */
2428         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2429                 btrfs_err(fs_info, "bytes_used is too small %llu",
2430                           btrfs_super_bytes_used(sb));
2431                 ret = -EINVAL;
2432         }
2433         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2434                 btrfs_err(fs_info, "invalid stripesize %u",
2435                           btrfs_super_stripesize(sb));
2436                 ret = -EINVAL;
2437         }
2438         if (btrfs_super_num_devices(sb) > (1UL << 31))
2439                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2440                            btrfs_super_num_devices(sb));
2441         if (btrfs_super_num_devices(sb) == 0) {
2442                 btrfs_err(fs_info, "number of devices is 0");
2443                 ret = -EINVAL;
2444         }
2445
2446         if (mirror_num >= 0 &&
2447             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2448                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2449                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2450                 ret = -EINVAL;
2451         }
2452
2453         /*
2454          * Obvious sys_chunk_array corruptions, it must hold at least one key
2455          * and one chunk
2456          */
2457         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2458                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2459                           btrfs_super_sys_array_size(sb),
2460                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2461                 ret = -EINVAL;
2462         }
2463         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2464                         + sizeof(struct btrfs_chunk)) {
2465                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2466                           btrfs_super_sys_array_size(sb),
2467                           sizeof(struct btrfs_disk_key)
2468                           + sizeof(struct btrfs_chunk));
2469                 ret = -EINVAL;
2470         }
2471
2472         /*
2473          * The generation is a global counter, we'll trust it more than the others
2474          * but it's still possible that it's the one that's wrong.
2475          */
2476         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2477                 btrfs_warn(fs_info,
2478                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2479                         btrfs_super_generation(sb),
2480                         btrfs_super_chunk_root_generation(sb));
2481         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2482             && btrfs_super_cache_generation(sb) != (u64)-1)
2483                 btrfs_warn(fs_info,
2484                         "suspicious: generation < cache_generation: %llu < %llu",
2485                         btrfs_super_generation(sb),
2486                         btrfs_super_cache_generation(sb));
2487
2488         return ret;
2489 }
2490
2491 /*
2492  * Validation of super block at mount time.
2493  * Some checks already done early at mount time, like csum type and incompat
2494  * flags will be skipped.
2495  */
2496 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2497 {
2498         return validate_super(fs_info, fs_info->super_copy, 0);
2499 }
2500
2501 /*
2502  * Validation of super block at write time.
2503  * Some checks like bytenr check will be skipped as their values will be
2504  * overwritten soon.
2505  * Extra checks like csum type and incompat flags will be done here.
2506  */
2507 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2508                                       struct btrfs_super_block *sb)
2509 {
2510         int ret;
2511
2512         ret = validate_super(fs_info, sb, -1);
2513         if (ret < 0)
2514                 goto out;
2515         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2516                 ret = -EUCLEAN;
2517                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2518                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2519                 goto out;
2520         }
2521         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2522                 ret = -EUCLEAN;
2523                 btrfs_err(fs_info,
2524                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2525                           btrfs_super_incompat_flags(sb),
2526                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2527                 goto out;
2528         }
2529 out:
2530         if (ret < 0)
2531                 btrfs_err(fs_info,
2532                 "super block corruption detected before writing it to disk");
2533         return ret;
2534 }
2535
2536 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2537 {
2538         int backup_index = find_newest_super_backup(fs_info);
2539         struct btrfs_super_block *sb = fs_info->super_copy;
2540         struct btrfs_root *tree_root = fs_info->tree_root;
2541         bool handle_error = false;
2542         int ret = 0;
2543         int i;
2544
2545         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2546                 u64 generation;
2547                 int level;
2548
2549                 if (handle_error) {
2550                         if (!IS_ERR(tree_root->node))
2551                                 free_extent_buffer(tree_root->node);
2552                         tree_root->node = NULL;
2553
2554                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2555                                 break;
2556
2557                         free_root_pointers(fs_info, 0);
2558
2559                         /*
2560                          * Don't use the log in recovery mode, it won't be
2561                          * valid
2562                          */
2563                         btrfs_set_super_log_root(sb, 0);
2564
2565                         /* We can't trust the free space cache either */
2566                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2567
2568                         ret = read_backup_root(fs_info, i);
2569                         backup_index = ret;
2570                         if (ret < 0)
2571                                 return ret;
2572                 }
2573                 generation = btrfs_super_generation(sb);
2574                 level = btrfs_super_root_level(sb);
2575                 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2576                                                   generation, level, NULL);
2577                 if (IS_ERR(tree_root->node) ||
2578                     !extent_buffer_uptodate(tree_root->node)) {
2579                         handle_error = true;
2580
2581                         if (IS_ERR(tree_root->node))
2582                                 ret = PTR_ERR(tree_root->node);
2583                         else if (!extent_buffer_uptodate(tree_root->node))
2584                                 ret = -EUCLEAN;
2585
2586                         btrfs_warn(fs_info, "failed to read tree root");
2587                         continue;
2588                 }
2589
2590                 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2591                 tree_root->commit_root = btrfs_root_node(tree_root);
2592                 btrfs_set_root_refs(&tree_root->root_item, 1);
2593
2594                 /*
2595                  * No need to hold btrfs_root::objectid_mutex since the fs
2596                  * hasn't been fully initialised and we are the only user
2597                  */
2598                 ret = btrfs_find_highest_objectid(tree_root,
2599                                                 &tree_root->highest_objectid);
2600                 if (ret < 0) {
2601                         handle_error = true;
2602                         continue;
2603                 }
2604
2605                 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2606
2607                 ret = btrfs_read_roots(fs_info);
2608                 if (ret < 0) {
2609                         handle_error = true;
2610                         continue;
2611                 }
2612
2613                 /* All successful */
2614                 fs_info->generation = generation;
2615                 fs_info->last_trans_committed = generation;
2616
2617                 /* Always begin writing backup roots after the one being used */
2618                 if (backup_index < 0) {
2619                         fs_info->backup_root_index = 0;
2620                 } else {
2621                         fs_info->backup_root_index = backup_index + 1;
2622                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2623                 }
2624                 break;
2625         }
2626
2627         return ret;
2628 }
2629
2630 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2631 {
2632         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2633         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2634         INIT_LIST_HEAD(&fs_info->trans_list);
2635         INIT_LIST_HEAD(&fs_info->dead_roots);
2636         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2637         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2638         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2639         spin_lock_init(&fs_info->delalloc_root_lock);
2640         spin_lock_init(&fs_info->trans_lock);
2641         spin_lock_init(&fs_info->fs_roots_radix_lock);
2642         spin_lock_init(&fs_info->delayed_iput_lock);
2643         spin_lock_init(&fs_info->defrag_inodes_lock);
2644         spin_lock_init(&fs_info->super_lock);
2645         spin_lock_init(&fs_info->buffer_lock);
2646         spin_lock_init(&fs_info->unused_bgs_lock);
2647         rwlock_init(&fs_info->tree_mod_log_lock);
2648         mutex_init(&fs_info->unused_bg_unpin_mutex);
2649         mutex_init(&fs_info->delete_unused_bgs_mutex);
2650         mutex_init(&fs_info->reloc_mutex);
2651         mutex_init(&fs_info->delalloc_root_mutex);
2652         seqlock_init(&fs_info->profiles_lock);
2653
2654         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2655         INIT_LIST_HEAD(&fs_info->space_info);
2656         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2657         INIT_LIST_HEAD(&fs_info->unused_bgs);
2658 #ifdef CONFIG_BTRFS_DEBUG
2659         INIT_LIST_HEAD(&fs_info->allocated_roots);
2660         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2661         spin_lock_init(&fs_info->eb_leak_lock);
2662 #endif
2663         extent_map_tree_init(&fs_info->mapping_tree);
2664         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2665                              BTRFS_BLOCK_RSV_GLOBAL);
2666         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2667         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2668         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2669         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2670                              BTRFS_BLOCK_RSV_DELOPS);
2671         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2672                              BTRFS_BLOCK_RSV_DELREFS);
2673
2674         atomic_set(&fs_info->async_delalloc_pages, 0);
2675         atomic_set(&fs_info->defrag_running, 0);
2676         atomic_set(&fs_info->reada_works_cnt, 0);
2677         atomic_set(&fs_info->nr_delayed_iputs, 0);
2678         atomic64_set(&fs_info->tree_mod_seq, 0);
2679         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2680         fs_info->metadata_ratio = 0;
2681         fs_info->defrag_inodes = RB_ROOT;
2682         atomic64_set(&fs_info->free_chunk_space, 0);
2683         fs_info->tree_mod_log = RB_ROOT;
2684         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2685         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2686         /* readahead state */
2687         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2688         spin_lock_init(&fs_info->reada_lock);
2689         btrfs_init_ref_verify(fs_info);
2690
2691         fs_info->thread_pool_size = min_t(unsigned long,
2692                                           num_online_cpus() + 2, 8);
2693
2694         INIT_LIST_HEAD(&fs_info->ordered_roots);
2695         spin_lock_init(&fs_info->ordered_root_lock);
2696
2697         btrfs_init_scrub(fs_info);
2698 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2699         fs_info->check_integrity_print_mask = 0;
2700 #endif
2701         btrfs_init_balance(fs_info);
2702         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2703
2704         spin_lock_init(&fs_info->block_group_cache_lock);
2705         fs_info->block_group_cache_tree = RB_ROOT;
2706         fs_info->first_logical_byte = (u64)-1;
2707
2708         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2709                             IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2710         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2711
2712         mutex_init(&fs_info->ordered_operations_mutex);
2713         mutex_init(&fs_info->tree_log_mutex);
2714         mutex_init(&fs_info->chunk_mutex);
2715         mutex_init(&fs_info->transaction_kthread_mutex);
2716         mutex_init(&fs_info->cleaner_mutex);
2717         mutex_init(&fs_info->ro_block_group_mutex);
2718         init_rwsem(&fs_info->commit_root_sem);
2719         init_rwsem(&fs_info->cleanup_work_sem);
2720         init_rwsem(&fs_info->subvol_sem);
2721         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2722
2723         btrfs_init_dev_replace_locks(fs_info);
2724         btrfs_init_qgroup(fs_info);
2725         btrfs_discard_init(fs_info);
2726
2727         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2728         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2729
2730         init_waitqueue_head(&fs_info->transaction_throttle);
2731         init_waitqueue_head(&fs_info->transaction_wait);
2732         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2733         init_waitqueue_head(&fs_info->async_submit_wait);
2734         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2735
2736         /* Usable values until the real ones are cached from the superblock */
2737         fs_info->nodesize = 4096;
2738         fs_info->sectorsize = 4096;
2739         fs_info->stripesize = 4096;
2740
2741         spin_lock_init(&fs_info->swapfile_pins_lock);
2742         fs_info->swapfile_pins = RB_ROOT;
2743
2744         fs_info->send_in_progress = 0;
2745 }
2746
2747 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2748 {
2749         int ret;
2750
2751         fs_info->sb = sb;
2752         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2753         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2754
2755         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2756         if (ret)
2757                 return ret;
2758
2759         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2760         if (ret)
2761                 return ret;
2762
2763         fs_info->dirty_metadata_batch = PAGE_SIZE *
2764                                         (1 + ilog2(nr_cpu_ids));
2765
2766         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2767         if (ret)
2768                 return ret;
2769
2770         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2771                         GFP_KERNEL);
2772         if (ret)
2773                 return ret;
2774
2775         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2776                                         GFP_KERNEL);
2777         if (!fs_info->delayed_root)
2778                 return -ENOMEM;
2779         btrfs_init_delayed_root(fs_info->delayed_root);
2780
2781         return btrfs_alloc_stripe_hash_table(fs_info);
2782 }
2783
2784 static int btrfs_uuid_rescan_kthread(void *data)
2785 {
2786         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2787         int ret;
2788
2789         /*
2790          * 1st step is to iterate through the existing UUID tree and
2791          * to delete all entries that contain outdated data.
2792          * 2nd step is to add all missing entries to the UUID tree.
2793          */
2794         ret = btrfs_uuid_tree_iterate(fs_info);
2795         if (ret < 0) {
2796                 if (ret != -EINTR)
2797                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2798                                    ret);
2799                 up(&fs_info->uuid_tree_rescan_sem);
2800                 return ret;
2801         }
2802         return btrfs_uuid_scan_kthread(data);
2803 }
2804
2805 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2806 {
2807         struct task_struct *task;
2808
2809         down(&fs_info->uuid_tree_rescan_sem);
2810         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2811         if (IS_ERR(task)) {
2812                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2813                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2814                 up(&fs_info->uuid_tree_rescan_sem);
2815                 return PTR_ERR(task);
2816         }
2817
2818         return 0;
2819 }
2820
2821 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2822                       char *options)
2823 {
2824         u32 sectorsize;
2825         u32 nodesize;
2826         u32 stripesize;
2827         u64 generation;
2828         u64 features;
2829         u16 csum_type;
2830         struct btrfs_key location;
2831         struct btrfs_super_block *disk_super;
2832         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2833         struct btrfs_root *tree_root;
2834         struct btrfs_root *chunk_root;
2835         int ret;
2836         int err = -EINVAL;
2837         int clear_free_space_tree = 0;
2838         int level;
2839
2840         ret = init_mount_fs_info(fs_info, sb);
2841         if (ret) {
2842                 err = ret;
2843                 goto fail;
2844         }
2845
2846         /* These need to be init'ed before we start creating inodes and such. */
2847         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2848                                      GFP_KERNEL);
2849         fs_info->tree_root = tree_root;
2850         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2851                                       GFP_KERNEL);
2852         fs_info->chunk_root = chunk_root;
2853         if (!tree_root || !chunk_root) {
2854                 err = -ENOMEM;
2855                 goto fail;
2856         }
2857
2858         fs_info->btree_inode = new_inode(sb);
2859         if (!fs_info->btree_inode) {
2860                 err = -ENOMEM;
2861                 goto fail;
2862         }
2863         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2864         btrfs_init_btree_inode(fs_info);
2865
2866         invalidate_bdev(fs_devices->latest_bdev);
2867
2868         /*
2869          * Read super block and check the signature bytes only
2870          */
2871         disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2872         if (IS_ERR(disk_super)) {
2873                 err = PTR_ERR(disk_super);
2874                 goto fail_alloc;
2875         }
2876
2877         /*
2878          * Verify the type first, if that or the the checksum value are
2879          * corrupted, we'll find out
2880          */
2881         csum_type = btrfs_super_csum_type(disk_super);
2882         if (!btrfs_supported_super_csum(csum_type)) {
2883                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2884                           csum_type);
2885                 err = -EINVAL;
2886                 btrfs_release_disk_super(disk_super);
2887                 goto fail_alloc;
2888         }
2889
2890         ret = btrfs_init_csum_hash(fs_info, csum_type);
2891         if (ret) {
2892                 err = ret;
2893                 btrfs_release_disk_super(disk_super);
2894                 goto fail_alloc;
2895         }
2896
2897         /*
2898          * We want to check superblock checksum, the type is stored inside.
2899          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2900          */
2901         if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2902                 btrfs_err(fs_info, "superblock checksum mismatch");
2903                 err = -EINVAL;
2904                 btrfs_release_disk_super(disk_super);
2905                 goto fail_alloc;
2906         }
2907
2908         /*
2909          * super_copy is zeroed at allocation time and we never touch the
2910          * following bytes up to INFO_SIZE, the checksum is calculated from
2911          * the whole block of INFO_SIZE
2912          */
2913         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2914         btrfs_release_disk_super(disk_super);
2915
2916         disk_super = fs_info->super_copy;
2917
2918         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2919                        BTRFS_FSID_SIZE));
2920
2921         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2922                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2923                                 fs_info->super_copy->metadata_uuid,
2924                                 BTRFS_FSID_SIZE));
2925         }
2926
2927         features = btrfs_super_flags(disk_super);
2928         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2929                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2930                 btrfs_set_super_flags(disk_super, features);
2931                 btrfs_info(fs_info,
2932                         "found metadata UUID change in progress flag, clearing");
2933         }
2934
2935         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2936                sizeof(*fs_info->super_for_commit));
2937
2938         ret = btrfs_validate_mount_super(fs_info);
2939         if (ret) {
2940                 btrfs_err(fs_info, "superblock contains fatal errors");
2941                 err = -EINVAL;
2942                 goto fail_alloc;
2943         }
2944
2945         if (!btrfs_super_root(disk_super))
2946                 goto fail_alloc;
2947
2948         /* check FS state, whether FS is broken. */
2949         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2950                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2951
2952         /*
2953          * In the long term, we'll store the compression type in the super
2954          * block, and it'll be used for per file compression control.
2955          */
2956         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2957
2958         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2959         if (ret) {
2960                 err = ret;
2961                 goto fail_alloc;
2962         }
2963
2964         features = btrfs_super_incompat_flags(disk_super) &
2965                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2966         if (features) {
2967                 btrfs_err(fs_info,
2968                     "cannot mount because of unsupported optional features (%llx)",
2969                     features);
2970                 err = -EINVAL;
2971                 goto fail_alloc;
2972         }
2973
2974         features = btrfs_super_incompat_flags(disk_super);
2975         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2976         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2977                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2978         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2979                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2980
2981         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2982                 btrfs_info(fs_info, "has skinny extents");
2983
2984         /*
2985          * flag our filesystem as having big metadata blocks if
2986          * they are bigger than the page size
2987          */
2988         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2989                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2990                         btrfs_info(fs_info,
2991                                 "flagging fs with big metadata feature");
2992                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2993         }
2994
2995         nodesize = btrfs_super_nodesize(disk_super);
2996         sectorsize = btrfs_super_sectorsize(disk_super);
2997         stripesize = sectorsize;
2998         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2999         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3000
3001         /* Cache block sizes */
3002         fs_info->nodesize = nodesize;
3003         fs_info->sectorsize = sectorsize;
3004         fs_info->stripesize = stripesize;
3005
3006         /*
3007          * mixed block groups end up with duplicate but slightly offset
3008          * extent buffers for the same range.  It leads to corruptions
3009          */
3010         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3011             (sectorsize != nodesize)) {
3012                 btrfs_err(fs_info,
3013 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3014                         nodesize, sectorsize);
3015                 goto fail_alloc;
3016         }
3017
3018         /*
3019          * Needn't use the lock because there is no other task which will
3020          * update the flag.
3021          */
3022         btrfs_set_super_incompat_flags(disk_super, features);
3023
3024         features = btrfs_super_compat_ro_flags(disk_super) &
3025                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3026         if (!sb_rdonly(sb) && features) {
3027                 btrfs_err(fs_info,
3028         "cannot mount read-write because of unsupported optional features (%llx)",
3029                        features);
3030                 err = -EINVAL;
3031                 goto fail_alloc;
3032         }
3033
3034         ret = btrfs_init_workqueues(fs_info, fs_devices);
3035         if (ret) {
3036                 err = ret;
3037                 goto fail_sb_buffer;
3038         }
3039
3040         sb->s_bdi->congested_fn = btrfs_congested_fn;
3041         sb->s_bdi->congested_data = fs_info;
3042         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3043         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3044         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3045         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3046
3047         sb->s_blocksize = sectorsize;
3048         sb->s_blocksize_bits = blksize_bits(sectorsize);
3049         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3050
3051         mutex_lock(&fs_info->chunk_mutex);
3052         ret = btrfs_read_sys_array(fs_info);
3053         mutex_unlock(&fs_info->chunk_mutex);
3054         if (ret) {
3055                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3056                 goto fail_sb_buffer;
3057         }
3058
3059         generation = btrfs_super_chunk_root_generation(disk_super);
3060         level = btrfs_super_chunk_root_level(disk_super);
3061
3062         chunk_root->node = read_tree_block(fs_info,
3063                                            btrfs_super_chunk_root(disk_super),
3064                                            generation, level, NULL);
3065         if (IS_ERR(chunk_root->node) ||
3066             !extent_buffer_uptodate(chunk_root->node)) {
3067                 btrfs_err(fs_info, "failed to read chunk root");
3068                 if (!IS_ERR(chunk_root->node))
3069                         free_extent_buffer(chunk_root->node);
3070                 chunk_root->node = NULL;
3071                 goto fail_tree_roots;
3072         }
3073         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3074         chunk_root->commit_root = btrfs_root_node(chunk_root);
3075
3076         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3077                            offsetof(struct btrfs_header, chunk_tree_uuid),
3078                            BTRFS_UUID_SIZE);
3079
3080         ret = btrfs_read_chunk_tree(fs_info);
3081         if (ret) {
3082                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3083                 goto fail_tree_roots;
3084         }
3085
3086         /*
3087          * Keep the devid that is marked to be the target device for the
3088          * device replace procedure
3089          */
3090         btrfs_free_extra_devids(fs_devices, 0);
3091
3092         if (!fs_devices->latest_bdev) {
3093                 btrfs_err(fs_info, "failed to read devices");
3094                 goto fail_tree_roots;
3095         }
3096
3097         ret = init_tree_roots(fs_info);
3098         if (ret)
3099                 goto fail_tree_roots;
3100
3101         /*
3102          * If we have a uuid root and we're not being told to rescan we need to
3103          * check the generation here so we can set the
3104          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3105          * transaction during a balance or the log replay without updating the
3106          * uuid generation, and then if we crash we would rescan the uuid tree,
3107          * even though it was perfectly fine.
3108          */
3109         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3110             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3111                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3112
3113         ret = btrfs_verify_dev_extents(fs_info);
3114         if (ret) {
3115                 btrfs_err(fs_info,
3116                           "failed to verify dev extents against chunks: %d",
3117                           ret);
3118                 goto fail_block_groups;
3119         }
3120         ret = btrfs_recover_balance(fs_info);
3121         if (ret) {
3122                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3123                 goto fail_block_groups;
3124         }
3125
3126         ret = btrfs_init_dev_stats(fs_info);
3127         if (ret) {
3128                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3129                 goto fail_block_groups;
3130         }
3131
3132         ret = btrfs_init_dev_replace(fs_info);
3133         if (ret) {
3134                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3135                 goto fail_block_groups;
3136         }
3137
3138         btrfs_free_extra_devids(fs_devices, 1);
3139
3140         ret = btrfs_sysfs_add_fsid(fs_devices);
3141         if (ret) {
3142                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3143                                 ret);
3144                 goto fail_block_groups;
3145         }
3146
3147         ret = btrfs_sysfs_add_mounted(fs_info);
3148         if (ret) {
3149                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3150                 goto fail_fsdev_sysfs;
3151         }
3152
3153         ret = btrfs_init_space_info(fs_info);
3154         if (ret) {
3155                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3156                 goto fail_sysfs;
3157         }
3158
3159         ret = btrfs_read_block_groups(fs_info);
3160         if (ret) {
3161                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3162                 goto fail_sysfs;
3163         }
3164
3165         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3166                 btrfs_warn(fs_info,
3167                 "writable mount is not allowed due to too many missing devices");
3168                 goto fail_sysfs;
3169         }
3170
3171         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3172                                                "btrfs-cleaner");
3173         if (IS_ERR(fs_info->cleaner_kthread))
3174                 goto fail_sysfs;
3175
3176         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3177                                                    tree_root,
3178                                                    "btrfs-transaction");
3179         if (IS_ERR(fs_info->transaction_kthread))
3180                 goto fail_cleaner;
3181
3182         if (!btrfs_test_opt(fs_info, NOSSD) &&
3183             !fs_info->fs_devices->rotating) {
3184                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3185         }
3186
3187         /*
3188          * Mount does not set all options immediately, we can do it now and do
3189          * not have to wait for transaction commit
3190          */
3191         btrfs_apply_pending_changes(fs_info);
3192
3193 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3194         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3195                 ret = btrfsic_mount(fs_info, fs_devices,
3196                                     btrfs_test_opt(fs_info,
3197                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3198                                     1 : 0,
3199                                     fs_info->check_integrity_print_mask);
3200                 if (ret)
3201                         btrfs_warn(fs_info,
3202                                 "failed to initialize integrity check module: %d",
3203                                 ret);
3204         }
3205 #endif
3206         ret = btrfs_read_qgroup_config(fs_info);
3207         if (ret)
3208                 goto fail_trans_kthread;
3209
3210         if (btrfs_build_ref_tree(fs_info))
3211                 btrfs_err(fs_info, "couldn't build ref tree");
3212
3213         /* do not make disk changes in broken FS or nologreplay is given */
3214         if (btrfs_super_log_root(disk_super) != 0 &&
3215             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3216                 btrfs_info(fs_info, "start tree-log replay");
3217                 ret = btrfs_replay_log(fs_info, fs_devices);
3218                 if (ret) {
3219                         err = ret;
3220                         goto fail_qgroup;
3221                 }
3222         }
3223
3224         ret = btrfs_find_orphan_roots(fs_info);
3225         if (ret)
3226                 goto fail_qgroup;
3227
3228         if (!sb_rdonly(sb)) {
3229                 ret = btrfs_cleanup_fs_roots(fs_info);
3230                 if (ret)
3231                         goto fail_qgroup;
3232
3233                 mutex_lock(&fs_info->cleaner_mutex);
3234                 ret = btrfs_recover_relocation(tree_root);
3235                 mutex_unlock(&fs_info->cleaner_mutex);
3236                 if (ret < 0) {
3237                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3238                                         ret);
3239                         err = -EINVAL;
3240                         goto fail_qgroup;
3241                 }
3242         }
3243
3244         location.objectid = BTRFS_FS_TREE_OBJECTID;
3245         location.type = BTRFS_ROOT_ITEM_KEY;
3246         location.offset = 0;
3247
3248         fs_info->fs_root = btrfs_get_fs_root(fs_info, &location, true);
3249         if (IS_ERR(fs_info->fs_root)) {
3250                 err = PTR_ERR(fs_info->fs_root);
3251                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3252                 fs_info->fs_root = NULL;
3253                 goto fail_qgroup;
3254         }
3255
3256         if (sb_rdonly(sb))
3257                 return 0;
3258
3259         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3260             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3261                 clear_free_space_tree = 1;
3262         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3263                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3264                 btrfs_warn(fs_info, "free space tree is invalid");
3265                 clear_free_space_tree = 1;
3266         }
3267
3268         if (clear_free_space_tree) {
3269                 btrfs_info(fs_info, "clearing free space tree");
3270                 ret = btrfs_clear_free_space_tree(fs_info);
3271                 if (ret) {
3272                         btrfs_warn(fs_info,
3273                                    "failed to clear free space tree: %d", ret);
3274                         close_ctree(fs_info);
3275                         return ret;
3276                 }
3277         }
3278
3279         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3280             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3281                 btrfs_info(fs_info, "creating free space tree");
3282                 ret = btrfs_create_free_space_tree(fs_info);
3283                 if (ret) {
3284                         btrfs_warn(fs_info,
3285                                 "failed to create free space tree: %d", ret);
3286                         close_ctree(fs_info);
3287                         return ret;
3288                 }
3289         }
3290
3291         down_read(&fs_info->cleanup_work_sem);
3292         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3293             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3294                 up_read(&fs_info->cleanup_work_sem);
3295                 close_ctree(fs_info);
3296                 return ret;
3297         }
3298         up_read(&fs_info->cleanup_work_sem);
3299
3300         ret = btrfs_resume_balance_async(fs_info);
3301         if (ret) {
3302                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3303                 close_ctree(fs_info);
3304                 return ret;
3305         }
3306
3307         ret = btrfs_resume_dev_replace_async(fs_info);
3308         if (ret) {
3309                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3310                 close_ctree(fs_info);
3311                 return ret;
3312         }
3313
3314         btrfs_qgroup_rescan_resume(fs_info);
3315         btrfs_discard_resume(fs_info);
3316
3317         if (!fs_info->uuid_root) {
3318                 btrfs_info(fs_info, "creating UUID tree");
3319                 ret = btrfs_create_uuid_tree(fs_info);
3320                 if (ret) {
3321                         btrfs_warn(fs_info,
3322                                 "failed to create the UUID tree: %d", ret);
3323                         close_ctree(fs_info);
3324                         return ret;
3325                 }
3326         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3327                    fs_info->generation !=
3328                                 btrfs_super_uuid_tree_generation(disk_super)) {
3329                 btrfs_info(fs_info, "checking UUID tree");
3330                 ret = btrfs_check_uuid_tree(fs_info);
3331                 if (ret) {
3332                         btrfs_warn(fs_info,
3333                                 "failed to check the UUID tree: %d", ret);
3334                         close_ctree(fs_info);
3335                         return ret;
3336                 }
3337         }
3338         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3339
3340         /*
3341          * backuproot only affect mount behavior, and if open_ctree succeeded,
3342          * no need to keep the flag
3343          */
3344         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3345
3346         return 0;
3347
3348 fail_qgroup:
3349         btrfs_free_qgroup_config(fs_info);
3350 fail_trans_kthread:
3351         kthread_stop(fs_info->transaction_kthread);
3352         btrfs_cleanup_transaction(fs_info);
3353         btrfs_free_fs_roots(fs_info);
3354 fail_cleaner:
3355         kthread_stop(fs_info->cleaner_kthread);
3356
3357         /*
3358          * make sure we're done with the btree inode before we stop our
3359          * kthreads
3360          */
3361         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3362
3363 fail_sysfs:
3364         btrfs_sysfs_remove_mounted(fs_info);
3365
3366 fail_fsdev_sysfs:
3367         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3368
3369 fail_block_groups:
3370         btrfs_put_block_group_cache(fs_info);
3371
3372 fail_tree_roots:
3373         free_root_pointers(fs_info, true);
3374         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3375
3376 fail_sb_buffer:
3377         btrfs_stop_all_workers(fs_info);
3378         btrfs_free_block_groups(fs_info);
3379 fail_alloc:
3380         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3381
3382         iput(fs_info->btree_inode);
3383 fail:
3384         btrfs_close_devices(fs_info->fs_devices);
3385         return err;
3386 }
3387 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3388
3389 static void btrfs_end_super_write(struct bio *bio)
3390 {
3391         struct btrfs_device *device = bio->bi_private;
3392         struct bio_vec *bvec;
3393         struct bvec_iter_all iter_all;
3394         struct page *page;
3395
3396         bio_for_each_segment_all(bvec, bio, iter_all) {
3397                 page = bvec->bv_page;
3398
3399                 if (bio->bi_status) {
3400                         btrfs_warn_rl_in_rcu(device->fs_info,
3401                                 "lost page write due to IO error on %s (%d)",
3402                                 rcu_str_deref(device->name),
3403                                 blk_status_to_errno(bio->bi_status));
3404                         ClearPageUptodate(page);
3405                         SetPageError(page);
3406                         btrfs_dev_stat_inc_and_print(device,
3407                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3408                 } else {
3409                         SetPageUptodate(page);
3410                 }
3411
3412                 put_page(page);
3413                 unlock_page(page);
3414         }
3415
3416         bio_put(bio);
3417 }
3418
3419 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3420                                                    int copy_num)
3421 {
3422         struct btrfs_super_block *super;
3423         struct page *page;
3424         u64 bytenr;
3425         struct address_space *mapping = bdev->bd_inode->i_mapping;
3426
3427         bytenr = btrfs_sb_offset(copy_num);
3428         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3429                 return ERR_PTR(-EINVAL);
3430
3431         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3432         if (IS_ERR(page))
3433                 return ERR_CAST(page);
3434
3435         super = page_address(page);
3436         if (btrfs_super_bytenr(super) != bytenr ||
3437                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3438                 btrfs_release_disk_super(super);
3439                 return ERR_PTR(-EINVAL);
3440         }
3441
3442         return super;
3443 }
3444
3445
3446 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3447 {
3448         struct btrfs_super_block *super, *latest = NULL;
3449         int i;
3450         u64 transid = 0;
3451
3452         /* we would like to check all the supers, but that would make
3453          * a btrfs mount succeed after a mkfs from a different FS.
3454          * So, we need to add a special mount option to scan for
3455          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3456          */
3457         for (i = 0; i < 1; i++) {
3458                 super = btrfs_read_dev_one_super(bdev, i);
3459                 if (IS_ERR(super))
3460                         continue;
3461
3462                 if (!latest || btrfs_super_generation(super) > transid) {
3463                         if (latest)
3464                                 btrfs_release_disk_super(super);
3465
3466                         latest = super;
3467                         transid = btrfs_super_generation(super);
3468                 }
3469         }
3470
3471         return super;
3472 }
3473
3474 /*
3475  * Write superblock @sb to the @device. Do not wait for completion, all the
3476  * pages we use for writing are locked.
3477  *
3478  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3479  * the expected device size at commit time. Note that max_mirrors must be
3480  * same for write and wait phases.
3481  *
3482  * Return number of errors when page is not found or submission fails.
3483  */
3484 static int write_dev_supers(struct btrfs_device *device,
3485                             struct btrfs_super_block *sb, int max_mirrors)
3486 {
3487         struct btrfs_fs_info *fs_info = device->fs_info;
3488         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3489         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3490         int i;
3491         int errors = 0;
3492         u64 bytenr;
3493
3494         if (max_mirrors == 0)
3495                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3496
3497         shash->tfm = fs_info->csum_shash;
3498
3499         for (i = 0; i < max_mirrors; i++) {
3500                 struct page *page;
3501                 struct bio *bio;
3502                 struct btrfs_super_block *disk_super;
3503
3504                 bytenr = btrfs_sb_offset(i);
3505                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3506                     device->commit_total_bytes)
3507                         break;
3508
3509                 btrfs_set_super_bytenr(sb, bytenr);
3510
3511                 crypto_shash_init(shash);
3512                 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3513                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3514                 crypto_shash_final(shash, sb->csum);
3515
3516                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3517                                            GFP_NOFS);
3518                 if (!page) {
3519                         btrfs_err(device->fs_info,
3520                             "couldn't get super block page for bytenr %llu",
3521                             bytenr);
3522                         errors++;
3523                         continue;
3524                 }
3525
3526                 /* Bump the refcount for wait_dev_supers() */
3527                 get_page(page);
3528
3529                 disk_super = page_address(page);
3530                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3531
3532                 /*
3533                  * Directly use bios here instead of relying on the page cache
3534                  * to do I/O, so we don't lose the ability to do integrity
3535                  * checking.
3536                  */
3537                 bio = bio_alloc(GFP_NOFS, 1);
3538                 bio_set_dev(bio, device->bdev);
3539                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3540                 bio->bi_private = device;
3541                 bio->bi_end_io = btrfs_end_super_write;
3542                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3543                                offset_in_page(bytenr));
3544
3545                 /*
3546                  * We FUA only the first super block.  The others we allow to
3547                  * go down lazy and there's a short window where the on-disk
3548                  * copies might still contain the older version.
3549                  */
3550                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3551                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3552                         bio->bi_opf |= REQ_FUA;
3553
3554                 btrfsic_submit_bio(bio);
3555         }
3556         return errors < i ? 0 : -1;
3557 }
3558
3559 /*
3560  * Wait for write completion of superblocks done by write_dev_supers,
3561  * @max_mirrors same for write and wait phases.
3562  *
3563  * Return number of errors when page is not found or not marked up to
3564  * date.
3565  */
3566 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3567 {
3568         int i;
3569         int errors = 0;
3570         bool primary_failed = false;
3571         u64 bytenr;
3572
3573         if (max_mirrors == 0)
3574                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3575
3576         for (i = 0; i < max_mirrors; i++) {
3577                 struct page *page;
3578
3579                 bytenr = btrfs_sb_offset(i);
3580                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3581                     device->commit_total_bytes)
3582                         break;
3583
3584                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3585                                      bytenr >> PAGE_SHIFT);
3586                 if (!page) {
3587                         errors++;
3588                         if (i == 0)
3589                                 primary_failed = true;
3590                         continue;
3591                 }
3592                 /* Page is submitted locked and unlocked once the IO completes */
3593                 wait_on_page_locked(page);
3594                 if (PageError(page)) {
3595                         errors++;
3596                         if (i == 0)
3597                                 primary_failed = true;
3598                 }
3599
3600                 /* Drop our reference */
3601                 put_page(page);
3602
3603                 /* Drop the reference from the writing run */
3604                 put_page(page);
3605         }
3606
3607         /* log error, force error return */
3608         if (primary_failed) {
3609                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3610                           device->devid);
3611                 return -1;
3612         }
3613
3614         return errors < i ? 0 : -1;
3615 }
3616
3617 /*
3618  * endio for the write_dev_flush, this will wake anyone waiting
3619  * for the barrier when it is done
3620  */
3621 static void btrfs_end_empty_barrier(struct bio *bio)
3622 {
3623         complete(bio->bi_private);
3624 }
3625
3626 /*
3627  * Submit a flush request to the device if it supports it. Error handling is
3628  * done in the waiting counterpart.
3629  */
3630 static void write_dev_flush(struct btrfs_device *device)
3631 {
3632         struct request_queue *q = bdev_get_queue(device->bdev);
3633         struct bio *bio = device->flush_bio;
3634
3635         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3636                 return;
3637
3638         bio_reset(bio);
3639         bio->bi_end_io = btrfs_end_empty_barrier;
3640         bio_set_dev(bio, device->bdev);
3641         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3642         init_completion(&device->flush_wait);
3643         bio->bi_private = &device->flush_wait;
3644
3645         btrfsic_submit_bio(bio);
3646         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3647 }
3648
3649 /*
3650  * If the flush bio has been submitted by write_dev_flush, wait for it.
3651  */
3652 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3653 {
3654         struct bio *bio = device->flush_bio;
3655
3656         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3657                 return BLK_STS_OK;
3658
3659         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3660         wait_for_completion_io(&device->flush_wait);
3661
3662         return bio->bi_status;
3663 }
3664
3665 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3666 {
3667         if (!btrfs_check_rw_degradable(fs_info, NULL))
3668                 return -EIO;
3669         return 0;
3670 }
3671
3672 /*
3673  * send an empty flush down to each device in parallel,
3674  * then wait for them
3675  */
3676 static int barrier_all_devices(struct btrfs_fs_info *info)
3677 {
3678         struct list_head *head;
3679         struct btrfs_device *dev;
3680         int errors_wait = 0;
3681         blk_status_t ret;
3682
3683         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3684         /* send down all the barriers */
3685         head = &info->fs_devices->devices;
3686         list_for_each_entry(dev, head, dev_list) {
3687                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3688                         continue;
3689                 if (!dev->bdev)
3690                         continue;
3691                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3692                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3693                         continue;
3694
3695                 write_dev_flush(dev);
3696                 dev->last_flush_error = BLK_STS_OK;
3697         }
3698
3699         /* wait for all the barriers */
3700         list_for_each_entry(dev, head, dev_list) {
3701                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3702                         continue;
3703                 if (!dev->bdev) {
3704                         errors_wait++;
3705                         continue;
3706                 }
3707                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3708                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3709                         continue;
3710
3711                 ret = wait_dev_flush(dev);
3712                 if (ret) {
3713                         dev->last_flush_error = ret;
3714                         btrfs_dev_stat_inc_and_print(dev,
3715                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3716                         errors_wait++;
3717                 }
3718         }
3719
3720         if (errors_wait) {
3721                 /*
3722                  * At some point we need the status of all disks
3723                  * to arrive at the volume status. So error checking
3724                  * is being pushed to a separate loop.
3725                  */
3726                 return check_barrier_error(info);
3727         }
3728         return 0;
3729 }
3730
3731 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3732 {
3733         int raid_type;
3734         int min_tolerated = INT_MAX;
3735
3736         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3737             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3738                 min_tolerated = min_t(int, min_tolerated,
3739                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3740                                     tolerated_failures);
3741
3742         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3743                 if (raid_type == BTRFS_RAID_SINGLE)
3744                         continue;
3745                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3746                         continue;
3747                 min_tolerated = min_t(int, min_tolerated,
3748                                     btrfs_raid_array[raid_type].
3749                                     tolerated_failures);
3750         }
3751
3752         if (min_tolerated == INT_MAX) {
3753                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3754                 min_tolerated = 0;
3755         }
3756
3757         return min_tolerated;
3758 }
3759
3760 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3761 {
3762         struct list_head *head;
3763         struct btrfs_device *dev;
3764         struct btrfs_super_block *sb;
3765         struct btrfs_dev_item *dev_item;
3766         int ret;
3767         int do_barriers;
3768         int max_errors;
3769         int total_errors = 0;
3770         u64 flags;
3771
3772         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3773
3774         /*
3775          * max_mirrors == 0 indicates we're from commit_transaction,
3776          * not from fsync where the tree roots in fs_info have not
3777          * been consistent on disk.
3778          */
3779         if (max_mirrors == 0)
3780                 backup_super_roots(fs_info);
3781
3782         sb = fs_info->super_for_commit;
3783         dev_item = &sb->dev_item;
3784
3785         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3786         head = &fs_info->fs_devices->devices;
3787         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3788
3789         if (do_barriers) {
3790                 ret = barrier_all_devices(fs_info);
3791                 if (ret) {
3792                         mutex_unlock(
3793                                 &fs_info->fs_devices->device_list_mutex);
3794                         btrfs_handle_fs_error(fs_info, ret,
3795                                               "errors while submitting device barriers.");
3796                         return ret;
3797                 }
3798         }
3799
3800         list_for_each_entry(dev, head, dev_list) {
3801                 if (!dev->bdev) {
3802                         total_errors++;
3803                         continue;
3804                 }
3805                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3806                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3807                         continue;
3808
3809                 btrfs_set_stack_device_generation(dev_item, 0);
3810                 btrfs_set_stack_device_type(dev_item, dev->type);
3811                 btrfs_set_stack_device_id(dev_item, dev->devid);
3812                 btrfs_set_stack_device_total_bytes(dev_item,
3813                                                    dev->commit_total_bytes);
3814                 btrfs_set_stack_device_bytes_used(dev_item,
3815                                                   dev->commit_bytes_used);
3816                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3817                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3818                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3819                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3820                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3821                        BTRFS_FSID_SIZE);
3822
3823                 flags = btrfs_super_flags(sb);
3824                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3825
3826                 ret = btrfs_validate_write_super(fs_info, sb);
3827                 if (ret < 0) {
3828                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3829                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3830                                 "unexpected superblock corruption detected");
3831                         return -EUCLEAN;
3832                 }
3833
3834                 ret = write_dev_supers(dev, sb, max_mirrors);
3835                 if (ret)
3836                         total_errors++;
3837         }
3838         if (total_errors > max_errors) {
3839                 btrfs_err(fs_info, "%d errors while writing supers",
3840                           total_errors);
3841                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3842
3843                 /* FUA is masked off if unsupported and can't be the reason */
3844                 btrfs_handle_fs_error(fs_info, -EIO,
3845                                       "%d errors while writing supers",
3846                                       total_errors);
3847                 return -EIO;
3848         }
3849
3850         total_errors = 0;
3851         list_for_each_entry(dev, head, dev_list) {
3852                 if (!dev->bdev)
3853                         continue;
3854                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3855                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3856                         continue;
3857
3858                 ret = wait_dev_supers(dev, max_mirrors);
3859                 if (ret)
3860                         total_errors++;
3861         }
3862         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3863         if (total_errors > max_errors) {
3864                 btrfs_handle_fs_error(fs_info, -EIO,
3865                                       "%d errors while writing supers",
3866                                       total_errors);
3867                 return -EIO;
3868         }
3869         return 0;
3870 }
3871
3872 /* Drop a fs root from the radix tree and free it. */
3873 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3874                                   struct btrfs_root *root)
3875 {
3876         bool drop_ref = false;
3877
3878         spin_lock(&fs_info->fs_roots_radix_lock);
3879         radix_tree_delete(&fs_info->fs_roots_radix,
3880                           (unsigned long)root->root_key.objectid);
3881         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3882                 drop_ref = true;
3883         spin_unlock(&fs_info->fs_roots_radix_lock);
3884
3885         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3886                 ASSERT(root->log_root == NULL);
3887                 if (root->reloc_root) {
3888                         btrfs_put_root(root->reloc_root);
3889                         root->reloc_root = NULL;
3890                 }
3891         }
3892
3893         if (root->free_ino_pinned)
3894                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3895         if (root->free_ino_ctl)
3896                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3897         if (root->ino_cache_inode) {
3898                 iput(root->ino_cache_inode);
3899                 root->ino_cache_inode = NULL;
3900         }
3901         if (drop_ref)
3902                 btrfs_put_root(root);
3903 }
3904
3905 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3906 {
3907         u64 root_objectid = 0;
3908         struct btrfs_root *gang[8];
3909         int i = 0;
3910         int err = 0;
3911         unsigned int ret = 0;
3912
3913         while (1) {
3914                 spin_lock(&fs_info->fs_roots_radix_lock);
3915                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3916                                              (void **)gang, root_objectid,
3917                                              ARRAY_SIZE(gang));
3918                 if (!ret) {
3919                         spin_unlock(&fs_info->fs_roots_radix_lock);
3920                         break;
3921                 }
3922                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3923
3924                 for (i = 0; i < ret; i++) {
3925                         /* Avoid to grab roots in dead_roots */
3926                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3927                                 gang[i] = NULL;
3928                                 continue;
3929                         }
3930                         /* grab all the search result for later use */
3931                         gang[i] = btrfs_grab_root(gang[i]);
3932                 }
3933                 spin_unlock(&fs_info->fs_roots_radix_lock);
3934
3935                 for (i = 0; i < ret; i++) {
3936                         if (!gang[i])
3937                                 continue;
3938                         root_objectid = gang[i]->root_key.objectid;
3939                         err = btrfs_orphan_cleanup(gang[i]);
3940                         if (err)
3941                                 break;
3942                         btrfs_put_root(gang[i]);
3943                 }
3944                 root_objectid++;
3945         }
3946
3947         /* release the uncleaned roots due to error */
3948         for (; i < ret; i++) {
3949                 if (gang[i])
3950                         btrfs_put_root(gang[i]);
3951         }
3952         return err;
3953 }
3954
3955 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3956 {
3957         struct btrfs_root *root = fs_info->tree_root;
3958         struct btrfs_trans_handle *trans;
3959
3960         mutex_lock(&fs_info->cleaner_mutex);
3961         btrfs_run_delayed_iputs(fs_info);
3962         mutex_unlock(&fs_info->cleaner_mutex);
3963         wake_up_process(fs_info->cleaner_kthread);
3964
3965         /* wait until ongoing cleanup work done */
3966         down_write(&fs_info->cleanup_work_sem);
3967         up_write(&fs_info->cleanup_work_sem);
3968
3969         trans = btrfs_join_transaction(root);
3970         if (IS_ERR(trans))
3971                 return PTR_ERR(trans);
3972         return btrfs_commit_transaction(trans);
3973 }
3974
3975 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3976 {
3977         int ret;
3978
3979         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3980         /*
3981          * We don't want the cleaner to start new transactions, add more delayed
3982          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3983          * because that frees the task_struct, and the transaction kthread might
3984          * still try to wake up the cleaner.
3985          */
3986         kthread_park(fs_info->cleaner_kthread);
3987
3988         /* wait for the qgroup rescan worker to stop */
3989         btrfs_qgroup_wait_for_completion(fs_info, false);
3990
3991         /* wait for the uuid_scan task to finish */
3992         down(&fs_info->uuid_tree_rescan_sem);
3993         /* avoid complains from lockdep et al., set sem back to initial state */
3994         up(&fs_info->uuid_tree_rescan_sem);
3995
3996         /* pause restriper - we want to resume on mount */
3997         btrfs_pause_balance(fs_info);
3998
3999         btrfs_dev_replace_suspend_for_unmount(fs_info);
4000
4001         btrfs_scrub_cancel(fs_info);
4002
4003         /* wait for any defraggers to finish */
4004         wait_event(fs_info->transaction_wait,
4005                    (atomic_read(&fs_info->defrag_running) == 0));
4006
4007         /* clear out the rbtree of defraggable inodes */
4008         btrfs_cleanup_defrag_inodes(fs_info);
4009
4010         cancel_work_sync(&fs_info->async_reclaim_work);
4011
4012         /* Cancel or finish ongoing discard work */
4013         btrfs_discard_cleanup(fs_info);
4014
4015         if (!sb_rdonly(fs_info->sb)) {
4016                 /*
4017                  * The cleaner kthread is stopped, so do one final pass over
4018                  * unused block groups.
4019                  */
4020                 btrfs_delete_unused_bgs(fs_info);
4021
4022                 /*
4023                  * There might be existing delayed inode workers still running
4024                  * and holding an empty delayed inode item. We must wait for
4025                  * them to complete first because they can create a transaction.
4026                  * This happens when someone calls btrfs_balance_delayed_items()
4027                  * and then a transaction commit runs the same delayed nodes
4028                  * before any delayed worker has done something with the nodes.
4029                  * We must wait for any worker here and not at transaction
4030                  * commit time since that could cause a deadlock.
4031                  * This is a very rare case.
4032                  */
4033                 btrfs_flush_workqueue(fs_info->delayed_workers);
4034
4035                 ret = btrfs_commit_super(fs_info);
4036                 if (ret)
4037                         btrfs_err(fs_info, "commit super ret %d", ret);
4038         }
4039
4040         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4041             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4042                 btrfs_error_commit_super(fs_info);
4043
4044         kthread_stop(fs_info->transaction_kthread);
4045         kthread_stop(fs_info->cleaner_kthread);
4046
4047         ASSERT(list_empty(&fs_info->delayed_iputs));
4048         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4049
4050         btrfs_free_qgroup_config(fs_info);
4051         ASSERT(list_empty(&fs_info->delalloc_roots));
4052
4053         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4054                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4055                        percpu_counter_sum(&fs_info->delalloc_bytes));
4056         }
4057
4058         if (percpu_counter_sum(&fs_info->dio_bytes))
4059                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4060                            percpu_counter_sum(&fs_info->dio_bytes));
4061
4062         btrfs_sysfs_remove_mounted(fs_info);
4063         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4064
4065         btrfs_put_block_group_cache(fs_info);
4066
4067         /*
4068          * we must make sure there is not any read request to
4069          * submit after we stopping all workers.
4070          */
4071         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4072         btrfs_stop_all_workers(fs_info);
4073
4074         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4075         free_root_pointers(fs_info, true);
4076         btrfs_free_fs_roots(fs_info);
4077
4078         /*
4079          * We must free the block groups after dropping the fs_roots as we could
4080          * have had an IO error and have left over tree log blocks that aren't
4081          * cleaned up until the fs roots are freed.  This makes the block group
4082          * accounting appear to be wrong because there's pending reserved bytes,
4083          * so make sure we do the block group cleanup afterwards.
4084          */
4085         btrfs_free_block_groups(fs_info);
4086
4087         iput(fs_info->btree_inode);
4088
4089 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4090         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4091                 btrfsic_unmount(fs_info->fs_devices);
4092 #endif
4093
4094         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4095         btrfs_close_devices(fs_info->fs_devices);
4096 }
4097
4098 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4099                           int atomic)
4100 {
4101         int ret;
4102         struct inode *btree_inode = buf->pages[0]->mapping->host;
4103
4104         ret = extent_buffer_uptodate(buf);
4105         if (!ret)
4106                 return ret;
4107
4108         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4109                                     parent_transid, atomic);
4110         if (ret == -EAGAIN)
4111                 return ret;
4112         return !ret;
4113 }
4114
4115 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4116 {
4117         struct btrfs_fs_info *fs_info;
4118         struct btrfs_root *root;
4119         u64 transid = btrfs_header_generation(buf);
4120         int was_dirty;
4121
4122 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4123         /*
4124          * This is a fast path so only do this check if we have sanity tests
4125          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4126          * outside of the sanity tests.
4127          */
4128         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4129                 return;
4130 #endif
4131         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4132         fs_info = root->fs_info;
4133         btrfs_assert_tree_locked(buf);
4134         if (transid != fs_info->generation)
4135                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4136                         buf->start, transid, fs_info->generation);
4137         was_dirty = set_extent_buffer_dirty(buf);
4138         if (!was_dirty)
4139                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4140                                          buf->len,
4141                                          fs_info->dirty_metadata_batch);
4142 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4143         /*
4144          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4145          * but item data not updated.
4146          * So here we should only check item pointers, not item data.
4147          */
4148         if (btrfs_header_level(buf) == 0 &&
4149             btrfs_check_leaf_relaxed(buf)) {
4150                 btrfs_print_leaf(buf);
4151                 ASSERT(0);
4152         }
4153 #endif
4154 }
4155
4156 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4157                                         int flush_delayed)
4158 {
4159         /*
4160          * looks as though older kernels can get into trouble with
4161          * this code, they end up stuck in balance_dirty_pages forever
4162          */
4163         int ret;
4164
4165         if (current->flags & PF_MEMALLOC)
4166                 return;
4167
4168         if (flush_delayed)
4169                 btrfs_balance_delayed_items(fs_info);
4170
4171         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4172                                      BTRFS_DIRTY_METADATA_THRESH,
4173                                      fs_info->dirty_metadata_batch);
4174         if (ret > 0) {
4175                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4176         }
4177 }
4178
4179 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4180 {
4181         __btrfs_btree_balance_dirty(fs_info, 1);
4182 }
4183
4184 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4185 {
4186         __btrfs_btree_balance_dirty(fs_info, 0);
4187 }
4188
4189 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4190                       struct btrfs_key *first_key)
4191 {
4192         return btree_read_extent_buffer_pages(buf, parent_transid,
4193                                               level, first_key);
4194 }
4195
4196 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4197 {
4198         /* cleanup FS via transaction */
4199         btrfs_cleanup_transaction(fs_info);
4200
4201         mutex_lock(&fs_info->cleaner_mutex);
4202         btrfs_run_delayed_iputs(fs_info);
4203         mutex_unlock(&fs_info->cleaner_mutex);
4204
4205         down_write(&fs_info->cleanup_work_sem);
4206         up_write(&fs_info->cleanup_work_sem);
4207 }
4208
4209 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4210 {
4211         struct btrfs_root *gang[8];
4212         u64 root_objectid = 0;
4213         int ret;
4214
4215         spin_lock(&fs_info->fs_roots_radix_lock);
4216         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4217                                              (void **)gang, root_objectid,
4218                                              ARRAY_SIZE(gang))) != 0) {
4219                 int i;
4220
4221                 for (i = 0; i < ret; i++)
4222                         gang[i] = btrfs_grab_root(gang[i]);
4223                 spin_unlock(&fs_info->fs_roots_radix_lock);
4224
4225                 for (i = 0; i < ret; i++) {
4226                         if (!gang[i])
4227                                 continue;
4228                         root_objectid = gang[i]->root_key.objectid;
4229                         btrfs_free_log(NULL, gang[i]);
4230                         btrfs_put_root(gang[i]);
4231                 }
4232                 root_objectid++;
4233                 spin_lock(&fs_info->fs_roots_radix_lock);
4234         }
4235         spin_unlock(&fs_info->fs_roots_radix_lock);
4236         btrfs_free_log_root_tree(NULL, fs_info);
4237 }
4238
4239 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4240 {
4241         struct btrfs_ordered_extent *ordered;
4242
4243         spin_lock(&root->ordered_extent_lock);
4244         /*
4245          * This will just short circuit the ordered completion stuff which will
4246          * make sure the ordered extent gets properly cleaned up.
4247          */
4248         list_for_each_entry(ordered, &root->ordered_extents,
4249                             root_extent_list)
4250                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4251         spin_unlock(&root->ordered_extent_lock);
4252 }
4253
4254 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4255 {
4256         struct btrfs_root *root;
4257         struct list_head splice;
4258
4259         INIT_LIST_HEAD(&splice);
4260
4261         spin_lock(&fs_info->ordered_root_lock);
4262         list_splice_init(&fs_info->ordered_roots, &splice);
4263         while (!list_empty(&splice)) {
4264                 root = list_first_entry(&splice, struct btrfs_root,
4265                                         ordered_root);
4266                 list_move_tail(&root->ordered_root,
4267                                &fs_info->ordered_roots);
4268
4269                 spin_unlock(&fs_info->ordered_root_lock);
4270                 btrfs_destroy_ordered_extents(root);
4271
4272                 cond_resched();
4273                 spin_lock(&fs_info->ordered_root_lock);
4274         }
4275         spin_unlock(&fs_info->ordered_root_lock);
4276
4277         /*
4278          * We need this here because if we've been flipped read-only we won't
4279          * get sync() from the umount, so we need to make sure any ordered
4280          * extents that haven't had their dirty pages IO start writeout yet
4281          * actually get run and error out properly.
4282          */
4283         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4284 }
4285
4286 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4287                                       struct btrfs_fs_info *fs_info)
4288 {
4289         struct rb_node *node;
4290         struct btrfs_delayed_ref_root *delayed_refs;
4291         struct btrfs_delayed_ref_node *ref;
4292         int ret = 0;
4293
4294         delayed_refs = &trans->delayed_refs;
4295
4296         spin_lock(&delayed_refs->lock);
4297         if (atomic_read(&delayed_refs->num_entries) == 0) {
4298                 spin_unlock(&delayed_refs->lock);
4299                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4300                 return ret;
4301         }
4302
4303         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4304                 struct btrfs_delayed_ref_head *head;
4305                 struct rb_node *n;
4306                 bool pin_bytes = false;
4307
4308                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4309                                 href_node);
4310                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4311                         continue;
4312
4313                 spin_lock(&head->lock);
4314                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4315                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4316                                        ref_node);
4317                         ref->in_tree = 0;
4318                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4319                         RB_CLEAR_NODE(&ref->ref_node);
4320                         if (!list_empty(&ref->add_list))
4321                                 list_del(&ref->add_list);
4322                         atomic_dec(&delayed_refs->num_entries);
4323                         btrfs_put_delayed_ref(ref);
4324                 }
4325                 if (head->must_insert_reserved)
4326                         pin_bytes = true;
4327                 btrfs_free_delayed_extent_op(head->extent_op);
4328                 btrfs_delete_ref_head(delayed_refs, head);
4329                 spin_unlock(&head->lock);
4330                 spin_unlock(&delayed_refs->lock);
4331                 mutex_unlock(&head->mutex);
4332
4333                 if (pin_bytes) {
4334                         struct btrfs_block_group *cache;
4335
4336                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4337                         BUG_ON(!cache);
4338
4339                         spin_lock(&cache->space_info->lock);
4340                         spin_lock(&cache->lock);
4341                         cache->pinned += head->num_bytes;
4342                         btrfs_space_info_update_bytes_pinned(fs_info,
4343                                 cache->space_info, head->num_bytes);
4344                         cache->reserved -= head->num_bytes;
4345                         cache->space_info->bytes_reserved -= head->num_bytes;
4346                         spin_unlock(&cache->lock);
4347                         spin_unlock(&cache->space_info->lock);
4348                         percpu_counter_add_batch(
4349                                 &cache->space_info->total_bytes_pinned,
4350                                 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4351
4352                         btrfs_put_block_group(cache);
4353
4354                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4355                                 head->bytenr + head->num_bytes - 1);
4356                 }
4357                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4358                 btrfs_put_delayed_ref_head(head);
4359                 cond_resched();
4360                 spin_lock(&delayed_refs->lock);
4361         }
4362         btrfs_qgroup_destroy_extent_records(trans);
4363
4364         spin_unlock(&delayed_refs->lock);
4365
4366         return ret;
4367 }
4368
4369 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4370 {
4371         struct btrfs_inode *btrfs_inode;
4372         struct list_head splice;
4373
4374         INIT_LIST_HEAD(&splice);
4375
4376         spin_lock(&root->delalloc_lock);
4377         list_splice_init(&root->delalloc_inodes, &splice);
4378
4379         while (!list_empty(&splice)) {
4380                 struct inode *inode = NULL;
4381                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4382                                                delalloc_inodes);
4383                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4384                 spin_unlock(&root->delalloc_lock);
4385
4386                 /*
4387                  * Make sure we get a live inode and that it'll not disappear
4388                  * meanwhile.
4389                  */
4390                 inode = igrab(&btrfs_inode->vfs_inode);
4391                 if (inode) {
4392                         invalidate_inode_pages2(inode->i_mapping);
4393                         iput(inode);
4394                 }
4395                 spin_lock(&root->delalloc_lock);
4396         }
4397         spin_unlock(&root->delalloc_lock);
4398 }
4399
4400 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4401 {
4402         struct btrfs_root *root;
4403         struct list_head splice;
4404
4405         INIT_LIST_HEAD(&splice);
4406
4407         spin_lock(&fs_info->delalloc_root_lock);
4408         list_splice_init(&fs_info->delalloc_roots, &splice);
4409         while (!list_empty(&splice)) {
4410                 root = list_first_entry(&splice, struct btrfs_root,
4411                                          delalloc_root);
4412                 root = btrfs_grab_root(root);
4413                 BUG_ON(!root);
4414                 spin_unlock(&fs_info->delalloc_root_lock);
4415
4416                 btrfs_destroy_delalloc_inodes(root);
4417                 btrfs_put_root(root);
4418
4419                 spin_lock(&fs_info->delalloc_root_lock);
4420         }
4421         spin_unlock(&fs_info->delalloc_root_lock);
4422 }
4423
4424 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4425                                         struct extent_io_tree *dirty_pages,
4426                                         int mark)
4427 {
4428         int ret;
4429         struct extent_buffer *eb;
4430         u64 start = 0;
4431         u64 end;
4432
4433         while (1) {
4434                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4435                                             mark, NULL);
4436                 if (ret)
4437                         break;
4438
4439                 clear_extent_bits(dirty_pages, start, end, mark);
4440                 while (start <= end) {
4441                         eb = find_extent_buffer(fs_info, start);
4442                         start += fs_info->nodesize;
4443                         if (!eb)
4444                                 continue;
4445                         wait_on_extent_buffer_writeback(eb);
4446
4447                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4448                                                &eb->bflags))
4449                                 clear_extent_buffer_dirty(eb);
4450                         free_extent_buffer_stale(eb);
4451                 }
4452         }
4453
4454         return ret;
4455 }
4456
4457 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4458                                        struct extent_io_tree *unpin)
4459 {
4460         u64 start;
4461         u64 end;
4462         int ret;
4463
4464         while (1) {
4465                 struct extent_state *cached_state = NULL;
4466
4467                 /*
4468                  * The btrfs_finish_extent_commit() may get the same range as
4469                  * ours between find_first_extent_bit and clear_extent_dirty.
4470                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4471                  * the same extent range.
4472                  */
4473                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4474                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4475                                             EXTENT_DIRTY, &cached_state);
4476                 if (ret) {
4477                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4478                         break;
4479                 }
4480
4481                 clear_extent_dirty(unpin, start, end, &cached_state);
4482                 free_extent_state(cached_state);
4483                 btrfs_error_unpin_extent_range(fs_info, start, end);
4484                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4485                 cond_resched();
4486         }
4487
4488         return 0;
4489 }
4490
4491 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4492 {
4493         struct inode *inode;
4494
4495         inode = cache->io_ctl.inode;
4496         if (inode) {
4497                 invalidate_inode_pages2(inode->i_mapping);
4498                 BTRFS_I(inode)->generation = 0;
4499                 cache->io_ctl.inode = NULL;
4500                 iput(inode);
4501         }
4502         btrfs_put_block_group(cache);
4503 }
4504
4505 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4506                              struct btrfs_fs_info *fs_info)
4507 {
4508         struct btrfs_block_group *cache;
4509
4510         spin_lock(&cur_trans->dirty_bgs_lock);
4511         while (!list_empty(&cur_trans->dirty_bgs)) {
4512                 cache = list_first_entry(&cur_trans->dirty_bgs,
4513                                          struct btrfs_block_group,
4514                                          dirty_list);
4515
4516                 if (!list_empty(&cache->io_list)) {
4517                         spin_unlock(&cur_trans->dirty_bgs_lock);
4518                         list_del_init(&cache->io_list);
4519                         btrfs_cleanup_bg_io(cache);
4520                         spin_lock(&cur_trans->dirty_bgs_lock);
4521                 }
4522
4523                 list_del_init(&cache->dirty_list);
4524                 spin_lock(&cache->lock);
4525                 cache->disk_cache_state = BTRFS_DC_ERROR;
4526                 spin_unlock(&cache->lock);
4527
4528                 spin_unlock(&cur_trans->dirty_bgs_lock);
4529                 btrfs_put_block_group(cache);
4530                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4531                 spin_lock(&cur_trans->dirty_bgs_lock);
4532         }
4533         spin_unlock(&cur_trans->dirty_bgs_lock);
4534
4535         /*
4536          * Refer to the definition of io_bgs member for details why it's safe
4537          * to use it without any locking
4538          */
4539         while (!list_empty(&cur_trans->io_bgs)) {
4540                 cache = list_first_entry(&cur_trans->io_bgs,
4541                                          struct btrfs_block_group,
4542                                          io_list);
4543
4544                 list_del_init(&cache->io_list);
4545                 spin_lock(&cache->lock);
4546                 cache->disk_cache_state = BTRFS_DC_ERROR;
4547                 spin_unlock(&cache->lock);
4548                 btrfs_cleanup_bg_io(cache);
4549         }
4550 }
4551
4552 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4553                                    struct btrfs_fs_info *fs_info)
4554 {
4555         struct btrfs_device *dev, *tmp;
4556
4557         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4558         ASSERT(list_empty(&cur_trans->dirty_bgs));
4559         ASSERT(list_empty(&cur_trans->io_bgs));
4560
4561         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4562                                  post_commit_list) {
4563                 list_del_init(&dev->post_commit_list);
4564         }
4565
4566         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4567
4568         cur_trans->state = TRANS_STATE_COMMIT_START;
4569         wake_up(&fs_info->transaction_blocked_wait);
4570
4571         cur_trans->state = TRANS_STATE_UNBLOCKED;
4572         wake_up(&fs_info->transaction_wait);
4573
4574         btrfs_destroy_delayed_inodes(fs_info);
4575
4576         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4577                                      EXTENT_DIRTY);
4578         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4579
4580         cur_trans->state =TRANS_STATE_COMPLETED;
4581         wake_up(&cur_trans->commit_wait);
4582 }
4583
4584 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4585 {
4586         struct btrfs_transaction *t;
4587
4588         mutex_lock(&fs_info->transaction_kthread_mutex);
4589
4590         spin_lock(&fs_info->trans_lock);
4591         while (!list_empty(&fs_info->trans_list)) {
4592                 t = list_first_entry(&fs_info->trans_list,
4593                                      struct btrfs_transaction, list);
4594                 if (t->state >= TRANS_STATE_COMMIT_START) {
4595                         refcount_inc(&t->use_count);
4596                         spin_unlock(&fs_info->trans_lock);
4597                         btrfs_wait_for_commit(fs_info, t->transid);
4598                         btrfs_put_transaction(t);
4599                         spin_lock(&fs_info->trans_lock);
4600                         continue;
4601                 }
4602                 if (t == fs_info->running_transaction) {
4603                         t->state = TRANS_STATE_COMMIT_DOING;
4604                         spin_unlock(&fs_info->trans_lock);
4605                         /*
4606                          * We wait for 0 num_writers since we don't hold a trans
4607                          * handle open currently for this transaction.
4608                          */
4609                         wait_event(t->writer_wait,
4610                                    atomic_read(&t->num_writers) == 0);
4611                 } else {
4612                         spin_unlock(&fs_info->trans_lock);
4613                 }
4614                 btrfs_cleanup_one_transaction(t, fs_info);
4615
4616                 spin_lock(&fs_info->trans_lock);
4617                 if (t == fs_info->running_transaction)
4618                         fs_info->running_transaction = NULL;
4619                 list_del_init(&t->list);
4620                 spin_unlock(&fs_info->trans_lock);
4621
4622                 btrfs_put_transaction(t);
4623                 trace_btrfs_transaction_commit(fs_info->tree_root);
4624                 spin_lock(&fs_info->trans_lock);
4625         }
4626         spin_unlock(&fs_info->trans_lock);
4627         btrfs_destroy_all_ordered_extents(fs_info);
4628         btrfs_destroy_delayed_inodes(fs_info);
4629         btrfs_assert_delayed_root_empty(fs_info);
4630         btrfs_destroy_all_delalloc_inodes(fs_info);
4631         btrfs_drop_all_logs(fs_info);
4632         mutex_unlock(&fs_info->transaction_kthread_mutex);
4633
4634         return 0;
4635 }
4636
4637 static const struct extent_io_ops btree_extent_io_ops = {
4638         /* mandatory callbacks */
4639         .submit_bio_hook = btree_submit_bio_hook,
4640         .readpage_end_io_hook = btree_readpage_end_io_hook,
4641 };