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/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <asm/unaligned.h>
  21 #include "ctree.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24 #include "btrfs_inode.h"
  25 #include "volumes.h"
  26 #include "print-tree.h"
  27 #include "locking.h"
  28 #include "tree-log.h"
  29 #include "free-space-cache.h"
  30 #include "free-space-tree.h"
  31 #include "inode-map.h"
  32 #include "check-integrity.h"
  33 #include "rcu-string.h"
  34 #include "dev-replace.h"
  35 #include "raid56.h"
  36 #include "sysfs.h"
  37 #include "qgroup.h"
  38 #include "compression.h"
  39 #include "tree-checker.h"
  40 #include "ref-verify.h"
  41
  42 #ifdef CONFIG_X86
  43 #include <asm/cpufeature.h>
  44 #endif
  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 /*
 102  * async submit bios are used to offload expensive checksumming
 103  * onto the worker threads.  They checksum file and metadata bios
 104  * just before they are sent down the IO stack.
 105  */
 106 struct async_submit_bio {
 107         void *private_data;
 108         struct bio *bio;
 109         extent_submit_bio_start_t *submit_bio_start;
 110         int mirror_num;
 111         /*
 112          * bio_offset is optional, can be used if the pages in the bio
 113          * can't tell us where in the file the bio should go
 114          */
 115         u64 bio_offset;
 116         struct btrfs_work work;
 117         blk_status_t status;
 118 };
 119
 120 /*
 121  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 122  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 123  * the level the eb occupies in the tree.
 124  *
 125  * Different roots are used for different purposes and may nest inside each
 126  * other and they require separate keysets.  As lockdep keys should be
 127  * static, assign keysets according to the purpose of the root as indicated
 128  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 129  * roots have separate keysets.
 130  *
 131  * Lock-nesting across peer nodes is always done with the immediate parent
 132  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 133  * subclass to avoid triggering lockdep warning in such cases.
 134  *
 135  * The key is set by the readpage_end_io_hook after the buffer has passed
 136  * csum validation but before the pages are unlocked.  It is also set by
 137  * btrfs_init_new_buffer on freshly allocated blocks.
 138  *
 139  * We also add a check to make sure the highest level of the tree is the
 140  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 141  * needs update as well.
 142  */
 143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 144 # if BTRFS_MAX_LEVEL != 8
 145 #  error
 146 # endif
 147
 148 static struct btrfs_lockdep_keyset {
 149         u64                     id;             /* root objectid */
 150         const char              *name_stem;     /* lock name stem */
 151         char                    names[BTRFS_MAX_LEVEL + 1][20];
 152         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 153 } btrfs_lockdep_keysets[] = {
 154         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 155         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 156         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 157         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 158         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 159         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 160         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 161         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 162         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 163         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 164         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 165         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 166         { .id = 0,                              .name_stem = "tree"     },
 167 };
 168
 169 void __init btrfs_init_lockdep(void)
 170 {
 171         int i, j;
 172
 173         /* initialize lockdep class names */
 174         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 175                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 176
 177                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 178                         snprintf(ks->names[j], sizeof(ks->names[j]),
 179                                  "btrfs-%s-%02d", ks->name_stem, j);
 180         }
 181 }
 182
 183 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 184                                     int level)
 185 {
 186         struct btrfs_lockdep_keyset *ks;
 187
 188         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 189
 190         /* find the matching keyset, id 0 is the default entry */
 191         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 192                 if (ks->id == objectid)
 193                         break;
 194
 195         lockdep_set_class_and_name(&eb->lock,
 196                                    &ks->keys[level], ks->names[level]);
 197 }
 198
 199 #endif
 200
 201 /*
 202  * extents on the btree inode are pretty simple, there's one extent
 203  * that covers the entire device
 204  */
 205 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 206                 struct page *page, size_t pg_offset, u64 start, u64 len,
 207                 int create)
 208 {
 209         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 210         struct extent_map_tree *em_tree = &inode->extent_tree;
 211         struct extent_map *em;
 212         int ret;
 213
 214         read_lock(&em_tree->lock);
 215         em = lookup_extent_mapping(em_tree, start, len);
 216         if (em) {
 217                 em->bdev = fs_info->fs_devices->latest_bdev;
 218                 read_unlock(&em_tree->lock);
 219                 goto out;
 220         }
 221         read_unlock(&em_tree->lock);
 222
 223         em = alloc_extent_map();
 224         if (!em) {
 225                 em = ERR_PTR(-ENOMEM);
 226                 goto out;
 227         }
 228         em->start = 0;
 229         em->len = (u64)-1;
 230         em->block_len = (u64)-1;
 231         em->block_start = 0;
 232         em->bdev = fs_info->fs_devices->latest_bdev;
 233
 234         write_lock(&em_tree->lock);
 235         ret = add_extent_mapping(em_tree, em, 0);
 236         if (ret == -EEXIST) {
 237                 free_extent_map(em);
 238                 em = lookup_extent_mapping(em_tree, start, len);
 239                 if (!em)
 240                         em = ERR_PTR(-EIO);
 241         } else if (ret) {
 242                 free_extent_map(em);
 243                 em = ERR_PTR(ret);
 244         }
 245         write_unlock(&em_tree->lock);
 246
 247 out:
 248         return em;
 249 }
 250
 251 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 252 {
 253         return crc32c(seed, data, len);
 254 }
 255
 256 void btrfs_csum_final(u32 crc, u8 *result)
 257 {
 258         put_unaligned_le32(~crc, result);
 259 }
 260
 261 /*
 262  * compute the csum for a btree block, and either verify it or write it
 263  * into the csum field of the block.
 264  */
 265 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 266                            struct extent_buffer *buf,
 267                            int verify)
 268 {
 269         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 270         char result[BTRFS_CSUM_SIZE];
 271         unsigned long len;
 272         unsigned long cur_len;
 273         unsigned long offset = BTRFS_CSUM_SIZE;
 274         char *kaddr;
 275         unsigned long map_start;
 276         unsigned long map_len;
 277         int err;
 278         u32 crc = ~(u32)0;
 279
 280         len = buf->len - offset;
 281         while (len > 0) {
 282                 /*
 283                  * Note: we don't need to check for the err == 1 case here, as
 284                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 285                  * and 'min_len = 32' and the currently implemented mapping
 286                  * algorithm we cannot cross a page boundary.
 287                  */
 288                 err = map_private_extent_buffer(buf, offset, 32,
 289                                         &kaddr, &map_start, &map_len);
 290                 if (err)
 291                         return err;
 292                 cur_len = min(len, map_len - (offset - map_start));
 293                 crc = btrfs_csum_data(kaddr + offset - map_start,
 294                                       crc, cur_len);
 295                 len -= cur_len;
 296                 offset += cur_len;
 297         }
 298         memset(result, 0, BTRFS_CSUM_SIZE);
 299
 300         btrfs_csum_final(crc, result);
 301
 302         if (verify) {
 303                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 304                         u32 val;
 305                         u32 found = 0;
 306                         memcpy(&found, result, csum_size);
 307
 308                         read_extent_buffer(buf, &val, 0, csum_size);
 309                         btrfs_warn_rl(fs_info,
 310                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 311                                 fs_info->sb->s_id, buf->start,
 312                                 val, found, btrfs_header_level(buf));
 313                         return -EUCLEAN;
 314                 }
 315         } else {
 316                 write_extent_buffer(buf, result, 0, csum_size);
 317         }
 318
 319         return 0;
 320 }
 321
 322 /*
 323  * we can't consider a given block up to date unless the transid of the
 324  * block matches the transid in the parent node's pointer.  This is how we
 325  * detect blocks that either didn't get written at all or got written
 326  * in the wrong place.
 327  */
 328 static int verify_parent_transid(struct extent_io_tree *io_tree,
 329                                  struct extent_buffer *eb, u64 parent_transid,
 330                                  int atomic)
 331 {
 332         struct extent_state *cached_state = NULL;
 333         int ret;
 334         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 335
 336         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 337                 return 0;
 338
 339         if (atomic)
 340                 return -EAGAIN;
 341
 342         if (need_lock) {
 343                 btrfs_tree_read_lock(eb);
 344                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 345         }
 346
 347         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 348                          &cached_state);
 349         if (extent_buffer_uptodate(eb) &&
 350             btrfs_header_generation(eb) == parent_transid) {
 351                 ret = 0;
 352                 goto out;
 353         }
 354         btrfs_err_rl(eb->fs_info,
 355                 "parent transid verify failed on %llu wanted %llu found %llu",
 356                         eb->start,
 357                         parent_transid, btrfs_header_generation(eb));
 358         ret = 1;
 359
 360         /*
 361          * Things reading via commit roots that don't have normal protection,
 362          * like send, can have a really old block in cache that may point at a
 363          * block that has been freed and re-allocated.  So don't clear uptodate
 364          * if we find an eb that is under IO (dirty/writeback) because we could
 365          * end up reading in the stale data and then writing it back out and
 366          * making everybody very sad.
 367          */
 368         if (!extent_buffer_under_io(eb))
 369                 clear_extent_buffer_uptodate(eb);
 370 out:
 371         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 372                              &cached_state);
 373         if (need_lock)
 374                 btrfs_tree_read_unlock_blocking(eb);
 375         return ret;
 376 }
 377
 378 /*
 379  * Return 0 if the superblock checksum type matches the checksum value of that
 380  * algorithm. Pass the raw disk superblock data.
 381  */
 382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 383                                   char *raw_disk_sb)
 384 {
 385         struct btrfs_super_block *disk_sb =
 386                 (struct btrfs_super_block *)raw_disk_sb;
 387         u16 csum_type = btrfs_super_csum_type(disk_sb);
 388         int ret = 0;
 389
 390         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 391                 u32 crc = ~(u32)0;
 392                 char result[sizeof(crc)];
 393
 394                 /*
 395                  * The super_block structure does not span the whole
 396                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 397                  * is filled with zeros and is included in the checksum.
 398                  */
 399                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 400                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 401                 btrfs_csum_final(crc, result);
 402
 403                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 404                         ret = 1;
 405         }
 406
 407         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 408                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 409                                 csum_type);
 410                 ret = 1;
 411         }
 412
 413         return ret;
 414 }
 415
 416 static int verify_level_key(struct btrfs_fs_info *fs_info,
 417                             struct extent_buffer *eb, int level,
 418                             struct btrfs_key *first_key, u64 parent_transid)
 419 {
 420         int found_level;
 421         struct btrfs_key found_key;
 422         int ret;
 423
 424         found_level = btrfs_header_level(eb);
 425         if (found_level != level) {
 426 #ifdef CONFIG_BTRFS_DEBUG
 427                 WARN_ON(1);
 428                 btrfs_err(fs_info,
 429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 430                           eb->start, level, found_level);
 431 #endif
 432                 return -EIO;
 433         }
 434
 435         if (!first_key)
 436                 return 0;
 437
 438         /*
 439          * For live tree block (new tree blocks in current transaction),
 440          * we need proper lock context to avoid race, which is impossible here.
 441          * So we only checks tree blocks which is read from disk, whose
 442          * generation <= fs_info->last_trans_committed.
 443          */
 444         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 445                 return 0;
 446         if (found_level)
 447                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 448         else
 449                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 450         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 451
 452 #ifdef CONFIG_BTRFS_DEBUG
 453         if (ret) {
 454                 WARN_ON(1);
 455                 btrfs_err(fs_info,
 456 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 457                           eb->start, parent_transid, first_key->objectid,
 458                           first_key->type, first_key->offset,
 459                           found_key.objectid, found_key.type,
 460                           found_key.offset);
 461         }
 462 #endif
 463         return ret;
 464 }
 465
 466 /*
 467  * helper to read a given tree block, doing retries as required when
 468  * the checksums don't match and we have alternate mirrors to try.
 469  *
 470  * @parent_transid:     expected transid, skip check if 0
 471  * @level:              expected level, mandatory check
 472  * @first_key:          expected key of first slot, skip check if NULL
 473  */
 474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 475                                           struct extent_buffer *eb,
 476                                           u64 parent_transid, int level,
 477                                           struct btrfs_key *first_key)
 478 {
 479         struct extent_io_tree *io_tree;
 480         int failed = 0;
 481         int ret;
 482         int num_copies = 0;
 483         int mirror_num = 0;
 484         int failed_mirror = 0;
 485
 486         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 487         while (1) {
 488                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 489                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 490                                                mirror_num);
 491                 if (!ret) {
 492                         if (verify_parent_transid(io_tree, eb,
 493                                                    parent_transid, 0))
 494                                 ret = -EIO;
 495                         else if (verify_level_key(fs_info, eb, level,
 496                                                   first_key, parent_transid))
 497                                 ret = -EUCLEAN;
 498                         else
 499                                 break;
 500                 }
 501
 502                 num_copies = btrfs_num_copies(fs_info,
 503                                               eb->start, eb->len);
 504                 if (num_copies == 1)
 505                         break;
 506
 507                 if (!failed_mirror) {
 508                         failed = 1;
 509                         failed_mirror = eb->read_mirror;
 510                 }
 511
 512                 mirror_num++;
 513                 if (mirror_num == failed_mirror)
 514                         mirror_num++;
 515
 516                 if (mirror_num > num_copies)
 517                         break;
 518         }
 519
 520         if (failed && !ret && failed_mirror)
 521                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 522
 523         return ret;
 524 }
 525
 526 /*
 527  * checksum a dirty tree block before IO.  This has extra checks to make sure
 528  * we only fill in the checksum field in the first page of a multi-page block
 529  */
 530
 531 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 532 {
 533         u64 start = page_offset(page);
 534         u64 found_start;
 535         struct extent_buffer *eb;
 536
 537         eb = (struct extent_buffer *)page->private;
 538         if (page != eb->pages[0])
 539                 return 0;
 540
 541         found_start = btrfs_header_bytenr(eb);
 542         /*
 543          * Please do not consolidate these warnings into a single if.
 544          * It is useful to know what went wrong.
 545          */
 546         if (WARN_ON(found_start != start))
 547                 return -EUCLEAN;
 548         if (WARN_ON(!PageUptodate(page)))
 549                 return -EUCLEAN;
 550
 551         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 552                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 553
 554         return csum_tree_block(fs_info, eb, 0);
 555 }
 556
 557 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 558                                  struct extent_buffer *eb)
 559 {
 560         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 561         u8 fsid[BTRFS_FSID_SIZE];
 562         int ret = 1;
 563
 564         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 565         while (fs_devices) {
 566                 u8 *metadata_uuid;
 567
 568                 /*
 569                  * Checking the incompat flag is only valid for the current
 570                  * fs. For seed devices it's forbidden to have their uuid
 571                  * changed so reading ->fsid in this case is fine
 572                  */
 573                 if (fs_devices == fs_info->fs_devices &&
 574                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 575                         metadata_uuid = fs_devices->metadata_uuid;
 576                 else
 577                         metadata_uuid = fs_devices->fsid;
 578
 579                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 580                         ret = 0;
 581                         break;
 582                 }
 583                 fs_devices = fs_devices->seed;
 584         }
 585         return ret;
 586 }
 587
 588 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 589                                       u64 phy_offset, struct page *page,
 590                                       u64 start, u64 end, int mirror)
 591 {
 592         u64 found_start;
 593         int found_level;
 594         struct extent_buffer *eb;
 595         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 596         struct btrfs_fs_info *fs_info = root->fs_info;
 597         int ret = 0;
 598         int reads_done;
 599
 600         if (!page->private)
 601                 goto out;
 602
 603         eb = (struct extent_buffer *)page->private;
 604
 605         /* the pending IO might have been the only thing that kept this buffer
 606          * in memory.  Make sure we have a ref for all this other checks
 607          */
 608         extent_buffer_get(eb);
 609
 610         reads_done = atomic_dec_and_test(&eb->io_pages);
 611         if (!reads_done)
 612                 goto err;
 613
 614         eb->read_mirror = mirror;
 615         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 616                 ret = -EIO;
 617                 goto err;
 618         }
 619
 620         found_start = btrfs_header_bytenr(eb);
 621         if (found_start != eb->start) {
 622                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 623                              eb->start, found_start);
 624                 ret = -EIO;
 625                 goto err;
 626         }
 627         if (check_tree_block_fsid(fs_info, eb)) {
 628                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 629                              eb->start);
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633         found_level = btrfs_header_level(eb);
 634         if (found_level >= BTRFS_MAX_LEVEL) {
 635                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 636                           (int)btrfs_header_level(eb), eb->start);
 637                 ret = -EIO;
 638                 goto err;
 639         }
 640
 641         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 642                                        eb, found_level);
 643
 644         ret = csum_tree_block(fs_info, eb, 1);
 645         if (ret)
 646                 goto err;
 647
 648         /*
 649          * If this is a leaf block and it is corrupt, set the corrupt bit so
 650          * that we don't try and read the other copies of this block, just
 651          * return -EIO.
 652          */
 653         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 654                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 655                 ret = -EIO;
 656         }
 657
 658         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 659                 ret = -EIO;
 660
 661         if (!ret)
 662                 set_extent_buffer_uptodate(eb);
 663 err:
 664         if (reads_done &&
 665             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 666                 btree_readahead_hook(eb, ret);
 667
 668         if (ret) {
 669                 /*
 670                  * our io error hook is going to dec the io pages
 671                  * again, we have to make sure it has something
 672                  * to decrement
 673                  */
 674                 atomic_inc(&eb->io_pages);
 675                 clear_extent_buffer_uptodate(eb);
 676         }
 677         free_extent_buffer(eb);
 678 out:
 679         return ret;
 680 }
 681
 682 static void end_workqueue_bio(struct bio *bio)
 683 {
 684         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 685         struct btrfs_fs_info *fs_info;
 686         struct btrfs_workqueue *wq;
 687         btrfs_work_func_t func;
 688
 689         fs_info = end_io_wq->info;
 690         end_io_wq->status = bio->bi_status;
 691
 692         if (bio_op(bio) == REQ_OP_WRITE) {
 693                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 694                         wq = fs_info->endio_meta_write_workers;
 695                         func = btrfs_endio_meta_write_helper;
 696                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 697                         wq = fs_info->endio_freespace_worker;
 698                         func = btrfs_freespace_write_helper;
 699                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 700                         wq = fs_info->endio_raid56_workers;
 701                         func = btrfs_endio_raid56_helper;
 702                 } else {
 703                         wq = fs_info->endio_write_workers;
 704                         func = btrfs_endio_write_helper;
 705                 }
 706         } else {
 707                 if (unlikely(end_io_wq->metadata ==
 708                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 709                         wq = fs_info->endio_repair_workers;
 710                         func = btrfs_endio_repair_helper;
 711                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 712                         wq = fs_info->endio_raid56_workers;
 713                         func = btrfs_endio_raid56_helper;
 714                 } else if (end_io_wq->metadata) {
 715                         wq = fs_info->endio_meta_workers;
 716                         func = btrfs_endio_meta_helper;
 717                 } else {
 718                         wq = fs_info->endio_workers;
 719                         func = btrfs_endio_helper;
 720                 }
 721         }
 722
 723         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 724         btrfs_queue_work(wq, &end_io_wq->work);
 725 }
 726
 727 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 728                         enum btrfs_wq_endio_type metadata)
 729 {
 730         struct btrfs_end_io_wq *end_io_wq;
 731
 732         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 733         if (!end_io_wq)
 734                 return BLK_STS_RESOURCE;
 735
 736         end_io_wq->private = bio->bi_private;
 737         end_io_wq->end_io = bio->bi_end_io;
 738         end_io_wq->info = info;
 739         end_io_wq->status = 0;
 740         end_io_wq->bio = bio;
 741         end_io_wq->metadata = metadata;
 742
 743         bio->bi_private = end_io_wq;
 744         bio->bi_end_io = end_workqueue_bio;
 745         return 0;
 746 }
 747
 748 static void run_one_async_start(struct btrfs_work *work)
 749 {
 750         struct async_submit_bio *async;
 751         blk_status_t ret;
 752
 753         async = container_of(work, struct  async_submit_bio, work);
 754         ret = async->submit_bio_start(async->private_data, async->bio,
 755                                       async->bio_offset);
 756         if (ret)
 757                 async->status = ret;
 758 }
 759
 760 /*
 761  * In order to insert checksums into the metadata in large chunks, we wait
 762  * until bio submission time.   All the pages in the bio are checksummed and
 763  * sums are attached onto the ordered extent record.
 764  *
 765  * At IO completion time the csums attached on the ordered extent record are
 766  * inserted into the tree.
 767  */
 768 static void run_one_async_done(struct btrfs_work *work)
 769 {
 770         struct async_submit_bio *async;
 771         struct inode *inode;
 772         blk_status_t ret;
 773
 774         async = container_of(work, struct  async_submit_bio, work);
 775         inode = async->private_data;
 776
 777         /* If an error occurred we just want to clean up the bio and move on */
 778         if (async->status) {
 779                 async->bio->bi_status = async->status;
 780                 bio_endio(async->bio);
 781                 return;
 782         }
 783
 784         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 785                         async->mirror_num, 1);
 786         if (ret) {
 787                 async->bio->bi_status = ret;
 788                 bio_endio(async->bio);
 789         }
 790 }
 791
 792 static void run_one_async_free(struct btrfs_work *work)
 793 {
 794         struct async_submit_bio *async;
 795
 796         async = container_of(work, struct  async_submit_bio, work);
 797         kfree(async);
 798 }
 799
 800 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 801                                  int mirror_num, unsigned long bio_flags,
 802                                  u64 bio_offset, void *private_data,
 803                                  extent_submit_bio_start_t *submit_bio_start)
 804 {
 805         struct async_submit_bio *async;
 806
 807         async = kmalloc(sizeof(*async), GFP_NOFS);
 808         if (!async)
 809                 return BLK_STS_RESOURCE;
 810
 811         async->private_data = private_data;
 812         async->bio = bio;
 813         async->mirror_num = mirror_num;
 814         async->submit_bio_start = submit_bio_start;
 815
 816         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 817                         run_one_async_done, run_one_async_free);
 818
 819         async->bio_offset = bio_offset;
 820
 821         async->status = 0;
 822
 823         if (op_is_sync(bio->bi_opf))
 824                 btrfs_set_work_high_priority(&async->work);
 825
 826         btrfs_queue_work(fs_info->workers, &async->work);
 827         return 0;
 828 }
 829
 830 static blk_status_t btree_csum_one_bio(struct bio *bio)
 831 {
 832         struct bio_vec *bvec;
 833         struct btrfs_root *root;
 834         int i, ret = 0;
 835
 836         ASSERT(!bio_flagged(bio, BIO_CLONED));
 837         bio_for_each_segment_all(bvec, bio, i) {
 838                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 839                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 840                 if (ret)
 841                         break;
 842         }
 843
 844         return errno_to_blk_status(ret);
 845 }
 846
 847 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 848                                              u64 bio_offset)
 849 {
 850         /*
 851          * when we're called for a write, we're already in the async
 852          * submission context.  Just jump into btrfs_map_bio
 853          */
 854         return btree_csum_one_bio(bio);
 855 }
 856
 857 static int check_async_write(struct btrfs_inode *bi)
 858 {
 859         if (atomic_read(&bi->sync_writers))
 860                 return 0;
 861 #ifdef CONFIG_X86
 862         if (static_cpu_has(X86_FEATURE_XMM4_2))
 863                 return 0;
 864 #endif
 865         return 1;
 866 }
 867
 868 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 869                                           int mirror_num, unsigned long bio_flags,
 870                                           u64 bio_offset)
 871 {
 872         struct inode *inode = private_data;
 873         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 874         int async = check_async_write(BTRFS_I(inode));
 875         blk_status_t ret;
 876
 877         if (bio_op(bio) != REQ_OP_WRITE) {
 878                 /*
 879                  * called for a read, do the setup so that checksum validation
 880                  * can happen in the async kernel threads
 881                  */
 882                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 883                                           BTRFS_WQ_ENDIO_METADATA);
 884                 if (ret)
 885                         goto out_w_error;
 886                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 887         } else if (!async) {
 888                 ret = btree_csum_one_bio(bio);
 889                 if (ret)
 890                         goto out_w_error;
 891                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 892         } else {
 893                 /*
 894                  * kthread helpers are used to submit writes so that
 895                  * checksumming can happen in parallel across all CPUs
 896                  */
 897                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 898                                           bio_offset, private_data,
 899                                           btree_submit_bio_start);
 900         }
 901
 902         if (ret)
 903                 goto out_w_error;
 904         return 0;
 905
 906 out_w_error:
 907         bio->bi_status = ret;
 908         bio_endio(bio);
 909         return ret;
 910 }
 911
 912 #ifdef CONFIG_MIGRATION
 913 static int btree_migratepage(struct address_space *mapping,
 914                         struct page *newpage, struct page *page,
 915                         enum migrate_mode mode)
 916 {
 917         /*
 918          * we can't safely write a btree page from here,
 919          * we haven't done the locking hook
 920          */
 921         if (PageDirty(page))
 922                 return -EAGAIN;
 923         /*
 924          * Buffers may be managed in a filesystem specific way.
 925          * We must have no buffers or drop them.
 926          */
 927         if (page_has_private(page) &&
 928             !try_to_release_page(page, GFP_KERNEL))
 929                 return -EAGAIN;
 930         return migrate_page(mapping, newpage, page, mode);
 931 }
 932 #endif
 933
 934
 935 static int btree_writepages(struct address_space *mapping,
 936                             struct writeback_control *wbc)
 937 {
 938         struct btrfs_fs_info *fs_info;
 939         int ret;
 940
 941         if (wbc->sync_mode == WB_SYNC_NONE) {
 942
 943                 if (wbc->for_kupdate)
 944                         return 0;
 945
 946                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 947                 /* this is a bit racy, but that's ok */
 948                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 949                                              BTRFS_DIRTY_METADATA_THRESH,
 950                                              fs_info->dirty_metadata_batch);
 951                 if (ret < 0)
 952                         return 0;
 953         }
 954         return btree_write_cache_pages(mapping, wbc);
 955 }
 956
 957 static int btree_readpage(struct file *file, struct page *page)
 958 {
 959         struct extent_io_tree *tree;
 960         tree = &BTRFS_I(page->mapping->host)->io_tree;
 961         return extent_read_full_page(tree, page, btree_get_extent, 0);
 962 }
 963
 964 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 965 {
 966         if (PageWriteback(page) || PageDirty(page))
 967                 return 0;
 968
 969         return try_release_extent_buffer(page);
 970 }
 971
 972 static void btree_invalidatepage(struct page *page, unsigned int offset,
 973                                  unsigned int length)
 974 {
 975         struct extent_io_tree *tree;
 976         tree = &BTRFS_I(page->mapping->host)->io_tree;
 977         extent_invalidatepage(tree, page, offset);
 978         btree_releasepage(page, GFP_NOFS);
 979         if (PagePrivate(page)) {
 980                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 981                            "page private not zero on page %llu",
 982                            (unsigned long long)page_offset(page));
 983                 ClearPagePrivate(page);
 984                 set_page_private(page, 0);
 985                 put_page(page);
 986         }
 987 }
 988
 989 static int btree_set_page_dirty(struct page *page)
 990 {
 991 #ifdef DEBUG
 992         struct extent_buffer *eb;
 993
 994         BUG_ON(!PagePrivate(page));
 995         eb = (struct extent_buffer *)page->private;
 996         BUG_ON(!eb);
 997         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 998         BUG_ON(!atomic_read(&eb->refs));
 999         btrfs_assert_tree_locked(eb);
1000 #endif
1001         return __set_page_dirty_nobuffers(page);
1002 }
1003
1004 static const struct address_space_operations btree_aops = {
1005         .readpage       = btree_readpage,
1006         .writepages     = btree_writepages,
1007         .releasepage    = btree_releasepage,
1008         .invalidatepage = btree_invalidatepage,
1009 #ifdef CONFIG_MIGRATION
1010         .migratepage    = btree_migratepage,
1011 #endif
1012         .set_page_dirty = btree_set_page_dirty,
1013 };
1014
1015 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1016 {
1017         struct extent_buffer *buf = NULL;
1018         struct inode *btree_inode = fs_info->btree_inode;
1019
1020         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1021         if (IS_ERR(buf))
1022                 return;
1023         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1024                                  buf, WAIT_NONE, 0);
1025         free_extent_buffer(buf);
1026 }
1027
1028 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1029                          int mirror_num, struct extent_buffer **eb)
1030 {
1031         struct extent_buffer *buf = NULL;
1032         struct inode *btree_inode = fs_info->btree_inode;
1033         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1034         int ret;
1035
1036         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1037         if (IS_ERR(buf))
1038                 return 0;
1039
1040         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1041
1042         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1043                                        mirror_num);
1044         if (ret) {
1045                 free_extent_buffer(buf);
1046                 return ret;
1047         }
1048
1049         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1050                 free_extent_buffer(buf);
1051                 return -EIO;
1052         } else if (extent_buffer_uptodate(buf)) {
1053                 *eb = buf;
1054         } else {
1055                 free_extent_buffer(buf);
1056         }
1057         return 0;
1058 }
1059
1060 struct extent_buffer *btrfs_find_create_tree_block(
1061                                                 struct btrfs_fs_info *fs_info,
1062                                                 u64 bytenr)
1063 {
1064         if (btrfs_is_testing(fs_info))
1065                 return alloc_test_extent_buffer(fs_info, bytenr);
1066         return alloc_extent_buffer(fs_info, bytenr);
1067 }
1068
1069
1070 int btrfs_write_tree_block(struct extent_buffer *buf)
1071 {
1072         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1073                                         buf->start + buf->len - 1);
1074 }
1075
1076 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1077 {
1078         filemap_fdatawait_range(buf->pages[0]->mapping,
1079                                 buf->start, buf->start + buf->len - 1);
1080 }
1081
1082 /*
1083  * Read tree block at logical address @bytenr and do variant basic but critical
1084  * verification.
1085  *
1086  * @parent_transid:     expected transid of this tree block, skip check if 0
1087  * @level:              expected level, mandatory check
1088  * @first_key:          expected key in slot 0, skip check if NULL
1089  */
1090 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1091                                       u64 parent_transid, int level,
1092                                       struct btrfs_key *first_key)
1093 {
1094         struct extent_buffer *buf = NULL;
1095         int ret;
1096
1097         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1098         if (IS_ERR(buf))
1099                 return buf;
1100
1101         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1102                                              level, first_key);
1103         if (ret) {
1104                 free_extent_buffer(buf);
1105                 return ERR_PTR(ret);
1106         }
1107         return buf;
1108
1109 }
1110
1111 void clean_tree_block(struct btrfs_fs_info *fs_info,
1112                       struct extent_buffer *buf)
1113 {
1114         if (btrfs_header_generation(buf) ==
1115             fs_info->running_transaction->transid) {
1116                 btrfs_assert_tree_locked(buf);
1117
1118                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1119                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1120                                                  -buf->len,
1121                                                  fs_info->dirty_metadata_batch);
1122                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1123                         btrfs_set_lock_blocking(buf);
1124                         clear_extent_buffer_dirty(buf);
1125                 }
1126         }
1127 }
1128
1129 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1130 {
1131         struct btrfs_subvolume_writers *writers;
1132         int ret;
1133
1134         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1135         if (!writers)
1136                 return ERR_PTR(-ENOMEM);
1137
1138         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1139         if (ret < 0) {
1140                 kfree(writers);
1141                 return ERR_PTR(ret);
1142         }
1143
1144         init_waitqueue_head(&writers->wait);
1145         return writers;
1146 }
1147
1148 static void
1149 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1150 {
1151         percpu_counter_destroy(&writers->counter);
1152         kfree(writers);
1153 }
1154
1155 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1156                          u64 objectid)
1157 {
1158         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1159         root->node = NULL;
1160         root->commit_root = NULL;
1161         root->state = 0;
1162         root->orphan_cleanup_state = 0;
1163
1164         root->last_trans = 0;
1165         root->highest_objectid = 0;
1166         root->nr_delalloc_inodes = 0;
1167         root->nr_ordered_extents = 0;
1168         root->inode_tree = RB_ROOT;
1169         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1170         root->block_rsv = NULL;
1171
1172         INIT_LIST_HEAD(&root->dirty_list);
1173         INIT_LIST_HEAD(&root->root_list);
1174         INIT_LIST_HEAD(&root->delalloc_inodes);
1175         INIT_LIST_HEAD(&root->delalloc_root);
1176         INIT_LIST_HEAD(&root->ordered_extents);
1177         INIT_LIST_HEAD(&root->ordered_root);
1178         INIT_LIST_HEAD(&root->logged_list[0]);
1179         INIT_LIST_HEAD(&root->logged_list[1]);
1180         spin_lock_init(&root->inode_lock);
1181         spin_lock_init(&root->delalloc_lock);
1182         spin_lock_init(&root->ordered_extent_lock);
1183         spin_lock_init(&root->accounting_lock);
1184         spin_lock_init(&root->log_extents_lock[0]);
1185         spin_lock_init(&root->log_extents_lock[1]);
1186         spin_lock_init(&root->qgroup_meta_rsv_lock);
1187         mutex_init(&root->objectid_mutex);
1188         mutex_init(&root->log_mutex);
1189         mutex_init(&root->ordered_extent_mutex);
1190         mutex_init(&root->delalloc_mutex);
1191         init_waitqueue_head(&root->log_writer_wait);
1192         init_waitqueue_head(&root->log_commit_wait[0]);
1193         init_waitqueue_head(&root->log_commit_wait[1]);
1194         INIT_LIST_HEAD(&root->log_ctxs[0]);
1195         INIT_LIST_HEAD(&root->log_ctxs[1]);
1196         atomic_set(&root->log_commit[0], 0);
1197         atomic_set(&root->log_commit[1], 0);
1198         atomic_set(&root->log_writers, 0);
1199         atomic_set(&root->log_batch, 0);
1200         refcount_set(&root->refs, 1);
1201         atomic_set(&root->will_be_snapshotted, 0);
1202         atomic_set(&root->snapshot_force_cow, 0);
1203         atomic_set(&root->nr_swapfiles, 0);
1204         root->log_transid = 0;
1205         root->log_transid_committed = -1;
1206         root->last_log_commit = 0;
1207         if (!dummy)
1208                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1209
1210         memset(&root->root_key, 0, sizeof(root->root_key));
1211         memset(&root->root_item, 0, sizeof(root->root_item));
1212         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1213         if (!dummy)
1214                 root->defrag_trans_start = fs_info->generation;
1215         else
1216                 root->defrag_trans_start = 0;
1217         root->root_key.objectid = objectid;
1218         root->anon_dev = 0;
1219
1220         spin_lock_init(&root->root_item_lock);
1221 }
1222
1223 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1224                 gfp_t flags)
1225 {
1226         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1227         if (root)
1228                 root->fs_info = fs_info;
1229         return root;
1230 }
1231
1232 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1233 /* Should only be used by the testing infrastructure */
1234 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1235 {
1236         struct btrfs_root *root;
1237
1238         if (!fs_info)
1239                 return ERR_PTR(-EINVAL);
1240
1241         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1242         if (!root)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         /* We don't use the stripesize in selftest, set it as sectorsize */
1246         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1247         root->alloc_bytenr = 0;
1248
1249         return root;
1250 }
1251 #endif
1252
1253 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1254                                      struct btrfs_fs_info *fs_info,
1255                                      u64 objectid)
1256 {
1257         struct extent_buffer *leaf;
1258         struct btrfs_root *tree_root = fs_info->tree_root;
1259         struct btrfs_root *root;
1260         struct btrfs_key key;
1261         int ret = 0;
1262         uuid_le uuid = NULL_UUID_LE;
1263
1264         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1265         if (!root)
1266                 return ERR_PTR(-ENOMEM);
1267
1268         __setup_root(root, fs_info, objectid);
1269         root->root_key.objectid = objectid;
1270         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1271         root->root_key.offset = 0;
1272
1273         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1274         if (IS_ERR(leaf)) {
1275                 ret = PTR_ERR(leaf);
1276                 leaf = NULL;
1277                 goto fail;
1278         }
1279
1280         root->node = leaf;
1281         btrfs_mark_buffer_dirty(leaf);
1282
1283         root->commit_root = btrfs_root_node(root);
1284         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1285
1286         root->root_item.flags = 0;
1287         root->root_item.byte_limit = 0;
1288         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1289         btrfs_set_root_generation(&root->root_item, trans->transid);
1290         btrfs_set_root_level(&root->root_item, 0);
1291         btrfs_set_root_refs(&root->root_item, 1);
1292         btrfs_set_root_used(&root->root_item, leaf->len);
1293         btrfs_set_root_last_snapshot(&root->root_item, 0);
1294         btrfs_set_root_dirid(&root->root_item, 0);
1295         if (is_fstree(objectid))
1296                 uuid_le_gen(&uuid);
1297         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1298         root->root_item.drop_level = 0;
1299
1300         key.objectid = objectid;
1301         key.type = BTRFS_ROOT_ITEM_KEY;
1302         key.offset = 0;
1303         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1304         if (ret)
1305                 goto fail;
1306
1307         btrfs_tree_unlock(leaf);
1308
1309         return root;
1310
1311 fail:
1312         if (leaf) {
1313                 btrfs_tree_unlock(leaf);
1314                 free_extent_buffer(root->commit_root);
1315                 free_extent_buffer(leaf);
1316         }
1317         kfree(root);
1318
1319         return ERR_PTR(ret);
1320 }
1321
1322 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1323                                          struct btrfs_fs_info *fs_info)
1324 {
1325         struct btrfs_root *root;
1326         struct extent_buffer *leaf;
1327
1328         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1329         if (!root)
1330                 return ERR_PTR(-ENOMEM);
1331
1332         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1333
1334         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1335         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1336         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1337
1338         /*
1339          * DON'T set REF_COWS for log trees
1340          *
1341          * log trees do not get reference counted because they go away
1342          * before a real commit is actually done.  They do store pointers
1343          * to file data extents, and those reference counts still get
1344          * updated (along with back refs to the log tree).
1345          */
1346
1347         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1348                         NULL, 0, 0, 0);
1349         if (IS_ERR(leaf)) {
1350                 kfree(root);
1351                 return ERR_CAST(leaf);
1352         }
1353
1354         root->node = leaf;
1355
1356         btrfs_mark_buffer_dirty(root->node);
1357         btrfs_tree_unlock(root->node);
1358         return root;
1359 }
1360
1361 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1362                              struct btrfs_fs_info *fs_info)
1363 {
1364         struct btrfs_root *log_root;
1365
1366         log_root = alloc_log_tree(trans, fs_info);
1367         if (IS_ERR(log_root))
1368                 return PTR_ERR(log_root);
1369         WARN_ON(fs_info->log_root_tree);
1370         fs_info->log_root_tree = log_root;
1371         return 0;
1372 }
1373
1374 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1375                        struct btrfs_root *root)
1376 {
1377         struct btrfs_fs_info *fs_info = root->fs_info;
1378         struct btrfs_root *log_root;
1379         struct btrfs_inode_item *inode_item;
1380
1381         log_root = alloc_log_tree(trans, fs_info);
1382         if (IS_ERR(log_root))
1383                 return PTR_ERR(log_root);
1384
1385         log_root->last_trans = trans->transid;
1386         log_root->root_key.offset = root->root_key.objectid;
1387
1388         inode_item = &log_root->root_item.inode;
1389         btrfs_set_stack_inode_generation(inode_item, 1);
1390         btrfs_set_stack_inode_size(inode_item, 3);
1391         btrfs_set_stack_inode_nlink(inode_item, 1);
1392         btrfs_set_stack_inode_nbytes(inode_item,
1393                                      fs_info->nodesize);
1394         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1395
1396         btrfs_set_root_node(&log_root->root_item, log_root->node);
1397
1398         WARN_ON(root->log_root);
1399         root->log_root = log_root;
1400         root->log_transid = 0;
1401         root->log_transid_committed = -1;
1402         root->last_log_commit = 0;
1403         return 0;
1404 }
1405
1406 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1407                                                struct btrfs_key *key)
1408 {
1409         struct btrfs_root *root;
1410         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1411         struct btrfs_path *path;
1412         u64 generation;
1413         int ret;
1414         int level;
1415
1416         path = btrfs_alloc_path();
1417         if (!path)
1418                 return ERR_PTR(-ENOMEM);
1419
1420         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1421         if (!root) {
1422                 ret = -ENOMEM;
1423                 goto alloc_fail;
1424         }
1425
1426         __setup_root(root, fs_info, key->objectid);
1427
1428         ret = btrfs_find_root(tree_root, key, path,
1429                               &root->root_item, &root->root_key);
1430         if (ret) {
1431                 if (ret > 0)
1432                         ret = -ENOENT;
1433                 goto find_fail;
1434         }
1435
1436         generation = btrfs_root_generation(&root->root_item);
1437         level = btrfs_root_level(&root->root_item);
1438         root->node = read_tree_block(fs_info,
1439                                      btrfs_root_bytenr(&root->root_item),
1440                                      generation, level, NULL);
1441         if (IS_ERR(root->node)) {
1442                 ret = PTR_ERR(root->node);
1443                 goto find_fail;
1444         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1445                 ret = -EIO;
1446                 free_extent_buffer(root->node);
1447                 goto find_fail;
1448         }
1449         root->commit_root = btrfs_root_node(root);
1450 out:
1451         btrfs_free_path(path);
1452         return root;
1453
1454 find_fail:
1455         kfree(root);
1456 alloc_fail:
1457         root = ERR_PTR(ret);
1458         goto out;
1459 }
1460
1461 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1462                                       struct btrfs_key *location)
1463 {
1464         struct btrfs_root *root;
1465
1466         root = btrfs_read_tree_root(tree_root, location);
1467         if (IS_ERR(root))
1468                 return root;
1469
1470         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1471                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1472                 btrfs_check_and_init_root_item(&root->root_item);
1473         }
1474
1475         return root;
1476 }
1477
1478 int btrfs_init_fs_root(struct btrfs_root *root)
1479 {
1480         int ret;
1481         struct btrfs_subvolume_writers *writers;
1482
1483         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1484         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1485                                         GFP_NOFS);
1486         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1487                 ret = -ENOMEM;
1488                 goto fail;
1489         }
1490
1491         writers = btrfs_alloc_subvolume_writers();
1492         if (IS_ERR(writers)) {
1493                 ret = PTR_ERR(writers);
1494                 goto fail;
1495         }
1496         root->subv_writers = writers;
1497
1498         btrfs_init_free_ino_ctl(root);
1499         spin_lock_init(&root->ino_cache_lock);
1500         init_waitqueue_head(&root->ino_cache_wait);
1501
1502         ret = get_anon_bdev(&root->anon_dev);
1503         if (ret)
1504                 goto fail;
1505
1506         mutex_lock(&root->objectid_mutex);
1507         ret = btrfs_find_highest_objectid(root,
1508                                         &root->highest_objectid);
1509         if (ret) {
1510                 mutex_unlock(&root->objectid_mutex);
1511                 goto fail;
1512         }
1513
1514         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1515
1516         mutex_unlock(&root->objectid_mutex);
1517
1518         return 0;
1519 fail:
1520         /* The caller is responsible to call btrfs_free_fs_root */
1521         return ret;
1522 }
1523
1524 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1525                                         u64 root_id)
1526 {
1527         struct btrfs_root *root;
1528
1529         spin_lock(&fs_info->fs_roots_radix_lock);
1530         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1531                                  (unsigned long)root_id);
1532         spin_unlock(&fs_info->fs_roots_radix_lock);
1533         return root;
1534 }
1535
1536 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1537                          struct btrfs_root *root)
1538 {
1539         int ret;
1540
1541         ret = radix_tree_preload(GFP_NOFS);
1542         if (ret)
1543                 return ret;
1544
1545         spin_lock(&fs_info->fs_roots_radix_lock);
1546         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1547                                 (unsigned long)root->root_key.objectid,
1548                                 root);
1549         if (ret == 0)
1550                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1551         spin_unlock(&fs_info->fs_roots_radix_lock);
1552         radix_tree_preload_end();
1553
1554         return ret;
1555 }
1556
1557 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1558                                      struct btrfs_key *location,
1559                                      bool check_ref)
1560 {
1561         struct btrfs_root *root;
1562         struct btrfs_path *path;
1563         struct btrfs_key key;
1564         int ret;
1565
1566         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1567                 return fs_info->tree_root;
1568         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1569                 return fs_info->extent_root;
1570         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1571                 return fs_info->chunk_root;
1572         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1573                 return fs_info->dev_root;
1574         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1575                 return fs_info->csum_root;
1576         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1577                 return fs_info->quota_root ? fs_info->quota_root :
1578                                              ERR_PTR(-ENOENT);
1579         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1580                 return fs_info->uuid_root ? fs_info->uuid_root :
1581                                             ERR_PTR(-ENOENT);
1582         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1583                 return fs_info->free_space_root ? fs_info->free_space_root :
1584                                                   ERR_PTR(-ENOENT);
1585 again:
1586         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1587         if (root) {
1588                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1589                         return ERR_PTR(-ENOENT);
1590                 return root;
1591         }
1592
1593         root = btrfs_read_fs_root(fs_info->tree_root, location);
1594         if (IS_ERR(root))
1595                 return root;
1596
1597         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1598                 ret = -ENOENT;
1599                 goto fail;
1600         }
1601
1602         ret = btrfs_init_fs_root(root);
1603         if (ret)
1604                 goto fail;
1605
1606         path = btrfs_alloc_path();
1607         if (!path) {
1608                 ret = -ENOMEM;
1609                 goto fail;
1610         }
1611         key.objectid = BTRFS_ORPHAN_OBJECTID;
1612         key.type = BTRFS_ORPHAN_ITEM_KEY;
1613         key.offset = location->objectid;
1614
1615         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1616         btrfs_free_path(path);
1617         if (ret < 0)
1618                 goto fail;
1619         if (ret == 0)
1620                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1621
1622         ret = btrfs_insert_fs_root(fs_info, root);
1623         if (ret) {
1624                 if (ret == -EEXIST) {
1625                         btrfs_free_fs_root(root);
1626                         goto again;
1627                 }
1628                 goto fail;
1629         }
1630         return root;
1631 fail:
1632         btrfs_free_fs_root(root);
1633         return ERR_PTR(ret);
1634 }
1635
1636 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1637 {
1638         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1639         int ret = 0;
1640         struct btrfs_device *device;
1641         struct backing_dev_info *bdi;
1642
1643         rcu_read_lock();
1644         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1645                 if (!device->bdev)
1646                         continue;
1647                 bdi = device->bdev->bd_bdi;
1648                 if (bdi_congested(bdi, bdi_bits)) {
1649                         ret = 1;
1650                         break;
1651                 }
1652         }
1653         rcu_read_unlock();
1654         return ret;
1655 }
1656
1657 /*
1658  * called by the kthread helper functions to finally call the bio end_io
1659  * functions.  This is where read checksum verification actually happens
1660  */
1661 static void end_workqueue_fn(struct btrfs_work *work)
1662 {
1663         struct bio *bio;
1664         struct btrfs_end_io_wq *end_io_wq;
1665
1666         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1667         bio = end_io_wq->bio;
1668
1669         bio->bi_status = end_io_wq->status;
1670         bio->bi_private = end_io_wq->private;
1671         bio->bi_end_io = end_io_wq->end_io;
1672         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1673         bio_endio(bio);
1674 }
1675
1676 static int cleaner_kthread(void *arg)
1677 {
1678         struct btrfs_root *root = arg;
1679         struct btrfs_fs_info *fs_info = root->fs_info;
1680         int again;
1681
1682         while (1) {
1683                 again = 0;
1684
1685                 /* Make the cleaner go to sleep early. */
1686                 if (btrfs_need_cleaner_sleep(fs_info))
1687                         goto sleep;
1688
1689                 /*
1690                  * Do not do anything if we might cause open_ctree() to block
1691                  * before we have finished mounting the filesystem.
1692                  */
1693                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1694                         goto sleep;
1695
1696                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1697                         goto sleep;
1698
1699                 /*
1700                  * Avoid the problem that we change the status of the fs
1701                  * during the above check and trylock.
1702                  */
1703                 if (btrfs_need_cleaner_sleep(fs_info)) {
1704                         mutex_unlock(&fs_info->cleaner_mutex);
1705                         goto sleep;
1706                 }
1707
1708                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1709                 btrfs_run_delayed_iputs(fs_info);
1710                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1711
1712                 again = btrfs_clean_one_deleted_snapshot(root);
1713                 mutex_unlock(&fs_info->cleaner_mutex);
1714
1715                 /*
1716                  * The defragger has dealt with the R/O remount and umount,
1717                  * needn't do anything special here.
1718                  */
1719                 btrfs_run_defrag_inodes(fs_info);
1720
1721                 /*
1722                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1723                  * with relocation (btrfs_relocate_chunk) and relocation
1724                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1725                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1726                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1727                  * unused block groups.
1728                  */
1729                 btrfs_delete_unused_bgs(fs_info);
1730 sleep:
1731                 if (kthread_should_park())
1732                         kthread_parkme();
1733                 if (kthread_should_stop())
1734                         return 0;
1735                 if (!again) {
1736                         set_current_state(TASK_INTERRUPTIBLE);
1737                         schedule();
1738                         __set_current_state(TASK_RUNNING);
1739                 }
1740         }
1741 }
1742
1743 static int transaction_kthread(void *arg)
1744 {
1745         struct btrfs_root *root = arg;
1746         struct btrfs_fs_info *fs_info = root->fs_info;
1747         struct btrfs_trans_handle *trans;
1748         struct btrfs_transaction *cur;
1749         u64 transid;
1750         time64_t now;
1751         unsigned long delay;
1752         bool cannot_commit;
1753
1754         do {
1755                 cannot_commit = false;
1756                 delay = HZ * fs_info->commit_interval;
1757                 mutex_lock(&fs_info->transaction_kthread_mutex);
1758
1759                 spin_lock(&fs_info->trans_lock);
1760                 cur = fs_info->running_transaction;
1761                 if (!cur) {
1762                         spin_unlock(&fs_info->trans_lock);
1763                         goto sleep;
1764                 }
1765
1766                 now = ktime_get_seconds();
1767                 if (cur->state < TRANS_STATE_BLOCKED &&
1768                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1769                     (now < cur->start_time ||
1770                      now - cur->start_time < fs_info->commit_interval)) {
1771                         spin_unlock(&fs_info->trans_lock);
1772                         delay = HZ * 5;
1773                         goto sleep;
1774                 }
1775                 transid = cur->transid;
1776                 spin_unlock(&fs_info->trans_lock);
1777
1778                 /* If the file system is aborted, this will always fail. */
1779                 trans = btrfs_attach_transaction(root);
1780                 if (IS_ERR(trans)) {
1781                         if (PTR_ERR(trans) != -ENOENT)
1782                                 cannot_commit = true;
1783                         goto sleep;
1784                 }
1785                 if (transid == trans->transid) {
1786                         btrfs_commit_transaction(trans);
1787                 } else {
1788                         btrfs_end_transaction(trans);
1789                 }
1790 sleep:
1791                 wake_up_process(fs_info->cleaner_kthread);
1792                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1793
1794                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1795                                       &fs_info->fs_state)))
1796                         btrfs_cleanup_transaction(fs_info);
1797                 if (!kthread_should_stop() &&
1798                                 (!btrfs_transaction_blocked(fs_info) ||
1799                                  cannot_commit))
1800                         schedule_timeout_interruptible(delay);
1801         } while (!kthread_should_stop());
1802         return 0;
1803 }
1804
1805 /*
1806  * this will find the highest generation in the array of
1807  * root backups.  The index of the highest array is returned,
1808  * or -1 if we can't find anything.
1809  *
1810  * We check to make sure the array is valid by comparing the
1811  * generation of the latest  root in the array with the generation
1812  * in the super block.  If they don't match we pitch it.
1813  */
1814 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1815 {
1816         u64 cur;
1817         int newest_index = -1;
1818         struct btrfs_root_backup *root_backup;
1819         int i;
1820
1821         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1822                 root_backup = info->super_copy->super_roots + i;
1823                 cur = btrfs_backup_tree_root_gen(root_backup);
1824                 if (cur == newest_gen)
1825                         newest_index = i;
1826         }
1827
1828         /* check to see if we actually wrapped around */
1829         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1830                 root_backup = info->super_copy->super_roots;
1831                 cur = btrfs_backup_tree_root_gen(root_backup);
1832                 if (cur == newest_gen)
1833                         newest_index = 0;
1834         }
1835         return newest_index;
1836 }
1837
1838
1839 /*
1840  * find the oldest backup so we know where to store new entries
1841  * in the backup array.  This will set the backup_root_index
1842  * field in the fs_info struct
1843  */
1844 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1845                                      u64 newest_gen)
1846 {
1847         int newest_index = -1;
1848
1849         newest_index = find_newest_super_backup(info, newest_gen);
1850         /* if there was garbage in there, just move along */
1851         if (newest_index == -1) {
1852                 info->backup_root_index = 0;
1853         } else {
1854                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1855         }
1856 }
1857
1858 /*
1859  * copy all the root pointers into the super backup array.
1860  * this will bump the backup pointer by one when it is
1861  * done
1862  */
1863 static void backup_super_roots(struct btrfs_fs_info *info)
1864 {
1865         int next_backup;
1866         struct btrfs_root_backup *root_backup;
1867         int last_backup;
1868
1869         next_backup = info->backup_root_index;
1870         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1871                 BTRFS_NUM_BACKUP_ROOTS;
1872
1873         /*
1874          * just overwrite the last backup if we're at the same generation
1875          * this happens only at umount
1876          */
1877         root_backup = info->super_for_commit->super_roots + last_backup;
1878         if (btrfs_backup_tree_root_gen(root_backup) ==
1879             btrfs_header_generation(info->tree_root->node))
1880                 next_backup = last_backup;
1881
1882         root_backup = info->super_for_commit->super_roots + next_backup;
1883
1884         /*
1885          * make sure all of our padding and empty slots get zero filled
1886          * regardless of which ones we use today
1887          */
1888         memset(root_backup, 0, sizeof(*root_backup));
1889
1890         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1891
1892         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1893         btrfs_set_backup_tree_root_gen(root_backup,
1894                                btrfs_header_generation(info->tree_root->node));
1895
1896         btrfs_set_backup_tree_root_level(root_backup,
1897                                btrfs_header_level(info->tree_root->node));
1898
1899         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1900         btrfs_set_backup_chunk_root_gen(root_backup,
1901                                btrfs_header_generation(info->chunk_root->node));
1902         btrfs_set_backup_chunk_root_level(root_backup,
1903                                btrfs_header_level(info->chunk_root->node));
1904
1905         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1906         btrfs_set_backup_extent_root_gen(root_backup,
1907                                btrfs_header_generation(info->extent_root->node));
1908         btrfs_set_backup_extent_root_level(root_backup,
1909                                btrfs_header_level(info->extent_root->node));
1910
1911         /*
1912          * we might commit during log recovery, which happens before we set
1913          * the fs_root.  Make sure it is valid before we fill it in.
1914          */
1915         if (info->fs_root && info->fs_root->node) {
1916                 btrfs_set_backup_fs_root(root_backup,
1917                                          info->fs_root->node->start);
1918                 btrfs_set_backup_fs_root_gen(root_backup,
1919                                btrfs_header_generation(info->fs_root->node));
1920                 btrfs_set_backup_fs_root_level(root_backup,
1921                                btrfs_header_level(info->fs_root->node));
1922         }
1923
1924         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1925         btrfs_set_backup_dev_root_gen(root_backup,
1926                                btrfs_header_generation(info->dev_root->node));
1927         btrfs_set_backup_dev_root_level(root_backup,
1928                                        btrfs_header_level(info->dev_root->node));
1929
1930         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1931         btrfs_set_backup_csum_root_gen(root_backup,
1932                                btrfs_header_generation(info->csum_root->node));
1933         btrfs_set_backup_csum_root_level(root_backup,
1934                                btrfs_header_level(info->csum_root->node));
1935
1936         btrfs_set_backup_total_bytes(root_backup,
1937                              btrfs_super_total_bytes(info->super_copy));
1938         btrfs_set_backup_bytes_used(root_backup,
1939                              btrfs_super_bytes_used(info->super_copy));
1940         btrfs_set_backup_num_devices(root_backup,
1941                              btrfs_super_num_devices(info->super_copy));
1942
1943         /*
1944          * if we don't copy this out to the super_copy, it won't get remembered
1945          * for the next commit
1946          */
1947         memcpy(&info->super_copy->super_roots,
1948                &info->super_for_commit->super_roots,
1949                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1950 }
1951
1952 /*
1953  * this copies info out of the root backup array and back into
1954  * the in-memory super block.  It is meant to help iterate through
1955  * the array, so you send it the number of backups you've already
1956  * tried and the last backup index you used.
1957  *
1958  * this returns -1 when it has tried all the backups
1959  */
1960 static noinline int next_root_backup(struct btrfs_fs_info *info,
1961                                      struct btrfs_super_block *super,
1962                                      int *num_backups_tried, int *backup_index)
1963 {
1964         struct btrfs_root_backup *root_backup;
1965         int newest = *backup_index;
1966
1967         if (*num_backups_tried == 0) {
1968                 u64 gen = btrfs_super_generation(super);
1969
1970                 newest = find_newest_super_backup(info, gen);
1971                 if (newest == -1)
1972                         return -1;
1973
1974                 *backup_index = newest;
1975                 *num_backups_tried = 1;
1976         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1977                 /* we've tried all the backups, all done */
1978                 return -1;
1979         } else {
1980                 /* jump to the next oldest backup */
1981                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1982                         BTRFS_NUM_BACKUP_ROOTS;
1983                 *backup_index = newest;
1984                 *num_backups_tried += 1;
1985         }
1986         root_backup = super->super_roots + newest;
1987
1988         btrfs_set_super_generation(super,
1989                                    btrfs_backup_tree_root_gen(root_backup));
1990         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1991         btrfs_set_super_root_level(super,
1992                                    btrfs_backup_tree_root_level(root_backup));
1993         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1994
1995         /*
1996          * fixme: the total bytes and num_devices need to match or we should
1997          * need a fsck
1998          */
1999         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2000         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2001         return 0;
2002 }
2003
2004 /* helper to cleanup workers */
2005 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2006 {
2007         btrfs_destroy_workqueue(fs_info->fixup_workers);
2008         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2009         btrfs_destroy_workqueue(fs_info->workers);
2010         btrfs_destroy_workqueue(fs_info->endio_workers);
2011         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2012         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2013         btrfs_destroy_workqueue(fs_info->rmw_workers);
2014         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2015         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2016         btrfs_destroy_workqueue(fs_info->submit_workers);
2017         btrfs_destroy_workqueue(fs_info->delayed_workers);
2018         btrfs_destroy_workqueue(fs_info->caching_workers);
2019         btrfs_destroy_workqueue(fs_info->readahead_workers);
2020         btrfs_destroy_workqueue(fs_info->flush_workers);
2021         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2022         btrfs_destroy_workqueue(fs_info->extent_workers);
2023         /*
2024          * Now that all other work queues are destroyed, we can safely destroy
2025          * the queues used for metadata I/O, since tasks from those other work
2026          * queues can do metadata I/O operations.
2027          */
2028         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2029         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2030 }
2031
2032 static void free_root_extent_buffers(struct btrfs_root *root)
2033 {
2034         if (root) {
2035                 free_extent_buffer(root->node);
2036                 free_extent_buffer(root->commit_root);
2037                 root->node = NULL;
2038                 root->commit_root = NULL;
2039         }
2040 }
2041
2042 /* helper to cleanup tree roots */
2043 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2044 {
2045         free_root_extent_buffers(info->tree_root);
2046
2047         free_root_extent_buffers(info->dev_root);
2048         free_root_extent_buffers(info->extent_root);
2049         free_root_extent_buffers(info->csum_root);
2050         free_root_extent_buffers(info->quota_root);
2051         free_root_extent_buffers(info->uuid_root);
2052         if (chunk_root)
2053                 free_root_extent_buffers(info->chunk_root);
2054         free_root_extent_buffers(info->free_space_root);
2055 }
2056
2057 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2058 {
2059         int ret;
2060         struct btrfs_root *gang[8];
2061         int i;
2062
2063         while (!list_empty(&fs_info->dead_roots)) {
2064                 gang[0] = list_entry(fs_info->dead_roots.next,
2065                                      struct btrfs_root, root_list);
2066                 list_del(&gang[0]->root_list);
2067
2068                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2069                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2070                 } else {
2071                         free_extent_buffer(gang[0]->node);
2072                         free_extent_buffer(gang[0]->commit_root);
2073                         btrfs_put_fs_root(gang[0]);
2074                 }
2075         }
2076
2077         while (1) {
2078                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2079                                              (void **)gang, 0,
2080                                              ARRAY_SIZE(gang));
2081                 if (!ret)
2082                         break;
2083                 for (i = 0; i < ret; i++)
2084                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2085         }
2086
2087         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2088                 btrfs_free_log_root_tree(NULL, fs_info);
2089                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2090         }
2091 }
2092
2093 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2094 {
2095         mutex_init(&fs_info->scrub_lock);
2096         atomic_set(&fs_info->scrubs_running, 0);
2097         atomic_set(&fs_info->scrub_pause_req, 0);
2098         atomic_set(&fs_info->scrubs_paused, 0);
2099         atomic_set(&fs_info->scrub_cancel_req, 0);
2100         init_waitqueue_head(&fs_info->scrub_pause_wait);
2101         fs_info->scrub_workers_refcnt = 0;
2102 }
2103
2104 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2105 {
2106         spin_lock_init(&fs_info->balance_lock);
2107         mutex_init(&fs_info->balance_mutex);
2108         atomic_set(&fs_info->balance_pause_req, 0);
2109         atomic_set(&fs_info->balance_cancel_req, 0);
2110         fs_info->balance_ctl = NULL;
2111         init_waitqueue_head(&fs_info->balance_wait_q);
2112 }
2113
2114 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2115 {
2116         struct inode *inode = fs_info->btree_inode;
2117
2118         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2119         set_nlink(inode, 1);
2120         /*
2121          * we set the i_size on the btree inode to the max possible int.
2122          * the real end of the address space is determined by all of
2123          * the devices in the system
2124          */
2125         inode->i_size = OFFSET_MAX;
2126         inode->i_mapping->a_ops = &btree_aops;
2127
2128         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2129         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2130         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2131         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2132
2133         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2134
2135         BTRFS_I(inode)->root = fs_info->tree_root;
2136         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2137         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2138         btrfs_insert_inode_hash(inode);
2139 }
2140
2141 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2142 {
2143         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2144         rwlock_init(&fs_info->dev_replace.lock);
2145         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2146         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2147         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2148 }
2149
2150 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2151 {
2152         spin_lock_init(&fs_info->qgroup_lock);
2153         mutex_init(&fs_info->qgroup_ioctl_lock);
2154         fs_info->qgroup_tree = RB_ROOT;
2155         fs_info->qgroup_op_tree = RB_ROOT;
2156         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2157         fs_info->qgroup_seq = 1;
2158         fs_info->qgroup_ulist = NULL;
2159         fs_info->qgroup_rescan_running = false;
2160         mutex_init(&fs_info->qgroup_rescan_lock);
2161 }
2162
2163 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2164                 struct btrfs_fs_devices *fs_devices)
2165 {
2166         u32 max_active = fs_info->thread_pool_size;
2167         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2168
2169         fs_info->workers =
2170                 btrfs_alloc_workqueue(fs_info, "worker",
2171                                       flags | WQ_HIGHPRI, max_active, 16);
2172
2173         fs_info->delalloc_workers =
2174                 btrfs_alloc_workqueue(fs_info, "delalloc",
2175                                       flags, max_active, 2);
2176
2177         fs_info->flush_workers =
2178                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2179                                       flags, max_active, 0);
2180
2181         fs_info->caching_workers =
2182                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2183
2184         /*
2185          * a higher idle thresh on the submit workers makes it much more
2186          * likely that bios will be send down in a sane order to the
2187          * devices
2188          */
2189         fs_info->submit_workers =
2190                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2191                                       min_t(u64, fs_devices->num_devices,
2192                                             max_active), 64);
2193
2194         fs_info->fixup_workers =
2195                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2196
2197         /*
2198          * endios are largely parallel and should have a very
2199          * low idle thresh
2200          */
2201         fs_info->endio_workers =
2202                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2203         fs_info->endio_meta_workers =
2204                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2205                                       max_active, 4);
2206         fs_info->endio_meta_write_workers =
2207                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2208                                       max_active, 2);
2209         fs_info->endio_raid56_workers =
2210                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2211                                       max_active, 4);
2212         fs_info->endio_repair_workers =
2213                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2214         fs_info->rmw_workers =
2215                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2216         fs_info->endio_write_workers =
2217                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2218                                       max_active, 2);
2219         fs_info->endio_freespace_worker =
2220                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2221                                       max_active, 0);
2222         fs_info->delayed_workers =
2223                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2224                                       max_active, 0);
2225         fs_info->readahead_workers =
2226                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2227                                       max_active, 2);
2228         fs_info->qgroup_rescan_workers =
2229                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2230         fs_info->extent_workers =
2231                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2232                                       min_t(u64, fs_devices->num_devices,
2233                                             max_active), 8);
2234
2235         if (!(fs_info->workers && fs_info->delalloc_workers &&
2236               fs_info->submit_workers && fs_info->flush_workers &&
2237               fs_info->endio_workers && fs_info->endio_meta_workers &&
2238               fs_info->endio_meta_write_workers &&
2239               fs_info->endio_repair_workers &&
2240               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2241               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2242               fs_info->caching_workers && fs_info->readahead_workers &&
2243               fs_info->fixup_workers && fs_info->delayed_workers &&
2244               fs_info->extent_workers &&
2245               fs_info->qgroup_rescan_workers)) {
2246                 return -ENOMEM;
2247         }
2248
2249         return 0;
2250 }
2251
2252 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2253                             struct btrfs_fs_devices *fs_devices)
2254 {
2255         int ret;
2256         struct btrfs_root *log_tree_root;
2257         struct btrfs_super_block *disk_super = fs_info->super_copy;
2258         u64 bytenr = btrfs_super_log_root(disk_super);
2259         int level = btrfs_super_log_root_level(disk_super);
2260
2261         if (fs_devices->rw_devices == 0) {
2262                 btrfs_warn(fs_info, "log replay required on RO media");
2263                 return -EIO;
2264         }
2265
2266         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2267         if (!log_tree_root)
2268                 return -ENOMEM;
2269
2270         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2271
2272         log_tree_root->node = read_tree_block(fs_info, bytenr,
2273                                               fs_info->generation + 1,
2274                                               level, NULL);
2275         if (IS_ERR(log_tree_root->node)) {
2276                 btrfs_warn(fs_info, "failed to read log tree");
2277                 ret = PTR_ERR(log_tree_root->node);
2278                 kfree(log_tree_root);
2279                 return ret;
2280         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2281                 btrfs_err(fs_info, "failed to read log tree");
2282                 free_extent_buffer(log_tree_root->node);
2283                 kfree(log_tree_root);
2284                 return -EIO;
2285         }
2286         /* returns with log_tree_root freed on success */
2287         ret = btrfs_recover_log_trees(log_tree_root);
2288         if (ret) {
2289                 btrfs_handle_fs_error(fs_info, ret,
2290                                       "Failed to recover log tree");
2291                 free_extent_buffer(log_tree_root->node);
2292                 kfree(log_tree_root);
2293                 return ret;
2294         }
2295
2296         if (sb_rdonly(fs_info->sb)) {
2297                 ret = btrfs_commit_super(fs_info);
2298                 if (ret)
2299                         return ret;
2300         }
2301
2302         return 0;
2303 }
2304
2305 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2306 {
2307         struct btrfs_root *tree_root = fs_info->tree_root;
2308         struct btrfs_root *root;
2309         struct btrfs_key location;
2310         int ret;
2311
2312         BUG_ON(!fs_info->tree_root);
2313
2314         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2315         location.type = BTRFS_ROOT_ITEM_KEY;
2316         location.offset = 0;
2317
2318         root = btrfs_read_tree_root(tree_root, &location);
2319         if (IS_ERR(root)) {
2320                 ret = PTR_ERR(root);
2321                 goto out;
2322         }
2323         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2324         fs_info->extent_root = root;
2325
2326         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2327         root = btrfs_read_tree_root(tree_root, &location);
2328         if (IS_ERR(root)) {
2329                 ret = PTR_ERR(root);
2330                 goto out;
2331         }
2332         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2333         fs_info->dev_root = root;
2334         btrfs_init_devices_late(fs_info);
2335
2336         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2337         root = btrfs_read_tree_root(tree_root, &location);
2338         if (IS_ERR(root)) {
2339                 ret = PTR_ERR(root);
2340                 goto out;
2341         }
2342         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343         fs_info->csum_root = root;
2344
2345         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2346         root = btrfs_read_tree_root(tree_root, &location);
2347         if (!IS_ERR(root)) {
2348                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2349                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2350                 fs_info->quota_root = root;
2351         }
2352
2353         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2354         root = btrfs_read_tree_root(tree_root, &location);
2355         if (IS_ERR(root)) {
2356                 ret = PTR_ERR(root);
2357                 if (ret != -ENOENT)
2358                         goto out;
2359         } else {
2360                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361                 fs_info->uuid_root = root;
2362         }
2363
2364         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2365                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2366                 root = btrfs_read_tree_root(tree_root, &location);
2367                 if (IS_ERR(root)) {
2368                         ret = PTR_ERR(root);
2369                         goto out;
2370                 }
2371                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2372                 fs_info->free_space_root = root;
2373         }
2374
2375         return 0;
2376 out:
2377         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2378                    location.objectid, ret);
2379         return ret;
2380 }
2381
2382 /*
2383  * Real super block validation
2384  * NOTE: super csum type and incompat features will not be checked here.
2385  *
2386  * @sb:         super block to check
2387  * @mirror_num: the super block number to check its bytenr:
2388  *              0       the primary (1st) sb
2389  *              1, 2    2nd and 3rd backup copy
2390  *             -1       skip bytenr check
2391  */
2392 static int validate_super(struct btrfs_fs_info *fs_info,
2393                             struct btrfs_super_block *sb, int mirror_num)
2394 {
2395         u64 nodesize = btrfs_super_nodesize(sb);
2396         u64 sectorsize = btrfs_super_sectorsize(sb);
2397         int ret = 0;
2398
2399         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2400                 btrfs_err(fs_info, "no valid FS found");
2401                 ret = -EINVAL;
2402         }
2403         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2404                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2405                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2406                 ret = -EINVAL;
2407         }
2408         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2410                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2411                 ret = -EINVAL;
2412         }
2413         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2414                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2415                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2416                 ret = -EINVAL;
2417         }
2418         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2420                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2421                 ret = -EINVAL;
2422         }
2423
2424         /*
2425          * Check sectorsize and nodesize first, other check will need it.
2426          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2427          */
2428         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2429             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2430                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2431                 ret = -EINVAL;
2432         }
2433         /* Only PAGE SIZE is supported yet */
2434         if (sectorsize != PAGE_SIZE) {
2435                 btrfs_err(fs_info,
2436                         "sectorsize %llu not supported yet, only support %lu",
2437                         sectorsize, PAGE_SIZE);
2438                 ret = -EINVAL;
2439         }
2440         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2441             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2443                 ret = -EINVAL;
2444         }
2445         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2446                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2447                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2448                 ret = -EINVAL;
2449         }
2450
2451         /* Root alignment check */
2452         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2453                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2454                            btrfs_super_root(sb));
2455                 ret = -EINVAL;
2456         }
2457         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2458                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2459                            btrfs_super_chunk_root(sb));
2460                 ret = -EINVAL;
2461         }
2462         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2463                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2464                            btrfs_super_log_root(sb));
2465                 ret = -EINVAL;
2466         }
2467
2468         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2469                    BTRFS_FSID_SIZE) != 0) {
2470                 btrfs_err(fs_info,
2471                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2472                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2473                 ret = -EINVAL;
2474         }
2475
2476         /*
2477          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2478          * done later
2479          */
2480         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2481                 btrfs_err(fs_info, "bytes_used is too small %llu",
2482                           btrfs_super_bytes_used(sb));
2483                 ret = -EINVAL;
2484         }
2485         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2486                 btrfs_err(fs_info, "invalid stripesize %u",
2487                           btrfs_super_stripesize(sb));
2488                 ret = -EINVAL;
2489         }
2490         if (btrfs_super_num_devices(sb) > (1UL << 31))
2491                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2492                            btrfs_super_num_devices(sb));
2493         if (btrfs_super_num_devices(sb) == 0) {
2494                 btrfs_err(fs_info, "number of devices is 0");
2495                 ret = -EINVAL;
2496         }
2497
2498         if (mirror_num >= 0 &&
2499             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2500                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2501                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2502                 ret = -EINVAL;
2503         }
2504
2505         /*
2506          * Obvious sys_chunk_array corruptions, it must hold at least one key
2507          * and one chunk
2508          */
2509         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2510                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2511                           btrfs_super_sys_array_size(sb),
2512                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2513                 ret = -EINVAL;
2514         }
2515         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2516                         + sizeof(struct btrfs_chunk)) {
2517                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2518                           btrfs_super_sys_array_size(sb),
2519                           sizeof(struct btrfs_disk_key)
2520                           + sizeof(struct btrfs_chunk));
2521                 ret = -EINVAL;
2522         }
2523
2524         /*
2525          * The generation is a global counter, we'll trust it more than the others
2526          * but it's still possible that it's the one that's wrong.
2527          */
2528         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2529                 btrfs_warn(fs_info,
2530                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2531                         btrfs_super_generation(sb),
2532                         btrfs_super_chunk_root_generation(sb));
2533         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2534             && btrfs_super_cache_generation(sb) != (u64)-1)
2535                 btrfs_warn(fs_info,
2536                         "suspicious: generation < cache_generation: %llu < %llu",
2537                         btrfs_super_generation(sb),
2538                         btrfs_super_cache_generation(sb));
2539
2540         return ret;
2541 }
2542
2543 /*
2544  * Validation of super block at mount time.
2545  * Some checks already done early at mount time, like csum type and incompat
2546  * flags will be skipped.
2547  */
2548 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2549 {
2550         return validate_super(fs_info, fs_info->super_copy, 0);
2551 }
2552
2553 /*
2554  * Validation of super block at write time.
2555  * Some checks like bytenr check will be skipped as their values will be
2556  * overwritten soon.
2557  * Extra checks like csum type and incompat flags will be done here.
2558  */
2559 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2560                                       struct btrfs_super_block *sb)
2561 {
2562         int ret;
2563
2564         ret = validate_super(fs_info, sb, -1);
2565         if (ret < 0)
2566                 goto out;
2567         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2568                 ret = -EUCLEAN;
2569                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2570                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2571                 goto out;
2572         }
2573         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2574                 ret = -EUCLEAN;
2575                 btrfs_err(fs_info,
2576                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2577                           btrfs_super_incompat_flags(sb),
2578                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2579                 goto out;
2580         }
2581 out:
2582         if (ret < 0)
2583                 btrfs_err(fs_info,
2584                 "super block corruption detected before writing it to disk");
2585         return ret;
2586 }
2587
2588 int open_ctree(struct super_block *sb,
2589                struct btrfs_fs_devices *fs_devices,
2590                char *options)
2591 {
2592         u32 sectorsize;
2593         u32 nodesize;
2594         u32 stripesize;
2595         u64 generation;
2596         u64 features;
2597         struct btrfs_key location;
2598         struct buffer_head *bh;
2599         struct btrfs_super_block *disk_super;
2600         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2601         struct btrfs_root *tree_root;
2602         struct btrfs_root *chunk_root;
2603         int ret;
2604         int err = -EINVAL;
2605         int num_backups_tried = 0;
2606         int backup_index = 0;
2607         int clear_free_space_tree = 0;
2608         int level;
2609
2610         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2611         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2612         if (!tree_root || !chunk_root) {
2613                 err = -ENOMEM;
2614                 goto fail;
2615         }
2616
2617         ret = init_srcu_struct(&fs_info->subvol_srcu);
2618         if (ret) {
2619                 err = ret;
2620                 goto fail;
2621         }
2622
2623         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2624         if (ret) {
2625                 err = ret;
2626                 goto fail_srcu;
2627         }
2628         fs_info->dirty_metadata_batch = PAGE_SIZE *
2629                                         (1 + ilog2(nr_cpu_ids));
2630
2631         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2632         if (ret) {
2633                 err = ret;
2634                 goto fail_dirty_metadata_bytes;
2635         }
2636
2637         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2638                         GFP_KERNEL);
2639         if (ret) {
2640                 err = ret;
2641                 goto fail_delalloc_bytes;
2642         }
2643
2644         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2645         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2646         INIT_LIST_HEAD(&fs_info->trans_list);
2647         INIT_LIST_HEAD(&fs_info->dead_roots);
2648         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2649         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2650         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2651         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2652         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2653         spin_lock_init(&fs_info->delalloc_root_lock);
2654         spin_lock_init(&fs_info->trans_lock);
2655         spin_lock_init(&fs_info->fs_roots_radix_lock);
2656         spin_lock_init(&fs_info->delayed_iput_lock);
2657         spin_lock_init(&fs_info->defrag_inodes_lock);
2658         spin_lock_init(&fs_info->tree_mod_seq_lock);
2659         spin_lock_init(&fs_info->super_lock);
2660         spin_lock_init(&fs_info->qgroup_op_lock);
2661         spin_lock_init(&fs_info->buffer_lock);
2662         spin_lock_init(&fs_info->unused_bgs_lock);
2663         rwlock_init(&fs_info->tree_mod_log_lock);
2664         mutex_init(&fs_info->unused_bg_unpin_mutex);
2665         mutex_init(&fs_info->delete_unused_bgs_mutex);
2666         mutex_init(&fs_info->reloc_mutex);
2667         mutex_init(&fs_info->delalloc_root_mutex);
2668         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2669         seqlock_init(&fs_info->profiles_lock);
2670
2671         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2672         INIT_LIST_HEAD(&fs_info->space_info);
2673         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2674         INIT_LIST_HEAD(&fs_info->unused_bgs);
2675         btrfs_mapping_init(&fs_info->mapping_tree);
2676         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2677                              BTRFS_BLOCK_RSV_GLOBAL);
2678         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2679         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2680         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2681         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2682                              BTRFS_BLOCK_RSV_DELOPS);
2683         atomic_set(&fs_info->async_delalloc_pages, 0);
2684         atomic_set(&fs_info->defrag_running, 0);
2685         atomic_set(&fs_info->qgroup_op_seq, 0);
2686         atomic_set(&fs_info->reada_works_cnt, 0);
2687         atomic64_set(&fs_info->tree_mod_seq, 0);
2688         fs_info->sb = sb;
2689         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2690         fs_info->metadata_ratio = 0;
2691         fs_info->defrag_inodes = RB_ROOT;
2692         atomic64_set(&fs_info->free_chunk_space, 0);
2693         fs_info->tree_mod_log = RB_ROOT;
2694         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2695         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2696         /* readahead state */
2697         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2698         spin_lock_init(&fs_info->reada_lock);
2699         btrfs_init_ref_verify(fs_info);
2700
2701         fs_info->thread_pool_size = min_t(unsigned long,
2702                                           num_online_cpus() + 2, 8);
2703
2704         INIT_LIST_HEAD(&fs_info->ordered_roots);
2705         spin_lock_init(&fs_info->ordered_root_lock);
2706
2707         fs_info->btree_inode = new_inode(sb);
2708         if (!fs_info->btree_inode) {
2709                 err = -ENOMEM;
2710                 goto fail_bio_counter;
2711         }
2712         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2713
2714         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2715                                         GFP_KERNEL);
2716         if (!fs_info->delayed_root) {
2717                 err = -ENOMEM;
2718                 goto fail_iput;
2719         }
2720         btrfs_init_delayed_root(fs_info->delayed_root);
2721
2722         btrfs_init_scrub(fs_info);
2723 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2724         fs_info->check_integrity_print_mask = 0;
2725 #endif
2726         btrfs_init_balance(fs_info);
2727         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2728
2729         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2730         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2731
2732         btrfs_init_btree_inode(fs_info);
2733
2734         spin_lock_init(&fs_info->block_group_cache_lock);
2735         fs_info->block_group_cache_tree = RB_ROOT;
2736         fs_info->first_logical_byte = (u64)-1;
2737
2738         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2739         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2740         fs_info->pinned_extents = &fs_info->freed_extents[0];
2741         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2742
2743         mutex_init(&fs_info->ordered_operations_mutex);
2744         mutex_init(&fs_info->tree_log_mutex);
2745         mutex_init(&fs_info->chunk_mutex);
2746         mutex_init(&fs_info->transaction_kthread_mutex);
2747         mutex_init(&fs_info->cleaner_mutex);
2748         mutex_init(&fs_info->ro_block_group_mutex);
2749         init_rwsem(&fs_info->commit_root_sem);
2750         init_rwsem(&fs_info->cleanup_work_sem);
2751         init_rwsem(&fs_info->subvol_sem);
2752         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2753
2754         btrfs_init_dev_replace_locks(fs_info);
2755         btrfs_init_qgroup(fs_info);
2756
2757         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2758         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2759
2760         init_waitqueue_head(&fs_info->transaction_throttle);
2761         init_waitqueue_head(&fs_info->transaction_wait);
2762         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2763         init_waitqueue_head(&fs_info->async_submit_wait);
2764
2765         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2766
2767         /* Usable values until the real ones are cached from the superblock */
2768         fs_info->nodesize = 4096;
2769         fs_info->sectorsize = 4096;
2770         fs_info->stripesize = 4096;
2771
2772         spin_lock_init(&fs_info->swapfile_pins_lock);
2773         fs_info->swapfile_pins = RB_ROOT;
2774
2775         ret = btrfs_alloc_stripe_hash_table(fs_info);
2776         if (ret) {
2777                 err = ret;
2778                 goto fail_alloc;
2779         }
2780
2781         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2782
2783         invalidate_bdev(fs_devices->latest_bdev);
2784
2785         /*
2786          * Read super block and check the signature bytes only
2787          */
2788         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2789         if (IS_ERR(bh)) {
2790                 err = PTR_ERR(bh);
2791                 goto fail_alloc;
2792         }
2793
2794         /*
2795          * We want to check superblock checksum, the type is stored inside.
2796          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2797          */
2798         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2799                 btrfs_err(fs_info, "superblock checksum mismatch");
2800                 err = -EINVAL;
2801                 brelse(bh);
2802                 goto fail_alloc;
2803         }
2804
2805         /*
2806          * super_copy is zeroed at allocation time and we never touch the
2807          * following bytes up to INFO_SIZE, the checksum is calculated from
2808          * the whole block of INFO_SIZE
2809          */
2810         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2811         brelse(bh);
2812
2813         disk_super = fs_info->super_copy;
2814
2815         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2816                        BTRFS_FSID_SIZE));
2817
2818         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2819                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2820                                 fs_info->super_copy->metadata_uuid,
2821                                 BTRFS_FSID_SIZE));
2822         }
2823
2824         features = btrfs_super_flags(disk_super);
2825         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2826                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2827                 btrfs_set_super_flags(disk_super, features);
2828                 btrfs_info(fs_info,
2829                         "found metadata UUID change in progress flag, clearing");
2830         }
2831
2832         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2833                sizeof(*fs_info->super_for_commit));
2834
2835         ret = btrfs_validate_mount_super(fs_info);
2836         if (ret) {
2837                 btrfs_err(fs_info, "superblock contains fatal errors");
2838                 err = -EINVAL;
2839                 goto fail_alloc;
2840         }
2841
2842         if (!btrfs_super_root(disk_super))
2843                 goto fail_alloc;
2844
2845         /* check FS state, whether FS is broken. */
2846         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2847                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2848
2849         /*
2850          * run through our array of backup supers and setup
2851          * our ring pointer to the oldest one
2852          */
2853         generation = btrfs_super_generation(disk_super);
2854         find_oldest_super_backup(fs_info, generation);
2855
2856         /*
2857          * In the long term, we'll store the compression type in the super
2858          * block, and it'll be used for per file compression control.
2859          */
2860         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2861
2862         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2863         if (ret) {
2864                 err = ret;
2865                 goto fail_alloc;
2866         }
2867
2868         features = btrfs_super_incompat_flags(disk_super) &
2869                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2870         if (features) {
2871                 btrfs_err(fs_info,
2872                     "cannot mount because of unsupported optional features (%llx)",
2873                     features);
2874                 err = -EINVAL;
2875                 goto fail_alloc;
2876         }
2877
2878         features = btrfs_super_incompat_flags(disk_super);
2879         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2880         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2881                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2882         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2883                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2884
2885         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2886                 btrfs_info(fs_info, "has skinny extents");
2887
2888         /*
2889          * flag our filesystem as having big metadata blocks if
2890          * they are bigger than the page size
2891          */
2892         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2893                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2894                         btrfs_info(fs_info,
2895                                 "flagging fs with big metadata feature");
2896                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2897         }
2898
2899         nodesize = btrfs_super_nodesize(disk_super);
2900         sectorsize = btrfs_super_sectorsize(disk_super);
2901         stripesize = sectorsize;
2902         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2903         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2904
2905         /* Cache block sizes */
2906         fs_info->nodesize = nodesize;
2907         fs_info->sectorsize = sectorsize;
2908         fs_info->stripesize = stripesize;
2909
2910         /*
2911          * mixed block groups end up with duplicate but slightly offset
2912          * extent buffers for the same range.  It leads to corruptions
2913          */
2914         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2915             (sectorsize != nodesize)) {
2916                 btrfs_err(fs_info,
2917 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2918                         nodesize, sectorsize);
2919                 goto fail_alloc;
2920         }
2921
2922         /*
2923          * Needn't use the lock because there is no other task which will
2924          * update the flag.
2925          */
2926         btrfs_set_super_incompat_flags(disk_super, features);
2927
2928         features = btrfs_super_compat_ro_flags(disk_super) &
2929                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2930         if (!sb_rdonly(sb) && features) {
2931                 btrfs_err(fs_info,
2932         "cannot mount read-write because of unsupported optional features (%llx)",
2933                        features);
2934                 err = -EINVAL;
2935                 goto fail_alloc;
2936         }
2937
2938         ret = btrfs_init_workqueues(fs_info, fs_devices);
2939         if (ret) {
2940                 err = ret;
2941                 goto fail_sb_buffer;
2942         }
2943
2944         sb->s_bdi->congested_fn = btrfs_congested_fn;
2945         sb->s_bdi->congested_data = fs_info;
2946         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2947         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2948         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2949         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2950
2951         sb->s_blocksize = sectorsize;
2952         sb->s_blocksize_bits = blksize_bits(sectorsize);
2953         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2954
2955         mutex_lock(&fs_info->chunk_mutex);
2956         ret = btrfs_read_sys_array(fs_info);
2957         mutex_unlock(&fs_info->chunk_mutex);
2958         if (ret) {
2959                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2960                 goto fail_sb_buffer;
2961         }
2962
2963         generation = btrfs_super_chunk_root_generation(disk_super);
2964         level = btrfs_super_chunk_root_level(disk_super);
2965
2966         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2967
2968         chunk_root->node = read_tree_block(fs_info,
2969                                            btrfs_super_chunk_root(disk_super),
2970                                            generation, level, NULL);
2971         if (IS_ERR(chunk_root->node) ||
2972             !extent_buffer_uptodate(chunk_root->node)) {
2973                 btrfs_err(fs_info, "failed to read chunk root");
2974                 if (!IS_ERR(chunk_root->node))
2975                         free_extent_buffer(chunk_root->node);
2976                 chunk_root->node = NULL;
2977                 goto fail_tree_roots;
2978         }
2979         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2980         chunk_root->commit_root = btrfs_root_node(chunk_root);
2981
2982         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2983            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2984
2985         ret = btrfs_read_chunk_tree(fs_info);
2986         if (ret) {
2987                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2988                 goto fail_tree_roots;
2989         }
2990
2991         /*
2992          * Keep the devid that is marked to be the target device for the
2993          * device replace procedure
2994          */
2995         btrfs_free_extra_devids(fs_devices, 0);
2996
2997         if (!fs_devices->latest_bdev) {
2998                 btrfs_err(fs_info, "failed to read devices");
2999                 goto fail_tree_roots;
3000         }
3001
3002 retry_root_backup:
3003         generation = btrfs_super_generation(disk_super);
3004         level = btrfs_super_root_level(disk_super);
3005
3006         tree_root->node = read_tree_block(fs_info,
3007                                           btrfs_super_root(disk_super),
3008                                           generation, level, NULL);
3009         if (IS_ERR(tree_root->node) ||
3010             !extent_buffer_uptodate(tree_root->node)) {
3011                 btrfs_warn(fs_info, "failed to read tree root");
3012                 if (!IS_ERR(tree_root->node))
3013                         free_extent_buffer(tree_root->node);
3014                 tree_root->node = NULL;
3015                 goto recovery_tree_root;
3016         }
3017
3018         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3019         tree_root->commit_root = btrfs_root_node(tree_root);
3020         btrfs_set_root_refs(&tree_root->root_item, 1);
3021
3022         mutex_lock(&tree_root->objectid_mutex);
3023         ret = btrfs_find_highest_objectid(tree_root,
3024                                         &tree_root->highest_objectid);
3025         if (ret) {
3026                 mutex_unlock(&tree_root->objectid_mutex);
3027                 goto recovery_tree_root;
3028         }
3029
3030         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3031
3032         mutex_unlock(&tree_root->objectid_mutex);
3033
3034         ret = btrfs_read_roots(fs_info);
3035         if (ret)
3036                 goto recovery_tree_root;
3037
3038         fs_info->generation = generation;
3039         fs_info->last_trans_committed = generation;
3040
3041         ret = btrfs_verify_dev_extents(fs_info);
3042         if (ret) {
3043                 btrfs_err(fs_info,
3044                           "failed to verify dev extents against chunks: %d",
3045                           ret);
3046                 goto fail_block_groups;
3047         }
3048         ret = btrfs_recover_balance(fs_info);
3049         if (ret) {
3050                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3051                 goto fail_block_groups;
3052         }
3053
3054         ret = btrfs_init_dev_stats(fs_info);
3055         if (ret) {
3056                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3057                 goto fail_block_groups;
3058         }
3059
3060         ret = btrfs_init_dev_replace(fs_info);
3061         if (ret) {
3062                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3063                 goto fail_block_groups;
3064         }
3065
3066         btrfs_free_extra_devids(fs_devices, 1);
3067
3068         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3069         if (ret) {
3070                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3071                                 ret);
3072                 goto fail_block_groups;
3073         }
3074
3075         ret = btrfs_sysfs_add_device(fs_devices);
3076         if (ret) {
3077                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3078                                 ret);
3079                 goto fail_fsdev_sysfs;
3080         }
3081
3082         ret = btrfs_sysfs_add_mounted(fs_info);
3083         if (ret) {
3084                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3085                 goto fail_fsdev_sysfs;
3086         }
3087
3088         ret = btrfs_init_space_info(fs_info);
3089         if (ret) {
3090                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3091                 goto fail_sysfs;
3092         }
3093
3094         ret = btrfs_read_block_groups(fs_info);
3095         if (ret) {
3096                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3097                 goto fail_sysfs;
3098         }
3099
3100         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3101                 btrfs_warn(fs_info,
3102                 "writeable mount is not allowed due to too many missing devices");
3103                 goto fail_sysfs;
3104         }
3105
3106         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3107                                                "btrfs-cleaner");
3108         if (IS_ERR(fs_info->cleaner_kthread))
3109                 goto fail_sysfs;
3110
3111         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3112                                                    tree_root,
3113                                                    "btrfs-transaction");
3114         if (IS_ERR(fs_info->transaction_kthread))
3115                 goto fail_cleaner;
3116
3117         if (!btrfs_test_opt(fs_info, NOSSD) &&
3118             !fs_info->fs_devices->rotating) {
3119                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3120         }
3121
3122         /*
3123          * Mount does not set all options immediately, we can do it now and do
3124          * not have to wait for transaction commit
3125          */
3126         btrfs_apply_pending_changes(fs_info);
3127
3128 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3129         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3130                 ret = btrfsic_mount(fs_info, fs_devices,
3131                                     btrfs_test_opt(fs_info,
3132                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3133                                     1 : 0,
3134                                     fs_info->check_integrity_print_mask);
3135                 if (ret)
3136                         btrfs_warn(fs_info,
3137                                 "failed to initialize integrity check module: %d",
3138                                 ret);
3139         }
3140 #endif
3141         ret = btrfs_read_qgroup_config(fs_info);
3142         if (ret)
3143                 goto fail_trans_kthread;
3144
3145         if (btrfs_build_ref_tree(fs_info))
3146                 btrfs_err(fs_info, "couldn't build ref tree");
3147
3148         /* do not make disk changes in broken FS or nologreplay is given */
3149         if (btrfs_super_log_root(disk_super) != 0 &&
3150             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3151                 ret = btrfs_replay_log(fs_info, fs_devices);
3152                 if (ret) {
3153                         err = ret;
3154                         goto fail_qgroup;
3155                 }
3156         }
3157
3158         ret = btrfs_find_orphan_roots(fs_info);
3159         if (ret)
3160                 goto fail_qgroup;
3161
3162         if (!sb_rdonly(sb)) {
3163                 ret = btrfs_cleanup_fs_roots(fs_info);
3164                 if (ret)
3165                         goto fail_qgroup;
3166
3167                 mutex_lock(&fs_info->cleaner_mutex);
3168                 ret = btrfs_recover_relocation(tree_root);
3169                 mutex_unlock(&fs_info->cleaner_mutex);
3170                 if (ret < 0) {
3171                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3172                                         ret);
3173                         err = -EINVAL;
3174                         goto fail_qgroup;
3175                 }
3176         }
3177
3178         location.objectid = BTRFS_FS_TREE_OBJECTID;
3179         location.type = BTRFS_ROOT_ITEM_KEY;
3180         location.offset = 0;
3181
3182         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3183         if (IS_ERR(fs_info->fs_root)) {
3184                 err = PTR_ERR(fs_info->fs_root);
3185                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3186                 goto fail_qgroup;
3187         }
3188
3189         if (sb_rdonly(sb))
3190                 return 0;
3191
3192         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3193             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3194                 clear_free_space_tree = 1;
3195         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3196                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3197                 btrfs_warn(fs_info, "free space tree is invalid");
3198                 clear_free_space_tree = 1;
3199         }
3200
3201         if (clear_free_space_tree) {
3202                 btrfs_info(fs_info, "clearing free space tree");
3203                 ret = btrfs_clear_free_space_tree(fs_info);
3204                 if (ret) {
3205                         btrfs_warn(fs_info,
3206                                    "failed to clear free space tree: %d", ret);
3207                         close_ctree(fs_info);
3208                         return ret;
3209                 }
3210         }
3211
3212         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3213             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3214                 btrfs_info(fs_info, "creating free space tree");
3215                 ret = btrfs_create_free_space_tree(fs_info);
3216                 if (ret) {
3217                         btrfs_warn(fs_info,
3218                                 "failed to create free space tree: %d", ret);
3219                         close_ctree(fs_info);
3220                         return ret;
3221                 }
3222         }
3223
3224         down_read(&fs_info->cleanup_work_sem);
3225         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3226             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3227                 up_read(&fs_info->cleanup_work_sem);
3228                 close_ctree(fs_info);
3229                 return ret;
3230         }
3231         up_read(&fs_info->cleanup_work_sem);
3232
3233         ret = btrfs_resume_balance_async(fs_info);
3234         if (ret) {
3235                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3236                 close_ctree(fs_info);
3237                 return ret;
3238         }
3239
3240         ret = btrfs_resume_dev_replace_async(fs_info);
3241         if (ret) {
3242                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3243                 close_ctree(fs_info);
3244                 return ret;
3245         }
3246
3247         btrfs_qgroup_rescan_resume(fs_info);
3248
3249         if (!fs_info->uuid_root) {
3250                 btrfs_info(fs_info, "creating UUID tree");
3251                 ret = btrfs_create_uuid_tree(fs_info);
3252                 if (ret) {
3253                         btrfs_warn(fs_info,
3254                                 "failed to create the UUID tree: %d", ret);
3255                         close_ctree(fs_info);
3256                         return ret;
3257                 }
3258         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3259                    fs_info->generation !=
3260                                 btrfs_super_uuid_tree_generation(disk_super)) {
3261                 btrfs_info(fs_info, "checking UUID tree");
3262                 ret = btrfs_check_uuid_tree(fs_info);
3263                 if (ret) {
3264                         btrfs_warn(fs_info,
3265                                 "failed to check the UUID tree: %d", ret);
3266                         close_ctree(fs_info);
3267                         return ret;
3268                 }
3269         } else {
3270                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3271         }
3272         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3273
3274         /*
3275          * backuproot only affect mount behavior, and if open_ctree succeeded,
3276          * no need to keep the flag
3277          */
3278         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3279
3280         return 0;
3281
3282 fail_qgroup:
3283         btrfs_free_qgroup_config(fs_info);
3284 fail_trans_kthread:
3285         kthread_stop(fs_info->transaction_kthread);
3286         btrfs_cleanup_transaction(fs_info);
3287         btrfs_free_fs_roots(fs_info);
3288 fail_cleaner:
3289         kthread_stop(fs_info->cleaner_kthread);
3290
3291         /*
3292          * make sure we're done with the btree inode before we stop our
3293          * kthreads
3294          */
3295         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3296
3297 fail_sysfs:
3298         btrfs_sysfs_remove_mounted(fs_info);
3299
3300 fail_fsdev_sysfs:
3301         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3302
3303 fail_block_groups:
3304         btrfs_put_block_group_cache(fs_info);
3305
3306 fail_tree_roots:
3307         free_root_pointers(fs_info, 1);
3308         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3309
3310 fail_sb_buffer:
3311         btrfs_stop_all_workers(fs_info);
3312         btrfs_free_block_groups(fs_info);
3313 fail_alloc:
3314 fail_iput:
3315         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3316
3317         iput(fs_info->btree_inode);
3318 fail_bio_counter:
3319         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3320 fail_delalloc_bytes:
3321         percpu_counter_destroy(&fs_info->delalloc_bytes);
3322 fail_dirty_metadata_bytes:
3323         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3324 fail_srcu:
3325         cleanup_srcu_struct(&fs_info->subvol_srcu);
3326 fail:
3327         btrfs_free_stripe_hash_table(fs_info);
3328         btrfs_close_devices(fs_info->fs_devices);
3329         return err;
3330
3331 recovery_tree_root:
3332         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3333                 goto fail_tree_roots;
3334
3335         free_root_pointers(fs_info, 0);
3336
3337         /* don't use the log in recovery mode, it won't be valid */
3338         btrfs_set_super_log_root(disk_super, 0);
3339
3340         /* we can't trust the free space cache either */
3341         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3342
3343         ret = next_root_backup(fs_info, fs_info->super_copy,
3344                                &num_backups_tried, &backup_index);
3345         if (ret == -1)
3346                 goto fail_block_groups;
3347         goto retry_root_backup;
3348 }
3349 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3350
3351 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3352 {
3353         if (uptodate) {
3354                 set_buffer_uptodate(bh);
3355         } else {
3356                 struct btrfs_device *device = (struct btrfs_device *)
3357                         bh->b_private;
3358
3359                 btrfs_warn_rl_in_rcu(device->fs_info,
3360                                 "lost page write due to IO error on %s",
3361                                           rcu_str_deref(device->name));
3362                 /* note, we don't set_buffer_write_io_error because we have
3363                  * our own ways of dealing with the IO errors
3364                  */
3365                 clear_buffer_uptodate(bh);
3366                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3367         }
3368         unlock_buffer(bh);
3369         put_bh(bh);
3370 }
3371
3372 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3373                         struct buffer_head **bh_ret)
3374 {
3375         struct buffer_head *bh;
3376         struct btrfs_super_block *super;
3377         u64 bytenr;
3378
3379         bytenr = btrfs_sb_offset(copy_num);
3380         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3381                 return -EINVAL;
3382
3383         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3384         /*
3385          * If we fail to read from the underlying devices, as of now
3386          * the best option we have is to mark it EIO.
3387          */
3388         if (!bh)
3389                 return -EIO;
3390
3391         super = (struct btrfs_super_block *)bh->b_data;
3392         if (btrfs_super_bytenr(super) != bytenr ||
3393                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3394                 brelse(bh);
3395                 return -EINVAL;
3396         }
3397
3398         *bh_ret = bh;
3399         return 0;
3400 }
3401
3402
3403 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3404 {
3405         struct buffer_head *bh;
3406         struct buffer_head *latest = NULL;
3407         struct btrfs_super_block *super;
3408         int i;
3409         u64 transid = 0;
3410         int ret = -EINVAL;
3411
3412         /* we would like to check all the supers, but that would make
3413          * a btrfs mount succeed after a mkfs from a different FS.
3414          * So, we need to add a special mount option to scan for
3415          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3416          */
3417         for (i = 0; i < 1; i++) {
3418                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3419                 if (ret)
3420                         continue;
3421
3422                 super = (struct btrfs_super_block *)bh->b_data;
3423
3424                 if (!latest || btrfs_super_generation(super) > transid) {
3425                         brelse(latest);
3426                         latest = bh;
3427                         transid = btrfs_super_generation(super);
3428                 } else {
3429                         brelse(bh);
3430                 }
3431         }
3432
3433         if (!latest)
3434                 return ERR_PTR(ret);
3435
3436         return latest;
3437 }
3438
3439 /*
3440  * Write superblock @sb to the @device. Do not wait for completion, all the
3441  * buffer heads we write are pinned.
3442  *
3443  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3444  * the expected device size at commit time. Note that max_mirrors must be
3445  * same for write and wait phases.
3446  *
3447  * Return number of errors when buffer head is not found or submission fails.
3448  */
3449 static int write_dev_supers(struct btrfs_device *device,
3450                             struct btrfs_super_block *sb, int max_mirrors)
3451 {
3452         struct buffer_head *bh;
3453         int i;
3454         int ret;
3455         int errors = 0;
3456         u32 crc;
3457         u64 bytenr;
3458         int op_flags;
3459
3460         if (max_mirrors == 0)
3461                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3462
3463         for (i = 0; i < max_mirrors; i++) {
3464                 bytenr = btrfs_sb_offset(i);
3465                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3466                     device->commit_total_bytes)
3467                         break;
3468
3469                 btrfs_set_super_bytenr(sb, bytenr);
3470
3471                 crc = ~(u32)0;
3472                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3473                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3474                 btrfs_csum_final(crc, sb->csum);
3475
3476                 /* One reference for us, and we leave it for the caller */
3477                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3478                               BTRFS_SUPER_INFO_SIZE);
3479                 if (!bh) {
3480                         btrfs_err(device->fs_info,
3481                             "couldn't get super buffer head for bytenr %llu",
3482                             bytenr);
3483                         errors++;
3484                         continue;
3485                 }
3486
3487                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3488
3489                 /* one reference for submit_bh */
3490                 get_bh(bh);
3491
3492                 set_buffer_uptodate(bh);
3493                 lock_buffer(bh);
3494                 bh->b_end_io = btrfs_end_buffer_write_sync;
3495                 bh->b_private = device;
3496
3497                 /*
3498                  * we fua the first super.  The others we allow
3499                  * to go down lazy.
3500                  */
3501                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3502                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3503                         op_flags |= REQ_FUA;
3504                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3505                 if (ret)
3506                         errors++;
3507         }
3508         return errors < i ? 0 : -1;
3509 }
3510
3511 /*
3512  * Wait for write completion of superblocks done by write_dev_supers,
3513  * @max_mirrors same for write and wait phases.
3514  *
3515  * Return number of errors when buffer head is not found or not marked up to
3516  * date.
3517  */
3518 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3519 {
3520         struct buffer_head *bh;
3521         int i;
3522         int errors = 0;
3523         bool primary_failed = false;
3524         u64 bytenr;
3525
3526         if (max_mirrors == 0)
3527                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3528
3529         for (i = 0; i < max_mirrors; i++) {
3530                 bytenr = btrfs_sb_offset(i);
3531                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3532                     device->commit_total_bytes)
3533                         break;
3534
3535                 bh = __find_get_block(device->bdev,
3536                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3537                                       BTRFS_SUPER_INFO_SIZE);
3538                 if (!bh) {
3539                         errors++;
3540                         if (i == 0)
3541                                 primary_failed = true;
3542                         continue;
3543                 }
3544                 wait_on_buffer(bh);
3545                 if (!buffer_uptodate(bh)) {
3546                         errors++;
3547                         if (i == 0)
3548                                 primary_failed = true;
3549                 }
3550
3551                 /* drop our reference */
3552                 brelse(bh);
3553
3554                 /* drop the reference from the writing run */
3555                 brelse(bh);
3556         }
3557
3558         /* log error, force error return */
3559         if (primary_failed) {
3560                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3561                           device->devid);
3562                 return -1;
3563         }
3564
3565         return errors < i ? 0 : -1;
3566 }
3567
3568 /*
3569  * endio for the write_dev_flush, this will wake anyone waiting
3570  * for the barrier when it is done
3571  */
3572 static void btrfs_end_empty_barrier(struct bio *bio)
3573 {
3574         complete(bio->bi_private);
3575 }
3576
3577 /*
3578  * Submit a flush request to the device if it supports it. Error handling is
3579  * done in the waiting counterpart.
3580  */
3581 static void write_dev_flush(struct btrfs_device *device)
3582 {
3583         struct request_queue *q = bdev_get_queue(device->bdev);
3584         struct bio *bio = device->flush_bio;
3585
3586         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3587                 return;
3588
3589         bio_reset(bio);
3590         bio->bi_end_io = btrfs_end_empty_barrier;
3591         bio_set_dev(bio, device->bdev);
3592         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3593         init_completion(&device->flush_wait);
3594         bio->bi_private = &device->flush_wait;
3595
3596         btrfsic_submit_bio(bio);
3597         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3598 }
3599
3600 /*
3601  * If the flush bio has been submitted by write_dev_flush, wait for it.
3602  */
3603 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3604 {
3605         struct bio *bio = device->flush_bio;
3606
3607         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3608                 return BLK_STS_OK;
3609
3610         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3611         wait_for_completion_io(&device->flush_wait);
3612
3613         return bio->bi_status;
3614 }
3615
3616 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3617 {
3618         if (!btrfs_check_rw_degradable(fs_info, NULL))
3619                 return -EIO;
3620         return 0;
3621 }
3622
3623 /*
3624  * send an empty flush down to each device in parallel,
3625  * then wait for them
3626  */
3627 static int barrier_all_devices(struct btrfs_fs_info *info)
3628 {
3629         struct list_head *head;
3630         struct btrfs_device *dev;
3631         int errors_wait = 0;
3632         blk_status_t ret;
3633
3634         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3635         /* send down all the barriers */
3636         head = &info->fs_devices->devices;
3637         list_for_each_entry(dev, head, dev_list) {
3638                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3639                         continue;
3640                 if (!dev->bdev)
3641                         continue;
3642                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3643                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3644                         continue;
3645
3646                 write_dev_flush(dev);
3647                 dev->last_flush_error = BLK_STS_OK;
3648         }
3649
3650         /* wait for all the barriers */
3651         list_for_each_entry(dev, head, dev_list) {
3652                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3653                         continue;
3654                 if (!dev->bdev) {
3655                         errors_wait++;
3656                         continue;
3657                 }
3658                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3659                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3660                         continue;
3661
3662                 ret = wait_dev_flush(dev);
3663                 if (ret) {
3664                         dev->last_flush_error = ret;
3665                         btrfs_dev_stat_inc_and_print(dev,
3666                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3667                         errors_wait++;
3668                 }
3669         }
3670
3671         if (errors_wait) {
3672                 /*
3673                  * At some point we need the status of all disks
3674                  * to arrive at the volume status. So error checking
3675                  * is being pushed to a separate loop.
3676                  */
3677                 return check_barrier_error(info);
3678         }
3679         return 0;
3680 }
3681
3682 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3683 {
3684         int raid_type;
3685         int min_tolerated = INT_MAX;
3686
3687         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3688             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3689                 min_tolerated = min(min_tolerated,
3690                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3691                                     tolerated_failures);
3692
3693         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3694                 if (raid_type == BTRFS_RAID_SINGLE)
3695                         continue;
3696                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3697                         continue;
3698                 min_tolerated = min(min_tolerated,
3699                                     btrfs_raid_array[raid_type].
3700                                     tolerated_failures);
3701         }
3702
3703         if (min_tolerated == INT_MAX) {
3704                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3705                 min_tolerated = 0;
3706         }
3707
3708         return min_tolerated;
3709 }
3710
3711 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3712 {
3713         struct list_head *head;
3714         struct btrfs_device *dev;
3715         struct btrfs_super_block *sb;
3716         struct btrfs_dev_item *dev_item;
3717         int ret;
3718         int do_barriers;
3719         int max_errors;
3720         int total_errors = 0;
3721         u64 flags;
3722
3723         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3724
3725         /*
3726          * max_mirrors == 0 indicates we're from commit_transaction,
3727          * not from fsync where the tree roots in fs_info have not
3728          * been consistent on disk.
3729          */
3730         if (max_mirrors == 0)
3731                 backup_super_roots(fs_info);
3732
3733         sb = fs_info->super_for_commit;
3734         dev_item = &sb->dev_item;
3735
3736         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3737         head = &fs_info->fs_devices->devices;
3738         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3739
3740         if (do_barriers) {
3741                 ret = barrier_all_devices(fs_info);
3742                 if (ret) {
3743                         mutex_unlock(
3744                                 &fs_info->fs_devices->device_list_mutex);
3745                         btrfs_handle_fs_error(fs_info, ret,
3746                                               "errors while submitting device barriers.");
3747                         return ret;
3748                 }
3749         }
3750
3751         list_for_each_entry(dev, head, dev_list) {
3752                 if (!dev->bdev) {
3753                         total_errors++;
3754                         continue;
3755                 }
3756                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3757                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3758                         continue;
3759
3760                 btrfs_set_stack_device_generation(dev_item, 0);
3761                 btrfs_set_stack_device_type(dev_item, dev->type);
3762                 btrfs_set_stack_device_id(dev_item, dev->devid);
3763                 btrfs_set_stack_device_total_bytes(dev_item,
3764                                                    dev->commit_total_bytes);
3765                 btrfs_set_stack_device_bytes_used(dev_item,
3766                                                   dev->commit_bytes_used);
3767                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3768                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3769                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3770                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3771                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3772                        BTRFS_FSID_SIZE);
3773
3774                 flags = btrfs_super_flags(sb);
3775                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3776
3777                 ret = btrfs_validate_write_super(fs_info, sb);
3778                 if (ret < 0) {
3779                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3780                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3781                                 "unexpected superblock corruption detected");
3782                         return -EUCLEAN;
3783                 }
3784
3785                 ret = write_dev_supers(dev, sb, max_mirrors);
3786                 if (ret)
3787                         total_errors++;
3788         }
3789         if (total_errors > max_errors) {
3790                 btrfs_err(fs_info, "%d errors while writing supers",
3791                           total_errors);
3792                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3793
3794                 /* FUA is masked off if unsupported and can't be the reason */
3795                 btrfs_handle_fs_error(fs_info, -EIO,
3796                                       "%d errors while writing supers",
3797                                       total_errors);
3798                 return -EIO;
3799         }
3800
3801         total_errors = 0;
3802         list_for_each_entry(dev, head, dev_list) {
3803                 if (!dev->bdev)
3804                         continue;
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                 ret = wait_dev_supers(dev, max_mirrors);
3810                 if (ret)
3811                         total_errors++;
3812         }
3813         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3814         if (total_errors > max_errors) {
3815                 btrfs_handle_fs_error(fs_info, -EIO,
3816                                       "%d errors while writing supers",
3817                                       total_errors);
3818                 return -EIO;
3819         }
3820         return 0;
3821 }
3822
3823 /* Drop a fs root from the radix tree and free it. */
3824 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3825                                   struct btrfs_root *root)
3826 {
3827         spin_lock(&fs_info->fs_roots_radix_lock);
3828         radix_tree_delete(&fs_info->fs_roots_radix,
3829                           (unsigned long)root->root_key.objectid);
3830         spin_unlock(&fs_info->fs_roots_radix_lock);
3831
3832         if (btrfs_root_refs(&root->root_item) == 0)
3833                 synchronize_srcu(&fs_info->subvol_srcu);
3834
3835         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3836                 btrfs_free_log(NULL, root);
3837                 if (root->reloc_root) {
3838                         free_extent_buffer(root->reloc_root->node);
3839                         free_extent_buffer(root->reloc_root->commit_root);
3840                         btrfs_put_fs_root(root->reloc_root);
3841                         root->reloc_root = NULL;
3842                 }
3843         }
3844
3845         if (root->free_ino_pinned)
3846                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3847         if (root->free_ino_ctl)
3848                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3849         btrfs_free_fs_root(root);
3850 }
3851
3852 void btrfs_free_fs_root(struct btrfs_root *root)
3853 {
3854         iput(root->ino_cache_inode);
3855         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3856         if (root->anon_dev)
3857                 free_anon_bdev(root->anon_dev);
3858         if (root->subv_writers)
3859                 btrfs_free_subvolume_writers(root->subv_writers);
3860         free_extent_buffer(root->node);
3861         free_extent_buffer(root->commit_root);
3862         kfree(root->free_ino_ctl);
3863         kfree(root->free_ino_pinned);
3864         btrfs_put_fs_root(root);
3865 }
3866
3867 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3868 {
3869         u64 root_objectid = 0;
3870         struct btrfs_root *gang[8];
3871         int i = 0;
3872         int err = 0;
3873         unsigned int ret = 0;
3874         int index;
3875
3876         while (1) {
3877                 index = srcu_read_lock(&fs_info->subvol_srcu);
3878                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3879                                              (void **)gang, root_objectid,
3880                                              ARRAY_SIZE(gang));
3881                 if (!ret) {
3882                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3883                         break;
3884                 }
3885                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3886
3887                 for (i = 0; i < ret; i++) {
3888                         /* Avoid to grab roots in dead_roots */
3889                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3890                                 gang[i] = NULL;
3891                                 continue;
3892                         }
3893                         /* grab all the search result for later use */
3894                         gang[i] = btrfs_grab_fs_root(gang[i]);
3895                 }
3896                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3897
3898                 for (i = 0; i < ret; i++) {
3899                         if (!gang[i])
3900                                 continue;
3901                         root_objectid = gang[i]->root_key.objectid;
3902                         err = btrfs_orphan_cleanup(gang[i]);
3903                         if (err)
3904                                 break;
3905                         btrfs_put_fs_root(gang[i]);
3906                 }
3907                 root_objectid++;
3908         }
3909
3910         /* release the uncleaned roots due to error */
3911         for (; i < ret; i++) {
3912                 if (gang[i])
3913                         btrfs_put_fs_root(gang[i]);
3914         }
3915         return err;
3916 }
3917
3918 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3919 {
3920         struct btrfs_root *root = fs_info->tree_root;
3921         struct btrfs_trans_handle *trans;
3922
3923         mutex_lock(&fs_info->cleaner_mutex);
3924         btrfs_run_delayed_iputs(fs_info);
3925         mutex_unlock(&fs_info->cleaner_mutex);
3926         wake_up_process(fs_info->cleaner_kthread);
3927
3928         /* wait until ongoing cleanup work done */
3929         down_write(&fs_info->cleanup_work_sem);
3930         up_write(&fs_info->cleanup_work_sem);
3931
3932         trans = btrfs_join_transaction(root);
3933         if (IS_ERR(trans))
3934                 return PTR_ERR(trans);
3935         return btrfs_commit_transaction(trans);
3936 }
3937
3938 void close_ctree(struct btrfs_fs_info *fs_info)
3939 {
3940         int ret;
3941
3942         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3943         /*
3944          * We don't want the cleaner to start new transactions, add more delayed
3945          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3946          * because that frees the task_struct, and the transaction kthread might
3947          * still try to wake up the cleaner.
3948          */
3949         kthread_park(fs_info->cleaner_kthread);
3950
3951         /* wait for the qgroup rescan worker to stop */
3952         btrfs_qgroup_wait_for_completion(fs_info, false);
3953
3954         /* wait for the uuid_scan task to finish */
3955         down(&fs_info->uuid_tree_rescan_sem);
3956         /* avoid complains from lockdep et al., set sem back to initial state */
3957         up(&fs_info->uuid_tree_rescan_sem);
3958
3959         /* pause restriper - we want to resume on mount */
3960         btrfs_pause_balance(fs_info);
3961
3962         btrfs_dev_replace_suspend_for_unmount(fs_info);
3963
3964         btrfs_scrub_cancel(fs_info);
3965
3966         /* wait for any defraggers to finish */
3967         wait_event(fs_info->transaction_wait,
3968                    (atomic_read(&fs_info->defrag_running) == 0));
3969
3970         /* clear out the rbtree of defraggable inodes */
3971         btrfs_cleanup_defrag_inodes(fs_info);
3972
3973         cancel_work_sync(&fs_info->async_reclaim_work);
3974
3975         if (!sb_rdonly(fs_info->sb)) {
3976                 /*
3977                  * The cleaner kthread is stopped, so do one final pass over
3978                  * unused block groups.
3979                  */
3980                 btrfs_delete_unused_bgs(fs_info);
3981
3982                 ret = btrfs_commit_super(fs_info);
3983                 if (ret)
3984                         btrfs_err(fs_info, "commit super ret %d", ret);
3985         }
3986
3987         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3988             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3989                 btrfs_error_commit_super(fs_info);
3990
3991         kthread_stop(fs_info->transaction_kthread);
3992         kthread_stop(fs_info->cleaner_kthread);
3993
3994         ASSERT(list_empty(&fs_info->delayed_iputs));
3995         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3996
3997         btrfs_free_qgroup_config(fs_info);
3998         ASSERT(list_empty(&fs_info->delalloc_roots));
3999
4000         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4001                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4002                        percpu_counter_sum(&fs_info->delalloc_bytes));
4003         }
4004
4005         btrfs_sysfs_remove_mounted(fs_info);
4006         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4007
4008         btrfs_free_fs_roots(fs_info);
4009
4010         btrfs_put_block_group_cache(fs_info);
4011
4012         /*
4013          * we must make sure there is not any read request to
4014          * submit after we stopping all workers.
4015          */
4016         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4017         btrfs_stop_all_workers(fs_info);
4018
4019         btrfs_free_block_groups(fs_info);
4020
4021         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4022         free_root_pointers(fs_info, 1);
4023
4024         iput(fs_info->btree_inode);
4025
4026 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4027         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4028                 btrfsic_unmount(fs_info->fs_devices);
4029 #endif
4030
4031         btrfs_close_devices(fs_info->fs_devices);
4032         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4033
4034         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4035         percpu_counter_destroy(&fs_info->delalloc_bytes);
4036         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4037         cleanup_srcu_struct(&fs_info->subvol_srcu);
4038
4039         btrfs_free_stripe_hash_table(fs_info);
4040         btrfs_free_ref_cache(fs_info);
4041
4042         while (!list_empty(&fs_info->pinned_chunks)) {
4043                 struct extent_map *em;
4044
4045                 em = list_first_entry(&fs_info->pinned_chunks,
4046                                       struct extent_map, list);
4047                 list_del_init(&em->list);
4048                 free_extent_map(em);
4049         }
4050 }
4051
4052 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4053                           int atomic)
4054 {
4055         int ret;
4056         struct inode *btree_inode = buf->pages[0]->mapping->host;
4057
4058         ret = extent_buffer_uptodate(buf);
4059         if (!ret)
4060                 return ret;
4061
4062         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4063                                     parent_transid, atomic);
4064         if (ret == -EAGAIN)
4065                 return ret;
4066         return !ret;
4067 }
4068
4069 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4070 {
4071         struct btrfs_fs_info *fs_info;
4072         struct btrfs_root *root;
4073         u64 transid = btrfs_header_generation(buf);
4074         int was_dirty;
4075
4076 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4077         /*
4078          * This is a fast path so only do this check if we have sanity tests
4079          * enabled.  Normal people shouldn't be using umapped buffers as dirty
4080          * outside of the sanity tests.
4081          */
4082         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4083                 return;
4084 #endif
4085         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4086         fs_info = root->fs_info;
4087         btrfs_assert_tree_locked(buf);
4088         if (transid != fs_info->generation)
4089                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4090                         buf->start, transid, fs_info->generation);
4091         was_dirty = set_extent_buffer_dirty(buf);
4092         if (!was_dirty)
4093                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4094                                          buf->len,
4095                                          fs_info->dirty_metadata_batch);
4096 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4097         /*
4098          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4099          * but item data not updated.
4100          * So here we should only check item pointers, not item data.
4101          */
4102         if (btrfs_header_level(buf) == 0 &&
4103             btrfs_check_leaf_relaxed(fs_info, buf)) {
4104                 btrfs_print_leaf(buf);
4105                 ASSERT(0);
4106         }
4107 #endif
4108 }
4109
4110 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4111                                         int flush_delayed)
4112 {
4113         /*
4114          * looks as though older kernels can get into trouble with
4115          * this code, they end up stuck in balance_dirty_pages forever
4116          */
4117         int ret;
4118
4119         if (current->flags & PF_MEMALLOC)
4120                 return;
4121
4122         if (flush_delayed)
4123                 btrfs_balance_delayed_items(fs_info);
4124
4125         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4126                                      BTRFS_DIRTY_METADATA_THRESH,
4127                                      fs_info->dirty_metadata_batch);
4128         if (ret > 0) {
4129                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4130         }
4131 }
4132
4133 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4134 {
4135         __btrfs_btree_balance_dirty(fs_info, 1);
4136 }
4137
4138 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4139 {
4140         __btrfs_btree_balance_dirty(fs_info, 0);
4141 }
4142
4143 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4144                       struct btrfs_key *first_key)
4145 {
4146         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4147         struct btrfs_fs_info *fs_info = root->fs_info;
4148
4149         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4150                                               level, first_key);
4151 }
4152
4153 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4154 {
4155         /* cleanup FS via transaction */
4156         btrfs_cleanup_transaction(fs_info);
4157
4158         mutex_lock(&fs_info->cleaner_mutex);
4159         btrfs_run_delayed_iputs(fs_info);
4160         mutex_unlock(&fs_info->cleaner_mutex);
4161
4162         down_write(&fs_info->cleanup_work_sem);
4163         up_write(&fs_info->cleanup_work_sem);
4164 }
4165
4166 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4167 {
4168         struct btrfs_ordered_extent *ordered;
4169
4170         spin_lock(&root->ordered_extent_lock);
4171         /*
4172          * This will just short circuit the ordered completion stuff which will
4173          * make sure the ordered extent gets properly cleaned up.
4174          */
4175         list_for_each_entry(ordered, &root->ordered_extents,
4176                             root_extent_list)
4177                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4178         spin_unlock(&root->ordered_extent_lock);
4179 }
4180
4181 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4182 {
4183         struct btrfs_root *root;
4184         struct list_head splice;
4185
4186         INIT_LIST_HEAD(&splice);
4187
4188         spin_lock(&fs_info->ordered_root_lock);
4189         list_splice_init(&fs_info->ordered_roots, &splice);
4190         while (!list_empty(&splice)) {
4191                 root = list_first_entry(&splice, struct btrfs_root,
4192                                         ordered_root);
4193                 list_move_tail(&root->ordered_root,
4194                                &fs_info->ordered_roots);
4195
4196                 spin_unlock(&fs_info->ordered_root_lock);
4197                 btrfs_destroy_ordered_extents(root);
4198
4199                 cond_resched();
4200                 spin_lock(&fs_info->ordered_root_lock);
4201         }
4202         spin_unlock(&fs_info->ordered_root_lock);
4203 }
4204
4205 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4206                                       struct btrfs_fs_info *fs_info)
4207 {
4208         struct rb_node *node;
4209         struct btrfs_delayed_ref_root *delayed_refs;
4210         struct btrfs_delayed_ref_node *ref;
4211         int ret = 0;
4212
4213         delayed_refs = &trans->delayed_refs;
4214
4215         spin_lock(&delayed_refs->lock);
4216         if (atomic_read(&delayed_refs->num_entries) == 0) {
4217                 spin_unlock(&delayed_refs->lock);
4218                 btrfs_info(fs_info, "delayed_refs has NO entry");
4219                 return ret;
4220         }
4221
4222         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4223                 struct btrfs_delayed_ref_head *head;
4224                 struct rb_node *n;
4225                 bool pin_bytes = false;
4226
4227                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4228                                 href_node);
4229                 if (!mutex_trylock(&head->mutex)) {
4230                         refcount_inc(&head->refs);
4231                         spin_unlock(&delayed_refs->lock);
4232
4233                         mutex_lock(&head->mutex);
4234                         mutex_unlock(&head->mutex);
4235                         btrfs_put_delayed_ref_head(head);
4236                         spin_lock(&delayed_refs->lock);
4237                         continue;
4238                 }
4239                 spin_lock(&head->lock);
4240                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4241                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4242                                        ref_node);
4243                         ref->in_tree = 0;
4244                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4245                         RB_CLEAR_NODE(&ref->ref_node);
4246                         if (!list_empty(&ref->add_list))
4247                                 list_del(&ref->add_list);
4248                         atomic_dec(&delayed_refs->num_entries);
4249                         btrfs_put_delayed_ref(ref);
4250                 }
4251                 if (head->must_insert_reserved)
4252                         pin_bytes = true;
4253                 btrfs_free_delayed_extent_op(head->extent_op);
4254                 delayed_refs->num_heads--;
4255                 if (head->processing == 0)
4256                         delayed_refs->num_heads_ready--;
4257                 atomic_dec(&delayed_refs->num_entries);
4258                 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
4259                 RB_CLEAR_NODE(&head->href_node);
4260                 spin_unlock(&head->lock);
4261                 spin_unlock(&delayed_refs->lock);
4262                 mutex_unlock(&head->mutex);
4263
4264                 if (pin_bytes)
4265                         btrfs_pin_extent(fs_info, head->bytenr,
4266                                          head->num_bytes, 1);
4267                 btrfs_put_delayed_ref_head(head);
4268                 cond_resched();
4269                 spin_lock(&delayed_refs->lock);
4270         }
4271
4272         spin_unlock(&delayed_refs->lock);
4273
4274         return ret;
4275 }
4276
4277 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4278 {
4279         struct btrfs_inode *btrfs_inode;
4280         struct list_head splice;
4281
4282         INIT_LIST_HEAD(&splice);
4283
4284         spin_lock(&root->delalloc_lock);
4285         list_splice_init(&root->delalloc_inodes, &splice);
4286
4287         while (!list_empty(&splice)) {
4288                 struct inode *inode = NULL;
4289                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4290                                                delalloc_inodes);
4291                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4292                 spin_unlock(&root->delalloc_lock);
4293
4294                 /*
4295                  * Make sure we get a live inode and that it'll not disappear
4296                  * meanwhile.
4297                  */
4298                 inode = igrab(&btrfs_inode->vfs_inode);
4299                 if (inode) {
4300                         invalidate_inode_pages2(inode->i_mapping);
4301                         iput(inode);
4302                 }
4303                 spin_lock(&root->delalloc_lock);
4304         }
4305         spin_unlock(&root->delalloc_lock);
4306 }
4307
4308 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4309 {
4310         struct btrfs_root *root;
4311         struct list_head splice;
4312
4313         INIT_LIST_HEAD(&splice);
4314
4315         spin_lock(&fs_info->delalloc_root_lock);
4316         list_splice_init(&fs_info->delalloc_roots, &splice);
4317         while (!list_empty(&splice)) {
4318                 root = list_first_entry(&splice, struct btrfs_root,
4319                                          delalloc_root);
4320                 root = btrfs_grab_fs_root(root);
4321                 BUG_ON(!root);
4322                 spin_unlock(&fs_info->delalloc_root_lock);
4323
4324                 btrfs_destroy_delalloc_inodes(root);
4325                 btrfs_put_fs_root(root);
4326
4327                 spin_lock(&fs_info->delalloc_root_lock);
4328         }
4329         spin_unlock(&fs_info->delalloc_root_lock);
4330 }
4331
4332 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4333                                         struct extent_io_tree *dirty_pages,
4334                                         int mark)
4335 {
4336         int ret;
4337         struct extent_buffer *eb;
4338         u64 start = 0;
4339         u64 end;
4340
4341         while (1) {
4342                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4343                                             mark, NULL);
4344                 if (ret)
4345                         break;
4346
4347                 clear_extent_bits(dirty_pages, start, end, mark);
4348                 while (start <= end) {
4349                         eb = find_extent_buffer(fs_info, start);
4350                         start += fs_info->nodesize;
4351                         if (!eb)
4352                                 continue;
4353                         wait_on_extent_buffer_writeback(eb);
4354
4355                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4356                                                &eb->bflags))
4357                                 clear_extent_buffer_dirty(eb);
4358                         free_extent_buffer_stale(eb);
4359                 }
4360         }
4361
4362         return ret;
4363 }
4364
4365 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4366                                        struct extent_io_tree *pinned_extents)
4367 {
4368         struct extent_io_tree *unpin;
4369         u64 start;
4370         u64 end;
4371         int ret;
4372         bool loop = true;
4373
4374         unpin = pinned_extents;
4375 again:
4376         while (1) {
4377                 struct extent_state *cached_state = NULL;
4378
4379                 /*
4380                  * The btrfs_finish_extent_commit() may get the same range as
4381                  * ours between find_first_extent_bit and clear_extent_dirty.
4382                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4383                  * the same extent range.
4384                  */
4385                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4386                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4387                                             EXTENT_DIRTY, &cached_state);
4388                 if (ret) {
4389                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4390                         break;
4391                 }
4392
4393                 clear_extent_dirty(unpin, start, end, &cached_state);
4394                 free_extent_state(cached_state);
4395                 btrfs_error_unpin_extent_range(fs_info, start, end);
4396                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4397                 cond_resched();
4398         }
4399
4400         if (loop) {
4401                 if (unpin == &fs_info->freed_extents[0])
4402                         unpin = &fs_info->freed_extents[1];
4403                 else
4404                         unpin = &fs_info->freed_extents[0];
4405                 loop = false;
4406                 goto again;
4407         }
4408
4409         return 0;
4410 }
4411
4412 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4413 {
4414         struct inode *inode;
4415
4416         inode = cache->io_ctl.inode;
4417         if (inode) {
4418                 invalidate_inode_pages2(inode->i_mapping);
4419                 BTRFS_I(inode)->generation = 0;
4420                 cache->io_ctl.inode = NULL;
4421                 iput(inode);
4422         }
4423         btrfs_put_block_group(cache);
4424 }
4425
4426 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4427                              struct btrfs_fs_info *fs_info)
4428 {
4429         struct btrfs_block_group_cache *cache;
4430
4431         spin_lock(&cur_trans->dirty_bgs_lock);
4432         while (!list_empty(&cur_trans->dirty_bgs)) {
4433                 cache = list_first_entry(&cur_trans->dirty_bgs,
4434                                          struct btrfs_block_group_cache,
4435                                          dirty_list);
4436
4437                 if (!list_empty(&cache->io_list)) {
4438                         spin_unlock(&cur_trans->dirty_bgs_lock);
4439                         list_del_init(&cache->io_list);
4440                         btrfs_cleanup_bg_io(cache);
4441                         spin_lock(&cur_trans->dirty_bgs_lock);
4442                 }
4443
4444                 list_del_init(&cache->dirty_list);
4445                 spin_lock(&cache->lock);
4446                 cache->disk_cache_state = BTRFS_DC_ERROR;
4447                 spin_unlock(&cache->lock);
4448
4449                 spin_unlock(&cur_trans->dirty_bgs_lock);
4450                 btrfs_put_block_group(cache);
4451                 spin_lock(&cur_trans->dirty_bgs_lock);
4452         }
4453         spin_unlock(&cur_trans->dirty_bgs_lock);
4454
4455         /*
4456          * Refer to the definition of io_bgs member for details why it's safe
4457          * to use it without any locking
4458          */
4459         while (!list_empty(&cur_trans->io_bgs)) {
4460                 cache = list_first_entry(&cur_trans->io_bgs,
4461                                          struct btrfs_block_group_cache,
4462                                          io_list);
4463
4464                 list_del_init(&cache->io_list);
4465                 spin_lock(&cache->lock);
4466                 cache->disk_cache_state = BTRFS_DC_ERROR;
4467                 spin_unlock(&cache->lock);
4468                 btrfs_cleanup_bg_io(cache);
4469         }
4470 }
4471
4472 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4473                                    struct btrfs_fs_info *fs_info)
4474 {
4475         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4476         ASSERT(list_empty(&cur_trans->dirty_bgs));
4477         ASSERT(list_empty(&cur_trans->io_bgs));
4478
4479         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4480
4481         cur_trans->state = TRANS_STATE_COMMIT_START;
4482         wake_up(&fs_info->transaction_blocked_wait);
4483
4484         cur_trans->state = TRANS_STATE_UNBLOCKED;
4485         wake_up(&fs_info->transaction_wait);
4486
4487         btrfs_destroy_delayed_inodes(fs_info);
4488         btrfs_assert_delayed_root_empty(fs_info);
4489
4490         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4491                                      EXTENT_DIRTY);
4492         btrfs_destroy_pinned_extent(fs_info,
4493                                     fs_info->pinned_extents);
4494
4495         cur_trans->state =TRANS_STATE_COMPLETED;
4496         wake_up(&cur_trans->commit_wait);
4497 }
4498
4499 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4500 {
4501         struct btrfs_transaction *t;
4502
4503         mutex_lock(&fs_info->transaction_kthread_mutex);
4504
4505         spin_lock(&fs_info->trans_lock);
4506         while (!list_empty(&fs_info->trans_list)) {
4507                 t = list_first_entry(&fs_info->trans_list,
4508                                      struct btrfs_transaction, list);
4509                 if (t->state >= TRANS_STATE_COMMIT_START) {
4510                         refcount_inc(&t->use_count);
4511                         spin_unlock(&fs_info->trans_lock);
4512                         btrfs_wait_for_commit(fs_info, t->transid);
4513                         btrfs_put_transaction(t);
4514                         spin_lock(&fs_info->trans_lock);
4515                         continue;
4516                 }
4517                 if (t == fs_info->running_transaction) {
4518                         t->state = TRANS_STATE_COMMIT_DOING;
4519                         spin_unlock(&fs_info->trans_lock);
4520                         /*
4521                          * We wait for 0 num_writers since we don't hold a trans
4522                          * handle open currently for this transaction.
4523                          */
4524                         wait_event(t->writer_wait,
4525                                    atomic_read(&t->num_writers) == 0);
4526                 } else {
4527                         spin_unlock(&fs_info->trans_lock);
4528                 }
4529                 btrfs_cleanup_one_transaction(t, fs_info);
4530
4531                 spin_lock(&fs_info->trans_lock);
4532                 if (t == fs_info->running_transaction)
4533                         fs_info->running_transaction = NULL;
4534                 list_del_init(&t->list);
4535                 spin_unlock(&fs_info->trans_lock);
4536
4537                 btrfs_put_transaction(t);
4538                 trace_btrfs_transaction_commit(fs_info->tree_root);
4539                 spin_lock(&fs_info->trans_lock);
4540         }
4541         spin_unlock(&fs_info->trans_lock);
4542         btrfs_destroy_all_ordered_extents(fs_info);
4543         btrfs_destroy_delayed_inodes(fs_info);
4544         btrfs_assert_delayed_root_empty(fs_info);
4545         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4546         btrfs_destroy_all_delalloc_inodes(fs_info);
4547         mutex_unlock(&fs_info->transaction_kthread_mutex);
4548
4549         return 0;
4550 }
4551
4552 static const struct extent_io_ops btree_extent_io_ops = {
4553         /* mandatory callbacks */
4554         .submit_bio_hook = btree_submit_bio_hook,
4555         .readpage_end_io_hook = btree_readpage_end_io_hook,
4556 };