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/scatterlist.h>
   9 #include <linux/swap.h>
  10 #include <linux/radix-tree.h>
  11 #include <linux/writeback.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/workqueue.h>
  14 #include <linux/kthread.h>
  15 #include <linux/slab.h>
  16 #include <linux/migrate.h>
  17 #include <linux/ratelimit.h>
  18 #include <linux/uuid.h>
  19 #include <linux/semaphore.h>
  20 #include <linux/error-injection.h>
  21 #include <linux/crc32c.h>
  22 #include <asm/unaligned.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "btrfs_inode.h"
  27 #include "volumes.h"
  28 #include "print-tree.h"
  29 #include "locking.h"
  30 #include "tree-log.h"
  31 #include "free-space-cache.h"
  32 #include "free-space-tree.h"
  33 #include "inode-map.h"
  34 #include "check-integrity.h"
  35 #include "rcu-string.h"
  36 #include "dev-replace.h"
  37 #include "raid56.h"
  38 #include "sysfs.h"
  39 #include "qgroup.h"
  40 #include "compression.h"
  41 #include "tree-checker.h"
  42 #include "ref-verify.h"
  43
  44 #ifdef CONFIG_X86
  45 #include <asm/cpufeature.h>
  46 #endif
  47
  48 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  49                                  BTRFS_HEADER_FLAG_RELOC |\
  50                                  BTRFS_SUPER_FLAG_ERROR |\
  51                                  BTRFS_SUPER_FLAG_SEEDING |\
  52                                  BTRFS_SUPER_FLAG_METADUMP |\
  53                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  54
  55 static const struct extent_io_ops btree_extent_io_ops;
  56 static void end_workqueue_fn(struct btrfs_work *work);
  57 static void free_fs_root(struct btrfs_root *root);
  58 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  59 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  60                                       struct btrfs_fs_info *fs_info);
  61 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  62 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  63                                         struct extent_io_tree *dirty_pages,
  64                                         int mark);
  65 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  66                                        struct extent_io_tree *pinned_extents);
  67 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  68 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  69
  70 /*
  71  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  72  * is complete.  This is used during reads to verify checksums, and it is used
  73  * by writes to insert metadata for new file extents after IO is complete.
  74  */
  75 struct btrfs_end_io_wq {
  76         struct bio *bio;
  77         bio_end_io_t *end_io;
  78         void *private;
  79         struct btrfs_fs_info *info;
  80         blk_status_t status;
  81         enum btrfs_wq_endio_type metadata;
  82         struct btrfs_work work;
  83 };
  84
  85 static struct kmem_cache *btrfs_end_io_wq_cache;
  86
  87 int __init btrfs_end_io_wq_init(void)
  88 {
  89         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  90                                         sizeof(struct btrfs_end_io_wq),
  91                                         0,
  92                                         SLAB_MEM_SPREAD,
  93                                         NULL);
  94         if (!btrfs_end_io_wq_cache)
  95                 return -ENOMEM;
  96         return 0;
  97 }
  98
  99 void __cold btrfs_end_io_wq_exit(void)
 100 {
 101         kmem_cache_destroy(btrfs_end_io_wq_cache);
 102 }
 103
 104 /*
 105  * async submit bios are used to offload expensive checksumming
 106  * onto the worker threads.  They checksum file and metadata bios
 107  * just before they are sent down the IO stack.
 108  */
 109 struct async_submit_bio {
 110         void *private_data;
 111         struct btrfs_fs_info *fs_info;
 112         struct bio *bio;
 113         extent_submit_bio_start_t *submit_bio_start;
 114         extent_submit_bio_done_t *submit_bio_done;
 115         int mirror_num;
 116         unsigned long bio_flags;
 117         /*
 118          * bio_offset is optional, can be used if the pages in the bio
 119          * can't tell us where in the file the bio should go
 120          */
 121         u64 bio_offset;
 122         struct btrfs_work work;
 123         blk_status_t status;
 124 };
 125
 126 /*
 127  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 128  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 129  * the level the eb occupies in the tree.
 130  *
 131  * Different roots are used for different purposes and may nest inside each
 132  * other and they require separate keysets.  As lockdep keys should be
 133  * static, assign keysets according to the purpose of the root as indicated
 134  * by btrfs_root->objectid.  This ensures that all special purpose roots
 135  * have separate keysets.
 136  *
 137  * Lock-nesting across peer nodes is always done with the immediate parent
 138  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 139  * subclass to avoid triggering lockdep warning in such cases.
 140  *
 141  * The key is set by the readpage_end_io_hook after the buffer has passed
 142  * csum validation but before the pages are unlocked.  It is also set by
 143  * btrfs_init_new_buffer on freshly allocated blocks.
 144  *
 145  * We also add a check to make sure the highest level of the tree is the
 146  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 147  * needs update as well.
 148  */
 149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 150 # if BTRFS_MAX_LEVEL != 8
 151 #  error
 152 # endif
 153
 154 static struct btrfs_lockdep_keyset {
 155         u64                     id;             /* root objectid */
 156         const char              *name_stem;     /* lock name stem */
 157         char                    names[BTRFS_MAX_LEVEL + 1][20];
 158         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 159 } btrfs_lockdep_keysets[] = {
 160         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 161         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 162         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 163         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 164         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 165         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 166         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 167         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 168         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 169         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 170         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 171         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 172         { .id = 0,                              .name_stem = "tree"     },
 173 };
 174
 175 void __init btrfs_init_lockdep(void)
 176 {
 177         int i, j;
 178
 179         /* initialize lockdep class names */
 180         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 181                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 182
 183                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 184                         snprintf(ks->names[j], sizeof(ks->names[j]),
 185                                  "btrfs-%s-%02d", ks->name_stem, j);
 186         }
 187 }
 188
 189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 190                                     int level)
 191 {
 192         struct btrfs_lockdep_keyset *ks;
 193
 194         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 195
 196         /* find the matching keyset, id 0 is the default entry */
 197         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 198                 if (ks->id == objectid)
 199                         break;
 200
 201         lockdep_set_class_and_name(&eb->lock,
 202                                    &ks->keys[level], ks->names[level]);
 203 }
 204
 205 #endif
 206
 207 /*
 208  * extents on the btree inode are pretty simple, there's one extent
 209  * that covers the entire device
 210  */
 211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 212                 struct page *page, size_t pg_offset, u64 start, u64 len,
 213                 int create)
 214 {
 215         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 216         struct extent_map_tree *em_tree = &inode->extent_tree;
 217         struct extent_map *em;
 218         int ret;
 219
 220         read_lock(&em_tree->lock);
 221         em = lookup_extent_mapping(em_tree, start, len);
 222         if (em) {
 223                 em->bdev = fs_info->fs_devices->latest_bdev;
 224                 read_unlock(&em_tree->lock);
 225                 goto out;
 226         }
 227         read_unlock(&em_tree->lock);
 228
 229         em = alloc_extent_map();
 230         if (!em) {
 231                 em = ERR_PTR(-ENOMEM);
 232                 goto out;
 233         }
 234         em->start = 0;
 235         em->len = (u64)-1;
 236         em->block_len = (u64)-1;
 237         em->block_start = 0;
 238         em->bdev = fs_info->fs_devices->latest_bdev;
 239
 240         write_lock(&em_tree->lock);
 241         ret = add_extent_mapping(em_tree, em, 0);
 242         if (ret == -EEXIST) {
 243                 free_extent_map(em);
 244                 em = lookup_extent_mapping(em_tree, start, len);
 245                 if (!em)
 246                         em = ERR_PTR(-EIO);
 247         } else if (ret) {
 248                 free_extent_map(em);
 249                 em = ERR_PTR(ret);
 250         }
 251         write_unlock(&em_tree->lock);
 252
 253 out:
 254         return em;
 255 }
 256
 257 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 258 {
 259         return crc32c(seed, data, len);
 260 }
 261
 262 void btrfs_csum_final(u32 crc, u8 *result)
 263 {
 264         put_unaligned_le32(~crc, result);
 265 }
 266
 267 /*
 268  * compute the csum for a btree block, and either verify it or write it
 269  * into the csum field of the block.
 270  */
 271 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 272                            struct extent_buffer *buf,
 273                            int verify)
 274 {
 275         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 276         char result[BTRFS_CSUM_SIZE];
 277         unsigned long len;
 278         unsigned long cur_len;
 279         unsigned long offset = BTRFS_CSUM_SIZE;
 280         char *kaddr;
 281         unsigned long map_start;
 282         unsigned long map_len;
 283         int err;
 284         u32 crc = ~(u32)0;
 285
 286         len = buf->len - offset;
 287         while (len > 0) {
 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)
 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 key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
 457                           eb->start, first_key->objectid, first_key->type,
 458                           first_key->offset, found_key.objectid,
 459                           found_key.type, found_key.offset);
 460         }
 461 #endif
 462         return ret;
 463 }
 464
 465 /*
 466  * helper to read a given tree block, doing retries as required when
 467  * the checksums don't match and we have alternate mirrors to try.
 468  *
 469  * @parent_transid:     expected transid, skip check if 0
 470  * @level:              expected level, mandatory check
 471  * @first_key:          expected key of first slot, skip check if NULL
 472  */
 473 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 474                                           struct extent_buffer *eb,
 475                                           u64 parent_transid, int level,
 476                                           struct btrfs_key *first_key)
 477 {
 478         struct extent_io_tree *io_tree;
 479         int failed = 0;
 480         int ret;
 481         int num_copies = 0;
 482         int mirror_num = 0;
 483         int failed_mirror = 0;
 484
 485         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 486         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 487         while (1) {
 488                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 489                                                mirror_num);
 490                 if (!ret) {
 491                         if (verify_parent_transid(io_tree, eb,
 492                                                    parent_transid, 0))
 493                                 ret = -EIO;
 494                         else if (verify_level_key(fs_info, eb, level,
 495                                                   first_key))
 496                                 ret = -EUCLEAN;
 497                         else
 498                                 break;
 499                 }
 500
 501                 /*
 502                  * This buffer's crc is fine, but its contents are corrupted, so
 503                  * there is no reason to read the other copies, they won't be
 504                  * any less wrong.
 505                  */
 506                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
 507                     ret == -EUCLEAN)
 508                         break;
 509
 510                 num_copies = btrfs_num_copies(fs_info,
 511                                               eb->start, eb->len);
 512                 if (num_copies == 1)
 513                         break;
 514
 515                 if (!failed_mirror) {
 516                         failed = 1;
 517                         failed_mirror = eb->read_mirror;
 518                 }
 519
 520                 mirror_num++;
 521                 if (mirror_num == failed_mirror)
 522                         mirror_num++;
 523
 524                 if (mirror_num > num_copies)
 525                         break;
 526         }
 527
 528         if (failed && !ret && failed_mirror)
 529                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 530
 531         return ret;
 532 }
 533
 534 /*
 535  * checksum a dirty tree block before IO.  This has extra checks to make sure
 536  * we only fill in the checksum field in the first page of a multi-page block
 537  */
 538
 539 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 540 {
 541         u64 start = page_offset(page);
 542         u64 found_start;
 543         struct extent_buffer *eb;
 544
 545         eb = (struct extent_buffer *)page->private;
 546         if (page != eb->pages[0])
 547                 return 0;
 548
 549         found_start = btrfs_header_bytenr(eb);
 550         /*
 551          * Please do not consolidate these warnings into a single if.
 552          * It is useful to know what went wrong.
 553          */
 554         if (WARN_ON(found_start != start))
 555                 return -EUCLEAN;
 556         if (WARN_ON(!PageUptodate(page)))
 557                 return -EUCLEAN;
 558
 559         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 560                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 561
 562         return csum_tree_block(fs_info, eb, 0);
 563 }
 564
 565 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 566                                  struct extent_buffer *eb)
 567 {
 568         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 569         u8 fsid[BTRFS_FSID_SIZE];
 570         int ret = 1;
 571
 572         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 573         while (fs_devices) {
 574                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 575                         ret = 0;
 576                         break;
 577                 }
 578                 fs_devices = fs_devices->seed;
 579         }
 580         return ret;
 581 }
 582
 583 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 584                                       u64 phy_offset, struct page *page,
 585                                       u64 start, u64 end, int mirror)
 586 {
 587         u64 found_start;
 588         int found_level;
 589         struct extent_buffer *eb;
 590         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 591         struct btrfs_fs_info *fs_info = root->fs_info;
 592         int ret = 0;
 593         int reads_done;
 594
 595         if (!page->private)
 596                 goto out;
 597
 598         eb = (struct extent_buffer *)page->private;
 599
 600         /* the pending IO might have been the only thing that kept this buffer
 601          * in memory.  Make sure we have a ref for all this other checks
 602          */
 603         extent_buffer_get(eb);
 604
 605         reads_done = atomic_dec_and_test(&eb->io_pages);
 606         if (!reads_done)
 607                 goto err;
 608
 609         eb->read_mirror = mirror;
 610         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 611                 ret = -EIO;
 612                 goto err;
 613         }
 614
 615         found_start = btrfs_header_bytenr(eb);
 616         if (found_start != eb->start) {
 617                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 618                              found_start, eb->start);
 619                 ret = -EIO;
 620                 goto err;
 621         }
 622         if (check_tree_block_fsid(fs_info, eb)) {
 623                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 624                              eb->start);
 625                 ret = -EIO;
 626                 goto err;
 627         }
 628         found_level = btrfs_header_level(eb);
 629         if (found_level >= BTRFS_MAX_LEVEL) {
 630                 btrfs_err(fs_info, "bad tree block level %d",
 631                           (int)btrfs_header_level(eb));
 632                 ret = -EIO;
 633                 goto err;
 634         }
 635
 636         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 637                                        eb, found_level);
 638
 639         ret = csum_tree_block(fs_info, eb, 1);
 640         if (ret)
 641                 goto err;
 642
 643         /*
 644          * If this is a leaf block and it is corrupt, set the corrupt bit so
 645          * that we don't try and read the other copies of this block, just
 646          * return -EIO.
 647          */
 648         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 649                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 650                 ret = -EIO;
 651         }
 652
 653         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 654                 ret = -EIO;
 655
 656         if (!ret)
 657                 set_extent_buffer_uptodate(eb);
 658 err:
 659         if (reads_done &&
 660             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 661                 btree_readahead_hook(eb, ret);
 662
 663         if (ret) {
 664                 /*
 665                  * our io error hook is going to dec the io pages
 666                  * again, we have to make sure it has something
 667                  * to decrement
 668                  */
 669                 atomic_inc(&eb->io_pages);
 670                 clear_extent_buffer_uptodate(eb);
 671         }
 672         free_extent_buffer(eb);
 673 out:
 674         return ret;
 675 }
 676
 677 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 678 {
 679         struct extent_buffer *eb;
 680
 681         eb = (struct extent_buffer *)page->private;
 682         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 683         eb->read_mirror = failed_mirror;
 684         atomic_dec(&eb->io_pages);
 685         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 686                 btree_readahead_hook(eb, -EIO);
 687         return -EIO;    /* we fixed nothing */
 688 }
 689
 690 static void end_workqueue_bio(struct bio *bio)
 691 {
 692         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 693         struct btrfs_fs_info *fs_info;
 694         struct btrfs_workqueue *wq;
 695         btrfs_work_func_t func;
 696
 697         fs_info = end_io_wq->info;
 698         end_io_wq->status = bio->bi_status;
 699
 700         if (bio_op(bio) == REQ_OP_WRITE) {
 701                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 702                         wq = fs_info->endio_meta_write_workers;
 703                         func = btrfs_endio_meta_write_helper;
 704                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 705                         wq = fs_info->endio_freespace_worker;
 706                         func = btrfs_freespace_write_helper;
 707                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 708                         wq = fs_info->endio_raid56_workers;
 709                         func = btrfs_endio_raid56_helper;
 710                 } else {
 711                         wq = fs_info->endio_write_workers;
 712                         func = btrfs_endio_write_helper;
 713                 }
 714         } else {
 715                 if (unlikely(end_io_wq->metadata ==
 716                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 717                         wq = fs_info->endio_repair_workers;
 718                         func = btrfs_endio_repair_helper;
 719                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 720                         wq = fs_info->endio_raid56_workers;
 721                         func = btrfs_endio_raid56_helper;
 722                 } else if (end_io_wq->metadata) {
 723                         wq = fs_info->endio_meta_workers;
 724                         func = btrfs_endio_meta_helper;
 725                 } else {
 726                         wq = fs_info->endio_workers;
 727                         func = btrfs_endio_helper;
 728                 }
 729         }
 730
 731         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 732         btrfs_queue_work(wq, &end_io_wq->work);
 733 }
 734
 735 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 736                         enum btrfs_wq_endio_type metadata)
 737 {
 738         struct btrfs_end_io_wq *end_io_wq;
 739
 740         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 741         if (!end_io_wq)
 742                 return BLK_STS_RESOURCE;
 743
 744         end_io_wq->private = bio->bi_private;
 745         end_io_wq->end_io = bio->bi_end_io;
 746         end_io_wq->info = info;
 747         end_io_wq->status = 0;
 748         end_io_wq->bio = bio;
 749         end_io_wq->metadata = metadata;
 750
 751         bio->bi_private = end_io_wq;
 752         bio->bi_end_io = end_workqueue_bio;
 753         return 0;
 754 }
 755
 756 static void run_one_async_start(struct btrfs_work *work)
 757 {
 758         struct async_submit_bio *async;
 759         blk_status_t ret;
 760
 761         async = container_of(work, struct  async_submit_bio, work);
 762         ret = async->submit_bio_start(async->private_data, async->bio,
 763                                       async->bio_offset);
 764         if (ret)
 765                 async->status = ret;
 766 }
 767
 768 static void run_one_async_done(struct btrfs_work *work)
 769 {
 770         struct async_submit_bio *async;
 771
 772         async = container_of(work, struct  async_submit_bio, work);
 773
 774         /* If an error occurred we just want to clean up the bio and move on */
 775         if (async->status) {
 776                 async->bio->bi_status = async->status;
 777                 bio_endio(async->bio);
 778                 return;
 779         }
 780
 781         async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
 782 }
 783
 784 static void run_one_async_free(struct btrfs_work *work)
 785 {
 786         struct async_submit_bio *async;
 787
 788         async = container_of(work, struct  async_submit_bio, work);
 789         kfree(async);
 790 }
 791
 792 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 793                                  int mirror_num, unsigned long bio_flags,
 794                                  u64 bio_offset, void *private_data,
 795                                  extent_submit_bio_start_t *submit_bio_start,
 796                                  extent_submit_bio_done_t *submit_bio_done)
 797 {
 798         struct async_submit_bio *async;
 799
 800         async = kmalloc(sizeof(*async), GFP_NOFS);
 801         if (!async)
 802                 return BLK_STS_RESOURCE;
 803
 804         async->private_data = private_data;
 805         async->fs_info = fs_info;
 806         async->bio = bio;
 807         async->mirror_num = mirror_num;
 808         async->submit_bio_start = submit_bio_start;
 809         async->submit_bio_done = submit_bio_done;
 810
 811         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 812                         run_one_async_done, run_one_async_free);
 813
 814         async->bio_flags = bio_flags;
 815         async->bio_offset = bio_offset;
 816
 817         async->status = 0;
 818
 819         if (op_is_sync(bio->bi_opf))
 820                 btrfs_set_work_high_priority(&async->work);
 821
 822         btrfs_queue_work(fs_info->workers, &async->work);
 823         return 0;
 824 }
 825
 826 static blk_status_t btree_csum_one_bio(struct bio *bio)
 827 {
 828         struct bio_vec *bvec;
 829         struct btrfs_root *root;
 830         int i, ret = 0;
 831
 832         ASSERT(!bio_flagged(bio, BIO_CLONED));
 833         bio_for_each_segment_all(bvec, bio, i) {
 834                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 835                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 836                 if (ret)
 837                         break;
 838         }
 839
 840         return errno_to_blk_status(ret);
 841 }
 842
 843 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 844                                              u64 bio_offset)
 845 {
 846         /*
 847          * when we're called for a write, we're already in the async
 848          * submission context.  Just jump into btrfs_map_bio
 849          */
 850         return btree_csum_one_bio(bio);
 851 }
 852
 853 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
 854                                             int mirror_num)
 855 {
 856         struct inode *inode = private_data;
 857         blk_status_t ret;
 858
 859         /*
 860          * when we're called for a write, we're already in the async
 861          * submission context.  Just jump into btrfs_map_bio
 862          */
 863         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
 864         if (ret) {
 865                 bio->bi_status = ret;
 866                 bio_endio(bio);
 867         }
 868         return ret;
 869 }
 870
 871 static int check_async_write(struct btrfs_inode *bi)
 872 {
 873         if (atomic_read(&bi->sync_writers))
 874                 return 0;
 875 #ifdef CONFIG_X86
 876         if (static_cpu_has(X86_FEATURE_XMM4_2))
 877                 return 0;
 878 #endif
 879         return 1;
 880 }
 881
 882 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 883                                           int mirror_num, unsigned long bio_flags,
 884                                           u64 bio_offset)
 885 {
 886         struct inode *inode = private_data;
 887         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 888         int async = check_async_write(BTRFS_I(inode));
 889         blk_status_t ret;
 890
 891         if (bio_op(bio) != REQ_OP_WRITE) {
 892                 /*
 893                  * called for a read, do the setup so that checksum validation
 894                  * can happen in the async kernel threads
 895                  */
 896                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 897                                           BTRFS_WQ_ENDIO_METADATA);
 898                 if (ret)
 899                         goto out_w_error;
 900                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 901         } else if (!async) {
 902                 ret = btree_csum_one_bio(bio);
 903                 if (ret)
 904                         goto out_w_error;
 905                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 906         } else {
 907                 /*
 908                  * kthread helpers are used to submit writes so that
 909                  * checksumming can happen in parallel across all CPUs
 910                  */
 911                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 912                                           bio_offset, private_data,
 913                                           btree_submit_bio_start,
 914                                           btree_submit_bio_done);
 915         }
 916
 917         if (ret)
 918                 goto out_w_error;
 919         return 0;
 920
 921 out_w_error:
 922         bio->bi_status = ret;
 923         bio_endio(bio);
 924         return ret;
 925 }
 926
 927 #ifdef CONFIG_MIGRATION
 928 static int btree_migratepage(struct address_space *mapping,
 929                         struct page *newpage, struct page *page,
 930                         enum migrate_mode mode)
 931 {
 932         /*
 933          * we can't safely write a btree page from here,
 934          * we haven't done the locking hook
 935          */
 936         if (PageDirty(page))
 937                 return -EAGAIN;
 938         /*
 939          * Buffers may be managed in a filesystem specific way.
 940          * We must have no buffers or drop them.
 941          */
 942         if (page_has_private(page) &&
 943             !try_to_release_page(page, GFP_KERNEL))
 944                 return -EAGAIN;
 945         return migrate_page(mapping, newpage, page, mode);
 946 }
 947 #endif
 948
 949
 950 static int btree_writepages(struct address_space *mapping,
 951                             struct writeback_control *wbc)
 952 {
 953         struct btrfs_fs_info *fs_info;
 954         int ret;
 955
 956         if (wbc->sync_mode == WB_SYNC_NONE) {
 957
 958                 if (wbc->for_kupdate)
 959                         return 0;
 960
 961                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 962                 /* this is a bit racy, but that's ok */
 963                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 964                                              BTRFS_DIRTY_METADATA_THRESH);
 965                 if (ret < 0)
 966                         return 0;
 967         }
 968         return btree_write_cache_pages(mapping, wbc);
 969 }
 970
 971 static int btree_readpage(struct file *file, struct page *page)
 972 {
 973         struct extent_io_tree *tree;
 974         tree = &BTRFS_I(page->mapping->host)->io_tree;
 975         return extent_read_full_page(tree, page, btree_get_extent, 0);
 976 }
 977
 978 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 979 {
 980         if (PageWriteback(page) || PageDirty(page))
 981                 return 0;
 982
 983         return try_release_extent_buffer(page);
 984 }
 985
 986 static void btree_invalidatepage(struct page *page, unsigned int offset,
 987                                  unsigned int length)
 988 {
 989         struct extent_io_tree *tree;
 990         tree = &BTRFS_I(page->mapping->host)->io_tree;
 991         extent_invalidatepage(tree, page, offset);
 992         btree_releasepage(page, GFP_NOFS);
 993         if (PagePrivate(page)) {
 994                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 995                            "page private not zero on page %llu",
 996                            (unsigned long long)page_offset(page));
 997                 ClearPagePrivate(page);
 998                 set_page_private(page, 0);
 999                 put_page(page);
1000         }
1001 }
1002
1003 static int btree_set_page_dirty(struct page *page)
1004 {
1005 #ifdef DEBUG
1006         struct extent_buffer *eb;
1007
1008         BUG_ON(!PagePrivate(page));
1009         eb = (struct extent_buffer *)page->private;
1010         BUG_ON(!eb);
1011         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1012         BUG_ON(!atomic_read(&eb->refs));
1013         btrfs_assert_tree_locked(eb);
1014 #endif
1015         return __set_page_dirty_nobuffers(page);
1016 }
1017
1018 static const struct address_space_operations btree_aops = {
1019         .readpage       = btree_readpage,
1020         .writepages     = btree_writepages,
1021         .releasepage    = btree_releasepage,
1022         .invalidatepage = btree_invalidatepage,
1023 #ifdef CONFIG_MIGRATION
1024         .migratepage    = btree_migratepage,
1025 #endif
1026         .set_page_dirty = btree_set_page_dirty,
1027 };
1028
1029 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1030 {
1031         struct extent_buffer *buf = NULL;
1032         struct inode *btree_inode = fs_info->btree_inode;
1033
1034         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1035         if (IS_ERR(buf))
1036                 return;
1037         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1038                                  buf, WAIT_NONE, 0);
1039         free_extent_buffer(buf);
1040 }
1041
1042 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1043                          int mirror_num, struct extent_buffer **eb)
1044 {
1045         struct extent_buffer *buf = NULL;
1046         struct inode *btree_inode = fs_info->btree_inode;
1047         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1048         int ret;
1049
1050         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1051         if (IS_ERR(buf))
1052                 return 0;
1053
1054         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1055
1056         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1057                                        mirror_num);
1058         if (ret) {
1059                 free_extent_buffer(buf);
1060                 return ret;
1061         }
1062
1063         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1064                 free_extent_buffer(buf);
1065                 return -EIO;
1066         } else if (extent_buffer_uptodate(buf)) {
1067                 *eb = buf;
1068         } else {
1069                 free_extent_buffer(buf);
1070         }
1071         return 0;
1072 }
1073
1074 struct extent_buffer *btrfs_find_create_tree_block(
1075                                                 struct btrfs_fs_info *fs_info,
1076                                                 u64 bytenr)
1077 {
1078         if (btrfs_is_testing(fs_info))
1079                 return alloc_test_extent_buffer(fs_info, bytenr);
1080         return alloc_extent_buffer(fs_info, bytenr);
1081 }
1082
1083
1084 int btrfs_write_tree_block(struct extent_buffer *buf)
1085 {
1086         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1087                                         buf->start + buf->len - 1);
1088 }
1089
1090 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1091 {
1092         filemap_fdatawait_range(buf->pages[0]->mapping,
1093                                 buf->start, buf->start + buf->len - 1);
1094 }
1095
1096 /*
1097  * Read tree block at logical address @bytenr and do variant basic but critical
1098  * verification.
1099  *
1100  * @parent_transid:     expected transid of this tree block, skip check if 0
1101  * @level:              expected level, mandatory check
1102  * @first_key:          expected key in slot 0, skip check if NULL
1103  */
1104 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1105                                       u64 parent_transid, int level,
1106                                       struct btrfs_key *first_key)
1107 {
1108         struct extent_buffer *buf = NULL;
1109         int ret;
1110
1111         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1112         if (IS_ERR(buf))
1113                 return buf;
1114
1115         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1116                                              level, first_key);
1117         if (ret) {
1118                 free_extent_buffer(buf);
1119                 return ERR_PTR(ret);
1120         }
1121         return buf;
1122
1123 }
1124
1125 void clean_tree_block(struct btrfs_fs_info *fs_info,
1126                       struct extent_buffer *buf)
1127 {
1128         if (btrfs_header_generation(buf) ==
1129             fs_info->running_transaction->transid) {
1130                 btrfs_assert_tree_locked(buf);
1131
1132                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1133                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1134                                                  -buf->len,
1135                                                  fs_info->dirty_metadata_batch);
1136                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1137                         btrfs_set_lock_blocking(buf);
1138                         clear_extent_buffer_dirty(buf);
1139                 }
1140         }
1141 }
1142
1143 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1144 {
1145         struct btrfs_subvolume_writers *writers;
1146         int ret;
1147
1148         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1149         if (!writers)
1150                 return ERR_PTR(-ENOMEM);
1151
1152         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1153         if (ret < 0) {
1154                 kfree(writers);
1155                 return ERR_PTR(ret);
1156         }
1157
1158         init_waitqueue_head(&writers->wait);
1159         return writers;
1160 }
1161
1162 static void
1163 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1164 {
1165         percpu_counter_destroy(&writers->counter);
1166         kfree(writers);
1167 }
1168
1169 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1170                          u64 objectid)
1171 {
1172         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1173         root->node = NULL;
1174         root->commit_root = NULL;
1175         root->state = 0;
1176         root->orphan_cleanup_state = 0;
1177
1178         root->objectid = objectid;
1179         root->last_trans = 0;
1180         root->highest_objectid = 0;
1181         root->nr_delalloc_inodes = 0;
1182         root->nr_ordered_extents = 0;
1183         root->name = NULL;
1184         root->inode_tree = RB_ROOT;
1185         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1186         root->block_rsv = NULL;
1187         root->orphan_block_rsv = NULL;
1188
1189         INIT_LIST_HEAD(&root->dirty_list);
1190         INIT_LIST_HEAD(&root->root_list);
1191         INIT_LIST_HEAD(&root->delalloc_inodes);
1192         INIT_LIST_HEAD(&root->delalloc_root);
1193         INIT_LIST_HEAD(&root->ordered_extents);
1194         INIT_LIST_HEAD(&root->ordered_root);
1195         INIT_LIST_HEAD(&root->logged_list[0]);
1196         INIT_LIST_HEAD(&root->logged_list[1]);
1197         spin_lock_init(&root->orphan_lock);
1198         spin_lock_init(&root->inode_lock);
1199         spin_lock_init(&root->delalloc_lock);
1200         spin_lock_init(&root->ordered_extent_lock);
1201         spin_lock_init(&root->accounting_lock);
1202         spin_lock_init(&root->log_extents_lock[0]);
1203         spin_lock_init(&root->log_extents_lock[1]);
1204         spin_lock_init(&root->qgroup_meta_rsv_lock);
1205         mutex_init(&root->objectid_mutex);
1206         mutex_init(&root->log_mutex);
1207         mutex_init(&root->ordered_extent_mutex);
1208         mutex_init(&root->delalloc_mutex);
1209         init_waitqueue_head(&root->log_writer_wait);
1210         init_waitqueue_head(&root->log_commit_wait[0]);
1211         init_waitqueue_head(&root->log_commit_wait[1]);
1212         INIT_LIST_HEAD(&root->log_ctxs[0]);
1213         INIT_LIST_HEAD(&root->log_ctxs[1]);
1214         atomic_set(&root->log_commit[0], 0);
1215         atomic_set(&root->log_commit[1], 0);
1216         atomic_set(&root->log_writers, 0);
1217         atomic_set(&root->log_batch, 0);
1218         atomic_set(&root->orphan_inodes, 0);
1219         refcount_set(&root->refs, 1);
1220         atomic_set(&root->will_be_snapshotted, 0);
1221         root->log_transid = 0;
1222         root->log_transid_committed = -1;
1223         root->last_log_commit = 0;
1224         if (!dummy)
1225                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1226
1227         memset(&root->root_key, 0, sizeof(root->root_key));
1228         memset(&root->root_item, 0, sizeof(root->root_item));
1229         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1230         if (!dummy)
1231                 root->defrag_trans_start = fs_info->generation;
1232         else
1233                 root->defrag_trans_start = 0;
1234         root->root_key.objectid = objectid;
1235         root->anon_dev = 0;
1236
1237         spin_lock_init(&root->root_item_lock);
1238 }
1239
1240 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1241                 gfp_t flags)
1242 {
1243         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1244         if (root)
1245                 root->fs_info = fs_info;
1246         return root;
1247 }
1248
1249 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1250 /* Should only be used by the testing infrastructure */
1251 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1252 {
1253         struct btrfs_root *root;
1254
1255         if (!fs_info)
1256                 return ERR_PTR(-EINVAL);
1257
1258         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1259         if (!root)
1260                 return ERR_PTR(-ENOMEM);
1261
1262         /* We don't use the stripesize in selftest, set it as sectorsize */
1263         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1264         root->alloc_bytenr = 0;
1265
1266         return root;
1267 }
1268 #endif
1269
1270 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1271                                      struct btrfs_fs_info *fs_info,
1272                                      u64 objectid)
1273 {
1274         struct extent_buffer *leaf;
1275         struct btrfs_root *tree_root = fs_info->tree_root;
1276         struct btrfs_root *root;
1277         struct btrfs_key key;
1278         int ret = 0;
1279         uuid_le uuid = NULL_UUID_LE;
1280
1281         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1282         if (!root)
1283                 return ERR_PTR(-ENOMEM);
1284
1285         __setup_root(root, fs_info, objectid);
1286         root->root_key.objectid = objectid;
1287         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1288         root->root_key.offset = 0;
1289
1290         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1291         if (IS_ERR(leaf)) {
1292                 ret = PTR_ERR(leaf);
1293                 leaf = NULL;
1294                 goto fail;
1295         }
1296
1297         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1298         btrfs_set_header_bytenr(leaf, leaf->start);
1299         btrfs_set_header_generation(leaf, trans->transid);
1300         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1301         btrfs_set_header_owner(leaf, objectid);
1302         root->node = leaf;
1303
1304         write_extent_buffer_fsid(leaf, fs_info->fsid);
1305         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1306         btrfs_mark_buffer_dirty(leaf);
1307
1308         root->commit_root = btrfs_root_node(root);
1309         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1310
1311         root->root_item.flags = 0;
1312         root->root_item.byte_limit = 0;
1313         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1314         btrfs_set_root_generation(&root->root_item, trans->transid);
1315         btrfs_set_root_level(&root->root_item, 0);
1316         btrfs_set_root_refs(&root->root_item, 1);
1317         btrfs_set_root_used(&root->root_item, leaf->len);
1318         btrfs_set_root_last_snapshot(&root->root_item, 0);
1319         btrfs_set_root_dirid(&root->root_item, 0);
1320         if (is_fstree(objectid))
1321                 uuid_le_gen(&uuid);
1322         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1323         root->root_item.drop_level = 0;
1324
1325         key.objectid = objectid;
1326         key.type = BTRFS_ROOT_ITEM_KEY;
1327         key.offset = 0;
1328         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1329         if (ret)
1330                 goto fail;
1331
1332         btrfs_tree_unlock(leaf);
1333
1334         return root;
1335
1336 fail:
1337         if (leaf) {
1338                 btrfs_tree_unlock(leaf);
1339                 free_extent_buffer(root->commit_root);
1340                 free_extent_buffer(leaf);
1341         }
1342         kfree(root);
1343
1344         return ERR_PTR(ret);
1345 }
1346
1347 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1348                                          struct btrfs_fs_info *fs_info)
1349 {
1350         struct btrfs_root *root;
1351         struct extent_buffer *leaf;
1352
1353         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1354         if (!root)
1355                 return ERR_PTR(-ENOMEM);
1356
1357         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1358
1359         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1360         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1361         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1362
1363         /*
1364          * DON'T set REF_COWS for log trees
1365          *
1366          * log trees do not get reference counted because they go away
1367          * before a real commit is actually done.  They do store pointers
1368          * to file data extents, and those reference counts still get
1369          * updated (along with back refs to the log tree).
1370          */
1371
1372         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1373                         NULL, 0, 0, 0);
1374         if (IS_ERR(leaf)) {
1375                 kfree(root);
1376                 return ERR_CAST(leaf);
1377         }
1378
1379         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1380         btrfs_set_header_bytenr(leaf, leaf->start);
1381         btrfs_set_header_generation(leaf, trans->transid);
1382         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1383         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1384         root->node = leaf;
1385
1386         write_extent_buffer_fsid(root->node, fs_info->fsid);
1387         btrfs_mark_buffer_dirty(root->node);
1388         btrfs_tree_unlock(root->node);
1389         return root;
1390 }
1391
1392 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1393                              struct btrfs_fs_info *fs_info)
1394 {
1395         struct btrfs_root *log_root;
1396
1397         log_root = alloc_log_tree(trans, fs_info);
1398         if (IS_ERR(log_root))
1399                 return PTR_ERR(log_root);
1400         WARN_ON(fs_info->log_root_tree);
1401         fs_info->log_root_tree = log_root;
1402         return 0;
1403 }
1404
1405 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1406                        struct btrfs_root *root)
1407 {
1408         struct btrfs_fs_info *fs_info = root->fs_info;
1409         struct btrfs_root *log_root;
1410         struct btrfs_inode_item *inode_item;
1411
1412         log_root = alloc_log_tree(trans, fs_info);
1413         if (IS_ERR(log_root))
1414                 return PTR_ERR(log_root);
1415
1416         log_root->last_trans = trans->transid;
1417         log_root->root_key.offset = root->root_key.objectid;
1418
1419         inode_item = &log_root->root_item.inode;
1420         btrfs_set_stack_inode_generation(inode_item, 1);
1421         btrfs_set_stack_inode_size(inode_item, 3);
1422         btrfs_set_stack_inode_nlink(inode_item, 1);
1423         btrfs_set_stack_inode_nbytes(inode_item,
1424                                      fs_info->nodesize);
1425         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1426
1427         btrfs_set_root_node(&log_root->root_item, log_root->node);
1428
1429         WARN_ON(root->log_root);
1430         root->log_root = log_root;
1431         root->log_transid = 0;
1432         root->log_transid_committed = -1;
1433         root->last_log_commit = 0;
1434         return 0;
1435 }
1436
1437 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1438                                                struct btrfs_key *key)
1439 {
1440         struct btrfs_root *root;
1441         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1442         struct btrfs_path *path;
1443         u64 generation;
1444         int ret;
1445         int level;
1446
1447         path = btrfs_alloc_path();
1448         if (!path)
1449                 return ERR_PTR(-ENOMEM);
1450
1451         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1452         if (!root) {
1453                 ret = -ENOMEM;
1454                 goto alloc_fail;
1455         }
1456
1457         __setup_root(root, fs_info, key->objectid);
1458
1459         ret = btrfs_find_root(tree_root, key, path,
1460                               &root->root_item, &root->root_key);
1461         if (ret) {
1462                 if (ret > 0)
1463                         ret = -ENOENT;
1464                 goto find_fail;
1465         }
1466
1467         generation = btrfs_root_generation(&root->root_item);
1468         level = btrfs_root_level(&root->root_item);
1469         root->node = read_tree_block(fs_info,
1470                                      btrfs_root_bytenr(&root->root_item),
1471                                      generation, level, NULL);
1472         if (IS_ERR(root->node)) {
1473                 ret = PTR_ERR(root->node);
1474                 goto find_fail;
1475         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1476                 ret = -EIO;
1477                 free_extent_buffer(root->node);
1478                 goto find_fail;
1479         }
1480         root->commit_root = btrfs_root_node(root);
1481 out:
1482         btrfs_free_path(path);
1483         return root;
1484
1485 find_fail:
1486         kfree(root);
1487 alloc_fail:
1488         root = ERR_PTR(ret);
1489         goto out;
1490 }
1491
1492 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1493                                       struct btrfs_key *location)
1494 {
1495         struct btrfs_root *root;
1496
1497         root = btrfs_read_tree_root(tree_root, location);
1498         if (IS_ERR(root))
1499                 return root;
1500
1501         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1502                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1503                 btrfs_check_and_init_root_item(&root->root_item);
1504         }
1505
1506         return root;
1507 }
1508
1509 int btrfs_init_fs_root(struct btrfs_root *root)
1510 {
1511         int ret;
1512         struct btrfs_subvolume_writers *writers;
1513
1514         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1516                                         GFP_NOFS);
1517         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1518                 ret = -ENOMEM;
1519                 goto fail;
1520         }
1521
1522         writers = btrfs_alloc_subvolume_writers();
1523         if (IS_ERR(writers)) {
1524                 ret = PTR_ERR(writers);
1525                 goto fail;
1526         }
1527         root->subv_writers = writers;
1528
1529         btrfs_init_free_ino_ctl(root);
1530         spin_lock_init(&root->ino_cache_lock);
1531         init_waitqueue_head(&root->ino_cache_wait);
1532
1533         ret = get_anon_bdev(&root->anon_dev);
1534         if (ret)
1535                 goto fail;
1536
1537         mutex_lock(&root->objectid_mutex);
1538         ret = btrfs_find_highest_objectid(root,
1539                                         &root->highest_objectid);
1540         if (ret) {
1541                 mutex_unlock(&root->objectid_mutex);
1542                 goto fail;
1543         }
1544
1545         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1546
1547         mutex_unlock(&root->objectid_mutex);
1548
1549         return 0;
1550 fail:
1551         /* the caller is responsible to call free_fs_root */
1552         return ret;
1553 }
1554
1555 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1556                                         u64 root_id)
1557 {
1558         struct btrfs_root *root;
1559
1560         spin_lock(&fs_info->fs_roots_radix_lock);
1561         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1562                                  (unsigned long)root_id);
1563         spin_unlock(&fs_info->fs_roots_radix_lock);
1564         return root;
1565 }
1566
1567 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1568                          struct btrfs_root *root)
1569 {
1570         int ret;
1571
1572         ret = radix_tree_preload(GFP_NOFS);
1573         if (ret)
1574                 return ret;
1575
1576         spin_lock(&fs_info->fs_roots_radix_lock);
1577         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1578                                 (unsigned long)root->root_key.objectid,
1579                                 root);
1580         if (ret == 0)
1581                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1582         spin_unlock(&fs_info->fs_roots_radix_lock);
1583         radix_tree_preload_end();
1584
1585         return ret;
1586 }
1587
1588 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1589                                      struct btrfs_key *location,
1590                                      bool check_ref)
1591 {
1592         struct btrfs_root *root;
1593         struct btrfs_path *path;
1594         struct btrfs_key key;
1595         int ret;
1596
1597         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1598                 return fs_info->tree_root;
1599         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1600                 return fs_info->extent_root;
1601         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1602                 return fs_info->chunk_root;
1603         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1604                 return fs_info->dev_root;
1605         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1606                 return fs_info->csum_root;
1607         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1608                 return fs_info->quota_root ? fs_info->quota_root :
1609                                              ERR_PTR(-ENOENT);
1610         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1611                 return fs_info->uuid_root ? fs_info->uuid_root :
1612                                             ERR_PTR(-ENOENT);
1613         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1614                 return fs_info->free_space_root ? fs_info->free_space_root :
1615                                                   ERR_PTR(-ENOENT);
1616 again:
1617         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1618         if (root) {
1619                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1620                         return ERR_PTR(-ENOENT);
1621                 return root;
1622         }
1623
1624         root = btrfs_read_fs_root(fs_info->tree_root, location);
1625         if (IS_ERR(root))
1626                 return root;
1627
1628         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1629                 ret = -ENOENT;
1630                 goto fail;
1631         }
1632
1633         ret = btrfs_init_fs_root(root);
1634         if (ret)
1635                 goto fail;
1636
1637         path = btrfs_alloc_path();
1638         if (!path) {
1639                 ret = -ENOMEM;
1640                 goto fail;
1641         }
1642         key.objectid = BTRFS_ORPHAN_OBJECTID;
1643         key.type = BTRFS_ORPHAN_ITEM_KEY;
1644         key.offset = location->objectid;
1645
1646         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1647         btrfs_free_path(path);
1648         if (ret < 0)
1649                 goto fail;
1650         if (ret == 0)
1651                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1652
1653         ret = btrfs_insert_fs_root(fs_info, root);
1654         if (ret) {
1655                 if (ret == -EEXIST) {
1656                         free_fs_root(root);
1657                         goto again;
1658                 }
1659                 goto fail;
1660         }
1661         return root;
1662 fail:
1663         free_fs_root(root);
1664         return ERR_PTR(ret);
1665 }
1666
1667 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1668 {
1669         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1670         int ret = 0;
1671         struct btrfs_device *device;
1672         struct backing_dev_info *bdi;
1673
1674         rcu_read_lock();
1675         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1676                 if (!device->bdev)
1677                         continue;
1678                 bdi = device->bdev->bd_bdi;
1679                 if (bdi_congested(bdi, bdi_bits)) {
1680                         ret = 1;
1681                         break;
1682                 }
1683         }
1684         rcu_read_unlock();
1685         return ret;
1686 }
1687
1688 /*
1689  * called by the kthread helper functions to finally call the bio end_io
1690  * functions.  This is where read checksum verification actually happens
1691  */
1692 static void end_workqueue_fn(struct btrfs_work *work)
1693 {
1694         struct bio *bio;
1695         struct btrfs_end_io_wq *end_io_wq;
1696
1697         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1698         bio = end_io_wq->bio;
1699
1700         bio->bi_status = end_io_wq->status;
1701         bio->bi_private = end_io_wq->private;
1702         bio->bi_end_io = end_io_wq->end_io;
1703         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1704         bio_endio(bio);
1705 }
1706
1707 static int cleaner_kthread(void *arg)
1708 {
1709         struct btrfs_root *root = arg;
1710         struct btrfs_fs_info *fs_info = root->fs_info;
1711         int again;
1712         struct btrfs_trans_handle *trans;
1713
1714         do {
1715                 again = 0;
1716
1717                 /* Make the cleaner go to sleep early. */
1718                 if (btrfs_need_cleaner_sleep(fs_info))
1719                         goto sleep;
1720
1721                 /*
1722                  * Do not do anything if we might cause open_ctree() to block
1723                  * before we have finished mounting the filesystem.
1724                  */
1725                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1726                         goto sleep;
1727
1728                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1729                         goto sleep;
1730
1731                 /*
1732                  * Avoid the problem that we change the status of the fs
1733                  * during the above check and trylock.
1734                  */
1735                 if (btrfs_need_cleaner_sleep(fs_info)) {
1736                         mutex_unlock(&fs_info->cleaner_mutex);
1737                         goto sleep;
1738                 }
1739
1740                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1741                 btrfs_run_delayed_iputs(fs_info);
1742                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1743
1744                 again = btrfs_clean_one_deleted_snapshot(root);
1745                 mutex_unlock(&fs_info->cleaner_mutex);
1746
1747                 /*
1748                  * The defragger has dealt with the R/O remount and umount,
1749                  * needn't do anything special here.
1750                  */
1751                 btrfs_run_defrag_inodes(fs_info);
1752
1753                 /*
1754                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1755                  * with relocation (btrfs_relocate_chunk) and relocation
1756                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1757                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1758                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1759                  * unused block groups.
1760                  */
1761                 btrfs_delete_unused_bgs(fs_info);
1762 sleep:
1763                 if (!again) {
1764                         set_current_state(TASK_INTERRUPTIBLE);
1765                         if (!kthread_should_stop())
1766                                 schedule();
1767                         __set_current_state(TASK_RUNNING);
1768                 }
1769         } while (!kthread_should_stop());
1770
1771         /*
1772          * Transaction kthread is stopped before us and wakes us up.
1773          * However we might have started a new transaction and COWed some
1774          * tree blocks when deleting unused block groups for example. So
1775          * make sure we commit the transaction we started to have a clean
1776          * shutdown when evicting the btree inode - if it has dirty pages
1777          * when we do the final iput() on it, eviction will trigger a
1778          * writeback for it which will fail with null pointer dereferences
1779          * since work queues and other resources were already released and
1780          * destroyed by the time the iput/eviction/writeback is made.
1781          */
1782         trans = btrfs_attach_transaction(root);
1783         if (IS_ERR(trans)) {
1784                 if (PTR_ERR(trans) != -ENOENT)
1785                         btrfs_err(fs_info,
1786                                   "cleaner transaction attach returned %ld",
1787                                   PTR_ERR(trans));
1788         } else {
1789                 int ret;
1790
1791                 ret = btrfs_commit_transaction(trans);
1792                 if (ret)
1793                         btrfs_err(fs_info,
1794                                   "cleaner open transaction commit returned %d",
1795                                   ret);
1796         }
1797
1798         return 0;
1799 }
1800
1801 static int transaction_kthread(void *arg)
1802 {
1803         struct btrfs_root *root = arg;
1804         struct btrfs_fs_info *fs_info = root->fs_info;
1805         struct btrfs_trans_handle *trans;
1806         struct btrfs_transaction *cur;
1807         u64 transid;
1808         unsigned long now;
1809         unsigned long delay;
1810         bool cannot_commit;
1811
1812         do {
1813                 cannot_commit = false;
1814                 delay = HZ * fs_info->commit_interval;
1815                 mutex_lock(&fs_info->transaction_kthread_mutex);
1816
1817                 spin_lock(&fs_info->trans_lock);
1818                 cur = fs_info->running_transaction;
1819                 if (!cur) {
1820                         spin_unlock(&fs_info->trans_lock);
1821                         goto sleep;
1822                 }
1823
1824                 now = get_seconds();
1825                 if (cur->state < TRANS_STATE_BLOCKED &&
1826                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1827                     (now < cur->start_time ||
1828                      now - cur->start_time < fs_info->commit_interval)) {
1829                         spin_unlock(&fs_info->trans_lock);
1830                         delay = HZ * 5;
1831                         goto sleep;
1832                 }
1833                 transid = cur->transid;
1834                 spin_unlock(&fs_info->trans_lock);
1835
1836                 /* If the file system is aborted, this will always fail. */
1837                 trans = btrfs_attach_transaction(root);
1838                 if (IS_ERR(trans)) {
1839                         if (PTR_ERR(trans) != -ENOENT)
1840                                 cannot_commit = true;
1841                         goto sleep;
1842                 }
1843                 if (transid == trans->transid) {
1844                         btrfs_commit_transaction(trans);
1845                 } else {
1846                         btrfs_end_transaction(trans);
1847                 }
1848 sleep:
1849                 wake_up_process(fs_info->cleaner_kthread);
1850                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1851
1852                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853                                       &fs_info->fs_state)))
1854                         btrfs_cleanup_transaction(fs_info);
1855                 if (!kthread_should_stop() &&
1856                                 (!btrfs_transaction_blocked(fs_info) ||
1857                                  cannot_commit))
1858                         schedule_timeout_interruptible(delay);
1859         } while (!kthread_should_stop());
1860         return 0;
1861 }
1862
1863 /*
1864  * this will find the highest generation in the array of
1865  * root backups.  The index of the highest array is returned,
1866  * or -1 if we can't find anything.
1867  *
1868  * We check to make sure the array is valid by comparing the
1869  * generation of the latest  root in the array with the generation
1870  * in the super block.  If they don't match we pitch it.
1871  */
1872 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1873 {
1874         u64 cur;
1875         int newest_index = -1;
1876         struct btrfs_root_backup *root_backup;
1877         int i;
1878
1879         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1880                 root_backup = info->super_copy->super_roots + i;
1881                 cur = btrfs_backup_tree_root_gen(root_backup);
1882                 if (cur == newest_gen)
1883                         newest_index = i;
1884         }
1885
1886         /* check to see if we actually wrapped around */
1887         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1888                 root_backup = info->super_copy->super_roots;
1889                 cur = btrfs_backup_tree_root_gen(root_backup);
1890                 if (cur == newest_gen)
1891                         newest_index = 0;
1892         }
1893         return newest_index;
1894 }
1895
1896
1897 /*
1898  * find the oldest backup so we know where to store new entries
1899  * in the backup array.  This will set the backup_root_index
1900  * field in the fs_info struct
1901  */
1902 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1903                                      u64 newest_gen)
1904 {
1905         int newest_index = -1;
1906
1907         newest_index = find_newest_super_backup(info, newest_gen);
1908         /* if there was garbage in there, just move along */
1909         if (newest_index == -1) {
1910                 info->backup_root_index = 0;
1911         } else {
1912                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1913         }
1914 }
1915
1916 /*
1917  * copy all the root pointers into the super backup array.
1918  * this will bump the backup pointer by one when it is
1919  * done
1920  */
1921 static void backup_super_roots(struct btrfs_fs_info *info)
1922 {
1923         int next_backup;
1924         struct btrfs_root_backup *root_backup;
1925         int last_backup;
1926
1927         next_backup = info->backup_root_index;
1928         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1929                 BTRFS_NUM_BACKUP_ROOTS;
1930
1931         /*
1932          * just overwrite the last backup if we're at the same generation
1933          * this happens only at umount
1934          */
1935         root_backup = info->super_for_commit->super_roots + last_backup;
1936         if (btrfs_backup_tree_root_gen(root_backup) ==
1937             btrfs_header_generation(info->tree_root->node))
1938                 next_backup = last_backup;
1939
1940         root_backup = info->super_for_commit->super_roots + next_backup;
1941
1942         /*
1943          * make sure all of our padding and empty slots get zero filled
1944          * regardless of which ones we use today
1945          */
1946         memset(root_backup, 0, sizeof(*root_backup));
1947
1948         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1949
1950         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1951         btrfs_set_backup_tree_root_gen(root_backup,
1952                                btrfs_header_generation(info->tree_root->node));
1953
1954         btrfs_set_backup_tree_root_level(root_backup,
1955                                btrfs_header_level(info->tree_root->node));
1956
1957         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1958         btrfs_set_backup_chunk_root_gen(root_backup,
1959                                btrfs_header_generation(info->chunk_root->node));
1960         btrfs_set_backup_chunk_root_level(root_backup,
1961                                btrfs_header_level(info->chunk_root->node));
1962
1963         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1964         btrfs_set_backup_extent_root_gen(root_backup,
1965                                btrfs_header_generation(info->extent_root->node));
1966         btrfs_set_backup_extent_root_level(root_backup,
1967                                btrfs_header_level(info->extent_root->node));
1968
1969         /*
1970          * we might commit during log recovery, which happens before we set
1971          * the fs_root.  Make sure it is valid before we fill it in.
1972          */
1973         if (info->fs_root && info->fs_root->node) {
1974                 btrfs_set_backup_fs_root(root_backup,
1975                                          info->fs_root->node->start);
1976                 btrfs_set_backup_fs_root_gen(root_backup,
1977                                btrfs_header_generation(info->fs_root->node));
1978                 btrfs_set_backup_fs_root_level(root_backup,
1979                                btrfs_header_level(info->fs_root->node));
1980         }
1981
1982         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1983         btrfs_set_backup_dev_root_gen(root_backup,
1984                                btrfs_header_generation(info->dev_root->node));
1985         btrfs_set_backup_dev_root_level(root_backup,
1986                                        btrfs_header_level(info->dev_root->node));
1987
1988         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1989         btrfs_set_backup_csum_root_gen(root_backup,
1990                                btrfs_header_generation(info->csum_root->node));
1991         btrfs_set_backup_csum_root_level(root_backup,
1992                                btrfs_header_level(info->csum_root->node));
1993
1994         btrfs_set_backup_total_bytes(root_backup,
1995                              btrfs_super_total_bytes(info->super_copy));
1996         btrfs_set_backup_bytes_used(root_backup,
1997                              btrfs_super_bytes_used(info->super_copy));
1998         btrfs_set_backup_num_devices(root_backup,
1999                              btrfs_super_num_devices(info->super_copy));
2000
2001         /*
2002          * if we don't copy this out to the super_copy, it won't get remembered
2003          * for the next commit
2004          */
2005         memcpy(&info->super_copy->super_roots,
2006                &info->super_for_commit->super_roots,
2007                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2008 }
2009
2010 /*
2011  * this copies info out of the root backup array and back into
2012  * the in-memory super block.  It is meant to help iterate through
2013  * the array, so you send it the number of backups you've already
2014  * tried and the last backup index you used.
2015  *
2016  * this returns -1 when it has tried all the backups
2017  */
2018 static noinline int next_root_backup(struct btrfs_fs_info *info,
2019                                      struct btrfs_super_block *super,
2020                                      int *num_backups_tried, int *backup_index)
2021 {
2022         struct btrfs_root_backup *root_backup;
2023         int newest = *backup_index;
2024
2025         if (*num_backups_tried == 0) {
2026                 u64 gen = btrfs_super_generation(super);
2027
2028                 newest = find_newest_super_backup(info, gen);
2029                 if (newest == -1)
2030                         return -1;
2031
2032                 *backup_index = newest;
2033                 *num_backups_tried = 1;
2034         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2035                 /* we've tried all the backups, all done */
2036                 return -1;
2037         } else {
2038                 /* jump to the next oldest backup */
2039                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2040                         BTRFS_NUM_BACKUP_ROOTS;
2041                 *backup_index = newest;
2042                 *num_backups_tried += 1;
2043         }
2044         root_backup = super->super_roots + newest;
2045
2046         btrfs_set_super_generation(super,
2047                                    btrfs_backup_tree_root_gen(root_backup));
2048         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2049         btrfs_set_super_root_level(super,
2050                                    btrfs_backup_tree_root_level(root_backup));
2051         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2052
2053         /*
2054          * fixme: the total bytes and num_devices need to match or we should
2055          * need a fsck
2056          */
2057         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2058         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2059         return 0;
2060 }
2061
2062 /* helper to cleanup workers */
2063 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2064 {
2065         btrfs_destroy_workqueue(fs_info->fixup_workers);
2066         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2067         btrfs_destroy_workqueue(fs_info->workers);
2068         btrfs_destroy_workqueue(fs_info->endio_workers);
2069         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2070         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2071         btrfs_destroy_workqueue(fs_info->rmw_workers);
2072         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2074         btrfs_destroy_workqueue(fs_info->submit_workers);
2075         btrfs_destroy_workqueue(fs_info->delayed_workers);
2076         btrfs_destroy_workqueue(fs_info->caching_workers);
2077         btrfs_destroy_workqueue(fs_info->readahead_workers);
2078         btrfs_destroy_workqueue(fs_info->flush_workers);
2079         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2080         btrfs_destroy_workqueue(fs_info->extent_workers);
2081         /*
2082          * Now that all other work queues are destroyed, we can safely destroy
2083          * the queues used for metadata I/O, since tasks from those other work
2084          * queues can do metadata I/O operations.
2085          */
2086         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2087         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2088 }
2089
2090 static void free_root_extent_buffers(struct btrfs_root *root)
2091 {
2092         if (root) {
2093                 free_extent_buffer(root->node);
2094                 free_extent_buffer(root->commit_root);
2095                 root->node = NULL;
2096                 root->commit_root = NULL;
2097         }
2098 }
2099
2100 /* helper to cleanup tree roots */
2101 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2102 {
2103         free_root_extent_buffers(info->tree_root);
2104
2105         free_root_extent_buffers(info->dev_root);
2106         free_root_extent_buffers(info->extent_root);
2107         free_root_extent_buffers(info->csum_root);
2108         free_root_extent_buffers(info->quota_root);
2109         free_root_extent_buffers(info->uuid_root);
2110         if (chunk_root)
2111                 free_root_extent_buffers(info->chunk_root);
2112         free_root_extent_buffers(info->free_space_root);
2113 }
2114
2115 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2116 {
2117         int ret;
2118         struct btrfs_root *gang[8];
2119         int i;
2120
2121         while (!list_empty(&fs_info->dead_roots)) {
2122                 gang[0] = list_entry(fs_info->dead_roots.next,
2123                                      struct btrfs_root, root_list);
2124                 list_del(&gang[0]->root_list);
2125
2126                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2127                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2128                 } else {
2129                         free_extent_buffer(gang[0]->node);
2130                         free_extent_buffer(gang[0]->commit_root);
2131                         btrfs_put_fs_root(gang[0]);
2132                 }
2133         }
2134
2135         while (1) {
2136                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2137                                              (void **)gang, 0,
2138                                              ARRAY_SIZE(gang));
2139                 if (!ret)
2140                         break;
2141                 for (i = 0; i < ret; i++)
2142                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2143         }
2144
2145         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2146                 btrfs_free_log_root_tree(NULL, fs_info);
2147                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2148         }
2149 }
2150
2151 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2152 {
2153         mutex_init(&fs_info->scrub_lock);
2154         atomic_set(&fs_info->scrubs_running, 0);
2155         atomic_set(&fs_info->scrub_pause_req, 0);
2156         atomic_set(&fs_info->scrubs_paused, 0);
2157         atomic_set(&fs_info->scrub_cancel_req, 0);
2158         init_waitqueue_head(&fs_info->scrub_pause_wait);
2159         fs_info->scrub_workers_refcnt = 0;
2160 }
2161
2162 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2163 {
2164         spin_lock_init(&fs_info->balance_lock);
2165         mutex_init(&fs_info->balance_mutex);
2166         atomic_set(&fs_info->balance_pause_req, 0);
2167         atomic_set(&fs_info->balance_cancel_req, 0);
2168         fs_info->balance_ctl = NULL;
2169         init_waitqueue_head(&fs_info->balance_wait_q);
2170 }
2171
2172 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2173 {
2174         struct inode *inode = fs_info->btree_inode;
2175
2176         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2177         set_nlink(inode, 1);
2178         /*
2179          * we set the i_size on the btree inode to the max possible int.
2180          * the real end of the address space is determined by all of
2181          * the devices in the system
2182          */
2183         inode->i_size = OFFSET_MAX;
2184         inode->i_mapping->a_ops = &btree_aops;
2185
2186         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2187         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2188         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2189         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2190
2191         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2192
2193         BTRFS_I(inode)->root = fs_info->tree_root;
2194         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2195         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2196         btrfs_insert_inode_hash(inode);
2197 }
2198
2199 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2200 {
2201         fs_info->dev_replace.lock_owner = 0;
2202         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2203         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2204         rwlock_init(&fs_info->dev_replace.lock);
2205         atomic_set(&fs_info->dev_replace.read_locks, 0);
2206         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2207         init_waitqueue_head(&fs_info->replace_wait);
2208         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2209 }
2210
2211 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2212 {
2213         spin_lock_init(&fs_info->qgroup_lock);
2214         mutex_init(&fs_info->qgroup_ioctl_lock);
2215         fs_info->qgroup_tree = RB_ROOT;
2216         fs_info->qgroup_op_tree = RB_ROOT;
2217         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2218         fs_info->qgroup_seq = 1;
2219         fs_info->qgroup_ulist = NULL;
2220         fs_info->qgroup_rescan_running = false;
2221         mutex_init(&fs_info->qgroup_rescan_lock);
2222 }
2223
2224 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2225                 struct btrfs_fs_devices *fs_devices)
2226 {
2227         u32 max_active = fs_info->thread_pool_size;
2228         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2229
2230         fs_info->workers =
2231                 btrfs_alloc_workqueue(fs_info, "worker",
2232                                       flags | WQ_HIGHPRI, max_active, 16);
2233
2234         fs_info->delalloc_workers =
2235                 btrfs_alloc_workqueue(fs_info, "delalloc",
2236                                       flags, max_active, 2);
2237
2238         fs_info->flush_workers =
2239                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2240                                       flags, max_active, 0);
2241
2242         fs_info->caching_workers =
2243                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2244
2245         /*
2246          * a higher idle thresh on the submit workers makes it much more
2247          * likely that bios will be send down in a sane order to the
2248          * devices
2249          */
2250         fs_info->submit_workers =
2251                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2252                                       min_t(u64, fs_devices->num_devices,
2253                                             max_active), 64);
2254
2255         fs_info->fixup_workers =
2256                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2257
2258         /*
2259          * endios are largely parallel and should have a very
2260          * low idle thresh
2261          */
2262         fs_info->endio_workers =
2263                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2264         fs_info->endio_meta_workers =
2265                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2266                                       max_active, 4);
2267         fs_info->endio_meta_write_workers =
2268                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2269                                       max_active, 2);
2270         fs_info->endio_raid56_workers =
2271                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2272                                       max_active, 4);
2273         fs_info->endio_repair_workers =
2274                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2275         fs_info->rmw_workers =
2276                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2277         fs_info->endio_write_workers =
2278                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2279                                       max_active, 2);
2280         fs_info->endio_freespace_worker =
2281                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2282                                       max_active, 0);
2283         fs_info->delayed_workers =
2284                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2285                                       max_active, 0);
2286         fs_info->readahead_workers =
2287                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2288                                       max_active, 2);
2289         fs_info->qgroup_rescan_workers =
2290                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2291         fs_info->extent_workers =
2292                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2293                                       min_t(u64, fs_devices->num_devices,
2294                                             max_active), 8);
2295
2296         if (!(fs_info->workers && fs_info->delalloc_workers &&
2297               fs_info->submit_workers && fs_info->flush_workers &&
2298               fs_info->endio_workers && fs_info->endio_meta_workers &&
2299               fs_info->endio_meta_write_workers &&
2300               fs_info->endio_repair_workers &&
2301               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2302               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2303               fs_info->caching_workers && fs_info->readahead_workers &&
2304               fs_info->fixup_workers && fs_info->delayed_workers &&
2305               fs_info->extent_workers &&
2306               fs_info->qgroup_rescan_workers)) {
2307                 return -ENOMEM;
2308         }
2309
2310         return 0;
2311 }
2312
2313 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2314                             struct btrfs_fs_devices *fs_devices)
2315 {
2316         int ret;
2317         struct btrfs_root *log_tree_root;
2318         struct btrfs_super_block *disk_super = fs_info->super_copy;
2319         u64 bytenr = btrfs_super_log_root(disk_super);
2320         int level = btrfs_super_log_root_level(disk_super);
2321
2322         if (fs_devices->rw_devices == 0) {
2323                 btrfs_warn(fs_info, "log replay required on RO media");
2324                 return -EIO;
2325         }
2326
2327         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2328         if (!log_tree_root)
2329                 return -ENOMEM;
2330
2331         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2332
2333         log_tree_root->node = read_tree_block(fs_info, bytenr,
2334                                               fs_info->generation + 1,
2335                                               level, NULL);
2336         if (IS_ERR(log_tree_root->node)) {
2337                 btrfs_warn(fs_info, "failed to read log tree");
2338                 ret = PTR_ERR(log_tree_root->node);
2339                 kfree(log_tree_root);
2340                 return ret;
2341         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2342                 btrfs_err(fs_info, "failed to read log tree");
2343                 free_extent_buffer(log_tree_root->node);
2344                 kfree(log_tree_root);
2345                 return -EIO;
2346         }
2347         /* returns with log_tree_root freed on success */
2348         ret = btrfs_recover_log_trees(log_tree_root);
2349         if (ret) {
2350                 btrfs_handle_fs_error(fs_info, ret,
2351                                       "Failed to recover log tree");
2352                 free_extent_buffer(log_tree_root->node);
2353                 kfree(log_tree_root);
2354                 return ret;
2355         }
2356
2357         if (sb_rdonly(fs_info->sb)) {
2358                 ret = btrfs_commit_super(fs_info);
2359                 if (ret)
2360                         return ret;
2361         }
2362
2363         return 0;
2364 }
2365
2366 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2367 {
2368         struct btrfs_root *tree_root = fs_info->tree_root;
2369         struct btrfs_root *root;
2370         struct btrfs_key location;
2371         int ret;
2372
2373         BUG_ON(!fs_info->tree_root);
2374
2375         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2376         location.type = BTRFS_ROOT_ITEM_KEY;
2377         location.offset = 0;
2378
2379         root = btrfs_read_tree_root(tree_root, &location);
2380         if (IS_ERR(root)) {
2381                 ret = PTR_ERR(root);
2382                 goto out;
2383         }
2384         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2385         fs_info->extent_root = root;
2386
2387         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2388         root = btrfs_read_tree_root(tree_root, &location);
2389         if (IS_ERR(root)) {
2390                 ret = PTR_ERR(root);
2391                 goto out;
2392         }
2393         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2394         fs_info->dev_root = root;
2395         btrfs_init_devices_late(fs_info);
2396
2397         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2398         root = btrfs_read_tree_root(tree_root, &location);
2399         if (IS_ERR(root)) {
2400                 ret = PTR_ERR(root);
2401                 goto out;
2402         }
2403         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2404         fs_info->csum_root = root;
2405
2406         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2407         root = btrfs_read_tree_root(tree_root, &location);
2408         if (!IS_ERR(root)) {
2409                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2410                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2411                 fs_info->quota_root = root;
2412         }
2413
2414         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2415         root = btrfs_read_tree_root(tree_root, &location);
2416         if (IS_ERR(root)) {
2417                 ret = PTR_ERR(root);
2418                 if (ret != -ENOENT)
2419                         goto out;
2420         } else {
2421                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2422                 fs_info->uuid_root = root;
2423         }
2424
2425         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2426                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2427                 root = btrfs_read_tree_root(tree_root, &location);
2428                 if (IS_ERR(root)) {
2429                         ret = PTR_ERR(root);
2430                         goto out;
2431                 }
2432                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2433                 fs_info->free_space_root = root;
2434         }
2435
2436         return 0;
2437 out:
2438         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2439                    location.objectid, ret);
2440         return ret;
2441 }
2442
2443 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
2444 {
2445         struct btrfs_super_block *sb = fs_info->super_copy;
2446         u64 nodesize = btrfs_super_nodesize(sb);
2447         u64 sectorsize = btrfs_super_sectorsize(sb);
2448         int ret = 0;
2449
2450         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2451                 btrfs_err(fs_info, "no valid FS found");
2452                 ret = -EINVAL;
2453         }
2454         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2455                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2456                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2457                 ret = -EINVAL;
2458         }
2459         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2460                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2461                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2462                 ret = -EINVAL;
2463         }
2464         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2465                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2466                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2467                 ret = -EINVAL;
2468         }
2469         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2470                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2471                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2472                 ret = -EINVAL;
2473         }
2474
2475         /*
2476          * Check sectorsize and nodesize first, other check will need it.
2477          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2478          */
2479         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2480             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2481                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2482                 ret = -EINVAL;
2483         }
2484         /* Only PAGE SIZE is supported yet */
2485         if (sectorsize != PAGE_SIZE) {
2486                 btrfs_err(fs_info,
2487                         "sectorsize %llu not supported yet, only support %lu",
2488                         sectorsize, PAGE_SIZE);
2489                 ret = -EINVAL;
2490         }
2491         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2492             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2493                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2494                 ret = -EINVAL;
2495         }
2496         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2497                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2498                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2499                 ret = -EINVAL;
2500         }
2501
2502         /* Root alignment check */
2503         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2504                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2505                            btrfs_super_root(sb));
2506                 ret = -EINVAL;
2507         }
2508         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2509                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2510                            btrfs_super_chunk_root(sb));
2511                 ret = -EINVAL;
2512         }
2513         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2514                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2515                            btrfs_super_log_root(sb));
2516                 ret = -EINVAL;
2517         }
2518
2519         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2520                 btrfs_err(fs_info,
2521                            "dev_item UUID does not match fsid: %pU != %pU",
2522                            fs_info->fsid, sb->dev_item.fsid);
2523                 ret = -EINVAL;
2524         }
2525
2526         /*
2527          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2528          * done later
2529          */
2530         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2531                 btrfs_err(fs_info, "bytes_used is too small %llu",
2532                           btrfs_super_bytes_used(sb));
2533                 ret = -EINVAL;
2534         }
2535         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2536                 btrfs_err(fs_info, "invalid stripesize %u",
2537                           btrfs_super_stripesize(sb));
2538                 ret = -EINVAL;
2539         }
2540         if (btrfs_super_num_devices(sb) > (1UL << 31))
2541                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2542                            btrfs_super_num_devices(sb));
2543         if (btrfs_super_num_devices(sb) == 0) {
2544                 btrfs_err(fs_info, "number of devices is 0");
2545                 ret = -EINVAL;
2546         }
2547
2548         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
2549                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2550                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2551                 ret = -EINVAL;
2552         }
2553
2554         /*
2555          * Obvious sys_chunk_array corruptions, it must hold at least one key
2556          * and one chunk
2557          */
2558         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2559                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2560                           btrfs_super_sys_array_size(sb),
2561                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2562                 ret = -EINVAL;
2563         }
2564         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2565                         + sizeof(struct btrfs_chunk)) {
2566                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2567                           btrfs_super_sys_array_size(sb),
2568                           sizeof(struct btrfs_disk_key)
2569                           + sizeof(struct btrfs_chunk));
2570                 ret = -EINVAL;
2571         }
2572
2573         /*
2574          * The generation is a global counter, we'll trust it more than the others
2575          * but it's still possible that it's the one that's wrong.
2576          */
2577         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2578                 btrfs_warn(fs_info,
2579                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2580                         btrfs_super_generation(sb),
2581                         btrfs_super_chunk_root_generation(sb));
2582         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2583             && btrfs_super_cache_generation(sb) != (u64)-1)
2584                 btrfs_warn(fs_info,
2585                         "suspicious: generation < cache_generation: %llu < %llu",
2586                         btrfs_super_generation(sb),
2587                         btrfs_super_cache_generation(sb));
2588
2589         return ret;
2590 }
2591
2592 int open_ctree(struct super_block *sb,
2593                struct btrfs_fs_devices *fs_devices,
2594                char *options)
2595 {
2596         u32 sectorsize;
2597         u32 nodesize;
2598         u32 stripesize;
2599         u64 generation;
2600         u64 features;
2601         struct btrfs_key location;
2602         struct buffer_head *bh;
2603         struct btrfs_super_block *disk_super;
2604         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2605         struct btrfs_root *tree_root;
2606         struct btrfs_root *chunk_root;
2607         int ret;
2608         int err = -EINVAL;
2609         int num_backups_tried = 0;
2610         int backup_index = 0;
2611         int clear_free_space_tree = 0;
2612         int level;
2613
2614         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2615         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2616         if (!tree_root || !chunk_root) {
2617                 err = -ENOMEM;
2618                 goto fail;
2619         }
2620
2621         ret = init_srcu_struct(&fs_info->subvol_srcu);
2622         if (ret) {
2623                 err = ret;
2624                 goto fail;
2625         }
2626
2627         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2628         if (ret) {
2629                 err = ret;
2630                 goto fail_srcu;
2631         }
2632         fs_info->dirty_metadata_batch = PAGE_SIZE *
2633                                         (1 + ilog2(nr_cpu_ids));
2634
2635         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2636         if (ret) {
2637                 err = ret;
2638                 goto fail_dirty_metadata_bytes;
2639         }
2640
2641         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2642         if (ret) {
2643                 err = ret;
2644                 goto fail_delalloc_bytes;
2645         }
2646
2647         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2648         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2649         INIT_LIST_HEAD(&fs_info->trans_list);
2650         INIT_LIST_HEAD(&fs_info->dead_roots);
2651         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2652         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2653         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2654         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2655         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2656         spin_lock_init(&fs_info->delalloc_root_lock);
2657         spin_lock_init(&fs_info->trans_lock);
2658         spin_lock_init(&fs_info->fs_roots_radix_lock);
2659         spin_lock_init(&fs_info->delayed_iput_lock);
2660         spin_lock_init(&fs_info->defrag_inodes_lock);
2661         spin_lock_init(&fs_info->tree_mod_seq_lock);
2662         spin_lock_init(&fs_info->super_lock);
2663         spin_lock_init(&fs_info->qgroup_op_lock);
2664         spin_lock_init(&fs_info->buffer_lock);
2665         spin_lock_init(&fs_info->unused_bgs_lock);
2666         rwlock_init(&fs_info->tree_mod_log_lock);
2667         mutex_init(&fs_info->unused_bg_unpin_mutex);
2668         mutex_init(&fs_info->delete_unused_bgs_mutex);
2669         mutex_init(&fs_info->reloc_mutex);
2670         mutex_init(&fs_info->delalloc_root_mutex);
2671         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2672         seqlock_init(&fs_info->profiles_lock);
2673
2674         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2675         INIT_LIST_HEAD(&fs_info->space_info);
2676         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2677         INIT_LIST_HEAD(&fs_info->unused_bgs);
2678         btrfs_mapping_init(&fs_info->mapping_tree);
2679         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2680                              BTRFS_BLOCK_RSV_GLOBAL);
2681         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2682         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2683         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2684         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2685                              BTRFS_BLOCK_RSV_DELOPS);
2686         atomic_set(&fs_info->async_delalloc_pages, 0);
2687         atomic_set(&fs_info->defrag_running, 0);
2688         atomic_set(&fs_info->qgroup_op_seq, 0);
2689         atomic_set(&fs_info->reada_works_cnt, 0);
2690         atomic64_set(&fs_info->tree_mod_seq, 0);
2691         fs_info->sb = sb;
2692         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2693         fs_info->metadata_ratio = 0;
2694         fs_info->defrag_inodes = RB_ROOT;
2695         atomic64_set(&fs_info->free_chunk_space, 0);
2696         fs_info->tree_mod_log = RB_ROOT;
2697         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2698         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2699         /* readahead state */
2700         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2701         spin_lock_init(&fs_info->reada_lock);
2702         btrfs_init_ref_verify(fs_info);
2703
2704         fs_info->thread_pool_size = min_t(unsigned long,
2705                                           num_online_cpus() + 2, 8);
2706
2707         INIT_LIST_HEAD(&fs_info->ordered_roots);
2708         spin_lock_init(&fs_info->ordered_root_lock);
2709
2710         fs_info->btree_inode = new_inode(sb);
2711         if (!fs_info->btree_inode) {
2712                 err = -ENOMEM;
2713                 goto fail_bio_counter;
2714         }
2715         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2716
2717         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2718                                         GFP_KERNEL);
2719         if (!fs_info->delayed_root) {
2720                 err = -ENOMEM;
2721                 goto fail_iput;
2722         }
2723         btrfs_init_delayed_root(fs_info->delayed_root);
2724
2725         btrfs_init_scrub(fs_info);
2726 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2727         fs_info->check_integrity_print_mask = 0;
2728 #endif
2729         btrfs_init_balance(fs_info);
2730         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2731
2732         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2733         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2734
2735         btrfs_init_btree_inode(fs_info);
2736
2737         spin_lock_init(&fs_info->block_group_cache_lock);
2738         fs_info->block_group_cache_tree = RB_ROOT;
2739         fs_info->first_logical_byte = (u64)-1;
2740
2741         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2742         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2743         fs_info->pinned_extents = &fs_info->freed_extents[0];
2744         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2745
2746         mutex_init(&fs_info->ordered_operations_mutex);
2747         mutex_init(&fs_info->tree_log_mutex);
2748         mutex_init(&fs_info->chunk_mutex);
2749         mutex_init(&fs_info->transaction_kthread_mutex);
2750         mutex_init(&fs_info->cleaner_mutex);
2751         mutex_init(&fs_info->ro_block_group_mutex);
2752         init_rwsem(&fs_info->commit_root_sem);
2753         init_rwsem(&fs_info->cleanup_work_sem);
2754         init_rwsem(&fs_info->subvol_sem);
2755         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2756
2757         btrfs_init_dev_replace_locks(fs_info);
2758         btrfs_init_qgroup(fs_info);
2759
2760         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2761         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2762
2763         init_waitqueue_head(&fs_info->transaction_throttle);
2764         init_waitqueue_head(&fs_info->transaction_wait);
2765         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2766         init_waitqueue_head(&fs_info->async_submit_wait);
2767
2768         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2769
2770         /* Usable values until the real ones are cached from the superblock */
2771         fs_info->nodesize = 4096;
2772         fs_info->sectorsize = 4096;
2773         fs_info->stripesize = 4096;
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         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2812                sizeof(*fs_info->super_for_commit));
2813         brelse(bh);
2814
2815         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2816
2817         ret = btrfs_check_super_valid(fs_info);
2818         if (ret) {
2819                 btrfs_err(fs_info, "superblock contains fatal errors");
2820                 err = -EINVAL;
2821                 goto fail_alloc;
2822         }
2823
2824         disk_super = fs_info->super_copy;
2825         if (!btrfs_super_root(disk_super))
2826                 goto fail_alloc;
2827
2828         /* check FS state, whether FS is broken. */
2829         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2830                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2831
2832         /*
2833          * run through our array of backup supers and setup
2834          * our ring pointer to the oldest one
2835          */
2836         generation = btrfs_super_generation(disk_super);
2837         find_oldest_super_backup(fs_info, generation);
2838
2839         /*
2840          * In the long term, we'll store the compression type in the super
2841          * block, and it'll be used for per file compression control.
2842          */
2843         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2844
2845         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2846         if (ret) {
2847                 err = ret;
2848                 goto fail_alloc;
2849         }
2850
2851         features = btrfs_super_incompat_flags(disk_super) &
2852                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2853         if (features) {
2854                 btrfs_err(fs_info,
2855                     "cannot mount because of unsupported optional features (%llx)",
2856                     features);
2857                 err = -EINVAL;
2858                 goto fail_alloc;
2859         }
2860
2861         features = btrfs_super_incompat_flags(disk_super);
2862         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2863         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2864                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2865         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2866                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2867
2868         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2869                 btrfs_info(fs_info, "has skinny extents");
2870
2871         /*
2872          * flag our filesystem as having big metadata blocks if
2873          * they are bigger than the page size
2874          */
2875         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2876                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2877                         btrfs_info(fs_info,
2878                                 "flagging fs with big metadata feature");
2879                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2880         }
2881
2882         nodesize = btrfs_super_nodesize(disk_super);
2883         sectorsize = btrfs_super_sectorsize(disk_super);
2884         stripesize = sectorsize;
2885         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2886         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2887
2888         /* Cache block sizes */
2889         fs_info->nodesize = nodesize;
2890         fs_info->sectorsize = sectorsize;
2891         fs_info->stripesize = stripesize;
2892
2893         /*
2894          * mixed block groups end up with duplicate but slightly offset
2895          * extent buffers for the same range.  It leads to corruptions
2896          */
2897         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2898             (sectorsize != nodesize)) {
2899                 btrfs_err(fs_info,
2900 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2901                         nodesize, sectorsize);
2902                 goto fail_alloc;
2903         }
2904
2905         /*
2906          * Needn't use the lock because there is no other task which will
2907          * update the flag.
2908          */
2909         btrfs_set_super_incompat_flags(disk_super, features);
2910
2911         features = btrfs_super_compat_ro_flags(disk_super) &
2912                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2913         if (!sb_rdonly(sb) && features) {
2914                 btrfs_err(fs_info,
2915         "cannot mount read-write because of unsupported optional features (%llx)",
2916                        features);
2917                 err = -EINVAL;
2918                 goto fail_alloc;
2919         }
2920
2921         ret = btrfs_init_workqueues(fs_info, fs_devices);
2922         if (ret) {
2923                 err = ret;
2924                 goto fail_sb_buffer;
2925         }
2926
2927         sb->s_bdi->congested_fn = btrfs_congested_fn;
2928         sb->s_bdi->congested_data = fs_info;
2929         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2930         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2931         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2932         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2933
2934         sb->s_blocksize = sectorsize;
2935         sb->s_blocksize_bits = blksize_bits(sectorsize);
2936         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2937
2938         mutex_lock(&fs_info->chunk_mutex);
2939         ret = btrfs_read_sys_array(fs_info);
2940         mutex_unlock(&fs_info->chunk_mutex);
2941         if (ret) {
2942                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2943                 goto fail_sb_buffer;
2944         }
2945
2946         generation = btrfs_super_chunk_root_generation(disk_super);
2947         level = btrfs_super_chunk_root_level(disk_super);
2948
2949         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2950
2951         chunk_root->node = read_tree_block(fs_info,
2952                                            btrfs_super_chunk_root(disk_super),
2953                                            generation, level, NULL);
2954         if (IS_ERR(chunk_root->node) ||
2955             !extent_buffer_uptodate(chunk_root->node)) {
2956                 btrfs_err(fs_info, "failed to read chunk root");
2957                 if (!IS_ERR(chunk_root->node))
2958                         free_extent_buffer(chunk_root->node);
2959                 chunk_root->node = NULL;
2960                 goto fail_tree_roots;
2961         }
2962         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2963         chunk_root->commit_root = btrfs_root_node(chunk_root);
2964
2965         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2966            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2967
2968         ret = btrfs_read_chunk_tree(fs_info);
2969         if (ret) {
2970                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2971                 goto fail_tree_roots;
2972         }
2973
2974         /*
2975          * Keep the devid that is marked to be the target device for the
2976          * device replace procedure
2977          */
2978         btrfs_free_extra_devids(fs_devices, 0);
2979
2980         if (!fs_devices->latest_bdev) {
2981                 btrfs_err(fs_info, "failed to read devices");
2982                 goto fail_tree_roots;
2983         }
2984
2985 retry_root_backup:
2986         generation = btrfs_super_generation(disk_super);
2987         level = btrfs_super_root_level(disk_super);
2988
2989         tree_root->node = read_tree_block(fs_info,
2990                                           btrfs_super_root(disk_super),
2991                                           generation, level, NULL);
2992         if (IS_ERR(tree_root->node) ||
2993             !extent_buffer_uptodate(tree_root->node)) {
2994                 btrfs_warn(fs_info, "failed to read tree root");
2995                 if (!IS_ERR(tree_root->node))
2996                         free_extent_buffer(tree_root->node);
2997                 tree_root->node = NULL;
2998                 goto recovery_tree_root;
2999         }
3000
3001         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3002         tree_root->commit_root = btrfs_root_node(tree_root);
3003         btrfs_set_root_refs(&tree_root->root_item, 1);
3004
3005         mutex_lock(&tree_root->objectid_mutex);
3006         ret = btrfs_find_highest_objectid(tree_root,
3007                                         &tree_root->highest_objectid);
3008         if (ret) {
3009                 mutex_unlock(&tree_root->objectid_mutex);
3010                 goto recovery_tree_root;
3011         }
3012
3013         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3014
3015         mutex_unlock(&tree_root->objectid_mutex);
3016
3017         ret = btrfs_read_roots(fs_info);
3018         if (ret)
3019                 goto recovery_tree_root;
3020
3021         fs_info->generation = generation;
3022         fs_info->last_trans_committed = generation;
3023
3024         ret = btrfs_recover_balance(fs_info);
3025         if (ret) {
3026                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3027                 goto fail_block_groups;
3028         }
3029
3030         ret = btrfs_init_dev_stats(fs_info);
3031         if (ret) {
3032                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3033                 goto fail_block_groups;
3034         }
3035
3036         ret = btrfs_init_dev_replace(fs_info);
3037         if (ret) {
3038                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3039                 goto fail_block_groups;
3040         }
3041
3042         btrfs_free_extra_devids(fs_devices, 1);
3043
3044         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3045         if (ret) {
3046                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3047                                 ret);
3048                 goto fail_block_groups;
3049         }
3050
3051         ret = btrfs_sysfs_add_device(fs_devices);
3052         if (ret) {
3053                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3054                                 ret);
3055                 goto fail_fsdev_sysfs;
3056         }
3057
3058         ret = btrfs_sysfs_add_mounted(fs_info);
3059         if (ret) {
3060                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3061                 goto fail_fsdev_sysfs;
3062         }
3063
3064         ret = btrfs_init_space_info(fs_info);
3065         if (ret) {
3066                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3067                 goto fail_sysfs;
3068         }
3069
3070         ret = btrfs_read_block_groups(fs_info);
3071         if (ret) {
3072                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3073                 goto fail_sysfs;
3074         }
3075
3076         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3077                 btrfs_warn(fs_info,
3078                 "writeable mount is not allowed due to too many missing devices");
3079                 goto fail_sysfs;
3080         }
3081
3082         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3083                                                "btrfs-cleaner");
3084         if (IS_ERR(fs_info->cleaner_kthread))
3085                 goto fail_sysfs;
3086
3087         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3088                                                    tree_root,
3089                                                    "btrfs-transaction");
3090         if (IS_ERR(fs_info->transaction_kthread))
3091                 goto fail_cleaner;
3092
3093         if (!btrfs_test_opt(fs_info, NOSSD) &&
3094             !fs_info->fs_devices->rotating) {
3095                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3096         }
3097
3098         /*
3099          * Mount does not set all options immediately, we can do it now and do
3100          * not have to wait for transaction commit
3101          */
3102         btrfs_apply_pending_changes(fs_info);
3103
3104 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3105         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3106                 ret = btrfsic_mount(fs_info, fs_devices,
3107                                     btrfs_test_opt(fs_info,
3108                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3109                                     1 : 0,
3110                                     fs_info->check_integrity_print_mask);
3111                 if (ret)
3112                         btrfs_warn(fs_info,
3113                                 "failed to initialize integrity check module: %d",
3114                                 ret);
3115         }
3116 #endif
3117         ret = btrfs_read_qgroup_config(fs_info);
3118         if (ret)
3119                 goto fail_trans_kthread;
3120
3121         if (btrfs_build_ref_tree(fs_info))
3122                 btrfs_err(fs_info, "couldn't build ref tree");
3123
3124         /* do not make disk changes in broken FS or nologreplay is given */
3125         if (btrfs_super_log_root(disk_super) != 0 &&
3126             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3127                 ret = btrfs_replay_log(fs_info, fs_devices);
3128                 if (ret) {
3129                         err = ret;
3130                         goto fail_qgroup;
3131                 }
3132         }
3133
3134         ret = btrfs_find_orphan_roots(fs_info);
3135         if (ret)
3136                 goto fail_qgroup;
3137
3138         if (!sb_rdonly(sb)) {
3139                 ret = btrfs_cleanup_fs_roots(fs_info);
3140                 if (ret)
3141                         goto fail_qgroup;
3142
3143                 mutex_lock(&fs_info->cleaner_mutex);
3144                 ret = btrfs_recover_relocation(tree_root);
3145                 mutex_unlock(&fs_info->cleaner_mutex);
3146                 if (ret < 0) {
3147                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3148                                         ret);
3149                         err = -EINVAL;
3150                         goto fail_qgroup;
3151                 }
3152         }
3153
3154         location.objectid = BTRFS_FS_TREE_OBJECTID;
3155         location.type = BTRFS_ROOT_ITEM_KEY;
3156         location.offset = 0;
3157
3158         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3159         if (IS_ERR(fs_info->fs_root)) {
3160                 err = PTR_ERR(fs_info->fs_root);
3161                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3162                 goto fail_qgroup;
3163         }
3164
3165         if (sb_rdonly(sb))
3166                 return 0;
3167
3168         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3169             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3170                 clear_free_space_tree = 1;
3171         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3172                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3173                 btrfs_warn(fs_info, "free space tree is invalid");
3174                 clear_free_space_tree = 1;
3175         }
3176
3177         if (clear_free_space_tree) {
3178                 btrfs_info(fs_info, "clearing free space tree");
3179                 ret = btrfs_clear_free_space_tree(fs_info);
3180                 if (ret) {
3181                         btrfs_warn(fs_info,
3182                                    "failed to clear free space tree: %d", ret);
3183                         close_ctree(fs_info);
3184                         return ret;
3185                 }
3186         }
3187
3188         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3189             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3190                 btrfs_info(fs_info, "creating free space tree");
3191                 ret = btrfs_create_free_space_tree(fs_info);
3192                 if (ret) {
3193                         btrfs_warn(fs_info,
3194                                 "failed to create free space tree: %d", ret);
3195                         close_ctree(fs_info);
3196                         return ret;
3197                 }
3198         }
3199
3200         down_read(&fs_info->cleanup_work_sem);
3201         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3202             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3203                 up_read(&fs_info->cleanup_work_sem);
3204                 close_ctree(fs_info);
3205                 return ret;
3206         }
3207         up_read(&fs_info->cleanup_work_sem);
3208
3209         ret = btrfs_resume_balance_async(fs_info);
3210         if (ret) {
3211                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3212                 close_ctree(fs_info);
3213                 return ret;
3214         }
3215
3216         ret = btrfs_resume_dev_replace_async(fs_info);
3217         if (ret) {
3218                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3219                 close_ctree(fs_info);
3220                 return ret;
3221         }
3222
3223         btrfs_qgroup_rescan_resume(fs_info);
3224
3225         if (!fs_info->uuid_root) {
3226                 btrfs_info(fs_info, "creating UUID tree");
3227                 ret = btrfs_create_uuid_tree(fs_info);
3228                 if (ret) {
3229                         btrfs_warn(fs_info,
3230                                 "failed to create the UUID tree: %d", ret);
3231                         close_ctree(fs_info);
3232                         return ret;
3233                 }
3234         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3235                    fs_info->generation !=
3236                                 btrfs_super_uuid_tree_generation(disk_super)) {
3237                 btrfs_info(fs_info, "checking UUID tree");
3238                 ret = btrfs_check_uuid_tree(fs_info);
3239                 if (ret) {
3240                         btrfs_warn(fs_info,
3241                                 "failed to check the UUID tree: %d", ret);
3242                         close_ctree(fs_info);
3243                         return ret;
3244                 }
3245         } else {
3246                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3247         }
3248         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3249
3250         /*
3251          * backuproot only affect mount behavior, and if open_ctree succeeded,
3252          * no need to keep the flag
3253          */
3254         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3255
3256         return 0;
3257
3258 fail_qgroup:
3259         btrfs_free_qgroup_config(fs_info);
3260 fail_trans_kthread:
3261         kthread_stop(fs_info->transaction_kthread);
3262         btrfs_cleanup_transaction(fs_info);
3263         btrfs_free_fs_roots(fs_info);
3264 fail_cleaner:
3265         kthread_stop(fs_info->cleaner_kthread);
3266
3267         /*
3268          * make sure we're done with the btree inode before we stop our
3269          * kthreads
3270          */
3271         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3272
3273 fail_sysfs:
3274         btrfs_sysfs_remove_mounted(fs_info);
3275
3276 fail_fsdev_sysfs:
3277         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3278
3279 fail_block_groups:
3280         btrfs_put_block_group_cache(fs_info);
3281
3282 fail_tree_roots:
3283         free_root_pointers(fs_info, 1);
3284         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3285
3286 fail_sb_buffer:
3287         btrfs_stop_all_workers(fs_info);
3288         btrfs_free_block_groups(fs_info);
3289 fail_alloc:
3290 fail_iput:
3291         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3292
3293         iput(fs_info->btree_inode);
3294 fail_bio_counter:
3295         percpu_counter_destroy(&fs_info->bio_counter);
3296 fail_delalloc_bytes:
3297         percpu_counter_destroy(&fs_info->delalloc_bytes);
3298 fail_dirty_metadata_bytes:
3299         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3300 fail_srcu:
3301         cleanup_srcu_struct(&fs_info->subvol_srcu);
3302 fail:
3303         btrfs_free_stripe_hash_table(fs_info);
3304         btrfs_close_devices(fs_info->fs_devices);
3305         return err;
3306
3307 recovery_tree_root:
3308         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3309                 goto fail_tree_roots;
3310
3311         free_root_pointers(fs_info, 0);
3312
3313         /* don't use the log in recovery mode, it won't be valid */
3314         btrfs_set_super_log_root(disk_super, 0);
3315
3316         /* we can't trust the free space cache either */
3317         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3318
3319         ret = next_root_backup(fs_info, fs_info->super_copy,
3320                                &num_backups_tried, &backup_index);
3321         if (ret == -1)
3322                 goto fail_block_groups;
3323         goto retry_root_backup;
3324 }
3325 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3326
3327 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3328 {
3329         if (uptodate) {
3330                 set_buffer_uptodate(bh);
3331         } else {
3332                 struct btrfs_device *device = (struct btrfs_device *)
3333                         bh->b_private;
3334
3335                 btrfs_warn_rl_in_rcu(device->fs_info,
3336                                 "lost page write due to IO error on %s",
3337                                           rcu_str_deref(device->name));
3338                 /* note, we don't set_buffer_write_io_error because we have
3339                  * our own ways of dealing with the IO errors
3340                  */
3341                 clear_buffer_uptodate(bh);
3342                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3343         }
3344         unlock_buffer(bh);
3345         put_bh(bh);
3346 }
3347
3348 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3349                         struct buffer_head **bh_ret)
3350 {
3351         struct buffer_head *bh;
3352         struct btrfs_super_block *super;
3353         u64 bytenr;
3354
3355         bytenr = btrfs_sb_offset(copy_num);
3356         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3357                 return -EINVAL;
3358
3359         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3360         /*
3361          * If we fail to read from the underlying devices, as of now
3362          * the best option we have is to mark it EIO.
3363          */
3364         if (!bh)
3365                 return -EIO;
3366
3367         super = (struct btrfs_super_block *)bh->b_data;
3368         if (btrfs_super_bytenr(super) != bytenr ||
3369                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3370                 brelse(bh);
3371                 return -EINVAL;
3372         }
3373
3374         *bh_ret = bh;
3375         return 0;
3376 }
3377
3378
3379 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3380 {
3381         struct buffer_head *bh;
3382         struct buffer_head *latest = NULL;
3383         struct btrfs_super_block *super;
3384         int i;
3385         u64 transid = 0;
3386         int ret = -EINVAL;
3387
3388         /* we would like to check all the supers, but that would make
3389          * a btrfs mount succeed after a mkfs from a different FS.
3390          * So, we need to add a special mount option to scan for
3391          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3392          */
3393         for (i = 0; i < 1; i++) {
3394                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3395                 if (ret)
3396                         continue;
3397
3398                 super = (struct btrfs_super_block *)bh->b_data;
3399
3400                 if (!latest || btrfs_super_generation(super) > transid) {
3401                         brelse(latest);
3402                         latest = bh;
3403                         transid = btrfs_super_generation(super);
3404                 } else {
3405                         brelse(bh);
3406                 }
3407         }
3408
3409         if (!latest)
3410                 return ERR_PTR(ret);
3411
3412         return latest;
3413 }
3414
3415 /*
3416  * Write superblock @sb to the @device. Do not wait for completion, all the
3417  * buffer heads we write are pinned.
3418  *
3419  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3420  * the expected device size at commit time. Note that max_mirrors must be
3421  * same for write and wait phases.
3422  *
3423  * Return number of errors when buffer head is not found or submission fails.
3424  */
3425 static int write_dev_supers(struct btrfs_device *device,
3426                             struct btrfs_super_block *sb, int max_mirrors)
3427 {
3428         struct buffer_head *bh;
3429         int i;
3430         int ret;
3431         int errors = 0;
3432         u32 crc;
3433         u64 bytenr;
3434         int op_flags;
3435
3436         if (max_mirrors == 0)
3437                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3438
3439         for (i = 0; i < max_mirrors; i++) {
3440                 bytenr = btrfs_sb_offset(i);
3441                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3442                     device->commit_total_bytes)
3443                         break;
3444
3445                 btrfs_set_super_bytenr(sb, bytenr);
3446
3447                 crc = ~(u32)0;
3448                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3449                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3450                 btrfs_csum_final(crc, sb->csum);
3451
3452                 /* One reference for us, and we leave it for the caller */
3453                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3454                               BTRFS_SUPER_INFO_SIZE);
3455                 if (!bh) {
3456                         btrfs_err(device->fs_info,
3457                             "couldn't get super buffer head for bytenr %llu",
3458                             bytenr);
3459                         errors++;
3460                         continue;
3461                 }
3462
3463                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3464
3465                 /* one reference for submit_bh */
3466                 get_bh(bh);
3467
3468                 set_buffer_uptodate(bh);
3469                 lock_buffer(bh);
3470                 bh->b_end_io = btrfs_end_buffer_write_sync;
3471                 bh->b_private = device;
3472
3473                 /*
3474                  * we fua the first super.  The others we allow
3475                  * to go down lazy.
3476                  */
3477                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3478                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3479                         op_flags |= REQ_FUA;
3480                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3481                 if (ret)
3482                         errors++;
3483         }
3484         return errors < i ? 0 : -1;
3485 }
3486
3487 /*
3488  * Wait for write completion of superblocks done by write_dev_supers,
3489  * @max_mirrors same for write and wait phases.
3490  *
3491  * Return number of errors when buffer head is not found or not marked up to
3492  * date.
3493  */
3494 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3495 {
3496         struct buffer_head *bh;
3497         int i;
3498         int errors = 0;
3499         bool primary_failed = false;
3500         u64 bytenr;
3501
3502         if (max_mirrors == 0)
3503                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3504
3505         for (i = 0; i < max_mirrors; i++) {
3506                 bytenr = btrfs_sb_offset(i);
3507                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3508                     device->commit_total_bytes)
3509                         break;
3510
3511                 bh = __find_get_block(device->bdev,
3512                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3513                                       BTRFS_SUPER_INFO_SIZE);
3514                 if (!bh) {
3515                         errors++;
3516                         if (i == 0)
3517                                 primary_failed = true;
3518                         continue;
3519                 }
3520                 wait_on_buffer(bh);
3521                 if (!buffer_uptodate(bh)) {
3522                         errors++;
3523                         if (i == 0)
3524                                 primary_failed = true;
3525                 }
3526
3527                 /* drop our reference */
3528                 brelse(bh);
3529
3530                 /* drop the reference from the writing run */
3531                 brelse(bh);
3532         }
3533
3534         /* log error, force error return */
3535         if (primary_failed) {
3536                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3537                           device->devid);
3538                 return -1;
3539         }
3540
3541         return errors < i ? 0 : -1;
3542 }
3543
3544 /*
3545  * endio for the write_dev_flush, this will wake anyone waiting
3546  * for the barrier when it is done
3547  */
3548 static void btrfs_end_empty_barrier(struct bio *bio)
3549 {
3550         complete(bio->bi_private);
3551 }
3552
3553 /*
3554  * Submit a flush request to the device if it supports it. Error handling is
3555  * done in the waiting counterpart.
3556  */
3557 static void write_dev_flush(struct btrfs_device *device)
3558 {
3559         struct request_queue *q = bdev_get_queue(device->bdev);
3560         struct bio *bio = device->flush_bio;
3561
3562         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3563                 return;
3564
3565         bio_reset(bio);
3566         bio->bi_end_io = btrfs_end_empty_barrier;
3567         bio_set_dev(bio, device->bdev);
3568         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3569         init_completion(&device->flush_wait);
3570         bio->bi_private = &device->flush_wait;
3571
3572         btrfsic_submit_bio(bio);
3573         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3574 }
3575
3576 /*
3577  * If the flush bio has been submitted by write_dev_flush, wait for it.
3578  */
3579 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3580 {
3581         struct bio *bio = device->flush_bio;
3582
3583         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3584                 return BLK_STS_OK;
3585
3586         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3587         wait_for_completion_io(&device->flush_wait);
3588
3589         return bio->bi_status;
3590 }
3591
3592 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3593 {
3594         if (!btrfs_check_rw_degradable(fs_info, NULL))
3595                 return -EIO;
3596         return 0;
3597 }
3598
3599 /*
3600  * send an empty flush down to each device in parallel,
3601  * then wait for them
3602  */
3603 static int barrier_all_devices(struct btrfs_fs_info *info)
3604 {
3605         struct list_head *head;
3606         struct btrfs_device *dev;
3607         int errors_wait = 0;
3608         blk_status_t ret;
3609
3610         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3611         /* send down all the barriers */
3612         head = &info->fs_devices->devices;
3613         list_for_each_entry(dev, head, dev_list) {
3614                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3615                         continue;
3616                 if (!dev->bdev)
3617                         continue;
3618                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3619                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3620                         continue;
3621
3622                 write_dev_flush(dev);
3623                 dev->last_flush_error = BLK_STS_OK;
3624         }
3625
3626         /* wait for all the barriers */
3627         list_for_each_entry(dev, head, dev_list) {
3628                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3629                         continue;
3630                 if (!dev->bdev) {
3631                         errors_wait++;
3632                         continue;
3633                 }
3634                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3635                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3636                         continue;
3637
3638                 ret = wait_dev_flush(dev);
3639                 if (ret) {
3640                         dev->last_flush_error = ret;
3641                         btrfs_dev_stat_inc_and_print(dev,
3642                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3643                         errors_wait++;
3644                 }
3645         }
3646
3647         if (errors_wait) {
3648                 /*
3649                  * At some point we need the status of all disks
3650                  * to arrive at the volume status. So error checking
3651                  * is being pushed to a separate loop.
3652                  */
3653                 return check_barrier_error(info);
3654         }
3655         return 0;
3656 }
3657
3658 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3659 {
3660         int raid_type;
3661         int min_tolerated = INT_MAX;
3662
3663         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3664             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3665                 min_tolerated = min(min_tolerated,
3666                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3667                                     tolerated_failures);
3668
3669         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3670                 if (raid_type == BTRFS_RAID_SINGLE)
3671                         continue;
3672                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3673                         continue;
3674                 min_tolerated = min(min_tolerated,
3675                                     btrfs_raid_array[raid_type].
3676                                     tolerated_failures);
3677         }
3678
3679         if (min_tolerated == INT_MAX) {
3680                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3681                 min_tolerated = 0;
3682         }
3683
3684         return min_tolerated;
3685 }
3686
3687 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3688 {
3689         struct list_head *head;
3690         struct btrfs_device *dev;
3691         struct btrfs_super_block *sb;
3692         struct btrfs_dev_item *dev_item;
3693         int ret;
3694         int do_barriers;
3695         int max_errors;
3696         int total_errors = 0;
3697         u64 flags;
3698
3699         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3700
3701         /*
3702          * max_mirrors == 0 indicates we're from commit_transaction,
3703          * not from fsync where the tree roots in fs_info have not
3704          * been consistent on disk.
3705          */
3706         if (max_mirrors == 0)
3707                 backup_super_roots(fs_info);
3708
3709         sb = fs_info->super_for_commit;
3710         dev_item = &sb->dev_item;
3711
3712         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3713         head = &fs_info->fs_devices->devices;
3714         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3715
3716         if (do_barriers) {
3717                 ret = barrier_all_devices(fs_info);
3718                 if (ret) {
3719                         mutex_unlock(
3720                                 &fs_info->fs_devices->device_list_mutex);
3721                         btrfs_handle_fs_error(fs_info, ret,
3722                                               "errors while submitting device barriers.");
3723                         return ret;
3724                 }
3725         }
3726
3727         list_for_each_entry(dev, head, dev_list) {
3728                 if (!dev->bdev) {
3729                         total_errors++;
3730                         continue;
3731                 }
3732                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3733                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3734                         continue;
3735
3736                 btrfs_set_stack_device_generation(dev_item, 0);
3737                 btrfs_set_stack_device_type(dev_item, dev->type);
3738                 btrfs_set_stack_device_id(dev_item, dev->devid);
3739                 btrfs_set_stack_device_total_bytes(dev_item,
3740                                                    dev->commit_total_bytes);
3741                 btrfs_set_stack_device_bytes_used(dev_item,
3742                                                   dev->commit_bytes_used);
3743                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3744                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3745                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3746                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3747                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3748
3749                 flags = btrfs_super_flags(sb);
3750                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3751
3752                 ret = write_dev_supers(dev, sb, max_mirrors);
3753                 if (ret)
3754                         total_errors++;
3755         }
3756         if (total_errors > max_errors) {
3757                 btrfs_err(fs_info, "%d errors while writing supers",
3758                           total_errors);
3759                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3760
3761                 /* FUA is masked off if unsupported and can't be the reason */
3762                 btrfs_handle_fs_error(fs_info, -EIO,
3763                                       "%d errors while writing supers",
3764                                       total_errors);
3765                 return -EIO;
3766         }
3767
3768         total_errors = 0;
3769         list_for_each_entry(dev, head, dev_list) {
3770                 if (!dev->bdev)
3771                         continue;
3772                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3773                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3774                         continue;
3775
3776                 ret = wait_dev_supers(dev, max_mirrors);
3777                 if (ret)
3778                         total_errors++;
3779         }
3780         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3781         if (total_errors > max_errors) {
3782                 btrfs_handle_fs_error(fs_info, -EIO,
3783                                       "%d errors while writing supers",
3784                                       total_errors);
3785                 return -EIO;
3786         }
3787         return 0;
3788 }
3789
3790 /* Drop a fs root from the radix tree and free it. */
3791 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3792                                   struct btrfs_root *root)
3793 {
3794         spin_lock(&fs_info->fs_roots_radix_lock);
3795         radix_tree_delete(&fs_info->fs_roots_radix,
3796                           (unsigned long)root->root_key.objectid);
3797         spin_unlock(&fs_info->fs_roots_radix_lock);
3798
3799         if (btrfs_root_refs(&root->root_item) == 0)
3800                 synchronize_srcu(&fs_info->subvol_srcu);
3801
3802         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3803                 btrfs_free_log(NULL, root);
3804                 if (root->reloc_root) {
3805                         free_extent_buffer(root->reloc_root->node);
3806                         free_extent_buffer(root->reloc_root->commit_root);
3807                         btrfs_put_fs_root(root->reloc_root);
3808                         root->reloc_root = NULL;
3809                 }
3810         }
3811
3812         if (root->free_ino_pinned)
3813                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3814         if (root->free_ino_ctl)
3815                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3816         free_fs_root(root);
3817 }
3818
3819 static void free_fs_root(struct btrfs_root *root)
3820 {
3821         iput(root->ino_cache_inode);
3822         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3823         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3824         root->orphan_block_rsv = NULL;
3825         if (root->anon_dev)
3826                 free_anon_bdev(root->anon_dev);
3827         if (root->subv_writers)
3828                 btrfs_free_subvolume_writers(root->subv_writers);
3829         free_extent_buffer(root->node);
3830         free_extent_buffer(root->commit_root);
3831         kfree(root->free_ino_ctl);
3832         kfree(root->free_ino_pinned);
3833         kfree(root->name);
3834         btrfs_put_fs_root(root);
3835 }
3836
3837 void btrfs_free_fs_root(struct btrfs_root *root)
3838 {
3839         free_fs_root(root);
3840 }
3841
3842 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3843 {
3844         u64 root_objectid = 0;
3845         struct btrfs_root *gang[8];
3846         int i = 0;
3847         int err = 0;
3848         unsigned int ret = 0;
3849         int index;
3850
3851         while (1) {
3852                 index = srcu_read_lock(&fs_info->subvol_srcu);
3853                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3854                                              (void **)gang, root_objectid,
3855                                              ARRAY_SIZE(gang));
3856                 if (!ret) {
3857                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3858                         break;
3859                 }
3860                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3861
3862                 for (i = 0; i < ret; i++) {
3863                         /* Avoid to grab roots in dead_roots */
3864                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3865                                 gang[i] = NULL;
3866                                 continue;
3867                         }
3868                         /* grab all the search result for later use */
3869                         gang[i] = btrfs_grab_fs_root(gang[i]);
3870                 }
3871                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3872
3873                 for (i = 0; i < ret; i++) {
3874                         if (!gang[i])
3875                                 continue;
3876                         root_objectid = gang[i]->root_key.objectid;
3877                         err = btrfs_orphan_cleanup(gang[i]);
3878                         if (err)
3879                                 break;
3880                         btrfs_put_fs_root(gang[i]);
3881                 }
3882                 root_objectid++;
3883         }
3884
3885         /* release the uncleaned roots due to error */
3886         for (; i < ret; i++) {
3887                 if (gang[i])
3888                         btrfs_put_fs_root(gang[i]);
3889         }
3890         return err;
3891 }
3892
3893 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3894 {
3895         struct btrfs_root *root = fs_info->tree_root;
3896         struct btrfs_trans_handle *trans;
3897
3898         mutex_lock(&fs_info->cleaner_mutex);
3899         btrfs_run_delayed_iputs(fs_info);
3900         mutex_unlock(&fs_info->cleaner_mutex);
3901         wake_up_process(fs_info->cleaner_kthread);
3902
3903         /* wait until ongoing cleanup work done */
3904         down_write(&fs_info->cleanup_work_sem);
3905         up_write(&fs_info->cleanup_work_sem);
3906
3907         trans = btrfs_join_transaction(root);
3908         if (IS_ERR(trans))
3909                 return PTR_ERR(trans);
3910         return btrfs_commit_transaction(trans);
3911 }
3912
3913 void close_ctree(struct btrfs_fs_info *fs_info)
3914 {
3915         struct btrfs_root *root = fs_info->tree_root;
3916         int ret;
3917
3918         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3919
3920         /* wait for the qgroup rescan worker to stop */
3921         btrfs_qgroup_wait_for_completion(fs_info, false);
3922
3923         /* wait for the uuid_scan task to finish */
3924         down(&fs_info->uuid_tree_rescan_sem);
3925         /* avoid complains from lockdep et al., set sem back to initial state */
3926         up(&fs_info->uuid_tree_rescan_sem);
3927
3928         /* pause restriper - we want to resume on mount */
3929         btrfs_pause_balance(fs_info);
3930
3931         btrfs_dev_replace_suspend_for_unmount(fs_info);
3932
3933         btrfs_scrub_cancel(fs_info);
3934
3935         /* wait for any defraggers to finish */
3936         wait_event(fs_info->transaction_wait,
3937                    (atomic_read(&fs_info->defrag_running) == 0));
3938
3939         /* clear out the rbtree of defraggable inodes */
3940         btrfs_cleanup_defrag_inodes(fs_info);
3941
3942         cancel_work_sync(&fs_info->async_reclaim_work);
3943
3944         if (!sb_rdonly(fs_info->sb)) {
3945                 /*
3946                  * If the cleaner thread is stopped and there are
3947                  * block groups queued for removal, the deletion will be
3948                  * skipped when we quit the cleaner thread.
3949                  */
3950                 btrfs_delete_unused_bgs(fs_info);
3951
3952                 ret = btrfs_commit_super(fs_info);
3953                 if (ret)
3954                         btrfs_err(fs_info, "commit super ret %d", ret);
3955         }
3956
3957         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3958             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3959                 btrfs_error_commit_super(fs_info);
3960
3961         kthread_stop(fs_info->transaction_kthread);
3962         kthread_stop(fs_info->cleaner_kthread);
3963
3964         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3965
3966         btrfs_free_qgroup_config(fs_info);
3967         ASSERT(list_empty(&fs_info->delalloc_roots));
3968
3969         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3970                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3971                        percpu_counter_sum(&fs_info->delalloc_bytes));
3972         }
3973
3974         btrfs_sysfs_remove_mounted(fs_info);
3975         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3976
3977         btrfs_free_fs_roots(fs_info);
3978
3979         btrfs_put_block_group_cache(fs_info);
3980
3981         /*
3982          * we must make sure there is not any read request to
3983          * submit after we stopping all workers.
3984          */
3985         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3986         btrfs_stop_all_workers(fs_info);
3987
3988         btrfs_free_block_groups(fs_info);
3989
3990         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3991         free_root_pointers(fs_info, 1);
3992
3993         iput(fs_info->btree_inode);
3994
3995 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3996         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3997                 btrfsic_unmount(fs_info->fs_devices);
3998 #endif
3999
4000         btrfs_close_devices(fs_info->fs_devices);
4001         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4002
4003         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4004         percpu_counter_destroy(&fs_info->delalloc_bytes);
4005         percpu_counter_destroy(&fs_info->bio_counter);
4006         cleanup_srcu_struct(&fs_info->subvol_srcu);
4007
4008         btrfs_free_stripe_hash_table(fs_info);
4009         btrfs_free_ref_cache(fs_info);
4010
4011         __btrfs_free_block_rsv(root->orphan_block_rsv);
4012         root->orphan_block_rsv = NULL;
4013
4014         while (!list_empty(&fs_info->pinned_chunks)) {
4015                 struct extent_map *em;
4016
4017                 em = list_first_entry(&fs_info->pinned_chunks,
4018                                       struct extent_map, list);
4019                 list_del_init(&em->list);
4020                 free_extent_map(em);
4021         }
4022 }
4023
4024 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4025                           int atomic)
4026 {
4027         int ret;
4028         struct inode *btree_inode = buf->pages[0]->mapping->host;
4029
4030         ret = extent_buffer_uptodate(buf);
4031         if (!ret)
4032                 return ret;
4033
4034         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4035                                     parent_transid, atomic);
4036         if (ret == -EAGAIN)
4037                 return ret;
4038         return !ret;
4039 }
4040
4041 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4042 {
4043         struct btrfs_fs_info *fs_info;
4044         struct btrfs_root *root;
4045         u64 transid = btrfs_header_generation(buf);
4046         int was_dirty;
4047
4048 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4049         /*
4050          * This is a fast path so only do this check if we have sanity tests
4051          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4052          * outside of the sanity tests.
4053          */
4054         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4055                 return;
4056 #endif
4057         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4058         fs_info = root->fs_info;
4059         btrfs_assert_tree_locked(buf);
4060         if (transid != fs_info->generation)
4061                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4062                         buf->start, transid, fs_info->generation);
4063         was_dirty = set_extent_buffer_dirty(buf);
4064         if (!was_dirty)
4065                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4066                                          buf->len,
4067                                          fs_info->dirty_metadata_batch);
4068 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4069         /*
4070          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4071          * but item data not updated.
4072          * So here we should only check item pointers, not item data.
4073          */
4074         if (btrfs_header_level(buf) == 0 &&
4075             btrfs_check_leaf_relaxed(fs_info, buf)) {
4076                 btrfs_print_leaf(buf);
4077                 ASSERT(0);
4078         }
4079 #endif
4080 }
4081
4082 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4083                                         int flush_delayed)
4084 {
4085         /*
4086          * looks as though older kernels can get into trouble with
4087          * this code, they end up stuck in balance_dirty_pages forever
4088          */
4089         int ret;
4090
4091         if (current->flags & PF_MEMALLOC)
4092                 return;
4093
4094         if (flush_delayed)
4095                 btrfs_balance_delayed_items(fs_info);
4096
4097         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4098                                      BTRFS_DIRTY_METADATA_THRESH);
4099         if (ret > 0) {
4100                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4101         }
4102 }
4103
4104 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4105 {
4106         __btrfs_btree_balance_dirty(fs_info, 1);
4107 }
4108
4109 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4110 {
4111         __btrfs_btree_balance_dirty(fs_info, 0);
4112 }
4113
4114 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4115                       struct btrfs_key *first_key)
4116 {
4117         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4118         struct btrfs_fs_info *fs_info = root->fs_info;
4119
4120         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4121                                               level, first_key);
4122 }
4123
4124 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4125 {
4126         /* cleanup FS via transaction */
4127         btrfs_cleanup_transaction(fs_info);
4128
4129         mutex_lock(&fs_info->cleaner_mutex);
4130         btrfs_run_delayed_iputs(fs_info);
4131         mutex_unlock(&fs_info->cleaner_mutex);
4132
4133         down_write(&fs_info->cleanup_work_sem);
4134         up_write(&fs_info->cleanup_work_sem);
4135 }
4136
4137 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4138 {
4139         struct btrfs_ordered_extent *ordered;
4140
4141         spin_lock(&root->ordered_extent_lock);
4142         /*
4143          * This will just short circuit the ordered completion stuff which will
4144          * make sure the ordered extent gets properly cleaned up.
4145          */
4146         list_for_each_entry(ordered, &root->ordered_extents,
4147                             root_extent_list)
4148                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4149         spin_unlock(&root->ordered_extent_lock);
4150 }
4151
4152 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4153 {
4154         struct btrfs_root *root;
4155         struct list_head splice;
4156
4157         INIT_LIST_HEAD(&splice);
4158
4159         spin_lock(&fs_info->ordered_root_lock);
4160         list_splice_init(&fs_info->ordered_roots, &splice);
4161         while (!list_empty(&splice)) {
4162                 root = list_first_entry(&splice, struct btrfs_root,
4163                                         ordered_root);
4164                 list_move_tail(&root->ordered_root,
4165                                &fs_info->ordered_roots);
4166
4167                 spin_unlock(&fs_info->ordered_root_lock);
4168                 btrfs_destroy_ordered_extents(root);
4169
4170                 cond_resched();
4171                 spin_lock(&fs_info->ordered_root_lock);
4172         }
4173         spin_unlock(&fs_info->ordered_root_lock);
4174 }
4175
4176 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4177                                       struct btrfs_fs_info *fs_info)
4178 {
4179         struct rb_node *node;
4180         struct btrfs_delayed_ref_root *delayed_refs;
4181         struct btrfs_delayed_ref_node *ref;
4182         int ret = 0;
4183
4184         delayed_refs = &trans->delayed_refs;
4185
4186         spin_lock(&delayed_refs->lock);
4187         if (atomic_read(&delayed_refs->num_entries) == 0) {
4188                 spin_unlock(&delayed_refs->lock);
4189                 btrfs_info(fs_info, "delayed_refs has NO entry");
4190                 return ret;
4191         }
4192
4193         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4194                 struct btrfs_delayed_ref_head *head;
4195                 struct rb_node *n;
4196                 bool pin_bytes = false;
4197
4198                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4199                                 href_node);
4200                 if (!mutex_trylock(&head->mutex)) {
4201                         refcount_inc(&head->refs);
4202                         spin_unlock(&delayed_refs->lock);
4203
4204                         mutex_lock(&head->mutex);
4205                         mutex_unlock(&head->mutex);
4206                         btrfs_put_delayed_ref_head(head);
4207                         spin_lock(&delayed_refs->lock);
4208                         continue;
4209                 }
4210                 spin_lock(&head->lock);
4211                 while ((n = rb_first(&head->ref_tree)) != NULL) {
4212                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4213                                        ref_node);
4214                         ref->in_tree = 0;
4215                         rb_erase(&ref->ref_node, &head->ref_tree);
4216                         RB_CLEAR_NODE(&ref->ref_node);
4217                         if (!list_empty(&ref->add_list))
4218                                 list_del(&ref->add_list);
4219                         atomic_dec(&delayed_refs->num_entries);
4220                         btrfs_put_delayed_ref(ref);
4221                 }
4222                 if (head->must_insert_reserved)
4223                         pin_bytes = true;
4224                 btrfs_free_delayed_extent_op(head->extent_op);
4225                 delayed_refs->num_heads--;
4226                 if (head->processing == 0)
4227                         delayed_refs->num_heads_ready--;
4228                 atomic_dec(&delayed_refs->num_entries);
4229                 rb_erase(&head->href_node, &delayed_refs->href_root);
4230                 RB_CLEAR_NODE(&head->href_node);
4231                 spin_unlock(&head->lock);
4232                 spin_unlock(&delayed_refs->lock);
4233                 mutex_unlock(&head->mutex);
4234
4235                 if (pin_bytes)
4236                         btrfs_pin_extent(fs_info, head->bytenr,
4237                                          head->num_bytes, 1);
4238                 btrfs_put_delayed_ref_head(head);
4239                 cond_resched();
4240                 spin_lock(&delayed_refs->lock);
4241         }
4242
4243         spin_unlock(&delayed_refs->lock);
4244
4245         return ret;
4246 }
4247
4248 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4249 {
4250         struct btrfs_inode *btrfs_inode;
4251         struct list_head splice;
4252
4253         INIT_LIST_HEAD(&splice);
4254
4255         spin_lock(&root->delalloc_lock);
4256         list_splice_init(&root->delalloc_inodes, &splice);
4257
4258         while (!list_empty(&splice)) {
4259                 struct inode *inode = NULL;
4260                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4261                                                delalloc_inodes);
4262                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4263                 spin_unlock(&root->delalloc_lock);
4264
4265                 /*
4266                  * Make sure we get a live inode and that it'll not disappear
4267                  * meanwhile.
4268                  */
4269                 inode = igrab(&btrfs_inode->vfs_inode);
4270                 if (inode) {
4271                         invalidate_inode_pages2(inode->i_mapping);
4272                         iput(inode);
4273                 }
4274                 spin_lock(&root->delalloc_lock);
4275         }
4276         spin_unlock(&root->delalloc_lock);
4277 }
4278
4279 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4280 {
4281         struct btrfs_root *root;
4282         struct list_head splice;
4283
4284         INIT_LIST_HEAD(&splice);
4285
4286         spin_lock(&fs_info->delalloc_root_lock);
4287         list_splice_init(&fs_info->delalloc_roots, &splice);
4288         while (!list_empty(&splice)) {
4289                 root = list_first_entry(&splice, struct btrfs_root,
4290                                          delalloc_root);
4291                 root = btrfs_grab_fs_root(root);
4292                 BUG_ON(!root);
4293                 spin_unlock(&fs_info->delalloc_root_lock);
4294
4295                 btrfs_destroy_delalloc_inodes(root);
4296                 btrfs_put_fs_root(root);
4297
4298                 spin_lock(&fs_info->delalloc_root_lock);
4299         }
4300         spin_unlock(&fs_info->delalloc_root_lock);
4301 }
4302
4303 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4304                                         struct extent_io_tree *dirty_pages,
4305                                         int mark)
4306 {
4307         int ret;
4308         struct extent_buffer *eb;
4309         u64 start = 0;
4310         u64 end;
4311
4312         while (1) {
4313                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4314                                             mark, NULL);
4315                 if (ret)
4316                         break;
4317
4318                 clear_extent_bits(dirty_pages, start, end, mark);
4319                 while (start <= end) {
4320                         eb = find_extent_buffer(fs_info, start);
4321                         start += fs_info->nodesize;
4322                         if (!eb)
4323                                 continue;
4324                         wait_on_extent_buffer_writeback(eb);
4325
4326                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4327                                                &eb->bflags))
4328                                 clear_extent_buffer_dirty(eb);
4329                         free_extent_buffer_stale(eb);
4330                 }
4331         }
4332
4333         return ret;
4334 }
4335
4336 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4337                                        struct extent_io_tree *pinned_extents)
4338 {
4339         struct extent_io_tree *unpin;
4340         u64 start;
4341         u64 end;
4342         int ret;
4343         bool loop = true;
4344
4345         unpin = pinned_extents;
4346 again:
4347         while (1) {
4348                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4349                                             EXTENT_DIRTY, NULL);
4350                 if (ret)
4351                         break;
4352
4353                 clear_extent_dirty(unpin, start, end);
4354                 btrfs_error_unpin_extent_range(fs_info, start, end);
4355                 cond_resched();
4356         }
4357
4358         if (loop) {
4359                 if (unpin == &fs_info->freed_extents[0])
4360                         unpin = &fs_info->freed_extents[1];
4361                 else
4362                         unpin = &fs_info->freed_extents[0];
4363                 loop = false;
4364                 goto again;
4365         }
4366
4367         return 0;
4368 }
4369
4370 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4371 {
4372         struct inode *inode;
4373
4374         inode = cache->io_ctl.inode;
4375         if (inode) {
4376                 invalidate_inode_pages2(inode->i_mapping);
4377                 BTRFS_I(inode)->generation = 0;
4378                 cache->io_ctl.inode = NULL;
4379                 iput(inode);
4380         }
4381         btrfs_put_block_group(cache);
4382 }
4383
4384 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4385                              struct btrfs_fs_info *fs_info)
4386 {
4387         struct btrfs_block_group_cache *cache;
4388
4389         spin_lock(&cur_trans->dirty_bgs_lock);
4390         while (!list_empty(&cur_trans->dirty_bgs)) {
4391                 cache = list_first_entry(&cur_trans->dirty_bgs,
4392                                          struct btrfs_block_group_cache,
4393                                          dirty_list);
4394
4395                 if (!list_empty(&cache->io_list)) {
4396                         spin_unlock(&cur_trans->dirty_bgs_lock);
4397                         list_del_init(&cache->io_list);
4398                         btrfs_cleanup_bg_io(cache);
4399                         spin_lock(&cur_trans->dirty_bgs_lock);
4400                 }
4401
4402                 list_del_init(&cache->dirty_list);
4403                 spin_lock(&cache->lock);
4404                 cache->disk_cache_state = BTRFS_DC_ERROR;
4405                 spin_unlock(&cache->lock);
4406
4407                 spin_unlock(&cur_trans->dirty_bgs_lock);
4408                 btrfs_put_block_group(cache);
4409                 spin_lock(&cur_trans->dirty_bgs_lock);
4410         }
4411         spin_unlock(&cur_trans->dirty_bgs_lock);
4412
4413         /*
4414          * Refer to the definition of io_bgs member for details why it's safe
4415          * to use it without any locking
4416          */
4417         while (!list_empty(&cur_trans->io_bgs)) {
4418                 cache = list_first_entry(&cur_trans->io_bgs,
4419                                          struct btrfs_block_group_cache,
4420                                          io_list);
4421
4422                 list_del_init(&cache->io_list);
4423                 spin_lock(&cache->lock);
4424                 cache->disk_cache_state = BTRFS_DC_ERROR;
4425                 spin_unlock(&cache->lock);
4426                 btrfs_cleanup_bg_io(cache);
4427         }
4428 }
4429
4430 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4431                                    struct btrfs_fs_info *fs_info)
4432 {
4433         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4434         ASSERT(list_empty(&cur_trans->dirty_bgs));
4435         ASSERT(list_empty(&cur_trans->io_bgs));
4436
4437         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4438
4439         cur_trans->state = TRANS_STATE_COMMIT_START;
4440         wake_up(&fs_info->transaction_blocked_wait);
4441
4442         cur_trans->state = TRANS_STATE_UNBLOCKED;
4443         wake_up(&fs_info->transaction_wait);
4444
4445         btrfs_destroy_delayed_inodes(fs_info);
4446         btrfs_assert_delayed_root_empty(fs_info);
4447
4448         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4449                                      EXTENT_DIRTY);
4450         btrfs_destroy_pinned_extent(fs_info,
4451                                     fs_info->pinned_extents);
4452
4453         cur_trans->state =TRANS_STATE_COMPLETED;
4454         wake_up(&cur_trans->commit_wait);
4455 }
4456
4457 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4458 {
4459         struct btrfs_transaction *t;
4460
4461         mutex_lock(&fs_info->transaction_kthread_mutex);
4462
4463         spin_lock(&fs_info->trans_lock);
4464         while (!list_empty(&fs_info->trans_list)) {
4465                 t = list_first_entry(&fs_info->trans_list,
4466                                      struct btrfs_transaction, list);
4467                 if (t->state >= TRANS_STATE_COMMIT_START) {
4468                         refcount_inc(&t->use_count);
4469                         spin_unlock(&fs_info->trans_lock);
4470                         btrfs_wait_for_commit(fs_info, t->transid);
4471                         btrfs_put_transaction(t);
4472                         spin_lock(&fs_info->trans_lock);
4473                         continue;
4474                 }
4475                 if (t == fs_info->running_transaction) {
4476                         t->state = TRANS_STATE_COMMIT_DOING;
4477                         spin_unlock(&fs_info->trans_lock);
4478                         /*
4479                          * We wait for 0 num_writers since we don't hold a trans
4480                          * handle open currently for this transaction.
4481                          */
4482                         wait_event(t->writer_wait,
4483                                    atomic_read(&t->num_writers) == 0);
4484                 } else {
4485                         spin_unlock(&fs_info->trans_lock);
4486                 }
4487                 btrfs_cleanup_one_transaction(t, fs_info);
4488
4489                 spin_lock(&fs_info->trans_lock);
4490                 if (t == fs_info->running_transaction)
4491                         fs_info->running_transaction = NULL;
4492                 list_del_init(&t->list);
4493                 spin_unlock(&fs_info->trans_lock);
4494
4495                 btrfs_put_transaction(t);
4496                 trace_btrfs_transaction_commit(fs_info->tree_root);
4497                 spin_lock(&fs_info->trans_lock);
4498         }
4499         spin_unlock(&fs_info->trans_lock);
4500         btrfs_destroy_all_ordered_extents(fs_info);
4501         btrfs_destroy_delayed_inodes(fs_info);
4502         btrfs_assert_delayed_root_empty(fs_info);
4503         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4504         btrfs_destroy_all_delalloc_inodes(fs_info);
4505         mutex_unlock(&fs_info->transaction_kthread_mutex);
4506
4507         return 0;
4508 }
4509
4510 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4511 {
4512         struct inode *inode = private_data;
4513         return btrfs_sb(inode->i_sb);
4514 }
4515
4516 static const struct extent_io_ops btree_extent_io_ops = {
4517         /* mandatory callbacks */
4518         .submit_bio_hook = btree_submit_bio_hook,
4519         .readpage_end_io_hook = btree_readpage_end_io_hook,
4520         /* note we're sharing with inode.c for the merge bio hook */
4521         .merge_bio_hook = btrfs_merge_bio_hook,
4522         .readpage_io_failed_hook = btree_io_failed_hook,
4523         .set_range_writeback = btrfs_set_range_writeback,
4524         .tree_fs_info = btree_fs_info,
4525
4526         /* optional callbacks */
4527 };