fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <asm/unaligned.h>
  21 #include "ctree.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24 #include "btrfs_inode.h"
  25 #include "volumes.h"
  26 #include "print-tree.h"
  27 #include "locking.h"
  28 #include "tree-log.h"
  29 #include "free-space-cache.h"
  30 #include "free-space-tree.h"
  31 #include "inode-map.h"
  32 #include "check-integrity.h"
  33 #include "rcu-string.h"
  34 #include "dev-replace.h"
  35 #include "raid56.h"
  36 #include "sysfs.h"
  37 #include "qgroup.h"
  38 #include "compression.h"
  39 #include "tree-checker.h"
  40 #include "ref-verify.h"
  41
  42 #ifdef CONFIG_X86
  43 #include <asm/cpufeature.h>
  44 #endif
  45
  46 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  47                                  BTRFS_HEADER_FLAG_RELOC |\
  48                                  BTRFS_SUPER_FLAG_ERROR |\
  49                                  BTRFS_SUPER_FLAG_SEEDING |\
  50                                  BTRFS_SUPER_FLAG_METADUMP |\
  51                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  52
  53 static const struct extent_io_ops btree_extent_io_ops;
  54 static void end_workqueue_fn(struct btrfs_work *work);
  55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  57                                       struct btrfs_fs_info *fs_info);
  58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  60                                         struct extent_io_tree *dirty_pages,
  61                                         int mark);
  62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  63                                        struct extent_io_tree *pinned_extents);
  64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  66
  67 /*
  68  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  69  * is complete.  This is used during reads to verify checksums, and it is used
  70  * by writes to insert metadata for new file extents after IO is complete.
  71  */
  72 struct btrfs_end_io_wq {
  73         struct bio *bio;
  74         bio_end_io_t *end_io;
  75         void *private;
  76         struct btrfs_fs_info *info;
  77         blk_status_t status;
  78         enum btrfs_wq_endio_type metadata;
  79         struct btrfs_work work;
  80 };
  81
  82 static struct kmem_cache *btrfs_end_io_wq_cache;
  83
  84 int __init btrfs_end_io_wq_init(void)
  85 {
  86         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  87                                         sizeof(struct btrfs_end_io_wq),
  88                                         0,
  89                                         SLAB_MEM_SPREAD,
  90                                         NULL);
  91         if (!btrfs_end_io_wq_cache)
  92                 return -ENOMEM;
  93         return 0;
  94 }
  95
  96 void __cold btrfs_end_io_wq_exit(void)
  97 {
  98         kmem_cache_destroy(btrfs_end_io_wq_cache);
  99 }
 100
 101 /*
 102  * async submit bios are used to offload expensive checksumming
 103  * onto the worker threads.  They checksum file and metadata bios
 104  * just before they are sent down the IO stack.
 105  */
 106 struct async_submit_bio {
 107         void *private_data;
 108         struct bio *bio;
 109         extent_submit_bio_start_t *submit_bio_start;
 110         int mirror_num;
 111         /*
 112          * bio_offset is optional, can be used if the pages in the bio
 113          * can't tell us where in the file the bio should go
 114          */
 115         u64 bio_offset;
 116         struct btrfs_work work;
 117         blk_status_t status;
 118 };
 119
 120 /*
 121  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 122  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 123  * the level the eb occupies in the tree.
 124  *
 125  * Different roots are used for different purposes and may nest inside each
 126  * other and they require separate keysets.  As lockdep keys should be
 127  * static, assign keysets according to the purpose of the root as indicated
 128  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 129  * roots have separate keysets.
 130  *
 131  * Lock-nesting across peer nodes is always done with the immediate parent
 132  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 133  * subclass to avoid triggering lockdep warning in such cases.
 134  *
 135  * The key is set by the readpage_end_io_hook after the buffer has passed
 136  * csum validation but before the pages are unlocked.  It is also set by
 137  * btrfs_init_new_buffer on freshly allocated blocks.
 138  *
 139  * We also add a check to make sure the highest level of the tree is the
 140  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 141  * needs update as well.
 142  */
 143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 144 # if BTRFS_MAX_LEVEL != 8
 145 #  error
 146 # endif
 147
 148 static struct btrfs_lockdep_keyset {
 149         u64                     id;             /* root objectid */
 150         const char              *name_stem;     /* lock name stem */
 151         char                    names[BTRFS_MAX_LEVEL + 1][20];
 152         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 153 } btrfs_lockdep_keysets[] = {
 154         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 155         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 156         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 157         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 158         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 159         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 160         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 161         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 162         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 163         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 164         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 165         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 166         { .id = 0,                              .name_stem = "tree"     },
 167 };
 168
 169 void __init btrfs_init_lockdep(void)
 170 {
 171         int i, j;
 172
 173         /* initialize lockdep class names */
 174         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 175                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 176
 177                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 178                         snprintf(ks->names[j], sizeof(ks->names[j]),
 179                                  "btrfs-%s-%02d", ks->name_stem, j);
 180         }
 181 }
 182
 183 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 184                                     int level)
 185 {
 186         struct btrfs_lockdep_keyset *ks;
 187
 188         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 189
 190         /* find the matching keyset, id 0 is the default entry */
 191         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 192                 if (ks->id == objectid)
 193                         break;
 194
 195         lockdep_set_class_and_name(&eb->lock,
 196                                    &ks->keys[level], ks->names[level]);
 197 }
 198
 199 #endif
 200
 201 /*
 202  * extents on the btree inode are pretty simple, there's one extent
 203  * that covers the entire device
 204  */
 205 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 206                 struct page *page, size_t pg_offset, u64 start, u64 len,
 207                 int create)
 208 {
 209         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 210         struct extent_map_tree *em_tree = &inode->extent_tree;
 211         struct extent_map *em;
 212         int ret;
 213
 214         read_lock(&em_tree->lock);
 215         em = lookup_extent_mapping(em_tree, start, len);
 216         if (em) {
 217                 em->bdev = fs_info->fs_devices->latest_bdev;
 218                 read_unlock(&em_tree->lock);
 219                 goto out;
 220         }
 221         read_unlock(&em_tree->lock);
 222
 223         em = alloc_extent_map();
 224         if (!em) {
 225                 em = ERR_PTR(-ENOMEM);
 226                 goto out;
 227         }
 228         em->start = 0;
 229         em->len = (u64)-1;
 230         em->block_len = (u64)-1;
 231         em->block_start = 0;
 232         em->bdev = fs_info->fs_devices->latest_bdev;
 233
 234         write_lock(&em_tree->lock);
 235         ret = add_extent_mapping(em_tree, em, 0);
 236         if (ret == -EEXIST) {
 237                 free_extent_map(em);
 238                 em = lookup_extent_mapping(em_tree, start, len);
 239                 if (!em)
 240                         em = ERR_PTR(-EIO);
 241         } else if (ret) {
 242                 free_extent_map(em);
 243                 em = ERR_PTR(ret);
 244         }
 245         write_unlock(&em_tree->lock);
 246
 247 out:
 248         return em;
 249 }
 250
 251 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 252 {
 253         return crc32c(seed, data, len);
 254 }
 255
 256 void btrfs_csum_final(u32 crc, u8 *result)
 257 {
 258         put_unaligned_le32(~crc, result);
 259 }
 260
 261 /*
 262  * compute the csum for a btree block, and either verify it or write it
 263  * into the csum field of the block.
 264  */
 265 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 266                            struct extent_buffer *buf,
 267                            int verify)
 268 {
 269         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 270         char result[BTRFS_CSUM_SIZE];
 271         unsigned long len;
 272         unsigned long cur_len;
 273         unsigned long offset = BTRFS_CSUM_SIZE;
 274         char *kaddr;
 275         unsigned long map_start;
 276         unsigned long map_len;
 277         int err;
 278         u32 crc = ~(u32)0;
 279
 280         len = buf->len - offset;
 281         while (len > 0) {
 282                 /*
 283                  * Note: we don't need to check for the err == 1 case here, as
 284                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 285                  * and 'min_len = 32' and the currently implemented mapping
 286                  * algorithm we cannot cross a page boundary.
 287                  */
 288                 err = map_private_extent_buffer(buf, offset, 32,
 289                                         &kaddr, &map_start, &map_len);
 290                 if (err)
 291                         return err;
 292                 cur_len = min(len, map_len - (offset - map_start));
 293                 crc = btrfs_csum_data(kaddr + offset - map_start,
 294                                       crc, cur_len);
 295                 len -= cur_len;
 296                 offset += cur_len;
 297         }
 298         memset(result, 0, BTRFS_CSUM_SIZE);
 299
 300         btrfs_csum_final(crc, result);
 301
 302         if (verify) {
 303                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 304                         u32 val;
 305                         u32 found = 0;
 306                         memcpy(&found, result, csum_size);
 307
 308                         read_extent_buffer(buf, &val, 0, csum_size);
 309                         btrfs_warn_rl(fs_info,
 310                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 311                                 fs_info->sb->s_id, buf->start,
 312                                 val, found, btrfs_header_level(buf));
 313                         return -EUCLEAN;
 314                 }
 315         } else {
 316                 write_extent_buffer(buf, result, 0, csum_size);
 317         }
 318
 319         return 0;
 320 }
 321
 322 /*
 323  * we can't consider a given block up to date unless the transid of the
 324  * block matches the transid in the parent node's pointer.  This is how we
 325  * detect blocks that either didn't get written at all or got written
 326  * in the wrong place.
 327  */
 328 static int verify_parent_transid(struct extent_io_tree *io_tree,
 329                                  struct extent_buffer *eb, u64 parent_transid,
 330                                  int atomic)
 331 {
 332         struct extent_state *cached_state = NULL;
 333         int ret;
 334         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 335
 336         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 337                 return 0;
 338
 339         if (atomic)
 340                 return -EAGAIN;
 341
 342         if (need_lock) {
 343                 btrfs_tree_read_lock(eb);
 344                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 345         }
 346
 347         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 348                          &cached_state);
 349         if (extent_buffer_uptodate(eb) &&
 350             btrfs_header_generation(eb) == parent_transid) {
 351                 ret = 0;
 352                 goto out;
 353         }
 354         btrfs_err_rl(eb->fs_info,
 355                 "parent transid verify failed on %llu wanted %llu found %llu",
 356                         eb->start,
 357                         parent_transid, btrfs_header_generation(eb));
 358         ret = 1;
 359
 360         /*
 361          * Things reading via commit roots that don't have normal protection,
 362          * like send, can have a really old block in cache that may point at a
 363          * block that has been freed and re-allocated.  So don't clear uptodate
 364          * if we find an eb that is under IO (dirty/writeback) because we could
 365          * end up reading in the stale data and then writing it back out and
 366          * making everybody very sad.
 367          */
 368         if (!extent_buffer_under_io(eb))
 369                 clear_extent_buffer_uptodate(eb);
 370 out:
 371         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 372                              &cached_state);
 373         if (need_lock)
 374                 btrfs_tree_read_unlock_blocking(eb);
 375         return ret;
 376 }
 377
 378 /*
 379  * Return 0 if the superblock checksum type matches the checksum value of that
 380  * algorithm. Pass the raw disk superblock data.
 381  */
 382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 383                                   char *raw_disk_sb)
 384 {
 385         struct btrfs_super_block *disk_sb =
 386                 (struct btrfs_super_block *)raw_disk_sb;
 387         u16 csum_type = btrfs_super_csum_type(disk_sb);
 388         int ret = 0;
 389
 390         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 391                 u32 crc = ~(u32)0;
 392                 char result[sizeof(crc)];
 393
 394                 /*
 395                  * The super_block structure does not span the whole
 396                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 397                  * is filled with zeros and is included in the checksum.
 398                  */
 399                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 400                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 401                 btrfs_csum_final(crc, result);
 402
 403                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 404                         ret = 1;
 405         }
 406
 407         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 408                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 409                                 csum_type);
 410                 ret = 1;
 411         }
 412
 413         return ret;
 414 }
 415
 416 static int verify_level_key(struct btrfs_fs_info *fs_info,
 417                             struct extent_buffer *eb, int level,
 418                             struct btrfs_key *first_key, u64 parent_transid)
 419 {
 420         int found_level;
 421         struct btrfs_key found_key;
 422         int ret;
 423
 424         found_level = btrfs_header_level(eb);
 425         if (found_level != level) {
 426 #ifdef CONFIG_BTRFS_DEBUG
 427                 WARN_ON(1);
 428                 btrfs_err(fs_info,
 429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 430                           eb->start, level, found_level);
 431 #endif
 432                 return -EIO;
 433         }
 434
 435         if (!first_key)
 436                 return 0;
 437
 438         /*
 439          * For live tree block (new tree blocks in current transaction),
 440          * we need proper lock context to avoid race, which is impossible here.
 441          * So we only checks tree blocks which is read from disk, whose
 442          * generation <= fs_info->last_trans_committed.
 443          */
 444         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 445                 return 0;
 446         if (found_level)
 447                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 448         else
 449                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 450         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 451
 452 #ifdef CONFIG_BTRFS_DEBUG
 453         if (ret) {
 454                 WARN_ON(1);
 455                 btrfs_err(fs_info,
 456 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 457                           eb->start, parent_transid, first_key->objectid,
 458                           first_key->type, first_key->offset,
 459                           found_key.objectid, found_key.type,
 460                           found_key.offset);
 461         }
 462 #endif
 463         return ret;
 464 }
 465
 466 /*
 467  * helper to read a given tree block, doing retries as required when
 468  * the checksums don't match and we have alternate mirrors to try.
 469  *
 470  * @parent_transid:     expected transid, skip check if 0
 471  * @level:              expected level, mandatory check
 472  * @first_key:          expected key of first slot, skip check if NULL
 473  */
 474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 475                                           struct extent_buffer *eb,
 476                                           u64 parent_transid, int level,
 477                                           struct btrfs_key *first_key)
 478 {
 479         struct extent_io_tree *io_tree;
 480         int failed = 0;
 481         int ret;
 482         int num_copies = 0;
 483         int mirror_num = 0;
 484         int failed_mirror = 0;
 485
 486         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 487         while (1) {
 488                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 489                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 490                                                mirror_num);
 491                 if (!ret) {
 492                         if (verify_parent_transid(io_tree, eb,
 493                                                    parent_transid, 0))
 494                                 ret = -EIO;
 495                         else if (verify_level_key(fs_info, eb, level,
 496                                                   first_key, parent_transid))
 497                                 ret = -EUCLEAN;
 498                         else
 499                                 break;
 500                 }
 501
 502                 num_copies = btrfs_num_copies(fs_info,
 503                                               eb->start, eb->len);
 504                 if (num_copies == 1)
 505                         break;
 506
 507                 if (!failed_mirror) {
 508                         failed = 1;
 509                         failed_mirror = eb->read_mirror;
 510                 }
 511
 512                 mirror_num++;
 513                 if (mirror_num == failed_mirror)
 514                         mirror_num++;
 515
 516                 if (mirror_num > num_copies)
 517                         break;
 518         }
 519
 520         if (failed && !ret && failed_mirror)
 521                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 522
 523         return ret;
 524 }
 525
 526 /*
 527  * checksum a dirty tree block before IO.  This has extra checks to make sure
 528  * we only fill in the checksum field in the first page of a multi-page block
 529  */
 530
 531 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 532 {
 533         u64 start = page_offset(page);
 534         u64 found_start;
 535         struct extent_buffer *eb;
 536
 537         eb = (struct extent_buffer *)page->private;
 538         if (page != eb->pages[0])
 539                 return 0;
 540
 541         found_start = btrfs_header_bytenr(eb);
 542         /*
 543          * Please do not consolidate these warnings into a single if.
 544          * It is useful to know what went wrong.
 545          */
 546         if (WARN_ON(found_start != start))
 547                 return -EUCLEAN;
 548         if (WARN_ON(!PageUptodate(page)))
 549                 return -EUCLEAN;
 550
 551         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 552                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 553
 554         return csum_tree_block(fs_info, eb, 0);
 555 }
 556
 557 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 558                                  struct extent_buffer *eb)
 559 {
 560         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 561         u8 fsid[BTRFS_FSID_SIZE];
 562         int ret = 1;
 563
 564         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 565         while (fs_devices) {
 566                 u8 *metadata_uuid;
 567
 568                 /*
 569                  * Checking the incompat flag is only valid for the current
 570                  * fs. For seed devices it's forbidden to have their uuid
 571                  * changed so reading ->fsid in this case is fine
 572                  */
 573                 if (fs_devices == fs_info->fs_devices &&
 574                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 575                         metadata_uuid = fs_devices->metadata_uuid;
 576                 else
 577                         metadata_uuid = fs_devices->fsid;
 578
 579                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 580                         ret = 0;
 581                         break;
 582                 }
 583                 fs_devices = fs_devices->seed;
 584         }
 585         return ret;
 586 }
 587
 588 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 589                                       u64 phy_offset, struct page *page,
 590                                       u64 start, u64 end, int mirror)
 591 {
 592         u64 found_start;
 593         int found_level;
 594         struct extent_buffer *eb;
 595         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 596         struct btrfs_fs_info *fs_info = root->fs_info;
 597         int ret = 0;
 598         int reads_done;
 599
 600         if (!page->private)
 601                 goto out;
 602
 603         eb = (struct extent_buffer *)page->private;
 604
 605         /* the pending IO might have been the only thing that kept this buffer
 606          * in memory.  Make sure we have a ref for all this other checks
 607          */
 608         extent_buffer_get(eb);
 609
 610         reads_done = atomic_dec_and_test(&eb->io_pages);
 611         if (!reads_done)
 612                 goto err;
 613
 614         eb->read_mirror = mirror;
 615         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 616                 ret = -EIO;
 617                 goto err;
 618         }
 619
 620         found_start = btrfs_header_bytenr(eb);
 621         if (found_start != eb->start) {
 622                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 623                              eb->start, found_start);
 624                 ret = -EIO;
 625                 goto err;
 626         }
 627         if (check_tree_block_fsid(fs_info, eb)) {
 628                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 629                              eb->start);
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633         found_level = btrfs_header_level(eb);
 634         if (found_level >= BTRFS_MAX_LEVEL) {
 635                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 636                           (int)btrfs_header_level(eb), eb->start);
 637                 ret = -EIO;
 638                 goto err;
 639         }
 640
 641         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 642                                        eb, found_level);
 643
 644         ret = csum_tree_block(fs_info, eb, 1);
 645         if (ret)
 646                 goto err;
 647
 648         /*
 649          * If this is a leaf block and it is corrupt, set the corrupt bit so
 650          * that we don't try and read the other copies of this block, just
 651          * return -EIO.
 652          */
 653         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 654                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 655                 ret = -EIO;
 656         }
 657
 658         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 659                 ret = -EIO;
 660
 661         if (!ret)
 662                 set_extent_buffer_uptodate(eb);
 663 err:
 664         if (reads_done &&
 665             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 666                 btree_readahead_hook(eb, ret);
 667
 668         if (ret) {
 669                 /*
 670                  * our io error hook is going to dec the io pages
 671                  * again, we have to make sure it has something
 672                  * to decrement
 673                  */
 674                 atomic_inc(&eb->io_pages);
 675                 clear_extent_buffer_uptodate(eb);
 676         }
 677         free_extent_buffer(eb);
 678 out:
 679         return ret;
 680 }
 681
 682 static void end_workqueue_bio(struct bio *bio)
 683 {
 684         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 685         struct btrfs_fs_info *fs_info;
 686         struct btrfs_workqueue *wq;
 687         btrfs_work_func_t func;
 688
 689         fs_info = end_io_wq->info;
 690         end_io_wq->status = bio->bi_status;
 691
 692         if (bio_op(bio) == REQ_OP_WRITE) {
 693                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 694                         wq = fs_info->endio_meta_write_workers;
 695                         func = btrfs_endio_meta_write_helper;
 696                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 697                         wq = fs_info->endio_freespace_worker;
 698                         func = btrfs_freespace_write_helper;
 699                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 700                         wq = fs_info->endio_raid56_workers;
 701                         func = btrfs_endio_raid56_helper;
 702                 } else {
 703                         wq = fs_info->endio_write_workers;
 704                         func = btrfs_endio_write_helper;
 705                 }
 706         } else {
 707                 if (unlikely(end_io_wq->metadata ==
 708                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 709                         wq = fs_info->endio_repair_workers;
 710                         func = btrfs_endio_repair_helper;
 711                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 712                         wq = fs_info->endio_raid56_workers;
 713                         func = btrfs_endio_raid56_helper;
 714                 } else if (end_io_wq->metadata) {
 715                         wq = fs_info->endio_meta_workers;
 716                         func = btrfs_endio_meta_helper;
 717                 } else {
 718                         wq = fs_info->endio_workers;
 719                         func = btrfs_endio_helper;
 720                 }
 721         }
 722
 723         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 724         btrfs_queue_work(wq, &end_io_wq->work);
 725 }
 726
 727 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 728                         enum btrfs_wq_endio_type metadata)
 729 {
 730         struct btrfs_end_io_wq *end_io_wq;
 731
 732         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 733         if (!end_io_wq)
 734                 return BLK_STS_RESOURCE;
 735
 736         end_io_wq->private = bio->bi_private;
 737         end_io_wq->end_io = bio->bi_end_io;
 738         end_io_wq->info = info;
 739         end_io_wq->status = 0;
 740         end_io_wq->bio = bio;
 741         end_io_wq->metadata = metadata;
 742
 743         bio->bi_private = end_io_wq;
 744         bio->bi_end_io = end_workqueue_bio;
 745         return 0;
 746 }
 747
 748 static void run_one_async_start(struct btrfs_work *work)
 749 {
 750         struct async_submit_bio *async;
 751         blk_status_t ret;
 752
 753         async = container_of(work, struct  async_submit_bio, work);
 754         ret = async->submit_bio_start(async->private_data, async->bio,
 755                                       async->bio_offset);
 756         if (ret)
 757                 async->status = ret;
 758 }
 759
 760 /*
 761  * In order to insert checksums into the metadata in large chunks, we wait
 762  * until bio submission time.   All the pages in the bio are checksummed and
 763  * sums are attached onto the ordered extent record.
 764  *
 765  * At IO completion time the csums attached on the ordered extent record are
 766  * inserted into the tree.
 767  */
 768 static void run_one_async_done(struct btrfs_work *work)
 769 {
 770         struct async_submit_bio *async;
 771         struct inode *inode;
 772         blk_status_t ret;
 773
 774         async = container_of(work, struct  async_submit_bio, work);
 775         inode = async->private_data;
 776
 777         /* If an error occurred we just want to clean up the bio and move on */
 778         if (async->status) {
 779                 async->bio->bi_status = async->status;
 780                 bio_endio(async->bio);
 781                 return;
 782         }
 783
 784         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 785                         async->mirror_num, 1);
 786         if (ret) {
 787                 async->bio->bi_status = ret;
 788                 bio_endio(async->bio);
 789         }
 790 }
 791
 792 static void run_one_async_free(struct btrfs_work *work)
 793 {
 794         struct async_submit_bio *async;
 795
 796         async = container_of(work, struct  async_submit_bio, work);
 797         kfree(async);
 798 }
 799
 800 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 801                                  int mirror_num, unsigned long bio_flags,
 802                                  u64 bio_offset, void *private_data,
 803                                  extent_submit_bio_start_t *submit_bio_start)
 804 {
 805         struct async_submit_bio *async;
 806
 807         async = kmalloc(sizeof(*async), GFP_NOFS);
 808         if (!async)
 809                 return BLK_STS_RESOURCE;
 810
 811         async->private_data = private_data;
 812         async->bio = bio;
 813         async->mirror_num = mirror_num;
 814         async->submit_bio_start = submit_bio_start;
 815
 816         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 817                         run_one_async_done, run_one_async_free);
 818
 819         async->bio_offset = bio_offset;
 820
 821         async->status = 0;
 822
 823         if (op_is_sync(bio->bi_opf))
 824                 btrfs_set_work_high_priority(&async->work);
 825
 826         btrfs_queue_work(fs_info->workers, &async->work);
 827         return 0;
 828 }
 829
 830 static blk_status_t btree_csum_one_bio(struct bio *bio)
 831 {
 832         struct bio_vec *bvec;
 833         struct btrfs_root *root;
 834         int i, ret = 0;
 835
 836         ASSERT(!bio_flagged(bio, BIO_CLONED));
 837         bio_for_each_segment_all(bvec, bio, i) {
 838                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 839                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 840                 if (ret)
 841                         break;
 842         }
 843
 844         return errno_to_blk_status(ret);
 845 }
 846
 847 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 848                                              u64 bio_offset)
 849 {
 850         /*
 851          * when we're called for a write, we're already in the async
 852          * submission context.  Just jump into btrfs_map_bio
 853          */
 854         return btree_csum_one_bio(bio);
 855 }
 856
 857 static int check_async_write(struct btrfs_inode *bi)
 858 {
 859         if (atomic_read(&bi->sync_writers))
 860                 return 0;
 861 #ifdef CONFIG_X86
 862         if (static_cpu_has(X86_FEATURE_XMM4_2))
 863                 return 0;
 864 #endif
 865         return 1;
 866 }
 867
 868 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 869                                           int mirror_num, unsigned long bio_flags,
 870                                           u64 bio_offset)
 871 {
 872         struct inode *inode = private_data;
 873         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 874         int async = check_async_write(BTRFS_I(inode));
 875         blk_status_t ret;
 876
 877         if (bio_op(bio) != REQ_OP_WRITE) {
 878                 /*
 879                  * called for a read, do the setup so that checksum validation
 880                  * can happen in the async kernel threads
 881                  */
 882                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 883                                           BTRFS_WQ_ENDIO_METADATA);
 884                 if (ret)
 885                         goto out_w_error;
 886                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 887         } else if (!async) {
 888                 ret = btree_csum_one_bio(bio);
 889                 if (ret)
 890                         goto out_w_error;
 891                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 892         } else {
 893                 /*
 894                  * kthread helpers are used to submit writes so that
 895                  * checksumming can happen in parallel across all CPUs
 896                  */
 897                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 898                                           bio_offset, private_data,
 899                                           btree_submit_bio_start);
 900         }
 901
 902         if (ret)
 903                 goto out_w_error;
 904         return 0;
 905
 906 out_w_error:
 907         bio->bi_status = ret;
 908         bio_endio(bio);
 909         return ret;
 910 }
 911
 912 #ifdef CONFIG_MIGRATION
 913 static int btree_migratepage(struct address_space *mapping,
 914                         struct page *newpage, struct page *page,
 915                         enum migrate_mode mode)
 916 {
 917         /*
 918          * we can't safely write a btree page from here,
 919          * we haven't done the locking hook
 920          */
 921         if (PageDirty(page))
 922                 return -EAGAIN;
 923         /*
 924          * Buffers may be managed in a filesystem specific way.
 925          * We must have no buffers or drop them.
 926          */
 927         if (page_has_private(page) &&
 928             !try_to_release_page(page, GFP_KERNEL))
 929                 return -EAGAIN;
 930         return migrate_page(mapping, newpage, page, mode);
 931 }
 932 #endif
 933
 934
 935 static int btree_writepages(struct address_space *mapping,
 936                             struct writeback_control *wbc)
 937 {
 938         struct btrfs_fs_info *fs_info;
 939         int ret;
 940
 941         if (wbc->sync_mode == WB_SYNC_NONE) {
 942
 943                 if (wbc->for_kupdate)
 944                         return 0;
 945
 946                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 947                 /* this is a bit racy, but that's ok */
 948                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 949                                              BTRFS_DIRTY_METADATA_THRESH,
 950                                              fs_info->dirty_metadata_batch);
 951                 if (ret < 0)
 952                         return 0;
 953         }
 954         return btree_write_cache_pages(mapping, wbc);
 955 }
 956
 957 static int btree_readpage(struct file *file, struct page *page)
 958 {
 959         struct extent_io_tree *tree;
 960         tree = &BTRFS_I(page->mapping->host)->io_tree;
 961         return extent_read_full_page(tree, page, btree_get_extent, 0);
 962 }
 963
 964 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 965 {
 966         if (PageWriteback(page) || PageDirty(page))
 967                 return 0;
 968
 969         return try_release_extent_buffer(page);
 970 }
 971
 972 static void btree_invalidatepage(struct page *page, unsigned int offset,
 973                                  unsigned int length)
 974 {
 975         struct extent_io_tree *tree;
 976         tree = &BTRFS_I(page->mapping->host)->io_tree;
 977         extent_invalidatepage(tree, page, offset);
 978         btree_releasepage(page, GFP_NOFS);
 979         if (PagePrivate(page)) {
 980                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 981                            "page private not zero on page %llu",
 982                            (unsigned long long)page_offset(page));
 983                 ClearPagePrivate(page);
 984                 set_page_private(page, 0);
 985                 put_page(page);
 986         }
 987 }
 988
 989 static int btree_set_page_dirty(struct page *page)
 990 {
 991 #ifdef DEBUG
 992         struct extent_buffer *eb;
 993
 994         BUG_ON(!PagePrivate(page));
 995         eb = (struct extent_buffer *)page->private;
 996         BUG_ON(!eb);
 997         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 998         BUG_ON(!atomic_read(&eb->refs));
 999         btrfs_assert_tree_locked(eb);
1000 #endif
1001         return __set_page_dirty_nobuffers(page);
1002 }
1003
1004 static const struct address_space_operations btree_aops = {
1005         .readpage       = btree_readpage,
1006         .writepages     = btree_writepages,
1007         .releasepage    = btree_releasepage,
1008         .invalidatepage = btree_invalidatepage,
1009 #ifdef CONFIG_MIGRATION
1010         .migratepage    = btree_migratepage,
1011 #endif
1012         .set_page_dirty = btree_set_page_dirty,
1013 };
1014
1015 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1016 {
1017         struct extent_buffer *buf = NULL;
1018         struct inode *btree_inode = fs_info->btree_inode;
1019
1020         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1021         if (IS_ERR(buf))
1022                 return;
1023         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1024                                  buf, WAIT_NONE, 0);
1025         free_extent_buffer(buf);
1026 }
1027
1028 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1029                          int mirror_num, struct extent_buffer **eb)
1030 {
1031         struct extent_buffer *buf = NULL;
1032         struct inode *btree_inode = fs_info->btree_inode;
1033         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1034         int ret;
1035
1036         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1037         if (IS_ERR(buf))
1038                 return 0;
1039
1040         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1041
1042         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1043                                        mirror_num);
1044         if (ret) {
1045                 free_extent_buffer(buf);
1046                 return ret;
1047         }
1048
1049         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1050                 free_extent_buffer(buf);
1051                 return -EIO;
1052         } else if (extent_buffer_uptodate(buf)) {
1053                 *eb = buf;
1054         } else {
1055                 free_extent_buffer(buf);
1056         }
1057         return 0;
1058 }
1059
1060 struct extent_buffer *btrfs_find_create_tree_block(
1061                                                 struct btrfs_fs_info *fs_info,
1062                                                 u64 bytenr)
1063 {
1064         if (btrfs_is_testing(fs_info))
1065                 return alloc_test_extent_buffer(fs_info, bytenr);
1066         return alloc_extent_buffer(fs_info, bytenr);
1067 }
1068
1069
1070 int btrfs_write_tree_block(struct extent_buffer *buf)
1071 {
1072         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1073                                         buf->start + buf->len - 1);
1074 }
1075
1076 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1077 {
1078         filemap_fdatawait_range(buf->pages[0]->mapping,
1079                                 buf->start, buf->start + buf->len - 1);
1080 }
1081
1082 /*
1083  * Read tree block at logical address @bytenr and do variant basic but critical
1084  * verification.
1085  *
1086  * @parent_transid:     expected transid of this tree block, skip check if 0
1087  * @level:              expected level, mandatory check
1088  * @first_key:          expected key in slot 0, skip check if NULL
1089  */
1090 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1091                                       u64 parent_transid, int level,
1092                                       struct btrfs_key *first_key)
1093 {
1094         struct extent_buffer *buf = NULL;
1095         int ret;
1096
1097         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1098         if (IS_ERR(buf))
1099                 return buf;
1100
1101         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1102                                              level, first_key);
1103         if (ret) {
1104                 free_extent_buffer(buf);
1105                 return ERR_PTR(ret);
1106         }
1107         return buf;
1108
1109 }
1110
1111 void clean_tree_block(struct btrfs_fs_info *fs_info,
1112                       struct extent_buffer *buf)
1113 {
1114         if (btrfs_header_generation(buf) ==
1115             fs_info->running_transaction->transid) {
1116                 btrfs_assert_tree_locked(buf);
1117
1118                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1119                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1120                                                  -buf->len,
1121                                                  fs_info->dirty_metadata_batch);
1122                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1123                         btrfs_set_lock_blocking(buf);
1124                         clear_extent_buffer_dirty(buf);
1125                 }
1126         }
1127 }
1128
1129 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1130 {
1131         struct btrfs_subvolume_writers *writers;
1132         int ret;
1133
1134         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1135         if (!writers)
1136                 return ERR_PTR(-ENOMEM);
1137
1138         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1139         if (ret < 0) {
1140                 kfree(writers);
1141                 return ERR_PTR(ret);
1142         }
1143
1144         init_waitqueue_head(&writers->wait);
1145         return writers;
1146 }
1147
1148 static void
1149 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1150 {
1151         percpu_counter_destroy(&writers->counter);
1152         kfree(writers);
1153 }
1154
1155 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1156                          u64 objectid)
1157 {
1158         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1159         root->node = NULL;
1160         root->commit_root = NULL;
1161         root->state = 0;
1162         root->orphan_cleanup_state = 0;
1163
1164         root->last_trans = 0;
1165         root->highest_objectid = 0;
1166         root->nr_delalloc_inodes = 0;
1167         root->nr_ordered_extents = 0;
1168         root->inode_tree = RB_ROOT;
1169         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1170         root->block_rsv = NULL;
1171
1172         INIT_LIST_HEAD(&root->dirty_list);
1173         INIT_LIST_HEAD(&root->root_list);
1174         INIT_LIST_HEAD(&root->delalloc_inodes);
1175         INIT_LIST_HEAD(&root->delalloc_root);
1176         INIT_LIST_HEAD(&root->ordered_extents);
1177         INIT_LIST_HEAD(&root->ordered_root);
1178         INIT_LIST_HEAD(&root->logged_list[0]);
1179         INIT_LIST_HEAD(&root->logged_list[1]);
1180         spin_lock_init(&root->inode_lock);
1181         spin_lock_init(&root->delalloc_lock);
1182         spin_lock_init(&root->ordered_extent_lock);
1183         spin_lock_init(&root->accounting_lock);
1184         spin_lock_init(&root->log_extents_lock[0]);
1185         spin_lock_init(&root->log_extents_lock[1]);
1186         spin_lock_init(&root->qgroup_meta_rsv_lock);
1187         mutex_init(&root->objectid_mutex);
1188         mutex_init(&root->log_mutex);
1189         mutex_init(&root->ordered_extent_mutex);
1190         mutex_init(&root->delalloc_mutex);
1191         init_waitqueue_head(&root->log_writer_wait);
1192         init_waitqueue_head(&root->log_commit_wait[0]);
1193         init_waitqueue_head(&root->log_commit_wait[1]);
1194         INIT_LIST_HEAD(&root->log_ctxs[0]);
1195         INIT_LIST_HEAD(&root->log_ctxs[1]);
1196         atomic_set(&root->log_commit[0], 0);
1197         atomic_set(&root->log_commit[1], 0);
1198         atomic_set(&root->log_writers, 0);
1199         atomic_set(&root->log_batch, 0);
1200         refcount_set(&root->refs, 1);
1201         atomic_set(&root->will_be_snapshotted, 0);
1202         atomic_set(&root->snapshot_force_cow, 0);
1203         atomic_set(&root->nr_swapfiles, 0);
1204         root->log_transid = 0;
1205         root->log_transid_committed = -1;
1206         root->last_log_commit = 0;
1207         if (!dummy)
1208                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1209
1210         memset(&root->root_key, 0, sizeof(root->root_key));
1211         memset(&root->root_item, 0, sizeof(root->root_item));
1212         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1213         if (!dummy)
1214                 root->defrag_trans_start = fs_info->generation;
1215         else
1216                 root->defrag_trans_start = 0;
1217         root->root_key.objectid = objectid;
1218         root->anon_dev = 0;
1219
1220         spin_lock_init(&root->root_item_lock);
1221 }
1222
1223 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1224                 gfp_t flags)
1225 {
1226         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1227         if (root)
1228                 root->fs_info = fs_info;
1229         return root;
1230 }
1231
1232 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1233 /* Should only be used by the testing infrastructure */
1234 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1235 {
1236         struct btrfs_root *root;
1237
1238         if (!fs_info)
1239                 return ERR_PTR(-EINVAL);
1240
1241         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1242         if (!root)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         /* We don't use the stripesize in selftest, set it as sectorsize */
1246         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1247         root->alloc_bytenr = 0;
1248
1249         return root;
1250 }
1251 #endif
1252
1253 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1254                                      struct btrfs_fs_info *fs_info,
1255                                      u64 objectid)
1256 {
1257         struct extent_buffer *leaf;
1258         struct btrfs_root *tree_root = fs_info->tree_root;
1259         struct btrfs_root *root;
1260         struct btrfs_key key;
1261         int ret = 0;
1262         uuid_le uuid = NULL_UUID_LE;
1263
1264         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1265         if (!root)
1266                 return ERR_PTR(-ENOMEM);
1267
1268         __setup_root(root, fs_info, objectid);
1269         root->root_key.objectid = objectid;
1270         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1271         root->root_key.offset = 0;
1272
1273         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1274         if (IS_ERR(leaf)) {
1275                 ret = PTR_ERR(leaf);
1276                 leaf = NULL;
1277                 goto fail;
1278         }
1279
1280         root->node = leaf;
1281         btrfs_mark_buffer_dirty(leaf);
1282
1283         root->commit_root = btrfs_root_node(root);
1284         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1285
1286         root->root_item.flags = 0;
1287         root->root_item.byte_limit = 0;
1288         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1289         btrfs_set_root_generation(&root->root_item, trans->transid);
1290         btrfs_set_root_level(&root->root_item, 0);
1291         btrfs_set_root_refs(&root->root_item, 1);
1292         btrfs_set_root_used(&root->root_item, leaf->len);
1293         btrfs_set_root_last_snapshot(&root->root_item, 0);
1294         btrfs_set_root_dirid(&root->root_item, 0);
1295         if (is_fstree(objectid))
1296                 uuid_le_gen(&uuid);
1297         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1298         root->root_item.drop_level = 0;
1299
1300         key.objectid = objectid;
1301         key.type = BTRFS_ROOT_ITEM_KEY;
1302         key.offset = 0;
1303         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1304         if (ret)
1305                 goto fail;
1306
1307         btrfs_tree_unlock(leaf);
1308
1309         return root;
1310
1311 fail:
1312         if (leaf) {
1313                 btrfs_tree_unlock(leaf);
1314                 free_extent_buffer(root->commit_root);
1315                 free_extent_buffer(leaf);
1316         }
1317         kfree(root);
1318
1319         return ERR_PTR(ret);
1320 }
1321
1322 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1323                                          struct btrfs_fs_info *fs_info)
1324 {
1325         struct btrfs_root *root;
1326         struct extent_buffer *leaf;
1327
1328         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1329         if (!root)
1330                 return ERR_PTR(-ENOMEM);
1331
1332         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1333
1334         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1335         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1336         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1337
1338         /*
1339          * DON'T set REF_COWS for log trees
1340          *
1341          * log trees do not get reference counted because they go away
1342          * before a real commit is actually done.  They do store pointers
1343          * to file data extents, and those reference counts still get
1344          * updated (along with back refs to the log tree).
1345          */
1346
1347         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1348                         NULL, 0, 0, 0);
1349         if (IS_ERR(leaf)) {
1350                 kfree(root);
1351                 return ERR_CAST(leaf);
1352         }
1353
1354         root->node = leaf;
1355
1356         btrfs_mark_buffer_dirty(root->node);
1357         btrfs_tree_unlock(root->node);
1358         return root;
1359 }
1360
1361 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1362                              struct btrfs_fs_info *fs_info)
1363 {
1364         struct btrfs_root *log_root;
1365
1366         log_root = alloc_log_tree(trans, fs_info);
1367         if (IS_ERR(log_root))
1368                 return PTR_ERR(log_root);
1369         WARN_ON(fs_info->log_root_tree);
1370         fs_info->log_root_tree = log_root;
1371         return 0;
1372 }
1373
1374 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1375                        struct btrfs_root *root)
1376 {
1377         struct btrfs_fs_info *fs_info = root->fs_info;
1378         struct btrfs_root *log_root;
1379         struct btrfs_inode_item *inode_item;
1380
1381         log_root = alloc_log_tree(trans, fs_info);
1382         if (IS_ERR(log_root))
1383                 return PTR_ERR(log_root);
1384
1385         log_root->last_trans = trans->transid;
1386         log_root->root_key.offset = root->root_key.objectid;
1387
1388         inode_item = &log_root->root_item.inode;
1389         btrfs_set_stack_inode_generation(inode_item, 1);
1390         btrfs_set_stack_inode_size(inode_item, 3);
1391         btrfs_set_stack_inode_nlink(inode_item, 1);
1392         btrfs_set_stack_inode_nbytes(inode_item,
1393                                      fs_info->nodesize);
1394         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1395
1396         btrfs_set_root_node(&log_root->root_item, log_root->node);
1397
1398         WARN_ON(root->log_root);
1399         root->log_root = log_root;
1400         root->log_transid = 0;
1401         root->log_transid_committed = -1;
1402         root->last_log_commit = 0;
1403         return 0;
1404 }
1405
1406 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1407                                                struct btrfs_key *key)
1408 {
1409         struct btrfs_root *root;
1410         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1411         struct btrfs_path *path;
1412         u64 generation;
1413         int ret;
1414         int level;
1415
1416         path = btrfs_alloc_path();
1417         if (!path)
1418                 return ERR_PTR(-ENOMEM);
1419
1420         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1421         if (!root) {
1422                 ret = -ENOMEM;
1423                 goto alloc_fail;
1424         }
1425
1426         __setup_root(root, fs_info, key->objectid);
1427
1428         ret = btrfs_find_root(tree_root, key, path,
1429                               &root->root_item, &root->root_key);
1430         if (ret) {
1431                 if (ret > 0)
1432                         ret = -ENOENT;
1433                 goto find_fail;
1434         }
1435
1436         generation = btrfs_root_generation(&root->root_item);
1437         level = btrfs_root_level(&root->root_item);
1438         root->node = read_tree_block(fs_info,
1439                                      btrfs_root_bytenr(&root->root_item),
1440                                      generation, level, NULL);
1441         if (IS_ERR(root->node)) {
1442                 ret = PTR_ERR(root->node);
1443                 goto find_fail;
1444         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1445                 ret = -EIO;
1446                 free_extent_buffer(root->node);
1447                 goto find_fail;
1448         }
1449         root->commit_root = btrfs_root_node(root);
1450 out:
1451         btrfs_free_path(path);
1452         return root;
1453
1454 find_fail:
1455         kfree(root);
1456 alloc_fail:
1457         root = ERR_PTR(ret);
1458         goto out;
1459 }
1460
1461 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1462                                       struct btrfs_key *location)
1463 {
1464         struct btrfs_root *root;
1465
1466         root = btrfs_read_tree_root(tree_root, location);
1467         if (IS_ERR(root))
1468                 return root;
1469
1470         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1471                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1472                 btrfs_check_and_init_root_item(&root->root_item);
1473         }
1474
1475         return root;
1476 }
1477
1478 int btrfs_init_fs_root(struct btrfs_root *root)
1479 {
1480         int ret;
1481         struct btrfs_subvolume_writers *writers;
1482
1483         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1484         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1485                                         GFP_NOFS);
1486         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1487                 ret = -ENOMEM;
1488                 goto fail;
1489         }
1490
1491         writers = btrfs_alloc_subvolume_writers();
1492         if (IS_ERR(writers)) {
1493                 ret = PTR_ERR(writers);
1494                 goto fail;
1495         }
1496         root->subv_writers = writers;
1497
1498         btrfs_init_free_ino_ctl(root);
1499         spin_lock_init(&root->ino_cache_lock);
1500         init_waitqueue_head(&root->ino_cache_wait);
1501
1502         ret = get_anon_bdev(&root->anon_dev);
1503         if (ret)
1504                 goto fail;
1505
1506         mutex_lock(&root->objectid_mutex);
1507         ret = btrfs_find_highest_objectid(root,
1508                                         &root->highest_objectid);
1509         if (ret) {
1510                 mutex_unlock(&root->objectid_mutex);
1511                 goto fail;
1512         }
1513
1514         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1515
1516         mutex_unlock(&root->objectid_mutex);
1517
1518         return 0;
1519 fail:
1520         /* The caller is responsible to call btrfs_free_fs_root */
1521         return ret;
1522 }
1523
1524 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1525                                         u64 root_id)
1526 {
1527         struct btrfs_root *root;
1528
1529         spin_lock(&fs_info->fs_roots_radix_lock);
1530         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1531                                  (unsigned long)root_id);
1532         spin_unlock(&fs_info->fs_roots_radix_lock);
1533         return root;
1534 }
1535
1536 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1537                          struct btrfs_root *root)
1538 {
1539         int ret;
1540
1541         ret = radix_tree_preload(GFP_NOFS);
1542         if (ret)
1543                 return ret;
1544
1545         spin_lock(&fs_info->fs_roots_radix_lock);
1546         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1547                                 (unsigned long)root->root_key.objectid,
1548                                 root);
1549         if (ret == 0)
1550                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1551         spin_unlock(&fs_info->fs_roots_radix_lock);
1552         radix_tree_preload_end();
1553
1554         return ret;
1555 }
1556
1557 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1558                                      struct btrfs_key *location,
1559                                      bool check_ref)
1560 {
1561         struct btrfs_root *root;
1562         struct btrfs_path *path;
1563         struct btrfs_key key;
1564         int ret;
1565
1566         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1567                 return fs_info->tree_root;
1568         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1569                 return fs_info->extent_root;
1570         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1571                 return fs_info->chunk_root;
1572         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1573                 return fs_info->dev_root;
1574         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1575                 return fs_info->csum_root;
1576         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1577                 return fs_info->quota_root ? fs_info->quota_root :
1578                                              ERR_PTR(-ENOENT);
1579         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1580                 return fs_info->uuid_root ? fs_info->uuid_root :
1581                                             ERR_PTR(-ENOENT);
1582         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1583                 return fs_info->free_space_root ? fs_info->free_space_root :
1584                                                   ERR_PTR(-ENOENT);
1585 again:
1586         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1587         if (root) {
1588                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1589                         return ERR_PTR(-ENOENT);
1590                 return root;
1591         }
1592
1593         root = btrfs_read_fs_root(fs_info->tree_root, location);
1594         if (IS_ERR(root))
1595                 return root;
1596
1597         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1598                 ret = -ENOENT;
1599                 goto fail;
1600         }
1601
1602         ret = btrfs_init_fs_root(root);
1603         if (ret)
1604                 goto fail;
1605
1606         path = btrfs_alloc_path();
1607         if (!path) {
1608                 ret = -ENOMEM;
1609                 goto fail;
1610         }
1611         key.objectid = BTRFS_ORPHAN_OBJECTID;
1612         key.type = BTRFS_ORPHAN_ITEM_KEY;
1613         key.offset = location->objectid;
1614
1615         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1616         btrfs_free_path(path);
1617         if (ret < 0)
1618                 goto fail;
1619         if (ret == 0)
1620                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1621
1622         ret = btrfs_insert_fs_root(fs_info, root);
1623         if (ret) {
1624                 if (ret == -EEXIST) {
1625                         btrfs_free_fs_root(root);
1626                         goto again;
1627                 }
1628                 goto fail;
1629         }
1630         return root;
1631 fail:
1632         btrfs_free_fs_root(root);
1633         return ERR_PTR(ret);
1634 }
1635
1636 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1637 {
1638         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1639         int ret = 0;
1640         struct btrfs_device *device;
1641         struct backing_dev_info *bdi;
1642
1643         rcu_read_lock();
1644         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1645                 if (!device->bdev)
1646                         continue;
1647                 bdi = device->bdev->bd_bdi;
1648                 if (bdi_congested(bdi, bdi_bits)) {
1649                         ret = 1;
1650                         break;
1651                 }
1652         }
1653         rcu_read_unlock();
1654         return ret;
1655 }
1656
1657 /*
1658  * called by the kthread helper functions to finally call the bio end_io
1659  * functions.  This is where read checksum verification actually happens
1660  */
1661 static void end_workqueue_fn(struct btrfs_work *work)
1662 {
1663         struct bio *bio;
1664         struct btrfs_end_io_wq *end_io_wq;
1665
1666         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1667         bio = end_io_wq->bio;
1668
1669         bio->bi_status = end_io_wq->status;
1670         bio->bi_private = end_io_wq->private;
1671         bio->bi_end_io = end_io_wq->end_io;
1672         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1673         bio_endio(bio);
1674 }
1675
1676 static int cleaner_kthread(void *arg)
1677 {
1678         struct btrfs_root *root = arg;
1679         struct btrfs_fs_info *fs_info = root->fs_info;
1680         int again;
1681
1682         while (1) {
1683                 again = 0;
1684
1685                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1686
1687                 /* Make the cleaner go to sleep early. */
1688                 if (btrfs_need_cleaner_sleep(fs_info))
1689                         goto sleep;
1690
1691                 /*
1692                  * Do not do anything if we might cause open_ctree() to block
1693                  * before we have finished mounting the filesystem.
1694                  */
1695                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1696                         goto sleep;
1697
1698                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1699                         goto sleep;
1700
1701                 /*
1702                  * Avoid the problem that we change the status of the fs
1703                  * during the above check and trylock.
1704                  */
1705                 if (btrfs_need_cleaner_sleep(fs_info)) {
1706                         mutex_unlock(&fs_info->cleaner_mutex);
1707                         goto sleep;
1708                 }
1709
1710                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1711                 btrfs_run_delayed_iputs(fs_info);
1712                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1713
1714                 again = btrfs_clean_one_deleted_snapshot(root);
1715                 mutex_unlock(&fs_info->cleaner_mutex);
1716
1717                 /*
1718                  * The defragger has dealt with the R/O remount and umount,
1719                  * needn't do anything special here.
1720                  */
1721                 btrfs_run_defrag_inodes(fs_info);
1722
1723                 /*
1724                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1725                  * with relocation (btrfs_relocate_chunk) and relocation
1726                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1727                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1728                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1729                  * unused block groups.
1730                  */
1731                 btrfs_delete_unused_bgs(fs_info);
1732 sleep:
1733                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1734                 if (kthread_should_park())
1735                         kthread_parkme();
1736                 if (kthread_should_stop())
1737                         return 0;
1738                 if (!again) {
1739                         set_current_state(TASK_INTERRUPTIBLE);
1740                         schedule();
1741                         __set_current_state(TASK_RUNNING);
1742                 }
1743         }
1744 }
1745
1746 static int transaction_kthread(void *arg)
1747 {
1748         struct btrfs_root *root = arg;
1749         struct btrfs_fs_info *fs_info = root->fs_info;
1750         struct btrfs_trans_handle *trans;
1751         struct btrfs_transaction *cur;
1752         u64 transid;
1753         time64_t now;
1754         unsigned long delay;
1755         bool cannot_commit;
1756
1757         do {
1758                 cannot_commit = false;
1759                 delay = HZ * fs_info->commit_interval;
1760                 mutex_lock(&fs_info->transaction_kthread_mutex);
1761
1762                 spin_lock(&fs_info->trans_lock);
1763                 cur = fs_info->running_transaction;
1764                 if (!cur) {
1765                         spin_unlock(&fs_info->trans_lock);
1766                         goto sleep;
1767                 }
1768
1769                 now = ktime_get_seconds();
1770                 if (cur->state < TRANS_STATE_BLOCKED &&
1771                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1772                     (now < cur->start_time ||
1773                      now - cur->start_time < fs_info->commit_interval)) {
1774                         spin_unlock(&fs_info->trans_lock);
1775                         delay = HZ * 5;
1776                         goto sleep;
1777                 }
1778                 transid = cur->transid;
1779                 spin_unlock(&fs_info->trans_lock);
1780
1781                 /* If the file system is aborted, this will always fail. */
1782                 trans = btrfs_attach_transaction(root);
1783                 if (IS_ERR(trans)) {
1784                         if (PTR_ERR(trans) != -ENOENT)
1785                                 cannot_commit = true;
1786                         goto sleep;
1787                 }
1788                 if (transid == trans->transid) {
1789                         btrfs_commit_transaction(trans);
1790                 } else {
1791                         btrfs_end_transaction(trans);
1792                 }
1793 sleep:
1794                 wake_up_process(fs_info->cleaner_kthread);
1795                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1796
1797                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1798                                       &fs_info->fs_state)))
1799                         btrfs_cleanup_transaction(fs_info);
1800                 if (!kthread_should_stop() &&
1801                                 (!btrfs_transaction_blocked(fs_info) ||
1802                                  cannot_commit))
1803                         schedule_timeout_interruptible(delay);
1804         } while (!kthread_should_stop());
1805         return 0;
1806 }
1807
1808 /*
1809  * this will find the highest generation in the array of
1810  * root backups.  The index of the highest array is returned,
1811  * or -1 if we can't find anything.
1812  *
1813  * We check to make sure the array is valid by comparing the
1814  * generation of the latest  root in the array with the generation
1815  * in the super block.  If they don't match we pitch it.
1816  */
1817 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1818 {
1819         u64 cur;
1820         int newest_index = -1;
1821         struct btrfs_root_backup *root_backup;
1822         int i;
1823
1824         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1825                 root_backup = info->super_copy->super_roots + i;
1826                 cur = btrfs_backup_tree_root_gen(root_backup);
1827                 if (cur == newest_gen)
1828                         newest_index = i;
1829         }
1830
1831         /* check to see if we actually wrapped around */
1832         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1833                 root_backup = info->super_copy->super_roots;
1834                 cur = btrfs_backup_tree_root_gen(root_backup);
1835                 if (cur == newest_gen)
1836                         newest_index = 0;
1837         }
1838         return newest_index;
1839 }
1840
1841
1842 /*
1843  * find the oldest backup so we know where to store new entries
1844  * in the backup array.  This will set the backup_root_index
1845  * field in the fs_info struct
1846  */
1847 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1848                                      u64 newest_gen)
1849 {
1850         int newest_index = -1;
1851
1852         newest_index = find_newest_super_backup(info, newest_gen);
1853         /* if there was garbage in there, just move along */
1854         if (newest_index == -1) {
1855                 info->backup_root_index = 0;
1856         } else {
1857                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1858         }
1859 }
1860
1861 /*
1862  * copy all the root pointers into the super backup array.
1863  * this will bump the backup pointer by one when it is
1864  * done
1865  */
1866 static void backup_super_roots(struct btrfs_fs_info *info)
1867 {
1868         int next_backup;
1869         struct btrfs_root_backup *root_backup;
1870         int last_backup;
1871
1872         next_backup = info->backup_root_index;
1873         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1874                 BTRFS_NUM_BACKUP_ROOTS;
1875
1876         /*
1877          * just overwrite the last backup if we're at the same generation
1878          * this happens only at umount
1879          */
1880         root_backup = info->super_for_commit->super_roots + last_backup;
1881         if (btrfs_backup_tree_root_gen(root_backup) ==
1882             btrfs_header_generation(info->tree_root->node))
1883                 next_backup = last_backup;
1884
1885         root_backup = info->super_for_commit->super_roots + next_backup;
1886
1887         /*
1888          * make sure all of our padding and empty slots get zero filled
1889          * regardless of which ones we use today
1890          */
1891         memset(root_backup, 0, sizeof(*root_backup));
1892
1893         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1894
1895         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1896         btrfs_set_backup_tree_root_gen(root_backup,
1897                                btrfs_header_generation(info->tree_root->node));
1898
1899         btrfs_set_backup_tree_root_level(root_backup,
1900                                btrfs_header_level(info->tree_root->node));
1901
1902         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1903         btrfs_set_backup_chunk_root_gen(root_backup,
1904                                btrfs_header_generation(info->chunk_root->node));
1905         btrfs_set_backup_chunk_root_level(root_backup,
1906                                btrfs_header_level(info->chunk_root->node));
1907
1908         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1909         btrfs_set_backup_extent_root_gen(root_backup,
1910                                btrfs_header_generation(info->extent_root->node));
1911         btrfs_set_backup_extent_root_level(root_backup,
1912                                btrfs_header_level(info->extent_root->node));
1913
1914         /*
1915          * we might commit during log recovery, which happens before we set
1916          * the fs_root.  Make sure it is valid before we fill it in.
1917          */
1918         if (info->fs_root && info->fs_root->node) {
1919                 btrfs_set_backup_fs_root(root_backup,
1920                                          info->fs_root->node->start);
1921                 btrfs_set_backup_fs_root_gen(root_backup,
1922                                btrfs_header_generation(info->fs_root->node));
1923                 btrfs_set_backup_fs_root_level(root_backup,
1924                                btrfs_header_level(info->fs_root->node));
1925         }
1926
1927         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1928         btrfs_set_backup_dev_root_gen(root_backup,
1929                                btrfs_header_generation(info->dev_root->node));
1930         btrfs_set_backup_dev_root_level(root_backup,
1931                                        btrfs_header_level(info->dev_root->node));
1932
1933         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1934         btrfs_set_backup_csum_root_gen(root_backup,
1935                                btrfs_header_generation(info->csum_root->node));
1936         btrfs_set_backup_csum_root_level(root_backup,
1937                                btrfs_header_level(info->csum_root->node));
1938
1939         btrfs_set_backup_total_bytes(root_backup,
1940                              btrfs_super_total_bytes(info->super_copy));
1941         btrfs_set_backup_bytes_used(root_backup,
1942                              btrfs_super_bytes_used(info->super_copy));
1943         btrfs_set_backup_num_devices(root_backup,
1944                              btrfs_super_num_devices(info->super_copy));
1945
1946         /*
1947          * if we don't copy this out to the super_copy, it won't get remembered
1948          * for the next commit
1949          */
1950         memcpy(&info->super_copy->super_roots,
1951                &info->super_for_commit->super_roots,
1952                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1953 }
1954
1955 /*
1956  * this copies info out of the root backup array and back into
1957  * the in-memory super block.  It is meant to help iterate through
1958  * the array, so you send it the number of backups you've already
1959  * tried and the last backup index you used.
1960  *
1961  * this returns -1 when it has tried all the backups
1962  */
1963 static noinline int next_root_backup(struct btrfs_fs_info *info,
1964                                      struct btrfs_super_block *super,
1965                                      int *num_backups_tried, int *backup_index)
1966 {
1967         struct btrfs_root_backup *root_backup;
1968         int newest = *backup_index;
1969
1970         if (*num_backups_tried == 0) {
1971                 u64 gen = btrfs_super_generation(super);
1972
1973                 newest = find_newest_super_backup(info, gen);
1974                 if (newest == -1)
1975                         return -1;
1976
1977                 *backup_index = newest;
1978                 *num_backups_tried = 1;
1979         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1980                 /* we've tried all the backups, all done */
1981                 return -1;
1982         } else {
1983                 /* jump to the next oldest backup */
1984                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1985                         BTRFS_NUM_BACKUP_ROOTS;
1986                 *backup_index = newest;
1987                 *num_backups_tried += 1;
1988         }
1989         root_backup = super->super_roots + newest;
1990
1991         btrfs_set_super_generation(super,
1992                                    btrfs_backup_tree_root_gen(root_backup));
1993         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1994         btrfs_set_super_root_level(super,
1995                                    btrfs_backup_tree_root_level(root_backup));
1996         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1997
1998         /*
1999          * fixme: the total bytes and num_devices need to match or we should
2000          * need a fsck
2001          */
2002         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2003         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2004         return 0;
2005 }
2006
2007 /* helper to cleanup workers */
2008 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2009 {
2010         btrfs_destroy_workqueue(fs_info->fixup_workers);
2011         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2012         btrfs_destroy_workqueue(fs_info->workers);
2013         btrfs_destroy_workqueue(fs_info->endio_workers);
2014         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2015         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2016         btrfs_destroy_workqueue(fs_info->rmw_workers);
2017         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2018         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2019         btrfs_destroy_workqueue(fs_info->submit_workers);
2020         btrfs_destroy_workqueue(fs_info->delayed_workers);
2021         btrfs_destroy_workqueue(fs_info->caching_workers);
2022         btrfs_destroy_workqueue(fs_info->readahead_workers);
2023         btrfs_destroy_workqueue(fs_info->flush_workers);
2024         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2025         btrfs_destroy_workqueue(fs_info->extent_workers);
2026         /*
2027          * Now that all other work queues are destroyed, we can safely destroy
2028          * the queues used for metadata I/O, since tasks from those other work
2029          * queues can do metadata I/O operations.
2030          */
2031         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2032         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2033 }
2034
2035 static void free_root_extent_buffers(struct btrfs_root *root)
2036 {
2037         if (root) {
2038                 free_extent_buffer(root->node);
2039                 free_extent_buffer(root->commit_root);
2040                 root->node = NULL;
2041                 root->commit_root = NULL;
2042         }
2043 }
2044
2045 /* helper to cleanup tree roots */
2046 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2047 {
2048         free_root_extent_buffers(info->tree_root);
2049
2050         free_root_extent_buffers(info->dev_root);
2051         free_root_extent_buffers(info->extent_root);
2052         free_root_extent_buffers(info->csum_root);
2053         free_root_extent_buffers(info->quota_root);
2054         free_root_extent_buffers(info->uuid_root);
2055         if (chunk_root)
2056                 free_root_extent_buffers(info->chunk_root);
2057         free_root_extent_buffers(info->free_space_root);
2058 }
2059
2060 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2061 {
2062         int ret;
2063         struct btrfs_root *gang[8];
2064         int i;
2065
2066         while (!list_empty(&fs_info->dead_roots)) {
2067                 gang[0] = list_entry(fs_info->dead_roots.next,
2068                                      struct btrfs_root, root_list);
2069                 list_del(&gang[0]->root_list);
2070
2071                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2072                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2073                 } else {
2074                         free_extent_buffer(gang[0]->node);
2075                         free_extent_buffer(gang[0]->commit_root);
2076                         btrfs_put_fs_root(gang[0]);
2077                 }
2078         }
2079
2080         while (1) {
2081                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2082                                              (void **)gang, 0,
2083                                              ARRAY_SIZE(gang));
2084                 if (!ret)
2085                         break;
2086                 for (i = 0; i < ret; i++)
2087                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2088         }
2089
2090         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2091                 btrfs_free_log_root_tree(NULL, fs_info);
2092                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2093         }
2094 }
2095
2096 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2097 {
2098         mutex_init(&fs_info->scrub_lock);
2099         atomic_set(&fs_info->scrubs_running, 0);
2100         atomic_set(&fs_info->scrub_pause_req, 0);
2101         atomic_set(&fs_info->scrubs_paused, 0);
2102         atomic_set(&fs_info->scrub_cancel_req, 0);
2103         init_waitqueue_head(&fs_info->scrub_pause_wait);
2104         fs_info->scrub_workers_refcnt = 0;
2105 }
2106
2107 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2108 {
2109         spin_lock_init(&fs_info->balance_lock);
2110         mutex_init(&fs_info->balance_mutex);
2111         atomic_set(&fs_info->balance_pause_req, 0);
2112         atomic_set(&fs_info->balance_cancel_req, 0);
2113         fs_info->balance_ctl = NULL;
2114         init_waitqueue_head(&fs_info->balance_wait_q);
2115 }
2116
2117 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2118 {
2119         struct inode *inode = fs_info->btree_inode;
2120
2121         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2122         set_nlink(inode, 1);
2123         /*
2124          * we set the i_size on the btree inode to the max possible int.
2125          * the real end of the address space is determined by all of
2126          * the devices in the system
2127          */
2128         inode->i_size = OFFSET_MAX;
2129         inode->i_mapping->a_ops = &btree_aops;
2130
2131         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2132         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2133         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2134         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2135
2136         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2137
2138         BTRFS_I(inode)->root = fs_info->tree_root;
2139         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2140         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2141         btrfs_insert_inode_hash(inode);
2142 }
2143
2144 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2145 {
2146         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2147         init_rwsem(&fs_info->dev_replace.rwsem);
2148         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2149 }
2150
2151 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2152 {
2153         spin_lock_init(&fs_info->qgroup_lock);
2154         mutex_init(&fs_info->qgroup_ioctl_lock);
2155         fs_info->qgroup_tree = RB_ROOT;
2156         fs_info->qgroup_op_tree = RB_ROOT;
2157         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2158         fs_info->qgroup_seq = 1;
2159         fs_info->qgroup_ulist = NULL;
2160         fs_info->qgroup_rescan_running = false;
2161         mutex_init(&fs_info->qgroup_rescan_lock);
2162 }
2163
2164 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2165                 struct btrfs_fs_devices *fs_devices)
2166 {
2167         u32 max_active = fs_info->thread_pool_size;
2168         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2169
2170         fs_info->workers =
2171                 btrfs_alloc_workqueue(fs_info, "worker",
2172                                       flags | WQ_HIGHPRI, max_active, 16);
2173
2174         fs_info->delalloc_workers =
2175                 btrfs_alloc_workqueue(fs_info, "delalloc",
2176                                       flags, max_active, 2);
2177
2178         fs_info->flush_workers =
2179                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2180                                       flags, max_active, 0);
2181
2182         fs_info->caching_workers =
2183                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2184
2185         /*
2186          * a higher idle thresh on the submit workers makes it much more
2187          * likely that bios will be send down in a sane order to the
2188          * devices
2189          */
2190         fs_info->submit_workers =
2191                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2192                                       min_t(u64, fs_devices->num_devices,
2193                                             max_active), 64);
2194
2195         fs_info->fixup_workers =
2196                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2197
2198         /*
2199          * endios are largely parallel and should have a very
2200          * low idle thresh
2201          */
2202         fs_info->endio_workers =
2203                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2204         fs_info->endio_meta_workers =
2205                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2206                                       max_active, 4);
2207         fs_info->endio_meta_write_workers =
2208                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2209                                       max_active, 2);
2210         fs_info->endio_raid56_workers =
2211                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2212                                       max_active, 4);
2213         fs_info->endio_repair_workers =
2214                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2215         fs_info->rmw_workers =
2216                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2217         fs_info->endio_write_workers =
2218                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2219                                       max_active, 2);
2220         fs_info->endio_freespace_worker =
2221                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2222                                       max_active, 0);
2223         fs_info->delayed_workers =
2224                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2225                                       max_active, 0);
2226         fs_info->readahead_workers =
2227                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2228                                       max_active, 2);
2229         fs_info->qgroup_rescan_workers =
2230                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2231         fs_info->extent_workers =
2232                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2233                                       min_t(u64, fs_devices->num_devices,
2234                                             max_active), 8);
2235
2236         if (!(fs_info->workers && fs_info->delalloc_workers &&
2237               fs_info->submit_workers && fs_info->flush_workers &&
2238               fs_info->endio_workers && fs_info->endio_meta_workers &&
2239               fs_info->endio_meta_write_workers &&
2240               fs_info->endio_repair_workers &&
2241               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2242               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2243               fs_info->caching_workers && fs_info->readahead_workers &&
2244               fs_info->fixup_workers && fs_info->delayed_workers &&
2245               fs_info->extent_workers &&
2246               fs_info->qgroup_rescan_workers)) {
2247                 return -ENOMEM;
2248         }
2249
2250         return 0;
2251 }
2252
2253 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2254                             struct btrfs_fs_devices *fs_devices)
2255 {
2256         int ret;
2257         struct btrfs_root *log_tree_root;
2258         struct btrfs_super_block *disk_super = fs_info->super_copy;
2259         u64 bytenr = btrfs_super_log_root(disk_super);
2260         int level = btrfs_super_log_root_level(disk_super);
2261
2262         if (fs_devices->rw_devices == 0) {
2263                 btrfs_warn(fs_info, "log replay required on RO media");
2264                 return -EIO;
2265         }
2266
2267         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2268         if (!log_tree_root)
2269                 return -ENOMEM;
2270
2271         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2272
2273         log_tree_root->node = read_tree_block(fs_info, bytenr,
2274                                               fs_info->generation + 1,
2275                                               level, NULL);
2276         if (IS_ERR(log_tree_root->node)) {
2277                 btrfs_warn(fs_info, "failed to read log tree");
2278                 ret = PTR_ERR(log_tree_root->node);
2279                 kfree(log_tree_root);
2280                 return ret;
2281         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2282                 btrfs_err(fs_info, "failed to read log tree");
2283                 free_extent_buffer(log_tree_root->node);
2284                 kfree(log_tree_root);
2285                 return -EIO;
2286         }
2287         /* returns with log_tree_root freed on success */
2288         ret = btrfs_recover_log_trees(log_tree_root);
2289         if (ret) {
2290                 btrfs_handle_fs_error(fs_info, ret,
2291                                       "Failed to recover log tree");
2292                 free_extent_buffer(log_tree_root->node);
2293                 kfree(log_tree_root);
2294                 return ret;
2295         }
2296
2297         if (sb_rdonly(fs_info->sb)) {
2298                 ret = btrfs_commit_super(fs_info);
2299                 if (ret)
2300                         return ret;
2301         }
2302
2303         return 0;
2304 }
2305
2306 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2307 {
2308         struct btrfs_root *tree_root = fs_info->tree_root;
2309         struct btrfs_root *root;
2310         struct btrfs_key location;
2311         int ret;
2312
2313         BUG_ON(!fs_info->tree_root);
2314
2315         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2316         location.type = BTRFS_ROOT_ITEM_KEY;
2317         location.offset = 0;
2318
2319         root = btrfs_read_tree_root(tree_root, &location);
2320         if (IS_ERR(root)) {
2321                 ret = PTR_ERR(root);
2322                 goto out;
2323         }
2324         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2325         fs_info->extent_root = root;
2326
2327         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2328         root = btrfs_read_tree_root(tree_root, &location);
2329         if (IS_ERR(root)) {
2330                 ret = PTR_ERR(root);
2331                 goto out;
2332         }
2333         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334         fs_info->dev_root = root;
2335         btrfs_init_devices_late(fs_info);
2336
2337         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2338         root = btrfs_read_tree_root(tree_root, &location);
2339         if (IS_ERR(root)) {
2340                 ret = PTR_ERR(root);
2341                 goto out;
2342         }
2343         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2344         fs_info->csum_root = root;
2345
2346         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2347         root = btrfs_read_tree_root(tree_root, &location);
2348         if (!IS_ERR(root)) {
2349                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2350                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2351                 fs_info->quota_root = root;
2352         }
2353
2354         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2355         root = btrfs_read_tree_root(tree_root, &location);
2356         if (IS_ERR(root)) {
2357                 ret = PTR_ERR(root);
2358                 if (ret != -ENOENT)
2359                         goto out;
2360         } else {
2361                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362                 fs_info->uuid_root = root;
2363         }
2364
2365         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2366                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2367                 root = btrfs_read_tree_root(tree_root, &location);
2368                 if (IS_ERR(root)) {
2369                         ret = PTR_ERR(root);
2370                         goto out;
2371                 }
2372                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2373                 fs_info->free_space_root = root;
2374         }
2375
2376         return 0;
2377 out:
2378         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2379                    location.objectid, ret);
2380         return ret;
2381 }
2382
2383 /*
2384  * Real super block validation
2385  * NOTE: super csum type and incompat features will not be checked here.
2386  *
2387  * @sb:         super block to check
2388  * @mirror_num: the super block number to check its bytenr:
2389  *              0       the primary (1st) sb
2390  *              1, 2    2nd and 3rd backup copy
2391  *             -1       skip bytenr check
2392  */
2393 static int validate_super(struct btrfs_fs_info *fs_info,
2394                             struct btrfs_super_block *sb, int mirror_num)
2395 {
2396         u64 nodesize = btrfs_super_nodesize(sb);
2397         u64 sectorsize = btrfs_super_sectorsize(sb);
2398         int ret = 0;
2399
2400         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2401                 btrfs_err(fs_info, "no valid FS found");
2402                 ret = -EINVAL;
2403         }
2404         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2405                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2406                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2407                 ret = -EINVAL;
2408         }
2409         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2410                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2411                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2412                 ret = -EINVAL;
2413         }
2414         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2415                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2416                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2417                 ret = -EINVAL;
2418         }
2419         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2420                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2421                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2422                 ret = -EINVAL;
2423         }
2424
2425         /*
2426          * Check sectorsize and nodesize first, other check will need it.
2427          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2428          */
2429         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2430             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2431                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2432                 ret = -EINVAL;
2433         }
2434         /* Only PAGE SIZE is supported yet */
2435         if (sectorsize != PAGE_SIZE) {
2436                 btrfs_err(fs_info,
2437                         "sectorsize %llu not supported yet, only support %lu",
2438                         sectorsize, PAGE_SIZE);
2439                 ret = -EINVAL;
2440         }
2441         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2442             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2443                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2444                 ret = -EINVAL;
2445         }
2446         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2447                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2448                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2449                 ret = -EINVAL;
2450         }
2451
2452         /* Root alignment check */
2453         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2454                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2455                            btrfs_super_root(sb));
2456                 ret = -EINVAL;
2457         }
2458         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2459                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2460                            btrfs_super_chunk_root(sb));
2461                 ret = -EINVAL;
2462         }
2463         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2464                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2465                            btrfs_super_log_root(sb));
2466                 ret = -EINVAL;
2467         }
2468
2469         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2470                    BTRFS_FSID_SIZE) != 0) {
2471                 btrfs_err(fs_info,
2472                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2473                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2474                 ret = -EINVAL;
2475         }
2476
2477         /*
2478          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2479          * done later
2480          */
2481         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2482                 btrfs_err(fs_info, "bytes_used is too small %llu",
2483                           btrfs_super_bytes_used(sb));
2484                 ret = -EINVAL;
2485         }
2486         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2487                 btrfs_err(fs_info, "invalid stripesize %u",
2488                           btrfs_super_stripesize(sb));
2489                 ret = -EINVAL;
2490         }
2491         if (btrfs_super_num_devices(sb) > (1UL << 31))
2492                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2493                            btrfs_super_num_devices(sb));
2494         if (btrfs_super_num_devices(sb) == 0) {
2495                 btrfs_err(fs_info, "number of devices is 0");
2496                 ret = -EINVAL;
2497         }
2498
2499         if (mirror_num >= 0 &&
2500             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2501                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2502                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2503                 ret = -EINVAL;
2504         }
2505
2506         /*
2507          * Obvious sys_chunk_array corruptions, it must hold at least one key
2508          * and one chunk
2509          */
2510         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2511                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2512                           btrfs_super_sys_array_size(sb),
2513                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2514                 ret = -EINVAL;
2515         }
2516         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2517                         + sizeof(struct btrfs_chunk)) {
2518                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2519                           btrfs_super_sys_array_size(sb),
2520                           sizeof(struct btrfs_disk_key)
2521                           + sizeof(struct btrfs_chunk));
2522                 ret = -EINVAL;
2523         }
2524
2525         /*
2526          * The generation is a global counter, we'll trust it more than the others
2527          * but it's still possible that it's the one that's wrong.
2528          */
2529         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2530                 btrfs_warn(fs_info,
2531                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2532                         btrfs_super_generation(sb),
2533                         btrfs_super_chunk_root_generation(sb));
2534         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2535             && btrfs_super_cache_generation(sb) != (u64)-1)
2536                 btrfs_warn(fs_info,
2537                         "suspicious: generation < cache_generation: %llu < %llu",
2538                         btrfs_super_generation(sb),
2539                         btrfs_super_cache_generation(sb));
2540
2541         return ret;
2542 }
2543
2544 /*
2545  * Validation of super block at mount time.
2546  * Some checks already done early at mount time, like csum type and incompat
2547  * flags will be skipped.
2548  */
2549 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2550 {
2551         return validate_super(fs_info, fs_info->super_copy, 0);
2552 }
2553
2554 /*
2555  * Validation of super block at write time.
2556  * Some checks like bytenr check will be skipped as their values will be
2557  * overwritten soon.
2558  * Extra checks like csum type and incompat flags will be done here.
2559  */
2560 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2561                                       struct btrfs_super_block *sb)
2562 {
2563         int ret;
2564
2565         ret = validate_super(fs_info, sb, -1);
2566         if (ret < 0)
2567                 goto out;
2568         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2569                 ret = -EUCLEAN;
2570                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2571                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2572                 goto out;
2573         }
2574         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2575                 ret = -EUCLEAN;
2576                 btrfs_err(fs_info,
2577                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2578                           btrfs_super_incompat_flags(sb),
2579                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2580                 goto out;
2581         }
2582 out:
2583         if (ret < 0)
2584                 btrfs_err(fs_info,
2585                 "super block corruption detected before writing it to disk");
2586         return ret;
2587 }
2588
2589 int open_ctree(struct super_block *sb,
2590                struct btrfs_fs_devices *fs_devices,
2591                char *options)
2592 {
2593         u32 sectorsize;
2594         u32 nodesize;
2595         u32 stripesize;
2596         u64 generation;
2597         u64 features;
2598         struct btrfs_key location;
2599         struct buffer_head *bh;
2600         struct btrfs_super_block *disk_super;
2601         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2602         struct btrfs_root *tree_root;
2603         struct btrfs_root *chunk_root;
2604         int ret;
2605         int err = -EINVAL;
2606         int num_backups_tried = 0;
2607         int backup_index = 0;
2608         int clear_free_space_tree = 0;
2609         int level;
2610
2611         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2612         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2613         if (!tree_root || !chunk_root) {
2614                 err = -ENOMEM;
2615                 goto fail;
2616         }
2617
2618         ret = init_srcu_struct(&fs_info->subvol_srcu);
2619         if (ret) {
2620                 err = ret;
2621                 goto fail;
2622         }
2623
2624         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2625         if (ret) {
2626                 err = ret;
2627                 goto fail_srcu;
2628         }
2629         fs_info->dirty_metadata_batch = PAGE_SIZE *
2630                                         (1 + ilog2(nr_cpu_ids));
2631
2632         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2633         if (ret) {
2634                 err = ret;
2635                 goto fail_dirty_metadata_bytes;
2636         }
2637
2638         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2639                         GFP_KERNEL);
2640         if (ret) {
2641                 err = ret;
2642                 goto fail_delalloc_bytes;
2643         }
2644
2645         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2646         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2647         INIT_LIST_HEAD(&fs_info->trans_list);
2648         INIT_LIST_HEAD(&fs_info->dead_roots);
2649         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2650         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2651         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2652         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2653         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2654         spin_lock_init(&fs_info->delalloc_root_lock);
2655         spin_lock_init(&fs_info->trans_lock);
2656         spin_lock_init(&fs_info->fs_roots_radix_lock);
2657         spin_lock_init(&fs_info->delayed_iput_lock);
2658         spin_lock_init(&fs_info->defrag_inodes_lock);
2659         spin_lock_init(&fs_info->tree_mod_seq_lock);
2660         spin_lock_init(&fs_info->super_lock);
2661         spin_lock_init(&fs_info->qgroup_op_lock);
2662         spin_lock_init(&fs_info->buffer_lock);
2663         spin_lock_init(&fs_info->unused_bgs_lock);
2664         rwlock_init(&fs_info->tree_mod_log_lock);
2665         mutex_init(&fs_info->unused_bg_unpin_mutex);
2666         mutex_init(&fs_info->delete_unused_bgs_mutex);
2667         mutex_init(&fs_info->reloc_mutex);
2668         mutex_init(&fs_info->delalloc_root_mutex);
2669         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2670         seqlock_init(&fs_info->profiles_lock);
2671
2672         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2673         INIT_LIST_HEAD(&fs_info->space_info);
2674         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2675         INIT_LIST_HEAD(&fs_info->unused_bgs);
2676         btrfs_mapping_init(&fs_info->mapping_tree);
2677         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2678                              BTRFS_BLOCK_RSV_GLOBAL);
2679         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2680         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2681         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2682         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2683                              BTRFS_BLOCK_RSV_DELOPS);
2684         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2685                              BTRFS_BLOCK_RSV_DELREFS);
2686
2687         atomic_set(&fs_info->async_delalloc_pages, 0);
2688         atomic_set(&fs_info->defrag_running, 0);
2689         atomic_set(&fs_info->qgroup_op_seq, 0);
2690         atomic_set(&fs_info->reada_works_cnt, 0);
2691         atomic64_set(&fs_info->tree_mod_seq, 0);
2692         fs_info->sb = sb;
2693         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2694         fs_info->metadata_ratio = 0;
2695         fs_info->defrag_inodes = RB_ROOT;
2696         atomic64_set(&fs_info->free_chunk_space, 0);
2697         fs_info->tree_mod_log = RB_ROOT;
2698         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2699         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2700         /* readahead state */
2701         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2702         spin_lock_init(&fs_info->reada_lock);
2703         btrfs_init_ref_verify(fs_info);
2704
2705         fs_info->thread_pool_size = min_t(unsigned long,
2706                                           num_online_cpus() + 2, 8);
2707
2708         INIT_LIST_HEAD(&fs_info->ordered_roots);
2709         spin_lock_init(&fs_info->ordered_root_lock);
2710
2711         fs_info->btree_inode = new_inode(sb);
2712         if (!fs_info->btree_inode) {
2713                 err = -ENOMEM;
2714                 goto fail_bio_counter;
2715         }
2716         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2717
2718         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2719                                         GFP_KERNEL);
2720         if (!fs_info->delayed_root) {
2721                 err = -ENOMEM;
2722                 goto fail_iput;
2723         }
2724         btrfs_init_delayed_root(fs_info->delayed_root);
2725
2726         btrfs_init_scrub(fs_info);
2727 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2728         fs_info->check_integrity_print_mask = 0;
2729 #endif
2730         btrfs_init_balance(fs_info);
2731         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2732
2733         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2734         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2735
2736         btrfs_init_btree_inode(fs_info);
2737
2738         spin_lock_init(&fs_info->block_group_cache_lock);
2739         fs_info->block_group_cache_tree = RB_ROOT;
2740         fs_info->first_logical_byte = (u64)-1;
2741
2742         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2743         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2744         fs_info->pinned_extents = &fs_info->freed_extents[0];
2745         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2746
2747         mutex_init(&fs_info->ordered_operations_mutex);
2748         mutex_init(&fs_info->tree_log_mutex);
2749         mutex_init(&fs_info->chunk_mutex);
2750         mutex_init(&fs_info->transaction_kthread_mutex);
2751         mutex_init(&fs_info->cleaner_mutex);
2752         mutex_init(&fs_info->ro_block_group_mutex);
2753         init_rwsem(&fs_info->commit_root_sem);
2754         init_rwsem(&fs_info->cleanup_work_sem);
2755         init_rwsem(&fs_info->subvol_sem);
2756         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2757
2758         btrfs_init_dev_replace_locks(fs_info);
2759         btrfs_init_qgroup(fs_info);
2760
2761         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2762         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2763
2764         init_waitqueue_head(&fs_info->transaction_throttle);
2765         init_waitqueue_head(&fs_info->transaction_wait);
2766         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2767         init_waitqueue_head(&fs_info->async_submit_wait);
2768
2769         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2770
2771         /* Usable values until the real ones are cached from the superblock */
2772         fs_info->nodesize = 4096;
2773         fs_info->sectorsize = 4096;
2774         fs_info->stripesize = 4096;
2775
2776         spin_lock_init(&fs_info->swapfile_pins_lock);
2777         fs_info->swapfile_pins = RB_ROOT;
2778
2779         ret = btrfs_alloc_stripe_hash_table(fs_info);
2780         if (ret) {
2781                 err = ret;
2782                 goto fail_alloc;
2783         }
2784
2785         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2786
2787         invalidate_bdev(fs_devices->latest_bdev);
2788
2789         /*
2790          * Read super block and check the signature bytes only
2791          */
2792         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2793         if (IS_ERR(bh)) {
2794                 err = PTR_ERR(bh);
2795                 goto fail_alloc;
2796         }
2797
2798         /*
2799          * We want to check superblock checksum, the type is stored inside.
2800          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2801          */
2802         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2803                 btrfs_err(fs_info, "superblock checksum mismatch");
2804                 err = -EINVAL;
2805                 brelse(bh);
2806                 goto fail_alloc;
2807         }
2808
2809         /*
2810          * super_copy is zeroed at allocation time and we never touch the
2811          * following bytes up to INFO_SIZE, the checksum is calculated from
2812          * the whole block of INFO_SIZE
2813          */
2814         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2815         brelse(bh);
2816
2817         disk_super = fs_info->super_copy;
2818
2819         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2820                        BTRFS_FSID_SIZE));
2821
2822         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2823                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2824                                 fs_info->super_copy->metadata_uuid,
2825                                 BTRFS_FSID_SIZE));
2826         }
2827
2828         features = btrfs_super_flags(disk_super);
2829         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2830                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2831                 btrfs_set_super_flags(disk_super, features);
2832                 btrfs_info(fs_info,
2833                         "found metadata UUID change in progress flag, clearing");
2834         }
2835
2836         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2837                sizeof(*fs_info->super_for_commit));
2838
2839         ret = btrfs_validate_mount_super(fs_info);
2840         if (ret) {
2841                 btrfs_err(fs_info, "superblock contains fatal errors");
2842                 err = -EINVAL;
2843                 goto fail_alloc;
2844         }
2845
2846         if (!btrfs_super_root(disk_super))
2847                 goto fail_alloc;
2848
2849         /* check FS state, whether FS is broken. */
2850         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2851                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2852
2853         /*
2854          * run through our array of backup supers and setup
2855          * our ring pointer to the oldest one
2856          */
2857         generation = btrfs_super_generation(disk_super);
2858         find_oldest_super_backup(fs_info, generation);
2859
2860         /*
2861          * In the long term, we'll store the compression type in the super
2862          * block, and it'll be used for per file compression control.
2863          */
2864         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2865
2866         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2867         if (ret) {
2868                 err = ret;
2869                 goto fail_alloc;
2870         }
2871
2872         features = btrfs_super_incompat_flags(disk_super) &
2873                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2874         if (features) {
2875                 btrfs_err(fs_info,
2876                     "cannot mount because of unsupported optional features (%llx)",
2877                     features);
2878                 err = -EINVAL;
2879                 goto fail_alloc;
2880         }
2881
2882         features = btrfs_super_incompat_flags(disk_super);
2883         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2884         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2885                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2886         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2887                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2888
2889         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2890                 btrfs_info(fs_info, "has skinny extents");
2891
2892         /*
2893          * flag our filesystem as having big metadata blocks if
2894          * they are bigger than the page size
2895          */
2896         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2897                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2898                         btrfs_info(fs_info,
2899                                 "flagging fs with big metadata feature");
2900                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2901         }
2902
2903         nodesize = btrfs_super_nodesize(disk_super);
2904         sectorsize = btrfs_super_sectorsize(disk_super);
2905         stripesize = sectorsize;
2906         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2907         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2908
2909         /* Cache block sizes */
2910         fs_info->nodesize = nodesize;
2911         fs_info->sectorsize = sectorsize;
2912         fs_info->stripesize = stripesize;
2913
2914         /*
2915          * mixed block groups end up with duplicate but slightly offset
2916          * extent buffers for the same range.  It leads to corruptions
2917          */
2918         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2919             (sectorsize != nodesize)) {
2920                 btrfs_err(fs_info,
2921 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2922                         nodesize, sectorsize);
2923                 goto fail_alloc;
2924         }
2925
2926         /*
2927          * Needn't use the lock because there is no other task which will
2928          * update the flag.
2929          */
2930         btrfs_set_super_incompat_flags(disk_super, features);
2931
2932         features = btrfs_super_compat_ro_flags(disk_super) &
2933                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2934         if (!sb_rdonly(sb) && features) {
2935                 btrfs_err(fs_info,
2936         "cannot mount read-write because of unsupported optional features (%llx)",
2937                        features);
2938                 err = -EINVAL;
2939                 goto fail_alloc;
2940         }
2941
2942         ret = btrfs_init_workqueues(fs_info, fs_devices);
2943         if (ret) {
2944                 err = ret;
2945                 goto fail_sb_buffer;
2946         }
2947
2948         sb->s_bdi->congested_fn = btrfs_congested_fn;
2949         sb->s_bdi->congested_data = fs_info;
2950         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2951         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2952         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2953         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2954
2955         sb->s_blocksize = sectorsize;
2956         sb->s_blocksize_bits = blksize_bits(sectorsize);
2957         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2958
2959         mutex_lock(&fs_info->chunk_mutex);
2960         ret = btrfs_read_sys_array(fs_info);
2961         mutex_unlock(&fs_info->chunk_mutex);
2962         if (ret) {
2963                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2964                 goto fail_sb_buffer;
2965         }
2966
2967         generation = btrfs_super_chunk_root_generation(disk_super);
2968         level = btrfs_super_chunk_root_level(disk_super);
2969
2970         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2971
2972         chunk_root->node = read_tree_block(fs_info,
2973                                            btrfs_super_chunk_root(disk_super),
2974                                            generation, level, NULL);
2975         if (IS_ERR(chunk_root->node) ||
2976             !extent_buffer_uptodate(chunk_root->node)) {
2977                 btrfs_err(fs_info, "failed to read chunk root");
2978                 if (!IS_ERR(chunk_root->node))
2979                         free_extent_buffer(chunk_root->node);
2980                 chunk_root->node = NULL;
2981                 goto fail_tree_roots;
2982         }
2983         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2984         chunk_root->commit_root = btrfs_root_node(chunk_root);
2985
2986         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2987            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2988
2989         ret = btrfs_read_chunk_tree(fs_info);
2990         if (ret) {
2991                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2992                 goto fail_tree_roots;
2993         }
2994
2995         /*
2996          * Keep the devid that is marked to be the target device for the
2997          * device replace procedure
2998          */
2999         btrfs_free_extra_devids(fs_devices, 0);
3000
3001         if (!fs_devices->latest_bdev) {
3002                 btrfs_err(fs_info, "failed to read devices");
3003                 goto fail_tree_roots;
3004         }
3005
3006 retry_root_backup:
3007         generation = btrfs_super_generation(disk_super);
3008         level = btrfs_super_root_level(disk_super);
3009
3010         tree_root->node = read_tree_block(fs_info,
3011                                           btrfs_super_root(disk_super),
3012                                           generation, level, NULL);
3013         if (IS_ERR(tree_root->node) ||
3014             !extent_buffer_uptodate(tree_root->node)) {
3015                 btrfs_warn(fs_info, "failed to read tree root");
3016                 if (!IS_ERR(tree_root->node))
3017                         free_extent_buffer(tree_root->node);
3018                 tree_root->node = NULL;
3019                 goto recovery_tree_root;
3020         }
3021
3022         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3023         tree_root->commit_root = btrfs_root_node(tree_root);
3024         btrfs_set_root_refs(&tree_root->root_item, 1);
3025
3026         mutex_lock(&tree_root->objectid_mutex);
3027         ret = btrfs_find_highest_objectid(tree_root,
3028                                         &tree_root->highest_objectid);
3029         if (ret) {
3030                 mutex_unlock(&tree_root->objectid_mutex);
3031                 goto recovery_tree_root;
3032         }
3033
3034         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3035
3036         mutex_unlock(&tree_root->objectid_mutex);
3037
3038         ret = btrfs_read_roots(fs_info);
3039         if (ret)
3040                 goto recovery_tree_root;
3041
3042         fs_info->generation = generation;
3043         fs_info->last_trans_committed = generation;
3044
3045         ret = btrfs_verify_dev_extents(fs_info);
3046         if (ret) {
3047                 btrfs_err(fs_info,
3048                           "failed to verify dev extents against chunks: %d",
3049                           ret);
3050                 goto fail_block_groups;
3051         }
3052         ret = btrfs_recover_balance(fs_info);
3053         if (ret) {
3054                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3055                 goto fail_block_groups;
3056         }
3057
3058         ret = btrfs_init_dev_stats(fs_info);
3059         if (ret) {
3060                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3061                 goto fail_block_groups;
3062         }
3063
3064         ret = btrfs_init_dev_replace(fs_info);
3065         if (ret) {
3066                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3067                 goto fail_block_groups;
3068         }
3069
3070         btrfs_free_extra_devids(fs_devices, 1);
3071
3072         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3073         if (ret) {
3074                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3075                                 ret);
3076                 goto fail_block_groups;
3077         }
3078
3079         ret = btrfs_sysfs_add_device(fs_devices);
3080         if (ret) {
3081                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3082                                 ret);
3083                 goto fail_fsdev_sysfs;
3084         }
3085
3086         ret = btrfs_sysfs_add_mounted(fs_info);
3087         if (ret) {
3088                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3089                 goto fail_fsdev_sysfs;
3090         }
3091
3092         ret = btrfs_init_space_info(fs_info);
3093         if (ret) {
3094                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3095                 goto fail_sysfs;
3096         }
3097
3098         ret = btrfs_read_block_groups(fs_info);
3099         if (ret) {
3100                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3101                 goto fail_sysfs;
3102         }
3103
3104         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3105                 btrfs_warn(fs_info,
3106                 "writable mount is not allowed due to too many missing devices");
3107                 goto fail_sysfs;
3108         }
3109
3110         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3111                                                "btrfs-cleaner");
3112         if (IS_ERR(fs_info->cleaner_kthread))
3113                 goto fail_sysfs;
3114
3115         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3116                                                    tree_root,
3117                                                    "btrfs-transaction");
3118         if (IS_ERR(fs_info->transaction_kthread))
3119                 goto fail_cleaner;
3120
3121         if (!btrfs_test_opt(fs_info, NOSSD) &&
3122             !fs_info->fs_devices->rotating) {
3123                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3124         }
3125
3126         /*
3127          * Mount does not set all options immediately, we can do it now and do
3128          * not have to wait for transaction commit
3129          */
3130         btrfs_apply_pending_changes(fs_info);
3131
3132 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3133         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3134                 ret = btrfsic_mount(fs_info, fs_devices,
3135                                     btrfs_test_opt(fs_info,
3136                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3137                                     1 : 0,
3138                                     fs_info->check_integrity_print_mask);
3139                 if (ret)
3140                         btrfs_warn(fs_info,
3141                                 "failed to initialize integrity check module: %d",
3142                                 ret);
3143         }
3144 #endif
3145         ret = btrfs_read_qgroup_config(fs_info);
3146         if (ret)
3147                 goto fail_trans_kthread;
3148
3149         if (btrfs_build_ref_tree(fs_info))
3150                 btrfs_err(fs_info, "couldn't build ref tree");
3151
3152         /* do not make disk changes in broken FS or nologreplay is given */
3153         if (btrfs_super_log_root(disk_super) != 0 &&
3154             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3155                 ret = btrfs_replay_log(fs_info, fs_devices);
3156                 if (ret) {
3157                         err = ret;
3158                         goto fail_qgroup;
3159                 }
3160         }
3161
3162         ret = btrfs_find_orphan_roots(fs_info);
3163         if (ret)
3164                 goto fail_qgroup;
3165
3166         if (!sb_rdonly(sb)) {
3167                 ret = btrfs_cleanup_fs_roots(fs_info);
3168                 if (ret)
3169                         goto fail_qgroup;
3170
3171                 mutex_lock(&fs_info->cleaner_mutex);
3172                 ret = btrfs_recover_relocation(tree_root);
3173                 mutex_unlock(&fs_info->cleaner_mutex);
3174                 if (ret < 0) {
3175                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3176                                         ret);
3177                         err = -EINVAL;
3178                         goto fail_qgroup;
3179                 }
3180         }
3181
3182         location.objectid = BTRFS_FS_TREE_OBJECTID;
3183         location.type = BTRFS_ROOT_ITEM_KEY;
3184         location.offset = 0;
3185
3186         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3187         if (IS_ERR(fs_info->fs_root)) {
3188                 err = PTR_ERR(fs_info->fs_root);
3189                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3190                 goto fail_qgroup;
3191         }
3192
3193         if (sb_rdonly(sb))
3194                 return 0;
3195
3196         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3197             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3198                 clear_free_space_tree = 1;
3199         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3200                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3201                 btrfs_warn(fs_info, "free space tree is invalid");
3202                 clear_free_space_tree = 1;
3203         }
3204
3205         if (clear_free_space_tree) {
3206                 btrfs_info(fs_info, "clearing free space tree");
3207                 ret = btrfs_clear_free_space_tree(fs_info);
3208                 if (ret) {
3209                         btrfs_warn(fs_info,
3210                                    "failed to clear free space tree: %d", ret);
3211                         close_ctree(fs_info);
3212                         return ret;
3213                 }
3214         }
3215
3216         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3217             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3218                 btrfs_info(fs_info, "creating free space tree");
3219                 ret = btrfs_create_free_space_tree(fs_info);
3220                 if (ret) {
3221                         btrfs_warn(fs_info,
3222                                 "failed to create free space tree: %d", ret);
3223                         close_ctree(fs_info);
3224                         return ret;
3225                 }
3226         }
3227
3228         down_read(&fs_info->cleanup_work_sem);
3229         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3230             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3231                 up_read(&fs_info->cleanup_work_sem);
3232                 close_ctree(fs_info);
3233                 return ret;
3234         }
3235         up_read(&fs_info->cleanup_work_sem);
3236
3237         ret = btrfs_resume_balance_async(fs_info);
3238         if (ret) {
3239                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3240                 close_ctree(fs_info);
3241                 return ret;
3242         }
3243
3244         ret = btrfs_resume_dev_replace_async(fs_info);
3245         if (ret) {
3246                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3247                 close_ctree(fs_info);
3248                 return ret;
3249         }
3250
3251         btrfs_qgroup_rescan_resume(fs_info);
3252
3253         if (!fs_info->uuid_root) {
3254                 btrfs_info(fs_info, "creating UUID tree");
3255                 ret = btrfs_create_uuid_tree(fs_info);
3256                 if (ret) {
3257                         btrfs_warn(fs_info,
3258                                 "failed to create the UUID tree: %d", ret);
3259                         close_ctree(fs_info);
3260                         return ret;
3261                 }
3262         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3263                    fs_info->generation !=
3264                                 btrfs_super_uuid_tree_generation(disk_super)) {
3265                 btrfs_info(fs_info, "checking UUID tree");
3266                 ret = btrfs_check_uuid_tree(fs_info);
3267                 if (ret) {
3268                         btrfs_warn(fs_info,
3269                                 "failed to check the UUID tree: %d", ret);
3270                         close_ctree(fs_info);
3271                         return ret;
3272                 }
3273         } else {
3274                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3275         }
3276         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3277
3278         /*
3279          * backuproot only affect mount behavior, and if open_ctree succeeded,
3280          * no need to keep the flag
3281          */
3282         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3283
3284         return 0;
3285
3286 fail_qgroup:
3287         btrfs_free_qgroup_config(fs_info);
3288 fail_trans_kthread:
3289         kthread_stop(fs_info->transaction_kthread);
3290         btrfs_cleanup_transaction(fs_info);
3291         btrfs_free_fs_roots(fs_info);
3292 fail_cleaner:
3293         kthread_stop(fs_info->cleaner_kthread);
3294
3295         /*
3296          * make sure we're done with the btree inode before we stop our
3297          * kthreads
3298          */
3299         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3300
3301 fail_sysfs:
3302         btrfs_sysfs_remove_mounted(fs_info);
3303
3304 fail_fsdev_sysfs:
3305         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3306
3307 fail_block_groups:
3308         btrfs_put_block_group_cache(fs_info);
3309
3310 fail_tree_roots:
3311         free_root_pointers(fs_info, 1);
3312         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3313
3314 fail_sb_buffer:
3315         btrfs_stop_all_workers(fs_info);
3316         btrfs_free_block_groups(fs_info);
3317 fail_alloc:
3318 fail_iput:
3319         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3320
3321         iput(fs_info->btree_inode);
3322 fail_bio_counter:
3323         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3324 fail_delalloc_bytes:
3325         percpu_counter_destroy(&fs_info->delalloc_bytes);
3326 fail_dirty_metadata_bytes:
3327         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3328 fail_srcu:
3329         cleanup_srcu_struct(&fs_info->subvol_srcu);
3330 fail:
3331         btrfs_free_stripe_hash_table(fs_info);
3332         btrfs_close_devices(fs_info->fs_devices);
3333         return err;
3334
3335 recovery_tree_root:
3336         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3337                 goto fail_tree_roots;
3338
3339         free_root_pointers(fs_info, 0);
3340
3341         /* don't use the log in recovery mode, it won't be valid */
3342         btrfs_set_super_log_root(disk_super, 0);
3343
3344         /* we can't trust the free space cache either */
3345         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3346
3347         ret = next_root_backup(fs_info, fs_info->super_copy,
3348                                &num_backups_tried, &backup_index);
3349         if (ret == -1)
3350                 goto fail_block_groups;
3351         goto retry_root_backup;
3352 }
3353 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3354
3355 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3356 {
3357         if (uptodate) {
3358                 set_buffer_uptodate(bh);
3359         } else {
3360                 struct btrfs_device *device = (struct btrfs_device *)
3361                         bh->b_private;
3362
3363                 btrfs_warn_rl_in_rcu(device->fs_info,
3364                                 "lost page write due to IO error on %s",
3365                                           rcu_str_deref(device->name));
3366                 /* note, we don't set_buffer_write_io_error because we have
3367                  * our own ways of dealing with the IO errors
3368                  */
3369                 clear_buffer_uptodate(bh);
3370                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3371         }
3372         unlock_buffer(bh);
3373         put_bh(bh);
3374 }
3375
3376 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3377                         struct buffer_head **bh_ret)
3378 {
3379         struct buffer_head *bh;
3380         struct btrfs_super_block *super;
3381         u64 bytenr;
3382
3383         bytenr = btrfs_sb_offset(copy_num);
3384         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3385                 return -EINVAL;
3386
3387         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3388         /*
3389          * If we fail to read from the underlying devices, as of now
3390          * the best option we have is to mark it EIO.
3391          */
3392         if (!bh)
3393                 return -EIO;
3394
3395         super = (struct btrfs_super_block *)bh->b_data;
3396         if (btrfs_super_bytenr(super) != bytenr ||
3397                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3398                 brelse(bh);
3399                 return -EINVAL;
3400         }
3401
3402         *bh_ret = bh;
3403         return 0;
3404 }
3405
3406
3407 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3408 {
3409         struct buffer_head *bh;
3410         struct buffer_head *latest = NULL;
3411         struct btrfs_super_block *super;
3412         int i;
3413         u64 transid = 0;
3414         int ret = -EINVAL;
3415
3416         /* we would like to check all the supers, but that would make
3417          * a btrfs mount succeed after a mkfs from a different FS.
3418          * So, we need to add a special mount option to scan for
3419          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3420          */
3421         for (i = 0; i < 1; i++) {
3422                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3423                 if (ret)
3424                         continue;
3425
3426                 super = (struct btrfs_super_block *)bh->b_data;
3427
3428                 if (!latest || btrfs_super_generation(super) > transid) {
3429                         brelse(latest);
3430                         latest = bh;
3431                         transid = btrfs_super_generation(super);
3432                 } else {
3433                         brelse(bh);
3434                 }
3435         }
3436
3437         if (!latest)
3438                 return ERR_PTR(ret);
3439
3440         return latest;
3441 }
3442
3443 /*
3444  * Write superblock @sb to the @device. Do not wait for completion, all the
3445  * buffer heads we write are pinned.
3446  *
3447  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3448  * the expected device size at commit time. Note that max_mirrors must be
3449  * same for write and wait phases.
3450  *
3451  * Return number of errors when buffer head is not found or submission fails.
3452  */
3453 static int write_dev_supers(struct btrfs_device *device,
3454                             struct btrfs_super_block *sb, int max_mirrors)
3455 {
3456         struct buffer_head *bh;
3457         int i;
3458         int ret;
3459         int errors = 0;
3460         u32 crc;
3461         u64 bytenr;
3462         int op_flags;
3463
3464         if (max_mirrors == 0)
3465                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3466
3467         for (i = 0; i < max_mirrors; i++) {
3468                 bytenr = btrfs_sb_offset(i);
3469                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3470                     device->commit_total_bytes)
3471                         break;
3472
3473                 btrfs_set_super_bytenr(sb, bytenr);
3474
3475                 crc = ~(u32)0;
3476                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3477                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3478                 btrfs_csum_final(crc, sb->csum);
3479
3480                 /* One reference for us, and we leave it for the caller */
3481                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3482                               BTRFS_SUPER_INFO_SIZE);
3483                 if (!bh) {
3484                         btrfs_err(device->fs_info,
3485                             "couldn't get super buffer head for bytenr %llu",
3486                             bytenr);
3487                         errors++;
3488                         continue;
3489                 }
3490
3491                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3492
3493                 /* one reference for submit_bh */
3494                 get_bh(bh);
3495
3496                 set_buffer_uptodate(bh);
3497                 lock_buffer(bh);
3498                 bh->b_end_io = btrfs_end_buffer_write_sync;
3499                 bh->b_private = device;
3500
3501                 /*
3502                  * we fua the first super.  The others we allow
3503                  * to go down lazy.
3504                  */
3505                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3506                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3507                         op_flags |= REQ_FUA;
3508                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3509                 if (ret)
3510                         errors++;
3511         }
3512         return errors < i ? 0 : -1;
3513 }
3514
3515 /*
3516  * Wait for write completion of superblocks done by write_dev_supers,
3517  * @max_mirrors same for write and wait phases.
3518  *
3519  * Return number of errors when buffer head is not found or not marked up to
3520  * date.
3521  */
3522 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3523 {
3524         struct buffer_head *bh;
3525         int i;
3526         int errors = 0;
3527         bool primary_failed = false;
3528         u64 bytenr;
3529
3530         if (max_mirrors == 0)
3531                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3532
3533         for (i = 0; i < max_mirrors; i++) {
3534                 bytenr = btrfs_sb_offset(i);
3535                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3536                     device->commit_total_bytes)
3537                         break;
3538
3539                 bh = __find_get_block(device->bdev,
3540                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3541                                       BTRFS_SUPER_INFO_SIZE);
3542                 if (!bh) {
3543                         errors++;
3544                         if (i == 0)
3545                                 primary_failed = true;
3546                         continue;
3547                 }
3548                 wait_on_buffer(bh);
3549                 if (!buffer_uptodate(bh)) {
3550                         errors++;
3551                         if (i == 0)
3552                                 primary_failed = true;
3553                 }
3554
3555                 /* drop our reference */
3556                 brelse(bh);
3557
3558                 /* drop the reference from the writing run */
3559                 brelse(bh);
3560         }
3561
3562         /* log error, force error return */
3563         if (primary_failed) {
3564                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3565                           device->devid);
3566                 return -1;
3567         }
3568
3569         return errors < i ? 0 : -1;
3570 }
3571
3572 /*
3573  * endio for the write_dev_flush, this will wake anyone waiting
3574  * for the barrier when it is done
3575  */
3576 static void btrfs_end_empty_barrier(struct bio *bio)
3577 {
3578         complete(bio->bi_private);
3579 }
3580
3581 /*
3582  * Submit a flush request to the device if it supports it. Error handling is
3583  * done in the waiting counterpart.
3584  */
3585 static void write_dev_flush(struct btrfs_device *device)
3586 {
3587         struct request_queue *q = bdev_get_queue(device->bdev);
3588         struct bio *bio = device->flush_bio;
3589
3590         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3591                 return;
3592
3593         bio_reset(bio);
3594         bio->bi_end_io = btrfs_end_empty_barrier;
3595         bio_set_dev(bio, device->bdev);
3596         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3597         init_completion(&device->flush_wait);
3598         bio->bi_private = &device->flush_wait;
3599
3600         btrfsic_submit_bio(bio);
3601         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3602 }
3603
3604 /*
3605  * If the flush bio has been submitted by write_dev_flush, wait for it.
3606  */
3607 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3608 {
3609         struct bio *bio = device->flush_bio;
3610
3611         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3612                 return BLK_STS_OK;
3613
3614         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3615         wait_for_completion_io(&device->flush_wait);
3616
3617         return bio->bi_status;
3618 }
3619
3620 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3621 {
3622         if (!btrfs_check_rw_degradable(fs_info, NULL))
3623                 return -EIO;
3624         return 0;
3625 }
3626
3627 /*
3628  * send an empty flush down to each device in parallel,
3629  * then wait for them
3630  */
3631 static int barrier_all_devices(struct btrfs_fs_info *info)
3632 {
3633         struct list_head *head;
3634         struct btrfs_device *dev;
3635         int errors_wait = 0;
3636         blk_status_t ret;
3637
3638         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3639         /* send down all the barriers */
3640         head = &info->fs_devices->devices;
3641         list_for_each_entry(dev, head, dev_list) {
3642                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3643                         continue;
3644                 if (!dev->bdev)
3645                         continue;
3646                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3647                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3648                         continue;
3649
3650                 write_dev_flush(dev);
3651                 dev->last_flush_error = BLK_STS_OK;
3652         }
3653
3654         /* wait for all the barriers */
3655         list_for_each_entry(dev, head, dev_list) {
3656                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3657                         continue;
3658                 if (!dev->bdev) {
3659                         errors_wait++;
3660                         continue;
3661                 }
3662                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3663                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3664                         continue;
3665
3666                 ret = wait_dev_flush(dev);
3667                 if (ret) {
3668                         dev->last_flush_error = ret;
3669                         btrfs_dev_stat_inc_and_print(dev,
3670                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3671                         errors_wait++;
3672                 }
3673         }
3674
3675         if (errors_wait) {
3676                 /*
3677                  * At some point we need the status of all disks
3678                  * to arrive at the volume status. So error checking
3679                  * is being pushed to a separate loop.
3680                  */
3681                 return check_barrier_error(info);
3682         }
3683         return 0;
3684 }
3685
3686 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3687 {
3688         int raid_type;
3689         int min_tolerated = INT_MAX;
3690
3691         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3692             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3693                 min_tolerated = min(min_tolerated,
3694                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3695                                     tolerated_failures);
3696
3697         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3698                 if (raid_type == BTRFS_RAID_SINGLE)
3699                         continue;
3700                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3701                         continue;
3702                 min_tolerated = min(min_tolerated,
3703                                     btrfs_raid_array[raid_type].
3704                                     tolerated_failures);
3705         }
3706
3707         if (min_tolerated == INT_MAX) {
3708                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3709                 min_tolerated = 0;
3710         }
3711
3712         return min_tolerated;
3713 }
3714
3715 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3716 {
3717         struct list_head *head;
3718         struct btrfs_device *dev;
3719         struct btrfs_super_block *sb;
3720         struct btrfs_dev_item *dev_item;
3721         int ret;
3722         int do_barriers;
3723         int max_errors;
3724         int total_errors = 0;
3725         u64 flags;
3726
3727         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3728
3729         /*
3730          * max_mirrors == 0 indicates we're from commit_transaction,
3731          * not from fsync where the tree roots in fs_info have not
3732          * been consistent on disk.
3733          */
3734         if (max_mirrors == 0)
3735                 backup_super_roots(fs_info);
3736
3737         sb = fs_info->super_for_commit;
3738         dev_item = &sb->dev_item;
3739
3740         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3741         head = &fs_info->fs_devices->devices;
3742         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3743
3744         if (do_barriers) {
3745                 ret = barrier_all_devices(fs_info);
3746                 if (ret) {
3747                         mutex_unlock(
3748                                 &fs_info->fs_devices->device_list_mutex);
3749                         btrfs_handle_fs_error(fs_info, ret,
3750                                               "errors while submitting device barriers.");
3751                         return ret;
3752                 }
3753         }
3754
3755         list_for_each_entry(dev, head, dev_list) {
3756                 if (!dev->bdev) {
3757                         total_errors++;
3758                         continue;
3759                 }
3760                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3761                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3762                         continue;
3763
3764                 btrfs_set_stack_device_generation(dev_item, 0);
3765                 btrfs_set_stack_device_type(dev_item, dev->type);
3766                 btrfs_set_stack_device_id(dev_item, dev->devid);
3767                 btrfs_set_stack_device_total_bytes(dev_item,
3768                                                    dev->commit_total_bytes);
3769                 btrfs_set_stack_device_bytes_used(dev_item,
3770                                                   dev->commit_bytes_used);
3771                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3772                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3773                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3774                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3775                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3776                        BTRFS_FSID_SIZE);
3777
3778                 flags = btrfs_super_flags(sb);
3779                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3780
3781                 ret = btrfs_validate_write_super(fs_info, sb);
3782                 if (ret < 0) {
3783                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3784                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3785                                 "unexpected superblock corruption detected");
3786                         return -EUCLEAN;
3787                 }
3788
3789                 ret = write_dev_supers(dev, sb, max_mirrors);
3790                 if (ret)
3791                         total_errors++;
3792         }
3793         if (total_errors > max_errors) {
3794                 btrfs_err(fs_info, "%d errors while writing supers",
3795                           total_errors);
3796                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3797
3798                 /* FUA is masked off if unsupported and can't be the reason */
3799                 btrfs_handle_fs_error(fs_info, -EIO,
3800                                       "%d errors while writing supers",
3801                                       total_errors);
3802                 return -EIO;
3803         }
3804
3805         total_errors = 0;
3806         list_for_each_entry(dev, head, dev_list) {
3807                 if (!dev->bdev)
3808                         continue;
3809                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3810                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3811                         continue;
3812
3813                 ret = wait_dev_supers(dev, max_mirrors);
3814                 if (ret)
3815                         total_errors++;
3816         }
3817         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3818         if (total_errors > max_errors) {
3819                 btrfs_handle_fs_error(fs_info, -EIO,
3820                                       "%d errors while writing supers",
3821                                       total_errors);
3822                 return -EIO;
3823         }
3824         return 0;
3825 }
3826
3827 /* Drop a fs root from the radix tree and free it. */
3828 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3829                                   struct btrfs_root *root)
3830 {
3831         spin_lock(&fs_info->fs_roots_radix_lock);
3832         radix_tree_delete(&fs_info->fs_roots_radix,
3833                           (unsigned long)root->root_key.objectid);
3834         spin_unlock(&fs_info->fs_roots_radix_lock);
3835
3836         if (btrfs_root_refs(&root->root_item) == 0)
3837                 synchronize_srcu(&fs_info->subvol_srcu);
3838
3839         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3840                 btrfs_free_log(NULL, root);
3841                 if (root->reloc_root) {
3842                         free_extent_buffer(root->reloc_root->node);
3843                         free_extent_buffer(root->reloc_root->commit_root);
3844                         btrfs_put_fs_root(root->reloc_root);
3845                         root->reloc_root = NULL;
3846                 }
3847         }
3848
3849         if (root->free_ino_pinned)
3850                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3851         if (root->free_ino_ctl)
3852                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3853         btrfs_free_fs_root(root);
3854 }
3855
3856 void btrfs_free_fs_root(struct btrfs_root *root)
3857 {
3858         iput(root->ino_cache_inode);
3859         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3860         if (root->anon_dev)
3861                 free_anon_bdev(root->anon_dev);
3862         if (root->subv_writers)
3863                 btrfs_free_subvolume_writers(root->subv_writers);
3864         free_extent_buffer(root->node);
3865         free_extent_buffer(root->commit_root);
3866         kfree(root->free_ino_ctl);
3867         kfree(root->free_ino_pinned);
3868         btrfs_put_fs_root(root);
3869 }
3870
3871 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3872 {
3873         u64 root_objectid = 0;
3874         struct btrfs_root *gang[8];
3875         int i = 0;
3876         int err = 0;
3877         unsigned int ret = 0;
3878         int index;
3879
3880         while (1) {
3881                 index = srcu_read_lock(&fs_info->subvol_srcu);
3882                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3883                                              (void **)gang, root_objectid,
3884                                              ARRAY_SIZE(gang));
3885                 if (!ret) {
3886                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3887                         break;
3888                 }
3889                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3890
3891                 for (i = 0; i < ret; i++) {
3892                         /* Avoid to grab roots in dead_roots */
3893                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3894                                 gang[i] = NULL;
3895                                 continue;
3896                         }
3897                         /* grab all the search result for later use */
3898                         gang[i] = btrfs_grab_fs_root(gang[i]);
3899                 }
3900                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3901
3902                 for (i = 0; i < ret; i++) {
3903                         if (!gang[i])
3904                                 continue;
3905                         root_objectid = gang[i]->root_key.objectid;
3906                         err = btrfs_orphan_cleanup(gang[i]);
3907                         if (err)
3908                                 break;
3909                         btrfs_put_fs_root(gang[i]);
3910                 }
3911                 root_objectid++;
3912         }
3913
3914         /* release the uncleaned roots due to error */
3915         for (; i < ret; i++) {
3916                 if (gang[i])
3917                         btrfs_put_fs_root(gang[i]);
3918         }
3919         return err;
3920 }
3921
3922 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3923 {
3924         struct btrfs_root *root = fs_info->tree_root;
3925         struct btrfs_trans_handle *trans;
3926
3927         mutex_lock(&fs_info->cleaner_mutex);
3928         btrfs_run_delayed_iputs(fs_info);
3929         mutex_unlock(&fs_info->cleaner_mutex);
3930         wake_up_process(fs_info->cleaner_kthread);
3931
3932         /* wait until ongoing cleanup work done */
3933         down_write(&fs_info->cleanup_work_sem);
3934         up_write(&fs_info->cleanup_work_sem);
3935
3936         trans = btrfs_join_transaction(root);
3937         if (IS_ERR(trans))
3938                 return PTR_ERR(trans);
3939         return btrfs_commit_transaction(trans);
3940 }
3941
3942 void close_ctree(struct btrfs_fs_info *fs_info)
3943 {
3944         int ret;
3945
3946         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3947         /*
3948          * We don't want the cleaner to start new transactions, add more delayed
3949          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3950          * because that frees the task_struct, and the transaction kthread might
3951          * still try to wake up the cleaner.
3952          */
3953         kthread_park(fs_info->cleaner_kthread);
3954
3955         /* wait for the qgroup rescan worker to stop */
3956         btrfs_qgroup_wait_for_completion(fs_info, false);
3957
3958         /* wait for the uuid_scan task to finish */
3959         down(&fs_info->uuid_tree_rescan_sem);
3960         /* avoid complains from lockdep et al., set sem back to initial state */
3961         up(&fs_info->uuid_tree_rescan_sem);
3962
3963         /* pause restriper - we want to resume on mount */
3964         btrfs_pause_balance(fs_info);
3965
3966         btrfs_dev_replace_suspend_for_unmount(fs_info);
3967
3968         btrfs_scrub_cancel(fs_info);
3969
3970         /* wait for any defraggers to finish */
3971         wait_event(fs_info->transaction_wait,
3972                    (atomic_read(&fs_info->defrag_running) == 0));
3973
3974         /* clear out the rbtree of defraggable inodes */
3975         btrfs_cleanup_defrag_inodes(fs_info);
3976
3977         cancel_work_sync(&fs_info->async_reclaim_work);
3978
3979         if (!sb_rdonly(fs_info->sb)) {
3980                 /*
3981                  * The cleaner kthread is stopped, so do one final pass over
3982                  * unused block groups.
3983                  */
3984                 btrfs_delete_unused_bgs(fs_info);
3985
3986                 ret = btrfs_commit_super(fs_info);
3987                 if (ret)
3988                         btrfs_err(fs_info, "commit super ret %d", ret);
3989         }
3990
3991         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3992             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3993                 btrfs_error_commit_super(fs_info);
3994
3995         kthread_stop(fs_info->transaction_kthread);
3996         kthread_stop(fs_info->cleaner_kthread);
3997
3998         ASSERT(list_empty(&fs_info->delayed_iputs));
3999         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4000
4001         btrfs_free_qgroup_config(fs_info);
4002         ASSERT(list_empty(&fs_info->delalloc_roots));
4003
4004         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4005                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4006                        percpu_counter_sum(&fs_info->delalloc_bytes));
4007         }
4008
4009         btrfs_sysfs_remove_mounted(fs_info);
4010         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4011
4012         btrfs_free_fs_roots(fs_info);
4013
4014         btrfs_put_block_group_cache(fs_info);
4015
4016         /*
4017          * we must make sure there is not any read request to
4018          * submit after we stopping all workers.
4019          */
4020         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4021         btrfs_stop_all_workers(fs_info);
4022
4023         btrfs_free_block_groups(fs_info);
4024
4025         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4026         free_root_pointers(fs_info, 1);
4027
4028         iput(fs_info->btree_inode);
4029
4030 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4031         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4032                 btrfsic_unmount(fs_info->fs_devices);
4033 #endif
4034
4035         btrfs_close_devices(fs_info->fs_devices);
4036         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4037
4038         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4039         percpu_counter_destroy(&fs_info->delalloc_bytes);
4040         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4041         cleanup_srcu_struct(&fs_info->subvol_srcu);
4042
4043         btrfs_free_stripe_hash_table(fs_info);
4044         btrfs_free_ref_cache(fs_info);
4045
4046         while (!list_empty(&fs_info->pinned_chunks)) {
4047                 struct extent_map *em;
4048
4049                 em = list_first_entry(&fs_info->pinned_chunks,
4050                                       struct extent_map, list);
4051                 list_del_init(&em->list);
4052                 free_extent_map(em);
4053         }
4054 }
4055
4056 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4057                           int atomic)
4058 {
4059         int ret;
4060         struct inode *btree_inode = buf->pages[0]->mapping->host;
4061
4062         ret = extent_buffer_uptodate(buf);
4063         if (!ret)
4064                 return ret;
4065
4066         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4067                                     parent_transid, atomic);
4068         if (ret == -EAGAIN)
4069                 return ret;
4070         return !ret;
4071 }
4072
4073 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4074 {
4075         struct btrfs_fs_info *fs_info;
4076         struct btrfs_root *root;
4077         u64 transid = btrfs_header_generation(buf);
4078         int was_dirty;
4079
4080 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4081         /*
4082          * This is a fast path so only do this check if we have sanity tests
4083          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4084          * outside of the sanity tests.
4085          */
4086         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4087                 return;
4088 #endif
4089         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4090         fs_info = root->fs_info;
4091         btrfs_assert_tree_locked(buf);
4092         if (transid != fs_info->generation)
4093                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4094                         buf->start, transid, fs_info->generation);
4095         was_dirty = set_extent_buffer_dirty(buf);
4096         if (!was_dirty)
4097                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4098                                          buf->len,
4099                                          fs_info->dirty_metadata_batch);
4100 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4101         /*
4102          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4103          * but item data not updated.
4104          * So here we should only check item pointers, not item data.
4105          */
4106         if (btrfs_header_level(buf) == 0 &&
4107             btrfs_check_leaf_relaxed(fs_info, buf)) {
4108                 btrfs_print_leaf(buf);
4109                 ASSERT(0);
4110         }
4111 #endif
4112 }
4113
4114 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4115                                         int flush_delayed)
4116 {
4117         /*
4118          * looks as though older kernels can get into trouble with
4119          * this code, they end up stuck in balance_dirty_pages forever
4120          */
4121         int ret;
4122
4123         if (current->flags & PF_MEMALLOC)
4124                 return;
4125
4126         if (flush_delayed)
4127                 btrfs_balance_delayed_items(fs_info);
4128
4129         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4130                                      BTRFS_DIRTY_METADATA_THRESH,
4131                                      fs_info->dirty_metadata_batch);
4132         if (ret > 0) {
4133                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4134         }
4135 }
4136
4137 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4138 {
4139         __btrfs_btree_balance_dirty(fs_info, 1);
4140 }
4141
4142 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4143 {
4144         __btrfs_btree_balance_dirty(fs_info, 0);
4145 }
4146
4147 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4148                       struct btrfs_key *first_key)
4149 {
4150         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4151         struct btrfs_fs_info *fs_info = root->fs_info;
4152
4153         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4154                                               level, first_key);
4155 }
4156
4157 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4158 {
4159         /* cleanup FS via transaction */
4160         btrfs_cleanup_transaction(fs_info);
4161
4162         mutex_lock(&fs_info->cleaner_mutex);
4163         btrfs_run_delayed_iputs(fs_info);
4164         mutex_unlock(&fs_info->cleaner_mutex);
4165
4166         down_write(&fs_info->cleanup_work_sem);
4167         up_write(&fs_info->cleanup_work_sem);
4168 }
4169
4170 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4171 {
4172         struct btrfs_ordered_extent *ordered;
4173
4174         spin_lock(&root->ordered_extent_lock);
4175         /*
4176          * This will just short circuit the ordered completion stuff which will
4177          * make sure the ordered extent gets properly cleaned up.
4178          */
4179         list_for_each_entry(ordered, &root->ordered_extents,
4180                             root_extent_list)
4181                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4182         spin_unlock(&root->ordered_extent_lock);
4183 }
4184
4185 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4186 {
4187         struct btrfs_root *root;
4188         struct list_head splice;
4189
4190         INIT_LIST_HEAD(&splice);
4191
4192         spin_lock(&fs_info->ordered_root_lock);
4193         list_splice_init(&fs_info->ordered_roots, &splice);
4194         while (!list_empty(&splice)) {
4195                 root = list_first_entry(&splice, struct btrfs_root,
4196                                         ordered_root);
4197                 list_move_tail(&root->ordered_root,
4198                                &fs_info->ordered_roots);
4199
4200                 spin_unlock(&fs_info->ordered_root_lock);
4201                 btrfs_destroy_ordered_extents(root);
4202
4203                 cond_resched();
4204                 spin_lock(&fs_info->ordered_root_lock);
4205         }
4206         spin_unlock(&fs_info->ordered_root_lock);
4207
4208         /*
4209          * We need this here because if we've been flipped read-only we won't
4210          * get sync() from the umount, so we need to make sure any ordered
4211          * extents that haven't had their dirty pages IO start writeout yet
4212          * actually get run and error out properly.
4213          */
4214         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4215 }
4216
4217 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4218                                       struct btrfs_fs_info *fs_info)
4219 {
4220         struct rb_node *node;
4221         struct btrfs_delayed_ref_root *delayed_refs;
4222         struct btrfs_delayed_ref_node *ref;
4223         int ret = 0;
4224
4225         delayed_refs = &trans->delayed_refs;
4226
4227         spin_lock(&delayed_refs->lock);
4228         if (atomic_read(&delayed_refs->num_entries) == 0) {
4229                 spin_unlock(&delayed_refs->lock);
4230                 btrfs_info(fs_info, "delayed_refs has NO entry");
4231                 return ret;
4232         }
4233
4234         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4235                 struct btrfs_delayed_ref_head *head;
4236                 struct rb_node *n;
4237                 bool pin_bytes = false;
4238
4239                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4240                                 href_node);
4241                 if (!mutex_trylock(&head->mutex)) {
4242                         refcount_inc(&head->refs);
4243                         spin_unlock(&delayed_refs->lock);
4244
4245                         mutex_lock(&head->mutex);
4246                         mutex_unlock(&head->mutex);
4247                         btrfs_put_delayed_ref_head(head);
4248                         spin_lock(&delayed_refs->lock);
4249                         continue;
4250                 }
4251                 spin_lock(&head->lock);
4252                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4253                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4254                                        ref_node);
4255                         ref->in_tree = 0;
4256                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4257                         RB_CLEAR_NODE(&ref->ref_node);
4258                         if (!list_empty(&ref->add_list))
4259                                 list_del(&ref->add_list);
4260                         atomic_dec(&delayed_refs->num_entries);
4261                         btrfs_put_delayed_ref(ref);
4262                 }
4263                 if (head->must_insert_reserved)
4264                         pin_bytes = true;
4265                 btrfs_free_delayed_extent_op(head->extent_op);
4266                 delayed_refs->num_heads--;
4267                 if (head->processing == 0)
4268                         delayed_refs->num_heads_ready--;
4269                 atomic_dec(&delayed_refs->num_entries);
4270                 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
4271                 RB_CLEAR_NODE(&head->href_node);
4272                 spin_unlock(&head->lock);
4273                 spin_unlock(&delayed_refs->lock);
4274                 mutex_unlock(&head->mutex);
4275
4276                 if (pin_bytes)
4277                         btrfs_pin_extent(fs_info, head->bytenr,
4278                                          head->num_bytes, 1);
4279                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4280                 btrfs_put_delayed_ref_head(head);
4281                 cond_resched();
4282                 spin_lock(&delayed_refs->lock);
4283         }
4284
4285         spin_unlock(&delayed_refs->lock);
4286
4287         return ret;
4288 }
4289
4290 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4291 {
4292         struct btrfs_inode *btrfs_inode;
4293         struct list_head splice;
4294
4295         INIT_LIST_HEAD(&splice);
4296
4297         spin_lock(&root->delalloc_lock);
4298         list_splice_init(&root->delalloc_inodes, &splice);
4299
4300         while (!list_empty(&splice)) {
4301                 struct inode *inode = NULL;
4302                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4303                                                delalloc_inodes);
4304                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4305                 spin_unlock(&root->delalloc_lock);
4306
4307                 /*
4308                  * Make sure we get a live inode and that it'll not disappear
4309                  * meanwhile.
4310                  */
4311                 inode = igrab(&btrfs_inode->vfs_inode);
4312                 if (inode) {
4313                         invalidate_inode_pages2(inode->i_mapping);
4314                         iput(inode);
4315                 }
4316                 spin_lock(&root->delalloc_lock);
4317         }
4318         spin_unlock(&root->delalloc_lock);
4319 }
4320
4321 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4322 {
4323         struct btrfs_root *root;
4324         struct list_head splice;
4325
4326         INIT_LIST_HEAD(&splice);
4327
4328         spin_lock(&fs_info->delalloc_root_lock);
4329         list_splice_init(&fs_info->delalloc_roots, &splice);
4330         while (!list_empty(&splice)) {
4331                 root = list_first_entry(&splice, struct btrfs_root,
4332                                          delalloc_root);
4333                 root = btrfs_grab_fs_root(root);
4334                 BUG_ON(!root);
4335                 spin_unlock(&fs_info->delalloc_root_lock);
4336
4337                 btrfs_destroy_delalloc_inodes(root);
4338                 btrfs_put_fs_root(root);
4339
4340                 spin_lock(&fs_info->delalloc_root_lock);
4341         }
4342         spin_unlock(&fs_info->delalloc_root_lock);
4343 }
4344
4345 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4346                                         struct extent_io_tree *dirty_pages,
4347                                         int mark)
4348 {
4349         int ret;
4350         struct extent_buffer *eb;
4351         u64 start = 0;
4352         u64 end;
4353
4354         while (1) {
4355                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4356                                             mark, NULL);
4357                 if (ret)
4358                         break;
4359
4360                 clear_extent_bits(dirty_pages, start, end, mark);
4361                 while (start <= end) {
4362                         eb = find_extent_buffer(fs_info, start);
4363                         start += fs_info->nodesize;
4364                         if (!eb)
4365                                 continue;
4366                         wait_on_extent_buffer_writeback(eb);
4367
4368                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4369                                                &eb->bflags))
4370                                 clear_extent_buffer_dirty(eb);
4371                         free_extent_buffer_stale(eb);
4372                 }
4373         }
4374
4375         return ret;
4376 }
4377
4378 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4379                                        struct extent_io_tree *pinned_extents)
4380 {
4381         struct extent_io_tree *unpin;
4382         u64 start;
4383         u64 end;
4384         int ret;
4385         bool loop = true;
4386
4387         unpin = pinned_extents;
4388 again:
4389         while (1) {
4390                 struct extent_state *cached_state = NULL;
4391
4392                 /*
4393                  * The btrfs_finish_extent_commit() may get the same range as
4394                  * ours between find_first_extent_bit and clear_extent_dirty.
4395                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4396                  * the same extent range.
4397                  */
4398                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4399                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4400                                             EXTENT_DIRTY, &cached_state);
4401                 if (ret) {
4402                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4403                         break;
4404                 }
4405
4406                 clear_extent_dirty(unpin, start, end, &cached_state);
4407                 free_extent_state(cached_state);
4408                 btrfs_error_unpin_extent_range(fs_info, start, end);
4409                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4410                 cond_resched();
4411         }
4412
4413         if (loop) {
4414                 if (unpin == &fs_info->freed_extents[0])
4415                         unpin = &fs_info->freed_extents[1];
4416                 else
4417                         unpin = &fs_info->freed_extents[0];
4418                 loop = false;
4419                 goto again;
4420         }
4421
4422         return 0;
4423 }
4424
4425 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4426 {
4427         struct inode *inode;
4428
4429         inode = cache->io_ctl.inode;
4430         if (inode) {
4431                 invalidate_inode_pages2(inode->i_mapping);
4432                 BTRFS_I(inode)->generation = 0;
4433                 cache->io_ctl.inode = NULL;
4434                 iput(inode);
4435         }
4436         btrfs_put_block_group(cache);
4437 }
4438
4439 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4440                              struct btrfs_fs_info *fs_info)
4441 {
4442         struct btrfs_block_group_cache *cache;
4443
4444         spin_lock(&cur_trans->dirty_bgs_lock);
4445         while (!list_empty(&cur_trans->dirty_bgs)) {
4446                 cache = list_first_entry(&cur_trans->dirty_bgs,
4447                                          struct btrfs_block_group_cache,
4448                                          dirty_list);
4449
4450                 if (!list_empty(&cache->io_list)) {
4451                         spin_unlock(&cur_trans->dirty_bgs_lock);
4452                         list_del_init(&cache->io_list);
4453                         btrfs_cleanup_bg_io(cache);
4454                         spin_lock(&cur_trans->dirty_bgs_lock);
4455                 }
4456
4457                 list_del_init(&cache->dirty_list);
4458                 spin_lock(&cache->lock);
4459                 cache->disk_cache_state = BTRFS_DC_ERROR;
4460                 spin_unlock(&cache->lock);
4461
4462                 spin_unlock(&cur_trans->dirty_bgs_lock);
4463                 btrfs_put_block_group(cache);
4464                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4465                 spin_lock(&cur_trans->dirty_bgs_lock);
4466         }
4467         spin_unlock(&cur_trans->dirty_bgs_lock);
4468
4469         /*
4470          * Refer to the definition of io_bgs member for details why it's safe
4471          * to use it without any locking
4472          */
4473         while (!list_empty(&cur_trans->io_bgs)) {
4474                 cache = list_first_entry(&cur_trans->io_bgs,
4475                                          struct btrfs_block_group_cache,
4476                                          io_list);
4477
4478                 list_del_init(&cache->io_list);
4479                 spin_lock(&cache->lock);
4480                 cache->disk_cache_state = BTRFS_DC_ERROR;
4481                 spin_unlock(&cache->lock);
4482                 btrfs_cleanup_bg_io(cache);
4483         }
4484 }
4485
4486 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4487                                    struct btrfs_fs_info *fs_info)
4488 {
4489         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4490         ASSERT(list_empty(&cur_trans->dirty_bgs));
4491         ASSERT(list_empty(&cur_trans->io_bgs));
4492
4493         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4494
4495         cur_trans->state = TRANS_STATE_COMMIT_START;
4496         wake_up(&fs_info->transaction_blocked_wait);
4497
4498         cur_trans->state = TRANS_STATE_UNBLOCKED;
4499         wake_up(&fs_info->transaction_wait);
4500
4501         btrfs_destroy_delayed_inodes(fs_info);
4502         btrfs_assert_delayed_root_empty(fs_info);
4503
4504         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4505                                      EXTENT_DIRTY);
4506         btrfs_destroy_pinned_extent(fs_info,
4507                                     fs_info->pinned_extents);
4508
4509         cur_trans->state =TRANS_STATE_COMPLETED;
4510         wake_up(&cur_trans->commit_wait);
4511 }
4512
4513 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4514 {
4515         struct btrfs_transaction *t;
4516
4517         mutex_lock(&fs_info->transaction_kthread_mutex);
4518
4519         spin_lock(&fs_info->trans_lock);
4520         while (!list_empty(&fs_info->trans_list)) {
4521                 t = list_first_entry(&fs_info->trans_list,
4522                                      struct btrfs_transaction, list);
4523                 if (t->state >= TRANS_STATE_COMMIT_START) {
4524                         refcount_inc(&t->use_count);
4525                         spin_unlock(&fs_info->trans_lock);
4526                         btrfs_wait_for_commit(fs_info, t->transid);
4527                         btrfs_put_transaction(t);
4528                         spin_lock(&fs_info->trans_lock);
4529                         continue;
4530                 }
4531                 if (t == fs_info->running_transaction) {
4532                         t->state = TRANS_STATE_COMMIT_DOING;
4533                         spin_unlock(&fs_info->trans_lock);
4534                         /*
4535                          * We wait for 0 num_writers since we don't hold a trans
4536                          * handle open currently for this transaction.
4537                          */
4538                         wait_event(t->writer_wait,
4539                                    atomic_read(&t->num_writers) == 0);
4540                 } else {
4541                         spin_unlock(&fs_info->trans_lock);
4542                 }
4543                 btrfs_cleanup_one_transaction(t, fs_info);
4544
4545                 spin_lock(&fs_info->trans_lock);
4546                 if (t == fs_info->running_transaction)
4547                         fs_info->running_transaction = NULL;
4548                 list_del_init(&t->list);
4549                 spin_unlock(&fs_info->trans_lock);
4550
4551                 btrfs_put_transaction(t);
4552                 trace_btrfs_transaction_commit(fs_info->tree_root);
4553                 spin_lock(&fs_info->trans_lock);
4554         }
4555         spin_unlock(&fs_info->trans_lock);
4556         btrfs_destroy_all_ordered_extents(fs_info);
4557         btrfs_destroy_delayed_inodes(fs_info);
4558         btrfs_assert_delayed_root_empty(fs_info);
4559         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4560         btrfs_destroy_all_delalloc_inodes(fs_info);
4561         mutex_unlock(&fs_info->transaction_kthread_mutex);
4562
4563         return 0;
4564 }
4565
4566 static const struct extent_io_ops btree_extent_io_ops = {
4567         /* mandatory callbacks */
4568         .submit_bio_hook = btree_submit_bio_hook,
4569         .readpage_end_io_hook = btree_readpage_end_io_hook,
4570 };