fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/scatterlist.h>
   9 #include <linux/swap.h>
  10 #include <linux/radix-tree.h>
  11 #include <linux/writeback.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/workqueue.h>
  14 #include <linux/kthread.h>
  15 #include <linux/slab.h>
  16 #include <linux/migrate.h>
  17 #include <linux/ratelimit.h>
  18 #include <linux/uuid.h>
  19 #include <linux/semaphore.h>
  20 #include <linux/error-injection.h>
  21 #include <linux/crc32c.h>
  22 #include <asm/unaligned.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "btrfs_inode.h"
  27 #include "volumes.h"
  28 #include "print-tree.h"
  29 #include "locking.h"
  30 #include "tree-log.h"
  31 #include "free-space-cache.h"
  32 #include "free-space-tree.h"
  33 #include "inode-map.h"
  34 #include "check-integrity.h"
  35 #include "rcu-string.h"
  36 #include "dev-replace.h"
  37 #include "raid56.h"
  38 #include "sysfs.h"
  39 #include "qgroup.h"
  40 #include "compression.h"
  41 #include "tree-checker.h"
  42 #include "ref-verify.h"
  43
  44 #ifdef CONFIG_X86
  45 #include <asm/cpufeature.h>
  46 #endif
  47
  48 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  49                                  BTRFS_HEADER_FLAG_RELOC |\
  50                                  BTRFS_SUPER_FLAG_ERROR |\
  51                                  BTRFS_SUPER_FLAG_SEEDING |\
  52                                  BTRFS_SUPER_FLAG_METADUMP |\
  53                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  54
  55 static const struct extent_io_ops btree_extent_io_ops;
  56 static void end_workqueue_fn(struct btrfs_work *work);
  57 static void free_fs_root(struct btrfs_root *root);
  58 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  59 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  60                                       struct btrfs_fs_info *fs_info);
  61 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  62 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  63                                         struct extent_io_tree *dirty_pages,
  64                                         int mark);
  65 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  66                                        struct extent_io_tree *pinned_extents);
  67 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  68 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  69
  70 /*
  71  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  72  * is complete.  This is used during reads to verify checksums, and it is used
  73  * by writes to insert metadata for new file extents after IO is complete.
  74  */
  75 struct btrfs_end_io_wq {
  76         struct bio *bio;
  77         bio_end_io_t *end_io;
  78         void *private;
  79         struct btrfs_fs_info *info;
  80         blk_status_t status;
  81         enum btrfs_wq_endio_type metadata;
  82         struct btrfs_work work;
  83 };
  84
  85 static struct kmem_cache *btrfs_end_io_wq_cache;
  86
  87 int __init btrfs_end_io_wq_init(void)
  88 {
  89         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  90                                         sizeof(struct btrfs_end_io_wq),
  91                                         0,
  92                                         SLAB_MEM_SPREAD,
  93                                         NULL);
  94         if (!btrfs_end_io_wq_cache)
  95                 return -ENOMEM;
  96         return 0;
  97 }
  98
  99 void __cold btrfs_end_io_wq_exit(void)
 100 {
 101         kmem_cache_destroy(btrfs_end_io_wq_cache);
 102 }
 103
 104 /*
 105  * async submit bios are used to offload expensive checksumming
 106  * onto the worker threads.  They checksum file and metadata bios
 107  * just before they are sent down the IO stack.
 108  */
 109 struct async_submit_bio {
 110         void *private_data;
 111         struct btrfs_fs_info *fs_info;
 112         struct bio *bio;
 113         extent_submit_bio_start_t *submit_bio_start;
 114         extent_submit_bio_done_t *submit_bio_done;
 115         int mirror_num;
 116         unsigned long bio_flags;
 117         /*
 118          * bio_offset is optional, can be used if the pages in the bio
 119          * can't tell us where in the file the bio should go
 120          */
 121         u64 bio_offset;
 122         struct btrfs_work work;
 123         blk_status_t status;
 124 };
 125
 126 /*
 127  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 128  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 129  * the level the eb occupies in the tree.
 130  *
 131  * Different roots are used for different purposes and may nest inside each
 132  * other and they require separate keysets.  As lockdep keys should be
 133  * static, assign keysets according to the purpose of the root as indicated
 134  * by btrfs_root->objectid.  This ensures that all special purpose roots
 135  * have separate keysets.
 136  *
 137  * Lock-nesting across peer nodes is always done with the immediate parent
 138  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 139  * subclass to avoid triggering lockdep warning in such cases.
 140  *
 141  * The key is set by the readpage_end_io_hook after the buffer has passed
 142  * csum validation but before the pages are unlocked.  It is also set by
 143  * btrfs_init_new_buffer on freshly allocated blocks.
 144  *
 145  * We also add a check to make sure the highest level of the tree is the
 146  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 147  * needs update as well.
 148  */
 149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 150 # if BTRFS_MAX_LEVEL != 8
 151 #  error
 152 # endif
 153
 154 static struct btrfs_lockdep_keyset {
 155         u64                     id;             /* root objectid */
 156         const char              *name_stem;     /* lock name stem */
 157         char                    names[BTRFS_MAX_LEVEL + 1][20];
 158         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 159 } btrfs_lockdep_keysets[] = {
 160         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 161         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 162         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 163         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 164         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 165         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 166         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 167         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 168         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 169         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 170         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 171         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 172         { .id = 0,                              .name_stem = "tree"     },
 173 };
 174
 175 void __init btrfs_init_lockdep(void)
 176 {
 177         int i, j;
 178
 179         /* initialize lockdep class names */
 180         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 181                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 182
 183                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 184                         snprintf(ks->names[j], sizeof(ks->names[j]),
 185                                  "btrfs-%s-%02d", ks->name_stem, j);
 186         }
 187 }
 188
 189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 190                                     int level)
 191 {
 192         struct btrfs_lockdep_keyset *ks;
 193
 194         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 195
 196         /* find the matching keyset, id 0 is the default entry */
 197         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 198                 if (ks->id == objectid)
 199                         break;
 200
 201         lockdep_set_class_and_name(&eb->lock,
 202                                    &ks->keys[level], ks->names[level]);
 203 }
 204
 205 #endif
 206
 207 /*
 208  * extents on the btree inode are pretty simple, there's one extent
 209  * that covers the entire device
 210  */
 211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 212                 struct page *page, size_t pg_offset, u64 start, u64 len,
 213                 int create)
 214 {
 215         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 216         struct extent_map_tree *em_tree = &inode->extent_tree;
 217         struct extent_map *em;
 218         int ret;
 219
 220         read_lock(&em_tree->lock);
 221         em = lookup_extent_mapping(em_tree, start, len);
 222         if (em) {
 223                 em->bdev = fs_info->fs_devices->latest_bdev;
 224                 read_unlock(&em_tree->lock);
 225                 goto out;
 226         }
 227         read_unlock(&em_tree->lock);
 228
 229         em = alloc_extent_map();
 230         if (!em) {
 231                 em = ERR_PTR(-ENOMEM);
 232                 goto out;
 233         }
 234         em->start = 0;
 235         em->len = (u64)-1;
 236         em->block_len = (u64)-1;
 237         em->block_start = 0;
 238         em->bdev = fs_info->fs_devices->latest_bdev;
 239
 240         write_lock(&em_tree->lock);
 241         ret = add_extent_mapping(em_tree, em, 0);
 242         if (ret == -EEXIST) {
 243                 free_extent_map(em);
 244                 em = lookup_extent_mapping(em_tree, start, len);
 245                 if (!em)
 246                         em = ERR_PTR(-EIO);
 247         } else if (ret) {
 248                 free_extent_map(em);
 249                 em = ERR_PTR(ret);
 250         }
 251         write_unlock(&em_tree->lock);
 252
 253 out:
 254         return em;
 255 }
 256
 257 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 258 {
 259         return crc32c(seed, data, len);
 260 }
 261
 262 void btrfs_csum_final(u32 crc, u8 *result)
 263 {
 264         put_unaligned_le32(~crc, result);
 265 }
 266
 267 /*
 268  * compute the csum for a btree block, and either verify it or write it
 269  * into the csum field of the block.
 270  */
 271 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 272                            struct extent_buffer *buf,
 273                            int verify)
 274 {
 275         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 276         char result[BTRFS_CSUM_SIZE];
 277         unsigned long len;
 278         unsigned long cur_len;
 279         unsigned long offset = BTRFS_CSUM_SIZE;
 280         char *kaddr;
 281         unsigned long map_start;
 282         unsigned long map_len;
 283         int err;
 284         u32 crc = ~(u32)0;
 285
 286         len = buf->len - offset;
 287         while (len > 0) {
 288                 err = map_private_extent_buffer(buf, offset, 32,
 289                                         &kaddr, &map_start, &map_len);
 290                 if (err)
 291                         return err;
 292                 cur_len = min(len, map_len - (offset - map_start));
 293                 crc = btrfs_csum_data(kaddr + offset - map_start,
 294                                       crc, cur_len);
 295                 len -= cur_len;
 296                 offset += cur_len;
 297         }
 298         memset(result, 0, BTRFS_CSUM_SIZE);
 299
 300         btrfs_csum_final(crc, result);
 301
 302         if (verify) {
 303                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 304                         u32 val;
 305                         u32 found = 0;
 306                         memcpy(&found, result, csum_size);
 307
 308                         read_extent_buffer(buf, &val, 0, csum_size);
 309                         btrfs_warn_rl(fs_info,
 310                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 311                                 fs_info->sb->s_id, buf->start,
 312                                 val, found, btrfs_header_level(buf));
 313                         return -EUCLEAN;
 314                 }
 315         } else {
 316                 write_extent_buffer(buf, result, 0, csum_size);
 317         }
 318
 319         return 0;
 320 }
 321
 322 /*
 323  * we can't consider a given block up to date unless the transid of the
 324  * block matches the transid in the parent node's pointer.  This is how we
 325  * detect blocks that either didn't get written at all or got written
 326  * in the wrong place.
 327  */
 328 static int verify_parent_transid(struct extent_io_tree *io_tree,
 329                                  struct extent_buffer *eb, u64 parent_transid,
 330                                  int atomic)
 331 {
 332         struct extent_state *cached_state = NULL;
 333         int ret;
 334         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 335
 336         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 337                 return 0;
 338
 339         if (atomic)
 340                 return -EAGAIN;
 341
 342         if (need_lock) {
 343                 btrfs_tree_read_lock(eb);
 344                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 345         }
 346
 347         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 348                          &cached_state);
 349         if (extent_buffer_uptodate(eb) &&
 350             btrfs_header_generation(eb) == parent_transid) {
 351                 ret = 0;
 352                 goto out;
 353         }
 354         btrfs_err_rl(eb->fs_info,
 355                 "parent transid verify failed on %llu wanted %llu found %llu",
 356                         eb->start,
 357                         parent_transid, btrfs_header_generation(eb));
 358         ret = 1;
 359
 360         /*
 361          * Things reading via commit roots that don't have normal protection,
 362          * like send, can have a really old block in cache that may point at a
 363          * block that has been freed and re-allocated.  So don't clear uptodate
 364          * if we find an eb that is under IO (dirty/writeback) because we could
 365          * end up reading in the stale data and then writing it back out and
 366          * making everybody very sad.
 367          */
 368         if (!extent_buffer_under_io(eb))
 369                 clear_extent_buffer_uptodate(eb);
 370 out:
 371         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 372                              &cached_state);
 373         if (need_lock)
 374                 btrfs_tree_read_unlock_blocking(eb);
 375         return ret;
 376 }
 377
 378 /*
 379  * Return 0 if the superblock checksum type matches the checksum value of that
 380  * algorithm. Pass the raw disk superblock data.
 381  */
 382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 383                                   char *raw_disk_sb)
 384 {
 385         struct btrfs_super_block *disk_sb =
 386                 (struct btrfs_super_block *)raw_disk_sb;
 387         u16 csum_type = btrfs_super_csum_type(disk_sb);
 388         int ret = 0;
 389
 390         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 391                 u32 crc = ~(u32)0;
 392                 char result[sizeof(crc)];
 393
 394                 /*
 395                  * The super_block structure does not span the whole
 396                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 397                  * is filled with zeros and is included in the checksum.
 398                  */
 399                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 400                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 401                 btrfs_csum_final(crc, result);
 402
 403                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 404                         ret = 1;
 405         }
 406
 407         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 408                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 409                                 csum_type);
 410                 ret = 1;
 411         }
 412
 413         return ret;
 414 }
 415
 416 static int verify_level_key(struct btrfs_fs_info *fs_info,
 417                             struct extent_buffer *eb, int level,
 418                             struct btrfs_key *first_key, 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         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 487         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 488         while (1) {
 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                 /*
 503                  * This buffer's crc is fine, but its contents are corrupted, so
 504                  * there is no reason to read the other copies, they won't be
 505                  * any less wrong.
 506                  */
 507                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
 508                     ret == -EUCLEAN)
 509                         break;
 510
 511                 num_copies = btrfs_num_copies(fs_info,
 512                                               eb->start, eb->len);
 513                 if (num_copies == 1)
 514                         break;
 515
 516                 if (!failed_mirror) {
 517                         failed = 1;
 518                         failed_mirror = eb->read_mirror;
 519                 }
 520
 521                 mirror_num++;
 522                 if (mirror_num == failed_mirror)
 523                         mirror_num++;
 524
 525                 if (mirror_num > num_copies)
 526                         break;
 527         }
 528
 529         if (failed && !ret && failed_mirror)
 530                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 531
 532         return ret;
 533 }
 534
 535 /*
 536  * checksum a dirty tree block before IO.  This has extra checks to make sure
 537  * we only fill in the checksum field in the first page of a multi-page block
 538  */
 539
 540 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 541 {
 542         u64 start = page_offset(page);
 543         u64 found_start;
 544         struct extent_buffer *eb;
 545
 546         eb = (struct extent_buffer *)page->private;
 547         if (page != eb->pages[0])
 548                 return 0;
 549
 550         found_start = btrfs_header_bytenr(eb);
 551         /*
 552          * Please do not consolidate these warnings into a single if.
 553          * It is useful to know what went wrong.
 554          */
 555         if (WARN_ON(found_start != start))
 556                 return -EUCLEAN;
 557         if (WARN_ON(!PageUptodate(page)))
 558                 return -EUCLEAN;
 559
 560         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 561                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 562
 563         return csum_tree_block(fs_info, eb, 0);
 564 }
 565
 566 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 567                                  struct extent_buffer *eb)
 568 {
 569         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 570         u8 fsid[BTRFS_FSID_SIZE];
 571         int ret = 1;
 572
 573         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 574         while (fs_devices) {
 575                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 576                         ret = 0;
 577                         break;
 578                 }
 579                 fs_devices = fs_devices->seed;
 580         }
 581         return ret;
 582 }
 583
 584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 585                                       u64 phy_offset, struct page *page,
 586                                       u64 start, u64 end, int mirror)
 587 {
 588         u64 found_start;
 589         int found_level;
 590         struct extent_buffer *eb;
 591         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 592         struct btrfs_fs_info *fs_info = root->fs_info;
 593         int ret = 0;
 594         int reads_done;
 595
 596         if (!page->private)
 597                 goto out;
 598
 599         eb = (struct extent_buffer *)page->private;
 600
 601         /* the pending IO might have been the only thing that kept this buffer
 602          * in memory.  Make sure we have a ref for all this other checks
 603          */
 604         extent_buffer_get(eb);
 605
 606         reads_done = atomic_dec_and_test(&eb->io_pages);
 607         if (!reads_done)
 608                 goto err;
 609
 610         eb->read_mirror = mirror;
 611         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 612                 ret = -EIO;
 613                 goto err;
 614         }
 615
 616         found_start = btrfs_header_bytenr(eb);
 617         if (found_start != eb->start) {
 618                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 619                              eb->start, found_start);
 620                 ret = -EIO;
 621                 goto err;
 622         }
 623         if (check_tree_block_fsid(fs_info, eb)) {
 624                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 625                              eb->start);
 626                 ret = -EIO;
 627                 goto err;
 628         }
 629         found_level = btrfs_header_level(eb);
 630         if (found_level >= BTRFS_MAX_LEVEL) {
 631                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 632                           (int)btrfs_header_level(eb), eb->start);
 633                 ret = -EIO;
 634                 goto err;
 635         }
 636
 637         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 638                                        eb, found_level);
 639
 640         ret = csum_tree_block(fs_info, eb, 1);
 641         if (ret)
 642                 goto err;
 643
 644         /*
 645          * If this is a leaf block and it is corrupt, set the corrupt bit so
 646          * that we don't try and read the other copies of this block, just
 647          * return -EIO.
 648          */
 649         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 650                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 651                 ret = -EIO;
 652         }
 653
 654         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 655                 ret = -EIO;
 656
 657         if (!ret)
 658                 set_extent_buffer_uptodate(eb);
 659 err:
 660         if (reads_done &&
 661             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 662                 btree_readahead_hook(eb, ret);
 663
 664         if (ret) {
 665                 /*
 666                  * our io error hook is going to dec the io pages
 667                  * again, we have to make sure it has something
 668                  * to decrement
 669                  */
 670                 atomic_inc(&eb->io_pages);
 671                 clear_extent_buffer_uptodate(eb);
 672         }
 673         free_extent_buffer(eb);
 674 out:
 675         return ret;
 676 }
 677
 678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 679 {
 680         struct extent_buffer *eb;
 681
 682         eb = (struct extent_buffer *)page->private;
 683         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 684         eb->read_mirror = failed_mirror;
 685         atomic_dec(&eb->io_pages);
 686         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 687                 btree_readahead_hook(eb, -EIO);
 688         return -EIO;    /* we fixed nothing */
 689 }
 690
 691 static void end_workqueue_bio(struct bio *bio)
 692 {
 693         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 694         struct btrfs_fs_info *fs_info;
 695         struct btrfs_workqueue *wq;
 696         btrfs_work_func_t func;
 697
 698         fs_info = end_io_wq->info;
 699         end_io_wq->status = bio->bi_status;
 700
 701         if (bio_op(bio) == REQ_OP_WRITE) {
 702                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 703                         wq = fs_info->endio_meta_write_workers;
 704                         func = btrfs_endio_meta_write_helper;
 705                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 706                         wq = fs_info->endio_freespace_worker;
 707                         func = btrfs_freespace_write_helper;
 708                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 709                         wq = fs_info->endio_raid56_workers;
 710                         func = btrfs_endio_raid56_helper;
 711                 } else {
 712                         wq = fs_info->endio_write_workers;
 713                         func = btrfs_endio_write_helper;
 714                 }
 715         } else {
 716                 if (unlikely(end_io_wq->metadata ==
 717                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 718                         wq = fs_info->endio_repair_workers;
 719                         func = btrfs_endio_repair_helper;
 720                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 721                         wq = fs_info->endio_raid56_workers;
 722                         func = btrfs_endio_raid56_helper;
 723                 } else if (end_io_wq->metadata) {
 724                         wq = fs_info->endio_meta_workers;
 725                         func = btrfs_endio_meta_helper;
 726                 } else {
 727                         wq = fs_info->endio_workers;
 728                         func = btrfs_endio_helper;
 729                 }
 730         }
 731
 732         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 733         btrfs_queue_work(wq, &end_io_wq->work);
 734 }
 735
 736 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 737                         enum btrfs_wq_endio_type metadata)
 738 {
 739         struct btrfs_end_io_wq *end_io_wq;
 740
 741         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 742         if (!end_io_wq)
 743                 return BLK_STS_RESOURCE;
 744
 745         end_io_wq->private = bio->bi_private;
 746         end_io_wq->end_io = bio->bi_end_io;
 747         end_io_wq->info = info;
 748         end_io_wq->status = 0;
 749         end_io_wq->bio = bio;
 750         end_io_wq->metadata = metadata;
 751
 752         bio->bi_private = end_io_wq;
 753         bio->bi_end_io = end_workqueue_bio;
 754         return 0;
 755 }
 756
 757 static void run_one_async_start(struct btrfs_work *work)
 758 {
 759         struct async_submit_bio *async;
 760         blk_status_t ret;
 761
 762         async = container_of(work, struct  async_submit_bio, work);
 763         ret = async->submit_bio_start(async->private_data, async->bio,
 764                                       async->bio_offset);
 765         if (ret)
 766                 async->status = ret;
 767 }
 768
 769 static void run_one_async_done(struct btrfs_work *work)
 770 {
 771         struct async_submit_bio *async;
 772
 773         async = container_of(work, struct  async_submit_bio, work);
 774
 775         /* If an error occurred we just want to clean up the bio and move on */
 776         if (async->status) {
 777                 async->bio->bi_status = async->status;
 778                 bio_endio(async->bio);
 779                 return;
 780         }
 781
 782         async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
 783 }
 784
 785 static void run_one_async_free(struct btrfs_work *work)
 786 {
 787         struct async_submit_bio *async;
 788
 789         async = container_of(work, struct  async_submit_bio, work);
 790         kfree(async);
 791 }
 792
 793 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 794                                  int mirror_num, unsigned long bio_flags,
 795                                  u64 bio_offset, void *private_data,
 796                                  extent_submit_bio_start_t *submit_bio_start,
 797                                  extent_submit_bio_done_t *submit_bio_done)
 798 {
 799         struct async_submit_bio *async;
 800
 801         async = kmalloc(sizeof(*async), GFP_NOFS);
 802         if (!async)
 803                 return BLK_STS_RESOURCE;
 804
 805         async->private_data = private_data;
 806         async->fs_info = fs_info;
 807         async->bio = bio;
 808         async->mirror_num = mirror_num;
 809         async->submit_bio_start = submit_bio_start;
 810         async->submit_bio_done = submit_bio_done;
 811
 812         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 813                         run_one_async_done, run_one_async_free);
 814
 815         async->bio_flags = bio_flags;
 816         async->bio_offset = bio_offset;
 817
 818         async->status = 0;
 819
 820         if (op_is_sync(bio->bi_opf))
 821                 btrfs_set_work_high_priority(&async->work);
 822
 823         btrfs_queue_work(fs_info->workers, &async->work);
 824         return 0;
 825 }
 826
 827 static blk_status_t btree_csum_one_bio(struct bio *bio)
 828 {
 829         struct bio_vec *bvec;
 830         struct btrfs_root *root;
 831         int i, ret = 0;
 832
 833         ASSERT(!bio_flagged(bio, BIO_CLONED));
 834         bio_for_each_segment_all(bvec, bio, i) {
 835                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 836                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 837                 if (ret)
 838                         break;
 839         }
 840
 841         return errno_to_blk_status(ret);
 842 }
 843
 844 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 845                                              u64 bio_offset)
 846 {
 847         /*
 848          * when we're called for a write, we're already in the async
 849          * submission context.  Just jump into btrfs_map_bio
 850          */
 851         return btree_csum_one_bio(bio);
 852 }
 853
 854 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
 855                                             int mirror_num)
 856 {
 857         struct inode *inode = private_data;
 858         blk_status_t ret;
 859
 860         /*
 861          * when we're called for a write, we're already in the async
 862          * submission context.  Just jump into btrfs_map_bio
 863          */
 864         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
 865         if (ret) {
 866                 bio->bi_status = ret;
 867                 bio_endio(bio);
 868         }
 869         return ret;
 870 }
 871
 872 static int check_async_write(struct btrfs_inode *bi)
 873 {
 874         if (atomic_read(&bi->sync_writers))
 875                 return 0;
 876 #ifdef CONFIG_X86
 877         if (static_cpu_has(X86_FEATURE_XMM4_2))
 878                 return 0;
 879 #endif
 880         return 1;
 881 }
 882
 883 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 884                                           int mirror_num, unsigned long bio_flags,
 885                                           u64 bio_offset)
 886 {
 887         struct inode *inode = private_data;
 888         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 889         int async = check_async_write(BTRFS_I(inode));
 890         blk_status_t ret;
 891
 892         if (bio_op(bio) != REQ_OP_WRITE) {
 893                 /*
 894                  * called for a read, do the setup so that checksum validation
 895                  * can happen in the async kernel threads
 896                  */
 897                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 898                                           BTRFS_WQ_ENDIO_METADATA);
 899                 if (ret)
 900                         goto out_w_error;
 901                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 902         } else if (!async) {
 903                 ret = btree_csum_one_bio(bio);
 904                 if (ret)
 905                         goto out_w_error;
 906                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 907         } else {
 908                 /*
 909                  * kthread helpers are used to submit writes so that
 910                  * checksumming can happen in parallel across all CPUs
 911                  */
 912                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 913                                           bio_offset, private_data,
 914                                           btree_submit_bio_start,
 915                                           btree_submit_bio_done);
 916         }
 917
 918         if (ret)
 919                 goto out_w_error;
 920         return 0;
 921
 922 out_w_error:
 923         bio->bi_status = ret;
 924         bio_endio(bio);
 925         return ret;
 926 }
 927
 928 #ifdef CONFIG_MIGRATION
 929 static int btree_migratepage(struct address_space *mapping,
 930                         struct page *newpage, struct page *page,
 931                         enum migrate_mode mode)
 932 {
 933         /*
 934          * we can't safely write a btree page from here,
 935          * we haven't done the locking hook
 936          */
 937         if (PageDirty(page))
 938                 return -EAGAIN;
 939         /*
 940          * Buffers may be managed in a filesystem specific way.
 941          * We must have no buffers or drop them.
 942          */
 943         if (page_has_private(page) &&
 944             !try_to_release_page(page, GFP_KERNEL))
 945                 return -EAGAIN;
 946         return migrate_page(mapping, newpage, page, mode);
 947 }
 948 #endif
 949
 950
 951 static int btree_writepages(struct address_space *mapping,
 952                             struct writeback_control *wbc)
 953 {
 954         struct btrfs_fs_info *fs_info;
 955         int ret;
 956
 957         if (wbc->sync_mode == WB_SYNC_NONE) {
 958
 959                 if (wbc->for_kupdate)
 960                         return 0;
 961
 962                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 963                 /* this is a bit racy, but that's ok */
 964                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 965                                              BTRFS_DIRTY_METADATA_THRESH);
 966                 if (ret < 0)
 967                         return 0;
 968         }
 969         return btree_write_cache_pages(mapping, wbc);
 970 }
 971
 972 static int btree_readpage(struct file *file, struct page *page)
 973 {
 974         struct extent_io_tree *tree;
 975         tree = &BTRFS_I(page->mapping->host)->io_tree;
 976         return extent_read_full_page(tree, page, btree_get_extent, 0);
 977 }
 978
 979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 980 {
 981         if (PageWriteback(page) || PageDirty(page))
 982                 return 0;
 983
 984         return try_release_extent_buffer(page);
 985 }
 986
 987 static void btree_invalidatepage(struct page *page, unsigned int offset,
 988                                  unsigned int length)
 989 {
 990         struct extent_io_tree *tree;
 991         tree = &BTRFS_I(page->mapping->host)->io_tree;
 992         extent_invalidatepage(tree, page, offset);
 993         btree_releasepage(page, GFP_NOFS);
 994         if (PagePrivate(page)) {
 995                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 996                            "page private not zero on page %llu",
 997                            (unsigned long long)page_offset(page));
 998                 ClearPagePrivate(page);
 999                 set_page_private(page, 0);
1000                 put_page(page);
1001         }
1002 }
1003
1004 static int btree_set_page_dirty(struct page *page)
1005 {
1006 #ifdef DEBUG
1007         struct extent_buffer *eb;
1008
1009         BUG_ON(!PagePrivate(page));
1010         eb = (struct extent_buffer *)page->private;
1011         BUG_ON(!eb);
1012         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013         BUG_ON(!atomic_read(&eb->refs));
1014         btrfs_assert_tree_locked(eb);
1015 #endif
1016         return __set_page_dirty_nobuffers(page);
1017 }
1018
1019 static const struct address_space_operations btree_aops = {
1020         .readpage       = btree_readpage,
1021         .writepages     = btree_writepages,
1022         .releasepage    = btree_releasepage,
1023         .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025         .migratepage    = btree_migratepage,
1026 #endif
1027         .set_page_dirty = btree_set_page_dirty,
1028 };
1029
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1031 {
1032         struct extent_buffer *buf = NULL;
1033         struct inode *btree_inode = fs_info->btree_inode;
1034
1035         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1036         if (IS_ERR(buf))
1037                 return;
1038         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039                                  buf, WAIT_NONE, 0);
1040         free_extent_buffer(buf);
1041 }
1042
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044                          int mirror_num, struct extent_buffer **eb)
1045 {
1046         struct extent_buffer *buf = NULL;
1047         struct inode *btree_inode = fs_info->btree_inode;
1048         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1049         int ret;
1050
1051         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052         if (IS_ERR(buf))
1053                 return 0;
1054
1055         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1056
1057         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1058                                        mirror_num);
1059         if (ret) {
1060                 free_extent_buffer(buf);
1061                 return ret;
1062         }
1063
1064         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065                 free_extent_buffer(buf);
1066                 return -EIO;
1067         } else if (extent_buffer_uptodate(buf)) {
1068                 *eb = buf;
1069         } else {
1070                 free_extent_buffer(buf);
1071         }
1072         return 0;
1073 }
1074
1075 struct extent_buffer *btrfs_find_create_tree_block(
1076                                                 struct btrfs_fs_info *fs_info,
1077                                                 u64 bytenr)
1078 {
1079         if (btrfs_is_testing(fs_info))
1080                 return alloc_test_extent_buffer(fs_info, bytenr);
1081         return alloc_extent_buffer(fs_info, bytenr);
1082 }
1083
1084
1085 int btrfs_write_tree_block(struct extent_buffer *buf)
1086 {
1087         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088                                         buf->start + buf->len - 1);
1089 }
1090
1091 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1092 {
1093         filemap_fdatawait_range(buf->pages[0]->mapping,
1094                                 buf->start, buf->start + buf->len - 1);
1095 }
1096
1097 /*
1098  * Read tree block at logical address @bytenr and do variant basic but critical
1099  * verification.
1100  *
1101  * @parent_transid:     expected transid of this tree block, skip check if 0
1102  * @level:              expected level, mandatory check
1103  * @first_key:          expected key in slot 0, skip check if NULL
1104  */
1105 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106                                       u64 parent_transid, int level,
1107                                       struct btrfs_key *first_key)
1108 {
1109         struct extent_buffer *buf = NULL;
1110         int ret;
1111
1112         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1113         if (IS_ERR(buf))
1114                 return buf;
1115
1116         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1117                                              level, first_key);
1118         if (ret) {
1119                 free_extent_buffer(buf);
1120                 return ERR_PTR(ret);
1121         }
1122         return buf;
1123
1124 }
1125
1126 void clean_tree_block(struct btrfs_fs_info *fs_info,
1127                       struct extent_buffer *buf)
1128 {
1129         if (btrfs_header_generation(buf) ==
1130             fs_info->running_transaction->transid) {
1131                 btrfs_assert_tree_locked(buf);
1132
1133                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1135                                                  -buf->len,
1136                                                  fs_info->dirty_metadata_batch);
1137                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1138                         btrfs_set_lock_blocking(buf);
1139                         clear_extent_buffer_dirty(buf);
1140                 }
1141         }
1142 }
1143
1144 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1145 {
1146         struct btrfs_subvolume_writers *writers;
1147         int ret;
1148
1149         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1150         if (!writers)
1151                 return ERR_PTR(-ENOMEM);
1152
1153         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1154         if (ret < 0) {
1155                 kfree(writers);
1156                 return ERR_PTR(ret);
1157         }
1158
1159         init_waitqueue_head(&writers->wait);
1160         return writers;
1161 }
1162
1163 static void
1164 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1165 {
1166         percpu_counter_destroy(&writers->counter);
1167         kfree(writers);
1168 }
1169
1170 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1171                          u64 objectid)
1172 {
1173         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1174         root->node = NULL;
1175         root->commit_root = NULL;
1176         root->state = 0;
1177         root->orphan_cleanup_state = 0;
1178
1179         root->objectid = objectid;
1180         root->last_trans = 0;
1181         root->highest_objectid = 0;
1182         root->nr_delalloc_inodes = 0;
1183         root->nr_ordered_extents = 0;
1184         root->name = NULL;
1185         root->inode_tree = RB_ROOT;
1186         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187         root->block_rsv = NULL;
1188
1189         INIT_LIST_HEAD(&root->dirty_list);
1190         INIT_LIST_HEAD(&root->root_list);
1191         INIT_LIST_HEAD(&root->delalloc_inodes);
1192         INIT_LIST_HEAD(&root->delalloc_root);
1193         INIT_LIST_HEAD(&root->ordered_extents);
1194         INIT_LIST_HEAD(&root->ordered_root);
1195         INIT_LIST_HEAD(&root->logged_list[0]);
1196         INIT_LIST_HEAD(&root->logged_list[1]);
1197         spin_lock_init(&root->inode_lock);
1198         spin_lock_init(&root->delalloc_lock);
1199         spin_lock_init(&root->ordered_extent_lock);
1200         spin_lock_init(&root->accounting_lock);
1201         spin_lock_init(&root->log_extents_lock[0]);
1202         spin_lock_init(&root->log_extents_lock[1]);
1203         spin_lock_init(&root->qgroup_meta_rsv_lock);
1204         mutex_init(&root->objectid_mutex);
1205         mutex_init(&root->log_mutex);
1206         mutex_init(&root->ordered_extent_mutex);
1207         mutex_init(&root->delalloc_mutex);
1208         init_waitqueue_head(&root->log_writer_wait);
1209         init_waitqueue_head(&root->log_commit_wait[0]);
1210         init_waitqueue_head(&root->log_commit_wait[1]);
1211         INIT_LIST_HEAD(&root->log_ctxs[0]);
1212         INIT_LIST_HEAD(&root->log_ctxs[1]);
1213         atomic_set(&root->log_commit[0], 0);
1214         atomic_set(&root->log_commit[1], 0);
1215         atomic_set(&root->log_writers, 0);
1216         atomic_set(&root->log_batch, 0);
1217         refcount_set(&root->refs, 1);
1218         atomic_set(&root->will_be_snapshotted, 0);
1219         root->log_transid = 0;
1220         root->log_transid_committed = -1;
1221         root->last_log_commit = 0;
1222         if (!dummy)
1223                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1224
1225         memset(&root->root_key, 0, sizeof(root->root_key));
1226         memset(&root->root_item, 0, sizeof(root->root_item));
1227         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1228         if (!dummy)
1229                 root->defrag_trans_start = fs_info->generation;
1230         else
1231                 root->defrag_trans_start = 0;
1232         root->root_key.objectid = objectid;
1233         root->anon_dev = 0;
1234
1235         spin_lock_init(&root->root_item_lock);
1236 }
1237
1238 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1239                 gfp_t flags)
1240 {
1241         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1242         if (root)
1243                 root->fs_info = fs_info;
1244         return root;
1245 }
1246
1247 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1248 /* Should only be used by the testing infrastructure */
1249 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1250 {
1251         struct btrfs_root *root;
1252
1253         if (!fs_info)
1254                 return ERR_PTR(-EINVAL);
1255
1256         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1257         if (!root)
1258                 return ERR_PTR(-ENOMEM);
1259
1260         /* We don't use the stripesize in selftest, set it as sectorsize */
1261         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1262         root->alloc_bytenr = 0;
1263
1264         return root;
1265 }
1266 #endif
1267
1268 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269                                      struct btrfs_fs_info *fs_info,
1270                                      u64 objectid)
1271 {
1272         struct extent_buffer *leaf;
1273         struct btrfs_root *tree_root = fs_info->tree_root;
1274         struct btrfs_root *root;
1275         struct btrfs_key key;
1276         int ret = 0;
1277         uuid_le uuid = NULL_UUID_LE;
1278
1279         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1280         if (!root)
1281                 return ERR_PTR(-ENOMEM);
1282
1283         __setup_root(root, fs_info, objectid);
1284         root->root_key.objectid = objectid;
1285         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1286         root->root_key.offset = 0;
1287
1288         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1289         if (IS_ERR(leaf)) {
1290                 ret = PTR_ERR(leaf);
1291                 leaf = NULL;
1292                 goto fail;
1293         }
1294
1295         root->node = leaf;
1296         btrfs_mark_buffer_dirty(leaf);
1297
1298         root->commit_root = btrfs_root_node(root);
1299         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1300
1301         root->root_item.flags = 0;
1302         root->root_item.byte_limit = 0;
1303         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1304         btrfs_set_root_generation(&root->root_item, trans->transid);
1305         btrfs_set_root_level(&root->root_item, 0);
1306         btrfs_set_root_refs(&root->root_item, 1);
1307         btrfs_set_root_used(&root->root_item, leaf->len);
1308         btrfs_set_root_last_snapshot(&root->root_item, 0);
1309         btrfs_set_root_dirid(&root->root_item, 0);
1310         if (is_fstree(objectid))
1311                 uuid_le_gen(&uuid);
1312         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1313         root->root_item.drop_level = 0;
1314
1315         key.objectid = objectid;
1316         key.type = BTRFS_ROOT_ITEM_KEY;
1317         key.offset = 0;
1318         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1319         if (ret)
1320                 goto fail;
1321
1322         btrfs_tree_unlock(leaf);
1323
1324         return root;
1325
1326 fail:
1327         if (leaf) {
1328                 btrfs_tree_unlock(leaf);
1329                 free_extent_buffer(root->commit_root);
1330                 free_extent_buffer(leaf);
1331         }
1332         kfree(root);
1333
1334         return ERR_PTR(ret);
1335 }
1336
1337 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1338                                          struct btrfs_fs_info *fs_info)
1339 {
1340         struct btrfs_root *root;
1341         struct extent_buffer *leaf;
1342
1343         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1344         if (!root)
1345                 return ERR_PTR(-ENOMEM);
1346
1347         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1348
1349         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1350         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1351         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1352
1353         /*
1354          * DON'T set REF_COWS for log trees
1355          *
1356          * log trees do not get reference counted because they go away
1357          * before a real commit is actually done.  They do store pointers
1358          * to file data extents, and those reference counts still get
1359          * updated (along with back refs to the log tree).
1360          */
1361
1362         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1363                         NULL, 0, 0, 0);
1364         if (IS_ERR(leaf)) {
1365                 kfree(root);
1366                 return ERR_CAST(leaf);
1367         }
1368
1369         root->node = leaf;
1370
1371         btrfs_mark_buffer_dirty(root->node);
1372         btrfs_tree_unlock(root->node);
1373         return root;
1374 }
1375
1376 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1377                              struct btrfs_fs_info *fs_info)
1378 {
1379         struct btrfs_root *log_root;
1380
1381         log_root = alloc_log_tree(trans, fs_info);
1382         if (IS_ERR(log_root))
1383                 return PTR_ERR(log_root);
1384         WARN_ON(fs_info->log_root_tree);
1385         fs_info->log_root_tree = log_root;
1386         return 0;
1387 }
1388
1389 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1390                        struct btrfs_root *root)
1391 {
1392         struct btrfs_fs_info *fs_info = root->fs_info;
1393         struct btrfs_root *log_root;
1394         struct btrfs_inode_item *inode_item;
1395
1396         log_root = alloc_log_tree(trans, fs_info);
1397         if (IS_ERR(log_root))
1398                 return PTR_ERR(log_root);
1399
1400         log_root->last_trans = trans->transid;
1401         log_root->root_key.offset = root->root_key.objectid;
1402
1403         inode_item = &log_root->root_item.inode;
1404         btrfs_set_stack_inode_generation(inode_item, 1);
1405         btrfs_set_stack_inode_size(inode_item, 3);
1406         btrfs_set_stack_inode_nlink(inode_item, 1);
1407         btrfs_set_stack_inode_nbytes(inode_item,
1408                                      fs_info->nodesize);
1409         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1410
1411         btrfs_set_root_node(&log_root->root_item, log_root->node);
1412
1413         WARN_ON(root->log_root);
1414         root->log_root = log_root;
1415         root->log_transid = 0;
1416         root->log_transid_committed = -1;
1417         root->last_log_commit = 0;
1418         return 0;
1419 }
1420
1421 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1422                                                struct btrfs_key *key)
1423 {
1424         struct btrfs_root *root;
1425         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1426         struct btrfs_path *path;
1427         u64 generation;
1428         int ret;
1429         int level;
1430
1431         path = btrfs_alloc_path();
1432         if (!path)
1433                 return ERR_PTR(-ENOMEM);
1434
1435         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1436         if (!root) {
1437                 ret = -ENOMEM;
1438                 goto alloc_fail;
1439         }
1440
1441         __setup_root(root, fs_info, key->objectid);
1442
1443         ret = btrfs_find_root(tree_root, key, path,
1444                               &root->root_item, &root->root_key);
1445         if (ret) {
1446                 if (ret > 0)
1447                         ret = -ENOENT;
1448                 goto find_fail;
1449         }
1450
1451         generation = btrfs_root_generation(&root->root_item);
1452         level = btrfs_root_level(&root->root_item);
1453         root->node = read_tree_block(fs_info,
1454                                      btrfs_root_bytenr(&root->root_item),
1455                                      generation, level, NULL);
1456         if (IS_ERR(root->node)) {
1457                 ret = PTR_ERR(root->node);
1458                 goto find_fail;
1459         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1460                 ret = -EIO;
1461                 free_extent_buffer(root->node);
1462                 goto find_fail;
1463         }
1464         root->commit_root = btrfs_root_node(root);
1465 out:
1466         btrfs_free_path(path);
1467         return root;
1468
1469 find_fail:
1470         kfree(root);
1471 alloc_fail:
1472         root = ERR_PTR(ret);
1473         goto out;
1474 }
1475
1476 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1477                                       struct btrfs_key *location)
1478 {
1479         struct btrfs_root *root;
1480
1481         root = btrfs_read_tree_root(tree_root, location);
1482         if (IS_ERR(root))
1483                 return root;
1484
1485         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1486                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1487                 btrfs_check_and_init_root_item(&root->root_item);
1488         }
1489
1490         return root;
1491 }
1492
1493 int btrfs_init_fs_root(struct btrfs_root *root)
1494 {
1495         int ret;
1496         struct btrfs_subvolume_writers *writers;
1497
1498         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1499         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1500                                         GFP_NOFS);
1501         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1502                 ret = -ENOMEM;
1503                 goto fail;
1504         }
1505
1506         writers = btrfs_alloc_subvolume_writers();
1507         if (IS_ERR(writers)) {
1508                 ret = PTR_ERR(writers);
1509                 goto fail;
1510         }
1511         root->subv_writers = writers;
1512
1513         btrfs_init_free_ino_ctl(root);
1514         spin_lock_init(&root->ino_cache_lock);
1515         init_waitqueue_head(&root->ino_cache_wait);
1516
1517         ret = get_anon_bdev(&root->anon_dev);
1518         if (ret)
1519                 goto fail;
1520
1521         mutex_lock(&root->objectid_mutex);
1522         ret = btrfs_find_highest_objectid(root,
1523                                         &root->highest_objectid);
1524         if (ret) {
1525                 mutex_unlock(&root->objectid_mutex);
1526                 goto fail;
1527         }
1528
1529         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1530
1531         mutex_unlock(&root->objectid_mutex);
1532
1533         return 0;
1534 fail:
1535         /* the caller is responsible to call free_fs_root */
1536         return ret;
1537 }
1538
1539 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1540                                         u64 root_id)
1541 {
1542         struct btrfs_root *root;
1543
1544         spin_lock(&fs_info->fs_roots_radix_lock);
1545         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1546                                  (unsigned long)root_id);
1547         spin_unlock(&fs_info->fs_roots_radix_lock);
1548         return root;
1549 }
1550
1551 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1552                          struct btrfs_root *root)
1553 {
1554         int ret;
1555
1556         ret = radix_tree_preload(GFP_NOFS);
1557         if (ret)
1558                 return ret;
1559
1560         spin_lock(&fs_info->fs_roots_radix_lock);
1561         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1562                                 (unsigned long)root->root_key.objectid,
1563                                 root);
1564         if (ret == 0)
1565                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1566         spin_unlock(&fs_info->fs_roots_radix_lock);
1567         radix_tree_preload_end();
1568
1569         return ret;
1570 }
1571
1572 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1573                                      struct btrfs_key *location,
1574                                      bool check_ref)
1575 {
1576         struct btrfs_root *root;
1577         struct btrfs_path *path;
1578         struct btrfs_key key;
1579         int ret;
1580
1581         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1582                 return fs_info->tree_root;
1583         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1584                 return fs_info->extent_root;
1585         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1586                 return fs_info->chunk_root;
1587         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1588                 return fs_info->dev_root;
1589         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1590                 return fs_info->csum_root;
1591         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1592                 return fs_info->quota_root ? fs_info->quota_root :
1593                                              ERR_PTR(-ENOENT);
1594         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1595                 return fs_info->uuid_root ? fs_info->uuid_root :
1596                                             ERR_PTR(-ENOENT);
1597         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1598                 return fs_info->free_space_root ? fs_info->free_space_root :
1599                                                   ERR_PTR(-ENOENT);
1600 again:
1601         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1602         if (root) {
1603                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1604                         return ERR_PTR(-ENOENT);
1605                 return root;
1606         }
1607
1608         root = btrfs_read_fs_root(fs_info->tree_root, location);
1609         if (IS_ERR(root))
1610                 return root;
1611
1612         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1613                 ret = -ENOENT;
1614                 goto fail;
1615         }
1616
1617         ret = btrfs_init_fs_root(root);
1618         if (ret)
1619                 goto fail;
1620
1621         path = btrfs_alloc_path();
1622         if (!path) {
1623                 ret = -ENOMEM;
1624                 goto fail;
1625         }
1626         key.objectid = BTRFS_ORPHAN_OBJECTID;
1627         key.type = BTRFS_ORPHAN_ITEM_KEY;
1628         key.offset = location->objectid;
1629
1630         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1631         btrfs_free_path(path);
1632         if (ret < 0)
1633                 goto fail;
1634         if (ret == 0)
1635                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1636
1637         ret = btrfs_insert_fs_root(fs_info, root);
1638         if (ret) {
1639                 if (ret == -EEXIST) {
1640                         free_fs_root(root);
1641                         goto again;
1642                 }
1643                 goto fail;
1644         }
1645         return root;
1646 fail:
1647         free_fs_root(root);
1648         return ERR_PTR(ret);
1649 }
1650
1651 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1652 {
1653         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1654         int ret = 0;
1655         struct btrfs_device *device;
1656         struct backing_dev_info *bdi;
1657
1658         rcu_read_lock();
1659         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1660                 if (!device->bdev)
1661                         continue;
1662                 bdi = device->bdev->bd_bdi;
1663                 if (bdi_congested(bdi, bdi_bits)) {
1664                         ret = 1;
1665                         break;
1666                 }
1667         }
1668         rcu_read_unlock();
1669         return ret;
1670 }
1671
1672 /*
1673  * called by the kthread helper functions to finally call the bio end_io
1674  * functions.  This is where read checksum verification actually happens
1675  */
1676 static void end_workqueue_fn(struct btrfs_work *work)
1677 {
1678         struct bio *bio;
1679         struct btrfs_end_io_wq *end_io_wq;
1680
1681         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1682         bio = end_io_wq->bio;
1683
1684         bio->bi_status = end_io_wq->status;
1685         bio->bi_private = end_io_wq->private;
1686         bio->bi_end_io = end_io_wq->end_io;
1687         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1688         bio_endio(bio);
1689 }
1690
1691 static int cleaner_kthread(void *arg)
1692 {
1693         struct btrfs_root *root = arg;
1694         struct btrfs_fs_info *fs_info = root->fs_info;
1695         int again;
1696         struct btrfs_trans_handle *trans;
1697
1698         do {
1699                 again = 0;
1700
1701                 /* Make the cleaner go to sleep early. */
1702                 if (btrfs_need_cleaner_sleep(fs_info))
1703                         goto sleep;
1704
1705                 /*
1706                  * Do not do anything if we might cause open_ctree() to block
1707                  * before we have finished mounting the filesystem.
1708                  */
1709                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1710                         goto sleep;
1711
1712                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1713                         goto sleep;
1714
1715                 /*
1716                  * Avoid the problem that we change the status of the fs
1717                  * during the above check and trylock.
1718                  */
1719                 if (btrfs_need_cleaner_sleep(fs_info)) {
1720                         mutex_unlock(&fs_info->cleaner_mutex);
1721                         goto sleep;
1722                 }
1723
1724                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1725                 btrfs_run_delayed_iputs(fs_info);
1726                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1727
1728                 again = btrfs_clean_one_deleted_snapshot(root);
1729                 mutex_unlock(&fs_info->cleaner_mutex);
1730
1731                 /*
1732                  * The defragger has dealt with the R/O remount and umount,
1733                  * needn't do anything special here.
1734                  */
1735                 btrfs_run_defrag_inodes(fs_info);
1736
1737                 /*
1738                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1739                  * with relocation (btrfs_relocate_chunk) and relocation
1740                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1741                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1742                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1743                  * unused block groups.
1744                  */
1745                 btrfs_delete_unused_bgs(fs_info);
1746 sleep:
1747                 if (!again) {
1748                         set_current_state(TASK_INTERRUPTIBLE);
1749                         if (!kthread_should_stop())
1750                                 schedule();
1751                         __set_current_state(TASK_RUNNING);
1752                 }
1753         } while (!kthread_should_stop());
1754
1755         /*
1756          * Transaction kthread is stopped before us and wakes us up.
1757          * However we might have started a new transaction and COWed some
1758          * tree blocks when deleting unused block groups for example. So
1759          * make sure we commit the transaction we started to have a clean
1760          * shutdown when evicting the btree inode - if it has dirty pages
1761          * when we do the final iput() on it, eviction will trigger a
1762          * writeback for it which will fail with null pointer dereferences
1763          * since work queues and other resources were already released and
1764          * destroyed by the time the iput/eviction/writeback is made.
1765          */
1766         trans = btrfs_attach_transaction(root);
1767         if (IS_ERR(trans)) {
1768                 if (PTR_ERR(trans) != -ENOENT)
1769                         btrfs_err(fs_info,
1770                                   "cleaner transaction attach returned %ld",
1771                                   PTR_ERR(trans));
1772         } else {
1773                 int ret;
1774
1775                 ret = btrfs_commit_transaction(trans);
1776                 if (ret)
1777                         btrfs_err(fs_info,
1778                                   "cleaner open transaction commit returned %d",
1779                                   ret);
1780         }
1781
1782         return 0;
1783 }
1784
1785 static int transaction_kthread(void *arg)
1786 {
1787         struct btrfs_root *root = arg;
1788         struct btrfs_fs_info *fs_info = root->fs_info;
1789         struct btrfs_trans_handle *trans;
1790         struct btrfs_transaction *cur;
1791         u64 transid;
1792         time64_t now;
1793         unsigned long delay;
1794         bool cannot_commit;
1795
1796         do {
1797                 cannot_commit = false;
1798                 delay = HZ * fs_info->commit_interval;
1799                 mutex_lock(&fs_info->transaction_kthread_mutex);
1800
1801                 spin_lock(&fs_info->trans_lock);
1802                 cur = fs_info->running_transaction;
1803                 if (!cur) {
1804                         spin_unlock(&fs_info->trans_lock);
1805                         goto sleep;
1806                 }
1807
1808                 now = ktime_get_seconds();
1809                 if (cur->state < TRANS_STATE_BLOCKED &&
1810                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1811                     (now < cur->start_time ||
1812                      now - cur->start_time < fs_info->commit_interval)) {
1813                         spin_unlock(&fs_info->trans_lock);
1814                         delay = HZ * 5;
1815                         goto sleep;
1816                 }
1817                 transid = cur->transid;
1818                 spin_unlock(&fs_info->trans_lock);
1819
1820                 /* If the file system is aborted, this will always fail. */
1821                 trans = btrfs_attach_transaction(root);
1822                 if (IS_ERR(trans)) {
1823                         if (PTR_ERR(trans) != -ENOENT)
1824                                 cannot_commit = true;
1825                         goto sleep;
1826                 }
1827                 if (transid == trans->transid) {
1828                         btrfs_commit_transaction(trans);
1829                 } else {
1830                         btrfs_end_transaction(trans);
1831                 }
1832 sleep:
1833                 wake_up_process(fs_info->cleaner_kthread);
1834                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1835
1836                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1837                                       &fs_info->fs_state)))
1838                         btrfs_cleanup_transaction(fs_info);
1839                 if (!kthread_should_stop() &&
1840                                 (!btrfs_transaction_blocked(fs_info) ||
1841                                  cannot_commit))
1842                         schedule_timeout_interruptible(delay);
1843         } while (!kthread_should_stop());
1844         return 0;
1845 }
1846
1847 /*
1848  * this will find the highest generation in the array of
1849  * root backups.  The index of the highest array is returned,
1850  * or -1 if we can't find anything.
1851  *
1852  * We check to make sure the array is valid by comparing the
1853  * generation of the latest  root in the array with the generation
1854  * in the super block.  If they don't match we pitch it.
1855  */
1856 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1857 {
1858         u64 cur;
1859         int newest_index = -1;
1860         struct btrfs_root_backup *root_backup;
1861         int i;
1862
1863         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1864                 root_backup = info->super_copy->super_roots + i;
1865                 cur = btrfs_backup_tree_root_gen(root_backup);
1866                 if (cur == newest_gen)
1867                         newest_index = i;
1868         }
1869
1870         /* check to see if we actually wrapped around */
1871         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1872                 root_backup = info->super_copy->super_roots;
1873                 cur = btrfs_backup_tree_root_gen(root_backup);
1874                 if (cur == newest_gen)
1875                         newest_index = 0;
1876         }
1877         return newest_index;
1878 }
1879
1880
1881 /*
1882  * find the oldest backup so we know where to store new entries
1883  * in the backup array.  This will set the backup_root_index
1884  * field in the fs_info struct
1885  */
1886 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1887                                      u64 newest_gen)
1888 {
1889         int newest_index = -1;
1890
1891         newest_index = find_newest_super_backup(info, newest_gen);
1892         /* if there was garbage in there, just move along */
1893         if (newest_index == -1) {
1894                 info->backup_root_index = 0;
1895         } else {
1896                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1897         }
1898 }
1899
1900 /*
1901  * copy all the root pointers into the super backup array.
1902  * this will bump the backup pointer by one when it is
1903  * done
1904  */
1905 static void backup_super_roots(struct btrfs_fs_info *info)
1906 {
1907         int next_backup;
1908         struct btrfs_root_backup *root_backup;
1909         int last_backup;
1910
1911         next_backup = info->backup_root_index;
1912         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1913                 BTRFS_NUM_BACKUP_ROOTS;
1914
1915         /*
1916          * just overwrite the last backup if we're at the same generation
1917          * this happens only at umount
1918          */
1919         root_backup = info->super_for_commit->super_roots + last_backup;
1920         if (btrfs_backup_tree_root_gen(root_backup) ==
1921             btrfs_header_generation(info->tree_root->node))
1922                 next_backup = last_backup;
1923
1924         root_backup = info->super_for_commit->super_roots + next_backup;
1925
1926         /*
1927          * make sure all of our padding and empty slots get zero filled
1928          * regardless of which ones we use today
1929          */
1930         memset(root_backup, 0, sizeof(*root_backup));
1931
1932         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1933
1934         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1935         btrfs_set_backup_tree_root_gen(root_backup,
1936                                btrfs_header_generation(info->tree_root->node));
1937
1938         btrfs_set_backup_tree_root_level(root_backup,
1939                                btrfs_header_level(info->tree_root->node));
1940
1941         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1942         btrfs_set_backup_chunk_root_gen(root_backup,
1943                                btrfs_header_generation(info->chunk_root->node));
1944         btrfs_set_backup_chunk_root_level(root_backup,
1945                                btrfs_header_level(info->chunk_root->node));
1946
1947         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1948         btrfs_set_backup_extent_root_gen(root_backup,
1949                                btrfs_header_generation(info->extent_root->node));
1950         btrfs_set_backup_extent_root_level(root_backup,
1951                                btrfs_header_level(info->extent_root->node));
1952
1953         /*
1954          * we might commit during log recovery, which happens before we set
1955          * the fs_root.  Make sure it is valid before we fill it in.
1956          */
1957         if (info->fs_root && info->fs_root->node) {
1958                 btrfs_set_backup_fs_root(root_backup,
1959                                          info->fs_root->node->start);
1960                 btrfs_set_backup_fs_root_gen(root_backup,
1961                                btrfs_header_generation(info->fs_root->node));
1962                 btrfs_set_backup_fs_root_level(root_backup,
1963                                btrfs_header_level(info->fs_root->node));
1964         }
1965
1966         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1967         btrfs_set_backup_dev_root_gen(root_backup,
1968                                btrfs_header_generation(info->dev_root->node));
1969         btrfs_set_backup_dev_root_level(root_backup,
1970                                        btrfs_header_level(info->dev_root->node));
1971
1972         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1973         btrfs_set_backup_csum_root_gen(root_backup,
1974                                btrfs_header_generation(info->csum_root->node));
1975         btrfs_set_backup_csum_root_level(root_backup,
1976                                btrfs_header_level(info->csum_root->node));
1977
1978         btrfs_set_backup_total_bytes(root_backup,
1979                              btrfs_super_total_bytes(info->super_copy));
1980         btrfs_set_backup_bytes_used(root_backup,
1981                              btrfs_super_bytes_used(info->super_copy));
1982         btrfs_set_backup_num_devices(root_backup,
1983                              btrfs_super_num_devices(info->super_copy));
1984
1985         /*
1986          * if we don't copy this out to the super_copy, it won't get remembered
1987          * for the next commit
1988          */
1989         memcpy(&info->super_copy->super_roots,
1990                &info->super_for_commit->super_roots,
1991                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1992 }
1993
1994 /*
1995  * this copies info out of the root backup array and back into
1996  * the in-memory super block.  It is meant to help iterate through
1997  * the array, so you send it the number of backups you've already
1998  * tried and the last backup index you used.
1999  *
2000  * this returns -1 when it has tried all the backups
2001  */
2002 static noinline int next_root_backup(struct btrfs_fs_info *info,
2003                                      struct btrfs_super_block *super,
2004                                      int *num_backups_tried, int *backup_index)
2005 {
2006         struct btrfs_root_backup *root_backup;
2007         int newest = *backup_index;
2008
2009         if (*num_backups_tried == 0) {
2010                 u64 gen = btrfs_super_generation(super);
2011
2012                 newest = find_newest_super_backup(info, gen);
2013                 if (newest == -1)
2014                         return -1;
2015
2016                 *backup_index = newest;
2017                 *num_backups_tried = 1;
2018         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2019                 /* we've tried all the backups, all done */
2020                 return -1;
2021         } else {
2022                 /* jump to the next oldest backup */
2023                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2024                         BTRFS_NUM_BACKUP_ROOTS;
2025                 *backup_index = newest;
2026                 *num_backups_tried += 1;
2027         }
2028         root_backup = super->super_roots + newest;
2029
2030         btrfs_set_super_generation(super,
2031                                    btrfs_backup_tree_root_gen(root_backup));
2032         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2033         btrfs_set_super_root_level(super,
2034                                    btrfs_backup_tree_root_level(root_backup));
2035         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2036
2037         /*
2038          * fixme: the total bytes and num_devices need to match or we should
2039          * need a fsck
2040          */
2041         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2042         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2043         return 0;
2044 }
2045
2046 /* helper to cleanup workers */
2047 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2048 {
2049         btrfs_destroy_workqueue(fs_info->fixup_workers);
2050         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2051         btrfs_destroy_workqueue(fs_info->workers);
2052         btrfs_destroy_workqueue(fs_info->endio_workers);
2053         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2054         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2055         btrfs_destroy_workqueue(fs_info->rmw_workers);
2056         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2057         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2058         btrfs_destroy_workqueue(fs_info->submit_workers);
2059         btrfs_destroy_workqueue(fs_info->delayed_workers);
2060         btrfs_destroy_workqueue(fs_info->caching_workers);
2061         btrfs_destroy_workqueue(fs_info->readahead_workers);
2062         btrfs_destroy_workqueue(fs_info->flush_workers);
2063         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2064         btrfs_destroy_workqueue(fs_info->extent_workers);
2065         /*
2066          * Now that all other work queues are destroyed, we can safely destroy
2067          * the queues used for metadata I/O, since tasks from those other work
2068          * queues can do metadata I/O operations.
2069          */
2070         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2071         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2072 }
2073
2074 static void free_root_extent_buffers(struct btrfs_root *root)
2075 {
2076         if (root) {
2077                 free_extent_buffer(root->node);
2078                 free_extent_buffer(root->commit_root);
2079                 root->node = NULL;
2080                 root->commit_root = NULL;
2081         }
2082 }
2083
2084 /* helper to cleanup tree roots */
2085 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2086 {
2087         free_root_extent_buffers(info->tree_root);
2088
2089         free_root_extent_buffers(info->dev_root);
2090         free_root_extent_buffers(info->extent_root);
2091         free_root_extent_buffers(info->csum_root);
2092         free_root_extent_buffers(info->quota_root);
2093         free_root_extent_buffers(info->uuid_root);
2094         if (chunk_root)
2095                 free_root_extent_buffers(info->chunk_root);
2096         free_root_extent_buffers(info->free_space_root);
2097 }
2098
2099 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2100 {
2101         int ret;
2102         struct btrfs_root *gang[8];
2103         int i;
2104
2105         while (!list_empty(&fs_info->dead_roots)) {
2106                 gang[0] = list_entry(fs_info->dead_roots.next,
2107                                      struct btrfs_root, root_list);
2108                 list_del(&gang[0]->root_list);
2109
2110                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2111                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2112                 } else {
2113                         free_extent_buffer(gang[0]->node);
2114                         free_extent_buffer(gang[0]->commit_root);
2115                         btrfs_put_fs_root(gang[0]);
2116                 }
2117         }
2118
2119         while (1) {
2120                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2121                                              (void **)gang, 0,
2122                                              ARRAY_SIZE(gang));
2123                 if (!ret)
2124                         break;
2125                 for (i = 0; i < ret; i++)
2126                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2127         }
2128
2129         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2130                 btrfs_free_log_root_tree(NULL, fs_info);
2131                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2132         }
2133 }
2134
2135 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2136 {
2137         mutex_init(&fs_info->scrub_lock);
2138         atomic_set(&fs_info->scrubs_running, 0);
2139         atomic_set(&fs_info->scrub_pause_req, 0);
2140         atomic_set(&fs_info->scrubs_paused, 0);
2141         atomic_set(&fs_info->scrub_cancel_req, 0);
2142         init_waitqueue_head(&fs_info->scrub_pause_wait);
2143         fs_info->scrub_workers_refcnt = 0;
2144 }
2145
2146 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2147 {
2148         spin_lock_init(&fs_info->balance_lock);
2149         mutex_init(&fs_info->balance_mutex);
2150         atomic_set(&fs_info->balance_pause_req, 0);
2151         atomic_set(&fs_info->balance_cancel_req, 0);
2152         fs_info->balance_ctl = NULL;
2153         init_waitqueue_head(&fs_info->balance_wait_q);
2154 }
2155
2156 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2157 {
2158         struct inode *inode = fs_info->btree_inode;
2159
2160         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2161         set_nlink(inode, 1);
2162         /*
2163          * we set the i_size on the btree inode to the max possible int.
2164          * the real end of the address space is determined by all of
2165          * the devices in the system
2166          */
2167         inode->i_size = OFFSET_MAX;
2168         inode->i_mapping->a_ops = &btree_aops;
2169
2170         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2171         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2172         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2173         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2174
2175         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2176
2177         BTRFS_I(inode)->root = fs_info->tree_root;
2178         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2179         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2180         btrfs_insert_inode_hash(inode);
2181 }
2182
2183 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2184 {
2185         fs_info->dev_replace.lock_owner = 0;
2186         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2187         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2188         rwlock_init(&fs_info->dev_replace.lock);
2189         atomic_set(&fs_info->dev_replace.read_locks, 0);
2190         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2191         init_waitqueue_head(&fs_info->replace_wait);
2192         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2193 }
2194
2195 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2196 {
2197         spin_lock_init(&fs_info->qgroup_lock);
2198         mutex_init(&fs_info->qgroup_ioctl_lock);
2199         fs_info->qgroup_tree = RB_ROOT;
2200         fs_info->qgroup_op_tree = RB_ROOT;
2201         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2202         fs_info->qgroup_seq = 1;
2203         fs_info->qgroup_ulist = NULL;
2204         fs_info->qgroup_rescan_running = false;
2205         mutex_init(&fs_info->qgroup_rescan_lock);
2206 }
2207
2208 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2209                 struct btrfs_fs_devices *fs_devices)
2210 {
2211         u32 max_active = fs_info->thread_pool_size;
2212         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2213
2214         fs_info->workers =
2215                 btrfs_alloc_workqueue(fs_info, "worker",
2216                                       flags | WQ_HIGHPRI, max_active, 16);
2217
2218         fs_info->delalloc_workers =
2219                 btrfs_alloc_workqueue(fs_info, "delalloc",
2220                                       flags, max_active, 2);
2221
2222         fs_info->flush_workers =
2223                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2224                                       flags, max_active, 0);
2225
2226         fs_info->caching_workers =
2227                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2228
2229         /*
2230          * a higher idle thresh on the submit workers makes it much more
2231          * likely that bios will be send down in a sane order to the
2232          * devices
2233          */
2234         fs_info->submit_workers =
2235                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2236                                       min_t(u64, fs_devices->num_devices,
2237                                             max_active), 64);
2238
2239         fs_info->fixup_workers =
2240                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2241
2242         /*
2243          * endios are largely parallel and should have a very
2244          * low idle thresh
2245          */
2246         fs_info->endio_workers =
2247                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2248         fs_info->endio_meta_workers =
2249                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2250                                       max_active, 4);
2251         fs_info->endio_meta_write_workers =
2252                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2253                                       max_active, 2);
2254         fs_info->endio_raid56_workers =
2255                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2256                                       max_active, 4);
2257         fs_info->endio_repair_workers =
2258                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2259         fs_info->rmw_workers =
2260                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2261         fs_info->endio_write_workers =
2262                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2263                                       max_active, 2);
2264         fs_info->endio_freespace_worker =
2265                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2266                                       max_active, 0);
2267         fs_info->delayed_workers =
2268                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2269                                       max_active, 0);
2270         fs_info->readahead_workers =
2271                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2272                                       max_active, 2);
2273         fs_info->qgroup_rescan_workers =
2274                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2275         fs_info->extent_workers =
2276                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2277                                       min_t(u64, fs_devices->num_devices,
2278                                             max_active), 8);
2279
2280         if (!(fs_info->workers && fs_info->delalloc_workers &&
2281               fs_info->submit_workers && fs_info->flush_workers &&
2282               fs_info->endio_workers && fs_info->endio_meta_workers &&
2283               fs_info->endio_meta_write_workers &&
2284               fs_info->endio_repair_workers &&
2285               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2286               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2287               fs_info->caching_workers && fs_info->readahead_workers &&
2288               fs_info->fixup_workers && fs_info->delayed_workers &&
2289               fs_info->extent_workers &&
2290               fs_info->qgroup_rescan_workers)) {
2291                 return -ENOMEM;
2292         }
2293
2294         return 0;
2295 }
2296
2297 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2298                             struct btrfs_fs_devices *fs_devices)
2299 {
2300         int ret;
2301         struct btrfs_root *log_tree_root;
2302         struct btrfs_super_block *disk_super = fs_info->super_copy;
2303         u64 bytenr = btrfs_super_log_root(disk_super);
2304         int level = btrfs_super_log_root_level(disk_super);
2305
2306         if (fs_devices->rw_devices == 0) {
2307                 btrfs_warn(fs_info, "log replay required on RO media");
2308                 return -EIO;
2309         }
2310
2311         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2312         if (!log_tree_root)
2313                 return -ENOMEM;
2314
2315         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2316
2317         log_tree_root->node = read_tree_block(fs_info, bytenr,
2318                                               fs_info->generation + 1,
2319                                               level, NULL);
2320         if (IS_ERR(log_tree_root->node)) {
2321                 btrfs_warn(fs_info, "failed to read log tree");
2322                 ret = PTR_ERR(log_tree_root->node);
2323                 kfree(log_tree_root);
2324                 return ret;
2325         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2326                 btrfs_err(fs_info, "failed to read log tree");
2327                 free_extent_buffer(log_tree_root->node);
2328                 kfree(log_tree_root);
2329                 return -EIO;
2330         }
2331         /* returns with log_tree_root freed on success */
2332         ret = btrfs_recover_log_trees(log_tree_root);
2333         if (ret) {
2334                 btrfs_handle_fs_error(fs_info, ret,
2335                                       "Failed to recover log tree");
2336                 free_extent_buffer(log_tree_root->node);
2337                 kfree(log_tree_root);
2338                 return ret;
2339         }
2340
2341         if (sb_rdonly(fs_info->sb)) {
2342                 ret = btrfs_commit_super(fs_info);
2343                 if (ret)
2344                         return ret;
2345         }
2346
2347         return 0;
2348 }
2349
2350 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2351 {
2352         struct btrfs_root *tree_root = fs_info->tree_root;
2353         struct btrfs_root *root;
2354         struct btrfs_key location;
2355         int ret;
2356
2357         BUG_ON(!fs_info->tree_root);
2358
2359         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2360         location.type = BTRFS_ROOT_ITEM_KEY;
2361         location.offset = 0;
2362
2363         root = btrfs_read_tree_root(tree_root, &location);
2364         if (IS_ERR(root)) {
2365                 ret = PTR_ERR(root);
2366                 goto out;
2367         }
2368         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369         fs_info->extent_root = root;
2370
2371         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2372         root = btrfs_read_tree_root(tree_root, &location);
2373         if (IS_ERR(root)) {
2374                 ret = PTR_ERR(root);
2375                 goto out;
2376         }
2377         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378         fs_info->dev_root = root;
2379         btrfs_init_devices_late(fs_info);
2380
2381         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2382         root = btrfs_read_tree_root(tree_root, &location);
2383         if (IS_ERR(root)) {
2384                 ret = PTR_ERR(root);
2385                 goto out;
2386         }
2387         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2388         fs_info->csum_root = root;
2389
2390         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2391         root = btrfs_read_tree_root(tree_root, &location);
2392         if (!IS_ERR(root)) {
2393                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2394                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2395                 fs_info->quota_root = root;
2396         }
2397
2398         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2399         root = btrfs_read_tree_root(tree_root, &location);
2400         if (IS_ERR(root)) {
2401                 ret = PTR_ERR(root);
2402                 if (ret != -ENOENT)
2403                         goto out;
2404         } else {
2405                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2406                 fs_info->uuid_root = root;
2407         }
2408
2409         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2410                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2411                 root = btrfs_read_tree_root(tree_root, &location);
2412                 if (IS_ERR(root)) {
2413                         ret = PTR_ERR(root);
2414                         goto out;
2415                 }
2416                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2417                 fs_info->free_space_root = root;
2418         }
2419
2420         return 0;
2421 out:
2422         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2423                    location.objectid, ret);
2424         return ret;
2425 }
2426
2427 /*
2428  * Real super block validation
2429  * NOTE: super csum type and incompat features will not be checked here.
2430  *
2431  * @sb:         super block to check
2432  * @mirror_num: the super block number to check its bytenr:
2433  *              0       the primary (1st) sb
2434  *              1, 2    2nd and 3rd backup copy
2435  *             -1       skip bytenr check
2436  */
2437 static int validate_super(struct btrfs_fs_info *fs_info,
2438                             struct btrfs_super_block *sb, int mirror_num)
2439 {
2440         u64 nodesize = btrfs_super_nodesize(sb);
2441         u64 sectorsize = btrfs_super_sectorsize(sb);
2442         int ret = 0;
2443
2444         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2445                 btrfs_err(fs_info, "no valid FS found");
2446                 ret = -EINVAL;
2447         }
2448         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2449                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2450                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2451                 ret = -EINVAL;
2452         }
2453         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2454                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2455                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2456                 ret = -EINVAL;
2457         }
2458         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2459                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2460                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2461                 ret = -EINVAL;
2462         }
2463         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2464                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2465                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2466                 ret = -EINVAL;
2467         }
2468
2469         /*
2470          * Check sectorsize and nodesize first, other check will need it.
2471          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2472          */
2473         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2474             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2475                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2476                 ret = -EINVAL;
2477         }
2478         /* Only PAGE SIZE is supported yet */
2479         if (sectorsize != PAGE_SIZE) {
2480                 btrfs_err(fs_info,
2481                         "sectorsize %llu not supported yet, only support %lu",
2482                         sectorsize, PAGE_SIZE);
2483                 ret = -EINVAL;
2484         }
2485         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2486             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2487                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2488                 ret = -EINVAL;
2489         }
2490         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2491                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2492                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2493                 ret = -EINVAL;
2494         }
2495
2496         /* Root alignment check */
2497         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2498                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2499                            btrfs_super_root(sb));
2500                 ret = -EINVAL;
2501         }
2502         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2503                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2504                            btrfs_super_chunk_root(sb));
2505                 ret = -EINVAL;
2506         }
2507         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2508                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2509                            btrfs_super_log_root(sb));
2510                 ret = -EINVAL;
2511         }
2512
2513         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2514                 btrfs_err(fs_info,
2515                            "dev_item UUID does not match fsid: %pU != %pU",
2516                            fs_info->fsid, sb->dev_item.fsid);
2517                 ret = -EINVAL;
2518         }
2519
2520         /*
2521          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2522          * done later
2523          */
2524         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2525                 btrfs_err(fs_info, "bytes_used is too small %llu",
2526                           btrfs_super_bytes_used(sb));
2527                 ret = -EINVAL;
2528         }
2529         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2530                 btrfs_err(fs_info, "invalid stripesize %u",
2531                           btrfs_super_stripesize(sb));
2532                 ret = -EINVAL;
2533         }
2534         if (btrfs_super_num_devices(sb) > (1UL << 31))
2535                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2536                            btrfs_super_num_devices(sb));
2537         if (btrfs_super_num_devices(sb) == 0) {
2538                 btrfs_err(fs_info, "number of devices is 0");
2539                 ret = -EINVAL;
2540         }
2541
2542         if (mirror_num >= 0 &&
2543             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2544                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2545                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2546                 ret = -EINVAL;
2547         }
2548
2549         /*
2550          * Obvious sys_chunk_array corruptions, it must hold at least one key
2551          * and one chunk
2552          */
2553         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2554                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2555                           btrfs_super_sys_array_size(sb),
2556                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2557                 ret = -EINVAL;
2558         }
2559         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2560                         + sizeof(struct btrfs_chunk)) {
2561                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2562                           btrfs_super_sys_array_size(sb),
2563                           sizeof(struct btrfs_disk_key)
2564                           + sizeof(struct btrfs_chunk));
2565                 ret = -EINVAL;
2566         }
2567
2568         /*
2569          * The generation is a global counter, we'll trust it more than the others
2570          * but it's still possible that it's the one that's wrong.
2571          */
2572         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2573                 btrfs_warn(fs_info,
2574                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2575                         btrfs_super_generation(sb),
2576                         btrfs_super_chunk_root_generation(sb));
2577         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2578             && btrfs_super_cache_generation(sb) != (u64)-1)
2579                 btrfs_warn(fs_info,
2580                         "suspicious: generation < cache_generation: %llu < %llu",
2581                         btrfs_super_generation(sb),
2582                         btrfs_super_cache_generation(sb));
2583
2584         return ret;
2585 }
2586
2587 /*
2588  * Validation of super block at mount time.
2589  * Some checks already done early at mount time, like csum type and incompat
2590  * flags will be skipped.
2591  */
2592 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2593 {
2594         return validate_super(fs_info, fs_info->super_copy, 0);
2595 }
2596
2597 /*
2598  * Validation of super block at write time.
2599  * Some checks like bytenr check will be skipped as their values will be
2600  * overwritten soon.
2601  * Extra checks like csum type and incompat flags will be done here.
2602  */
2603 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2604                                       struct btrfs_super_block *sb)
2605 {
2606         int ret;
2607
2608         ret = validate_super(fs_info, sb, -1);
2609         if (ret < 0)
2610                 goto out;
2611         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2612                 ret = -EUCLEAN;
2613                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2614                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2615                 goto out;
2616         }
2617         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2618                 ret = -EUCLEAN;
2619                 btrfs_err(fs_info,
2620                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2621                           btrfs_super_incompat_flags(sb),
2622                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2623                 goto out;
2624         }
2625 out:
2626         if (ret < 0)
2627                 btrfs_err(fs_info,
2628                 "super block corruption detected before writing it to disk");
2629         return ret;
2630 }
2631
2632 int open_ctree(struct super_block *sb,
2633                struct btrfs_fs_devices *fs_devices,
2634                char *options)
2635 {
2636         u32 sectorsize;
2637         u32 nodesize;
2638         u32 stripesize;
2639         u64 generation;
2640         u64 features;
2641         struct btrfs_key location;
2642         struct buffer_head *bh;
2643         struct btrfs_super_block *disk_super;
2644         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2645         struct btrfs_root *tree_root;
2646         struct btrfs_root *chunk_root;
2647         int ret;
2648         int err = -EINVAL;
2649         int num_backups_tried = 0;
2650         int backup_index = 0;
2651         int clear_free_space_tree = 0;
2652         int level;
2653
2654         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2655         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2656         if (!tree_root || !chunk_root) {
2657                 err = -ENOMEM;
2658                 goto fail;
2659         }
2660
2661         ret = init_srcu_struct(&fs_info->subvol_srcu);
2662         if (ret) {
2663                 err = ret;
2664                 goto fail;
2665         }
2666
2667         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2668         if (ret) {
2669                 err = ret;
2670                 goto fail_srcu;
2671         }
2672         fs_info->dirty_metadata_batch = PAGE_SIZE *
2673                                         (1 + ilog2(nr_cpu_ids));
2674
2675         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2676         if (ret) {
2677                 err = ret;
2678                 goto fail_dirty_metadata_bytes;
2679         }
2680
2681         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2682         if (ret) {
2683                 err = ret;
2684                 goto fail_delalloc_bytes;
2685         }
2686
2687         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2688         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2689         INIT_LIST_HEAD(&fs_info->trans_list);
2690         INIT_LIST_HEAD(&fs_info->dead_roots);
2691         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2692         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2693         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2694         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2695         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2696         spin_lock_init(&fs_info->delalloc_root_lock);
2697         spin_lock_init(&fs_info->trans_lock);
2698         spin_lock_init(&fs_info->fs_roots_radix_lock);
2699         spin_lock_init(&fs_info->delayed_iput_lock);
2700         spin_lock_init(&fs_info->defrag_inodes_lock);
2701         spin_lock_init(&fs_info->tree_mod_seq_lock);
2702         spin_lock_init(&fs_info->super_lock);
2703         spin_lock_init(&fs_info->qgroup_op_lock);
2704         spin_lock_init(&fs_info->buffer_lock);
2705         spin_lock_init(&fs_info->unused_bgs_lock);
2706         rwlock_init(&fs_info->tree_mod_log_lock);
2707         mutex_init(&fs_info->unused_bg_unpin_mutex);
2708         mutex_init(&fs_info->delete_unused_bgs_mutex);
2709         mutex_init(&fs_info->reloc_mutex);
2710         mutex_init(&fs_info->delalloc_root_mutex);
2711         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2712         seqlock_init(&fs_info->profiles_lock);
2713
2714         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2715         INIT_LIST_HEAD(&fs_info->space_info);
2716         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2717         INIT_LIST_HEAD(&fs_info->unused_bgs);
2718         btrfs_mapping_init(&fs_info->mapping_tree);
2719         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2720                              BTRFS_BLOCK_RSV_GLOBAL);
2721         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2722         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2723         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2724         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2725                              BTRFS_BLOCK_RSV_DELOPS);
2726         atomic_set(&fs_info->async_delalloc_pages, 0);
2727         atomic_set(&fs_info->defrag_running, 0);
2728         atomic_set(&fs_info->qgroup_op_seq, 0);
2729         atomic_set(&fs_info->reada_works_cnt, 0);
2730         atomic64_set(&fs_info->tree_mod_seq, 0);
2731         fs_info->sb = sb;
2732         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2733         fs_info->metadata_ratio = 0;
2734         fs_info->defrag_inodes = RB_ROOT;
2735         atomic64_set(&fs_info->free_chunk_space, 0);
2736         fs_info->tree_mod_log = RB_ROOT;
2737         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2738         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2739         /* readahead state */
2740         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2741         spin_lock_init(&fs_info->reada_lock);
2742         btrfs_init_ref_verify(fs_info);
2743
2744         fs_info->thread_pool_size = min_t(unsigned long,
2745                                           num_online_cpus() + 2, 8);
2746
2747         INIT_LIST_HEAD(&fs_info->ordered_roots);
2748         spin_lock_init(&fs_info->ordered_root_lock);
2749
2750         fs_info->btree_inode = new_inode(sb);
2751         if (!fs_info->btree_inode) {
2752                 err = -ENOMEM;
2753                 goto fail_bio_counter;
2754         }
2755         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2756
2757         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2758                                         GFP_KERNEL);
2759         if (!fs_info->delayed_root) {
2760                 err = -ENOMEM;
2761                 goto fail_iput;
2762         }
2763         btrfs_init_delayed_root(fs_info->delayed_root);
2764
2765         btrfs_init_scrub(fs_info);
2766 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2767         fs_info->check_integrity_print_mask = 0;
2768 #endif
2769         btrfs_init_balance(fs_info);
2770         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2771
2772         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2773         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2774
2775         btrfs_init_btree_inode(fs_info);
2776
2777         spin_lock_init(&fs_info->block_group_cache_lock);
2778         fs_info->block_group_cache_tree = RB_ROOT;
2779         fs_info->first_logical_byte = (u64)-1;
2780
2781         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2782         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2783         fs_info->pinned_extents = &fs_info->freed_extents[0];
2784         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2785
2786         mutex_init(&fs_info->ordered_operations_mutex);
2787         mutex_init(&fs_info->tree_log_mutex);
2788         mutex_init(&fs_info->chunk_mutex);
2789         mutex_init(&fs_info->transaction_kthread_mutex);
2790         mutex_init(&fs_info->cleaner_mutex);
2791         mutex_init(&fs_info->ro_block_group_mutex);
2792         init_rwsem(&fs_info->commit_root_sem);
2793         init_rwsem(&fs_info->cleanup_work_sem);
2794         init_rwsem(&fs_info->subvol_sem);
2795         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2796
2797         btrfs_init_dev_replace_locks(fs_info);
2798         btrfs_init_qgroup(fs_info);
2799
2800         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2801         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2802
2803         init_waitqueue_head(&fs_info->transaction_throttle);
2804         init_waitqueue_head(&fs_info->transaction_wait);
2805         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2806         init_waitqueue_head(&fs_info->async_submit_wait);
2807
2808         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2809
2810         /* Usable values until the real ones are cached from the superblock */
2811         fs_info->nodesize = 4096;
2812         fs_info->sectorsize = 4096;
2813         fs_info->stripesize = 4096;
2814
2815         ret = btrfs_alloc_stripe_hash_table(fs_info);
2816         if (ret) {
2817                 err = ret;
2818                 goto fail_alloc;
2819         }
2820
2821         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2822
2823         invalidate_bdev(fs_devices->latest_bdev);
2824
2825         /*
2826          * Read super block and check the signature bytes only
2827          */
2828         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2829         if (IS_ERR(bh)) {
2830                 err = PTR_ERR(bh);
2831                 goto fail_alloc;
2832         }
2833
2834         /*
2835          * We want to check superblock checksum, the type is stored inside.
2836          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2837          */
2838         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2839                 btrfs_err(fs_info, "superblock checksum mismatch");
2840                 err = -EINVAL;
2841                 brelse(bh);
2842                 goto fail_alloc;
2843         }
2844
2845         /*
2846          * super_copy is zeroed at allocation time and we never touch the
2847          * following bytes up to INFO_SIZE, the checksum is calculated from
2848          * the whole block of INFO_SIZE
2849          */
2850         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2851         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2852                sizeof(*fs_info->super_for_commit));
2853         brelse(bh);
2854
2855         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2856
2857         ret = btrfs_validate_mount_super(fs_info);
2858         if (ret) {
2859                 btrfs_err(fs_info, "superblock contains fatal errors");
2860                 err = -EINVAL;
2861                 goto fail_alloc;
2862         }
2863
2864         disk_super = fs_info->super_copy;
2865         if (!btrfs_super_root(disk_super))
2866                 goto fail_alloc;
2867
2868         /* check FS state, whether FS is broken. */
2869         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2870                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2871
2872         /*
2873          * run through our array of backup supers and setup
2874          * our ring pointer to the oldest one
2875          */
2876         generation = btrfs_super_generation(disk_super);
2877         find_oldest_super_backup(fs_info, generation);
2878
2879         /*
2880          * In the long term, we'll store the compression type in the super
2881          * block, and it'll be used for per file compression control.
2882          */
2883         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2884
2885         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2886         if (ret) {
2887                 err = ret;
2888                 goto fail_alloc;
2889         }
2890
2891         features = btrfs_super_incompat_flags(disk_super) &
2892                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2893         if (features) {
2894                 btrfs_err(fs_info,
2895                     "cannot mount because of unsupported optional features (%llx)",
2896                     features);
2897                 err = -EINVAL;
2898                 goto fail_alloc;
2899         }
2900
2901         features = btrfs_super_incompat_flags(disk_super);
2902         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2903         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2904                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2905         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2906                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2907
2908         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2909                 btrfs_info(fs_info, "has skinny extents");
2910
2911         /*
2912          * flag our filesystem as having big metadata blocks if
2913          * they are bigger than the page size
2914          */
2915         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2916                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2917                         btrfs_info(fs_info,
2918                                 "flagging fs with big metadata feature");
2919                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2920         }
2921
2922         nodesize = btrfs_super_nodesize(disk_super);
2923         sectorsize = btrfs_super_sectorsize(disk_super);
2924         stripesize = sectorsize;
2925         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2926         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2927
2928         /* Cache block sizes */
2929         fs_info->nodesize = nodesize;
2930         fs_info->sectorsize = sectorsize;
2931         fs_info->stripesize = stripesize;
2932
2933         /*
2934          * mixed block groups end up with duplicate but slightly offset
2935          * extent buffers for the same range.  It leads to corruptions
2936          */
2937         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2938             (sectorsize != nodesize)) {
2939                 btrfs_err(fs_info,
2940 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2941                         nodesize, sectorsize);
2942                 goto fail_alloc;
2943         }
2944
2945         /*
2946          * Needn't use the lock because there is no other task which will
2947          * update the flag.
2948          */
2949         btrfs_set_super_incompat_flags(disk_super, features);
2950
2951         features = btrfs_super_compat_ro_flags(disk_super) &
2952                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2953         if (!sb_rdonly(sb) && features) {
2954                 btrfs_err(fs_info,
2955         "cannot mount read-write because of unsupported optional features (%llx)",
2956                        features);
2957                 err = -EINVAL;
2958                 goto fail_alloc;
2959         }
2960
2961         ret = btrfs_init_workqueues(fs_info, fs_devices);
2962         if (ret) {
2963                 err = ret;
2964                 goto fail_sb_buffer;
2965         }
2966
2967         sb->s_bdi->congested_fn = btrfs_congested_fn;
2968         sb->s_bdi->congested_data = fs_info;
2969         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2970         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2971         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2972         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2973
2974         sb->s_blocksize = sectorsize;
2975         sb->s_blocksize_bits = blksize_bits(sectorsize);
2976         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2977
2978         mutex_lock(&fs_info->chunk_mutex);
2979         ret = btrfs_read_sys_array(fs_info);
2980         mutex_unlock(&fs_info->chunk_mutex);
2981         if (ret) {
2982                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2983                 goto fail_sb_buffer;
2984         }
2985
2986         generation = btrfs_super_chunk_root_generation(disk_super);
2987         level = btrfs_super_chunk_root_level(disk_super);
2988
2989         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2990
2991         chunk_root->node = read_tree_block(fs_info,
2992                                            btrfs_super_chunk_root(disk_super),
2993                                            generation, level, NULL);
2994         if (IS_ERR(chunk_root->node) ||
2995             !extent_buffer_uptodate(chunk_root->node)) {
2996                 btrfs_err(fs_info, "failed to read chunk root");
2997                 if (!IS_ERR(chunk_root->node))
2998                         free_extent_buffer(chunk_root->node);
2999                 chunk_root->node = NULL;
3000                 goto fail_tree_roots;
3001         }
3002         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3003         chunk_root->commit_root = btrfs_root_node(chunk_root);
3004
3005         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3006            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3007
3008         ret = btrfs_read_chunk_tree(fs_info);
3009         if (ret) {
3010                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3011                 goto fail_tree_roots;
3012         }
3013
3014         /*
3015          * Keep the devid that is marked to be the target device for the
3016          * device replace procedure
3017          */
3018         btrfs_free_extra_devids(fs_devices, 0);
3019
3020         if (!fs_devices->latest_bdev) {
3021                 btrfs_err(fs_info, "failed to read devices");
3022                 goto fail_tree_roots;
3023         }
3024
3025 retry_root_backup:
3026         generation = btrfs_super_generation(disk_super);
3027         level = btrfs_super_root_level(disk_super);
3028
3029         tree_root->node = read_tree_block(fs_info,
3030                                           btrfs_super_root(disk_super),
3031                                           generation, level, NULL);
3032         if (IS_ERR(tree_root->node) ||
3033             !extent_buffer_uptodate(tree_root->node)) {
3034                 btrfs_warn(fs_info, "failed to read tree root");
3035                 if (!IS_ERR(tree_root->node))
3036                         free_extent_buffer(tree_root->node);
3037                 tree_root->node = NULL;
3038                 goto recovery_tree_root;
3039         }
3040
3041         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3042         tree_root->commit_root = btrfs_root_node(tree_root);
3043         btrfs_set_root_refs(&tree_root->root_item, 1);
3044
3045         mutex_lock(&tree_root->objectid_mutex);
3046         ret = btrfs_find_highest_objectid(tree_root,
3047                                         &tree_root->highest_objectid);
3048         if (ret) {
3049                 mutex_unlock(&tree_root->objectid_mutex);
3050                 goto recovery_tree_root;
3051         }
3052
3053         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3054
3055         mutex_unlock(&tree_root->objectid_mutex);
3056
3057         ret = btrfs_read_roots(fs_info);
3058         if (ret)
3059                 goto recovery_tree_root;
3060
3061         fs_info->generation = generation;
3062         fs_info->last_trans_committed = generation;
3063
3064         ret = btrfs_recover_balance(fs_info);
3065         if (ret) {
3066                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3067                 goto fail_block_groups;
3068         }
3069
3070         ret = btrfs_init_dev_stats(fs_info);
3071         if (ret) {
3072                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3073                 goto fail_block_groups;
3074         }
3075
3076         ret = btrfs_init_dev_replace(fs_info);
3077         if (ret) {
3078                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3079                 goto fail_block_groups;
3080         }
3081
3082         btrfs_free_extra_devids(fs_devices, 1);
3083
3084         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3085         if (ret) {
3086                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3087                                 ret);
3088                 goto fail_block_groups;
3089         }
3090
3091         ret = btrfs_sysfs_add_device(fs_devices);
3092         if (ret) {
3093                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3094                                 ret);
3095                 goto fail_fsdev_sysfs;
3096         }
3097
3098         ret = btrfs_sysfs_add_mounted(fs_info);
3099         if (ret) {
3100                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3101                 goto fail_fsdev_sysfs;
3102         }
3103
3104         ret = btrfs_init_space_info(fs_info);
3105         if (ret) {
3106                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3107                 goto fail_sysfs;
3108         }
3109
3110         ret = btrfs_read_block_groups(fs_info);
3111         if (ret) {
3112                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3113                 goto fail_sysfs;
3114         }
3115
3116         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3117                 btrfs_warn(fs_info,
3118                 "writeable mount is not allowed due to too many missing devices");
3119                 goto fail_sysfs;
3120         }
3121
3122         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3123                                                "btrfs-cleaner");
3124         if (IS_ERR(fs_info->cleaner_kthread))
3125                 goto fail_sysfs;
3126
3127         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3128                                                    tree_root,
3129                                                    "btrfs-transaction");
3130         if (IS_ERR(fs_info->transaction_kthread))
3131                 goto fail_cleaner;
3132
3133         if (!btrfs_test_opt(fs_info, NOSSD) &&
3134             !fs_info->fs_devices->rotating) {
3135                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3136         }
3137
3138         /*
3139          * Mount does not set all options immediately, we can do it now and do
3140          * not have to wait for transaction commit
3141          */
3142         btrfs_apply_pending_changes(fs_info);
3143
3144 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3145         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3146                 ret = btrfsic_mount(fs_info, fs_devices,
3147                                     btrfs_test_opt(fs_info,
3148                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3149                                     1 : 0,
3150                                     fs_info->check_integrity_print_mask);
3151                 if (ret)
3152                         btrfs_warn(fs_info,
3153                                 "failed to initialize integrity check module: %d",
3154                                 ret);
3155         }
3156 #endif
3157         ret = btrfs_read_qgroup_config(fs_info);
3158         if (ret)
3159                 goto fail_trans_kthread;
3160
3161         if (btrfs_build_ref_tree(fs_info))
3162                 btrfs_err(fs_info, "couldn't build ref tree");
3163
3164         /* do not make disk changes in broken FS or nologreplay is given */
3165         if (btrfs_super_log_root(disk_super) != 0 &&
3166             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3167                 ret = btrfs_replay_log(fs_info, fs_devices);
3168                 if (ret) {
3169                         err = ret;
3170                         goto fail_qgroup;
3171                 }
3172         }
3173
3174         ret = btrfs_find_orphan_roots(fs_info);
3175         if (ret)
3176                 goto fail_qgroup;
3177
3178         if (!sb_rdonly(sb)) {
3179                 ret = btrfs_cleanup_fs_roots(fs_info);
3180                 if (ret)
3181                         goto fail_qgroup;
3182
3183                 mutex_lock(&fs_info->cleaner_mutex);
3184                 ret = btrfs_recover_relocation(tree_root);
3185                 mutex_unlock(&fs_info->cleaner_mutex);
3186                 if (ret < 0) {
3187                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3188                                         ret);
3189                         err = -EINVAL;
3190                         goto fail_qgroup;
3191                 }
3192         }
3193
3194         location.objectid = BTRFS_FS_TREE_OBJECTID;
3195         location.type = BTRFS_ROOT_ITEM_KEY;
3196         location.offset = 0;
3197
3198         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3199         if (IS_ERR(fs_info->fs_root)) {
3200                 err = PTR_ERR(fs_info->fs_root);
3201                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3202                 goto fail_qgroup;
3203         }
3204
3205         if (sb_rdonly(sb))
3206                 return 0;
3207
3208         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3209             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3210                 clear_free_space_tree = 1;
3211         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3212                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3213                 btrfs_warn(fs_info, "free space tree is invalid");
3214                 clear_free_space_tree = 1;
3215         }
3216
3217         if (clear_free_space_tree) {
3218                 btrfs_info(fs_info, "clearing free space tree");
3219                 ret = btrfs_clear_free_space_tree(fs_info);
3220                 if (ret) {
3221                         btrfs_warn(fs_info,
3222                                    "failed to clear free space tree: %d", ret);
3223                         close_ctree(fs_info);
3224                         return ret;
3225                 }
3226         }
3227
3228         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3229             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3230                 btrfs_info(fs_info, "creating free space tree");
3231                 ret = btrfs_create_free_space_tree(fs_info);
3232                 if (ret) {
3233                         btrfs_warn(fs_info,
3234                                 "failed to create free space tree: %d", ret);
3235                         close_ctree(fs_info);
3236                         return ret;
3237                 }
3238         }
3239
3240         down_read(&fs_info->cleanup_work_sem);
3241         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3242             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3243                 up_read(&fs_info->cleanup_work_sem);
3244                 close_ctree(fs_info);
3245                 return ret;
3246         }
3247         up_read(&fs_info->cleanup_work_sem);
3248
3249         ret = btrfs_resume_balance_async(fs_info);
3250         if (ret) {
3251                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3252                 close_ctree(fs_info);
3253                 return ret;
3254         }
3255
3256         ret = btrfs_resume_dev_replace_async(fs_info);
3257         if (ret) {
3258                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);