fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <linux/sched/mm.h>
  21 #include <asm/unaligned.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42
  43 #ifdef CONFIG_X86
  44 #include <asm/cpufeature.h>
  45 #endif
  46
  47 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  48                                  BTRFS_HEADER_FLAG_RELOC |\
  49                                  BTRFS_SUPER_FLAG_ERROR |\
  50                                  BTRFS_SUPER_FLAG_SEEDING |\
  51                                  BTRFS_SUPER_FLAG_METADUMP |\
  52                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  53
  54 static const struct extent_io_ops btree_extent_io_ops;
  55 static void end_workqueue_fn(struct btrfs_work *work);
  56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  58                                       struct btrfs_fs_info *fs_info);
  59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  61                                         struct extent_io_tree *dirty_pages,
  62                                         int mark);
  63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  64                                        struct extent_io_tree *pinned_extents);
  65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  67
  68 /*
  69  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  70  * is complete.  This is used during reads to verify checksums, and it is used
  71  * by writes to insert metadata for new file extents after IO is complete.
  72  */
  73 struct btrfs_end_io_wq {
  74         struct bio *bio;
  75         bio_end_io_t *end_io;
  76         void *private;
  77         struct btrfs_fs_info *info;
  78         blk_status_t status;
  79         enum btrfs_wq_endio_type metadata;
  80         struct btrfs_work work;
  81 };
  82
  83 static struct kmem_cache *btrfs_end_io_wq_cache;
  84
  85 int __init btrfs_end_io_wq_init(void)
  86 {
  87         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  88                                         sizeof(struct btrfs_end_io_wq),
  89                                         0,
  90                                         SLAB_MEM_SPREAD,
  91                                         NULL);
  92         if (!btrfs_end_io_wq_cache)
  93                 return -ENOMEM;
  94         return 0;
  95 }
  96
  97 void __cold btrfs_end_io_wq_exit(void)
  98 {
  99         kmem_cache_destroy(btrfs_end_io_wq_cache);
 100 }
 101
 102 /*
 103  * async submit bios are used to offload expensive checksumming
 104  * onto the worker threads.  They checksum file and metadata bios
 105  * just before they are sent down the IO stack.
 106  */
 107 struct async_submit_bio {
 108         void *private_data;
 109         struct bio *bio;
 110         extent_submit_bio_start_t *submit_bio_start;
 111         int mirror_num;
 112         /*
 113          * bio_offset is optional, can be used if the pages in the bio
 114          * can't tell us where in the file the bio should go
 115          */
 116         u64 bio_offset;
 117         struct btrfs_work work;
 118         blk_status_t status;
 119 };
 120
 121 /*
 122  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 123  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 124  * the level the eb occupies in the tree.
 125  *
 126  * Different roots are used for different purposes and may nest inside each
 127  * other and they require separate keysets.  As lockdep keys should be
 128  * static, assign keysets according to the purpose of the root as indicated
 129  * by btrfs_root->objectid.  This ensures that all special purpose roots
 130  * have separate keysets.
 131  *
 132  * Lock-nesting across peer nodes is always done with the immediate parent
 133  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 134  * subclass to avoid triggering lockdep warning in such cases.
 135  *
 136  * The key is set by the readpage_end_io_hook after the buffer has passed
 137  * csum validation but before the pages are unlocked.  It is also set by
 138  * btrfs_init_new_buffer on freshly allocated blocks.
 139  *
 140  * We also add a check to make sure the highest level of the tree is the
 141  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 142  * needs update as well.
 143  */
 144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 145 # if BTRFS_MAX_LEVEL != 8
 146 #  error
 147 # endif
 148
 149 static struct btrfs_lockdep_keyset {
 150         u64                     id;             /* root objectid */
 151         const char              *name_stem;     /* lock name stem */
 152         char                    names[BTRFS_MAX_LEVEL + 1][20];
 153         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 154 } btrfs_lockdep_keysets[] = {
 155         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 156         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 157         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 158         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 159         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 160         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 161         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 162         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 163         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 164         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 165         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 166         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 167         { .id = 0,                              .name_stem = "tree"     },
 168 };
 169
 170 void __init btrfs_init_lockdep(void)
 171 {
 172         int i, j;
 173
 174         /* initialize lockdep class names */
 175         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 176                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 177
 178                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 179                         snprintf(ks->names[j], sizeof(ks->names[j]),
 180                                  "btrfs-%s-%02d", ks->name_stem, j);
 181         }
 182 }
 183
 184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 185                                     int level)
 186 {
 187         struct btrfs_lockdep_keyset *ks;
 188
 189         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 190
 191         /* find the matching keyset, id 0 is the default entry */
 192         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 193                 if (ks->id == objectid)
 194                         break;
 195
 196         lockdep_set_class_and_name(&eb->lock,
 197                                    &ks->keys[level], ks->names[level]);
 198 }
 199
 200 #endif
 201
 202 /*
 203  * extents on the btree inode are pretty simple, there's one extent
 204  * that covers the entire device
 205  */
 206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 207                 struct page *page, size_t pg_offset, u64 start, u64 len,
 208                 int create)
 209 {
 210         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 211         struct extent_map_tree *em_tree = &inode->extent_tree;
 212         struct extent_map *em;
 213         int ret;
 214
 215         read_lock(&em_tree->lock);
 216         em = lookup_extent_mapping(em_tree, start, len);
 217         if (em) {
 218                 em->bdev = fs_info->fs_devices->latest_bdev;
 219                 read_unlock(&em_tree->lock);
 220                 goto out;
 221         }
 222         read_unlock(&em_tree->lock);
 223
 224         em = alloc_extent_map();
 225         if (!em) {
 226                 em = ERR_PTR(-ENOMEM);
 227                 goto out;
 228         }
 229         em->start = 0;
 230         em->len = (u64)-1;
 231         em->block_len = (u64)-1;
 232         em->block_start = 0;
 233         em->bdev = fs_info->fs_devices->latest_bdev;
 234
 235         write_lock(&em_tree->lock);
 236         ret = add_extent_mapping(em_tree, em, 0);
 237         if (ret == -EEXIST) {
 238                 free_extent_map(em);
 239                 em = lookup_extent_mapping(em_tree, start, len);
 240                 if (!em)
 241                         em = ERR_PTR(-EIO);
 242         } else if (ret) {
 243                 free_extent_map(em);
 244                 em = ERR_PTR(ret);
 245         }
 246         write_unlock(&em_tree->lock);
 247
 248 out:
 249         return em;
 250 }
 251
 252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 253 {
 254         return crc32c(seed, data, len);
 255 }
 256
 257 void btrfs_csum_final(u32 crc, u8 *result)
 258 {
 259         put_unaligned_le32(~crc, result);
 260 }
 261
 262 /*
 263  * compute the csum for a btree block, and either verify it or write it
 264  * into the csum field of the block.
 265  */
 266 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 267                            struct extent_buffer *buf,
 268                            int verify)
 269 {
 270         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 271         char result[BTRFS_CSUM_SIZE];
 272         unsigned long len;
 273         unsigned long cur_len;
 274         unsigned long offset = BTRFS_CSUM_SIZE;
 275         char *kaddr;
 276         unsigned long map_start;
 277         unsigned long map_len;
 278         int err;
 279         u32 crc = ~(u32)0;
 280
 281         len = buf->len - offset;
 282         while (len > 0) {
 283                 err = map_private_extent_buffer(buf, offset, 32,
 284                                         &kaddr, &map_start, &map_len);
 285                 if (err)
 286                         return err;
 287                 cur_len = min(len, map_len - (offset - map_start));
 288                 crc = btrfs_csum_data(kaddr + offset - map_start,
 289                                       crc, cur_len);
 290                 len -= cur_len;
 291                 offset += cur_len;
 292         }
 293         memset(result, 0, BTRFS_CSUM_SIZE);
 294
 295         btrfs_csum_final(crc, result);
 296
 297         if (verify) {
 298                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 299                         u32 val;
 300                         u32 found = 0;
 301                         memcpy(&found, result, csum_size);
 302
 303                         read_extent_buffer(buf, &val, 0, csum_size);
 304                         btrfs_warn_rl(fs_info,
 305                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 306                                 fs_info->sb->s_id, buf->start,
 307                                 val, found, btrfs_header_level(buf));
 308                         return -EUCLEAN;
 309                 }
 310         } else {
 311                 write_extent_buffer(buf, result, 0, csum_size);
 312         }
 313
 314         return 0;
 315 }
 316
 317 /*
 318  * we can't consider a given block up to date unless the transid of the
 319  * block matches the transid in the parent node's pointer.  This is how we
 320  * detect blocks that either didn't get written at all or got written
 321  * in the wrong place.
 322  */
 323 static int verify_parent_transid(struct extent_io_tree *io_tree,
 324                                  struct extent_buffer *eb, u64 parent_transid,
 325                                  int atomic)
 326 {
 327         struct extent_state *cached_state = NULL;
 328         int ret;
 329         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 330
 331         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 332                 return 0;
 333
 334         if (atomic)
 335                 return -EAGAIN;
 336
 337         if (need_lock) {
 338                 btrfs_tree_read_lock(eb);
 339                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 340         }
 341
 342         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 343                          &cached_state);
 344         if (extent_buffer_uptodate(eb) &&
 345             btrfs_header_generation(eb) == parent_transid) {
 346                 ret = 0;
 347                 goto out;
 348         }
 349         btrfs_err_rl(eb->fs_info,
 350                 "parent transid verify failed on %llu wanted %llu found %llu",
 351                         eb->start,
 352                         parent_transid, btrfs_header_generation(eb));
 353         ret = 1;
 354
 355         /*
 356          * Things reading via commit roots that don't have normal protection,
 357          * like send, can have a really old block in cache that may point at a
 358          * block that has been freed and re-allocated.  So don't clear uptodate
 359          * if we find an eb that is under IO (dirty/writeback) because we could
 360          * end up reading in the stale data and then writing it back out and
 361          * making everybody very sad.
 362          */
 363         if (!extent_buffer_under_io(eb))
 364                 clear_extent_buffer_uptodate(eb);
 365 out:
 366         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 367                              &cached_state);
 368         if (need_lock)
 369                 btrfs_tree_read_unlock_blocking(eb);
 370         return ret;
 371 }
 372
 373 /*
 374  * Return 0 if the superblock checksum type matches the checksum value of that
 375  * algorithm. Pass the raw disk superblock data.
 376  */
 377 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 378                                   char *raw_disk_sb)
 379 {
 380         struct btrfs_super_block *disk_sb =
 381                 (struct btrfs_super_block *)raw_disk_sb;
 382         u16 csum_type = btrfs_super_csum_type(disk_sb);
 383         int ret = 0;
 384
 385         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 386                 u32 crc = ~(u32)0;
 387                 char result[sizeof(crc)];
 388
 389                 /*
 390                  * The super_block structure does not span the whole
 391                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 392                  * is filled with zeros and is included in the checksum.
 393                  */
 394                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 395                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 396                 btrfs_csum_final(crc, result);
 397
 398                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 399                         ret = 1;
 400         }
 401
 402         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 403                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 404                                 csum_type);
 405                 ret = 1;
 406         }
 407
 408         return ret;
 409 }
 410
 411 int btrfs_verify_level_key(struct btrfs_fs_info *fs_info,
 412                            struct extent_buffer *eb, int level,
 413                            struct btrfs_key *first_key, u64 parent_transid)
 414 {
 415         int found_level;
 416         struct btrfs_key found_key;
 417         int ret;
 418
 419         found_level = btrfs_header_level(eb);
 420         if (found_level != level) {
 421 #ifdef CONFIG_BTRFS_DEBUG
 422                 WARN_ON(1);
 423                 btrfs_err(fs_info,
 424 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 425                           eb->start, level, found_level);
 426 #endif
 427                 return -EIO;
 428         }
 429
 430         if (!first_key)
 431                 return 0;
 432
 433         /*
 434          * For live tree block (new tree blocks in current transaction),
 435          * we need proper lock context to avoid race, which is impossible here.
 436          * So we only checks tree blocks which is read from disk, whose
 437          * generation <= fs_info->last_trans_committed.
 438          */
 439         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 440                 return 0;
 441
 442         /* We have @first_key, so this @eb must have at least one item */
 443         if (btrfs_header_nritems(eb) == 0) {
 444                 btrfs_err(fs_info,
 445                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 446                           eb->start);
 447                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 448                 return -EUCLEAN;
 449         }
 450
 451         if (found_level)
 452                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 453         else
 454                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 455         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 456
 457 #ifdef CONFIG_BTRFS_DEBUG
 458         if (ret) {
 459                 WARN_ON(1);
 460                 btrfs_err(fs_info,
 461 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 462                           eb->start, parent_transid, first_key->objectid,
 463                           first_key->type, first_key->offset,
 464                           found_key.objectid, found_key.type,
 465                           found_key.offset);
 466         }
 467 #endif
 468         return ret;
 469 }
 470
 471 /*
 472  * helper to read a given tree block, doing retries as required when
 473  * the checksums don't match and we have alternate mirrors to try.
 474  *
 475  * @parent_transid:     expected transid, skip check if 0
 476  * @level:              expected level, mandatory check
 477  * @first_key:          expected key of first slot, skip check if NULL
 478  */
 479 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 480                                           struct extent_buffer *eb,
 481                                           u64 parent_transid, int level,
 482                                           struct btrfs_key *first_key)
 483 {
 484         struct extent_io_tree *io_tree;
 485         int failed = 0;
 486         int ret;
 487         int num_copies = 0;
 488         int mirror_num = 0;
 489         int failed_mirror = 0;
 490
 491         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 492         while (1) {
 493                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 494                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 495                                                mirror_num);
 496                 if (!ret) {
 497                         if (verify_parent_transid(io_tree, eb,
 498                                                    parent_transid, 0))
 499                                 ret = -EIO;
 500                         else if (btrfs_verify_level_key(fs_info, eb, level,
 501                                                 first_key, parent_transid))
 502                                 ret = -EUCLEAN;
 503                         else
 504                                 break;
 505                 }
 506
 507                 num_copies = btrfs_num_copies(fs_info,
 508                                               eb->start, eb->len);
 509                 if (num_copies == 1)
 510                         break;
 511
 512                 if (!failed_mirror) {
 513                         failed = 1;
 514                         failed_mirror = eb->read_mirror;
 515                 }
 516
 517                 mirror_num++;
 518                 if (mirror_num == failed_mirror)
 519                         mirror_num++;
 520
 521                 if (mirror_num > num_copies)
 522                         break;
 523         }
 524
 525         if (failed && !ret && failed_mirror)
 526                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 527
 528         return ret;
 529 }
 530
 531 /*
 532  * checksum a dirty tree block before IO.  This has extra checks to make sure
 533  * we only fill in the checksum field in the first page of a multi-page block
 534  */
 535
 536 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 537 {
 538         u64 start = page_offset(page);
 539         u64 found_start;
 540         struct extent_buffer *eb;
 541
 542         eb = (struct extent_buffer *)page->private;
 543         if (page != eb->pages[0])
 544                 return 0;
 545
 546         found_start = btrfs_header_bytenr(eb);
 547         /*
 548          * Please do not consolidate these warnings into a single if.
 549          * It is useful to know what went wrong.
 550          */
 551         if (WARN_ON(found_start != start))
 552                 return -EUCLEAN;
 553         if (WARN_ON(!PageUptodate(page)))
 554                 return -EUCLEAN;
 555
 556         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 557                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 558
 559         return csum_tree_block(fs_info, eb, 0);
 560 }
 561
 562 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 563                                  struct extent_buffer *eb)
 564 {
 565         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 566         u8 fsid[BTRFS_FSID_SIZE];
 567         int ret = 1;
 568
 569         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 570         while (fs_devices) {
 571                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 572                         ret = 0;
 573                         break;
 574                 }
 575                 fs_devices = fs_devices->seed;
 576         }
 577         return ret;
 578 }
 579
 580 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 581                                       u64 phy_offset, struct page *page,
 582                                       u64 start, u64 end, int mirror)
 583 {
 584         u64 found_start;
 585         int found_level;
 586         struct extent_buffer *eb;
 587         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 588         struct btrfs_fs_info *fs_info = root->fs_info;
 589         int ret = 0;
 590         int reads_done;
 591
 592         if (!page->private)
 593                 goto out;
 594
 595         eb = (struct extent_buffer *)page->private;
 596
 597         /* the pending IO might have been the only thing that kept this buffer
 598          * in memory.  Make sure we have a ref for all this other checks
 599          */
 600         extent_buffer_get(eb);
 601
 602         reads_done = atomic_dec_and_test(&eb->io_pages);
 603         if (!reads_done)
 604                 goto err;
 605
 606         eb->read_mirror = mirror;
 607         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 608                 ret = -EIO;
 609                 goto err;
 610         }
 611
 612         found_start = btrfs_header_bytenr(eb);
 613         if (found_start != eb->start) {
 614                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 615                              eb->start, found_start);
 616                 ret = -EIO;
 617                 goto err;
 618         }
 619         if (check_tree_block_fsid(fs_info, eb)) {
 620                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 621                              eb->start);
 622                 ret = -EIO;
 623                 goto err;
 624         }
 625         found_level = btrfs_header_level(eb);
 626         if (found_level >= BTRFS_MAX_LEVEL) {
 627                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 628                           (int)btrfs_header_level(eb), eb->start);
 629                 ret = -EIO;
 630                 goto err;
 631         }
 632
 633         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 634                                        eb, found_level);
 635
 636         ret = csum_tree_block(fs_info, eb, 1);
 637         if (ret)
 638                 goto err;
 639
 640         /*
 641          * If this is a leaf block and it is corrupt, set the corrupt bit so
 642          * that we don't try and read the other copies of this block, just
 643          * return -EIO.
 644          */
 645         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 646                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 647                 ret = -EIO;
 648         }
 649
 650         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 651                 ret = -EIO;
 652
 653         if (!ret)
 654                 set_extent_buffer_uptodate(eb);
 655 err:
 656         if (reads_done &&
 657             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 658                 btree_readahead_hook(eb, ret);
 659
 660         if (ret) {
 661                 /*
 662                  * our io error hook is going to dec the io pages
 663                  * again, we have to make sure it has something
 664                  * to decrement
 665                  */
 666                 atomic_inc(&eb->io_pages);
 667                 clear_extent_buffer_uptodate(eb);
 668         }
 669         free_extent_buffer(eb);
 670 out:
 671         return ret;
 672 }
 673
 674 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 675 {
 676         struct extent_buffer *eb;
 677
 678         eb = (struct extent_buffer *)page->private;
 679         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 680         eb->read_mirror = failed_mirror;
 681         atomic_dec(&eb->io_pages);
 682         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 683                 btree_readahead_hook(eb, -EIO);
 684         return -EIO;    /* we fixed nothing */
 685 }
 686
 687 static void end_workqueue_bio(struct bio *bio)
 688 {
 689         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 690         struct btrfs_fs_info *fs_info;
 691         struct btrfs_workqueue *wq;
 692         btrfs_work_func_t func;
 693
 694         fs_info = end_io_wq->info;
 695         end_io_wq->status = bio->bi_status;
 696
 697         if (bio_op(bio) == REQ_OP_WRITE) {
 698                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 699                         wq = fs_info->endio_meta_write_workers;
 700                         func = btrfs_endio_meta_write_helper;
 701                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 702                         wq = fs_info->endio_freespace_worker;
 703                         func = btrfs_freespace_write_helper;
 704                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 705                         wq = fs_info->endio_raid56_workers;
 706                         func = btrfs_endio_raid56_helper;
 707                 } else {
 708                         wq = fs_info->endio_write_workers;
 709                         func = btrfs_endio_write_helper;
 710                 }
 711         } else {
 712                 if (unlikely(end_io_wq->metadata ==
 713                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 714                         wq = fs_info->endio_repair_workers;
 715                         func = btrfs_endio_repair_helper;
 716                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 717                         wq = fs_info->endio_raid56_workers;
 718                         func = btrfs_endio_raid56_helper;
 719                 } else if (end_io_wq->metadata) {
 720                         wq = fs_info->endio_meta_workers;
 721                         func = btrfs_endio_meta_helper;
 722                 } else {
 723                         wq = fs_info->endio_workers;
 724                         func = btrfs_endio_helper;
 725                 }
 726         }
 727
 728         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 729         btrfs_queue_work(wq, &end_io_wq->work);
 730 }
 731
 732 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 733                         enum btrfs_wq_endio_type metadata)
 734 {
 735         struct btrfs_end_io_wq *end_io_wq;
 736
 737         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 738         if (!end_io_wq)
 739                 return BLK_STS_RESOURCE;
 740
 741         end_io_wq->private = bio->bi_private;
 742         end_io_wq->end_io = bio->bi_end_io;
 743         end_io_wq->info = info;
 744         end_io_wq->status = 0;
 745         end_io_wq->bio = bio;
 746         end_io_wq->metadata = metadata;
 747
 748         bio->bi_private = end_io_wq;
 749         bio->bi_end_io = end_workqueue_bio;
 750         return 0;
 751 }
 752
 753 static void run_one_async_start(struct btrfs_work *work)
 754 {
 755         struct async_submit_bio *async;
 756         blk_status_t ret;
 757
 758         async = container_of(work, struct  async_submit_bio, work);
 759         ret = async->submit_bio_start(async->private_data, async->bio,
 760                                       async->bio_offset);
 761         if (ret)
 762                 async->status = ret;
 763 }
 764
 765 static void run_one_async_done(struct btrfs_work *work)
 766 {
 767         struct async_submit_bio *async;
 768
 769         async = container_of(work, struct  async_submit_bio, work);
 770
 771         /* If an error occurred we just want to clean up the bio and move on */
 772         if (async->status) {
 773                 async->bio->bi_status = async->status;
 774                 bio_endio(async->bio);
 775                 return;
 776         }
 777
 778         btrfs_submit_bio_done(async->private_data, async->bio, async->mirror_num);
 779 }
 780
 781 static void run_one_async_free(struct btrfs_work *work)
 782 {
 783         struct async_submit_bio *async;
 784
 785         async = container_of(work, struct  async_submit_bio, work);
 786         kfree(async);
 787 }
 788
 789 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 790                                  int mirror_num, unsigned long bio_flags,
 791                                  u64 bio_offset, void *private_data,
 792                                  extent_submit_bio_start_t *submit_bio_start)
 793 {
 794         struct async_submit_bio *async;
 795
 796         async = kmalloc(sizeof(*async), GFP_NOFS);
 797         if (!async)
 798                 return BLK_STS_RESOURCE;
 799
 800         async->private_data = private_data;
 801         async->bio = bio;
 802         async->mirror_num = mirror_num;
 803         async->submit_bio_start = submit_bio_start;
 804
 805         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 806                         run_one_async_done, run_one_async_free);
 807
 808         async->bio_offset = bio_offset;
 809
 810         async->status = 0;
 811
 812         if (op_is_sync(bio->bi_opf))
 813                 btrfs_set_work_high_priority(&async->work);
 814
 815         btrfs_queue_work(fs_info->workers, &async->work);
 816         return 0;
 817 }
 818
 819 static blk_status_t btree_csum_one_bio(struct bio *bio)
 820 {
 821         struct bio_vec *bvec;
 822         struct btrfs_root *root;
 823         int i, ret = 0;
 824
 825         ASSERT(!bio_flagged(bio, BIO_CLONED));
 826         bio_for_each_segment_all(bvec, bio, i) {
 827                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 828                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 829                 if (ret)
 830                         break;
 831         }
 832
 833         return errno_to_blk_status(ret);
 834 }
 835
 836 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 837                                              u64 bio_offset)
 838 {
 839         /*
 840          * when we're called for a write, we're already in the async
 841          * submission context.  Just jump into btrfs_map_bio
 842          */
 843         return btree_csum_one_bio(bio);
 844 }
 845
 846 static int check_async_write(struct btrfs_inode *bi)
 847 {
 848         if (atomic_read(&bi->sync_writers))
 849                 return 0;
 850 #ifdef CONFIG_X86
 851         if (static_cpu_has(X86_FEATURE_XMM4_2))
 852                 return 0;
 853 #endif
 854         return 1;
 855 }
 856
 857 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 858                                           int mirror_num, unsigned long bio_flags,
 859                                           u64 bio_offset)
 860 {
 861         struct inode *inode = private_data;
 862         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 863         int async = check_async_write(BTRFS_I(inode));
 864         blk_status_t ret;
 865
 866         if (bio_op(bio) != REQ_OP_WRITE) {
 867                 /*
 868                  * called for a read, do the setup so that checksum validation
 869                  * can happen in the async kernel threads
 870                  */
 871                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 872                                           BTRFS_WQ_ENDIO_METADATA);
 873                 if (ret)
 874                         goto out_w_error;
 875                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 876         } else if (!async) {
 877                 ret = btree_csum_one_bio(bio);
 878                 if (ret)
 879                         goto out_w_error;
 880                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 881         } else {
 882                 /*
 883                  * kthread helpers are used to submit writes so that
 884                  * checksumming can happen in parallel across all CPUs
 885                  */
 886                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 887                                           bio_offset, private_data,
 888                                           btree_submit_bio_start);
 889         }
 890
 891         if (ret)
 892                 goto out_w_error;
 893         return 0;
 894
 895 out_w_error:
 896         bio->bi_status = ret;
 897         bio_endio(bio);
 898         return ret;
 899 }
 900
 901 #ifdef CONFIG_MIGRATION
 902 static int btree_migratepage(struct address_space *mapping,
 903                         struct page *newpage, struct page *page,
 904                         enum migrate_mode mode)
 905 {
 906         /*
 907          * we can't safely write a btree page from here,
 908          * we haven't done the locking hook
 909          */
 910         if (PageDirty(page))
 911                 return -EAGAIN;
 912         /*
 913          * Buffers may be managed in a filesystem specific way.
 914          * We must have no buffers or drop them.
 915          */
 916         if (page_has_private(page) &&
 917             !try_to_release_page(page, GFP_KERNEL))
 918                 return -EAGAIN;
 919         return migrate_page(mapping, newpage, page, mode);
 920 }
 921 #endif
 922
 923
 924 static int btree_writepages(struct address_space *mapping,
 925                             struct writeback_control *wbc)
 926 {
 927         struct btrfs_fs_info *fs_info;
 928         int ret;
 929
 930         if (wbc->sync_mode == WB_SYNC_NONE) {
 931
 932                 if (wbc->for_kupdate)
 933                         return 0;
 934
 935                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 936                 /* this is a bit racy, but that's ok */
 937                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 938                                              BTRFS_DIRTY_METADATA_THRESH,
 939                                              fs_info->dirty_metadata_batch);
 940                 if (ret < 0)
 941                         return 0;
 942         }
 943         return btree_write_cache_pages(mapping, wbc);
 944 }
 945
 946 static int btree_readpage(struct file *file, struct page *page)
 947 {
 948         struct extent_io_tree *tree;
 949         tree = &BTRFS_I(page->mapping->host)->io_tree;
 950         return extent_read_full_page(tree, page, btree_get_extent, 0);
 951 }
 952
 953 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 954 {
 955         if (PageWriteback(page) || PageDirty(page))
 956                 return 0;
 957
 958         return try_release_extent_buffer(page);
 959 }
 960
 961 static void btree_invalidatepage(struct page *page, unsigned int offset,
 962                                  unsigned int length)
 963 {
 964         struct extent_io_tree *tree;
 965         tree = &BTRFS_I(page->mapping->host)->io_tree;
 966         extent_invalidatepage(tree, page, offset);
 967         btree_releasepage(page, GFP_NOFS);
 968         if (PagePrivate(page)) {
 969                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 970                            "page private not zero on page %llu",
 971                            (unsigned long long)page_offset(page));
 972                 ClearPagePrivate(page);
 973                 set_page_private(page, 0);
 974                 put_page(page);
 975         }
 976 }
 977
 978 static int btree_set_page_dirty(struct page *page)
 979 {
 980 #ifdef DEBUG
 981         struct extent_buffer *eb;
 982
 983         BUG_ON(!PagePrivate(page));
 984         eb = (struct extent_buffer *)page->private;
 985         BUG_ON(!eb);
 986         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 987         BUG_ON(!atomic_read(&eb->refs));
 988         btrfs_assert_tree_locked(eb);
 989 #endif
 990         return __set_page_dirty_nobuffers(page);
 991 }
 992
 993 static const struct address_space_operations btree_aops = {
 994         .readpage       = btree_readpage,
 995         .writepages     = btree_writepages,
 996         .releasepage    = btree_releasepage,
 997         .invalidatepage = btree_invalidatepage,
 998 #ifdef CONFIG_MIGRATION
 999         .migratepage    = btree_migratepage,
1000 #endif
1001         .set_page_dirty = btree_set_page_dirty,
1002 };
1003
1004 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1005 {
1006         struct extent_buffer *buf = NULL;
1007         struct inode *btree_inode = fs_info->btree_inode;
1008         int ret;
1009
1010         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1011         if (IS_ERR(buf))
1012                 return;
1013
1014         ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf,
1015                         WAIT_NONE, 0);
1016         if (ret < 0)
1017                 free_extent_buffer_stale(buf);
1018         else
1019                 free_extent_buffer(buf);
1020 }
1021
1022 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1023                          int mirror_num, struct extent_buffer **eb)
1024 {
1025         struct extent_buffer *buf = NULL;
1026         struct inode *btree_inode = fs_info->btree_inode;
1027         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1028         int ret;
1029
1030         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1031         if (IS_ERR(buf))
1032                 return 0;
1033
1034         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1035
1036         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1037                                        mirror_num);
1038         if (ret) {
1039                 free_extent_buffer_stale(buf);
1040                 return ret;
1041         }
1042
1043         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1044                 free_extent_buffer_stale(buf);
1045                 return -EIO;
1046         } else if (extent_buffer_uptodate(buf)) {
1047                 *eb = buf;
1048         } else {
1049                 free_extent_buffer(buf);
1050         }
1051         return 0;
1052 }
1053
1054 struct extent_buffer *btrfs_find_create_tree_block(
1055                                                 struct btrfs_fs_info *fs_info,
1056                                                 u64 bytenr)
1057 {
1058         if (btrfs_is_testing(fs_info))
1059                 return alloc_test_extent_buffer(fs_info, bytenr);
1060         return alloc_extent_buffer(fs_info, bytenr);
1061 }
1062
1063
1064 int btrfs_write_tree_block(struct extent_buffer *buf)
1065 {
1066         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1067                                         buf->start + buf->len - 1);
1068 }
1069
1070 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1071 {
1072         filemap_fdatawait_range(buf->pages[0]->mapping,
1073                                 buf->start, buf->start + buf->len - 1);
1074 }
1075
1076 /*
1077  * Read tree block at logical address @bytenr and do variant basic but critical
1078  * verification.
1079  *
1080  * @parent_transid:     expected transid of this tree block, skip check if 0
1081  * @level:              expected level, mandatory check
1082  * @first_key:          expected key in slot 0, skip check if NULL
1083  */
1084 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1085                                       u64 parent_transid, int level,
1086                                       struct btrfs_key *first_key)
1087 {
1088         struct extent_buffer *buf = NULL;
1089         int ret;
1090
1091         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1092         if (IS_ERR(buf))
1093                 return buf;
1094
1095         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1096                                              level, first_key);
1097         if (ret) {
1098                 free_extent_buffer_stale(buf);
1099                 return ERR_PTR(ret);
1100         }
1101         return buf;
1102
1103 }
1104
1105 void clean_tree_block(struct btrfs_fs_info *fs_info,
1106                       struct extent_buffer *buf)
1107 {
1108         if (btrfs_header_generation(buf) ==
1109             fs_info->running_transaction->transid) {
1110                 btrfs_assert_tree_locked(buf);
1111
1112                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1113                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1114                                                  -buf->len,
1115                                                  fs_info->dirty_metadata_batch);
1116                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1117                         btrfs_set_lock_blocking(buf);
1118                         clear_extent_buffer_dirty(buf);
1119                 }
1120         }
1121 }
1122
1123 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1124 {
1125         struct btrfs_subvolume_writers *writers;
1126         int ret;
1127
1128         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1129         if (!writers)
1130                 return ERR_PTR(-ENOMEM);
1131
1132         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1133         if (ret < 0) {
1134                 kfree(writers);
1135                 return ERR_PTR(ret);
1136         }
1137
1138         init_waitqueue_head(&writers->wait);
1139         return writers;
1140 }
1141
1142 static void
1143 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1144 {
1145         percpu_counter_destroy(&writers->counter);
1146         kfree(writers);
1147 }
1148
1149 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1150                          u64 objectid)
1151 {
1152         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1153         root->node = NULL;
1154         root->commit_root = NULL;
1155         root->state = 0;
1156         root->orphan_cleanup_state = 0;
1157
1158         root->objectid = objectid;
1159         root->last_trans = 0;
1160         root->highest_objectid = 0;
1161         root->nr_delalloc_inodes = 0;
1162         root->nr_ordered_extents = 0;
1163         root->inode_tree = RB_ROOT;
1164         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1165         root->block_rsv = NULL;
1166
1167         INIT_LIST_HEAD(&root->dirty_list);
1168         INIT_LIST_HEAD(&root->root_list);
1169         INIT_LIST_HEAD(&root->delalloc_inodes);
1170         INIT_LIST_HEAD(&root->delalloc_root);
1171         INIT_LIST_HEAD(&root->ordered_extents);
1172         INIT_LIST_HEAD(&root->ordered_root);
1173         INIT_LIST_HEAD(&root->logged_list[0]);
1174         INIT_LIST_HEAD(&root->logged_list[1]);
1175         spin_lock_init(&root->inode_lock);
1176         spin_lock_init(&root->delalloc_lock);
1177         spin_lock_init(&root->ordered_extent_lock);
1178         spin_lock_init(&root->accounting_lock);
1179         spin_lock_init(&root->log_extents_lock[0]);
1180         spin_lock_init(&root->log_extents_lock[1]);
1181         spin_lock_init(&root->qgroup_meta_rsv_lock);
1182         mutex_init(&root->objectid_mutex);
1183         mutex_init(&root->log_mutex);
1184         mutex_init(&root->ordered_extent_mutex);
1185         mutex_init(&root->delalloc_mutex);
1186         init_waitqueue_head(&root->log_writer_wait);
1187         init_waitqueue_head(&root->log_commit_wait[0]);
1188         init_waitqueue_head(&root->log_commit_wait[1]);
1189         INIT_LIST_HEAD(&root->log_ctxs[0]);
1190         INIT_LIST_HEAD(&root->log_ctxs[1]);
1191         atomic_set(&root->log_commit[0], 0);
1192         atomic_set(&root->log_commit[1], 0);
1193         atomic_set(&root->log_writers, 0);
1194         atomic_set(&root->log_batch, 0);
1195         refcount_set(&root->refs, 1);
1196         atomic_set(&root->will_be_snapshotted, 0);
1197         atomic_set(&root->snapshot_force_cow, 0);
1198         root->log_transid = 0;
1199         root->log_transid_committed = -1;
1200         root->last_log_commit = 0;
1201         if (!dummy)
1202                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1203
1204         memset(&root->root_key, 0, sizeof(root->root_key));
1205         memset(&root->root_item, 0, sizeof(root->root_item));
1206         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1207         if (!dummy)
1208                 root->defrag_trans_start = fs_info->generation;
1209         else
1210                 root->defrag_trans_start = 0;
1211         root->root_key.objectid = objectid;
1212         root->anon_dev = 0;
1213
1214         spin_lock_init(&root->root_item_lock);
1215 }
1216
1217 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1218                 gfp_t flags)
1219 {
1220         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1221         if (root)
1222                 root->fs_info = fs_info;
1223         return root;
1224 }
1225
1226 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1227 /* Should only be used by the testing infrastructure */
1228 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1229 {
1230         struct btrfs_root *root;
1231
1232         if (!fs_info)
1233                 return ERR_PTR(-EINVAL);
1234
1235         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1236         if (!root)
1237                 return ERR_PTR(-ENOMEM);
1238
1239         /* We don't use the stripesize in selftest, set it as sectorsize */
1240         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1241         root->alloc_bytenr = 0;
1242
1243         return root;
1244 }
1245 #endif
1246
1247 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1248                                      struct btrfs_fs_info *fs_info,
1249                                      u64 objectid)
1250 {
1251         struct extent_buffer *leaf;
1252         struct btrfs_root *tree_root = fs_info->tree_root;
1253         struct btrfs_root *root;
1254         struct btrfs_key key;
1255         unsigned int nofs_flag;
1256         int ret = 0;
1257         uuid_le uuid = NULL_UUID_LE;
1258
1259         /*
1260          * We're holding a transaction handle, so use a NOFS memory allocation
1261          * context to avoid deadlock if reclaim happens.
1262          */
1263         nofs_flag = memalloc_nofs_save();
1264         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1265         memalloc_nofs_restore(nofs_flag);
1266         if (!root)
1267                 return ERR_PTR(-ENOMEM);
1268
1269         __setup_root(root, fs_info, objectid);
1270         root->root_key.objectid = objectid;
1271         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1272         root->root_key.offset = 0;
1273
1274         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1275         if (IS_ERR(leaf)) {
1276                 ret = PTR_ERR(leaf);
1277                 leaf = NULL;
1278                 goto fail;
1279         }
1280
1281         root->node = leaf;
1282         btrfs_mark_buffer_dirty(leaf);
1283
1284         root->commit_root = btrfs_root_node(root);
1285         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1286
1287         root->root_item.flags = 0;
1288         root->root_item.byte_limit = 0;
1289         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1290         btrfs_set_root_generation(&root->root_item, trans->transid);
1291         btrfs_set_root_level(&root->root_item, 0);
1292         btrfs_set_root_refs(&root->root_item, 1);
1293         btrfs_set_root_used(&root->root_item, leaf->len);
1294         btrfs_set_root_last_snapshot(&root->root_item, 0);
1295         btrfs_set_root_dirid(&root->root_item, 0);
1296         if (is_fstree(objectid))
1297                 uuid_le_gen(&uuid);
1298         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1299         root->root_item.drop_level = 0;
1300
1301         key.objectid = objectid;
1302         key.type = BTRFS_ROOT_ITEM_KEY;
1303         key.offset = 0;
1304         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305         if (ret)
1306                 goto fail;
1307
1308         btrfs_tree_unlock(leaf);
1309
1310         return root;
1311
1312 fail:
1313         if (leaf) {
1314                 btrfs_tree_unlock(leaf);
1315                 free_extent_buffer(root->commit_root);
1316                 free_extent_buffer(leaf);
1317         }
1318         kfree(root);
1319
1320         return ERR_PTR(ret);
1321 }
1322
1323 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1324                                          struct btrfs_fs_info *fs_info)
1325 {
1326         struct btrfs_root *root;
1327         struct extent_buffer *leaf;
1328
1329         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1330         if (!root)
1331                 return ERR_PTR(-ENOMEM);
1332
1333         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1334
1335         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1336         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1337         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1338
1339         /*
1340          * DON'T set REF_COWS for log trees
1341          *
1342          * log trees do not get reference counted because they go away
1343          * before a real commit is actually done.  They do store pointers
1344          * to file data extents, and those reference counts still get
1345          * updated (along with back refs to the log tree).
1346          */
1347
1348         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1349                         NULL, 0, 0, 0);
1350         if (IS_ERR(leaf)) {
1351                 kfree(root);
1352                 return ERR_CAST(leaf);
1353         }
1354
1355         root->node = leaf;
1356
1357         btrfs_mark_buffer_dirty(root->node);
1358         btrfs_tree_unlock(root->node);
1359         return root;
1360 }
1361
1362 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1363                              struct btrfs_fs_info *fs_info)
1364 {
1365         struct btrfs_root *log_root;
1366
1367         log_root = alloc_log_tree(trans, fs_info);
1368         if (IS_ERR(log_root))
1369                 return PTR_ERR(log_root);
1370         WARN_ON(fs_info->log_root_tree);
1371         fs_info->log_root_tree = log_root;
1372         return 0;
1373 }
1374
1375 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1376                        struct btrfs_root *root)
1377 {
1378         struct btrfs_fs_info *fs_info = root->fs_info;
1379         struct btrfs_root *log_root;
1380         struct btrfs_inode_item *inode_item;
1381
1382         log_root = alloc_log_tree(trans, fs_info);
1383         if (IS_ERR(log_root))
1384                 return PTR_ERR(log_root);
1385
1386         log_root->last_trans = trans->transid;
1387         log_root->root_key.offset = root->root_key.objectid;
1388
1389         inode_item = &log_root->root_item.inode;
1390         btrfs_set_stack_inode_generation(inode_item, 1);
1391         btrfs_set_stack_inode_size(inode_item, 3);
1392         btrfs_set_stack_inode_nlink(inode_item, 1);
1393         btrfs_set_stack_inode_nbytes(inode_item,
1394                                      fs_info->nodesize);
1395         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1396
1397         btrfs_set_root_node(&log_root->root_item, log_root->node);
1398
1399         WARN_ON(root->log_root);
1400         root->log_root = log_root;
1401         root->log_transid = 0;
1402         root->log_transid_committed = -1;
1403         root->last_log_commit = 0;
1404         return 0;
1405 }
1406
1407 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1408                                                struct btrfs_key *key)
1409 {
1410         struct btrfs_root *root;
1411         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1412         struct btrfs_path *path;
1413         u64 generation;
1414         int ret;
1415         int level;
1416
1417         path = btrfs_alloc_path();
1418         if (!path)
1419                 return ERR_PTR(-ENOMEM);
1420
1421         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1422         if (!root) {
1423                 ret = -ENOMEM;
1424                 goto alloc_fail;
1425         }
1426
1427         __setup_root(root, fs_info, key->objectid);
1428
1429         ret = btrfs_find_root(tree_root, key, path,
1430                               &root->root_item, &root->root_key);
1431         if (ret) {
1432                 if (ret > 0)
1433                         ret = -ENOENT;
1434                 goto find_fail;
1435         }
1436
1437         generation = btrfs_root_generation(&root->root_item);
1438         level = btrfs_root_level(&root->root_item);
1439         root->node = read_tree_block(fs_info,
1440                                      btrfs_root_bytenr(&root->root_item),
1441                                      generation, level, NULL);
1442         if (IS_ERR(root->node)) {
1443                 ret = PTR_ERR(root->node);
1444                 goto find_fail;
1445         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1446                 ret = -EIO;
1447                 free_extent_buffer(root->node);
1448                 goto find_fail;
1449         }
1450         root->commit_root = btrfs_root_node(root);
1451 out:
1452         btrfs_free_path(path);
1453         return root;
1454
1455 find_fail:
1456         kfree(root);
1457 alloc_fail:
1458         root = ERR_PTR(ret);
1459         goto out;
1460 }
1461
1462 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1463                                       struct btrfs_key *location)
1464 {
1465         struct btrfs_root *root;
1466
1467         root = btrfs_read_tree_root(tree_root, location);
1468         if (IS_ERR(root))
1469                 return root;
1470
1471         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1472                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1473                 btrfs_check_and_init_root_item(&root->root_item);
1474         }
1475
1476         return root;
1477 }
1478
1479 int btrfs_init_fs_root(struct btrfs_root *root)
1480 {
1481         int ret;
1482         struct btrfs_subvolume_writers *writers;
1483
1484         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1485         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1486                                         GFP_NOFS);
1487         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1488                 ret = -ENOMEM;
1489                 goto fail;
1490         }
1491
1492         writers = btrfs_alloc_subvolume_writers();
1493         if (IS_ERR(writers)) {
1494                 ret = PTR_ERR(writers);
1495                 goto fail;
1496         }
1497         root->subv_writers = writers;
1498
1499         btrfs_init_free_ino_ctl(root);
1500         spin_lock_init(&root->ino_cache_lock);
1501         init_waitqueue_head(&root->ino_cache_wait);
1502
1503         /*
1504          * Don't assign anonymous block device to roots that are not exposed to
1505          * userspace, the id pool is limited to 1M
1506          */
1507         if (is_fstree(root->root_key.objectid) &&
1508             btrfs_root_refs(&root->root_item) > 0) {
1509                 ret = get_anon_bdev(&root->anon_dev);
1510                 if (ret)
1511                         goto fail;
1512         }
1513
1514         mutex_lock(&root->objectid_mutex);
1515         ret = btrfs_find_highest_objectid(root,
1516                                         &root->highest_objectid);
1517         if (ret) {
1518                 mutex_unlock(&root->objectid_mutex);
1519                 goto fail;
1520         }
1521
1522         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1523
1524         mutex_unlock(&root->objectid_mutex);
1525
1526         return 0;
1527 fail:
1528         /* The caller is responsible to call btrfs_free_fs_root */
1529         return ret;
1530 }
1531
1532 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1533                                         u64 root_id)
1534 {
1535         struct btrfs_root *root;
1536
1537         spin_lock(&fs_info->fs_roots_radix_lock);
1538         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1539                                  (unsigned long)root_id);
1540         spin_unlock(&fs_info->fs_roots_radix_lock);
1541         return root;
1542 }
1543
1544 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1545                          struct btrfs_root *root)
1546 {
1547         int ret;
1548
1549         ret = radix_tree_preload(GFP_NOFS);
1550         if (ret)
1551                 return ret;
1552
1553         spin_lock(&fs_info->fs_roots_radix_lock);
1554         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1555                                 (unsigned long)root->root_key.objectid,
1556                                 root);
1557         if (ret == 0)
1558                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1559         spin_unlock(&fs_info->fs_roots_radix_lock);
1560         radix_tree_preload_end();
1561
1562         return ret;
1563 }
1564
1565 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1566                                      struct btrfs_key *location,
1567                                      bool check_ref)
1568 {
1569         struct btrfs_root *root;
1570         struct btrfs_path *path;
1571         struct btrfs_key key;
1572         int ret;
1573
1574         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1575                 return fs_info->tree_root;
1576         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1577                 return fs_info->extent_root;
1578         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1579                 return fs_info->chunk_root;
1580         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1581                 return fs_info->dev_root;
1582         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1583                 return fs_info->csum_root;
1584         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1585                 return fs_info->quota_root ? fs_info->quota_root :
1586                                              ERR_PTR(-ENOENT);
1587         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1588                 return fs_info->uuid_root ? fs_info->uuid_root :
1589                                             ERR_PTR(-ENOENT);
1590         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1591                 return fs_info->free_space_root ? fs_info->free_space_root :
1592                                                   ERR_PTR(-ENOENT);
1593 again:
1594         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1595         if (root) {
1596                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1597                         return ERR_PTR(-ENOENT);
1598                 return root;
1599         }
1600
1601         root = btrfs_read_fs_root(fs_info->tree_root, location);
1602         if (IS_ERR(root))
1603                 return root;
1604
1605         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1606                 ret = -ENOENT;
1607                 goto fail;
1608         }
1609
1610         ret = btrfs_init_fs_root(root);
1611         if (ret)
1612                 goto fail;
1613
1614         path = btrfs_alloc_path();
1615         if (!path) {
1616                 ret = -ENOMEM;
1617                 goto fail;
1618         }
1619         key.objectid = BTRFS_ORPHAN_OBJECTID;
1620         key.type = BTRFS_ORPHAN_ITEM_KEY;
1621         key.offset = location->objectid;
1622
1623         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1624         btrfs_free_path(path);
1625         if (ret < 0)
1626                 goto fail;
1627         if (ret == 0)
1628                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1629
1630         ret = btrfs_insert_fs_root(fs_info, root);
1631         if (ret) {
1632                 if (ret == -EEXIST) {
1633                         btrfs_free_fs_root(root);
1634                         goto again;
1635                 }
1636                 goto fail;
1637         }
1638         return root;
1639 fail:
1640         btrfs_free_fs_root(root);
1641         return ERR_PTR(ret);
1642 }
1643
1644 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1645 {
1646         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1647         int ret = 0;
1648         struct btrfs_device *device;
1649         struct backing_dev_info *bdi;
1650
1651         rcu_read_lock();
1652         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1653                 if (!device->bdev)
1654                         continue;
1655                 bdi = device->bdev->bd_bdi;
1656                 if (bdi_congested(bdi, bdi_bits)) {
1657                         ret = 1;
1658                         break;
1659                 }
1660         }
1661         rcu_read_unlock();
1662         return ret;
1663 }
1664
1665 /*
1666  * called by the kthread helper functions to finally call the bio end_io
1667  * functions.  This is where read checksum verification actually happens
1668  */
1669 static void end_workqueue_fn(struct btrfs_work *work)
1670 {
1671         struct bio *bio;
1672         struct btrfs_end_io_wq *end_io_wq;
1673
1674         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1675         bio = end_io_wq->bio;
1676
1677         bio->bi_status = end_io_wq->status;
1678         bio->bi_private = end_io_wq->private;
1679         bio->bi_end_io = end_io_wq->end_io;
1680         bio_endio(bio);
1681         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1682 }
1683
1684 static int cleaner_kthread(void *arg)
1685 {
1686         struct btrfs_root *root = arg;
1687         struct btrfs_fs_info *fs_info = root->fs_info;
1688         int again;
1689
1690         while (1) {
1691                 again = 0;
1692
1693                 /* Make the cleaner go to sleep early. */
1694                 if (btrfs_need_cleaner_sleep(fs_info))
1695                         goto sleep;
1696
1697                 /*
1698                  * Do not do anything if we might cause open_ctree() to block
1699                  * before we have finished mounting the filesystem.
1700                  */
1701                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1702                         goto sleep;
1703
1704                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1705                         goto sleep;
1706
1707                 /*
1708                  * Avoid the problem that we change the status of the fs
1709                  * during the above check and trylock.
1710                  */
1711                 if (btrfs_need_cleaner_sleep(fs_info)) {
1712                         mutex_unlock(&fs_info->cleaner_mutex);
1713                         goto sleep;
1714                 }
1715
1716                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1717                 btrfs_run_delayed_iputs(fs_info);
1718                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1719
1720                 again = btrfs_clean_one_deleted_snapshot(root);
1721                 mutex_unlock(&fs_info->cleaner_mutex);
1722
1723                 /*
1724                  * The defragger has dealt with the R/O remount and umount,
1725                  * needn't do anything special here.
1726                  */
1727                 btrfs_run_defrag_inodes(fs_info);
1728
1729                 /*
1730                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1731                  * with relocation (btrfs_relocate_chunk) and relocation
1732                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1733                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1734                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1735                  * unused block groups.
1736                  */
1737                 btrfs_delete_unused_bgs(fs_info);
1738 sleep:
1739                 if (kthread_should_park())
1740                         kthread_parkme();
1741                 if (kthread_should_stop())
1742                         return 0;
1743                 if (!again) {
1744                         set_current_state(TASK_INTERRUPTIBLE);
1745                         schedule();
1746                         __set_current_state(TASK_RUNNING);
1747                 }
1748         }
1749 }
1750
1751 static int transaction_kthread(void *arg)
1752 {
1753         struct btrfs_root *root = arg;
1754         struct btrfs_fs_info *fs_info = root->fs_info;
1755         struct btrfs_trans_handle *trans;
1756         struct btrfs_transaction *cur;
1757         u64 transid;
1758         time64_t now;
1759         unsigned long delay;
1760         bool cannot_commit;
1761
1762         do {
1763                 cannot_commit = false;
1764                 delay = HZ * fs_info->commit_interval;
1765                 mutex_lock(&fs_info->transaction_kthread_mutex);
1766
1767                 spin_lock(&fs_info->trans_lock);
1768                 cur = fs_info->running_transaction;
1769                 if (!cur) {
1770                         spin_unlock(&fs_info->trans_lock);
1771                         goto sleep;
1772                 }
1773
1774                 now = ktime_get_seconds();
1775                 if (cur->state < TRANS_STATE_BLOCKED &&
1776                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1777                     (now < cur->start_time ||
1778                      now - cur->start_time < fs_info->commit_interval)) {
1779                         spin_unlock(&fs_info->trans_lock);
1780                         delay = HZ * 5;
1781                         goto sleep;
1782                 }
1783                 transid = cur->transid;
1784                 spin_unlock(&fs_info->trans_lock);
1785
1786                 /* If the file system is aborted, this will always fail. */
1787                 trans = btrfs_attach_transaction(root);
1788                 if (IS_ERR(trans)) {
1789                         if (PTR_ERR(trans) != -ENOENT)
1790                                 cannot_commit = true;
1791                         goto sleep;
1792                 }
1793                 if (transid == trans->transid) {
1794                         btrfs_commit_transaction(trans);
1795                 } else {
1796                         btrfs_end_transaction(trans);
1797                 }
1798 sleep:
1799                 wake_up_process(fs_info->cleaner_kthread);
1800                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1801
1802                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1803                                       &fs_info->fs_state)))
1804                         btrfs_cleanup_transaction(fs_info);
1805                 if (!kthread_should_stop() &&
1806                                 (!btrfs_transaction_blocked(fs_info) ||
1807                                  cannot_commit))
1808                         schedule_timeout_interruptible(delay);
1809         } while (!kthread_should_stop());
1810         return 0;
1811 }
1812
1813 /*
1814  * this will find the highest generation in the array of
1815  * root backups.  The index of the highest array is returned,
1816  * or -1 if we can't find anything.
1817  *
1818  * We check to make sure the array is valid by comparing the
1819  * generation of the latest  root in the array with the generation
1820  * in the super block.  If they don't match we pitch it.
1821  */
1822 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1823 {
1824         u64 cur;
1825         int newest_index = -1;
1826         struct btrfs_root_backup *root_backup;
1827         int i;
1828
1829         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1830                 root_backup = info->super_copy->super_roots + i;
1831                 cur = btrfs_backup_tree_root_gen(root_backup);
1832                 if (cur == newest_gen)
1833                         newest_index = i;
1834         }
1835
1836         /* check to see if we actually wrapped around */
1837         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1838                 root_backup = info->super_copy->super_roots;
1839                 cur = btrfs_backup_tree_root_gen(root_backup);
1840                 if (cur == newest_gen)
1841                         newest_index = 0;
1842         }
1843         return newest_index;
1844 }
1845
1846
1847 /*
1848  * find the oldest backup so we know where to store new entries
1849  * in the backup array.  This will set the backup_root_index
1850  * field in the fs_info struct
1851  */
1852 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1853                                      u64 newest_gen)
1854 {
1855         int newest_index = -1;
1856
1857         newest_index = find_newest_super_backup(info, newest_gen);
1858         /* if there was garbage in there, just move along */
1859         if (newest_index == -1) {
1860                 info->backup_root_index = 0;
1861         } else {
1862                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1863         }
1864 }
1865
1866 /*
1867  * copy all the root pointers into the super backup array.
1868  * this will bump the backup pointer by one when it is
1869  * done
1870  */
1871 static void backup_super_roots(struct btrfs_fs_info *info)
1872 {
1873         int next_backup;
1874         struct btrfs_root_backup *root_backup;
1875         int last_backup;
1876
1877         next_backup = info->backup_root_index;
1878         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1879                 BTRFS_NUM_BACKUP_ROOTS;
1880
1881         /*
1882          * just overwrite the last backup if we're at the same generation
1883          * this happens only at umount
1884          */
1885         root_backup = info->super_for_commit->super_roots + last_backup;
1886         if (btrfs_backup_tree_root_gen(root_backup) ==
1887             btrfs_header_generation(info->tree_root->node))
1888                 next_backup = last_backup;
1889
1890         root_backup = info->super_for_commit->super_roots + next_backup;
1891
1892         /*
1893          * make sure all of our padding and empty slots get zero filled
1894          * regardless of which ones we use today
1895          */
1896         memset(root_backup, 0, sizeof(*root_backup));
1897
1898         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1899
1900         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1901         btrfs_set_backup_tree_root_gen(root_backup,
1902                                btrfs_header_generation(info->tree_root->node));
1903
1904         btrfs_set_backup_tree_root_level(root_backup,
1905                                btrfs_header_level(info->tree_root->node));
1906
1907         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1908         btrfs_set_backup_chunk_root_gen(root_backup,
1909                                btrfs_header_generation(info->chunk_root->node));
1910         btrfs_set_backup_chunk_root_level(root_backup,
1911                                btrfs_header_level(info->chunk_root->node));
1912
1913         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1914         btrfs_set_backup_extent_root_gen(root_backup,
1915                                btrfs_header_generation(info->extent_root->node));
1916         btrfs_set_backup_extent_root_level(root_backup,
1917                                btrfs_header_level(info->extent_root->node));
1918
1919         /*
1920          * we might commit during log recovery, which happens before we set
1921          * the fs_root.  Make sure it is valid before we fill it in.
1922          */
1923         if (info->fs_root && info->fs_root->node) {
1924                 btrfs_set_backup_fs_root(root_backup,
1925                                          info->fs_root->node->start);
1926                 btrfs_set_backup_fs_root_gen(root_backup,
1927                                btrfs_header_generation(info->fs_root->node));
1928                 btrfs_set_backup_fs_root_level(root_backup,
1929                                btrfs_header_level(info->fs_root->node));
1930         }
1931
1932         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1933         btrfs_set_backup_dev_root_gen(root_backup,
1934                                btrfs_header_generation(info->dev_root->node));
1935         btrfs_set_backup_dev_root_level(root_backup,
1936                                        btrfs_header_level(info->dev_root->node));
1937
1938         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1939         btrfs_set_backup_csum_root_gen(root_backup,
1940                                btrfs_header_generation(info->csum_root->node));
1941         btrfs_set_backup_csum_root_level(root_backup,
1942                                btrfs_header_level(info->csum_root->node));
1943
1944         btrfs_set_backup_total_bytes(root_backup,
1945                              btrfs_super_total_bytes(info->super_copy));
1946         btrfs_set_backup_bytes_used(root_backup,
1947                              btrfs_super_bytes_used(info->super_copy));
1948         btrfs_set_backup_num_devices(root_backup,
1949                              btrfs_super_num_devices(info->super_copy));
1950
1951         /*
1952          * if we don't copy this out to the super_copy, it won't get remembered
1953          * for the next commit
1954          */
1955         memcpy(&info->super_copy->super_roots,
1956                &info->super_for_commit->super_roots,
1957                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1958 }
1959
1960 /*
1961  * this copies info out of the root backup array and back into
1962  * the in-memory super block.  It is meant to help iterate through
1963  * the array, so you send it the number of backups you've already
1964  * tried and the last backup index you used.
1965  *
1966  * this returns -1 when it has tried all the backups
1967  */
1968 static noinline int next_root_backup(struct btrfs_fs_info *info,
1969                                      struct btrfs_super_block *super,
1970                                      int *num_backups_tried, int *backup_index)
1971 {
1972         struct btrfs_root_backup *root_backup;
1973         int newest = *backup_index;
1974
1975         if (*num_backups_tried == 0) {
1976                 u64 gen = btrfs_super_generation(super);
1977
1978                 newest = find_newest_super_backup(info, gen);
1979                 if (newest == -1)
1980                         return -1;
1981
1982                 *backup_index = newest;
1983                 *num_backups_tried = 1;
1984         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1985                 /* we've tried all the backups, all done */
1986                 return -1;
1987         } else {
1988                 /* jump to the next oldest backup */
1989                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1990                         BTRFS_NUM_BACKUP_ROOTS;
1991                 *backup_index = newest;
1992                 *num_backups_tried += 1;
1993         }
1994         root_backup = super->super_roots + newest;
1995
1996         btrfs_set_super_generation(super,
1997                                    btrfs_backup_tree_root_gen(root_backup));
1998         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1999         btrfs_set_super_root_level(super,
2000                                    btrfs_backup_tree_root_level(root_backup));
2001         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2002
2003         /*
2004          * fixme: the total bytes and num_devices need to match or we should
2005          * need a fsck
2006          */
2007         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2008         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2009         return 0;
2010 }
2011
2012 /* helper to cleanup workers */
2013 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2014 {
2015         btrfs_destroy_workqueue(fs_info->fixup_workers);
2016         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2017         btrfs_destroy_workqueue(fs_info->workers);
2018         btrfs_destroy_workqueue(fs_info->endio_workers);
2019         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2020         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2021         btrfs_destroy_workqueue(fs_info->rmw_workers);
2022         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2023         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2024         btrfs_destroy_workqueue(fs_info->submit_workers);
2025         btrfs_destroy_workqueue(fs_info->delayed_workers);
2026         btrfs_destroy_workqueue(fs_info->caching_workers);
2027         btrfs_destroy_workqueue(fs_info->readahead_workers);
2028         btrfs_destroy_workqueue(fs_info->flush_workers);
2029         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2030         btrfs_destroy_workqueue(fs_info->extent_workers);
2031         /*
2032          * Now that all other work queues are destroyed, we can safely destroy
2033          * the queues used for metadata I/O, since tasks from those other work
2034          * queues can do metadata I/O operations.
2035          */
2036         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2037         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2038 }
2039
2040 static void free_root_extent_buffers(struct btrfs_root *root)
2041 {
2042         if (root) {
2043                 free_extent_buffer(root->node);
2044                 free_extent_buffer(root->commit_root);
2045                 root->node = NULL;
2046                 root->commit_root = NULL;
2047         }
2048 }
2049
2050 /* helper to cleanup tree roots */
2051 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2052 {
2053         free_root_extent_buffers(info->tree_root);
2054
2055         free_root_extent_buffers(info->dev_root);
2056         free_root_extent_buffers(info->extent_root);
2057         free_root_extent_buffers(info->csum_root);
2058         free_root_extent_buffers(info->quota_root);
2059         free_root_extent_buffers(info->uuid_root);
2060         if (free_chunk_root)
2061                 free_root_extent_buffers(info->chunk_root);
2062         free_root_extent_buffers(info->free_space_root);
2063 }
2064
2065 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2066 {
2067         int ret;
2068         struct btrfs_root *gang[8];
2069         int i;
2070
2071         while (!list_empty(&fs_info->dead_roots)) {
2072                 gang[0] = list_entry(fs_info->dead_roots.next,
2073                                      struct btrfs_root, root_list);
2074                 list_del(&gang[0]->root_list);
2075
2076                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2077                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2078                 } else {
2079                         free_extent_buffer(gang[0]->node);
2080                         free_extent_buffer(gang[0]->commit_root);
2081                         btrfs_put_fs_root(gang[0]);
2082                 }
2083         }
2084
2085         while (1) {
2086                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2087                                              (void **)gang, 0,
2088                                              ARRAY_SIZE(gang));
2089                 if (!ret)
2090                         break;
2091                 for (i = 0; i < ret; i++)
2092                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2093         }
2094
2095         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2096                 btrfs_free_log_root_tree(NULL, fs_info);
2097                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2098         }
2099 }
2100
2101 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2102 {
2103         mutex_init(&fs_info->scrub_lock);
2104         atomic_set(&fs_info->scrubs_running, 0);
2105         atomic_set(&fs_info->scrub_pause_req, 0);
2106         atomic_set(&fs_info->scrubs_paused, 0);
2107         atomic_set(&fs_info->scrub_cancel_req, 0);
2108         init_waitqueue_head(&fs_info->scrub_pause_wait);
2109         fs_info->scrub_workers_refcnt = 0;
2110 }
2111
2112 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2113 {
2114         spin_lock_init(&fs_info->balance_lock);
2115         mutex_init(&fs_info->balance_mutex);
2116         atomic_set(&fs_info->balance_pause_req, 0);
2117         atomic_set(&fs_info->balance_cancel_req, 0);
2118         fs_info->balance_ctl = NULL;
2119         init_waitqueue_head(&fs_info->balance_wait_q);
2120 }
2121
2122 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2123 {
2124         struct inode *inode = fs_info->btree_inode;
2125
2126         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2127         set_nlink(inode, 1);
2128         /*
2129          * we set the i_size on the btree inode to the max possible int.
2130          * the real end of the address space is determined by all of
2131          * the devices in the system
2132          */
2133         inode->i_size = OFFSET_MAX;
2134         inode->i_mapping->a_ops = &btree_aops;
2135
2136         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2137         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2138         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2139         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2140
2141         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2142
2143         BTRFS_I(inode)->root = fs_info->tree_root;
2144         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2145         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2146         btrfs_insert_inode_hash(inode);
2147 }
2148
2149 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2150 {
2151         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2152         rwlock_init(&fs_info->dev_replace.lock);
2153         atomic_set(&fs_info->dev_replace.read_locks, 0);
2154         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2155         init_waitqueue_head(&fs_info->replace_wait);
2156         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2157 }
2158
2159 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2160 {
2161         spin_lock_init(&fs_info->qgroup_lock);
2162         mutex_init(&fs_info->qgroup_ioctl_lock);
2163         fs_info->qgroup_tree = RB_ROOT;
2164         fs_info->qgroup_op_tree = RB_ROOT;
2165         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2166         fs_info->qgroup_seq = 1;
2167         fs_info->qgroup_ulist = NULL;
2168         fs_info->qgroup_rescan_running = false;
2169         mutex_init(&fs_info->qgroup_rescan_lock);
2170 }
2171
2172 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2173                 struct btrfs_fs_devices *fs_devices)
2174 {
2175         u32 max_active = fs_info->thread_pool_size;
2176         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2177
2178         fs_info->workers =
2179                 btrfs_alloc_workqueue(fs_info, "worker",
2180                                       flags | WQ_HIGHPRI, max_active, 16);
2181
2182         fs_info->delalloc_workers =
2183                 btrfs_alloc_workqueue(fs_info, "delalloc",
2184                                       flags, max_active, 2);
2185
2186         fs_info->flush_workers =
2187                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2188                                       flags, max_active, 0);
2189
2190         fs_info->caching_workers =
2191                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2192
2193         /*
2194          * a higher idle thresh on the submit workers makes it much more
2195          * likely that bios will be send down in a sane order to the
2196          * devices
2197          */
2198         fs_info->submit_workers =
2199                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2200                                       min_t(u64, fs_devices->num_devices,
2201                                             max_active), 64);
2202
2203         fs_info->fixup_workers =
2204                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2205
2206         /*
2207          * endios are largely parallel and should have a very
2208          * low idle thresh
2209          */
2210         fs_info->endio_workers =
2211                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2212         fs_info->endio_meta_workers =
2213                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2214                                       max_active, 4);
2215         fs_info->endio_meta_write_workers =
2216                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2217                                       max_active, 2);
2218         fs_info->endio_raid56_workers =
2219                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2220                                       max_active, 4);
2221         fs_info->endio_repair_workers =
2222                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2223         fs_info->rmw_workers =
2224                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2225         fs_info->endio_write_workers =
2226                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2227                                       max_active, 2);
2228         fs_info->endio_freespace_worker =
2229                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2230                                       max_active, 0);
2231         fs_info->delayed_workers =
2232                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2233                                       max_active, 0);
2234         fs_info->readahead_workers =
2235                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2236                                       max_active, 2);
2237         fs_info->qgroup_rescan_workers =
2238                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2239         fs_info->extent_workers =
2240                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2241                                       min_t(u64, fs_devices->num_devices,
2242                                             max_active), 8);
2243
2244         if (!(fs_info->workers && fs_info->delalloc_workers &&
2245               fs_info->submit_workers && fs_info->flush_workers &&
2246               fs_info->endio_workers && fs_info->endio_meta_workers &&
2247               fs_info->endio_meta_write_workers &&
2248               fs_info->endio_repair_workers &&
2249               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2250               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2251               fs_info->caching_workers && fs_info->readahead_workers &&
2252               fs_info->fixup_workers && fs_info->delayed_workers &&
2253               fs_info->extent_workers &&
2254               fs_info->qgroup_rescan_workers)) {
2255                 return -ENOMEM;
2256         }
2257
2258         return 0;
2259 }
2260
2261 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2262                             struct btrfs_fs_devices *fs_devices)
2263 {
2264         int ret;
2265         struct btrfs_root *log_tree_root;
2266         struct btrfs_super_block *disk_super = fs_info->super_copy;
2267         u64 bytenr = btrfs_super_log_root(disk_super);
2268         int level = btrfs_super_log_root_level(disk_super);
2269
2270         if (fs_devices->rw_devices == 0) {
2271                 btrfs_warn(fs_info, "log replay required on RO media");
2272                 return -EIO;
2273         }
2274
2275         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2276         if (!log_tree_root)
2277                 return -ENOMEM;
2278
2279         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2280
2281         log_tree_root->node = read_tree_block(fs_info, bytenr,
2282                                               fs_info->generation + 1,
2283                                               level, NULL);
2284         if (IS_ERR(log_tree_root->node)) {
2285                 btrfs_warn(fs_info, "failed to read log tree");
2286                 ret = PTR_ERR(log_tree_root->node);
2287                 kfree(log_tree_root);
2288                 return ret;
2289         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2290                 btrfs_err(fs_info, "failed to read log tree");
2291                 free_extent_buffer(log_tree_root->node);
2292                 kfree(log_tree_root);
2293                 return -EIO;
2294         }
2295         /* returns with log_tree_root freed on success */
2296         ret = btrfs_recover_log_trees(log_tree_root);
2297         if (ret) {
2298                 btrfs_handle_fs_error(fs_info, ret,
2299                                       "Failed to recover log tree");
2300                 free_extent_buffer(log_tree_root->node);
2301                 kfree(log_tree_root);
2302                 return ret;
2303         }
2304
2305         if (sb_rdonly(fs_info->sb)) {
2306                 ret = btrfs_commit_super(fs_info);
2307                 if (ret)
2308                         return ret;
2309         }
2310
2311         return 0;
2312 }
2313
2314 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2315 {
2316         struct btrfs_root *tree_root = fs_info->tree_root;
2317         struct btrfs_root *root;
2318         struct btrfs_key location;
2319         int ret;
2320
2321         BUG_ON(!fs_info->tree_root);
2322
2323         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2324         location.type = BTRFS_ROOT_ITEM_KEY;
2325         location.offset = 0;
2326
2327         root = btrfs_read_tree_root(tree_root, &location);
2328         if (IS_ERR(root)) {
2329                 ret = PTR_ERR(root);
2330                 goto out;
2331         }
2332         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2333         fs_info->extent_root = root;
2334
2335         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2336         root = btrfs_read_tree_root(tree_root, &location);
2337         if (IS_ERR(root)) {
2338                 ret = PTR_ERR(root);
2339                 goto out;
2340         }
2341         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2342         fs_info->dev_root = root;
2343         btrfs_init_devices_late(fs_info);
2344
2345         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2346         root = btrfs_read_tree_root(tree_root, &location);
2347         if (IS_ERR(root)) {
2348                 ret = PTR_ERR(root);
2349                 goto out;
2350         }
2351         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2352         fs_info->csum_root = root;
2353
2354         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2355         root = btrfs_read_tree_root(tree_root, &location);
2356         if (!IS_ERR(root)) {
2357                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2359                 fs_info->quota_root = root;
2360         }
2361
2362         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2363         root = btrfs_read_tree_root(tree_root, &location);
2364         if (IS_ERR(root)) {
2365                 ret = PTR_ERR(root);
2366                 if (ret != -ENOENT)
2367                         goto out;
2368         } else {
2369                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2370                 fs_info->uuid_root = root;
2371         }
2372
2373         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2374                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2375                 root = btrfs_read_tree_root(tree_root, &location);
2376                 if (IS_ERR(root)) {
2377                         ret = PTR_ERR(root);
2378                         goto out;
2379                 }
2380                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381                 fs_info->free_space_root = root;
2382         }
2383
2384         return 0;
2385 out:
2386         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2387                    location.objectid, ret);
2388         return ret;
2389 }
2390
2391 /*
2392  * Real super block validation
2393  * NOTE: super csum type and incompat features will not be checked here.
2394  *
2395  * @sb:         super block to check
2396  * @mirror_num: the super block number to check its bytenr:
2397  *              0       the primary (1st) sb
2398  *              1, 2    2nd and 3rd backup copy
2399  *             -1       skip bytenr check
2400  */
2401 static int validate_super(struct btrfs_fs_info *fs_info,
2402                             struct btrfs_super_block *sb, int mirror_num)
2403 {
2404         u64 nodesize = btrfs_super_nodesize(sb);
2405         u64 sectorsize = btrfs_super_sectorsize(sb);
2406         int ret = 0;
2407
2408         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2409                 btrfs_err(fs_info, "no valid FS found");
2410                 ret = -EINVAL;
2411         }
2412         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2413                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2414                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2415                 ret = -EINVAL;
2416         }
2417         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2419                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2420                 ret = -EINVAL;
2421         }
2422         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2424                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2425                 ret = -EINVAL;
2426         }
2427         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2428                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2429                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2430                 ret = -EINVAL;
2431         }
2432
2433         /*
2434          * Check sectorsize and nodesize first, other check will need it.
2435          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2436          */
2437         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2438             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2439                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2440                 ret = -EINVAL;
2441         }
2442         /* Only PAGE SIZE is supported yet */
2443         if (sectorsize != PAGE_SIZE) {
2444                 btrfs_err(fs_info,
2445                         "sectorsize %llu not supported yet, only support %lu",
2446                         sectorsize, PAGE_SIZE);
2447                 ret = -EINVAL;
2448         }
2449         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2450             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2451                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2452                 ret = -EINVAL;
2453         }
2454         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2455                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2456                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2457                 ret = -EINVAL;
2458         }
2459
2460         /* Root alignment check */
2461         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2462                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2463                            btrfs_super_root(sb));
2464                 ret = -EINVAL;
2465         }
2466         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2467                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2468                            btrfs_super_chunk_root(sb));
2469                 ret = -EINVAL;
2470         }
2471         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2472                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2473                            btrfs_super_log_root(sb));
2474                 ret = -EINVAL;
2475         }
2476
2477         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2478                 btrfs_err(fs_info,
2479                            "dev_item UUID does not match fsid: %pU != %pU",
2480                            fs_info->fsid, sb->dev_item.fsid);
2481                 ret = -EINVAL;
2482         }
2483
2484         /*
2485          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2486          * done later
2487          */
2488         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2489                 btrfs_err(fs_info, "bytes_used is too small %llu",
2490                           btrfs_super_bytes_used(sb));
2491                 ret = -EINVAL;
2492         }
2493         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2494                 btrfs_err(fs_info, "invalid stripesize %u",
2495                           btrfs_super_stripesize(sb));
2496                 ret = -EINVAL;
2497         }
2498         if (btrfs_super_num_devices(sb) > (1UL << 31))
2499                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2500                            btrfs_super_num_devices(sb));
2501         if (btrfs_super_num_devices(sb) == 0) {
2502                 btrfs_err(fs_info, "number of devices is 0");
2503                 ret = -EINVAL;
2504         }
2505
2506         if (mirror_num >= 0 &&
2507             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2508                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2509                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2510                 ret = -EINVAL;
2511         }
2512
2513         /*
2514          * Obvious sys_chunk_array corruptions, it must hold at least one key
2515          * and one chunk
2516          */
2517         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2518                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2519                           btrfs_super_sys_array_size(sb),
2520                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2521                 ret = -EINVAL;
2522         }
2523         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2524                         + sizeof(struct btrfs_chunk)) {
2525                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2526                           btrfs_super_sys_array_size(sb),
2527                           sizeof(struct btrfs_disk_key)
2528                           + sizeof(struct btrfs_chunk));
2529                 ret = -EINVAL;
2530         }
2531
2532         /*
2533          * The generation is a global counter, we'll trust it more than the others
2534          * but it's still possible that it's the one that's wrong.
2535          */
2536         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2537                 btrfs_warn(fs_info,
2538                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2539                         btrfs_super_generation(sb),
2540                         btrfs_super_chunk_root_generation(sb));
2541         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2542             && btrfs_super_cache_generation(sb) != (u64)-1)
2543                 btrfs_warn(fs_info,
2544                         "suspicious: generation < cache_generation: %llu < %llu",
2545                         btrfs_super_generation(sb),
2546                         btrfs_super_cache_generation(sb));
2547
2548         return ret;
2549 }
2550
2551 /*
2552  * Validation of super block at mount time.
2553  * Some checks already done early at mount time, like csum type and incompat
2554  * flags will be skipped.
2555  */
2556 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2557 {
2558         return validate_super(fs_info, fs_info->super_copy, 0);
2559 }
2560
2561 /*
2562  * Validation of super block at write time.
2563  * Some checks like bytenr check will be skipped as their values will be
2564  * overwritten soon.
2565  * Extra checks like csum type and incompat flags will be done here.
2566  */
2567 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2568                                       struct btrfs_super_block *sb)
2569 {
2570         int ret;
2571
2572         ret = validate_super(fs_info, sb, -1);
2573         if (ret < 0)
2574                 goto out;
2575         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2576                 ret = -EUCLEAN;
2577                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2578                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2579                 goto out;
2580         }
2581         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2582                 ret = -EUCLEAN;
2583                 btrfs_err(fs_info,
2584                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2585                           btrfs_super_incompat_flags(sb),
2586                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2587                 goto out;
2588         }
2589 out:
2590         if (ret < 0)
2591                 btrfs_err(fs_info,
2592                 "super block corruption detected before writing it to disk");
2593         return ret;
2594 }
2595
2596 int open_ctree(struct super_block *sb,
2597                struct btrfs_fs_devices *fs_devices,
2598                char *options)
2599 {
2600         u32 sectorsize;
2601         u32 nodesize;
2602         u32 stripesize;
2603         u64 generation;
2604         u64 features;
2605         struct btrfs_key location;
2606         struct buffer_head *bh;
2607         struct btrfs_super_block *disk_super;
2608         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2609         struct btrfs_root *tree_root;
2610         struct btrfs_root *chunk_root;
2611         int ret;
2612         int err = -EINVAL;
2613         int num_backups_tried = 0;
2614         int backup_index = 0;
2615         int clear_free_space_tree = 0;
2616         int level;
2617
2618         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2619         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2620         if (!tree_root || !chunk_root) {
2621                 err = -ENOMEM;
2622                 goto fail;
2623         }
2624
2625         ret = init_srcu_struct(&fs_info->subvol_srcu);
2626         if (ret) {
2627                 err = ret;
2628                 goto fail;
2629         }
2630
2631         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2632         if (ret) {
2633                 err = ret;
2634                 goto fail_srcu;
2635         }
2636         fs_info->dirty_metadata_batch = PAGE_SIZE *
2637                                         (1 + ilog2(nr_cpu_ids));
2638
2639         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2640         if (ret) {
2641                 err = ret;
2642                 goto fail_dirty_metadata_bytes;
2643         }
2644
2645         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2646         if (ret) {
2647                 err = ret;
2648                 goto fail_delalloc_bytes;
2649         }
2650
2651         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2652         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2653         INIT_LIST_HEAD(&fs_info->trans_list);
2654         INIT_LIST_HEAD(&fs_info->dead_roots);
2655         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2656         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2657         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2658         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2659         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2660         spin_lock_init(&fs_info->delalloc_root_lock);
2661         spin_lock_init(&fs_info->trans_lock);
2662         spin_lock_init(&fs_info->fs_roots_radix_lock);
2663         spin_lock_init(&fs_info->delayed_iput_lock);
2664         spin_lock_init(&fs_info->defrag_inodes_lock);
2665         spin_lock_init(&fs_info->super_lock);
2666         spin_lock_init(&fs_info->qgroup_op_lock);
2667         spin_lock_init(&fs_info->buffer_lock);
2668         spin_lock_init(&fs_info->unused_bgs_lock);
2669         rwlock_init(&fs_info->tree_mod_log_lock);
2670         mutex_init(&fs_info->unused_bg_unpin_mutex);
2671         mutex_init(&fs_info->delete_unused_bgs_mutex);
2672         mutex_init(&fs_info->reloc_mutex);
2673         mutex_init(&fs_info->delalloc_root_mutex);
2674         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2675         seqlock_init(&fs_info->profiles_lock);
2676
2677         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2678         INIT_LIST_HEAD(&fs_info->space_info);
2679         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2680         INIT_LIST_HEAD(&fs_info->unused_bgs);
2681         btrfs_mapping_init(&fs_info->mapping_tree);
2682         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2683                              BTRFS_BLOCK_RSV_GLOBAL);
2684         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2685         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2686         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2687         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2688                              BTRFS_BLOCK_RSV_DELOPS);
2689         atomic_set(&fs_info->async_delalloc_pages, 0);
2690         atomic_set(&fs_info->defrag_running, 0);
2691         atomic_set(&fs_info->qgroup_op_seq, 0);
2692         atomic_set(&fs_info->reada_works_cnt, 0);
2693         atomic64_set(&fs_info->tree_mod_seq, 0);
2694         fs_info->sb = sb;
2695         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2696         fs_info->metadata_ratio = 0;
2697         fs_info->defrag_inodes = RB_ROOT;
2698         atomic64_set(&fs_info->free_chunk_space, 0);
2699         fs_info->tree_mod_log = RB_ROOT;
2700         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2701         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2702         /* readahead state */
2703         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2704         spin_lock_init(&fs_info->reada_lock);
2705         btrfs_init_ref_verify(fs_info);
2706
2707         fs_info->thread_pool_size = min_t(unsigned long,
2708                                           num_online_cpus() + 2, 8);
2709
2710         INIT_LIST_HEAD(&fs_info->ordered_roots);
2711         spin_lock_init(&fs_info->ordered_root_lock);
2712
2713         fs_info->btree_inode = new_inode(sb);
2714         if (!fs_info->btree_inode) {
2715                 err = -ENOMEM;
2716                 goto fail_bio_counter;
2717         }
2718         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2719
2720         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2721                                         GFP_KERNEL);
2722         if (!fs_info->delayed_root) {
2723                 err = -ENOMEM;
2724                 goto fail_iput;
2725         }
2726         btrfs_init_delayed_root(fs_info->delayed_root);
2727
2728         btrfs_init_scrub(fs_info);
2729 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2730         fs_info->check_integrity_print_mask = 0;
2731 #endif
2732         btrfs_init_balance(fs_info);
2733         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2734
2735         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2736         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2737
2738         btrfs_init_btree_inode(fs_info);
2739
2740         spin_lock_init(&fs_info->block_group_cache_lock);
2741         fs_info->block_group_cache_tree = RB_ROOT;
2742         fs_info->first_logical_byte = (u64)-1;
2743
2744         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2745         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2746         fs_info->pinned_extents = &fs_info->freed_extents[0];
2747         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2748
2749         mutex_init(&fs_info->ordered_operations_mutex);
2750         mutex_init(&fs_info->tree_log_mutex);
2751         mutex_init(&fs_info->chunk_mutex);
2752         mutex_init(&fs_info->transaction_kthread_mutex);
2753         mutex_init(&fs_info->cleaner_mutex);
2754         mutex_init(&fs_info->ro_block_group_mutex);
2755         init_rwsem(&fs_info->commit_root_sem);
2756         init_rwsem(&fs_info->cleanup_work_sem);
2757         init_rwsem(&fs_info->subvol_sem);
2758         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2759
2760         btrfs_init_dev_replace_locks(fs_info);
2761         btrfs_init_qgroup(fs_info);
2762
2763         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2764         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2765
2766         init_waitqueue_head(&fs_info->transaction_throttle);
2767         init_waitqueue_head(&fs_info->transaction_wait);
2768         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2769         init_waitqueue_head(&fs_info->async_submit_wait);
2770
2771         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2772
2773         /* Usable values until the real ones are cached from the superblock */
2774         fs_info->nodesize = 4096;
2775         fs_info->sectorsize = 4096;
2776         fs_info->stripesize = 4096;
2777
2778         ret = btrfs_alloc_stripe_hash_table(fs_info);
2779         if (ret) {
2780                 err = ret;
2781                 goto fail_alloc;
2782         }
2783
2784         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2785
2786         invalidate_bdev(fs_devices->latest_bdev);
2787
2788         /*
2789          * Read super block and check the signature bytes only
2790          */
2791         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2792         if (IS_ERR(bh)) {
2793                 err = PTR_ERR(bh);
2794                 goto fail_alloc;
2795         }
2796
2797         /*
2798          * We want to check superblock checksum, the type is stored inside.
2799          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2800          */
2801         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2802                 btrfs_err(fs_info, "superblock checksum mismatch");
2803                 err = -EINVAL;
2804                 brelse(bh);
2805                 goto fail_alloc;
2806         }
2807
2808         /*
2809          * super_copy is zeroed at allocation time and we never touch the
2810          * following bytes up to INFO_SIZE, the checksum is calculated from
2811          * the whole block of INFO_SIZE
2812          */
2813         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2814         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2815                sizeof(*fs_info->super_for_commit));
2816         brelse(bh);
2817
2818         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2819
2820         ret = btrfs_validate_mount_super(fs_info);
2821         if (ret) {
2822                 btrfs_err(fs_info, "superblock contains fatal errors");
2823                 err = -EINVAL;
2824                 goto fail_alloc;
2825         }
2826
2827         disk_super = fs_info->super_copy;
2828         if (!btrfs_super_root(disk_super))
2829                 goto fail_alloc;
2830
2831         /* check FS state, whether FS is broken. */
2832         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2833                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2834
2835         /*
2836          * run through our array of backup supers and setup
2837          * our ring pointer to the oldest one
2838          */
2839         generation = btrfs_super_generation(disk_super);
2840         find_oldest_super_backup(fs_info, generation);
2841
2842         /*
2843          * In the long term, we'll store the compression type in the super
2844          * block, and it'll be used for per file compression control.
2845          */
2846         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2847
2848         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2849         if (ret) {
2850                 err = ret;
2851                 goto fail_alloc;
2852         }
2853
2854         features = btrfs_super_incompat_flags(disk_super) &
2855                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2856         if (features) {
2857                 btrfs_err(fs_info,
2858                     "cannot mount because of unsupported optional features (%llx)",
2859                     features);
2860                 err = -EINVAL;
2861                 goto fail_alloc;
2862         }
2863
2864         features = btrfs_super_incompat_flags(disk_super);
2865         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2866         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2867                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2868         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2869                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2870
2871         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2872                 btrfs_info(fs_info, "has skinny extents");
2873
2874         /*
2875          * flag our filesystem as having big metadata blocks if
2876          * they are bigger than the page size
2877          */
2878         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2879                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2880                         btrfs_info(fs_info,
2881                                 "flagging fs with big metadata feature");
2882                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2883         }
2884
2885         nodesize = btrfs_super_nodesize(disk_super);
2886         sectorsize = btrfs_super_sectorsize(disk_super);
2887         stripesize = sectorsize;
2888         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2889         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2890
2891         /* Cache block sizes */
2892         fs_info->nodesize = nodesize;
2893         fs_info->sectorsize = sectorsize;
2894         fs_info->stripesize = stripesize;
2895
2896         /*
2897          * mixed block groups end up with duplicate but slightly offset
2898          * extent buffers for the same range.  It leads to corruptions
2899          */
2900         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2901             (sectorsize != nodesize)) {
2902                 btrfs_err(fs_info,
2903 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2904                         nodesize, sectorsize);
2905                 goto fail_alloc;
2906         }
2907
2908         /*
2909          * Needn't use the lock because there is no other task which will
2910          * update the flag.
2911          */
2912         btrfs_set_super_incompat_flags(disk_super, features);
2913
2914         features = btrfs_super_compat_ro_flags(disk_super) &
2915                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2916         if (!sb_rdonly(sb) && features) {
2917                 btrfs_err(fs_info,
2918         "cannot mount read-write because of unsupported optional features (%llx)",
2919                        features);
2920                 err = -EINVAL;
2921                 goto fail_alloc;
2922         }
2923
2924         ret = btrfs_init_workqueues(fs_info, fs_devices);
2925         if (ret) {
2926                 err = ret;
2927                 goto fail_sb_buffer;
2928         }
2929
2930         sb->s_bdi->congested_fn = btrfs_congested_fn;
2931         sb->s_bdi->congested_data = fs_info;
2932         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2933         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2934         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2935         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2936
2937         sb->s_blocksize = sectorsize;
2938         sb->s_blocksize_bits = blksize_bits(sectorsize);
2939         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2940
2941         mutex_lock(&fs_info->chunk_mutex);
2942         ret = btrfs_read_sys_array(fs_info);
2943         mutex_unlock(&fs_info->chunk_mutex);
2944         if (ret) {
2945                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2946                 goto fail_sb_buffer;
2947         }
2948
2949         generation = btrfs_super_chunk_root_generation(disk_super);
2950         level = btrfs_super_chunk_root_level(disk_super);
2951
2952         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2953
2954         chunk_root->node = read_tree_block(fs_info,
2955                                            btrfs_super_chunk_root(disk_super),
2956                                            generation, level, NULL);
2957         if (IS_ERR(chunk_root->node) ||
2958             !extent_buffer_uptodate(chunk_root->node)) {
2959                 btrfs_err(fs_info, "failed to read chunk root");
2960                 if (!IS_ERR(chunk_root->node))
2961                         free_extent_buffer(chunk_root->node);
2962                 chunk_root->node = NULL;
2963                 goto fail_tree_roots;
2964         }
2965         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2966         chunk_root->commit_root = btrfs_root_node(chunk_root);
2967
2968         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2969            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2970
2971         ret = btrfs_read_chunk_tree(fs_info);
2972         if (ret) {
2973                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2974                 goto fail_tree_roots;
2975         }
2976
2977         /*
2978          * Keep the devid that is marked to be the target device for the
2979          * device replace procedure
2980          */
2981         btrfs_free_extra_devids(fs_devices, 0);
2982
2983         if (!fs_devices->latest_bdev) {
2984                 btrfs_err(fs_info, "failed to read devices");
2985                 goto fail_tree_roots;
2986         }
2987
2988 retry_root_backup:
2989         generation = btrfs_super_generation(disk_super);
2990         level = btrfs_super_root_level(disk_super);
2991
2992         tree_root->node = read_tree_block(fs_info,
2993                                           btrfs_super_root(disk_super),
2994                                           generation, level, NULL);
2995         if (IS_ERR(tree_root->node) ||
2996             !extent_buffer_uptodate(tree_root->node)) {
2997                 btrfs_warn(fs_info, "failed to read tree root");
2998                 if (!IS_ERR(tree_root->node))
2999                         free_extent_buffer(tree_root->node);
3000                 tree_root->node = NULL;
3001                 goto recovery_tree_root;
3002         }
3003
3004         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3005         tree_root->commit_root = btrfs_root_node(tree_root);
3006         btrfs_set_root_refs(&tree_root->root_item, 1);
3007
3008         mutex_lock(&tree_root->objectid_mutex);
3009         ret = btrfs_find_highest_objectid(tree_root,
3010                                         &tree_root->highest_objectid);
3011         if (ret) {
3012                 mutex_unlock(&tree_root->objectid_mutex);
3013                 goto recovery_tree_root;
3014         }
3015
3016         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3017
3018         mutex_unlock(&tree_root->objectid_mutex);
3019
3020         ret = btrfs_read_roots(fs_info);
3021         if (ret)
3022                 goto recovery_tree_root;
3023
3024         fs_info->generation = generation;
3025         fs_info->last_trans_committed = generation;
3026
3027         /*
3028          * If we have a uuid root and we're not being told to rescan we need to
3029          * check the generation here so we can set the
3030          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3031          * transaction during a balance or the log replay without updating the
3032          * uuid generation, and then if we crash we would rescan the uuid tree,
3033          * even though it was perfectly fine.
3034          */
3035         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3036             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3037                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3038
3039         ret = btrfs_verify_dev_extents(fs_info);
3040         if (ret) {
3041                 btrfs_err(fs_info,
3042                           "failed to verify dev extents against chunks: %d",
3043                           ret);
3044                 goto fail_block_groups;
3045         }
3046         ret = btrfs_recover_balance(fs_info);
3047         if (ret) {
3048                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3049                 goto fail_block_groups;
3050         }
3051
3052         ret = btrfs_init_dev_stats(fs_info);
3053         if (ret) {
3054                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3055                 goto fail_block_groups;
3056         }
3057
3058         ret = btrfs_init_dev_replace(fs_info);
3059         if (ret) {
3060                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3061                 goto fail_block_groups;
3062         }
3063
3064         btrfs_free_extra_devids(fs_devices, 1);
3065
3066         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3067         if (ret) {
3068                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3069                                 ret);
3070                 goto fail_block_groups;
3071         }
3072
3073         ret = btrfs_sysfs_add_device(fs_devices);
3074         if (ret) {
3075                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3076                                 ret);
3077                 goto fail_fsdev_sysfs;
3078         }
3079
3080         ret = btrfs_sysfs_add_mounted(fs_info);
3081         if (ret) {
3082                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3083                 goto fail_fsdev_sysfs;
3084         }
3085
3086         ret = btrfs_init_space_info(fs_info);
3087         if (ret) {
3088                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3089                 goto fail_sysfs;
3090         }
3091
3092         ret = btrfs_read_block_groups(fs_info);
3093         if (ret) {
3094                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3095                 goto fail_sysfs;
3096         }
3097
3098         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3099             !btrfs_check_rw_degradable(fs_info, NULL)) {
3100                 btrfs_warn(fs_info,
3101                 "writeable mount is not allowed due to too many missing devices");
3102                 goto fail_sysfs;
3103         }
3104
3105         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3106                                                "btrfs-cleaner");
3107         if (IS_ERR(fs_info->cleaner_kthread))
3108                 goto fail_sysfs;
3109
3110         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3111                                                    tree_root,
3112                                                    "btrfs-transaction");
3113         if (IS_ERR(fs_info->transaction_kthread))
3114                 goto fail_cleaner;
3115
3116         if (!btrfs_test_opt(fs_info, NOSSD) &&
3117             !fs_info->fs_devices->rotating) {
3118                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3119         }
3120
3121         /*
3122          * Mount does not set all options immediately, we can do it now and do
3123          * not have to wait for transaction commit
3124          */
3125         btrfs_apply_pending_changes(fs_info);
3126
3127 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3128         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3129                 ret = btrfsic_mount(fs_info, fs_devices,
3130                                     btrfs_test_opt(fs_info,
3131                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3132                                     1 : 0,
3133                                     fs_info->check_integrity_print_mask);
3134                 if (ret)
3135                         btrfs_warn(fs_info,
3136                                 "failed to initialize integrity check module: %d",
3137                                 ret);
3138         }
3139 #endif
3140         ret = btrfs_read_qgroup_config(fs_info);
3141         if (ret)
3142                 goto fail_trans_kthread;
3143
3144         if (btrfs_build_ref_tree(fs_info))
3145                 btrfs_err(fs_info, "couldn't build ref tree");
3146
3147         /* do not make disk changes in broken FS or nologreplay is given */
3148         if (btrfs_super_log_root(disk_super) != 0 &&
3149             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3150                 btrfs_info(fs_info, "start tree-log replay");
3151                 ret = btrfs_replay_log(fs_info, fs_devices);
3152                 if (ret) {
3153                         err = ret;
3154                         goto fail_qgroup;
3155                 }
3156         }
3157
3158         ret = btrfs_find_orphan_roots(fs_info);
3159         if (ret)
3160                 goto fail_qgroup;
3161
3162         if (!sb_rdonly(sb)) {
3163                 ret = btrfs_cleanup_fs_roots(fs_info);
3164                 if (ret)
3165                         goto fail_qgroup;
3166
3167                 mutex_lock(&fs_info->cleaner_mutex);
3168                 ret = btrfs_recover_relocation(tree_root);
3169                 mutex_unlock(&fs_info->cleaner_mutex);
3170                 if (ret < 0) {
3171                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3172                                         ret);
3173                         err = -EINVAL;
3174                         goto fail_qgroup;
3175                 }
3176         }
3177
3178         location.objectid = BTRFS_FS_TREE_OBJECTID;
3179         location.type = BTRFS_ROOT_ITEM_KEY;
3180         location.offset = 0;
3181
3182         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3183         if (IS_ERR(fs_info->fs_root)) {
3184                 err = PTR_ERR(fs_info->fs_root);
3185                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3186                 fs_info->fs_root = NULL;
3187                 goto fail_qgroup;
3188         }
3189
3190         if (sb_rdonly(sb))
3191                 return 0;
3192
3193         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3194             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3195                 clear_free_space_tree = 1;
3196         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3197                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3198                 btrfs_warn(fs_info, "free space tree is invalid");
3199                 clear_free_space_tree = 1;
3200         }
3201
3202         if (clear_free_space_tree) {
3203                 btrfs_info(fs_info, "clearing free space tree");
3204                 ret = btrfs_clear_free_space_tree(fs_info);
3205                 if (ret) {
3206                         btrfs_warn(fs_info,
3207                                    "failed to clear free space tree: %d", ret);
3208                         close_ctree(fs_info);
3209                         return ret;
3210                 }
3211         }
3212
3213         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3214             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3215                 btrfs_info(fs_info, "creating free space tree");
3216                 ret = btrfs_create_free_space_tree(fs_info);
3217                 if (ret) {
3218                         btrfs_warn(fs_info,
3219                                 "failed to create free space tree: %d", ret);
3220                         close_ctree(fs_info);
3221                         return ret;
3222                 }
3223         }
3224
3225         down_read(&fs_info->cleanup_work_sem);
3226         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3227             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3228                 up_read(&fs_info->cleanup_work_sem);
3229                 close_ctree(fs_info);
3230                 return ret;
3231         }
3232         up_read(&fs_info->cleanup_work_sem);
3233
3234         ret = btrfs_resume_balance_async(fs_info);
3235         if (ret) {
3236                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3237                 close_ctree(fs_info);
3238                 return ret;
3239         }
3240
3241         ret = btrfs_resume_dev_replace_async(fs_info);
3242         if (ret) {
3243                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3244                 close_ctree(fs_info);
3245                 return ret;
3246         }
3247
3248         btrfs_qgroup_rescan_resume(fs_info);
3249
3250         if (!fs_info->uuid_root) {
3251                 btrfs_info(fs_info, "creating UUID tree");
3252                 ret = btrfs_create_uuid_tree(fs_info);
3253                 if (ret) {
3254                         btrfs_warn(fs_info,
3255                                 "failed to create the UUID tree: %d", ret);
3256                         close_ctree(fs_info);
3257                         return ret;
3258                 }
3259         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3260                    fs_info->generation !=
3261                                 btrfs_super_uuid_tree_generation(disk_super)) {
3262                 btrfs_info(fs_info, "checking UUID tree");
3263                 ret = btrfs_check_uuid_tree(fs_info);
3264                 if (ret) {
3265                         btrfs_warn(fs_info,
3266                                 "failed to check the UUID tree: %d", ret);
3267                         close_ctree(fs_info);
3268                         return ret;
3269                 }
3270         }
3271         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3272
3273         /*
3274          * backuproot only affect mount behavior, and if open_ctree succeeded,
3275          * no need to keep the flag
3276          */
3277         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3278
3279         return 0;
3280
3281 fail_qgroup:
3282         btrfs_free_qgroup_config(fs_info);
3283 fail_trans_kthread:
3284         kthread_stop(fs_info->transaction_kthread);
3285         btrfs_cleanup_transaction(fs_info);
3286         btrfs_free_fs_roots(fs_info);
3287 fail_cleaner:
3288         kthread_stop(fs_info->cleaner_kthread);
3289
3290         /*
3291          * make sure we're done with the btree inode before we stop our
3292          * kthreads
3293          */
3294         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3295
3296 fail_sysfs:
3297         btrfs_sysfs_remove_mounted(fs_info);
3298
3299 fail_fsdev_sysfs:
3300         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3301
3302 fail_block_groups:
3303         btrfs_put_block_group_cache(fs_info);
3304
3305 fail_tree_roots:
3306         free_root_pointers(fs_info, true);
3307         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3308
3309 fail_sb_buffer:
3310         btrfs_stop_all_workers(fs_info);
3311         btrfs_free_block_groups(fs_info);
3312 fail_alloc:
3313 fail_iput:
3314         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3315
3316         iput(fs_info->btree_inode);
3317 fail_bio_counter:
3318         percpu_counter_destroy(&fs_info->bio_counter);
3319 fail_delalloc_bytes:
3320         percpu_counter_destroy(&fs_info->delalloc_bytes);
3321 fail_dirty_metadata_bytes:
3322         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3323 fail_srcu:
3324         cleanup_srcu_struct(&fs_info->subvol_srcu);
3325 fail:
3326         btrfs_free_stripe_hash_table(fs_info);
3327         btrfs_close_devices(fs_info->fs_devices);
3328         return err;
3329
3330 recovery_tree_root:
3331         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3332                 goto fail_tree_roots;
3333
3334         free_root_pointers(fs_info, false);
3335
3336         /* don't use the log in recovery mode, it won't be valid */
3337         btrfs_set_super_log_root(disk_super, 0);
3338
3339         /* we can't trust the free space cache either */
3340         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3341
3342         ret = next_root_backup(fs_info, fs_info->super_copy,
3343                                &num_backups_tried, &backup_index);
3344         if (ret == -1)
3345                 goto fail_block_groups;
3346         goto retry_root_backup;
3347 }
3348 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3349
3350 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3351 {
3352         if (uptodate) {
3353                 set_buffer_uptodate(bh);
3354         } else {
3355                 struct btrfs_device *device = (struct btrfs_device *)
3356                         bh->b_private;
3357
3358                 btrfs_warn_rl_in_rcu(device->fs_info,
3359                                 "lost page write due to IO error on %s",
3360                                           rcu_str_deref(device->name));
3361                 /* note, we don't set_buffer_write_io_error because we have
3362                  * our own ways of dealing with the IO errors
3363                  */
3364                 clear_buffer_uptodate(bh);
3365                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3366         }
3367         unlock_buffer(bh);
3368         put_bh(bh);
3369 }
3370
3371 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3372                         struct buffer_head **bh_ret)
3373 {
3374         struct buffer_head *bh;
3375         struct btrfs_super_block *super;
3376         u64 bytenr;
3377
3378         bytenr = btrfs_sb_offset(copy_num);
3379         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3380                 return -EINVAL;
3381
3382         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3383         /*
3384          * If we fail to read from the underlying devices, as of now
3385          * the best option we have is to mark it EIO.
3386          */
3387         if (!bh)
3388                 return -EIO;
3389
3390         super = (struct btrfs_super_block *)bh->b_data;
3391         if (btrfs_super_bytenr(super) != bytenr ||
3392                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3393                 brelse(bh);
3394                 return -EINVAL;
3395         }
3396
3397         *bh_ret = bh;
3398         return 0;
3399 }
3400
3401
3402 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3403 {
3404         struct buffer_head *bh;
3405         struct buffer_head *latest = NULL;
3406         struct btrfs_super_block *super;
3407         int i;
3408         u64 transid = 0;
3409         int ret = -EINVAL;
3410
3411         /* we would like to check all the supers, but that would make
3412          * a btrfs mount succeed after a mkfs from a different FS.
3413          * So, we need to add a special mount option to scan for
3414          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3415          */
3416         for (i = 0; i < 1; i++) {
3417                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3418                 if (ret)
3419                         continue;
3420
3421                 super = (struct btrfs_super_block *)bh->b_data;
3422
3423                 if (!latest || btrfs_super_generation(super) > transid) {
3424                         brelse(latest);
3425                         latest = bh;
3426                         transid = btrfs_super_generation(super);
3427                 } else {
3428                         brelse(bh);
3429                 }
3430         }
3431
3432         if (!latest)
3433                 return ERR_PTR(ret);
3434
3435         return latest;
3436 }
3437
3438 /*
3439  * Write superblock @sb to the @device. Do not wait for completion, all the
3440  * buffer heads we write are pinned.
3441  *
3442  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3443  * the expected device size at commit time. Note that max_mirrors must be
3444  * same for write and wait phases.
3445  *
3446  * Return number of errors when buffer head is not found or submission fails.
3447  */
3448 static int write_dev_supers(struct btrfs_device *device,
3449                             struct btrfs_super_block *sb, int max_mirrors)
3450 {
3451         struct buffer_head *bh;
3452         int i;
3453         int ret;
3454         int errors = 0;
3455         u32 crc;
3456         u64 bytenr;
3457         int op_flags;
3458
3459         if (max_mirrors == 0)
3460                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3461
3462         for (i = 0; i < max_mirrors; i++) {
3463                 bytenr = btrfs_sb_offset(i);
3464                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3465                     device->commit_total_bytes)
3466                         break;
3467
3468                 btrfs_set_super_bytenr(sb, bytenr);
3469
3470                 crc = ~(u32)0;
3471                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3472                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3473                 btrfs_csum_final(crc, sb->csum);
3474
3475                 /* One reference for us, and we leave it for the caller */
3476                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3477                               BTRFS_SUPER_INFO_SIZE);
3478                 if (!bh) {
3479                         btrfs_err(device->fs_info,
3480                             "couldn't get super buffer head for bytenr %llu",
3481                             bytenr);
3482                         errors++;
3483                         continue;
3484                 }
3485
3486                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3487
3488                 /* one reference for submit_bh */
3489                 get_bh(bh);
3490
3491                 set_buffer_uptodate(bh);
3492                 lock_buffer(bh);
3493                 bh->b_end_io = btrfs_end_buffer_write_sync;
3494                 bh->b_private = device;
3495
3496                 /*
3497                  * we fua the first super.  The others we allow
3498                  * to go down lazy.
3499                  */
3500                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3501                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3502                         op_flags |= REQ_FUA;
3503                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3504                 if (ret)
3505                         errors++;
3506         }
3507         return errors < i ? 0 : -1;
3508 }
3509
3510 /*
3511  * Wait for write completion of superblocks done by write_dev_supers,
3512  * @max_mirrors same for write and wait phases.
3513  *
3514  * Return number of errors when buffer head is not found or not marked up to
3515  * date.
3516  */
3517 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3518 {
3519         struct buffer_head *bh;
3520         int i;
3521         int errors = 0;
3522         bool primary_failed = false;
3523         u64 bytenr;
3524
3525         if (max_mirrors == 0)
3526                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3527
3528         for (i = 0; i < max_mirrors; i++) {
3529                 bytenr = btrfs_sb_offset(i);
3530                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3531                     device->commit_total_bytes)
3532                         break;
3533
3534                 bh = __find_get_block(device->bdev,
3535                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3536                                       BTRFS_SUPER_INFO_SIZE);
3537                 if (!bh) {
3538                         errors++;
3539                         if (i == 0)
3540                                 primary_failed = true;
3541                         continue;
3542                 }
3543                 wait_on_buffer(bh);
3544                 if (!buffer_uptodate(bh)) {
3545                         errors++;
3546                         if (i == 0)
3547                                 primary_failed = true;
3548                 }
3549
3550                 /* drop our reference */
3551                 brelse(bh);
3552
3553                 /* drop the reference from the writing run */
3554                 brelse(bh);
3555         }
3556
3557         /* log error, force error return */
3558         if (primary_failed) {
3559                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3560                           device->devid);
3561                 return -1;
3562         }
3563
3564         return errors < i ? 0 : -1;
3565 }
3566
3567 /*
3568  * endio for the write_dev_flush, this will wake anyone waiting
3569  * for the barrier when it is done
3570  */
3571 static void btrfs_end_empty_barrier(struct bio *bio)
3572 {
3573         complete(bio->bi_private);
3574 }
3575
3576 /*
3577  * Submit a flush request to the device if it supports it. Error handling is
3578  * done in the waiting counterpart.
3579  */
3580 static void write_dev_flush(struct btrfs_device *device)
3581 {
3582         struct bio *bio = device->flush_bio;
3583
3584 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3585         /*
3586          * When a disk has write caching disabled, we skip submission of a bio
3587          * with flush and sync requests before writing the superblock, since
3588          * it's not needed. However when the integrity checker is enabled, this
3589          * results in reports that there are metadata blocks referred by a
3590          * superblock that were not properly flushed. So don't skip the bio
3591          * submission only when the integrity checker is enabled for the sake
3592          * of simplicity, since this is a debug tool and not meant for use in
3593          * non-debug builds.
3594          */
3595         struct request_queue *q = bdev_get_queue(device->bdev);
3596         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3597                 return;
3598 #endif
3599
3600         bio_reset(bio);
3601         bio->bi_end_io = btrfs_end_empty_barrier;
3602         bio_set_dev(bio, device->bdev);
3603         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3604         init_completion(&device->flush_wait);
3605         bio->bi_private = &device->flush_wait;
3606
3607         btrfsic_submit_bio(bio);
3608         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3609 }
3610
3611 /*
3612  * If the flush bio has been submitted by write_dev_flush, wait for it.
3613  */
3614 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3615 {
3616         struct bio *bio = device->flush_bio;
3617
3618         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3619                 return BLK_STS_OK;
3620
3621         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3622         wait_for_completion_io(&device->flush_wait);
3623
3624         return bio->bi_status;
3625 }
3626
3627 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3628 {
3629         if (!btrfs_check_rw_degradable(fs_info, NULL))
3630                 return -EIO;
3631         return 0;
3632 }
3633
3634 /*
3635  * send an empty flush down to each device in parallel,
3636  * then wait for them
3637  */
3638 static int barrier_all_devices(struct btrfs_fs_info *info)
3639 {
3640         struct list_head *head;
3641         struct btrfs_device *dev;
3642         int errors_wait = 0;
3643         blk_status_t ret;
3644
3645         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3646         /* send down all the barriers */
3647         head = &info->fs_devices->devices;
3648         list_for_each_entry(dev, head, dev_list) {
3649                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3650                         continue;
3651                 if (!dev->bdev)
3652                         continue;
3653                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3654                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3655                         continue;
3656
3657                 write_dev_flush(dev);
3658                 dev->last_flush_error = BLK_STS_OK;
3659         }
3660
3661         /* wait for all the barriers */
3662         list_for_each_entry(dev, head, dev_list) {
3663                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3664                         continue;
3665                 if (!dev->bdev) {
3666                         errors_wait++;
3667                         continue;
3668                 }
3669                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3670                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3671                         continue;
3672
3673                 ret = wait_dev_flush(dev);
3674                 if (ret) {
3675                         dev->last_flush_error = ret;
3676                         btrfs_dev_stat_inc_and_print(dev,
3677                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3678                         errors_wait++;
3679                 }
3680         }
3681
3682         if (errors_wait) {
3683                 /*
3684                  * At some point we need the status of all disks
3685                  * to arrive at the volume status. So error checking
3686                  * is being pushed to a separate loop.
3687                  */
3688                 return check_barrier_error(info);
3689         }
3690         return 0;
3691 }
3692
3693 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3694 {
3695         int raid_type;
3696         int min_tolerated = INT_MAX;
3697
3698         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3699             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3700                 min_tolerated = min(min_tolerated,
3701                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3702                                     tolerated_failures);
3703
3704         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3705                 if (raid_type == BTRFS_RAID_SINGLE)
3706                         continue;
3707                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3708                         continue;
3709                 min_tolerated = min(min_tolerated,
3710                                     btrfs_raid_array[raid_type].
3711                                     tolerated_failures);
3712         }
3713
3714         if (min_tolerated == INT_MAX) {
3715                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3716                 min_tolerated = 0;
3717         }
3718
3719         return min_tolerated;
3720 }
3721
3722 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3723 {
3724         struct list_head *head;
3725         struct btrfs_device *dev;
3726         struct btrfs_super_block *sb;
3727         struct btrfs_dev_item *dev_item;
3728         int ret;
3729         int do_barriers;
3730         int max_errors;
3731         int total_errors = 0;
3732         u64 flags;
3733
3734         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3735
3736         /*
3737          * max_mirrors == 0 indicates we're from commit_transaction,
3738          * not from fsync where the tree roots in fs_info have not
3739          * been consistent on disk.
3740          */
3741         if (max_mirrors == 0)
3742                 backup_super_roots(fs_info);
3743
3744         sb = fs_info->super_for_commit;
3745         dev_item = &sb->dev_item;
3746
3747         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3748         head = &fs_info->fs_devices->devices;
3749         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3750
3751         if (do_barriers) {
3752                 ret = barrier_all_devices(fs_info);
3753                 if (ret) {
3754                         mutex_unlock(
3755                                 &fs_info->fs_devices->device_list_mutex);
3756                         btrfs_handle_fs_error(fs_info, ret,
3757                                               "errors while submitting device barriers.");
3758                         return ret;
3759                 }
3760         }
3761
3762         list_for_each_entry(dev, head, dev_list) {
3763                 if (!dev->bdev) {
3764                         total_errors++;
3765                         continue;
3766                 }
3767                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3768                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3769                         continue;
3770
3771                 btrfs_set_stack_device_generation(dev_item, 0);
3772                 btrfs_set_stack_device_type(dev_item, dev->type);
3773                 btrfs_set_stack_device_id(dev_item, dev->devid);
3774                 btrfs_set_stack_device_total_bytes(dev_item,
3775                                                    dev->commit_total_bytes);
3776                 btrfs_set_stack_device_bytes_used(dev_item,
3777                                                   dev->commit_bytes_used);
3778                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3779                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3780                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3781                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3782                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3783
3784                 flags = btrfs_super_flags(sb);
3785                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3786
3787                 ret = btrfs_validate_write_super(fs_info, sb);
3788                 if (ret < 0) {
3789                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3790                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3791                                 "unexpected superblock corruption detected");
3792                         return -EUCLEAN;
3793                 }
3794
3795                 ret = write_dev_supers(dev, sb, max_mirrors);
3796                 if (ret)
3797                         total_errors++;
3798         }
3799         if (total_errors > max_errors) {
3800                 btrfs_err(fs_info, "%d errors while writing supers",
3801                           total_errors);
3802                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3803
3804                 /* FUA is masked off if unsupported and can't be the reason */
3805                 btrfs_handle_fs_error(fs_info, -EIO,
3806                                       "%d errors while writing supers",
3807                                       total_errors);
3808                 return -EIO;
3809         }
3810
3811         total_errors = 0;
3812         list_for_each_entry(dev, head, dev_list) {
3813                 if (!dev->bdev)
3814                         continue;
3815                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3816                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3817                         continue;
3818
3819                 ret = wait_dev_supers(dev, max_mirrors);
3820                 if (ret)
3821                         total_errors++;
3822         }
3823         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3824         if (total_errors > max_errors) {
3825                 btrfs_handle_fs_error(fs_info, -EIO,
3826                                       "%d errors while writing supers",
3827                                       total_errors);
3828                 return -EIO;
3829         }
3830         return 0;
3831 }
3832
3833 /* Drop a fs root from the radix tree and free it. */
3834 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3835                                   struct btrfs_root *root)
3836 {
3837         spin_lock(&fs_info->fs_roots_radix_lock);
3838         radix_tree_delete(&fs_info->fs_roots_radix,
3839                           (unsigned long)root->root_key.objectid);
3840         spin_unlock(&fs_info->fs_roots_radix_lock);
3841
3842         if (btrfs_root_refs(&root->root_item) == 0)
3843                 synchronize_srcu(&fs_info->subvol_srcu);
3844
3845         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3846                 btrfs_free_log(NULL, root);
3847                 if (root->reloc_root) {
3848                         free_extent_buffer(root->reloc_root->node);
3849                         free_extent_buffer(root->reloc_root->commit_root);
3850                         btrfs_put_fs_root(root->reloc_root);
3851                         root->reloc_root = NULL;
3852                 }
3853         }
3854
3855         if (root->free_ino_pinned)
3856                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3857         if (root->free_ino_ctl)
3858                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3859         btrfs_free_fs_root(root);
3860 }
3861
3862 void btrfs_free_fs_root(struct btrfs_root *root)
3863 {
3864         iput(root->ino_cache_inode);
3865         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3866         if (root->anon_dev)
3867                 free_anon_bdev(root->anon_dev);
3868         if (root->subv_writers)
3869                 btrfs_free_subvolume_writers(root->subv_writers);
3870         free_extent_buffer(root->node);
3871         free_extent_buffer(root->commit_root);
3872         kfree(root->free_ino_ctl);
3873         kfree(root->free_ino_pinned);
3874         btrfs_put_fs_root(root);
3875 }
3876
3877 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3878 {
3879         u64 root_objectid = 0;
3880         struct btrfs_root *gang[8];
3881         int i = 0;
3882         int err = 0;
3883         unsigned int ret = 0;
3884         int index;
3885
3886         while (1) {
3887                 index = srcu_read_lock(&fs_info->subvol_srcu);
3888                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3889                                              (void **)gang, root_objectid,
3890                                              ARRAY_SIZE(gang));
3891                 if (!ret) {
3892                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3893                         break;
3894                 }
3895                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3896
3897                 for (i = 0; i < ret; i++) {
3898                         /* Avoid to grab roots in dead_roots */
3899                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3900                                 gang[i] = NULL;
3901                                 continue;
3902                         }
3903                         /* grab all the search result for later use */
3904                         gang[i] = btrfs_grab_fs_root(gang[i]);
3905                 }
3906                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3907
3908                 for (i = 0; i < ret; i++) {
3909                         if (!gang[i])
3910                                 continue;
3911                         root_objectid = gang[i]->root_key.objectid;
3912                         err = btrfs_orphan_cleanup(gang[i]);
3913                         if (err)
3914                                 break;
3915                         btrfs_put_fs_root(gang[i]);
3916                 }
3917                 root_objectid++;
3918         }
3919
3920         /* release the uncleaned roots due to error */
3921         for (; i < ret; i++) {
3922                 if (gang[i])
3923                         btrfs_put_fs_root(gang[i]);
3924         }
3925         return err;
3926 }
3927
3928 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3929 {
3930         struct btrfs_root *root = fs_info->tree_root;
3931         struct btrfs_trans_handle *trans;
3932
3933         mutex_lock(&fs_info->cleaner_mutex);
3934         btrfs_run_delayed_iputs(fs_info);
3935         mutex_unlock(&fs_info->cleaner_mutex);
3936         wake_up_process(fs_info->cleaner_kthread);
3937
3938         /* wait until ongoing cleanup work done */
3939         down_write(&fs_info->cleanup_work_sem);
3940         up_write(&fs_info->cleanup_work_sem);
3941
3942         trans = btrfs_join_transaction(root);
3943         if (IS_ERR(trans))
3944                 return PTR_ERR(trans);
3945         return btrfs_commit_transaction(trans);
3946 }
3947
3948 void close_ctree(struct btrfs_fs_info *fs_info)
3949 {
3950         int ret;
3951
3952         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3953         /*
3954          * We don't want the cleaner to start new transactions, add more delayed
3955          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3956          * because that frees the task_struct, and the transaction kthread might
3957          * still try to wake up the cleaner.
3958          */
3959         kthread_park(fs_info->cleaner_kthread);
3960
3961         /* wait for the qgroup rescan worker to stop */
3962         btrfs_qgroup_wait_for_completion(fs_info, false);
3963
3964         /* wait for the uuid_scan task to finish */
3965         down(&fs_info->uuid_tree_rescan_sem);
3966         /* avoid complains from lockdep et al., set sem back to initial state */
3967         up(&fs_info->uuid_tree_rescan_sem);
3968
3969         /* pause restriper - we want to resume on mount */
3970         btrfs_pause_balance(fs_info);
3971
3972         btrfs_dev_replace_suspend_for_unmount(fs_info);
3973
3974         btrfs_scrub_cancel(fs_info);
3975
3976         /* wait for any defraggers to finish */
3977         wait_event(fs_info->transaction_wait,
3978                    (atomic_read(&fs_info->defrag_running) == 0));
3979
3980         /* clear out the rbtree of defraggable inodes */
3981         btrfs_cleanup_defrag_inodes(fs_info);
3982
3983         cancel_work_sync(&fs_info->async_reclaim_work);
3984
3985         if (!sb_rdonly(fs_info->sb)) {
3986                 /*
3987                  * The cleaner kthread is stopped, so do one final pass over
3988                  * unused block groups.
3989                  */
3990                 btrfs_delete_unused_bgs(fs_info);
3991
3992                 /*
3993                  * There might be existing delayed inode workers still running
3994                  * and holding an empty delayed inode item. We must wait for
3995                  * them to complete first because they can create a transaction.
3996                  * This happens when someone calls btrfs_balance_delayed_items()
3997                  * and then a transaction commit runs the same delayed nodes
3998                  * before any delayed worker has done something with the nodes.
3999                  * We must wait for any worker here and not at transaction
4000                  * commit time since that could cause a deadlock.
4001                  * This is a very rare case.
4002                  */
4003                 btrfs_flush_workqueue(fs_info->delayed_workers);
4004
4005                 ret = btrfs_commit_super(fs_info);
4006                 if (ret)
4007                         btrfs_err(fs_info, "commit super ret %d", ret);
4008         }
4009
4010         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4011             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4012                 btrfs_error_commit_super(fs_info);
4013
4014         kthread_stop(fs_info->transaction_kthread);
4015         kthread_stop(fs_info->cleaner_kthread);
4016
4017         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4018
4019         btrfs_free_qgroup_config(fs_info);
4020         ASSERT(list_empty(&fs_info->delalloc_roots));
4021
4022         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4023                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4024                        percpu_counter_sum(&fs_info->delalloc_bytes));
4025         }
4026
4027         btrfs_sysfs_remove_mounted(fs_info);
4028         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4029
4030         btrfs_free_fs_roots(fs_info);
4031
4032         btrfs_put_block_group_cache(fs_info);
4033
4034         /*
4035          * we must make sure there is not any read request to
4036          * submit after we stopping all workers.
4037          */
4038         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4039         btrfs_stop_all_workers(fs_info);
4040
4041         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4042         free_root_pointers(fs_info, true);
4043
4044         /*
4045          * We must free the block groups after dropping the fs_roots as we could
4046          * have had an IO error and have left over tree log blocks that aren't
4047          * cleaned up until the fs roots are freed.  This makes the block group
4048          * accounting appear to be wrong because there's pending reserved bytes,
4049          * so make sure we do the block group cleanup afterwards.
4050          */
4051         btrfs_free_block_groups(fs_info);
4052
4053         iput(fs_info->btree_inode);
4054
4055 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4056         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4057                 btrfsic_unmount(fs_info->fs_devices);
4058 #endif
4059
4060         btrfs_close_devices(fs_info->fs_devices);
4061         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4062
4063         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4064         percpu_counter_destroy(&fs_info->delalloc_bytes);
4065         percpu_counter_destroy(&fs_info->bio_counter);
4066         cleanup_srcu_struct(&fs_info->subvol_srcu);
4067
4068         btrfs_free_stripe_hash_table(fs_info);
4069         btrfs_free_ref_cache(fs_info);
4070
4071         while (!list_empty(&fs_info->pinned_chunks)) {
4072                 struct extent_map *em;
4073
4074                 em = list_first_entry(&fs_info->pinned_chunks,
4075                                       struct extent_map, list);
4076                 list_del_init(&em->list);
4077                 free_extent_map(em);
4078         }
4079 }
4080
4081 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4082                           int atomic)
4083 {
4084         int ret;
4085         struct inode *btree_inode = buf->pages[0]->mapping->host;
4086
4087         ret = extent_buffer_uptodate(buf);
4088         if (!ret)
4089                 return ret;
4090
4091         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4092                                     parent_transid, atomic);
4093         if (ret == -EAGAIN)
4094                 return ret;
4095         return !ret;
4096 }
4097
4098 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4099 {
4100         struct btrfs_fs_info *fs_info;
4101         struct btrfs_root *root;
4102         u64 transid = btrfs_header_generation(buf);
4103         int was_dirty;
4104
4105 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4106         /*
4107          * This is a fast path so only do this check if we have sanity tests
4108          * enabled.  Normal people shouldn't be using umapped buffers as dirty
4109          * outside of the sanity tests.
4110          */
4111         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4112                 return;
4113 #endif
4114         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4115         fs_info = root->fs_info;
4116         btrfs_assert_tree_locked(buf);
4117         if (transid != fs_info->generation)
4118                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4119                         buf->start, transid, fs_info->generation);
4120         was_dirty = set_extent_buffer_dirty(buf);
4121         if (!was_dirty)
4122                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4123                                          buf->len,
4124                                          fs_info->dirty_metadata_batch);
4125 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4126         /*
4127          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4128          * but item data not updated.
4129          * So here we should only check item pointers, not item data.
4130          */
4131         if (btrfs_header_level(buf) == 0 &&
4132             btrfs_check_leaf_relaxed(fs_info, buf)) {
4133                 btrfs_print_leaf(buf);
4134                 ASSERT(0);
4135         }
4136 #endif
4137 }
4138
4139 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4140                                         int flush_delayed)
4141 {
4142         /*
4143          * looks as though older kernels can get into trouble with
4144          * this code, they end up stuck in balance_dirty_pages forever
4145          */
4146         int ret;
4147
4148         if (current->flags & PF_MEMALLOC)
4149                 return;
4150
4151         if (flush_delayed)
4152                 btrfs_balance_delayed_items(fs_info);
4153
4154         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4155                                      BTRFS_DIRTY_METADATA_THRESH,
4156                                      fs_info->dirty_metadata_batch);
4157         if (ret > 0) {
4158                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4159         }
4160 }
4161
4162 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4163 {
4164         __btrfs_btree_balance_dirty(fs_info, 1);
4165 }
4166
4167 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4168 {
4169         __btrfs_btree_balance_dirty(fs_info, 0);
4170 }
4171
4172 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4173                       struct btrfs_key *first_key)
4174 {
4175         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4176         struct btrfs_fs_info *fs_info = root->fs_info;
4177
4178         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4179                                               level, first_key);
4180 }
4181
4182 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4183 {
4184         /* cleanup FS via transaction */
4185         btrfs_cleanup_transaction(fs_info);
4186
4187         mutex_lock(&fs_info->cleaner_mutex);
4188         btrfs_run_delayed_iputs(fs_info);
4189         mutex_unlock(&fs_info->cleaner_mutex);
4190
4191         down_write(&fs_info->cleanup_work_sem);
4192         up_write(&fs_info->cleanup_work_sem);
4193 }
4194
4195 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4196 {
4197         struct btrfs_ordered_extent *ordered;
4198
4199         spin_lock(&root->ordered_extent_lock);
4200         /*
4201          * This will just short circuit the ordered completion stuff which will
4202          * make sure the ordered extent gets properly cleaned up.
4203          */
4204         list_for_each_entry(ordered, &root->ordered_extents,
4205                             root_extent_list)
4206                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4207         spin_unlock(&root->ordered_extent_lock);
4208 }
4209
4210 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4211 {
4212         struct btrfs_root *root;
4213         struct list_head splice;
4214
4215         INIT_LIST_HEAD(&splice);
4216
4217         spin_lock(&fs_info->ordered_root_lock);
4218         list_splice_init(&fs_info->ordered_roots, &splice);
4219         while (!list_empty(&splice)) {
4220                 root = list_first_entry(&splice, struct btrfs_root,
4221                                         ordered_root);
4222                 list_move_tail(&root->ordered_root,
4223                                &fs_info->ordered_roots);
4224
4225                 spin_unlock(&fs_info->ordered_root_lock);
4226                 btrfs_destroy_ordered_extents(root);
4227
4228                 cond_resched();
4229                 spin_lock(&fs_info->ordered_root_lock);
4230         }
4231         spin_unlock(&fs_info->ordered_root_lock);
4232
4233         /*
4234          * We need this here because if we've been flipped read-only we won't
4235          * get sync() from the umount, so we need to make sure any ordered
4236          * extents that haven't had their dirty pages IO start writeout yet
4237          * actually get run and error out properly.
4238          */
4239         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4240 }
4241
4242 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4243                                       struct btrfs_fs_info *fs_info)
4244 {
4245         struct rb_node *node;
4246         struct btrfs_delayed_ref_root *delayed_refs;
4247         struct btrfs_delayed_ref_node *ref;
4248         int ret = 0;
4249
4250         delayed_refs = &trans->delayed_refs;
4251
4252         spin_lock(&delayed_refs->lock);
4253         if (atomic_read(&delayed_refs->num_entries) == 0) {
4254                 spin_unlock(&delayed_refs->lock);
4255                 btrfs_info(fs_info, "delayed_refs has NO entry");
4256                 return ret;
4257         }
4258
4259         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4260                 struct btrfs_delayed_ref_head *head;
4261                 struct rb_node *n;
4262                 bool pin_bytes = false;
4263
4264                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4265                                 href_node);
4266                 if (!mutex_trylock(&head->mutex)) {
4267                         refcount_inc(&head->refs);
4268                         spin_unlock(&delayed_refs->lock);
4269
4270                         mutex_lock(&head->mutex);
4271                         mutex_unlock(&head->mutex);
4272                         btrfs_put_delayed_ref_head(head);
4273                         spin_lock(&delayed_refs->lock);
4274                         continue;
4275                 }
4276                 spin_lock(&head->lock);
4277                 while ((n = rb_first(&head->ref_tree)) != NULL) {
4278                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4279                                        ref_node);
4280                         ref->in_tree = 0;
4281                         rb_erase(&ref->ref_node, &head->ref_tree);
4282                         RB_CLEAR_NODE(&ref->ref_node);
4283                         if (!list_empty(&ref->add_list))
4284                                 list_del(&ref->add_list);
4285                         atomic_dec(&delayed_refs->num_entries);
4286                         btrfs_put_delayed_ref(ref);
4287                 }
4288                 if (head->must_insert_reserved)
4289                         pin_bytes = true;
4290                 btrfs_free_delayed_extent_op(head->extent_op);
4291                 delayed_refs->num_heads--;
4292                 if (head->processing == 0)
4293                         delayed_refs->num_heads_ready--;
4294                 atomic_dec(&delayed_refs->num_entries);
4295                 rb_erase(&head->href_node, &delayed_refs->href_root);
4296                 RB_CLEAR_NODE(&head->href_node);
4297                 spin_unlock(&head->lock);
4298                 spin_unlock(&delayed_refs->lock);
4299                 mutex_unlock(&head->mutex);
4300
4301                 if (pin_bytes)
4302                         btrfs_pin_extent(fs_info, head->bytenr,
4303                                          head->num_bytes, 1);
4304                 btrfs_put_delayed_ref_head(head);
4305                 cond_resched();
4306                 spin_lock(&delayed_refs->lock);
4307         }
4308
4309         spin_unlock(&delayed_refs->lock);
4310
4311         return ret;
4312 }
4313
4314 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4315 {
4316         struct btrfs_inode *btrfs_inode;
4317         struct list_head splice;
4318
4319         INIT_LIST_HEAD(&splice);
4320
4321         spin_lock(&root->delalloc_lock);
4322         list_splice_init(&root->delalloc_inodes, &splice);
4323
4324         while (!list_empty(&splice)) {
4325                 struct inode *inode = NULL;
4326                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4327                                                delalloc_inodes);
4328                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4329                 spin_unlock(&root->delalloc_lock);
4330
4331                 /*
4332                  * Make sure we get a live inode and that it'll not disappear
4333                  * meanwhile.
4334                  */
4335                 inode = igrab(&btrfs_inode->vfs_inode);
4336                 if (inode) {
4337                         invalidate_inode_pages2(inode->i_mapping);
4338                         iput(inode);
4339                 }
4340                 spin_lock(&root->delalloc_lock);
4341         }
4342         spin_unlock(&root->delalloc_lock);
4343 }
4344
4345 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4346 {
4347         struct btrfs_root *root;
4348         struct list_head splice;
4349
4350         INIT_LIST_HEAD(&splice);
4351
4352         spin_lock(&fs_info->delalloc_root_lock);
4353         list_splice_init(&fs_info->delalloc_roots, &splice);
4354         while (!list_empty(&splice)) {
4355                 root = list_first_entry(&splice, struct btrfs_root,
4356                                          delalloc_root);
4357                 root = btrfs_grab_fs_root(root);
4358                 BUG_ON(!root);
4359                 spin_unlock(&fs_info->delalloc_root_lock);
4360
4361                 btrfs_destroy_delalloc_inodes(root);
4362                 btrfs_put_fs_root(root);
4363
4364                 spin_lock(&fs_info->delalloc_root_lock);
4365         }
4366         spin_unlock(&fs_info->delalloc_root_lock);
4367 }
4368
4369 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4370                                         struct extent_io_tree *dirty_pages,
4371                                         int mark)
4372 {
4373         int ret;
4374         struct extent_buffer *eb;
4375         u64 start = 0;
4376         u64 end;
4377
4378         while (1) {
4379                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4380                                             mark, NULL);
4381                 if (ret)
4382                         break;
4383
4384                 clear_extent_bits(dirty_pages, start, end, mark);
4385                 while (start <= end) {
4386                         eb = find_extent_buffer(fs_info, start);
4387                         start += fs_info->nodesize;
4388                         if (!eb)
4389                                 continue;
4390                         wait_on_extent_buffer_writeback(eb);
4391
4392                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4393                                                &eb->bflags))
4394                                 clear_extent_buffer_dirty(eb);
4395                         free_extent_buffer_stale(eb);
4396                 }
4397         }
4398
4399         return ret;
4400 }
4401
4402 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4403                                        struct extent_io_tree *pinned_extents)
4404 {
4405         struct extent_io_tree *unpin;
4406         u64 start;
4407         u64 end;
4408         int ret;
4409         bool loop = true;
4410
4411         unpin = pinned_extents;
4412 again:
4413         while (1) {
4414                 struct extent_state *cached_state = NULL;
4415
4416                 /*
4417                  * The btrfs_finish_extent_commit() may get the same range as
4418                  * ours between find_first_extent_bit and clear_extent_dirty.
4419                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4420                  * the same extent range.
4421                  */
4422                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4423                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4424                                             EXTENT_DIRTY, &cached_state);
4425                 if (ret) {
4426                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4427                         break;
4428                 }
4429
4430                 clear_extent_dirty(unpin, start, end, &cached_state);
4431                 free_extent_state(cached_state);
4432                 btrfs_error_unpin_extent_range(fs_info, start, end);
4433                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4434                 cond_resched();
4435         }
4436
4437         if (loop) {
4438                 if (unpin == &fs_info->freed_extents[0])
4439                         unpin = &fs_info->freed_extents[1];
4440                 else
4441                         unpin = &fs_info->freed_extents[0];
4442                 loop = false;
4443                 goto again;
4444         }
4445
4446         return 0;
4447 }
4448
4449 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4450 {
4451         struct inode *inode;
4452
4453         inode = cache->io_ctl.inode;
4454         if (inode) {
4455                 invalidate_inode_pages2(inode->i_mapping);
4456                 BTRFS_I(inode)->generation = 0;
4457                 cache->io_ctl.inode = NULL;
4458                 iput(inode);
4459         }
4460         ASSERT(cache->io_ctl.pages == NULL);
4461         btrfs_put_block_group(cache);
4462 }
4463
4464 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4465                              struct btrfs_fs_info *fs_info)
4466 {
4467         struct btrfs_block_group_cache *cache;
4468
4469         spin_lock(&cur_trans->dirty_bgs_lock);
4470         while (!list_empty(&cur_trans->dirty_bgs)) {
4471                 cache = list_first_entry(&cur_trans->dirty_bgs,
4472                                          struct btrfs_block_group_cache,
4473                                          dirty_list);
4474
4475                 if (!list_empty(&cache->io_list)) {
4476                         spin_unlock(&cur_trans->dirty_bgs_lock);
4477                         list_del_init(&cache->io_list);
4478                         btrfs_cleanup_bg_io(cache);
4479                         spin_lock(&cur_trans->dirty_bgs_lock);
4480                 }
4481
4482                 list_del_init(&cache->dirty_list);
4483                 spin_lock(&cache->lock);
4484                 cache->disk_cache_state = BTRFS_DC_ERROR;
4485                 spin_unlock(&cache->lock);
4486
4487                 spin_unlock(&cur_trans->dirty_bgs_lock);
4488                 btrfs_put_block_group(cache);
4489                 spin_lock(&cur_trans->dirty_bgs_lock);
4490         }
4491         spin_unlock(&cur_trans->dirty_bgs_lock);
4492
4493         /*
4494          * Refer to the definition of io_bgs member for details why it's safe
4495          * to use it without any locking
4496          */
4497         while (!list_empty(&cur_trans->io_bgs)) {
4498                 cache = list_first_entry(&cur_trans->io_bgs,
4499                                          struct btrfs_block_group_cache,
4500                                          io_list);
4501
4502                 list_del_init(&cache->io_list);
4503                 spin_lock(&cache->lock);
4504                 cache->disk_cache_state = BTRFS_DC_ERROR;
4505                 spin_unlock(&cache->lock);
4506                 btrfs_cleanup_bg_io(cache);
4507         }
4508 }
4509
4510 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4511                                    struct btrfs_fs_info *fs_info)
4512 {
4513         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4514         ASSERT(list_empty(&cur_trans->dirty_bgs));
4515         ASSERT(list_empty(&cur_trans->io_bgs));
4516
4517         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4518
4519         cur_trans->state = TRANS_STATE_COMMIT_START;
4520         wake_up(&fs_info->transaction_blocked_wait);
4521
4522         cur_trans->state = TRANS_STATE_UNBLOCKED;
4523         wake_up(&fs_info->transaction_wait);
4524
4525         btrfs_destroy_delayed_inodes(fs_info);
4526
4527         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4528                                      EXTENT_DIRTY);
4529         btrfs_destroy_pinned_extent(fs_info,
4530                                     fs_info->pinned_extents);
4531
4532         cur_trans->state =TRANS_STATE_COMPLETED;
4533         wake_up(&cur_trans->commit_wait);
4534 }
4535
4536 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4537 {
4538         struct btrfs_transaction *t;
4539
4540         mutex_lock(&fs_info->transaction_kthread_mutex);
4541
4542         spin_lock(&fs_info->trans_lock);
4543         while (!list_empty(&fs_info->trans_list)) {
4544                 t = list_first_entry(&fs_info->trans_list,
4545                                      struct btrfs_transaction, list);
4546                 if (t->state >= TRANS_STATE_COMMIT_START) {
4547                         refcount_inc(&t->use_count);
4548                         spin_unlock(&fs_info->trans_lock);
4549                         btrfs_wait_for_commit(fs_info, t->transid);
4550                         btrfs_put_transaction(t);
4551                         spin_lock(&fs_info->trans_lock);
4552                         continue;
4553                 }
4554                 if (t == fs_info->running_transaction) {
4555                         t->state = TRANS_STATE_COMMIT_DOING;
4556                         spin_unlock(&fs_info->trans_lock);
4557                         /*
4558                          * We wait for 0 num_writers since we don't hold a trans
4559                          * handle open currently for this transaction.
4560                          */
4561                         wait_event(t->writer_wait,
4562                                    atomic_read(&t->num_writers) == 0);
4563                 } else {
4564                         spin_unlock(&fs_info->trans_lock);
4565                 }
4566                 btrfs_cleanup_one_transaction(t, fs_info);
4567
4568                 spin_lock(&fs_info->trans_lock);
4569                 if (t == fs_info->running_transaction)
4570                         fs_info->running_transaction = NULL;
4571                 list_del_init(&t->list);
4572                 spin_unlock(&fs_info->trans_lock);
4573
4574                 btrfs_put_transaction(t);
4575                 trace_btrfs_transaction_commit(fs_info->tree_root);
4576                 spin_lock(&fs_info->trans_lock);
4577         }
4578         spin_unlock(&fs_info->trans_lock);
4579         btrfs_destroy_all_ordered_extents(fs_info);
4580         btrfs_destroy_delayed_inodes(fs_info);
4581         btrfs_assert_delayed_root_empty(fs_info);
4582         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4583         btrfs_destroy_all_delalloc_inodes(fs_info);
4584         mutex_unlock(&fs_info->transaction_kthread_mutex);
4585
4586         return 0;
4587 }
4588
4589 static const struct extent_io_ops btree_extent_io_ops = {
4590         /* mandatory callbacks */
4591         .submit_bio_hook = btree_submit_bio_hook,
4592         .readpage_end_io_hook = btree_readpage_end_io_hook,
4593         .readpage_io_failed_hook = btree_io_failed_hook,
4594
4595         /* optional callbacks */
4596 };