fs/btrfs/disk-io.c

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