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/workqueue.h>
  11 #include <linux/kthread.h>
  12 #include <linux/slab.h>
  13 #include <linux/migrate.h>
  14 #include <linux/ratelimit.h>
  15 #include <linux/uuid.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/crc32c.h>
  19 #include <linux/sched/mm.h>
  20 #include <asm/unaligned.h>
  21 #include <crypto/hash.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42 #include "block-group.h"
  43 #include "discard.h"
  44 #include "space-info.h"
  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 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
 102 {
 103         if (fs_info->csum_shash)
 104                 crypto_free_shash(fs_info->csum_shash);
 105 }
 106
 107 /*
 108  * async submit bios are used to offload expensive checksumming
 109  * onto the worker threads.  They checksum file and metadata bios
 110  * just before they are sent down the IO stack.
 111  */
 112 struct async_submit_bio {
 113         void *private_data;
 114         struct bio *bio;
 115         extent_submit_bio_start_t *submit_bio_start;
 116         int mirror_num;
 117         /*
 118          * bio_offset is optional, can be used if the pages in the bio
 119          * can't tell us where in the file the bio should go
 120          */
 121         u64 bio_offset;
 122         struct btrfs_work work;
 123         blk_status_t status;
 124 };
 125
 126 /*
 127  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 128  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 129  * the level the eb occupies in the tree.
 130  *
 131  * Different roots are used for different purposes and may nest inside each
 132  * other and they require separate keysets.  As lockdep keys should be
 133  * static, assign keysets according to the purpose of the root as indicated
 134  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 135  * roots have separate keysets.
 136  *
 137  * Lock-nesting across peer nodes is always done with the immediate parent
 138  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 139  * subclass to avoid triggering lockdep warning in such cases.
 140  *
 141  * The key is set by the readpage_end_io_hook after the buffer has passed
 142  * csum validation but before the pages are unlocked.  It is also set by
 143  * btrfs_init_new_buffer on freshly allocated blocks.
 144  *
 145  * We also add a check to make sure the highest level of the tree is the
 146  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 147  * needs update as well.
 148  */
 149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 150 # if BTRFS_MAX_LEVEL != 8
 151 #  error
 152 # endif
 153
 154 static struct btrfs_lockdep_keyset {
 155         u64                     id;             /* root objectid */
 156         const char              *name_stem;     /* lock name stem */
 157         char                    names[BTRFS_MAX_LEVEL + 1][20];
 158         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 159 } btrfs_lockdep_keysets[] = {
 160         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 161         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 162         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 163         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 164         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 165         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 166         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 167         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 168         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 169         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 170         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 171         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 172         { .id = 0,                              .name_stem = "tree"     },
 173 };
 174
 175 void __init btrfs_init_lockdep(void)
 176 {
 177         int i, j;
 178
 179         /* initialize lockdep class names */
 180         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 181                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 182
 183                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 184                         snprintf(ks->names[j], sizeof(ks->names[j]),
 185                                  "btrfs-%s-%02d", ks->name_stem, j);
 186         }
 187 }
 188
 189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 190                                     int level)
 191 {
 192         struct btrfs_lockdep_keyset *ks;
 193
 194         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 195
 196         /* find the matching keyset, id 0 is the default entry */
 197         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 198                 if (ks->id == objectid)
 199                         break;
 200
 201         lockdep_set_class_and_name(&eb->lock,
 202                                    &ks->keys[level], ks->names[level]);
 203 }
 204
 205 #endif
 206
 207 /*
 208  * extents on the btree inode are pretty simple, there's one extent
 209  * that covers the entire device
 210  */
 211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 212                                     struct page *page, size_t pg_offset,
 213                                     u64 start, u64 len)
 214 {
 215         struct extent_map_tree *em_tree = &inode->extent_tree;
 216         struct extent_map *em;
 217         int ret;
 218
 219         read_lock(&em_tree->lock);
 220         em = lookup_extent_mapping(em_tree, start, len);
 221         if (em) {
 222                 read_unlock(&em_tree->lock);
 223                 goto out;
 224         }
 225         read_unlock(&em_tree->lock);
 226
 227         em = alloc_extent_map();
 228         if (!em) {
 229                 em = ERR_PTR(-ENOMEM);
 230                 goto out;
 231         }
 232         em->start = 0;
 233         em->len = (u64)-1;
 234         em->block_len = (u64)-1;
 235         em->block_start = 0;
 236
 237         write_lock(&em_tree->lock);
 238         ret = add_extent_mapping(em_tree, em, 0);
 239         if (ret == -EEXIST) {
 240                 free_extent_map(em);
 241                 em = lookup_extent_mapping(em_tree, start, len);
 242                 if (!em)
 243                         em = ERR_PTR(-EIO);
 244         } else if (ret) {
 245                 free_extent_map(em);
 246                 em = ERR_PTR(ret);
 247         }
 248         write_unlock(&em_tree->lock);
 249
 250 out:
 251         return em;
 252 }
 253
 254 /*
 255  * Compute the csum of a btree block and store the result to provided buffer.
 256  */
 257 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
 258 {
 259         struct btrfs_fs_info *fs_info = buf->fs_info;
 260         const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
 261         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 262         char *kaddr;
 263         int i;
 264
 265         shash->tfm = fs_info->csum_shash;
 266         crypto_shash_init(shash);
 267         kaddr = page_address(buf->pages[0]);
 268         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
 269                             PAGE_SIZE - BTRFS_CSUM_SIZE);
 270
 271         for (i = 1; i < num_pages; i++) {
 272                 kaddr = page_address(buf->pages[i]);
 273                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 274         }
 275         memset(result, 0, BTRFS_CSUM_SIZE);
 276         crypto_shash_final(shash, result);
 277 }
 278
 279 /*
 280  * we can't consider a given block up to date unless the transid of the
 281  * block matches the transid in the parent node's pointer.  This is how we
 282  * detect blocks that either didn't get written at all or got written
 283  * in the wrong place.
 284  */
 285 static int verify_parent_transid(struct extent_io_tree *io_tree,
 286                                  struct extent_buffer *eb, u64 parent_transid,
 287                                  int atomic)
 288 {
 289         struct extent_state *cached_state = NULL;
 290         int ret;
 291         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 292
 293         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 294                 return 0;
 295
 296         if (atomic)
 297                 return -EAGAIN;
 298
 299         if (need_lock) {
 300                 btrfs_tree_read_lock(eb);
 301                 btrfs_set_lock_blocking_read(eb);
 302         }
 303
 304         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 305                          &cached_state);
 306         if (extent_buffer_uptodate(eb) &&
 307             btrfs_header_generation(eb) == parent_transid) {
 308                 ret = 0;
 309                 goto out;
 310         }
 311         btrfs_err_rl(eb->fs_info,
 312                 "parent transid verify failed on %llu wanted %llu found %llu",
 313                         eb->start,
 314                         parent_transid, btrfs_header_generation(eb));
 315         ret = 1;
 316
 317         /*
 318          * Things reading via commit roots that don't have normal protection,
 319          * like send, can have a really old block in cache that may point at a
 320          * block that has been freed and re-allocated.  So don't clear uptodate
 321          * if we find an eb that is under IO (dirty/writeback) because we could
 322          * end up reading in the stale data and then writing it back out and
 323          * making everybody very sad.
 324          */
 325         if (!extent_buffer_under_io(eb))
 326                 clear_extent_buffer_uptodate(eb);
 327 out:
 328         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 329                              &cached_state);
 330         if (need_lock)
 331                 btrfs_tree_read_unlock_blocking(eb);
 332         return ret;
 333 }
 334
 335 static bool btrfs_supported_super_csum(u16 csum_type)
 336 {
 337         switch (csum_type) {
 338         case BTRFS_CSUM_TYPE_CRC32:
 339         case BTRFS_CSUM_TYPE_XXHASH:
 340         case BTRFS_CSUM_TYPE_SHA256:
 341         case BTRFS_CSUM_TYPE_BLAKE2:
 342                 return true;
 343         default:
 344                 return false;
 345         }
 346 }
 347
 348 /*
 349  * Return 0 if the superblock checksum type matches the checksum value of that
 350  * algorithm. Pass the raw disk superblock data.
 351  */
 352 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 353                                   char *raw_disk_sb)
 354 {
 355         struct btrfs_super_block *disk_sb =
 356                 (struct btrfs_super_block *)raw_disk_sb;
 357         char result[BTRFS_CSUM_SIZE];
 358         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 359
 360         shash->tfm = fs_info->csum_shash;
 361
 362         /*
 363          * The super_block structure does not span the whole
 364          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 365          * filled with zeros and is included in the checksum.
 366          */
 367         crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
 368                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 369
 370         if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
 371                 return 1;
 372
 373         return 0;
 374 }
 375
 376 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 377                            struct btrfs_key *first_key, u64 parent_transid)
 378 {
 379         struct btrfs_fs_info *fs_info = eb->fs_info;
 380         int found_level;
 381         struct btrfs_key found_key;
 382         int ret;
 383
 384         found_level = btrfs_header_level(eb);
 385         if (found_level != level) {
 386                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 387                      KERN_ERR "BTRFS: tree level check failed\n");
 388                 btrfs_err(fs_info,
 389 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 390                           eb->start, level, found_level);
 391                 return -EIO;
 392         }
 393
 394         if (!first_key)
 395                 return 0;
 396
 397         /*
 398          * For live tree block (new tree blocks in current transaction),
 399          * we need proper lock context to avoid race, which is impossible here.
 400          * So we only checks tree blocks which is read from disk, whose
 401          * generation <= fs_info->last_trans_committed.
 402          */
 403         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 404                 return 0;
 405
 406         /* We have @first_key, so this @eb must have at least one item */
 407         if (btrfs_header_nritems(eb) == 0) {
 408                 btrfs_err(fs_info,
 409                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 410                           eb->start);
 411                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 412                 return -EUCLEAN;
 413         }
 414
 415         if (found_level)
 416                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 417         else
 418                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 419         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 420
 421         if (ret) {
 422                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 423                      KERN_ERR "BTRFS: tree first key check failed\n");
 424                 btrfs_err(fs_info,
 425 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 426                           eb->start, parent_transid, first_key->objectid,
 427                           first_key->type, first_key->offset,
 428                           found_key.objectid, found_key.type,
 429                           found_key.offset);
 430         }
 431         return ret;
 432 }
 433
 434 /*
 435  * helper to read a given tree block, doing retries as required when
 436  * the checksums don't match and we have alternate mirrors to try.
 437  *
 438  * @parent_transid:     expected transid, skip check if 0
 439  * @level:              expected level, mandatory check
 440  * @first_key:          expected key of first slot, skip check if NULL
 441  */
 442 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 443                                           u64 parent_transid, int level,
 444                                           struct btrfs_key *first_key)
 445 {
 446         struct btrfs_fs_info *fs_info = eb->fs_info;
 447         struct extent_io_tree *io_tree;
 448         int failed = 0;
 449         int ret;
 450         int num_copies = 0;
 451         int mirror_num = 0;
 452         int failed_mirror = 0;
 453
 454         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 455         while (1) {
 456                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 457                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 458                 if (!ret) {
 459                         if (verify_parent_transid(io_tree, eb,
 460                                                    parent_transid, 0))
 461                                 ret = -EIO;
 462                         else if (btrfs_verify_level_key(eb, level,
 463                                                 first_key, parent_transid))
 464                                 ret = -EUCLEAN;
 465                         else
 466                                 break;
 467                 }
 468
 469                 num_copies = btrfs_num_copies(fs_info,
 470                                               eb->start, eb->len);
 471                 if (num_copies == 1)
 472                         break;
 473
 474                 if (!failed_mirror) {
 475                         failed = 1;
 476                         failed_mirror = eb->read_mirror;
 477                 }
 478
 479                 mirror_num++;
 480                 if (mirror_num == failed_mirror)
 481                         mirror_num++;
 482
 483                 if (mirror_num > num_copies)
 484                         break;
 485         }
 486
 487         if (failed && !ret && failed_mirror)
 488                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 489
 490         return ret;
 491 }
 492
 493 /*
 494  * checksum a dirty tree block before IO.  This has extra checks to make sure
 495  * we only fill in the checksum field in the first page of a multi-page block
 496  */
 497
 498 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 499 {
 500         u64 start = page_offset(page);
 501         u64 found_start;
 502         u8 result[BTRFS_CSUM_SIZE];
 503         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 504         struct extent_buffer *eb;
 505         int ret;
 506
 507         eb = (struct extent_buffer *)page->private;
 508         if (page != eb->pages[0])
 509                 return 0;
 510
 511         found_start = btrfs_header_bytenr(eb);
 512         /*
 513          * Please do not consolidate these warnings into a single if.
 514          * It is useful to know what went wrong.
 515          */
 516         if (WARN_ON(found_start != start))
 517                 return -EUCLEAN;
 518         if (WARN_ON(!PageUptodate(page)))
 519                 return -EUCLEAN;
 520
 521         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 522                                     offsetof(struct btrfs_header, fsid),
 523                                     BTRFS_FSID_SIZE) == 0);
 524
 525         csum_tree_block(eb, result);
 526
 527         if (btrfs_header_level(eb))
 528                 ret = btrfs_check_node(eb);
 529         else
 530                 ret = btrfs_check_leaf_full(eb);
 531
 532         if (ret < 0) {
 533                 btrfs_print_tree(eb, 0);
 534                 btrfs_err(fs_info,
 535                 "block=%llu write time tree block corruption detected",
 536                           eb->start);
 537                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 538                 return ret;
 539         }
 540         write_extent_buffer(eb, result, 0, csum_size);
 541
 542         return 0;
 543 }
 544
 545 static int check_tree_block_fsid(struct extent_buffer *eb)
 546 {
 547         struct btrfs_fs_info *fs_info = eb->fs_info;
 548         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 549         u8 fsid[BTRFS_FSID_SIZE];
 550         int ret = 1;
 551
 552         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 553                            BTRFS_FSID_SIZE);
 554         while (fs_devices) {
 555                 u8 *metadata_uuid;
 556
 557                 /*
 558                  * Checking the incompat flag is only valid for the current
 559                  * fs. For seed devices it's forbidden to have their uuid
 560                  * changed so reading ->fsid in this case is fine
 561                  */
 562                 if (fs_devices == fs_info->fs_devices &&
 563                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 564                         metadata_uuid = fs_devices->metadata_uuid;
 565                 else
 566                         metadata_uuid = fs_devices->fsid;
 567
 568                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 569                         ret = 0;
 570                         break;
 571                 }
 572                 fs_devices = fs_devices->seed;
 573         }
 574         return ret;
 575 }
 576
 577 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 578                                       u64 phy_offset, struct page *page,
 579                                       u64 start, u64 end, int mirror)
 580 {
 581         u64 found_start;
 582         int found_level;
 583         struct extent_buffer *eb;
 584         struct btrfs_fs_info *fs_info;
 585         u16 csum_size;
 586         int ret = 0;
 587         u8 result[BTRFS_CSUM_SIZE];
 588         int reads_done;
 589
 590         if (!page->private)
 591                 goto out;
 592
 593         eb = (struct extent_buffer *)page->private;
 594         fs_info = eb->fs_info;
 595         csum_size = btrfs_super_csum_size(fs_info->super_copy);
 596
 597         /* the pending IO might have been the only thing that kept this buffer
 598          * in memory.  Make sure we have a ref for all this other checks
 599          */
 600         atomic_inc(&eb->refs);
 601
 602         reads_done = atomic_dec_and_test(&eb->io_pages);
 603         if (!reads_done)
 604                 goto err;
 605
 606         eb->read_mirror = mirror;
 607         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 608                 ret = -EIO;
 609                 goto err;
 610         }
 611
 612         found_start = btrfs_header_bytenr(eb);
 613         if (found_start != eb->start) {
 614                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 615                              eb->start, found_start);
 616                 ret = -EIO;
 617                 goto err;
 618         }
 619         if (check_tree_block_fsid(eb)) {
 620                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 621                              eb->start);
 622                 ret = -EIO;
 623                 goto err;
 624         }
 625         found_level = btrfs_header_level(eb);
 626         if (found_level >= BTRFS_MAX_LEVEL) {
 627                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 628                           (int)btrfs_header_level(eb), eb->start);
 629                 ret = -EIO;
 630                 goto err;
 631         }
 632
 633         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 634                                        eb, found_level);
 635
 636         csum_tree_block(eb, result);
 637
 638         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 639                 u32 val;
 640                 u32 found = 0;
 641
 642                 memcpy(&found, result, csum_size);
 643
 644                 read_extent_buffer(eb, &val, 0, csum_size);
 645                 btrfs_warn_rl(fs_info,
 646                 "%s checksum verify failed on %llu wanted %x found %x level %d",
 647                               fs_info->sb->s_id, eb->start,
 648                               val, found, btrfs_header_level(eb));
 649                 ret = -EUCLEAN;
 650                 goto err;
 651         }
 652
 653         /*
 654          * If this is a leaf block and it is corrupt, set the corrupt bit so
 655          * that we don't try and read the other copies of this block, just
 656          * return -EIO.
 657          */
 658         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 659                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 660                 ret = -EIO;
 661         }
 662
 663         if (found_level > 0 && btrfs_check_node(eb))
 664                 ret = -EIO;
 665
 666         if (!ret)
 667                 set_extent_buffer_uptodate(eb);
 668         else
 669                 btrfs_err(fs_info,
 670                           "block=%llu read time tree block corruption detected",
 671                           eb->start);
 672 err:
 673         if (reads_done &&
 674             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 675                 btree_readahead_hook(eb, ret);
 676
 677         if (ret) {
 678                 /*
 679                  * our io error hook is going to dec the io pages
 680                  * again, we have to make sure it has something
 681                  * to decrement
 682                  */
 683                 atomic_inc(&eb->io_pages);
 684                 clear_extent_buffer_uptodate(eb);
 685         }
 686         free_extent_buffer(eb);
 687 out:
 688         return ret;
 689 }
 690
 691 static void end_workqueue_bio(struct bio *bio)
 692 {
 693         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 694         struct btrfs_fs_info *fs_info;
 695         struct btrfs_workqueue *wq;
 696
 697         fs_info = end_io_wq->info;
 698         end_io_wq->status = bio->bi_status;
 699
 700         if (bio_op(bio) == REQ_OP_WRITE) {
 701                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 702                         wq = fs_info->endio_meta_write_workers;
 703                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 704                         wq = fs_info->endio_freespace_worker;
 705                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 706                         wq = fs_info->endio_raid56_workers;
 707                 else
 708                         wq = fs_info->endio_write_workers;
 709         } else {
 710                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 711                         wq = fs_info->endio_raid56_workers;
 712                 else if (end_io_wq->metadata)
 713                         wq = fs_info->endio_meta_workers;
 714                 else
 715                         wq = fs_info->endio_workers;
 716         }
 717
 718         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 719         btrfs_queue_work(wq, &end_io_wq->work);
 720 }
 721
 722 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 723                         enum btrfs_wq_endio_type metadata)
 724 {
 725         struct btrfs_end_io_wq *end_io_wq;
 726
 727         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 728         if (!end_io_wq)
 729                 return BLK_STS_RESOURCE;
 730
 731         end_io_wq->private = bio->bi_private;
 732         end_io_wq->end_io = bio->bi_end_io;
 733         end_io_wq->info = info;
 734         end_io_wq->status = 0;
 735         end_io_wq->bio = bio;
 736         end_io_wq->metadata = metadata;
 737
 738         bio->bi_private = end_io_wq;
 739         bio->bi_end_io = end_workqueue_bio;
 740         return 0;
 741 }
 742
 743 static void run_one_async_start(struct btrfs_work *work)
 744 {
 745         struct async_submit_bio *async;
 746         blk_status_t ret;
 747
 748         async = container_of(work, struct  async_submit_bio, work);
 749         ret = async->submit_bio_start(async->private_data, async->bio,
 750                                       async->bio_offset);
 751         if (ret)
 752                 async->status = ret;
 753 }
 754
 755 /*
 756  * In order to insert checksums into the metadata in large chunks, we wait
 757  * until bio submission time.   All the pages in the bio are checksummed and
 758  * sums are attached onto the ordered extent record.
 759  *
 760  * At IO completion time the csums attached on the ordered extent record are
 761  * inserted into the tree.
 762  */
 763 static void run_one_async_done(struct btrfs_work *work)
 764 {
 765         struct async_submit_bio *async;
 766         struct inode *inode;
 767         blk_status_t ret;
 768
 769         async = container_of(work, struct  async_submit_bio, work);
 770         inode = async->private_data;
 771
 772         /* If an error occurred we just want to clean up the bio and move on */
 773         if (async->status) {
 774                 async->bio->bi_status = async->status;
 775                 bio_endio(async->bio);
 776                 return;
 777         }
 778
 779         /*
 780          * All of the bios that pass through here are from async helpers.
 781          * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
 782          * This changes nothing when cgroups aren't in use.
 783          */
 784         async->bio->bi_opf |= REQ_CGROUP_PUNT;
 785         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
 786         if (ret) {
 787                 async->bio->bi_status = ret;
 788                 bio_endio(async->bio);
 789         }
 790 }
 791
 792 static void run_one_async_free(struct btrfs_work *work)
 793 {
 794         struct async_submit_bio *async;
 795
 796         async = container_of(work, struct  async_submit_bio, work);
 797         kfree(async);
 798 }
 799
 800 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 801                                  int mirror_num, unsigned long bio_flags,
 802                                  u64 bio_offset, void *private_data,
 803                                  extent_submit_bio_start_t *submit_bio_start)
 804 {
 805         struct async_submit_bio *async;
 806
 807         async = kmalloc(sizeof(*async), GFP_NOFS);
 808         if (!async)
 809                 return BLK_STS_RESOURCE;
 810
 811         async->private_data = private_data;
 812         async->bio = bio;
 813         async->mirror_num = mirror_num;
 814         async->submit_bio_start = submit_bio_start;
 815
 816         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 817                         run_one_async_free);
 818
 819         async->bio_offset = bio_offset;
 820
 821         async->status = 0;
 822
 823         if (op_is_sync(bio->bi_opf))
 824                 btrfs_set_work_high_priority(&async->work);
 825
 826         btrfs_queue_work(fs_info->workers, &async->work);
 827         return 0;
 828 }
 829
 830 static blk_status_t btree_csum_one_bio(struct bio *bio)
 831 {
 832         struct bio_vec *bvec;
 833         struct btrfs_root *root;
 834         int ret = 0;
 835         struct bvec_iter_all iter_all;
 836
 837         ASSERT(!bio_flagged(bio, BIO_CLONED));
 838         bio_for_each_segment_all(bvec, bio, iter_all) {
 839                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 840                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 841                 if (ret)
 842                         break;
 843         }
 844
 845         return errno_to_blk_status(ret);
 846 }
 847
 848 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 849                                              u64 bio_offset)
 850 {
 851         /*
 852          * when we're called for a write, we're already in the async
 853          * submission context.  Just jump into btrfs_map_bio
 854          */
 855         return btree_csum_one_bio(bio);
 856 }
 857
 858 static int check_async_write(struct btrfs_fs_info *fs_info,
 859                              struct btrfs_inode *bi)
 860 {
 861         if (atomic_read(&bi->sync_writers))
 862                 return 0;
 863         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 864                 return 0;
 865         return 1;
 866 }
 867
 868 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 869                                           int mirror_num,
 870                                           unsigned long bio_flags)
 871 {
 872         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 873         int async = check_async_write(fs_info, BTRFS_I(inode));
 874         blk_status_t ret;
 875
 876         if (bio_op(bio) != REQ_OP_WRITE) {
 877                 /*
 878                  * called for a read, do the setup so that checksum validation
 879                  * can happen in the async kernel threads
 880                  */
 881                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 882                                           BTRFS_WQ_ENDIO_METADATA);
 883                 if (ret)
 884                         goto out_w_error;
 885                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 886         } else if (!async) {
 887                 ret = btree_csum_one_bio(bio);
 888                 if (ret)
 889                         goto out_w_error;
 890                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 891         } else {
 892                 /*
 893                  * kthread helpers are used to submit writes so that
 894                  * checksumming can happen in parallel across all CPUs
 895                  */
 896                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 897                                           0, inode, btree_submit_bio_start);
 898         }
 899
 900         if (ret)
 901                 goto out_w_error;
 902         return 0;
 903
 904 out_w_error:
 905         bio->bi_status = ret;
 906         bio_endio(bio);
 907         return ret;
 908 }
 909
 910 #ifdef CONFIG_MIGRATION
 911 static int btree_migratepage(struct address_space *mapping,
 912                         struct page *newpage, struct page *page,
 913                         enum migrate_mode mode)
 914 {
 915         /*
 916          * we can't safely write a btree page from here,
 917          * we haven't done the locking hook
 918          */
 919         if (PageDirty(page))
 920                 return -EAGAIN;
 921         /*
 922          * Buffers may be managed in a filesystem specific way.
 923          * We must have no buffers or drop them.
 924          */
 925         if (page_has_private(page) &&
 926             !try_to_release_page(page, GFP_KERNEL))
 927                 return -EAGAIN;
 928         return migrate_page(mapping, newpage, page, mode);
 929 }
 930 #endif
 931
 932
 933 static int btree_writepages(struct address_space *mapping,
 934                             struct writeback_control *wbc)
 935 {
 936         struct btrfs_fs_info *fs_info;
 937         int ret;
 938
 939         if (wbc->sync_mode == WB_SYNC_NONE) {
 940
 941                 if (wbc->for_kupdate)
 942                         return 0;
 943
 944                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 945                 /* this is a bit racy, but that's ok */
 946                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 947                                              BTRFS_DIRTY_METADATA_THRESH,
 948                                              fs_info->dirty_metadata_batch);
 949                 if (ret < 0)
 950                         return 0;
 951         }
 952         return btree_write_cache_pages(mapping, wbc);
 953 }
 954
 955 static int btree_readpage(struct file *file, struct page *page)
 956 {
 957         return extent_read_full_page(page, btree_get_extent, 0);
 958 }
 959
 960 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 961 {
 962         if (PageWriteback(page) || PageDirty(page))
 963                 return 0;
 964
 965         return try_release_extent_buffer(page);
 966 }
 967
 968 static void btree_invalidatepage(struct page *page, unsigned int offset,
 969                                  unsigned int length)
 970 {
 971         struct extent_io_tree *tree;
 972         tree = &BTRFS_I(page->mapping->host)->io_tree;
 973         extent_invalidatepage(tree, page, offset);
 974         btree_releasepage(page, GFP_NOFS);
 975         if (PagePrivate(page)) {
 976                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 977                            "page private not zero on page %llu",
 978                            (unsigned long long)page_offset(page));
 979                 detach_page_private(page);
 980         }
 981 }
 982
 983 static int btree_set_page_dirty(struct page *page)
 984 {
 985 #ifdef DEBUG
 986         struct extent_buffer *eb;
 987
 988         BUG_ON(!PagePrivate(page));
 989         eb = (struct extent_buffer *)page->private;
 990         BUG_ON(!eb);
 991         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 992         BUG_ON(!atomic_read(&eb->refs));
 993         btrfs_assert_tree_locked(eb);
 994 #endif
 995         return __set_page_dirty_nobuffers(page);
 996 }
 997
 998 static const struct address_space_operations btree_aops = {
 999         .readpage       = btree_readpage,
1000         .writepages     = btree_writepages,
1001         .releasepage    = btree_releasepage,
1002         .invalidatepage = btree_invalidatepage,
1003 #ifdef CONFIG_MIGRATION
1004         .migratepage    = btree_migratepage,
1005 #endif
1006         .set_page_dirty = btree_set_page_dirty,
1007 };
1008
1009 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1010 {
1011         struct extent_buffer *buf = NULL;
1012         int ret;
1013
1014         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1015         if (IS_ERR(buf))
1016                 return;
1017
1018         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1019         if (ret < 0)
1020                 free_extent_buffer_stale(buf);
1021         else
1022                 free_extent_buffer(buf);
1023 }
1024
1025 struct extent_buffer *btrfs_find_create_tree_block(
1026                                                 struct btrfs_fs_info *fs_info,
1027                                                 u64 bytenr)
1028 {
1029         if (btrfs_is_testing(fs_info))
1030                 return alloc_test_extent_buffer(fs_info, bytenr);
1031         return alloc_extent_buffer(fs_info, bytenr);
1032 }
1033
1034 /*
1035  * Read tree block at logical address @bytenr and do variant basic but critical
1036  * verification.
1037  *
1038  * @parent_transid:     expected transid of this tree block, skip check if 0
1039  * @level:              expected level, mandatory check
1040  * @first_key:          expected key in slot 0, skip check if NULL
1041  */
1042 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1043                                       u64 parent_transid, int level,
1044                                       struct btrfs_key *first_key)
1045 {
1046         struct extent_buffer *buf = NULL;
1047         int ret;
1048
1049         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1050         if (IS_ERR(buf))
1051                 return buf;
1052
1053         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1054                                              level, first_key);
1055         if (ret) {
1056                 free_extent_buffer_stale(buf);
1057                 return ERR_PTR(ret);
1058         }
1059         return buf;
1060
1061 }
1062
1063 void btrfs_clean_tree_block(struct extent_buffer *buf)
1064 {
1065         struct btrfs_fs_info *fs_info = buf->fs_info;
1066         if (btrfs_header_generation(buf) ==
1067             fs_info->running_transaction->transid) {
1068                 btrfs_assert_tree_locked(buf);
1069
1070                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1071                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1072                                                  -buf->len,
1073                                                  fs_info->dirty_metadata_batch);
1074                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1075                         btrfs_set_lock_blocking_write(buf);
1076                         clear_extent_buffer_dirty(buf);
1077                 }
1078         }
1079 }
1080
1081 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1082                          u64 objectid)
1083 {
1084         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1085         root->fs_info = fs_info;
1086         root->node = NULL;
1087         root->commit_root = NULL;
1088         root->state = 0;
1089         root->orphan_cleanup_state = 0;
1090
1091         root->last_trans = 0;
1092         root->highest_objectid = 0;
1093         root->nr_delalloc_inodes = 0;
1094         root->nr_ordered_extents = 0;
1095         root->inode_tree = RB_ROOT;
1096         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1097         root->block_rsv = NULL;
1098
1099         INIT_LIST_HEAD(&root->dirty_list);
1100         INIT_LIST_HEAD(&root->root_list);
1101         INIT_LIST_HEAD(&root->delalloc_inodes);
1102         INIT_LIST_HEAD(&root->delalloc_root);
1103         INIT_LIST_HEAD(&root->ordered_extents);
1104         INIT_LIST_HEAD(&root->ordered_root);
1105         INIT_LIST_HEAD(&root->reloc_dirty_list);
1106         INIT_LIST_HEAD(&root->logged_list[0]);
1107         INIT_LIST_HEAD(&root->logged_list[1]);
1108         spin_lock_init(&root->inode_lock);
1109         spin_lock_init(&root->delalloc_lock);
1110         spin_lock_init(&root->ordered_extent_lock);
1111         spin_lock_init(&root->accounting_lock);
1112         spin_lock_init(&root->log_extents_lock[0]);
1113         spin_lock_init(&root->log_extents_lock[1]);
1114         spin_lock_init(&root->qgroup_meta_rsv_lock);
1115         mutex_init(&root->objectid_mutex);
1116         mutex_init(&root->log_mutex);
1117         mutex_init(&root->ordered_extent_mutex);
1118         mutex_init(&root->delalloc_mutex);
1119         init_waitqueue_head(&root->log_writer_wait);
1120         init_waitqueue_head(&root->log_commit_wait[0]);
1121         init_waitqueue_head(&root->log_commit_wait[1]);
1122         INIT_LIST_HEAD(&root->log_ctxs[0]);
1123         INIT_LIST_HEAD(&root->log_ctxs[1]);
1124         atomic_set(&root->log_commit[0], 0);
1125         atomic_set(&root->log_commit[1], 0);
1126         atomic_set(&root->log_writers, 0);
1127         atomic_set(&root->log_batch, 0);
1128         refcount_set(&root->refs, 1);
1129         atomic_set(&root->snapshot_force_cow, 0);
1130         atomic_set(&root->nr_swapfiles, 0);
1131         root->log_transid = 0;
1132         root->log_transid_committed = -1;
1133         root->last_log_commit = 0;
1134         if (!dummy) {
1135                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1136                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1137                 extent_io_tree_init(fs_info, &root->log_csum_range,
1138                                     IO_TREE_LOG_CSUM_RANGE, NULL);
1139         }
1140
1141         memset(&root->root_key, 0, sizeof(root->root_key));
1142         memset(&root->root_item, 0, sizeof(root->root_item));
1143         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1144         if (!dummy)
1145                 root->defrag_trans_start = fs_info->generation;
1146         else
1147                 root->defrag_trans_start = 0;
1148         root->root_key.objectid = objectid;
1149         root->anon_dev = 0;
1150
1151         spin_lock_init(&root->root_item_lock);
1152         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1153 #ifdef CONFIG_BTRFS_DEBUG
1154         INIT_LIST_HEAD(&root->leak_list);
1155         spin_lock(&fs_info->fs_roots_radix_lock);
1156         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1157         spin_unlock(&fs_info->fs_roots_radix_lock);
1158 #endif
1159 }
1160
1161 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1162                                            u64 objectid, gfp_t flags)
1163 {
1164         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1165         if (root)
1166                 __setup_root(root, fs_info, objectid);
1167         return root;
1168 }
1169
1170 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1171 /* Should only be used by the testing infrastructure */
1172 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1173 {
1174         struct btrfs_root *root;
1175
1176         if (!fs_info)
1177                 return ERR_PTR(-EINVAL);
1178
1179         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1180         if (!root)
1181                 return ERR_PTR(-ENOMEM);
1182
1183         /* We don't use the stripesize in selftest, set it as sectorsize */
1184         root->alloc_bytenr = 0;
1185
1186         return root;
1187 }
1188 #endif
1189
1190 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1191                                      u64 objectid)
1192 {
1193         struct btrfs_fs_info *fs_info = trans->fs_info;
1194         struct extent_buffer *leaf;
1195         struct btrfs_root *tree_root = fs_info->tree_root;
1196         struct btrfs_root *root;
1197         struct btrfs_key key;
1198         unsigned int nofs_flag;
1199         int ret = 0;
1200
1201         /*
1202          * We're holding a transaction handle, so use a NOFS memory allocation
1203          * context to avoid deadlock if reclaim happens.
1204          */
1205         nofs_flag = memalloc_nofs_save();
1206         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1207         memalloc_nofs_restore(nofs_flag);
1208         if (!root)
1209                 return ERR_PTR(-ENOMEM);
1210
1211         root->root_key.objectid = objectid;
1212         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1213         root->root_key.offset = 0;
1214
1215         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1216         if (IS_ERR(leaf)) {
1217                 ret = PTR_ERR(leaf);
1218                 leaf = NULL;
1219                 goto fail;
1220         }
1221
1222         root->node = leaf;
1223         btrfs_mark_buffer_dirty(leaf);
1224
1225         root->commit_root = btrfs_root_node(root);
1226         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1227
1228         root->root_item.flags = 0;
1229         root->root_item.byte_limit = 0;
1230         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1231         btrfs_set_root_generation(&root->root_item, trans->transid);
1232         btrfs_set_root_level(&root->root_item, 0);
1233         btrfs_set_root_refs(&root->root_item, 1);
1234         btrfs_set_root_used(&root->root_item, leaf->len);
1235         btrfs_set_root_last_snapshot(&root->root_item, 0);
1236         btrfs_set_root_dirid(&root->root_item, 0);
1237         if (is_fstree(objectid))
1238                 generate_random_guid(root->root_item.uuid);
1239         else
1240                 export_guid(root->root_item.uuid, &guid_null);
1241         root->root_item.drop_level = 0;
1242
1243         key.objectid = objectid;
1244         key.type = BTRFS_ROOT_ITEM_KEY;
1245         key.offset = 0;
1246         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1247         if (ret)
1248                 goto fail;
1249
1250         btrfs_tree_unlock(leaf);
1251
1252         return root;
1253
1254 fail:
1255         if (leaf)
1256                 btrfs_tree_unlock(leaf);
1257         btrfs_put_root(root);
1258
1259         return ERR_PTR(ret);
1260 }
1261
1262 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1263                                          struct btrfs_fs_info *fs_info)
1264 {
1265         struct btrfs_root *root;
1266         struct extent_buffer *leaf;
1267
1268         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1269         if (!root)
1270                 return ERR_PTR(-ENOMEM);
1271
1272         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1273         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1274         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1275
1276         /*
1277          * DON'T set SHAREABLE bit for log trees.
1278          *
1279          * Log trees are not exposed to user space thus can't be snapshotted,
1280          * and they go away before a real commit is actually done.
1281          *
1282          * They do store pointers to file data extents, and those reference
1283          * counts still get updated (along with back refs to the log tree).
1284          */
1285
1286         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1287                         NULL, 0, 0, 0);
1288         if (IS_ERR(leaf)) {
1289                 btrfs_put_root(root);
1290                 return ERR_CAST(leaf);
1291         }
1292
1293         root->node = leaf;
1294
1295         btrfs_mark_buffer_dirty(root->node);
1296         btrfs_tree_unlock(root->node);
1297         return root;
1298 }
1299
1300 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1301                              struct btrfs_fs_info *fs_info)
1302 {
1303         struct btrfs_root *log_root;
1304
1305         log_root = alloc_log_tree(trans, fs_info);
1306         if (IS_ERR(log_root))
1307                 return PTR_ERR(log_root);
1308         WARN_ON(fs_info->log_root_tree);
1309         fs_info->log_root_tree = log_root;
1310         return 0;
1311 }
1312
1313 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1314                        struct btrfs_root *root)
1315 {
1316         struct btrfs_fs_info *fs_info = root->fs_info;
1317         struct btrfs_root *log_root;
1318         struct btrfs_inode_item *inode_item;
1319
1320         log_root = alloc_log_tree(trans, fs_info);
1321         if (IS_ERR(log_root))
1322                 return PTR_ERR(log_root);
1323
1324         log_root->last_trans = trans->transid;
1325         log_root->root_key.offset = root->root_key.objectid;
1326
1327         inode_item = &log_root->root_item.inode;
1328         btrfs_set_stack_inode_generation(inode_item, 1);
1329         btrfs_set_stack_inode_size(inode_item, 3);
1330         btrfs_set_stack_inode_nlink(inode_item, 1);
1331         btrfs_set_stack_inode_nbytes(inode_item,
1332                                      fs_info->nodesize);
1333         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1334
1335         btrfs_set_root_node(&log_root->root_item, log_root->node);
1336
1337         WARN_ON(root->log_root);
1338         root->log_root = log_root;
1339         root->log_transid = 0;
1340         root->log_transid_committed = -1;
1341         root->last_log_commit = 0;
1342         return 0;
1343 }
1344
1345 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1346                                         struct btrfs_key *key)
1347 {
1348         struct btrfs_root *root;
1349         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1350         struct btrfs_path *path;
1351         u64 generation;
1352         int ret;
1353         int level;
1354
1355         path = btrfs_alloc_path();
1356         if (!path)
1357                 return ERR_PTR(-ENOMEM);
1358
1359         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1360         if (!root) {
1361                 ret = -ENOMEM;
1362                 goto alloc_fail;
1363         }
1364
1365         ret = btrfs_find_root(tree_root, key, path,
1366                               &root->root_item, &root->root_key);
1367         if (ret) {
1368                 if (ret > 0)
1369                         ret = -ENOENT;
1370                 goto find_fail;
1371         }
1372
1373         generation = btrfs_root_generation(&root->root_item);
1374         level = btrfs_root_level(&root->root_item);
1375         root->node = read_tree_block(fs_info,
1376                                      btrfs_root_bytenr(&root->root_item),
1377                                      generation, level, NULL);
1378         if (IS_ERR(root->node)) {
1379                 ret = PTR_ERR(root->node);
1380                 root->node = NULL;
1381                 goto find_fail;
1382         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1383                 ret = -EIO;
1384                 goto find_fail;
1385         }
1386         root->commit_root = btrfs_root_node(root);
1387 out:
1388         btrfs_free_path(path);
1389         return root;
1390
1391 find_fail:
1392         btrfs_put_root(root);
1393 alloc_fail:
1394         root = ERR_PTR(ret);
1395         goto out;
1396 }
1397
1398 static int btrfs_init_fs_root(struct btrfs_root *root)
1399 {
1400         int ret;
1401         unsigned int nofs_flag;
1402
1403         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1404         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1405                                         GFP_NOFS);
1406         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1407                 ret = -ENOMEM;
1408                 goto fail;
1409         }
1410
1411         /*
1412          * We might be called under a transaction (e.g. indirect backref
1413          * resolution) which could deadlock if it triggers memory reclaim
1414          */
1415         nofs_flag = memalloc_nofs_save();
1416         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1417         memalloc_nofs_restore(nofs_flag);
1418         if (ret)
1419                 goto fail;
1420
1421         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1422             root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1423                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1424                 btrfs_check_and_init_root_item(&root->root_item);
1425         }
1426
1427         btrfs_init_free_ino_ctl(root);
1428         spin_lock_init(&root->ino_cache_lock);
1429         init_waitqueue_head(&root->ino_cache_wait);
1430
1431         ret = get_anon_bdev(&root->anon_dev);
1432         if (ret)
1433                 goto fail;
1434
1435         mutex_lock(&root->objectid_mutex);
1436         ret = btrfs_find_highest_objectid(root,
1437                                         &root->highest_objectid);
1438         if (ret) {
1439                 mutex_unlock(&root->objectid_mutex);
1440                 goto fail;
1441         }
1442
1443         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1444
1445         mutex_unlock(&root->objectid_mutex);
1446
1447         return 0;
1448 fail:
1449         /* The caller is responsible to call btrfs_free_fs_root */
1450         return ret;
1451 }
1452
1453 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1454                                                u64 root_id)
1455 {
1456         struct btrfs_root *root;
1457
1458         spin_lock(&fs_info->fs_roots_radix_lock);
1459         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1460                                  (unsigned long)root_id);
1461         if (root)
1462                 root = btrfs_grab_root(root);
1463         spin_unlock(&fs_info->fs_roots_radix_lock);
1464         return root;
1465 }
1466
1467 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1468                          struct btrfs_root *root)
1469 {
1470         int ret;
1471
1472         ret = radix_tree_preload(GFP_NOFS);
1473         if (ret)
1474                 return ret;
1475
1476         spin_lock(&fs_info->fs_roots_radix_lock);
1477         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1478                                 (unsigned long)root->root_key.objectid,
1479                                 root);
1480         if (ret == 0) {
1481                 btrfs_grab_root(root);
1482                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1483         }
1484         spin_unlock(&fs_info->fs_roots_radix_lock);
1485         radix_tree_preload_end();
1486
1487         return ret;
1488 }
1489
1490 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1491 {
1492 #ifdef CONFIG_BTRFS_DEBUG
1493         struct btrfs_root *root;
1494
1495         while (!list_empty(&fs_info->allocated_roots)) {
1496                 root = list_first_entry(&fs_info->allocated_roots,
1497                                         struct btrfs_root, leak_list);
1498                 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1499                           root->root_key.objectid, root->root_key.offset,
1500                           refcount_read(&root->refs));
1501                 while (refcount_read(&root->refs) > 1)
1502                         btrfs_put_root(root);
1503                 btrfs_put_root(root);
1504         }
1505 #endif
1506 }
1507
1508 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1509 {
1510         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1511         percpu_counter_destroy(&fs_info->delalloc_bytes);
1512         percpu_counter_destroy(&fs_info->dio_bytes);
1513         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1514         btrfs_free_csum_hash(fs_info);
1515         btrfs_free_stripe_hash_table(fs_info);
1516         btrfs_free_ref_cache(fs_info);
1517         kfree(fs_info->balance_ctl);
1518         kfree(fs_info->delayed_root);
1519         btrfs_put_root(fs_info->extent_root);
1520         btrfs_put_root(fs_info->tree_root);
1521         btrfs_put_root(fs_info->chunk_root);
1522         btrfs_put_root(fs_info->dev_root);
1523         btrfs_put_root(fs_info->csum_root);
1524         btrfs_put_root(fs_info->quota_root);
1525         btrfs_put_root(fs_info->uuid_root);
1526         btrfs_put_root(fs_info->free_space_root);
1527         btrfs_put_root(fs_info->fs_root);
1528         btrfs_put_root(fs_info->data_reloc_root);
1529         btrfs_check_leaked_roots(fs_info);
1530         btrfs_extent_buffer_leak_debug_check(fs_info);
1531         kfree(fs_info->super_copy);
1532         kfree(fs_info->super_for_commit);
1533         kvfree(fs_info);
1534 }
1535
1536
1537 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1538                                      u64 objectid, bool check_ref)
1539 {
1540         struct btrfs_root *root;
1541         struct btrfs_path *path;
1542         struct btrfs_key key;
1543         int ret;
1544
1545         if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1546                 return btrfs_grab_root(fs_info->tree_root);
1547         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1548                 return btrfs_grab_root(fs_info->extent_root);
1549         if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1550                 return btrfs_grab_root(fs_info->chunk_root);
1551         if (objectid == BTRFS_DEV_TREE_OBJECTID)
1552                 return btrfs_grab_root(fs_info->dev_root);
1553         if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1554                 return btrfs_grab_root(fs_info->csum_root);
1555         if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1556                 return btrfs_grab_root(fs_info->quota_root) ?
1557                         fs_info->quota_root : ERR_PTR(-ENOENT);
1558         if (objectid == BTRFS_UUID_TREE_OBJECTID)
1559                 return btrfs_grab_root(fs_info->uuid_root) ?
1560                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1561         if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1562                 return btrfs_grab_root(fs_info->free_space_root) ?
1563                         fs_info->free_space_root : ERR_PTR(-ENOENT);
1564 again:
1565         root = btrfs_lookup_fs_root(fs_info, objectid);
1566         if (root) {
1567                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1568                         btrfs_put_root(root);
1569                         return ERR_PTR(-ENOENT);
1570                 }
1571                 return root;
1572         }
1573
1574         key.objectid = objectid;
1575         key.type = BTRFS_ROOT_ITEM_KEY;
1576         key.offset = (u64)-1;
1577         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1578         if (IS_ERR(root))
1579                 return root;
1580
1581         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1582                 ret = -ENOENT;
1583                 goto fail;
1584         }
1585
1586         ret = btrfs_init_fs_root(root);
1587         if (ret)
1588                 goto fail;
1589
1590         path = btrfs_alloc_path();
1591         if (!path) {
1592                 ret = -ENOMEM;
1593                 goto fail;
1594         }
1595         key.objectid = BTRFS_ORPHAN_OBJECTID;
1596         key.type = BTRFS_ORPHAN_ITEM_KEY;
1597         key.offset = objectid;
1598
1599         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1600         btrfs_free_path(path);
1601         if (ret < 0)
1602                 goto fail;
1603         if (ret == 0)
1604                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1605
1606         ret = btrfs_insert_fs_root(fs_info, root);
1607         if (ret) {
1608                 btrfs_put_root(root);
1609                 if (ret == -EEXIST)
1610                         goto again;
1611                 goto fail;
1612         }
1613         return root;
1614 fail:
1615         btrfs_put_root(root);
1616         return ERR_PTR(ret);
1617 }
1618
1619 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1620 {
1621         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1622         int ret = 0;
1623         struct btrfs_device *device;
1624         struct backing_dev_info *bdi;
1625
1626         rcu_read_lock();
1627         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1628                 if (!device->bdev)
1629                         continue;
1630                 bdi = device->bdev->bd_bdi;
1631                 if (bdi_congested(bdi, bdi_bits)) {
1632                         ret = 1;
1633                         break;
1634                 }
1635         }
1636         rcu_read_unlock();
1637         return ret;
1638 }
1639
1640 /*
1641  * called by the kthread helper functions to finally call the bio end_io
1642  * functions.  This is where read checksum verification actually happens
1643  */
1644 static void end_workqueue_fn(struct btrfs_work *work)
1645 {
1646         struct bio *bio;
1647         struct btrfs_end_io_wq *end_io_wq;
1648
1649         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1650         bio = end_io_wq->bio;
1651
1652         bio->bi_status = end_io_wq->status;
1653         bio->bi_private = end_io_wq->private;
1654         bio->bi_end_io = end_io_wq->end_io;
1655         bio_endio(bio);
1656         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1657 }
1658
1659 static int cleaner_kthread(void *arg)
1660 {
1661         struct btrfs_root *root = arg;
1662         struct btrfs_fs_info *fs_info = root->fs_info;
1663         int again;
1664
1665         while (1) {
1666                 again = 0;
1667
1668                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1669
1670                 /* Make the cleaner go to sleep early. */
1671                 if (btrfs_need_cleaner_sleep(fs_info))
1672                         goto sleep;
1673
1674                 /*
1675                  * Do not do anything if we might cause open_ctree() to block
1676                  * before we have finished mounting the filesystem.
1677                  */
1678                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1679                         goto sleep;
1680
1681                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1682                         goto sleep;
1683
1684                 /*
1685                  * Avoid the problem that we change the status of the fs
1686                  * during the above check and trylock.
1687                  */
1688                 if (btrfs_need_cleaner_sleep(fs_info)) {
1689                         mutex_unlock(&fs_info->cleaner_mutex);
1690                         goto sleep;
1691                 }
1692
1693                 btrfs_run_delayed_iputs(fs_info);
1694
1695                 again = btrfs_clean_one_deleted_snapshot(root);
1696                 mutex_unlock(&fs_info->cleaner_mutex);
1697
1698                 /*
1699                  * The defragger has dealt with the R/O remount and umount,
1700                  * needn't do anything special here.
1701                  */
1702                 btrfs_run_defrag_inodes(fs_info);
1703
1704                 /*
1705                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1706                  * with relocation (btrfs_relocate_chunk) and relocation
1707                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1708                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1709                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1710                  * unused block groups.
1711                  */
1712                 btrfs_delete_unused_bgs(fs_info);
1713 sleep:
1714                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1715                 if (kthread_should_park())
1716                         kthread_parkme();
1717                 if (kthread_should_stop())
1718                         return 0;
1719                 if (!again) {
1720                         set_current_state(TASK_INTERRUPTIBLE);
1721                         schedule();
1722                         __set_current_state(TASK_RUNNING);
1723                 }
1724         }
1725 }
1726
1727 static int transaction_kthread(void *arg)
1728 {
1729         struct btrfs_root *root = arg;
1730         struct btrfs_fs_info *fs_info = root->fs_info;
1731         struct btrfs_trans_handle *trans;
1732         struct btrfs_transaction *cur;
1733         u64 transid;
1734         time64_t now;
1735         unsigned long delay;
1736         bool cannot_commit;
1737
1738         do {
1739                 cannot_commit = false;
1740                 delay = HZ * fs_info->commit_interval;
1741                 mutex_lock(&fs_info->transaction_kthread_mutex);
1742
1743                 spin_lock(&fs_info->trans_lock);
1744                 cur = fs_info->running_transaction;
1745                 if (!cur) {
1746                         spin_unlock(&fs_info->trans_lock);
1747                         goto sleep;
1748                 }
1749
1750                 now = ktime_get_seconds();
1751                 if (cur->state < TRANS_STATE_COMMIT_START &&
1752                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1753                     (now < cur->start_time ||
1754                      now - cur->start_time < fs_info->commit_interval)) {
1755                         spin_unlock(&fs_info->trans_lock);
1756                         delay = HZ * 5;
1757                         goto sleep;
1758                 }
1759                 transid = cur->transid;
1760                 spin_unlock(&fs_info->trans_lock);
1761
1762                 /* If the file system is aborted, this will always fail. */
1763                 trans = btrfs_attach_transaction(root);
1764                 if (IS_ERR(trans)) {
1765                         if (PTR_ERR(trans) != -ENOENT)
1766                                 cannot_commit = true;
1767                         goto sleep;
1768                 }
1769                 if (transid == trans->transid) {
1770                         btrfs_commit_transaction(trans);
1771                 } else {
1772                         btrfs_end_transaction(trans);
1773                 }
1774 sleep:
1775                 wake_up_process(fs_info->cleaner_kthread);
1776                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1777
1778                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1779                                       &fs_info->fs_state)))
1780                         btrfs_cleanup_transaction(fs_info);
1781                 if (!kthread_should_stop() &&
1782                                 (!btrfs_transaction_blocked(fs_info) ||
1783                                  cannot_commit))
1784                         schedule_timeout_interruptible(delay);
1785         } while (!kthread_should_stop());
1786         return 0;
1787 }
1788
1789 /*
1790  * This will find the highest generation in the array of root backups.  The
1791  * index of the highest array is returned, or -EINVAL if we can't find
1792  * anything.
1793  *
1794  * We check to make sure the array is valid by comparing the
1795  * generation of the latest  root in the array with the generation
1796  * in the super block.  If they don't match we pitch it.
1797  */
1798 static int find_newest_super_backup(struct btrfs_fs_info *info)
1799 {
1800         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1801         u64 cur;
1802         struct btrfs_root_backup *root_backup;
1803         int i;
1804
1805         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1806                 root_backup = info->super_copy->super_roots + i;
1807                 cur = btrfs_backup_tree_root_gen(root_backup);
1808                 if (cur == newest_gen)
1809                         return i;
1810         }
1811
1812         return -EINVAL;
1813 }
1814
1815 /*
1816  * copy all the root pointers into the super backup array.
1817  * this will bump the backup pointer by one when it is
1818  * done
1819  */
1820 static void backup_super_roots(struct btrfs_fs_info *info)
1821 {
1822         const int next_backup = info->backup_root_index;
1823         struct btrfs_root_backup *root_backup;
1824
1825         root_backup = info->super_for_commit->super_roots + next_backup;
1826
1827         /*
1828          * make sure all of our padding and empty slots get zero filled
1829          * regardless of which ones we use today
1830          */
1831         memset(root_backup, 0, sizeof(*root_backup));
1832
1833         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1834
1835         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1836         btrfs_set_backup_tree_root_gen(root_backup,
1837                                btrfs_header_generation(info->tree_root->node));
1838
1839         btrfs_set_backup_tree_root_level(root_backup,
1840                                btrfs_header_level(info->tree_root->node));
1841
1842         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1843         btrfs_set_backup_chunk_root_gen(root_backup,
1844                                btrfs_header_generation(info->chunk_root->node));
1845         btrfs_set_backup_chunk_root_level(root_backup,
1846                                btrfs_header_level(info->chunk_root->node));
1847
1848         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1849         btrfs_set_backup_extent_root_gen(root_backup,
1850                                btrfs_header_generation(info->extent_root->node));
1851         btrfs_set_backup_extent_root_level(root_backup,
1852                                btrfs_header_level(info->extent_root->node));
1853
1854         /*
1855          * we might commit during log recovery, which happens before we set
1856          * the fs_root.  Make sure it is valid before we fill it in.
1857          */
1858         if (info->fs_root && info->fs_root->node) {
1859                 btrfs_set_backup_fs_root(root_backup,
1860                                          info->fs_root->node->start);
1861                 btrfs_set_backup_fs_root_gen(root_backup,
1862                                btrfs_header_generation(info->fs_root->node));
1863                 btrfs_set_backup_fs_root_level(root_backup,
1864                                btrfs_header_level(info->fs_root->node));
1865         }
1866
1867         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1868         btrfs_set_backup_dev_root_gen(root_backup,
1869                                btrfs_header_generation(info->dev_root->node));
1870         btrfs_set_backup_dev_root_level(root_backup,
1871                                        btrfs_header_level(info->dev_root->node));
1872
1873         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1874         btrfs_set_backup_csum_root_gen(root_backup,
1875                                btrfs_header_generation(info->csum_root->node));
1876         btrfs_set_backup_csum_root_level(root_backup,
1877                                btrfs_header_level(info->csum_root->node));
1878
1879         btrfs_set_backup_total_bytes(root_backup,
1880                              btrfs_super_total_bytes(info->super_copy));
1881         btrfs_set_backup_bytes_used(root_backup,
1882                              btrfs_super_bytes_used(info->super_copy));
1883         btrfs_set_backup_num_devices(root_backup,
1884                              btrfs_super_num_devices(info->super_copy));
1885
1886         /*
1887          * if we don't copy this out to the super_copy, it won't get remembered
1888          * for the next commit
1889          */
1890         memcpy(&info->super_copy->super_roots,
1891                &info->super_for_commit->super_roots,
1892                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1893 }
1894
1895 /*
1896  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1897  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1898  *
1899  * fs_info - filesystem whose backup roots need to be read
1900  * priority - priority of backup root required
1901  *
1902  * Returns backup root index on success and -EINVAL otherwise.
1903  */
1904 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1905 {
1906         int backup_index = find_newest_super_backup(fs_info);
1907         struct btrfs_super_block *super = fs_info->super_copy;
1908         struct btrfs_root_backup *root_backup;
1909
1910         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1911                 if (priority == 0)
1912                         return backup_index;
1913
1914                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1915                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1916         } else {
1917                 return -EINVAL;
1918         }
1919
1920         root_backup = super->super_roots + backup_index;
1921
1922         btrfs_set_super_generation(super,
1923                                    btrfs_backup_tree_root_gen(root_backup));
1924         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1925         btrfs_set_super_root_level(super,
1926                                    btrfs_backup_tree_root_level(root_backup));
1927         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1928
1929         /*
1930          * Fixme: the total bytes and num_devices need to match or we should
1931          * need a fsck
1932          */
1933         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1934         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1935
1936         return backup_index;
1937 }
1938
1939 /* helper to cleanup workers */
1940 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1941 {
1942         btrfs_destroy_workqueue(fs_info->fixup_workers);
1943         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1944         btrfs_destroy_workqueue(fs_info->workers);
1945         btrfs_destroy_workqueue(fs_info->endio_workers);
1946         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1947         btrfs_destroy_workqueue(fs_info->rmw_workers);
1948         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1949         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1950         btrfs_destroy_workqueue(fs_info->delayed_workers);
1951         btrfs_destroy_workqueue(fs_info->caching_workers);
1952         btrfs_destroy_workqueue(fs_info->readahead_workers);
1953         btrfs_destroy_workqueue(fs_info->flush_workers);
1954         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1955         if (fs_info->discard_ctl.discard_workers)
1956                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1957         /*
1958          * Now that all other work queues are destroyed, we can safely destroy
1959          * the queues used for metadata I/O, since tasks from those other work
1960          * queues can do metadata I/O operations.
1961          */
1962         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1963         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1964 }
1965
1966 static void free_root_extent_buffers(struct btrfs_root *root)
1967 {
1968         if (root) {
1969                 free_extent_buffer(root->node);
1970                 free_extent_buffer(root->commit_root);
1971                 root->node = NULL;
1972                 root->commit_root = NULL;
1973         }
1974 }
1975
1976 /* helper to cleanup tree roots */
1977 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1978 {
1979         free_root_extent_buffers(info->tree_root);
1980
1981         free_root_extent_buffers(info->dev_root);
1982         free_root_extent_buffers(info->extent_root);
1983         free_root_extent_buffers(info->csum_root);
1984         free_root_extent_buffers(info->quota_root);
1985         free_root_extent_buffers(info->uuid_root);
1986         free_root_extent_buffers(info->fs_root);
1987         free_root_extent_buffers(info->data_reloc_root);
1988         if (free_chunk_root)
1989                 free_root_extent_buffers(info->chunk_root);
1990         free_root_extent_buffers(info->free_space_root);
1991 }
1992
1993 void btrfs_put_root(struct btrfs_root *root)
1994 {
1995         if (!root)
1996                 return;
1997
1998         if (refcount_dec_and_test(&root->refs)) {
1999                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2000                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2001                 if (root->anon_dev)
2002                         free_anon_bdev(root->anon_dev);
2003                 btrfs_drew_lock_destroy(&root->snapshot_lock);
2004                 free_extent_buffer(root->node);
2005                 free_extent_buffer(root->commit_root);
2006                 kfree(root->free_ino_ctl);
2007                 kfree(root->free_ino_pinned);
2008 #ifdef CONFIG_BTRFS_DEBUG
2009                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2010                 list_del_init(&root->leak_list);
2011                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2012 #endif
2013                 kfree(root);
2014         }
2015 }
2016
2017 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2018 {
2019         int ret;
2020         struct btrfs_root *gang[8];
2021         int i;
2022
2023         while (!list_empty(&fs_info->dead_roots)) {
2024                 gang[0] = list_entry(fs_info->dead_roots.next,
2025                                      struct btrfs_root, root_list);
2026                 list_del(&gang[0]->root_list);
2027
2028                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2029                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2030                 btrfs_put_root(gang[0]);
2031         }
2032
2033         while (1) {
2034                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2035                                              (void **)gang, 0,
2036                                              ARRAY_SIZE(gang));
2037                 if (!ret)
2038                         break;
2039                 for (i = 0; i < ret; i++)
2040                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2041         }
2042 }
2043
2044 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2045 {
2046         mutex_init(&fs_info->scrub_lock);
2047         atomic_set(&fs_info->scrubs_running, 0);
2048         atomic_set(&fs_info->scrub_pause_req, 0);
2049         atomic_set(&fs_info->scrubs_paused, 0);
2050         atomic_set(&fs_info->scrub_cancel_req, 0);
2051         init_waitqueue_head(&fs_info->scrub_pause_wait);
2052         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2053 }
2054
2055 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2056 {
2057         spin_lock_init(&fs_info->balance_lock);
2058         mutex_init(&fs_info->balance_mutex);
2059         atomic_set(&fs_info->balance_pause_req, 0);
2060         atomic_set(&fs_info->balance_cancel_req, 0);
2061         fs_info->balance_ctl = NULL;
2062         init_waitqueue_head(&fs_info->balance_wait_q);
2063 }
2064
2065 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2066 {
2067         struct inode *inode = fs_info->btree_inode;
2068
2069         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2070         set_nlink(inode, 1);
2071         /*
2072          * we set the i_size on the btree inode to the max possible int.
2073          * the real end of the address space is determined by all of
2074          * the devices in the system
2075          */
2076         inode->i_size = OFFSET_MAX;
2077         inode->i_mapping->a_ops = &btree_aops;
2078
2079         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2080         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2081                             IO_TREE_INODE_IO, inode);
2082         BTRFS_I(inode)->io_tree.track_uptodate = false;
2083         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2084
2085         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2086
2087         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2088         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2089         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2090         btrfs_insert_inode_hash(inode);
2091 }
2092
2093 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2094 {
2095         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2096         init_rwsem(&fs_info->dev_replace.rwsem);
2097         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2098 }
2099
2100 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2101 {
2102         spin_lock_init(&fs_info->qgroup_lock);
2103         mutex_init(&fs_info->qgroup_ioctl_lock);
2104         fs_info->qgroup_tree = RB_ROOT;
2105         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2106         fs_info->qgroup_seq = 1;
2107         fs_info->qgroup_ulist = NULL;
2108         fs_info->qgroup_rescan_running = false;
2109         mutex_init(&fs_info->qgroup_rescan_lock);
2110 }
2111
2112 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2113                 struct btrfs_fs_devices *fs_devices)
2114 {
2115         u32 max_active = fs_info->thread_pool_size;
2116         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2117
2118         fs_info->workers =
2119                 btrfs_alloc_workqueue(fs_info, "worker",
2120                                       flags | WQ_HIGHPRI, max_active, 16);
2121
2122         fs_info->delalloc_workers =
2123                 btrfs_alloc_workqueue(fs_info, "delalloc",
2124                                       flags, max_active, 2);
2125
2126         fs_info->flush_workers =
2127                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2128                                       flags, max_active, 0);
2129
2130         fs_info->caching_workers =
2131                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2132
2133         fs_info->fixup_workers =
2134                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2135
2136         /*
2137          * endios are largely parallel and should have a very
2138          * low idle thresh
2139          */
2140         fs_info->endio_workers =
2141                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2142         fs_info->endio_meta_workers =
2143                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2144                                       max_active, 4);
2145         fs_info->endio_meta_write_workers =
2146                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2147                                       max_active, 2);
2148         fs_info->endio_raid56_workers =
2149                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2150                                       max_active, 4);
2151         fs_info->rmw_workers =
2152                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2153         fs_info->endio_write_workers =
2154                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2155                                       max_active, 2);
2156         fs_info->endio_freespace_worker =
2157                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2158                                       max_active, 0);
2159         fs_info->delayed_workers =
2160                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2161                                       max_active, 0);
2162         fs_info->readahead_workers =
2163                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2164                                       max_active, 2);
2165         fs_info->qgroup_rescan_workers =
2166                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2167         fs_info->discard_ctl.discard_workers =
2168                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2169
2170         if (!(fs_info->workers && fs_info->delalloc_workers &&
2171               fs_info->flush_workers &&
2172               fs_info->endio_workers && fs_info->endio_meta_workers &&
2173               fs_info->endio_meta_write_workers &&
2174               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2175               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2176               fs_info->caching_workers && fs_info->readahead_workers &&
2177               fs_info->fixup_workers && fs_info->delayed_workers &&
2178               fs_info->qgroup_rescan_workers &&
2179               fs_info->discard_ctl.discard_workers)) {
2180                 return -ENOMEM;
2181         }
2182
2183         return 0;
2184 }
2185
2186 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2187 {
2188         struct crypto_shash *csum_shash;
2189         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2190
2191         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2192
2193         if (IS_ERR(csum_shash)) {
2194                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2195                           csum_driver);
2196                 return PTR_ERR(csum_shash);
2197         }
2198
2199         fs_info->csum_shash = csum_shash;
2200
2201         return 0;
2202 }
2203
2204 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2205                             struct btrfs_fs_devices *fs_devices)
2206 {
2207         int ret;
2208         struct btrfs_root *log_tree_root;
2209         struct btrfs_super_block *disk_super = fs_info->super_copy;
2210         u64 bytenr = btrfs_super_log_root(disk_super);
2211         int level = btrfs_super_log_root_level(disk_super);
2212
2213         if (fs_devices->rw_devices == 0) {
2214                 btrfs_warn(fs_info, "log replay required on RO media");
2215                 return -EIO;
2216         }
2217
2218         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2219                                          GFP_KERNEL);
2220         if (!log_tree_root)
2221                 return -ENOMEM;
2222
2223         log_tree_root->node = read_tree_block(fs_info, bytenr,
2224                                               fs_info->generation + 1,
2225                                               level, NULL);
2226         if (IS_ERR(log_tree_root->node)) {
2227                 btrfs_warn(fs_info, "failed to read log tree");
2228                 ret = PTR_ERR(log_tree_root->node);
2229                 log_tree_root->node = NULL;
2230                 btrfs_put_root(log_tree_root);
2231                 return ret;
2232         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2233                 btrfs_err(fs_info, "failed to read log tree");
2234                 btrfs_put_root(log_tree_root);
2235                 return -EIO;
2236         }
2237         /* returns with log_tree_root freed on success */
2238         ret = btrfs_recover_log_trees(log_tree_root);
2239         if (ret) {
2240                 btrfs_handle_fs_error(fs_info, ret,
2241                                       "Failed to recover log tree");
2242                 btrfs_put_root(log_tree_root);
2243                 return ret;
2244         }
2245
2246         if (sb_rdonly(fs_info->sb)) {
2247                 ret = btrfs_commit_super(fs_info);
2248                 if (ret)
2249                         return ret;
2250         }
2251
2252         return 0;
2253 }
2254
2255 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2256 {
2257         struct btrfs_root *tree_root = fs_info->tree_root;
2258         struct btrfs_root *root;
2259         struct btrfs_key location;
2260         int ret;
2261
2262         BUG_ON(!fs_info->tree_root);
2263
2264         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2265         location.type = BTRFS_ROOT_ITEM_KEY;
2266         location.offset = 0;
2267
2268         root = btrfs_read_tree_root(tree_root, &location);
2269         if (IS_ERR(root)) {
2270                 ret = PTR_ERR(root);
2271                 goto out;
2272         }
2273         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274         fs_info->extent_root = root;
2275
2276         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2277         root = btrfs_read_tree_root(tree_root, &location);
2278         if (IS_ERR(root)) {
2279                 ret = PTR_ERR(root);
2280                 goto out;
2281         }
2282         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2283         fs_info->dev_root = root;
2284         btrfs_init_devices_late(fs_info);
2285
2286         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2287         root = btrfs_read_tree_root(tree_root, &location);
2288         if (IS_ERR(root)) {
2289                 ret = PTR_ERR(root);
2290                 goto out;
2291         }
2292         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2293         fs_info->csum_root = root;
2294
2295         /*
2296          * This tree can share blocks with some other fs tree during relocation
2297          * and we need a proper setup by btrfs_get_fs_root
2298          */
2299         root = btrfs_get_fs_root(tree_root->fs_info,
2300                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2301         if (IS_ERR(root)) {
2302                 ret = PTR_ERR(root);
2303                 goto out;
2304         }
2305         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2306         fs_info->data_reloc_root = root;
2307
2308         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2309         root = btrfs_read_tree_root(tree_root, &location);
2310         if (!IS_ERR(root)) {
2311                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2312                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2313                 fs_info->quota_root = root;
2314         }
2315
2316         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2317         root = btrfs_read_tree_root(tree_root, &location);
2318         if (IS_ERR(root)) {
2319                 ret = PTR_ERR(root);
2320                 if (ret != -ENOENT)
2321                         goto out;
2322         } else {
2323                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2324                 fs_info->uuid_root = root;
2325         }
2326
2327         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2328                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2329                 root = btrfs_read_tree_root(tree_root, &location);
2330                 if (IS_ERR(root)) {
2331                         ret = PTR_ERR(root);
2332                         goto out;
2333                 }
2334                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2335                 fs_info->free_space_root = root;
2336         }
2337
2338         return 0;
2339 out:
2340         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2341                    location.objectid, ret);
2342         return ret;
2343 }
2344
2345 /*
2346  * Real super block validation
2347  * NOTE: super csum type and incompat features will not be checked here.
2348  *
2349  * @sb:         super block to check
2350  * @mirror_num: the super block number to check its bytenr:
2351  *              0       the primary (1st) sb
2352  *              1, 2    2nd and 3rd backup copy
2353  *             -1       skip bytenr check
2354  */
2355 static int validate_super(struct btrfs_fs_info *fs_info,
2356                             struct btrfs_super_block *sb, int mirror_num)
2357 {
2358         u64 nodesize = btrfs_super_nodesize(sb);
2359         u64 sectorsize = btrfs_super_sectorsize(sb);
2360         int ret = 0;
2361
2362         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2363                 btrfs_err(fs_info, "no valid FS found");
2364                 ret = -EINVAL;
2365         }
2366         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2367                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2368                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2369                 ret = -EINVAL;
2370         }
2371         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2372                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2373                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2374                 ret = -EINVAL;
2375         }
2376         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2377                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2378                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2379                 ret = -EINVAL;
2380         }
2381         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2382                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2383                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2384                 ret = -EINVAL;
2385         }
2386
2387         /*
2388          * Check sectorsize and nodesize first, other check will need it.
2389          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2390          */
2391         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2392             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2393                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2394                 ret = -EINVAL;
2395         }
2396         /* Only PAGE SIZE is supported yet */
2397         if (sectorsize != PAGE_SIZE) {
2398                 btrfs_err(fs_info,
2399                         "sectorsize %llu not supported yet, only support %lu",
2400                         sectorsize, PAGE_SIZE);
2401                 ret = -EINVAL;
2402         }
2403         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2404             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2405                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2406                 ret = -EINVAL;
2407         }
2408         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2409                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2410                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2411                 ret = -EINVAL;
2412         }
2413
2414         /* Root alignment check */
2415         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2416                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2417                            btrfs_super_root(sb));
2418                 ret = -EINVAL;
2419         }
2420         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2421                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2422                            btrfs_super_chunk_root(sb));
2423                 ret = -EINVAL;
2424         }
2425         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2426                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2427                            btrfs_super_log_root(sb));
2428                 ret = -EINVAL;
2429         }
2430
2431         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2432                    BTRFS_FSID_SIZE) != 0) {
2433                 btrfs_err(fs_info,
2434                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2435                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2436                 ret = -EINVAL;
2437         }
2438
2439         /*
2440          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2441          * done later
2442          */
2443         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2444                 btrfs_err(fs_info, "bytes_used is too small %llu",
2445                           btrfs_super_bytes_used(sb));
2446                 ret = -EINVAL;
2447         }
2448         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2449                 btrfs_err(fs_info, "invalid stripesize %u",
2450                           btrfs_super_stripesize(sb));
2451                 ret = -EINVAL;
2452         }
2453         if (btrfs_super_num_devices(sb) > (1UL << 31))
2454                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2455                            btrfs_super_num_devices(sb));
2456         if (btrfs_super_num_devices(sb) == 0) {
2457                 btrfs_err(fs_info, "number of devices is 0");
2458                 ret = -EINVAL;
2459         }
2460
2461         if (mirror_num >= 0 &&
2462             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2463                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2464                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2465                 ret = -EINVAL;
2466         }
2467
2468         /*
2469          * Obvious sys_chunk_array corruptions, it must hold at least one key
2470          * and one chunk
2471          */
2472         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2473                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2474                           btrfs_super_sys_array_size(sb),
2475                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2476                 ret = -EINVAL;
2477         }
2478         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2479                         + sizeof(struct btrfs_chunk)) {
2480                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2481                           btrfs_super_sys_array_size(sb),
2482                           sizeof(struct btrfs_disk_key)
2483                           + sizeof(struct btrfs_chunk));
2484                 ret = -EINVAL;
2485         }
2486
2487         /*
2488          * The generation is a global counter, we'll trust it more than the others
2489          * but it's still possible that it's the one that's wrong.
2490          */
2491         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2492                 btrfs_warn(fs_info,
2493                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2494                         btrfs_super_generation(sb),
2495                         btrfs_super_chunk_root_generation(sb));
2496         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2497             && btrfs_super_cache_generation(sb) != (u64)-1)
2498                 btrfs_warn(fs_info,
2499                         "suspicious: generation < cache_generation: %llu < %llu",
2500                         btrfs_super_generation(sb),
2501                         btrfs_super_cache_generation(sb));
2502
2503         return ret;
2504 }
2505
2506 /*
2507  * Validation of super block at mount time.
2508  * Some checks already done early at mount time, like csum type and incompat
2509  * flags will be skipped.
2510  */
2511 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2512 {
2513         return validate_super(fs_info, fs_info->super_copy, 0);
2514 }
2515
2516 /*
2517  * Validation of super block at write time.
2518  * Some checks like bytenr check will be skipped as their values will be
2519  * overwritten soon.
2520  * Extra checks like csum type and incompat flags will be done here.
2521  */
2522 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2523                                       struct btrfs_super_block *sb)
2524 {
2525         int ret;
2526
2527         ret = validate_super(fs_info, sb, -1);
2528         if (ret < 0)
2529                 goto out;
2530         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2531                 ret = -EUCLEAN;
2532                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2533                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2534                 goto out;
2535         }
2536         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2537                 ret = -EUCLEAN;
2538                 btrfs_err(fs_info,
2539                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2540                           btrfs_super_incompat_flags(sb),
2541                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2542                 goto out;
2543         }
2544 out:
2545         if (ret < 0)
2546                 btrfs_err(fs_info,
2547                 "super block corruption detected before writing it to disk");
2548         return ret;
2549 }
2550
2551 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2552 {
2553         int backup_index = find_newest_super_backup(fs_info);
2554         struct btrfs_super_block *sb = fs_info->super_copy;
2555         struct btrfs_root *tree_root = fs_info->tree_root;
2556         bool handle_error = false;
2557         int ret = 0;
2558         int i;
2559
2560         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2561                 u64 generation;
2562                 int level;
2563
2564                 if (handle_error) {
2565                         if (!IS_ERR(tree_root->node))
2566                                 free_extent_buffer(tree_root->node);
2567                         tree_root->node = NULL;
2568
2569                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2570                                 break;
2571
2572                         free_root_pointers(fs_info, 0);
2573
2574                         /*
2575                          * Don't use the log in recovery mode, it won't be
2576                          * valid
2577                          */
2578                         btrfs_set_super_log_root(sb, 0);
2579
2580                         /* We can't trust the free space cache either */
2581                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2582
2583                         ret = read_backup_root(fs_info, i);
2584                         backup_index = ret;
2585                         if (ret < 0)
2586                                 return ret;
2587                 }
2588                 generation = btrfs_super_generation(sb);
2589                 level = btrfs_super_root_level(sb);
2590                 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2591                                                   generation, level, NULL);
2592                 if (IS_ERR(tree_root->node) ||
2593                     !extent_buffer_uptodate(tree_root->node)) {
2594                         handle_error = true;
2595
2596                         if (IS_ERR(tree_root->node)) {
2597                                 ret = PTR_ERR(tree_root->node);
2598                                 tree_root->node = NULL;
2599                         } else if (!extent_buffer_uptodate(tree_root->node)) {
2600                                 ret = -EUCLEAN;
2601                         }
2602
2603                         btrfs_warn(fs_info, "failed to read tree root");
2604                         continue;
2605                 }
2606
2607                 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2608                 tree_root->commit_root = btrfs_root_node(tree_root);
2609                 btrfs_set_root_refs(&tree_root->root_item, 1);
2610
2611                 /*
2612                  * No need to hold btrfs_root::objectid_mutex since the fs
2613                  * hasn't been fully initialised and we are the only user
2614                  */
2615                 ret = btrfs_find_highest_objectid(tree_root,
2616                                                 &tree_root->highest_objectid);
2617                 if (ret < 0) {
2618                         handle_error = true;
2619                         continue;
2620                 }
2621
2622                 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2623
2624                 ret = btrfs_read_roots(fs_info);
2625                 if (ret < 0) {
2626                         handle_error = true;
2627                         continue;
2628                 }
2629
2630                 /* All successful */
2631                 fs_info->generation = generation;
2632                 fs_info->last_trans_committed = generation;
2633
2634                 /* Always begin writing backup roots after the one being used */
2635                 if (backup_index < 0) {
2636                         fs_info->backup_root_index = 0;
2637                 } else {
2638                         fs_info->backup_root_index = backup_index + 1;
2639                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2640                 }
2641                 break;
2642         }
2643
2644         return ret;
2645 }
2646
2647 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2648 {
2649         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2650         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2651         INIT_LIST_HEAD(&fs_info->trans_list);
2652         INIT_LIST_HEAD(&fs_info->dead_roots);
2653         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2654         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2655         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2656         spin_lock_init(&fs_info->delalloc_root_lock);
2657         spin_lock_init(&fs_info->trans_lock);
2658         spin_lock_init(&fs_info->fs_roots_radix_lock);
2659         spin_lock_init(&fs_info->delayed_iput_lock);
2660         spin_lock_init(&fs_info->defrag_inodes_lock);
2661         spin_lock_init(&fs_info->super_lock);
2662         spin_lock_init(&fs_info->buffer_lock);
2663         spin_lock_init(&fs_info->unused_bgs_lock);
2664         rwlock_init(&fs_info->tree_mod_log_lock);
2665         mutex_init(&fs_info->unused_bg_unpin_mutex);
2666         mutex_init(&fs_info->delete_unused_bgs_mutex);
2667         mutex_init(&fs_info->reloc_mutex);
2668         mutex_init(&fs_info->delalloc_root_mutex);
2669         seqlock_init(&fs_info->profiles_lock);
2670
2671         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2672         INIT_LIST_HEAD(&fs_info->space_info);
2673         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2674         INIT_LIST_HEAD(&fs_info->unused_bgs);
2675 #ifdef CONFIG_BTRFS_DEBUG
2676         INIT_LIST_HEAD(&fs_info->allocated_roots);
2677         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2678         spin_lock_init(&fs_info->eb_leak_lock);
2679 #endif
2680         extent_map_tree_init(&fs_info->mapping_tree);
2681         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2682                              BTRFS_BLOCK_RSV_GLOBAL);
2683         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2684         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2685         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2686         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2687                              BTRFS_BLOCK_RSV_DELOPS);
2688         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2689                              BTRFS_BLOCK_RSV_DELREFS);
2690
2691         atomic_set(&fs_info->async_delalloc_pages, 0);
2692         atomic_set(&fs_info->defrag_running, 0);
2693         atomic_set(&fs_info->reada_works_cnt, 0);
2694         atomic_set(&fs_info->nr_delayed_iputs, 0);
2695         atomic64_set(&fs_info->tree_mod_seq, 0);
2696         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2697         fs_info->metadata_ratio = 0;
2698         fs_info->defrag_inodes = RB_ROOT;
2699         atomic64_set(&fs_info->free_chunk_space, 0);
2700         fs_info->tree_mod_log = RB_ROOT;
2701         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2702         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2703         /* readahead state */
2704         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2705         spin_lock_init(&fs_info->reada_lock);
2706         btrfs_init_ref_verify(fs_info);
2707
2708         fs_info->thread_pool_size = min_t(unsigned long,
2709                                           num_online_cpus() + 2, 8);
2710
2711         INIT_LIST_HEAD(&fs_info->ordered_roots);
2712         spin_lock_init(&fs_info->ordered_root_lock);
2713
2714         btrfs_init_scrub(fs_info);
2715 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2716         fs_info->check_integrity_print_mask = 0;
2717 #endif
2718         btrfs_init_balance(fs_info);
2719         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2720
2721         spin_lock_init(&fs_info->block_group_cache_lock);
2722         fs_info->block_group_cache_tree = RB_ROOT;
2723         fs_info->first_logical_byte = (u64)-1;
2724
2725         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2726                             IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2727         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2728
2729         mutex_init(&fs_info->ordered_operations_mutex);
2730         mutex_init(&fs_info->tree_log_mutex);
2731         mutex_init(&fs_info->chunk_mutex);
2732         mutex_init(&fs_info->transaction_kthread_mutex);
2733         mutex_init(&fs_info->cleaner_mutex);
2734         mutex_init(&fs_info->ro_block_group_mutex);
2735         init_rwsem(&fs_info->commit_root_sem);
2736         init_rwsem(&fs_info->cleanup_work_sem);
2737         init_rwsem(&fs_info->subvol_sem);
2738         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2739
2740         btrfs_init_dev_replace_locks(fs_info);
2741         btrfs_init_qgroup(fs_info);
2742         btrfs_discard_init(fs_info);
2743
2744         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2745         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2746
2747         init_waitqueue_head(&fs_info->transaction_throttle);
2748         init_waitqueue_head(&fs_info->transaction_wait);
2749         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2750         init_waitqueue_head(&fs_info->async_submit_wait);
2751         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2752
2753         /* Usable values until the real ones are cached from the superblock */
2754         fs_info->nodesize = 4096;
2755         fs_info->sectorsize = 4096;
2756         fs_info->stripesize = 4096;
2757
2758         spin_lock_init(&fs_info->swapfile_pins_lock);
2759         fs_info->swapfile_pins = RB_ROOT;
2760
2761         fs_info->send_in_progress = 0;
2762 }
2763
2764 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2765 {
2766         int ret;
2767
2768         fs_info->sb = sb;
2769         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2770         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2771
2772         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2773         if (ret)
2774                 return ret;
2775
2776         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2777         if (ret)
2778                 return ret;
2779
2780         fs_info->dirty_metadata_batch = PAGE_SIZE *
2781                                         (1 + ilog2(nr_cpu_ids));
2782
2783         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2784         if (ret)
2785                 return ret;
2786
2787         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2788                         GFP_KERNEL);
2789         if (ret)
2790                 return ret;
2791
2792         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2793                                         GFP_KERNEL);
2794         if (!fs_info->delayed_root)
2795                 return -ENOMEM;
2796         btrfs_init_delayed_root(fs_info->delayed_root);
2797
2798         return btrfs_alloc_stripe_hash_table(fs_info);
2799 }
2800
2801 static int btrfs_uuid_rescan_kthread(void *data)
2802 {
2803         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2804         int ret;
2805
2806         /*
2807          * 1st step is to iterate through the existing UUID tree and
2808          * to delete all entries that contain outdated data.
2809          * 2nd step is to add all missing entries to the UUID tree.
2810          */
2811         ret = btrfs_uuid_tree_iterate(fs_info);
2812         if (ret < 0) {
2813                 if (ret != -EINTR)
2814                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2815                                    ret);
2816                 up(&fs_info->uuid_tree_rescan_sem);
2817                 return ret;
2818         }
2819         return btrfs_uuid_scan_kthread(data);
2820 }
2821
2822 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2823 {
2824         struct task_struct *task;
2825
2826         down(&fs_info->uuid_tree_rescan_sem);
2827         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2828         if (IS_ERR(task)) {
2829                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2830                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2831                 up(&fs_info->uuid_tree_rescan_sem);
2832                 return PTR_ERR(task);
2833         }
2834
2835         return 0;
2836 }
2837
2838 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2839                       char *options)
2840 {
2841         u32 sectorsize;
2842         u32 nodesize;
2843         u32 stripesize;
2844         u64 generation;
2845         u64 features;
2846         u16 csum_type;
2847         struct btrfs_super_block *disk_super;
2848         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2849         struct btrfs_root *tree_root;
2850         struct btrfs_root *chunk_root;
2851         int ret;
2852         int err = -EINVAL;
2853         int clear_free_space_tree = 0;
2854         int level;
2855
2856         ret = init_mount_fs_info(fs_info, sb);
2857         if (ret) {
2858                 err = ret;
2859                 goto fail;
2860         }
2861
2862         /* These need to be init'ed before we start creating inodes and such. */
2863         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2864                                      GFP_KERNEL);
2865         fs_info->tree_root = tree_root;
2866         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2867                                       GFP_KERNEL);
2868         fs_info->chunk_root = chunk_root;
2869         if (!tree_root || !chunk_root) {
2870                 err = -ENOMEM;
2871                 goto fail;
2872         }
2873
2874         fs_info->btree_inode = new_inode(sb);
2875         if (!fs_info->btree_inode) {
2876                 err = -ENOMEM;
2877                 goto fail;
2878         }
2879         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2880         btrfs_init_btree_inode(fs_info);
2881
2882         invalidate_bdev(fs_devices->latest_bdev);
2883
2884         /*
2885          * Read super block and check the signature bytes only
2886          */
2887         disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2888         if (IS_ERR(disk_super)) {
2889                 err = PTR_ERR(disk_super);
2890                 goto fail_alloc;
2891         }
2892
2893         /*
2894          * Verify the type first, if that or the the checksum value are
2895          * corrupted, we'll find out
2896          */
2897         csum_type = btrfs_super_csum_type(disk_super);
2898         if (!btrfs_supported_super_csum(csum_type)) {
2899                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2900                           csum_type);
2901                 err = -EINVAL;
2902                 btrfs_release_disk_super(disk_super);
2903                 goto fail_alloc;
2904         }
2905
2906         ret = btrfs_init_csum_hash(fs_info, csum_type);
2907         if (ret) {
2908                 err = ret;
2909                 btrfs_release_disk_super(disk_super);
2910                 goto fail_alloc;
2911         }
2912
2913         /*
2914          * We want to check superblock checksum, the type is stored inside.
2915          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2916          */
2917         if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2918                 btrfs_err(fs_info, "superblock checksum mismatch");
2919                 err = -EINVAL;
2920                 btrfs_release_disk_super(disk_super);
2921                 goto fail_alloc;
2922         }
2923
2924         /*
2925          * super_copy is zeroed at allocation time and we never touch the
2926          * following bytes up to INFO_SIZE, the checksum is calculated from
2927          * the whole block of INFO_SIZE
2928          */
2929         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2930         btrfs_release_disk_super(disk_super);
2931
2932         disk_super = fs_info->super_copy;
2933
2934         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2935                        BTRFS_FSID_SIZE));
2936
2937         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2938                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2939                                 fs_info->super_copy->metadata_uuid,
2940                                 BTRFS_FSID_SIZE));
2941         }
2942
2943         features = btrfs_super_flags(disk_super);
2944         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2945                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2946                 btrfs_set_super_flags(disk_super, features);
2947                 btrfs_info(fs_info,
2948                         "found metadata UUID change in progress flag, clearing");
2949         }
2950
2951         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2952                sizeof(*fs_info->super_for_commit));
2953
2954         ret = btrfs_validate_mount_super(fs_info);
2955         if (ret) {
2956                 btrfs_err(fs_info, "superblock contains fatal errors");
2957                 err = -EINVAL;
2958                 goto fail_alloc;
2959         }
2960
2961         if (!btrfs_super_root(disk_super))
2962                 goto fail_alloc;
2963
2964         /* check FS state, whether FS is broken. */
2965         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2966                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2967
2968         /*
2969          * In the long term, we'll store the compression type in the super
2970          * block, and it'll be used for per file compression control.
2971          */
2972         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2973
2974         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2975         if (ret) {
2976                 err = ret;
2977                 goto fail_alloc;
2978         }
2979
2980         features = btrfs_super_incompat_flags(disk_super) &
2981                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2982         if (features) {
2983                 btrfs_err(fs_info,
2984                     "cannot mount because of unsupported optional features (%llx)",
2985                     features);
2986                 err = -EINVAL;
2987                 goto fail_alloc;
2988         }
2989
2990         features = btrfs_super_incompat_flags(disk_super);
2991         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2992         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2993                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2994         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2995                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2996
2997         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2998                 btrfs_info(fs_info, "has skinny extents");
2999
3000         /*
3001          * flag our filesystem as having big metadata blocks if
3002          * they are bigger than the page size
3003          */
3004         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3005                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3006                         btrfs_info(fs_info,
3007                                 "flagging fs with big metadata feature");
3008                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3009         }
3010
3011         nodesize = btrfs_super_nodesize(disk_super);
3012         sectorsize = btrfs_super_sectorsize(disk_super);
3013         stripesize = sectorsize;
3014         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3015         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3016
3017         /* Cache block sizes */
3018         fs_info->nodesize = nodesize;
3019         fs_info->sectorsize = sectorsize;
3020         fs_info->stripesize = stripesize;
3021
3022         /*
3023          * mixed block groups end up with duplicate but slightly offset
3024          * extent buffers for the same range.  It leads to corruptions
3025          */
3026         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3027             (sectorsize != nodesize)) {
3028                 btrfs_err(fs_info,
3029 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3030                         nodesize, sectorsize);
3031                 goto fail_alloc;
3032         }
3033
3034         /*
3035          * Needn't use the lock because there is no other task which will
3036          * update the flag.
3037          */
3038         btrfs_set_super_incompat_flags(disk_super, features);
3039
3040         features = btrfs_super_compat_ro_flags(disk_super) &
3041                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3042         if (!sb_rdonly(sb) && features) {
3043                 btrfs_err(fs_info,
3044         "cannot mount read-write because of unsupported optional features (%llx)",
3045                        features);
3046                 err = -EINVAL;
3047                 goto fail_alloc;
3048         }
3049
3050         ret = btrfs_init_workqueues(fs_info, fs_devices);
3051         if (ret) {
3052                 err = ret;
3053                 goto fail_sb_buffer;
3054         }
3055
3056         sb->s_bdi->congested_fn = btrfs_congested_fn;
3057         sb->s_bdi->congested_data = fs_info;
3058         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3059         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3060         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3061         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3062
3063         sb->s_blocksize = sectorsize;
3064         sb->s_blocksize_bits = blksize_bits(sectorsize);
3065         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3066
3067         mutex_lock(&fs_info->chunk_mutex);
3068         ret = btrfs_read_sys_array(fs_info);
3069         mutex_unlock(&fs_info->chunk_mutex);
3070         if (ret) {
3071                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3072                 goto fail_sb_buffer;
3073         }
3074
3075         generation = btrfs_super_chunk_root_generation(disk_super);
3076         level = btrfs_super_chunk_root_level(disk_super);
3077
3078         chunk_root->node = read_tree_block(fs_info,
3079                                            btrfs_super_chunk_root(disk_super),
3080                                            generation, level, NULL);
3081         if (IS_ERR(chunk_root->node) ||
3082             !extent_buffer_uptodate(chunk_root->node)) {
3083                 btrfs_err(fs_info, "failed to read chunk root");
3084                 if (!IS_ERR(chunk_root->node))
3085                         free_extent_buffer(chunk_root->node);
3086                 chunk_root->node = NULL;
3087                 goto fail_tree_roots;
3088         }
3089         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3090         chunk_root->commit_root = btrfs_root_node(chunk_root);
3091
3092         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3093                            offsetof(struct btrfs_header, chunk_tree_uuid),
3094                            BTRFS_UUID_SIZE);
3095
3096         ret = btrfs_read_chunk_tree(fs_info);
3097         if (ret) {
3098                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3099                 goto fail_tree_roots;
3100         }
3101
3102         /*
3103          * Keep the devid that is marked to be the target device for the
3104          * device replace procedure
3105          */
3106         btrfs_free_extra_devids(fs_devices, 0);
3107
3108         if (!fs_devices->latest_bdev) {
3109                 btrfs_err(fs_info, "failed to read devices");
3110                 goto fail_tree_roots;
3111         }
3112
3113         ret = init_tree_roots(fs_info);
3114         if (ret)
3115                 goto fail_tree_roots;
3116
3117         /*
3118          * If we have a uuid root and we're not being told to rescan we need to
3119          * check the generation here so we can set the
3120          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3121          * transaction during a balance or the log replay without updating the
3122          * uuid generation, and then if we crash we would rescan the uuid tree,
3123          * even though it was perfectly fine.
3124          */
3125         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3126             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3127                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3128
3129         ret = btrfs_verify_dev_extents(fs_info);
3130         if (ret) {
3131                 btrfs_err(fs_info,
3132                           "failed to verify dev extents against chunks: %d",
3133                           ret);
3134                 goto fail_block_groups;
3135         }
3136         ret = btrfs_recover_balance(fs_info);
3137         if (ret) {
3138                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3139                 goto fail_block_groups;
3140         }
3141
3142         ret = btrfs_init_dev_stats(fs_info);
3143         if (ret) {
3144                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3145                 goto fail_block_groups;
3146         }
3147
3148         ret = btrfs_init_dev_replace(fs_info);
3149         if (ret) {
3150                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3151                 goto fail_block_groups;
3152         }
3153
3154         btrfs_free_extra_devids(fs_devices, 1);
3155
3156         ret = btrfs_sysfs_add_fsid(fs_devices);
3157         if (ret) {
3158                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3159                                 ret);
3160                 goto fail_block_groups;
3161         }
3162
3163         ret = btrfs_sysfs_add_mounted(fs_info);
3164         if (ret) {
3165                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3166                 goto fail_fsdev_sysfs;
3167         }
3168
3169         ret = btrfs_init_space_info(fs_info);
3170         if (ret) {
3171                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3172                 goto fail_sysfs;
3173         }
3174
3175         ret = btrfs_read_block_groups(fs_info);
3176         if (ret) {
3177                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3178                 goto fail_sysfs;
3179         }
3180
3181         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3182                 btrfs_warn(fs_info,
3183                 "writable mount is not allowed due to too many missing devices");
3184                 goto fail_sysfs;
3185         }
3186
3187         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3188                                                "btrfs-cleaner");
3189         if (IS_ERR(fs_info->cleaner_kthread))
3190                 goto fail_sysfs;
3191
3192         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3193                                                    tree_root,
3194                                                    "btrfs-transaction");
3195         if (IS_ERR(fs_info->transaction_kthread))
3196                 goto fail_cleaner;
3197
3198         if (!btrfs_test_opt(fs_info, NOSSD) &&
3199             !fs_info->fs_devices->rotating) {
3200                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3201         }
3202
3203         /*
3204          * Mount does not set all options immediately, we can do it now and do
3205          * not have to wait for transaction commit
3206          */
3207         btrfs_apply_pending_changes(fs_info);
3208
3209 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3210         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3211                 ret = btrfsic_mount(fs_info, fs_devices,
3212                                     btrfs_test_opt(fs_info,
3213                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3214                                     1 : 0,
3215                                     fs_info->check_integrity_print_mask);
3216                 if (ret)
3217                         btrfs_warn(fs_info,
3218                                 "failed to initialize integrity check module: %d",
3219                                 ret);
3220         }
3221 #endif
3222         ret = btrfs_read_qgroup_config(fs_info);
3223         if (ret)
3224                 goto fail_trans_kthread;
3225
3226         if (btrfs_build_ref_tree(fs_info))
3227                 btrfs_err(fs_info, "couldn't build ref tree");
3228
3229         /* do not make disk changes in broken FS or nologreplay is given */
3230         if (btrfs_super_log_root(disk_super) != 0 &&
3231             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3232                 btrfs_info(fs_info, "start tree-log replay");
3233                 ret = btrfs_replay_log(fs_info, fs_devices);
3234                 if (ret) {
3235                         err = ret;
3236                         goto fail_qgroup;
3237                 }
3238         }
3239
3240         ret = btrfs_find_orphan_roots(fs_info);
3241         if (ret)
3242                 goto fail_qgroup;
3243
3244         if (!sb_rdonly(sb)) {
3245                 ret = btrfs_cleanup_fs_roots(fs_info);
3246                 if (ret)
3247                         goto fail_qgroup;
3248
3249                 mutex_lock(&fs_info->cleaner_mutex);
3250                 ret = btrfs_recover_relocation(tree_root);
3251                 mutex_unlock(&fs_info->cleaner_mutex);
3252                 if (ret < 0) {
3253                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3254                                         ret);
3255                         err = -EINVAL;
3256                         goto fail_qgroup;
3257                 }
3258         }
3259
3260         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3261         if (IS_ERR(fs_info->fs_root)) {
3262                 err = PTR_ERR(fs_info->fs_root);
3263                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3264                 fs_info->fs_root = NULL;
3265                 goto fail_qgroup;
3266         }
3267
3268         if (sb_rdonly(sb))
3269                 return 0;
3270
3271         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3272             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3273                 clear_free_space_tree = 1;
3274         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3275                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3276                 btrfs_warn(fs_info, "free space tree is invalid");
3277                 clear_free_space_tree = 1;
3278         }
3279
3280         if (clear_free_space_tree) {
3281                 btrfs_info(fs_info, "clearing free space tree");
3282                 ret = btrfs_clear_free_space_tree(fs_info);
3283                 if (ret) {
3284                         btrfs_warn(fs_info,
3285                                    "failed to clear free space tree: %d", ret);
3286                         close_ctree(fs_info);
3287                         return ret;
3288                 }
3289         }
3290
3291         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3292             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3293                 btrfs_info(fs_info, "creating free space tree");
3294                 ret = btrfs_create_free_space_tree(fs_info);
3295                 if (ret) {
3296                         btrfs_warn(fs_info,
3297                                 "failed to create free space tree: %d", ret);
3298                         close_ctree(fs_info);
3299                         return ret;
3300                 }
3301         }
3302
3303         down_read(&fs_info->cleanup_work_sem);
3304         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3305             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3306                 up_read(&fs_info->cleanup_work_sem);
3307                 close_ctree(fs_info);
3308                 return ret;
3309         }
3310         up_read(&fs_info->cleanup_work_sem);
3311
3312         ret = btrfs_resume_balance_async(fs_info);
3313         if (ret) {
3314                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3315                 close_ctree(fs_info);
3316                 return ret;
3317         }
3318
3319         ret = btrfs_resume_dev_replace_async(fs_info);
3320         if (ret) {
3321                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3322                 close_ctree(fs_info);
3323                 return ret;
3324         }
3325
3326         btrfs_qgroup_rescan_resume(fs_info);
3327         btrfs_discard_resume(fs_info);
3328
3329         if (!fs_info->uuid_root) {
3330                 btrfs_info(fs_info, "creating UUID tree");
3331                 ret = btrfs_create_uuid_tree(fs_info);
3332                 if (ret) {
3333                         btrfs_warn(fs_info,
3334                                 "failed to create the UUID tree: %d", ret);
3335                         close_ctree(fs_info);
3336                         return ret;
3337                 }
3338         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3339                    fs_info->generation !=
3340                                 btrfs_super_uuid_tree_generation(disk_super)) {
3341                 btrfs_info(fs_info, "checking UUID tree");
3342                 ret = btrfs_check_uuid_tree(fs_info);
3343                 if (ret) {
3344                         btrfs_warn(fs_info,
3345                                 "failed to check the UUID tree: %d", ret);
3346                         close_ctree(fs_info);
3347                         return ret;
3348                 }
3349         }
3350         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3351
3352         /*
3353          * backuproot only affect mount behavior, and if open_ctree succeeded,
3354          * no need to keep the flag
3355          */
3356         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3357
3358         return 0;
3359
3360 fail_qgroup:
3361         btrfs_free_qgroup_config(fs_info);
3362 fail_trans_kthread:
3363         kthread_stop(fs_info->transaction_kthread);
3364         btrfs_cleanup_transaction(fs_info);
3365         btrfs_free_fs_roots(fs_info);
3366 fail_cleaner:
3367         kthread_stop(fs_info->cleaner_kthread);
3368
3369         /*
3370          * make sure we're done with the btree inode before we stop our
3371          * kthreads
3372          */
3373         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3374
3375 fail_sysfs:
3376         btrfs_sysfs_remove_mounted(fs_info);
3377
3378 fail_fsdev_sysfs:
3379         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3380
3381 fail_block_groups:
3382         btrfs_put_block_group_cache(fs_info);
3383
3384 fail_tree_roots:
3385         free_root_pointers(fs_info, true);
3386         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3387
3388 fail_sb_buffer:
3389         btrfs_stop_all_workers(fs_info);
3390         btrfs_free_block_groups(fs_info);
3391 fail_alloc:
3392         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3393
3394         iput(fs_info->btree_inode);
3395 fail:
3396         btrfs_close_devices(fs_info->fs_devices);
3397         return err;
3398 }
3399 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3400
3401 static void btrfs_end_super_write(struct bio *bio)
3402 {
3403         struct btrfs_device *device = bio->bi_private;
3404         struct bio_vec *bvec;
3405         struct bvec_iter_all iter_all;
3406         struct page *page;
3407
3408         bio_for_each_segment_all(bvec, bio, iter_all) {
3409                 page = bvec->bv_page;
3410
3411                 if (bio->bi_status) {
3412                         btrfs_warn_rl_in_rcu(device->fs_info,
3413                                 "lost page write due to IO error on %s (%d)",
3414                                 rcu_str_deref(device->name),
3415                                 blk_status_to_errno(bio->bi_status));
3416                         ClearPageUptodate(page);
3417                         SetPageError(page);
3418                         btrfs_dev_stat_inc_and_print(device,
3419                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3420                 } else {
3421                         SetPageUptodate(page);
3422                 }
3423
3424                 put_page(page);
3425                 unlock_page(page);
3426         }
3427
3428         bio_put(bio);
3429 }
3430
3431 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3432                                                    int copy_num)
3433 {
3434         struct btrfs_super_block *super;
3435         struct page *page;
3436         u64 bytenr;
3437         struct address_space *mapping = bdev->bd_inode->i_mapping;
3438
3439         bytenr = btrfs_sb_offset(copy_num);
3440         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3441                 return ERR_PTR(-EINVAL);
3442
3443         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3444         if (IS_ERR(page))
3445                 return ERR_CAST(page);
3446
3447         super = page_address(page);
3448         if (btrfs_super_bytenr(super) != bytenr ||
3449                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3450                 btrfs_release_disk_super(super);
3451                 return ERR_PTR(-EINVAL);
3452         }
3453
3454         return super;
3455 }
3456
3457
3458 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3459 {
3460         struct btrfs_super_block *super, *latest = NULL;
3461         int i;
3462         u64 transid = 0;
3463
3464         /* we would like to check all the supers, but that would make
3465          * a btrfs mount succeed after a mkfs from a different FS.
3466          * So, we need to add a special mount option to scan for
3467          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3468          */
3469         for (i = 0; i < 1; i++) {
3470                 super = btrfs_read_dev_one_super(bdev, i);
3471                 if (IS_ERR(super))
3472                         continue;
3473
3474                 if (!latest || btrfs_super_generation(super) > transid) {
3475                         if (latest)
3476                                 btrfs_release_disk_super(super);
3477
3478                         latest = super;
3479                         transid = btrfs_super_generation(super);
3480                 }
3481         }
3482
3483         return super;
3484 }
3485
3486 /*
3487  * Write superblock @sb to the @device. Do not wait for completion, all the
3488  * pages we use for writing are locked.
3489  *
3490  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3491  * the expected device size at commit time. Note that max_mirrors must be
3492  * same for write and wait phases.
3493  *
3494  * Return number of errors when page is not found or submission fails.
3495  */
3496 static int write_dev_supers(struct btrfs_device *device,
3497                             struct btrfs_super_block *sb, int max_mirrors)
3498 {
3499         struct btrfs_fs_info *fs_info = device->fs_info;
3500         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3501         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3502         int i;
3503         int errors = 0;
3504         u64 bytenr;
3505
3506         if (max_mirrors == 0)
3507                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3508
3509         shash->tfm = fs_info->csum_shash;
3510
3511         for (i = 0; i < max_mirrors; i++) {
3512                 struct page *page;
3513                 struct bio *bio;
3514                 struct btrfs_super_block *disk_super;
3515
3516                 bytenr = btrfs_sb_offset(i);
3517                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3518                     device->commit_total_bytes)
3519                         break;
3520
3521                 btrfs_set_super_bytenr(sb, bytenr);
3522
3523                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3524                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3525                                     sb->csum);
3526
3527                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3528                                            GFP_NOFS);
3529                 if (!page) {
3530                         btrfs_err(device->fs_info,
3531                             "couldn't get super block page for bytenr %llu",
3532                             bytenr);
3533                         errors++;
3534                         continue;
3535                 }
3536
3537                 /* Bump the refcount for wait_dev_supers() */
3538                 get_page(page);
3539
3540                 disk_super = page_address(page);
3541                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3542
3543                 /*
3544                  * Directly use bios here instead of relying on the page cache
3545                  * to do I/O, so we don't lose the ability to do integrity
3546                  * checking.
3547                  */
3548                 bio = bio_alloc(GFP_NOFS, 1);
3549                 bio_set_dev(bio, device->bdev);
3550                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3551                 bio->bi_private = device;
3552                 bio->bi_end_io = btrfs_end_super_write;
3553                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3554                                offset_in_page(bytenr));
3555
3556                 /*
3557                  * We FUA only the first super block.  The others we allow to
3558                  * go down lazy and there's a short window where the on-disk
3559                  * copies might still contain the older version.
3560                  */
3561                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3562                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3563                         bio->bi_opf |= REQ_FUA;
3564
3565                 btrfsic_submit_bio(bio);
3566         }
3567         return errors < i ? 0 : -1;
3568 }
3569
3570 /*
3571  * Wait for write completion of superblocks done by write_dev_supers,
3572  * @max_mirrors same for write and wait phases.
3573  *
3574  * Return number of errors when page is not found or not marked up to
3575  * date.
3576  */
3577 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3578 {
3579         int i;
3580         int errors = 0;
3581         bool primary_failed = false;
3582         u64 bytenr;
3583
3584         if (max_mirrors == 0)
3585                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3586
3587         for (i = 0; i < max_mirrors; i++) {
3588                 struct page *page;
3589
3590                 bytenr = btrfs_sb_offset(i);
3591                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3592                     device->commit_total_bytes)
3593                         break;
3594
3595                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3596                                      bytenr >> PAGE_SHIFT);
3597                 if (!page) {
3598                         errors++;
3599                         if (i == 0)
3600                                 primary_failed = true;
3601                         continue;
3602                 }
3603                 /* Page is submitted locked and unlocked once the IO completes */
3604                 wait_on_page_locked(page);
3605                 if (PageError(page)) {
3606                         errors++;
3607                         if (i == 0)
3608                                 primary_failed = true;
3609                 }
3610
3611                 /* Drop our reference */
3612                 put_page(page);
3613
3614                 /* Drop the reference from the writing run */
3615                 put_page(page);
3616         }
3617
3618         /* log error, force error return */
3619         if (primary_failed) {
3620                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3621                           device->devid);
3622                 return -1;
3623         }
3624
3625         return errors < i ? 0 : -1;
3626 }
3627
3628 /*
3629  * endio for the write_dev_flush, this will wake anyone waiting
3630  * for the barrier when it is done
3631  */
3632 static void btrfs_end_empty_barrier(struct bio *bio)
3633 {
3634         complete(bio->bi_private);
3635 }
3636
3637 /*
3638  * Submit a flush request to the device if it supports it. Error handling is
3639  * done in the waiting counterpart.
3640  */
3641 static void write_dev_flush(struct btrfs_device *device)
3642 {
3643         struct request_queue *q = bdev_get_queue(device->bdev);
3644         struct bio *bio = device->flush_bio;
3645
3646         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3647                 return;
3648
3649         bio_reset(bio);
3650         bio->bi_end_io = btrfs_end_empty_barrier;
3651         bio_set_dev(bio, device->bdev);
3652         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3653         init_completion(&device->flush_wait);
3654         bio->bi_private = &device->flush_wait;
3655
3656         btrfsic_submit_bio(bio);
3657         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3658 }
3659
3660 /*
3661  * If the flush bio has been submitted by write_dev_flush, wait for it.
3662  */
3663 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3664 {
3665         struct bio *bio = device->flush_bio;
3666
3667         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3668                 return BLK_STS_OK;
3669
3670         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3671         wait_for_completion_io(&device->flush_wait);
3672
3673         return bio->bi_status;
3674 }
3675
3676 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3677 {
3678         if (!btrfs_check_rw_degradable(fs_info, NULL))
3679                 return -EIO;
3680         return 0;
3681 }
3682
3683 /*
3684  * send an empty flush down to each device in parallel,
3685  * then wait for them
3686  */
3687 static int barrier_all_devices(struct btrfs_fs_info *info)
3688 {
3689         struct list_head *head;
3690         struct btrfs_device *dev;
3691         int errors_wait = 0;
3692         blk_status_t ret;
3693
3694         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3695         /* send down all the barriers */
3696         head = &info->fs_devices->devices;
3697         list_for_each_entry(dev, head, dev_list) {
3698                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3699                         continue;
3700                 if (!dev->bdev)
3701                         continue;
3702                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3703                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3704                         continue;
3705
3706                 write_dev_flush(dev);
3707                 dev->last_flush_error = BLK_STS_OK;
3708         }
3709
3710         /* wait for all the barriers */
3711         list_for_each_entry(dev, head, dev_list) {
3712                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3713                         continue;
3714                 if (!dev->bdev) {
3715                         errors_wait++;
3716                         continue;
3717                 }
3718                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3719                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3720                         continue;
3721
3722                 ret = wait_dev_flush(dev);
3723                 if (ret) {
3724                         dev->last_flush_error = ret;
3725                         btrfs_dev_stat_inc_and_print(dev,
3726                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3727                         errors_wait++;
3728                 }
3729         }
3730
3731         if (errors_wait) {
3732                 /*
3733                  * At some point we need the status of all disks
3734                  * to arrive at the volume status. So error checking
3735                  * is being pushed to a separate loop.
3736                  */
3737                 return check_barrier_error(info);
3738         }
3739         return 0;
3740 }
3741
3742 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3743 {
3744         int raid_type;
3745         int min_tolerated = INT_MAX;
3746
3747         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3748             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3749                 min_tolerated = min_t(int, min_tolerated,
3750                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3751                                     tolerated_failures);
3752
3753         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3754                 if (raid_type == BTRFS_RAID_SINGLE)
3755                         continue;
3756                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3757                         continue;
3758                 min_tolerated = min_t(int, min_tolerated,
3759                                     btrfs_raid_array[raid_type].
3760                                     tolerated_failures);
3761         }
3762
3763         if (min_tolerated == INT_MAX) {
3764                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3765                 min_tolerated = 0;
3766         }
3767
3768         return min_tolerated;
3769 }
3770
3771 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3772 {
3773         struct list_head *head;
3774         struct btrfs_device *dev;
3775         struct btrfs_super_block *sb;
3776         struct btrfs_dev_item *dev_item;
3777         int ret;
3778         int do_barriers;
3779         int max_errors;
3780         int total_errors = 0;
3781         u64 flags;
3782
3783         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3784
3785         /*
3786          * max_mirrors == 0 indicates we're from commit_transaction,
3787          * not from fsync where the tree roots in fs_info have not
3788          * been consistent on disk.
3789          */
3790         if (max_mirrors == 0)
3791                 backup_super_roots(fs_info);
3792
3793         sb = fs_info->super_for_commit;
3794         dev_item = &sb->dev_item;
3795
3796         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3797         head = &fs_info->fs_devices->devices;
3798         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3799
3800         if (do_barriers) {
3801                 ret = barrier_all_devices(fs_info);
3802                 if (ret) {
3803                         mutex_unlock(
3804                                 &fs_info->fs_devices->device_list_mutex);
3805                         btrfs_handle_fs_error(fs_info, ret,
3806                                               "errors while submitting device barriers.");
3807                         return ret;
3808                 }
3809         }
3810
3811         list_for_each_entry(dev, head, dev_list) {
3812                 if (!dev->bdev) {
3813                         total_errors++;
3814                         continue;
3815                 }
3816                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3817                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3818                         continue;
3819
3820                 btrfs_set_stack_device_generation(dev_item, 0);
3821                 btrfs_set_stack_device_type(dev_item, dev->type);
3822                 btrfs_set_stack_device_id(dev_item, dev->devid);
3823                 btrfs_set_stack_device_total_bytes(dev_item,
3824                                                    dev->commit_total_bytes);
3825                 btrfs_set_stack_device_bytes_used(dev_item,
3826                                                   dev->commit_bytes_used);
3827                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3828                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3829                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3830                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3831                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3832                        BTRFS_FSID_SIZE);
3833
3834                 flags = btrfs_super_flags(sb);
3835                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3836
3837                 ret = btrfs_validate_write_super(fs_info, sb);
3838                 if (ret < 0) {
3839                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3840                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3841                                 "unexpected superblock corruption detected");
3842                         return -EUCLEAN;
3843                 }
3844
3845                 ret = write_dev_supers(dev, sb, max_mirrors);
3846                 if (ret)
3847                         total_errors++;
3848         }
3849         if (total_errors > max_errors) {
3850                 btrfs_err(fs_info, "%d errors while writing supers",
3851                           total_errors);
3852                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3853
3854                 /* FUA is masked off if unsupported and can't be the reason */
3855                 btrfs_handle_fs_error(fs_info, -EIO,
3856                                       "%d errors while writing supers",
3857                                       total_errors);
3858                 return -EIO;
3859         }
3860
3861         total_errors = 0;
3862         list_for_each_entry(dev, head, dev_list) {
3863                 if (!dev->bdev)
3864                         continue;
3865                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3866                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3867                         continue;
3868
3869                 ret = wait_dev_supers(dev, max_mirrors);
3870                 if (ret)
3871                         total_errors++;
3872         }
3873         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3874         if (total_errors > max_errors) {
3875                 btrfs_handle_fs_error(fs_info, -EIO,
3876                                       "%d errors while writing supers",
3877                                       total_errors);
3878                 return -EIO;
3879         }
3880         return 0;
3881 }
3882
3883 /* Drop a fs root from the radix tree and free it. */
3884 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3885                                   struct btrfs_root *root)
3886 {
3887         bool drop_ref = false;
3888
3889         spin_lock(&fs_info->fs_roots_radix_lock);
3890         radix_tree_delete(&fs_info->fs_roots_radix,
3891                           (unsigned long)root->root_key.objectid);
3892         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3893                 drop_ref = true;
3894         spin_unlock(&fs_info->fs_roots_radix_lock);
3895
3896         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3897                 ASSERT(root->log_root == NULL);
3898                 if (root->reloc_root) {
3899                         btrfs_put_root(root->reloc_root);
3900                         root->reloc_root = NULL;
3901                 }
3902         }
3903
3904         if (root->free_ino_pinned)
3905                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3906         if (root->free_ino_ctl)
3907                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3908         if (root->ino_cache_inode) {
3909                 iput(root->ino_cache_inode);
3910                 root->ino_cache_inode = NULL;
3911         }
3912         if (drop_ref)
3913                 btrfs_put_root(root);
3914 }
3915
3916 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3917 {
3918         u64 root_objectid = 0;
3919         struct btrfs_root *gang[8];
3920         int i = 0;
3921         int err = 0;
3922         unsigned int ret = 0;
3923
3924         while (1) {
3925                 spin_lock(&fs_info->fs_roots_radix_lock);
3926                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3927                                              (void **)gang, root_objectid,
3928                                              ARRAY_SIZE(gang));
3929                 if (!ret) {
3930                         spin_unlock(&fs_info->fs_roots_radix_lock);
3931                         break;
3932                 }
3933                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3934
3935                 for (i = 0; i < ret; i++) {
3936                         /* Avoid to grab roots in dead_roots */
3937                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3938                                 gang[i] = NULL;
3939                                 continue;
3940                         }
3941                         /* grab all the search result for later use */
3942                         gang[i] = btrfs_grab_root(gang[i]);
3943                 }
3944                 spin_unlock(&fs_info->fs_roots_radix_lock);
3945
3946                 for (i = 0; i < ret; i++) {
3947                         if (!gang[i])
3948                                 continue;
3949                         root_objectid = gang[i]->root_key.objectid;
3950                         err = btrfs_orphan_cleanup(gang[i]);
3951                         if (err)
3952                                 break;
3953                         btrfs_put_root(gang[i]);
3954                 }
3955                 root_objectid++;
3956         }
3957
3958         /* release the uncleaned roots due to error */
3959         for (; i < ret; i++) {
3960                 if (gang[i])
3961                         btrfs_put_root(gang[i]);
3962         }
3963         return err;
3964 }
3965
3966 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3967 {
3968         struct btrfs_root *root = fs_info->tree_root;
3969         struct btrfs_trans_handle *trans;
3970
3971         mutex_lock(&fs_info->cleaner_mutex);
3972         btrfs_run_delayed_iputs(fs_info);
3973         mutex_unlock(&fs_info->cleaner_mutex);
3974         wake_up_process(fs_info->cleaner_kthread);
3975
3976         /* wait until ongoing cleanup work done */
3977         down_write(&fs_info->cleanup_work_sem);
3978         up_write(&fs_info->cleanup_work_sem);
3979
3980         trans = btrfs_join_transaction(root);
3981         if (IS_ERR(trans))
3982                 return PTR_ERR(trans);
3983         return btrfs_commit_transaction(trans);
3984 }
3985
3986 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3987 {
3988         int ret;
3989
3990         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3991         /*
3992          * We don't want the cleaner to start new transactions, add more delayed
3993          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3994          * because that frees the task_struct, and the transaction kthread might
3995          * still try to wake up the cleaner.
3996          */
3997         kthread_park(fs_info->cleaner_kthread);
3998
3999         /* wait for the qgroup rescan worker to stop */
4000         btrfs_qgroup_wait_for_completion(fs_info, false);
4001
4002         /* wait for the uuid_scan task to finish */
4003         down(&fs_info->uuid_tree_rescan_sem);
4004         /* avoid complains from lockdep et al., set sem back to initial state */
4005         up(&fs_info->uuid_tree_rescan_sem);
4006
4007         /* pause restriper - we want to resume on mount */
4008         btrfs_pause_balance(fs_info);
4009
4010         btrfs_dev_replace_suspend_for_unmount(fs_info);
4011
4012         btrfs_scrub_cancel(fs_info);
4013
4014         /* wait for any defraggers to finish */
4015         wait_event(fs_info->transaction_wait,
4016                    (atomic_read(&fs_info->defrag_running) == 0));
4017
4018         /* clear out the rbtree of defraggable inodes */
4019         btrfs_cleanup_defrag_inodes(fs_info);
4020
4021         cancel_work_sync(&fs_info->async_reclaim_work);
4022
4023         /* Cancel or finish ongoing discard work */
4024         btrfs_discard_cleanup(fs_info);
4025
4026         if (!sb_rdonly(fs_info->sb)) {
4027                 /*
4028                  * The cleaner kthread is stopped, so do one final pass over
4029                  * unused block groups.
4030                  */
4031                 btrfs_delete_unused_bgs(fs_info);
4032
4033                 /*
4034                  * There might be existing delayed inode workers still running
4035                  * and holding an empty delayed inode item. We must wait for
4036                  * them to complete first because they can create a transaction.
4037                  * This happens when someone calls btrfs_balance_delayed_items()
4038                  * and then a transaction commit runs the same delayed nodes
4039                  * before any delayed worker has done something with the nodes.
4040                  * We must wait for any worker here and not at transaction
4041                  * commit time since that could cause a deadlock.
4042                  * This is a very rare case.
4043                  */
4044                 btrfs_flush_workqueue(fs_info->delayed_workers);
4045
4046                 ret = btrfs_commit_super(fs_info);
4047                 if (ret)
4048                         btrfs_err(fs_info, "commit super ret %d", ret);
4049         }
4050
4051         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4052             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4053                 btrfs_error_commit_super(fs_info);
4054
4055         kthread_stop(fs_info->transaction_kthread);
4056         kthread_stop(fs_info->cleaner_kthread);
4057
4058         ASSERT(list_empty(&fs_info->delayed_iputs));
4059         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4060
4061         btrfs_free_qgroup_config(fs_info);
4062         ASSERT(list_empty(&fs_info->delalloc_roots));
4063
4064         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4065                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4066                        percpu_counter_sum(&fs_info->delalloc_bytes));
4067         }
4068
4069         if (percpu_counter_sum(&fs_info->dio_bytes))
4070                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4071                            percpu_counter_sum(&fs_info->dio_bytes));
4072
4073         btrfs_sysfs_remove_mounted(fs_info);
4074         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4075
4076         btrfs_put_block_group_cache(fs_info);
4077
4078         /*
4079          * we must make sure there is not any read request to
4080          * submit after we stopping all workers.
4081          */
4082         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4083         btrfs_stop_all_workers(fs_info);
4084
4085         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4086         free_root_pointers(fs_info, true);
4087         btrfs_free_fs_roots(fs_info);
4088
4089         /*
4090          * We must free the block groups after dropping the fs_roots as we could
4091          * have had an IO error and have left over tree log blocks that aren't
4092          * cleaned up until the fs roots are freed.  This makes the block group
4093          * accounting appear to be wrong because there's pending reserved bytes,
4094          * so make sure we do the block group cleanup afterwards.
4095          */
4096         btrfs_free_block_groups(fs_info);
4097
4098         iput(fs_info->btree_inode);
4099
4100 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4101         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4102                 btrfsic_unmount(fs_info->fs_devices);
4103 #endif
4104
4105         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4106         btrfs_close_devices(fs_info->fs_devices);
4107 }
4108
4109 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4110                           int atomic)
4111 {
4112         int ret;
4113         struct inode *btree_inode = buf->pages[0]->mapping->host;
4114
4115         ret = extent_buffer_uptodate(buf);
4116         if (!ret)
4117                 return ret;
4118
4119         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4120                                     parent_transid, atomic);
4121         if (ret == -EAGAIN)
4122                 return ret;
4123         return !ret;
4124 }
4125
4126 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4127 {
4128         struct btrfs_fs_info *fs_info;
4129         struct btrfs_root *root;
4130         u64 transid = btrfs_header_generation(buf);
4131         int was_dirty;
4132
4133 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4134         /*
4135          * This is a fast path so only do this check if we have sanity tests
4136          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4137          * outside of the sanity tests.
4138          */
4139         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4140                 return;
4141 #endif
4142         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4143         fs_info = root->fs_info;
4144         btrfs_assert_tree_locked(buf);
4145         if (transid != fs_info->generation)
4146                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4147                         buf->start, transid, fs_info->generation);
4148         was_dirty = set_extent_buffer_dirty(buf);
4149         if (!was_dirty)
4150                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4151                                          buf->len,
4152                                          fs_info->dirty_metadata_batch);
4153 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4154         /*
4155          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4156          * but item data not updated.
4157          * So here we should only check item pointers, not item data.
4158          */
4159         if (btrfs_header_level(buf) == 0 &&
4160             btrfs_check_leaf_relaxed(buf)) {
4161                 btrfs_print_leaf(buf);
4162                 ASSERT(0);
4163         }
4164 #endif
4165 }
4166
4167 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4168                                         int flush_delayed)
4169 {
4170         /*
4171          * looks as though older kernels can get into trouble with
4172          * this code, they end up stuck in balance_dirty_pages forever
4173          */
4174         int ret;
4175
4176         if (current->flags & PF_MEMALLOC)
4177                 return;
4178
4179         if (flush_delayed)
4180                 btrfs_balance_delayed_items(fs_info);
4181
4182         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4183                                      BTRFS_DIRTY_METADATA_THRESH,
4184                                      fs_info->dirty_metadata_batch);
4185         if (ret > 0) {
4186                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4187         }
4188 }
4189
4190 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4191 {
4192         __btrfs_btree_balance_dirty(fs_info, 1);
4193 }
4194
4195 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4196 {
4197         __btrfs_btree_balance_dirty(fs_info, 0);
4198 }
4199
4200 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4201                       struct btrfs_key *first_key)
4202 {
4203         return btree_read_extent_buffer_pages(buf, parent_transid,
4204                                               level, first_key);
4205 }
4206
4207 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4208 {
4209         /* cleanup FS via transaction */
4210         btrfs_cleanup_transaction(fs_info);
4211
4212         mutex_lock(&fs_info->cleaner_mutex);
4213         btrfs_run_delayed_iputs(fs_info);
4214         mutex_unlock(&fs_info->cleaner_mutex);
4215
4216         down_write(&fs_info->cleanup_work_sem);
4217         up_write(&fs_info->cleanup_work_sem);
4218 }
4219
4220 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4221 {
4222         struct btrfs_root *gang[8];
4223         u64 root_objectid = 0;
4224         int ret;
4225
4226         spin_lock(&fs_info->fs_roots_radix_lock);
4227         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4228                                              (void **)gang, root_objectid,
4229                                              ARRAY_SIZE(gang))) != 0) {
4230                 int i;
4231
4232                 for (i = 0; i < ret; i++)
4233                         gang[i] = btrfs_grab_root(gang[i]);
4234                 spin_unlock(&fs_info->fs_roots_radix_lock);
4235
4236                 for (i = 0; i < ret; i++) {
4237                         if (!gang[i])
4238                                 continue;
4239                         root_objectid = gang[i]->root_key.objectid;
4240                         btrfs_free_log(NULL, gang[i]);
4241                         btrfs_put_root(gang[i]);
4242                 }
4243                 root_objectid++;
4244                 spin_lock(&fs_info->fs_roots_radix_lock);
4245         }
4246         spin_unlock(&fs_info->fs_roots_radix_lock);
4247         btrfs_free_log_root_tree(NULL, fs_info);
4248 }
4249
4250 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4251 {
4252         struct btrfs_ordered_extent *ordered;
4253
4254         spin_lock(&root->ordered_extent_lock);
4255         /*
4256          * This will just short circuit the ordered completion stuff which will
4257          * make sure the ordered extent gets properly cleaned up.
4258          */
4259         list_for_each_entry(ordered, &root->ordered_extents,
4260                             root_extent_list)
4261                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4262         spin_unlock(&root->ordered_extent_lock);
4263 }
4264
4265 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4266 {
4267         struct btrfs_root *root;
4268         struct list_head splice;
4269
4270         INIT_LIST_HEAD(&splice);
4271
4272         spin_lock(&fs_info->ordered_root_lock);
4273         list_splice_init(&fs_info->ordered_roots, &splice);
4274         while (!list_empty(&splice)) {
4275                 root = list_first_entry(&splice, struct btrfs_root,
4276                                         ordered_root);
4277                 list_move_tail(&root->ordered_root,
4278                                &fs_info->ordered_roots);
4279
4280                 spin_unlock(&fs_info->ordered_root_lock);
4281                 btrfs_destroy_ordered_extents(root);
4282
4283                 cond_resched();
4284                 spin_lock(&fs_info->ordered_root_lock);
4285         }
4286         spin_unlock(&fs_info->ordered_root_lock);
4287
4288         /*
4289          * We need this here because if we've been flipped read-only we won't
4290          * get sync() from the umount, so we need to make sure any ordered
4291          * extents that haven't had their dirty pages IO start writeout yet
4292          * actually get run and error out properly.
4293          */
4294         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4295 }
4296
4297 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4298                                       struct btrfs_fs_info *fs_info)
4299 {
4300         struct rb_node *node;
4301         struct btrfs_delayed_ref_root *delayed_refs;
4302         struct btrfs_delayed_ref_node *ref;
4303         int ret = 0;
4304
4305         delayed_refs = &trans->delayed_refs;
4306
4307         spin_lock(&delayed_refs->lock);
4308         if (atomic_read(&delayed_refs->num_entries) == 0) {
4309                 spin_unlock(&delayed_refs->lock);
4310                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4311                 return ret;
4312         }
4313
4314         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4315                 struct btrfs_delayed_ref_head *head;
4316                 struct rb_node *n;
4317                 bool pin_bytes = false;
4318
4319                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4320                                 href_node);
4321                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4322                         continue;
4323
4324                 spin_lock(&head->lock);
4325                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4326                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4327                                        ref_node);
4328                         ref->in_tree = 0;
4329                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4330                         RB_CLEAR_NODE(&ref->ref_node);
4331                         if (!list_empty(&ref->add_list))
4332                                 list_del(&ref->add_list);
4333                         atomic_dec(&delayed_refs->num_entries);
4334                         btrfs_put_delayed_ref(ref);
4335                 }
4336                 if (head->must_insert_reserved)
4337                         pin_bytes = true;
4338                 btrfs_free_delayed_extent_op(head->extent_op);
4339                 btrfs_delete_ref_head(delayed_refs, head);
4340                 spin_unlock(&head->lock);
4341                 spin_unlock(&delayed_refs->lock);
4342                 mutex_unlock(&head->mutex);
4343
4344                 if (pin_bytes) {
4345                         struct btrfs_block_group *cache;
4346
4347                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4348                         BUG_ON(!cache);
4349
4350                         spin_lock(&cache->space_info->lock);
4351                         spin_lock(&cache->lock);
4352                         cache->pinned += head->num_bytes;
4353                         btrfs_space_info_update_bytes_pinned(fs_info,
4354                                 cache->space_info, head->num_bytes);
4355                         cache->reserved -= head->num_bytes;
4356                         cache->space_info->bytes_reserved -= head->num_bytes;
4357                         spin_unlock(&cache->lock);
4358                         spin_unlock(&cache->space_info->lock);
4359                         percpu_counter_add_batch(
4360                                 &cache->space_info->total_bytes_pinned,
4361                                 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4362
4363                         btrfs_put_block_group(cache);
4364
4365                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4366                                 head->bytenr + head->num_bytes - 1);
4367                 }
4368                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4369                 btrfs_put_delayed_ref_head(head);
4370                 cond_resched();
4371                 spin_lock(&delayed_refs->lock);
4372         }
4373         btrfs_qgroup_destroy_extent_records(trans);
4374
4375         spin_unlock(&delayed_refs->lock);
4376
4377         return ret;
4378 }
4379
4380 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4381 {
4382         struct btrfs_inode *btrfs_inode;
4383         struct list_head splice;
4384
4385         INIT_LIST_HEAD(&splice);
4386
4387         spin_lock(&root->delalloc_lock);
4388         list_splice_init(&root->delalloc_inodes, &splice);
4389
4390         while (!list_empty(&splice)) {
4391                 struct inode *inode = NULL;
4392                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4393                                                delalloc_inodes);
4394                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4395                 spin_unlock(&root->delalloc_lock);
4396
4397                 /*
4398                  * Make sure we get a live inode and that it'll not disappear
4399                  * meanwhile.
4400                  */
4401                 inode = igrab(&btrfs_inode->vfs_inode);
4402                 if (inode) {
4403                         invalidate_inode_pages2(inode->i_mapping);
4404                         iput(inode);
4405                 }
4406                 spin_lock(&root->delalloc_lock);
4407         }
4408         spin_unlock(&root->delalloc_lock);
4409 }
4410
4411 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4412 {
4413         struct btrfs_root *root;
4414         struct list_head splice;
4415
4416         INIT_LIST_HEAD(&splice);
4417
4418         spin_lock(&fs_info->delalloc_root_lock);
4419         list_splice_init(&fs_info->delalloc_roots, &splice);
4420         while (!list_empty(&splice)) {
4421                 root = list_first_entry(&splice, struct btrfs_root,
4422                                          delalloc_root);
4423                 root = btrfs_grab_root(root);
4424                 BUG_ON(!root);
4425                 spin_unlock(&fs_info->delalloc_root_lock);
4426
4427                 btrfs_destroy_delalloc_inodes(root);
4428                 btrfs_put_root(root);
4429
4430                 spin_lock(&fs_info->delalloc_root_lock);
4431         }
4432         spin_unlock(&fs_info->delalloc_root_lock);
4433 }
4434
4435 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4436                                         struct extent_io_tree *dirty_pages,
4437                                         int mark)
4438 {
4439         int ret;
4440         struct extent_buffer *eb;
4441         u64 start = 0;
4442         u64 end;
4443
4444         while (1) {
4445                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4446                                             mark, NULL);
4447                 if (ret)
4448                         break;
4449
4450                 clear_extent_bits(dirty_pages, start, end, mark);
4451                 while (start <= end) {
4452                         eb = find_extent_buffer(fs_info, start);
4453                         start += fs_info->nodesize;
4454                         if (!eb)
4455                                 continue;
4456                         wait_on_extent_buffer_writeback(eb);
4457
4458                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4459                                                &eb->bflags))
4460                                 clear_extent_buffer_dirty(eb);
4461                         free_extent_buffer_stale(eb);
4462                 }
4463         }
4464
4465         return ret;
4466 }
4467
4468 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4469                                        struct extent_io_tree *unpin)
4470 {
4471         u64 start;
4472         u64 end;
4473         int ret;
4474
4475         while (1) {
4476                 struct extent_state *cached_state = NULL;
4477
4478                 /*
4479                  * The btrfs_finish_extent_commit() may get the same range as
4480                  * ours between find_first_extent_bit and clear_extent_dirty.
4481                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4482                  * the same extent range.
4483                  */
4484                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4485                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4486                                             EXTENT_DIRTY, &cached_state);
4487                 if (ret) {
4488                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4489                         break;
4490                 }
4491
4492                 clear_extent_dirty(unpin, start, end, &cached_state);
4493                 free_extent_state(cached_state);
4494                 btrfs_error_unpin_extent_range(fs_info, start, end);
4495                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4496                 cond_resched();
4497         }
4498
4499         return 0;
4500 }
4501
4502 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4503 {
4504         struct inode *inode;
4505
4506         inode = cache->io_ctl.inode;
4507         if (inode) {
4508                 invalidate_inode_pages2(inode->i_mapping);
4509                 BTRFS_I(inode)->generation = 0;
4510                 cache->io_ctl.inode = NULL;
4511                 iput(inode);
4512         }
4513         btrfs_put_block_group(cache);
4514 }
4515
4516 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4517                              struct btrfs_fs_info *fs_info)
4518 {
4519         struct btrfs_block_group *cache;
4520
4521         spin_lock(&cur_trans->dirty_bgs_lock);
4522         while (!list_empty(&cur_trans->dirty_bgs)) {
4523                 cache = list_first_entry(&cur_trans->dirty_bgs,
4524                                          struct btrfs_block_group,
4525                                          dirty_list);
4526
4527                 if (!list_empty(&cache->io_list)) {
4528                         spin_unlock(&cur_trans->dirty_bgs_lock);
4529                         list_del_init(&cache->io_list);
4530                         btrfs_cleanup_bg_io(cache);
4531                         spin_lock(&cur_trans->dirty_bgs_lock);
4532                 }
4533
4534                 list_del_init(&cache->dirty_list);
4535                 spin_lock(&cache->lock);
4536                 cache->disk_cache_state = BTRFS_DC_ERROR;
4537                 spin_unlock(&cache->lock);
4538
4539                 spin_unlock(&cur_trans->dirty_bgs_lock);
4540                 btrfs_put_block_group(cache);
4541                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4542                 spin_lock(&cur_trans->dirty_bgs_lock);
4543         }
4544         spin_unlock(&cur_trans->dirty_bgs_lock);
4545
4546         /*
4547          * Refer to the definition of io_bgs member for details why it's safe
4548          * to use it without any locking
4549          */
4550         while (!list_empty(&cur_trans->io_bgs)) {
4551                 cache = list_first_entry(&cur_trans->io_bgs,
4552                                          struct btrfs_block_group,
4553                                          io_list);
4554
4555                 list_del_init(&cache->io_list);
4556                 spin_lock(&cache->lock);
4557                 cache->disk_cache_state = BTRFS_DC_ERROR;
4558                 spin_unlock(&cache->lock);
4559                 btrfs_cleanup_bg_io(cache);
4560         }
4561 }
4562
4563 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4564                                    struct btrfs_fs_info *fs_info)
4565 {
4566         struct btrfs_device *dev, *tmp;
4567
4568         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4569         ASSERT(list_empty(&cur_trans->dirty_bgs));
4570         ASSERT(list_empty(&cur_trans->io_bgs));
4571
4572         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4573                                  post_commit_list) {
4574                 list_del_init(&dev->post_commit_list);
4575         }
4576
4577         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4578
4579         cur_trans->state = TRANS_STATE_COMMIT_START;
4580         wake_up(&fs_info->transaction_blocked_wait);
4581
4582         cur_trans->state = TRANS_STATE_UNBLOCKED;
4583         wake_up(&fs_info->transaction_wait);
4584
4585         btrfs_destroy_delayed_inodes(fs_info);
4586
4587         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4588                                      EXTENT_DIRTY);
4589         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4590
4591         cur_trans->state =TRANS_STATE_COMPLETED;
4592         wake_up(&cur_trans->commit_wait);
4593 }
4594
4595 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4596 {
4597         struct btrfs_transaction *t;
4598
4599         mutex_lock(&fs_info->transaction_kthread_mutex);
4600
4601         spin_lock(&fs_info->trans_lock);
4602         while (!list_empty(&fs_info->trans_list)) {
4603                 t = list_first_entry(&fs_info->trans_list,
4604                                      struct btrfs_transaction, list);
4605                 if (t->state >= TRANS_STATE_COMMIT_START) {
4606                         refcount_inc(&t->use_count);
4607                         spin_unlock(&fs_info->trans_lock);
4608                         btrfs_wait_for_commit(fs_info, t->transid);
4609                         btrfs_put_transaction(t);
4610                         spin_lock(&fs_info->trans_lock);
4611                         continue;
4612                 }
4613                 if (t == fs_info->running_transaction) {
4614                         t->state = TRANS_STATE_COMMIT_DOING;
4615                         spin_unlock(&fs_info->trans_lock);
4616                         /*
4617                          * We wait for 0 num_writers since we don't hold a trans
4618                          * handle open currently for this transaction.
4619                          */
4620                         wait_event(t->writer_wait,
4621                                    atomic_read(&t->num_writers) == 0);
4622                 } else {
4623                         spin_unlock(&fs_info->trans_lock);
4624                 }
4625                 btrfs_cleanup_one_transaction(t, fs_info);
4626
4627                 spin_lock(&fs_info->trans_lock);
4628                 if (t == fs_info->running_transaction)
4629                         fs_info->running_transaction = NULL;
4630                 list_del_init(&t->list);
4631                 spin_unlock(&fs_info->trans_lock);
4632
4633                 btrfs_put_transaction(t);
4634                 trace_btrfs_transaction_commit(fs_info->tree_root);
4635                 spin_lock(&fs_info->trans_lock);
4636         }
4637         spin_unlock(&fs_info->trans_lock);
4638         btrfs_destroy_all_ordered_extents(fs_info);
4639         btrfs_destroy_delayed_inodes(fs_info);
4640         btrfs_assert_delayed_root_empty(fs_info);
4641         btrfs_destroy_all_delalloc_inodes(fs_info);
4642         btrfs_drop_all_logs(fs_info);
4643         mutex_unlock(&fs_info->transaction_kthread_mutex);
4644
4645         return 0;
4646 }
4647
4648 static const struct extent_io_ops btree_extent_io_ops = {
4649         /* mandatory callbacks */
4650         .submit_bio_hook = btree_submit_bio_hook,
4651         .readpage_end_io_hook = btree_readpage_end_io_hook,
4652 };