fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <linux/sched/mm.h>
  21 #include <asm/unaligned.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42
  43 #ifdef CONFIG_X86
  44 #include <asm/cpufeature.h>
  45 #endif
  46
  47 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  48                                  BTRFS_HEADER_FLAG_RELOC |\
  49                                  BTRFS_SUPER_FLAG_ERROR |\
  50                                  BTRFS_SUPER_FLAG_SEEDING |\
  51                                  BTRFS_SUPER_FLAG_METADUMP |\
  52                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  53
  54 static const struct extent_io_ops btree_extent_io_ops;
  55 static void end_workqueue_fn(struct btrfs_work *work);
  56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  58                                       struct btrfs_fs_info *fs_info);
  59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  61                                         struct extent_io_tree *dirty_pages,
  62                                         int mark);
  63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  64                                        struct extent_io_tree *pinned_extents);
  65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  67
  68 /*
  69  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  70  * is complete.  This is used during reads to verify checksums, and it is used
  71  * by writes to insert metadata for new file extents after IO is complete.
  72  */
  73 struct btrfs_end_io_wq {
  74         struct bio *bio;
  75         bio_end_io_t *end_io;
  76         void *private;
  77         struct btrfs_fs_info *info;
  78         blk_status_t status;
  79         enum btrfs_wq_endio_type metadata;
  80         struct btrfs_work work;
  81 };
  82
  83 static struct kmem_cache *btrfs_end_io_wq_cache;
  84
  85 int __init btrfs_end_io_wq_init(void)
  86 {
  87         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  88                                         sizeof(struct btrfs_end_io_wq),
  89                                         0,
  90                                         SLAB_MEM_SPREAD,
  91                                         NULL);
  92         if (!btrfs_end_io_wq_cache)
  93                 return -ENOMEM;
  94         return 0;
  95 }
  96
  97 void __cold btrfs_end_io_wq_exit(void)
  98 {
  99         kmem_cache_destroy(btrfs_end_io_wq_cache);
 100 }
 101
 102 /*
 103  * async submit bios are used to offload expensive checksumming
 104  * onto the worker threads.  They checksum file and metadata bios
 105  * just before they are sent down the IO stack.
 106  */
 107 struct async_submit_bio {
 108         void *private_data;
 109         struct bio *bio;
 110         extent_submit_bio_start_t *submit_bio_start;
 111         int mirror_num;
 112         /*
 113          * bio_offset is optional, can be used if the pages in the bio
 114          * can't tell us where in the file the bio should go
 115          */
 116         u64 bio_offset;
 117         struct btrfs_work work;
 118         blk_status_t status;
 119 };
 120
 121 /*
 122  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 123  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 124  * the level the eb occupies in the tree.
 125  *
 126  * Different roots are used for different purposes and may nest inside each
 127  * other and they require separate keysets.  As lockdep keys should be
 128  * static, assign keysets according to the purpose of the root as indicated
 129  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 130  * roots have separate keysets.
 131  *
 132  * Lock-nesting across peer nodes is always done with the immediate parent
 133  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 134  * subclass to avoid triggering lockdep warning in such cases.
 135  *
 136  * The key is set by the readpage_end_io_hook after the buffer has passed
 137  * csum validation but before the pages are unlocked.  It is also set by
 138  * btrfs_init_new_buffer on freshly allocated blocks.
 139  *
 140  * We also add a check to make sure the highest level of the tree is the
 141  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 142  * needs update as well.
 143  */
 144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 145 # if BTRFS_MAX_LEVEL != 8
 146 #  error
 147 # endif
 148
 149 static struct btrfs_lockdep_keyset {
 150         u64                     id;             /* root objectid */
 151         const char              *name_stem;     /* lock name stem */
 152         char                    names[BTRFS_MAX_LEVEL + 1][20];
 153         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 154 } btrfs_lockdep_keysets[] = {
 155         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 156         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 157         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 158         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 159         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 160         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 161         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 162         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 163         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 164         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 165         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 166         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 167         { .id = 0,                              .name_stem = "tree"     },
 168 };
 169
 170 void __init btrfs_init_lockdep(void)
 171 {
 172         int i, j;
 173
 174         /* initialize lockdep class names */
 175         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 176                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 177
 178                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 179                         snprintf(ks->names[j], sizeof(ks->names[j]),
 180                                  "btrfs-%s-%02d", ks->name_stem, j);
 181         }
 182 }
 183
 184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 185                                     int level)
 186 {
 187         struct btrfs_lockdep_keyset *ks;
 188
 189         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 190
 191         /* find the matching keyset, id 0 is the default entry */
 192         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 193                 if (ks->id == objectid)
 194                         break;
 195
 196         lockdep_set_class_and_name(&eb->lock,
 197                                    &ks->keys[level], ks->names[level]);
 198 }
 199
 200 #endif
 201
 202 /*
 203  * extents on the btree inode are pretty simple, there's one extent
 204  * that covers the entire device
 205  */
 206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 207                 struct page *page, size_t pg_offset, u64 start, u64 len,
 208                 int create)
 209 {
 210         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 211         struct extent_map_tree *em_tree = &inode->extent_tree;
 212         struct extent_map *em;
 213         int ret;
 214
 215         read_lock(&em_tree->lock);
 216         em = lookup_extent_mapping(em_tree, start, len);
 217         if (em) {
 218                 em->bdev = fs_info->fs_devices->latest_bdev;
 219                 read_unlock(&em_tree->lock);
 220                 goto out;
 221         }
 222         read_unlock(&em_tree->lock);
 223
 224         em = alloc_extent_map();
 225         if (!em) {
 226                 em = ERR_PTR(-ENOMEM);
 227                 goto out;
 228         }
 229         em->start = 0;
 230         em->len = (u64)-1;
 231         em->block_len = (u64)-1;
 232         em->block_start = 0;
 233         em->bdev = fs_info->fs_devices->latest_bdev;
 234
 235         write_lock(&em_tree->lock);
 236         ret = add_extent_mapping(em_tree, em, 0);
 237         if (ret == -EEXIST) {
 238                 free_extent_map(em);
 239                 em = lookup_extent_mapping(em_tree, start, len);
 240                 if (!em)
 241                         em = ERR_PTR(-EIO);
 242         } else if (ret) {
 243                 free_extent_map(em);
 244                 em = ERR_PTR(ret);
 245         }
 246         write_unlock(&em_tree->lock);
 247
 248 out:
 249         return em;
 250 }
 251
 252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 253 {
 254         return crc32c(seed, data, len);
 255 }
 256
 257 void btrfs_csum_final(u32 crc, u8 *result)
 258 {
 259         put_unaligned_le32(~crc, result);
 260 }
 261
 262 /*
 263  * Compute the csum of a btree block and store the result to provided buffer.
 264  *
 265  * Returns error if the extent buffer cannot be mapped.
 266  */
 267 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
 268 {
 269         unsigned long len;
 270         unsigned long cur_len;
 271         unsigned long offset = BTRFS_CSUM_SIZE;
 272         char *kaddr;
 273         unsigned long map_start;
 274         unsigned long map_len;
 275         int err;
 276         u32 crc = ~(u32)0;
 277
 278         len = buf->len - offset;
 279         while (len > 0) {
 280                 /*
 281                  * Note: we don't need to check for the err == 1 case here, as
 282                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 283                  * and 'min_len = 32' and the currently implemented mapping
 284                  * algorithm we cannot cross a page boundary.
 285                  */
 286                 err = map_private_extent_buffer(buf, offset, 32,
 287                                         &kaddr, &map_start, &map_len);
 288                 if (WARN_ON(err))
 289                         return err;
 290                 cur_len = min(len, map_len - (offset - map_start));
 291                 crc = btrfs_csum_data(kaddr + offset - map_start,
 292                                       crc, cur_len);
 293                 len -= cur_len;
 294                 offset += cur_len;
 295         }
 296         memset(result, 0, BTRFS_CSUM_SIZE);
 297
 298         btrfs_csum_final(crc, result);
 299
 300         return 0;
 301 }
 302
 303 /*
 304  * we can't consider a given block up to date unless the transid of the
 305  * block matches the transid in the parent node's pointer.  This is how we
 306  * detect blocks that either didn't get written at all or got written
 307  * in the wrong place.
 308  */
 309 static int verify_parent_transid(struct extent_io_tree *io_tree,
 310                                  struct extent_buffer *eb, u64 parent_transid,
 311                                  int atomic)
 312 {
 313         struct extent_state *cached_state = NULL;
 314         int ret;
 315         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 316
 317         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 318                 return 0;
 319
 320         if (atomic)
 321                 return -EAGAIN;
 322
 323         if (need_lock) {
 324                 btrfs_tree_read_lock(eb);
 325                 btrfs_set_lock_blocking_read(eb);
 326         }
 327
 328         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 329                          &cached_state);
 330         if (extent_buffer_uptodate(eb) &&
 331             btrfs_header_generation(eb) == parent_transid) {
 332                 ret = 0;
 333                 goto out;
 334         }
 335         btrfs_err_rl(eb->fs_info,
 336                 "parent transid verify failed on %llu wanted %llu found %llu",
 337                         eb->start,
 338                         parent_transid, btrfs_header_generation(eb));
 339         ret = 1;
 340
 341         /*
 342          * Things reading via commit roots that don't have normal protection,
 343          * like send, can have a really old block in cache that may point at a
 344          * block that has been freed and re-allocated.  So don't clear uptodate
 345          * if we find an eb that is under IO (dirty/writeback) because we could
 346          * end up reading in the stale data and then writing it back out and
 347          * making everybody very sad.
 348          */
 349         if (!extent_buffer_under_io(eb))
 350                 clear_extent_buffer_uptodate(eb);
 351 out:
 352         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 353                              &cached_state);
 354         if (need_lock)
 355                 btrfs_tree_read_unlock_blocking(eb);
 356         return ret;
 357 }
 358
 359 /*
 360  * Return 0 if the superblock checksum type matches the checksum value of that
 361  * algorithm. Pass the raw disk superblock data.
 362  */
 363 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 364                                   char *raw_disk_sb)
 365 {
 366         struct btrfs_super_block *disk_sb =
 367                 (struct btrfs_super_block *)raw_disk_sb;
 368         u16 csum_type = btrfs_super_csum_type(disk_sb);
 369         int ret = 0;
 370
 371         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 372                 u32 crc = ~(u32)0;
 373                 char result[sizeof(crc)];
 374
 375                 /*
 376                  * The super_block structure does not span the whole
 377                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 378                  * is filled with zeros and is included in the checksum.
 379                  */
 380                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 381                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 382                 btrfs_csum_final(crc, result);
 383
 384                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 385                         ret = 1;
 386         }
 387
 388         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 389                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 390                                 csum_type);
 391                 ret = 1;
 392         }
 393
 394         return ret;
 395 }
 396
 397 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 398                            struct btrfs_key *first_key, u64 parent_transid)
 399 {
 400         struct btrfs_fs_info *fs_info = eb->fs_info;
 401         int found_level;
 402         struct btrfs_key found_key;
 403         int ret;
 404
 405         found_level = btrfs_header_level(eb);
 406         if (found_level != level) {
 407                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 408                      KERN_ERR "BTRFS: tree level check failed\n");
 409                 btrfs_err(fs_info,
 410 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 411                           eb->start, level, found_level);
 412                 return -EIO;
 413         }
 414
 415         if (!first_key)
 416                 return 0;
 417
 418         /*
 419          * For live tree block (new tree blocks in current transaction),
 420          * we need proper lock context to avoid race, which is impossible here.
 421          * So we only checks tree blocks which is read from disk, whose
 422          * generation <= fs_info->last_trans_committed.
 423          */
 424         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 425                 return 0;
 426         if (found_level)
 427                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 428         else
 429                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 430         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 431
 432         if (ret) {
 433                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 434                      KERN_ERR "BTRFS: tree first key check failed\n");
 435                 btrfs_err(fs_info,
 436 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 437                           eb->start, parent_transid, first_key->objectid,
 438                           first_key->type, first_key->offset,
 439                           found_key.objectid, found_key.type,
 440                           found_key.offset);
 441         }
 442         return ret;
 443 }
 444
 445 /*
 446  * helper to read a given tree block, doing retries as required when
 447  * the checksums don't match and we have alternate mirrors to try.
 448  *
 449  * @parent_transid:     expected transid, skip check if 0
 450  * @level:              expected level, mandatory check
 451  * @first_key:          expected key of first slot, skip check if NULL
 452  */
 453 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 454                                           u64 parent_transid, int level,
 455                                           struct btrfs_key *first_key)
 456 {
 457         struct btrfs_fs_info *fs_info = eb->fs_info;
 458         struct extent_io_tree *io_tree;
 459         int failed = 0;
 460         int ret;
 461         int num_copies = 0;
 462         int mirror_num = 0;
 463         int failed_mirror = 0;
 464
 465         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 466         while (1) {
 467                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 468                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 469                 if (!ret) {
 470                         if (verify_parent_transid(io_tree, eb,
 471                                                    parent_transid, 0))
 472                                 ret = -EIO;
 473                         else if (btrfs_verify_level_key(eb, level,
 474                                                 first_key, parent_transid))
 475                                 ret = -EUCLEAN;
 476                         else
 477                                 break;
 478                 }
 479
 480                 num_copies = btrfs_num_copies(fs_info,
 481                                               eb->start, eb->len);
 482                 if (num_copies == 1)
 483                         break;
 484
 485                 if (!failed_mirror) {
 486                         failed = 1;
 487                         failed_mirror = eb->read_mirror;
 488                 }
 489
 490                 mirror_num++;
 491                 if (mirror_num == failed_mirror)
 492                         mirror_num++;
 493
 494                 if (mirror_num > num_copies)
 495                         break;
 496         }
 497
 498         if (failed && !ret && failed_mirror)
 499                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 500
 501         return ret;
 502 }
 503
 504 /*
 505  * checksum a dirty tree block before IO.  This has extra checks to make sure
 506  * we only fill in the checksum field in the first page of a multi-page block
 507  */
 508
 509 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 510 {
 511         u64 start = page_offset(page);
 512         u64 found_start;
 513         u8 result[BTRFS_CSUM_SIZE];
 514         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 515         struct extent_buffer *eb;
 516         int ret;
 517
 518         eb = (struct extent_buffer *)page->private;
 519         if (page != eb->pages[0])
 520                 return 0;
 521
 522         found_start = btrfs_header_bytenr(eb);
 523         /*
 524          * Please do not consolidate these warnings into a single if.
 525          * It is useful to know what went wrong.
 526          */
 527         if (WARN_ON(found_start != start))
 528                 return -EUCLEAN;
 529         if (WARN_ON(!PageUptodate(page)))
 530                 return -EUCLEAN;
 531
 532         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 533                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 534
 535         if (csum_tree_block(eb, result))
 536                 return -EINVAL;
 537
 538         if (btrfs_header_level(eb))
 539                 ret = btrfs_check_node(eb);
 540         else
 541                 ret = btrfs_check_leaf_full(eb);
 542
 543         if (ret < 0) {
 544                 btrfs_err(fs_info,
 545                 "block=%llu write time tree block corruption detected",
 546                           eb->start);
 547                 return ret;
 548         }
 549         write_extent_buffer(eb, result, 0, csum_size);
 550
 551         return 0;
 552 }
 553
 554 static int check_tree_block_fsid(struct extent_buffer *eb)
 555 {
 556         struct btrfs_fs_info *fs_info = eb->fs_info;
 557         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 558         u8 fsid[BTRFS_FSID_SIZE];
 559         int ret = 1;
 560
 561         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 562         while (fs_devices) {
 563                 u8 *metadata_uuid;
 564
 565                 /*
 566                  * Checking the incompat flag is only valid for the current
 567                  * fs. For seed devices it's forbidden to have their uuid
 568                  * changed so reading ->fsid in this case is fine
 569                  */
 570                 if (fs_devices == fs_info->fs_devices &&
 571                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 572                         metadata_uuid = fs_devices->metadata_uuid;
 573                 else
 574                         metadata_uuid = fs_devices->fsid;
 575
 576                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 577                         ret = 0;
 578                         break;
 579                 }
 580                 fs_devices = fs_devices->seed;
 581         }
 582         return ret;
 583 }
 584
 585 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 586                                       u64 phy_offset, struct page *page,
 587                                       u64 start, u64 end, int mirror)
 588 {
 589         u64 found_start;
 590         int found_level;
 591         struct extent_buffer *eb;
 592         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 593         struct btrfs_fs_info *fs_info = root->fs_info;
 594         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 595         int ret = 0;
 596         u8 result[BTRFS_CSUM_SIZE];
 597         int reads_done;
 598
 599         if (!page->private)
 600                 goto out;
 601
 602         eb = (struct extent_buffer *)page->private;
 603
 604         /* the pending IO might have been the only thing that kept this buffer
 605          * in memory.  Make sure we have a ref for all this other checks
 606          */
 607         extent_buffer_get(eb);
 608
 609         reads_done = atomic_dec_and_test(&eb->io_pages);
 610         if (!reads_done)
 611                 goto err;
 612
 613         eb->read_mirror = mirror;
 614         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 615                 ret = -EIO;
 616                 goto err;
 617         }
 618
 619         found_start = btrfs_header_bytenr(eb);
 620         if (found_start != eb->start) {
 621                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 622                              eb->start, found_start);
 623                 ret = -EIO;
 624                 goto err;
 625         }
 626         if (check_tree_block_fsid(eb)) {
 627                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 628                              eb->start);
 629                 ret = -EIO;
 630                 goto err;
 631         }
 632         found_level = btrfs_header_level(eb);
 633         if (found_level >= BTRFS_MAX_LEVEL) {
 634                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 635                           (int)btrfs_header_level(eb), eb->start);
 636                 ret = -EIO;
 637                 goto err;
 638         }
 639
 640         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 641                                        eb, found_level);
 642
 643         ret = csum_tree_block(eb, result);
 644         if (ret)
 645                 goto err;
 646
 647         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 648                 u32 val;
 649                 u32 found = 0;
 650
 651                 memcpy(&found, result, csum_size);
 652
 653                 read_extent_buffer(eb, &val, 0, csum_size);
 654                 btrfs_warn_rl(fs_info,
 655                 "%s checksum verify failed on %llu wanted %x found %x level %d",
 656                               fs_info->sb->s_id, eb->start,
 657                               val, found, btrfs_header_level(eb));
 658                 ret = -EUCLEAN;
 659                 goto err;
 660         }
 661
 662         /*
 663          * If this is a leaf block and it is corrupt, set the corrupt bit so
 664          * that we don't try and read the other copies of this block, just
 665          * return -EIO.
 666          */
 667         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 668                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 669                 ret = -EIO;
 670         }
 671
 672         if (found_level > 0 && btrfs_check_node(eb))
 673                 ret = -EIO;
 674
 675         if (!ret)
 676                 set_extent_buffer_uptodate(eb);
 677         else
 678                 btrfs_err(fs_info,
 679                           "block=%llu read time tree block corruption detected",
 680                           eb->start);
 681 err:
 682         if (reads_done &&
 683             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 684                 btree_readahead_hook(eb, ret);
 685
 686         if (ret) {
 687                 /*
 688                  * our io error hook is going to dec the io pages
 689                  * again, we have to make sure it has something
 690                  * to decrement
 691                  */
 692                 atomic_inc(&eb->io_pages);
 693                 clear_extent_buffer_uptodate(eb);
 694         }
 695         free_extent_buffer(eb);
 696 out:
 697         return ret;
 698 }
 699
 700 static void end_workqueue_bio(struct bio *bio)
 701 {
 702         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 703         struct btrfs_fs_info *fs_info;
 704         struct btrfs_workqueue *wq;
 705         btrfs_work_func_t func;
 706
 707         fs_info = end_io_wq->info;
 708         end_io_wq->status = bio->bi_status;
 709
 710         if (bio_op(bio) == REQ_OP_WRITE) {
 711                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 712                         wq = fs_info->endio_meta_write_workers;
 713                         func = btrfs_endio_meta_write_helper;
 714                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 715                         wq = fs_info->endio_freespace_worker;
 716                         func = btrfs_freespace_write_helper;
 717                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 718                         wq = fs_info->endio_raid56_workers;
 719                         func = btrfs_endio_raid56_helper;
 720                 } else {
 721                         wq = fs_info->endio_write_workers;
 722                         func = btrfs_endio_write_helper;
 723                 }
 724         } else {
 725                 if (unlikely(end_io_wq->metadata ==
 726                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 727                         wq = fs_info->endio_repair_workers;
 728                         func = btrfs_endio_repair_helper;
 729                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 730                         wq = fs_info->endio_raid56_workers;
 731                         func = btrfs_endio_raid56_helper;
 732                 } else if (end_io_wq->metadata) {
 733                         wq = fs_info->endio_meta_workers;
 734                         func = btrfs_endio_meta_helper;
 735                 } else {
 736                         wq = fs_info->endio_workers;
 737                         func = btrfs_endio_helper;
 738                 }
 739         }
 740
 741         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 742         btrfs_queue_work(wq, &end_io_wq->work);
 743 }
 744
 745 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 746                         enum btrfs_wq_endio_type metadata)
 747 {
 748         struct btrfs_end_io_wq *end_io_wq;
 749
 750         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 751         if (!end_io_wq)
 752                 return BLK_STS_RESOURCE;
 753
 754         end_io_wq->private = bio->bi_private;
 755         end_io_wq->end_io = bio->bi_end_io;
 756         end_io_wq->info = info;
 757         end_io_wq->status = 0;
 758         end_io_wq->bio = bio;
 759         end_io_wq->metadata = metadata;
 760
 761         bio->bi_private = end_io_wq;
 762         bio->bi_end_io = end_workqueue_bio;
 763         return 0;
 764 }
 765
 766 static void run_one_async_start(struct btrfs_work *work)
 767 {
 768         struct async_submit_bio *async;
 769         blk_status_t ret;
 770
 771         async = container_of(work, struct  async_submit_bio, work);
 772         ret = async->submit_bio_start(async->private_data, async->bio,
 773                                       async->bio_offset);
 774         if (ret)
 775                 async->status = ret;
 776 }
 777
 778 /*
 779  * In order to insert checksums into the metadata in large chunks, we wait
 780  * until bio submission time.   All the pages in the bio are checksummed and
 781  * sums are attached onto the ordered extent record.
 782  *
 783  * At IO completion time the csums attached on the ordered extent record are
 784  * inserted into the tree.
 785  */
 786 static void run_one_async_done(struct btrfs_work *work)
 787 {
 788         struct async_submit_bio *async;
 789         struct inode *inode;
 790         blk_status_t ret;
 791
 792         async = container_of(work, struct  async_submit_bio, work);
 793         inode = async->private_data;
 794
 795         /* If an error occurred we just want to clean up the bio and move on */
 796         if (async->status) {
 797                 async->bio->bi_status = async->status;
 798                 bio_endio(async->bio);
 799                 return;
 800         }
 801
 802         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 803                         async->mirror_num, 1);
 804         if (ret) {
 805                 async->bio->bi_status = ret;
 806                 bio_endio(async->bio);
 807         }
 808 }
 809
 810 static void run_one_async_free(struct btrfs_work *work)
 811 {
 812         struct async_submit_bio *async;
 813
 814         async = container_of(work, struct  async_submit_bio, work);
 815         kfree(async);
 816 }
 817
 818 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 819                                  int mirror_num, unsigned long bio_flags,
 820                                  u64 bio_offset, void *private_data,
 821                                  extent_submit_bio_start_t *submit_bio_start)
 822 {
 823         struct async_submit_bio *async;
 824
 825         async = kmalloc(sizeof(*async), GFP_NOFS);
 826         if (!async)
 827                 return BLK_STS_RESOURCE;
 828
 829         async->private_data = private_data;
 830         async->bio = bio;
 831         async->mirror_num = mirror_num;
 832         async->submit_bio_start = submit_bio_start;
 833
 834         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 835                         run_one_async_done, run_one_async_free);
 836
 837         async->bio_offset = bio_offset;
 838
 839         async->status = 0;
 840
 841         if (op_is_sync(bio->bi_opf))
 842                 btrfs_set_work_high_priority(&async->work);
 843
 844         btrfs_queue_work(fs_info->workers, &async->work);
 845         return 0;
 846 }
 847
 848 static blk_status_t btree_csum_one_bio(struct bio *bio)
 849 {
 850         struct bio_vec *bvec;
 851         struct btrfs_root *root;
 852         int ret = 0;
 853         struct bvec_iter_all iter_all;
 854
 855         ASSERT(!bio_flagged(bio, BIO_CLONED));
 856         bio_for_each_segment_all(bvec, bio, iter_all) {
 857                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 858                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 859                 if (ret)
 860                         break;
 861         }
 862
 863         return errno_to_blk_status(ret);
 864 }
 865
 866 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 867                                              u64 bio_offset)
 868 {
 869         /*
 870          * when we're called for a write, we're already in the async
 871          * submission context.  Just jump into btrfs_map_bio
 872          */
 873         return btree_csum_one_bio(bio);
 874 }
 875
 876 static int check_async_write(struct btrfs_inode *bi)
 877 {
 878         if (atomic_read(&bi->sync_writers))
 879                 return 0;
 880 #ifdef CONFIG_X86
 881         if (static_cpu_has(X86_FEATURE_XMM4_2))
 882                 return 0;
 883 #endif
 884         return 1;
 885 }
 886
 887 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 888                                           int mirror_num,
 889                                           unsigned long bio_flags)
 890 {
 891         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 892         int async = check_async_write(BTRFS_I(inode));
 893         blk_status_t ret;
 894
 895         if (bio_op(bio) != REQ_OP_WRITE) {
 896                 /*
 897                  * called for a read, do the setup so that checksum validation
 898                  * can happen in the async kernel threads
 899                  */
 900                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 901                                           BTRFS_WQ_ENDIO_METADATA);
 902                 if (ret)
 903                         goto out_w_error;
 904                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 905         } else if (!async) {
 906                 ret = btree_csum_one_bio(bio);
 907                 if (ret)
 908                         goto out_w_error;
 909                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 910         } else {
 911                 /*
 912                  * kthread helpers are used to submit writes so that
 913                  * checksumming can happen in parallel across all CPUs
 914                  */
 915                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 916                                           0, inode, btree_submit_bio_start);
 917         }
 918
 919         if (ret)
 920                 goto out_w_error;
 921         return 0;
 922
 923 out_w_error:
 924         bio->bi_status = ret;
 925         bio_endio(bio);
 926         return ret;
 927 }
 928
 929 #ifdef CONFIG_MIGRATION
 930 static int btree_migratepage(struct address_space *mapping,
 931                         struct page *newpage, struct page *page,
 932                         enum migrate_mode mode)
 933 {
 934         /*
 935          * we can't safely write a btree page from here,
 936          * we haven't done the locking hook
 937          */
 938         if (PageDirty(page))
 939                 return -EAGAIN;
 940         /*
 941          * Buffers may be managed in a filesystem specific way.
 942          * We must have no buffers or drop them.
 943          */
 944         if (page_has_private(page) &&
 945             !try_to_release_page(page, GFP_KERNEL))
 946                 return -EAGAIN;
 947         return migrate_page(mapping, newpage, page, mode);
 948 }
 949 #endif
 950
 951
 952 static int btree_writepages(struct address_space *mapping,
 953                             struct writeback_control *wbc)
 954 {
 955         struct btrfs_fs_info *fs_info;
 956         int ret;
 957
 958         if (wbc->sync_mode == WB_SYNC_NONE) {
 959
 960                 if (wbc->for_kupdate)
 961                         return 0;
 962
 963                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 964                 /* this is a bit racy, but that's ok */
 965                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 966                                              BTRFS_DIRTY_METADATA_THRESH,
 967                                              fs_info->dirty_metadata_batch);
 968                 if (ret < 0)
 969                         return 0;
 970         }
 971         return btree_write_cache_pages(mapping, wbc);
 972 }
 973
 974 static int btree_readpage(struct file *file, struct page *page)
 975 {
 976         struct extent_io_tree *tree;
 977         tree = &BTRFS_I(page->mapping->host)->io_tree;
 978         return extent_read_full_page(tree, page, btree_get_extent, 0);
 979 }
 980
 981 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 982 {
 983         if (PageWriteback(page) || PageDirty(page))
 984                 return 0;
 985
 986         return try_release_extent_buffer(page);
 987 }
 988
 989 static void btree_invalidatepage(struct page *page, unsigned int offset,
 990                                  unsigned int length)
 991 {
 992         struct extent_io_tree *tree;
 993         tree = &BTRFS_I(page->mapping->host)->io_tree;
 994         extent_invalidatepage(tree, page, offset);
 995         btree_releasepage(page, GFP_NOFS);
 996         if (PagePrivate(page)) {
 997                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 998                            "page private not zero on page %llu",
 999                            (unsigned long long)page_offset(page));
1000                 ClearPagePrivate(page);
1001                 set_page_private(page, 0);
1002                 put_page(page);
1003         }
1004 }
1005
1006 static int btree_set_page_dirty(struct page *page)
1007 {
1008 #ifdef DEBUG
1009         struct extent_buffer *eb;
1010
1011         BUG_ON(!PagePrivate(page));
1012         eb = (struct extent_buffer *)page->private;
1013         BUG_ON(!eb);
1014         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1015         BUG_ON(!atomic_read(&eb->refs));
1016         btrfs_assert_tree_locked(eb);
1017 #endif
1018         return __set_page_dirty_nobuffers(page);
1019 }
1020
1021 static const struct address_space_operations btree_aops = {
1022         .readpage       = btree_readpage,
1023         .writepages     = btree_writepages,
1024         .releasepage    = btree_releasepage,
1025         .invalidatepage = btree_invalidatepage,
1026 #ifdef CONFIG_MIGRATION
1027         .migratepage    = btree_migratepage,
1028 #endif
1029         .set_page_dirty = btree_set_page_dirty,
1030 };
1031
1032 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1033 {
1034         struct extent_buffer *buf = NULL;
1035         int ret;
1036
1037         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1038         if (IS_ERR(buf))
1039                 return;
1040
1041         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1042         if (ret < 0)
1043                 free_extent_buffer_stale(buf);
1044         else
1045                 free_extent_buffer(buf);
1046 }
1047
1048 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1049                          int mirror_num, struct extent_buffer **eb)
1050 {
1051         struct extent_buffer *buf = NULL;
1052         int ret;
1053
1054         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1055         if (IS_ERR(buf))
1056                 return 0;
1057
1058         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1059
1060         ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num);
1061         if (ret) {
1062                 free_extent_buffer_stale(buf);
1063                 return ret;
1064         }
1065
1066         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1067                 free_extent_buffer_stale(buf);
1068                 return -EIO;
1069         } else if (extent_buffer_uptodate(buf)) {
1070                 *eb = buf;
1071         } else {
1072                 free_extent_buffer(buf);
1073         }
1074         return 0;
1075 }
1076
1077 struct extent_buffer *btrfs_find_create_tree_block(
1078                                                 struct btrfs_fs_info *fs_info,
1079                                                 u64 bytenr)
1080 {
1081         if (btrfs_is_testing(fs_info))
1082                 return alloc_test_extent_buffer(fs_info, bytenr);
1083         return alloc_extent_buffer(fs_info, bytenr);
1084 }
1085
1086 /*
1087  * Read tree block at logical address @bytenr and do variant basic but critical
1088  * verification.
1089  *
1090  * @parent_transid:     expected transid of this tree block, skip check if 0
1091  * @level:              expected level, mandatory check
1092  * @first_key:          expected key in slot 0, skip check if NULL
1093  */
1094 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1095                                       u64 parent_transid, int level,
1096                                       struct btrfs_key *first_key)
1097 {
1098         struct extent_buffer *buf = NULL;
1099         int ret;
1100
1101         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1102         if (IS_ERR(buf))
1103                 return buf;
1104
1105         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1106                                              level, first_key);
1107         if (ret) {
1108                 free_extent_buffer_stale(buf);
1109                 return ERR_PTR(ret);
1110         }
1111         return buf;
1112
1113 }
1114
1115 void btrfs_clean_tree_block(struct extent_buffer *buf)
1116 {
1117         struct btrfs_fs_info *fs_info = buf->fs_info;
1118         if (btrfs_header_generation(buf) ==
1119             fs_info->running_transaction->transid) {
1120                 btrfs_assert_tree_locked(buf);
1121
1122                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1123                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1124                                                  -buf->len,
1125                                                  fs_info->dirty_metadata_batch);
1126                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1127                         btrfs_set_lock_blocking_write(buf);
1128                         clear_extent_buffer_dirty(buf);
1129                 }
1130         }
1131 }
1132
1133 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1134 {
1135         struct btrfs_subvolume_writers *writers;
1136         int ret;
1137
1138         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1139         if (!writers)
1140                 return ERR_PTR(-ENOMEM);
1141
1142         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1143         if (ret < 0) {
1144                 kfree(writers);
1145                 return ERR_PTR(ret);
1146         }
1147
1148         init_waitqueue_head(&writers->wait);
1149         return writers;
1150 }
1151
1152 static void
1153 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1154 {
1155         percpu_counter_destroy(&writers->counter);
1156         kfree(writers);
1157 }
1158
1159 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1160                          u64 objectid)
1161 {
1162         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1163         root->node = NULL;
1164         root->commit_root = NULL;
1165         root->state = 0;
1166         root->orphan_cleanup_state = 0;
1167
1168         root->last_trans = 0;
1169         root->highest_objectid = 0;
1170         root->nr_delalloc_inodes = 0;
1171         root->nr_ordered_extents = 0;
1172         root->inode_tree = RB_ROOT;
1173         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1174         root->block_rsv = NULL;
1175
1176         INIT_LIST_HEAD(&root->dirty_list);
1177         INIT_LIST_HEAD(&root->root_list);
1178         INIT_LIST_HEAD(&root->delalloc_inodes);
1179         INIT_LIST_HEAD(&root->delalloc_root);
1180         INIT_LIST_HEAD(&root->ordered_extents);
1181         INIT_LIST_HEAD(&root->ordered_root);
1182         INIT_LIST_HEAD(&root->reloc_dirty_list);
1183         INIT_LIST_HEAD(&root->logged_list[0]);
1184         INIT_LIST_HEAD(&root->logged_list[1]);
1185         spin_lock_init(&root->inode_lock);
1186         spin_lock_init(&root->delalloc_lock);
1187         spin_lock_init(&root->ordered_extent_lock);
1188         spin_lock_init(&root->accounting_lock);
1189         spin_lock_init(&root->log_extents_lock[0]);
1190         spin_lock_init(&root->log_extents_lock[1]);
1191         spin_lock_init(&root->qgroup_meta_rsv_lock);
1192         mutex_init(&root->objectid_mutex);
1193         mutex_init(&root->log_mutex);
1194         mutex_init(&root->ordered_extent_mutex);
1195         mutex_init(&root->delalloc_mutex);
1196         init_waitqueue_head(&root->log_writer_wait);
1197         init_waitqueue_head(&root->log_commit_wait[0]);
1198         init_waitqueue_head(&root->log_commit_wait[1]);
1199         INIT_LIST_HEAD(&root->log_ctxs[0]);
1200         INIT_LIST_HEAD(&root->log_ctxs[1]);
1201         atomic_set(&root->log_commit[0], 0);
1202         atomic_set(&root->log_commit[1], 0);
1203         atomic_set(&root->log_writers, 0);
1204         atomic_set(&root->log_batch, 0);
1205         refcount_set(&root->refs, 1);
1206         atomic_set(&root->will_be_snapshotted, 0);
1207         atomic_set(&root->snapshot_force_cow, 0);
1208         atomic_set(&root->nr_swapfiles, 0);
1209         root->log_transid = 0;
1210         root->log_transid_committed = -1;
1211         root->last_log_commit = 0;
1212         if (!dummy)
1213                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1214                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1215
1216         memset(&root->root_key, 0, sizeof(root->root_key));
1217         memset(&root->root_item, 0, sizeof(root->root_item));
1218         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1219         if (!dummy)
1220                 root->defrag_trans_start = fs_info->generation;
1221         else
1222                 root->defrag_trans_start = 0;
1223         root->root_key.objectid = objectid;
1224         root->anon_dev = 0;
1225
1226         spin_lock_init(&root->root_item_lock);
1227         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1228 }
1229
1230 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1231                 gfp_t flags)
1232 {
1233         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1234         if (root)
1235                 root->fs_info = fs_info;
1236         return root;
1237 }
1238
1239 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1240 /* Should only be used by the testing infrastructure */
1241 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1242 {
1243         struct btrfs_root *root;
1244
1245         if (!fs_info)
1246                 return ERR_PTR(-EINVAL);
1247
1248         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1249         if (!root)
1250                 return ERR_PTR(-ENOMEM);
1251
1252         /* We don't use the stripesize in selftest, set it as sectorsize */
1253         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1254         root->alloc_bytenr = 0;
1255
1256         return root;
1257 }
1258 #endif
1259
1260 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1261                                      u64 objectid)
1262 {
1263         struct btrfs_fs_info *fs_info = trans->fs_info;
1264         struct extent_buffer *leaf;
1265         struct btrfs_root *tree_root = fs_info->tree_root;
1266         struct btrfs_root *root;
1267         struct btrfs_key key;
1268         unsigned int nofs_flag;
1269         int ret = 0;
1270         uuid_le uuid = NULL_UUID_LE;
1271
1272         /*
1273          * We're holding a transaction handle, so use a NOFS memory allocation
1274          * context to avoid deadlock if reclaim happens.
1275          */
1276         nofs_flag = memalloc_nofs_save();
1277         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1278         memalloc_nofs_restore(nofs_flag);
1279         if (!root)
1280                 return ERR_PTR(-ENOMEM);
1281
1282         __setup_root(root, fs_info, objectid);
1283         root->root_key.objectid = objectid;
1284         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1285         root->root_key.offset = 0;
1286
1287         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1288         if (IS_ERR(leaf)) {
1289                 ret = PTR_ERR(leaf);
1290                 leaf = NULL;
1291                 goto fail;
1292         }
1293
1294         root->node = leaf;
1295         btrfs_mark_buffer_dirty(leaf);
1296
1297         root->commit_root = btrfs_root_node(root);
1298         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1299
1300         root->root_item.flags = 0;
1301         root->root_item.byte_limit = 0;
1302         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1303         btrfs_set_root_generation(&root->root_item, trans->transid);
1304         btrfs_set_root_level(&root->root_item, 0);
1305         btrfs_set_root_refs(&root->root_item, 1);
1306         btrfs_set_root_used(&root->root_item, leaf->len);
1307         btrfs_set_root_last_snapshot(&root->root_item, 0);
1308         btrfs_set_root_dirid(&root->root_item, 0);
1309         if (is_fstree(objectid))
1310                 uuid_le_gen(&uuid);
1311         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1312         root->root_item.drop_level = 0;
1313
1314         key.objectid = objectid;
1315         key.type = BTRFS_ROOT_ITEM_KEY;
1316         key.offset = 0;
1317         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1318         if (ret)
1319                 goto fail;
1320
1321         btrfs_tree_unlock(leaf);
1322
1323         return root;
1324
1325 fail:
1326         if (leaf) {
1327                 btrfs_tree_unlock(leaf);
1328                 free_extent_buffer(root->commit_root);
1329                 free_extent_buffer(leaf);
1330         }
1331         kfree(root);
1332
1333         return ERR_PTR(ret);
1334 }
1335
1336 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1337                                          struct btrfs_fs_info *fs_info)
1338 {
1339         struct btrfs_root *root;
1340         struct extent_buffer *leaf;
1341
1342         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1343         if (!root)
1344                 return ERR_PTR(-ENOMEM);
1345
1346         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1347
1348         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1349         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1350         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1351
1352         /*
1353          * DON'T set REF_COWS for log trees
1354          *
1355          * log trees do not get reference counted because they go away
1356          * before a real commit is actually done.  They do store pointers
1357          * to file data extents, and those reference counts still get
1358          * updated (along with back refs to the log tree).
1359          */
1360
1361         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1362                         NULL, 0, 0, 0);
1363         if (IS_ERR(leaf)) {
1364                 kfree(root);
1365                 return ERR_CAST(leaf);
1366         }
1367
1368         root->node = leaf;
1369
1370         btrfs_mark_buffer_dirty(root->node);
1371         btrfs_tree_unlock(root->node);
1372         return root;
1373 }
1374
1375 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1376                              struct btrfs_fs_info *fs_info)
1377 {
1378         struct btrfs_root *log_root;
1379
1380         log_root = alloc_log_tree(trans, fs_info);
1381         if (IS_ERR(log_root))
1382                 return PTR_ERR(log_root);
1383         WARN_ON(fs_info->log_root_tree);
1384         fs_info->log_root_tree = log_root;
1385         return 0;
1386 }
1387
1388 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1389                        struct btrfs_root *root)
1390 {
1391         struct btrfs_fs_info *fs_info = root->fs_info;
1392         struct btrfs_root *log_root;
1393         struct btrfs_inode_item *inode_item;
1394
1395         log_root = alloc_log_tree(trans, fs_info);
1396         if (IS_ERR(log_root))
1397                 return PTR_ERR(log_root);
1398
1399         log_root->last_trans = trans->transid;
1400         log_root->root_key.offset = root->root_key.objectid;
1401
1402         inode_item = &log_root->root_item.inode;
1403         btrfs_set_stack_inode_generation(inode_item, 1);
1404         btrfs_set_stack_inode_size(inode_item, 3);
1405         btrfs_set_stack_inode_nlink(inode_item, 1);
1406         btrfs_set_stack_inode_nbytes(inode_item,
1407                                      fs_info->nodesize);
1408         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1409
1410         btrfs_set_root_node(&log_root->root_item, log_root->node);
1411
1412         WARN_ON(root->log_root);
1413         root->log_root = log_root;
1414         root->log_transid = 0;
1415         root->log_transid_committed = -1;
1416         root->last_log_commit = 0;
1417         return 0;
1418 }
1419
1420 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1421                                                struct btrfs_key *key)
1422 {
1423         struct btrfs_root *root;
1424         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1425         struct btrfs_path *path;
1426         u64 generation;
1427         int ret;
1428         int level;
1429
1430         path = btrfs_alloc_path();
1431         if (!path)
1432                 return ERR_PTR(-ENOMEM);
1433
1434         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1435         if (!root) {
1436                 ret = -ENOMEM;
1437                 goto alloc_fail;
1438         }
1439
1440         __setup_root(root, fs_info, key->objectid);
1441
1442         ret = btrfs_find_root(tree_root, key, path,
1443                               &root->root_item, &root->root_key);
1444         if (ret) {
1445                 if (ret > 0)
1446                         ret = -ENOENT;
1447                 goto find_fail;
1448         }
1449
1450         generation = btrfs_root_generation(&root->root_item);
1451         level = btrfs_root_level(&root->root_item);
1452         root->node = read_tree_block(fs_info,
1453                                      btrfs_root_bytenr(&root->root_item),
1454                                      generation, level, NULL);
1455         if (IS_ERR(root->node)) {
1456                 ret = PTR_ERR(root->node);
1457                 goto find_fail;
1458         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1459                 ret = -EIO;
1460                 free_extent_buffer(root->node);
1461                 goto find_fail;
1462         }
1463         root->commit_root = btrfs_root_node(root);
1464 out:
1465         btrfs_free_path(path);
1466         return root;
1467
1468 find_fail:
1469         kfree(root);
1470 alloc_fail:
1471         root = ERR_PTR(ret);
1472         goto out;
1473 }
1474
1475 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1476                                       struct btrfs_key *location)
1477 {
1478         struct btrfs_root *root;
1479
1480         root = btrfs_read_tree_root(tree_root, location);
1481         if (IS_ERR(root))
1482                 return root;
1483
1484         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1485                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1486                 btrfs_check_and_init_root_item(&root->root_item);
1487         }
1488
1489         return root;
1490 }
1491
1492 int btrfs_init_fs_root(struct btrfs_root *root)
1493 {
1494         int ret;
1495         struct btrfs_subvolume_writers *writers;
1496
1497         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1498         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1499                                         GFP_NOFS);
1500         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1501                 ret = -ENOMEM;
1502                 goto fail;
1503         }
1504
1505         writers = btrfs_alloc_subvolume_writers();
1506         if (IS_ERR(writers)) {
1507                 ret = PTR_ERR(writers);
1508                 goto fail;
1509         }
1510         root->subv_writers = writers;
1511
1512         btrfs_init_free_ino_ctl(root);
1513         spin_lock_init(&root->ino_cache_lock);
1514         init_waitqueue_head(&root->ino_cache_wait);
1515
1516         ret = get_anon_bdev(&root->anon_dev);
1517         if (ret)
1518                 goto fail;
1519
1520         mutex_lock(&root->objectid_mutex);
1521         ret = btrfs_find_highest_objectid(root,
1522                                         &root->highest_objectid);
1523         if (ret) {
1524                 mutex_unlock(&root->objectid_mutex);
1525                 goto fail;
1526         }
1527
1528         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1529
1530         mutex_unlock(&root->objectid_mutex);
1531
1532         return 0;
1533 fail:
1534         /* The caller is responsible to call btrfs_free_fs_root */
1535         return ret;
1536 }
1537
1538 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1539                                         u64 root_id)
1540 {
1541         struct btrfs_root *root;
1542
1543         spin_lock(&fs_info->fs_roots_radix_lock);
1544         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1545                                  (unsigned long)root_id);
1546         spin_unlock(&fs_info->fs_roots_radix_lock);
1547         return root;
1548 }
1549
1550 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1551                          struct btrfs_root *root)
1552 {
1553         int ret;
1554
1555         ret = radix_tree_preload(GFP_NOFS);
1556         if (ret)
1557                 return ret;
1558
1559         spin_lock(&fs_info->fs_roots_radix_lock);
1560         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1561                                 (unsigned long)root->root_key.objectid,
1562                                 root);
1563         if (ret == 0)
1564                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1565         spin_unlock(&fs_info->fs_roots_radix_lock);
1566         radix_tree_preload_end();
1567
1568         return ret;
1569 }
1570
1571 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1572                                      struct btrfs_key *location,
1573                                      bool check_ref)
1574 {
1575         struct btrfs_root *root;
1576         struct btrfs_path *path;
1577         struct btrfs_key key;
1578         int ret;
1579
1580         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1581                 return fs_info->tree_root;
1582         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1583                 return fs_info->extent_root;
1584         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1585                 return fs_info->chunk_root;
1586         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1587                 return fs_info->dev_root;
1588         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1589                 return fs_info->csum_root;
1590         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1591                 return fs_info->quota_root ? fs_info->quota_root :
1592                                              ERR_PTR(-ENOENT);
1593         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1594                 return fs_info->uuid_root ? fs_info->uuid_root :
1595                                             ERR_PTR(-ENOENT);
1596         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1597                 return fs_info->free_space_root ? fs_info->free_space_root :
1598                                                   ERR_PTR(-ENOENT);
1599 again:
1600         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1601         if (root) {
1602                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1603                         return ERR_PTR(-ENOENT);
1604                 return root;
1605         }
1606
1607         root = btrfs_read_fs_root(fs_info->tree_root, location);
1608         if (IS_ERR(root))
1609                 return root;
1610
1611         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1612                 ret = -ENOENT;
1613                 goto fail;
1614         }
1615
1616         ret = btrfs_init_fs_root(root);
1617         if (ret)
1618                 goto fail;
1619
1620         path = btrfs_alloc_path();
1621         if (!path) {
1622                 ret = -ENOMEM;
1623                 goto fail;
1624         }
1625         key.objectid = BTRFS_ORPHAN_OBJECTID;
1626         key.type = BTRFS_ORPHAN_ITEM_KEY;
1627         key.offset = location->objectid;
1628
1629         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1630         btrfs_free_path(path);
1631         if (ret < 0)
1632                 goto fail;
1633         if (ret == 0)
1634                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1635
1636         ret = btrfs_insert_fs_root(fs_info, root);
1637         if (ret) {
1638                 if (ret == -EEXIST) {
1639                         btrfs_free_fs_root(root);
1640                         goto again;
1641                 }
1642                 goto fail;
1643         }
1644         return root;
1645 fail:
1646         btrfs_free_fs_root(root);
1647         return ERR_PTR(ret);
1648 }
1649
1650 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1651 {
1652         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1653         int ret = 0;
1654         struct btrfs_device *device;
1655         struct backing_dev_info *bdi;
1656
1657         rcu_read_lock();
1658         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1659                 if (!device->bdev)
1660                         continue;
1661                 bdi = device->bdev->bd_bdi;
1662                 if (bdi_congested(bdi, bdi_bits)) {
1663                         ret = 1;
1664                         break;
1665                 }
1666         }
1667         rcu_read_unlock();
1668         return ret;
1669 }
1670
1671 /*
1672  * called by the kthread helper functions to finally call the bio end_io
1673  * functions.  This is where read checksum verification actually happens
1674  */
1675 static void end_workqueue_fn(struct btrfs_work *work)
1676 {
1677         struct bio *bio;
1678         struct btrfs_end_io_wq *end_io_wq;
1679
1680         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1681         bio = end_io_wq->bio;
1682
1683         bio->bi_status = end_io_wq->status;
1684         bio->bi_private = end_io_wq->private;
1685         bio->bi_end_io = end_io_wq->end_io;
1686         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1687         bio_endio(bio);
1688 }
1689
1690 static int cleaner_kthread(void *arg)
1691 {
1692         struct btrfs_root *root = arg;
1693         struct btrfs_fs_info *fs_info = root->fs_info;
1694         int again;
1695
1696         while (1) {
1697                 again = 0;
1698
1699                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1700
1701                 /* Make the cleaner go to sleep early. */
1702                 if (btrfs_need_cleaner_sleep(fs_info))
1703                         goto sleep;
1704
1705                 /*
1706                  * Do not do anything if we might cause open_ctree() to block
1707                  * before we have finished mounting the filesystem.
1708                  */
1709                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1710                         goto sleep;
1711
1712                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1713                         goto sleep;
1714
1715                 /*
1716                  * Avoid the problem that we change the status of the fs
1717                  * during the above check and trylock.
1718                  */
1719                 if (btrfs_need_cleaner_sleep(fs_info)) {
1720                         mutex_unlock(&fs_info->cleaner_mutex);
1721                         goto sleep;
1722                 }
1723
1724                 btrfs_run_delayed_iputs(fs_info);
1725
1726                 again = btrfs_clean_one_deleted_snapshot(root);
1727                 mutex_unlock(&fs_info->cleaner_mutex);
1728
1729                 /*
1730                  * The defragger has dealt with the R/O remount and umount,
1731                  * needn't do anything special here.
1732                  */
1733                 btrfs_run_defrag_inodes(fs_info);
1734
1735                 /*
1736                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1737                  * with relocation (btrfs_relocate_chunk) and relocation
1738                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1739                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1740                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1741                  * unused block groups.
1742                  */
1743                 btrfs_delete_unused_bgs(fs_info);
1744 sleep:
1745                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1746                 if (kthread_should_park())
1747                         kthread_parkme();
1748                 if (kthread_should_stop())
1749                         return 0;
1750                 if (!again) {
1751                         set_current_state(TASK_INTERRUPTIBLE);
1752                         schedule();
1753                         __set_current_state(TASK_RUNNING);
1754                 }
1755         }
1756 }
1757
1758 static int transaction_kthread(void *arg)
1759 {
1760         struct btrfs_root *root = arg;
1761         struct btrfs_fs_info *fs_info = root->fs_info;
1762         struct btrfs_trans_handle *trans;
1763         struct btrfs_transaction *cur;
1764         u64 transid;
1765         time64_t now;
1766         unsigned long delay;
1767         bool cannot_commit;
1768
1769         do {
1770                 cannot_commit = false;
1771                 delay = HZ * fs_info->commit_interval;
1772                 mutex_lock(&fs_info->transaction_kthread_mutex);
1773
1774                 spin_lock(&fs_info->trans_lock);
1775                 cur = fs_info->running_transaction;
1776                 if (!cur) {
1777                         spin_unlock(&fs_info->trans_lock);
1778                         goto sleep;
1779                 }
1780
1781                 now = ktime_get_seconds();
1782                 if (cur->state < TRANS_STATE_BLOCKED &&
1783                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1784                     (now < cur->start_time ||
1785                      now - cur->start_time < fs_info->commit_interval)) {
1786                         spin_unlock(&fs_info->trans_lock);
1787                         delay = HZ * 5;
1788                         goto sleep;
1789                 }
1790                 transid = cur->transid;
1791                 spin_unlock(&fs_info->trans_lock);
1792
1793                 /* If the file system is aborted, this will always fail. */
1794                 trans = btrfs_attach_transaction(root);
1795                 if (IS_ERR(trans)) {
1796                         if (PTR_ERR(trans) != -ENOENT)
1797                                 cannot_commit = true;
1798                         goto sleep;
1799                 }
1800                 if (transid == trans->transid) {
1801                         btrfs_commit_transaction(trans);
1802                 } else {
1803                         btrfs_end_transaction(trans);
1804                 }
1805 sleep:
1806                 wake_up_process(fs_info->cleaner_kthread);
1807                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1808
1809                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1810                                       &fs_info->fs_state)))
1811                         btrfs_cleanup_transaction(fs_info);
1812                 if (!kthread_should_stop() &&
1813                                 (!btrfs_transaction_blocked(fs_info) ||
1814                                  cannot_commit))
1815                         schedule_timeout_interruptible(delay);
1816         } while (!kthread_should_stop());
1817         return 0;
1818 }
1819
1820 /*
1821  * this will find the highest generation in the array of
1822  * root backups.  The index of the highest array is returned,
1823  * or -1 if we can't find anything.
1824  *
1825  * We check to make sure the array is valid by comparing the
1826  * generation of the latest  root in the array with the generation
1827  * in the super block.  If they don't match we pitch it.
1828  */
1829 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1830 {
1831         u64 cur;
1832         int newest_index = -1;
1833         struct btrfs_root_backup *root_backup;
1834         int i;
1835
1836         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1837                 root_backup = info->super_copy->super_roots + i;
1838                 cur = btrfs_backup_tree_root_gen(root_backup);
1839                 if (cur == newest_gen)
1840                         newest_index = i;
1841         }
1842
1843         /* check to see if we actually wrapped around */
1844         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1845                 root_backup = info->super_copy->super_roots;
1846                 cur = btrfs_backup_tree_root_gen(root_backup);
1847                 if (cur == newest_gen)
1848                         newest_index = 0;
1849         }
1850         return newest_index;
1851 }
1852
1853
1854 /*
1855  * find the oldest backup so we know where to store new entries
1856  * in the backup array.  This will set the backup_root_index
1857  * field in the fs_info struct
1858  */
1859 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1860                                      u64 newest_gen)
1861 {
1862         int newest_index = -1;
1863
1864         newest_index = find_newest_super_backup(info, newest_gen);
1865         /* if there was garbage in there, just move along */
1866         if (newest_index == -1) {
1867                 info->backup_root_index = 0;
1868         } else {
1869                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1870         }
1871 }
1872
1873 /*
1874  * copy all the root pointers into the super backup array.
1875  * this will bump the backup pointer by one when it is
1876  * done
1877  */
1878 static void backup_super_roots(struct btrfs_fs_info *info)
1879 {
1880         int next_backup;
1881         struct btrfs_root_backup *root_backup;
1882         int last_backup;
1883
1884         next_backup = info->backup_root_index;
1885         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1886                 BTRFS_NUM_BACKUP_ROOTS;
1887
1888         /*
1889          * just overwrite the last backup if we're at the same generation
1890          * this happens only at umount
1891          */
1892         root_backup = info->super_for_commit->super_roots + last_backup;
1893         if (btrfs_backup_tree_root_gen(root_backup) ==
1894             btrfs_header_generation(info->tree_root->node))
1895                 next_backup = last_backup;
1896
1897         root_backup = info->super_for_commit->super_roots + next_backup;
1898
1899         /*
1900          * make sure all of our padding and empty slots get zero filled
1901          * regardless of which ones we use today
1902          */
1903         memset(root_backup, 0, sizeof(*root_backup));
1904
1905         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1906
1907         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1908         btrfs_set_backup_tree_root_gen(root_backup,
1909                                btrfs_header_generation(info->tree_root->node));
1910
1911         btrfs_set_backup_tree_root_level(root_backup,
1912                                btrfs_header_level(info->tree_root->node));
1913
1914         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1915         btrfs_set_backup_chunk_root_gen(root_backup,
1916                                btrfs_header_generation(info->chunk_root->node));
1917         btrfs_set_backup_chunk_root_level(root_backup,
1918                                btrfs_header_level(info->chunk_root->node));
1919
1920         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1921         btrfs_set_backup_extent_root_gen(root_backup,
1922                                btrfs_header_generation(info->extent_root->node));
1923         btrfs_set_backup_extent_root_level(root_backup,
1924                                btrfs_header_level(info->extent_root->node));
1925
1926         /*
1927          * we might commit during log recovery, which happens before we set
1928          * the fs_root.  Make sure it is valid before we fill it in.
1929          */
1930         if (info->fs_root && info->fs_root->node) {
1931                 btrfs_set_backup_fs_root(root_backup,
1932                                          info->fs_root->node->start);
1933                 btrfs_set_backup_fs_root_gen(root_backup,
1934                                btrfs_header_generation(info->fs_root->node));
1935                 btrfs_set_backup_fs_root_level(root_backup,
1936                                btrfs_header_level(info->fs_root->node));
1937         }
1938
1939         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1940         btrfs_set_backup_dev_root_gen(root_backup,
1941                                btrfs_header_generation(info->dev_root->node));
1942         btrfs_set_backup_dev_root_level(root_backup,
1943                                        btrfs_header_level(info->dev_root->node));
1944
1945         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1946         btrfs_set_backup_csum_root_gen(root_backup,
1947                                btrfs_header_generation(info->csum_root->node));
1948         btrfs_set_backup_csum_root_level(root_backup,
1949                                btrfs_header_level(info->csum_root->node));
1950
1951         btrfs_set_backup_total_bytes(root_backup,
1952                              btrfs_super_total_bytes(info->super_copy));
1953         btrfs_set_backup_bytes_used(root_backup,
1954                              btrfs_super_bytes_used(info->super_copy));
1955         btrfs_set_backup_num_devices(root_backup,
1956                              btrfs_super_num_devices(info->super_copy));
1957
1958         /*
1959          * if we don't copy this out to the super_copy, it won't get remembered
1960          * for the next commit
1961          */
1962         memcpy(&info->super_copy->super_roots,
1963                &info->super_for_commit->super_roots,
1964                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1965 }
1966
1967 /*
1968  * this copies info out of the root backup array and back into
1969  * the in-memory super block.  It is meant to help iterate through
1970  * the array, so you send it the number of backups you've already
1971  * tried and the last backup index you used.
1972  *
1973  * this returns -1 when it has tried all the backups
1974  */
1975 static noinline int next_root_backup(struct btrfs_fs_info *info,
1976                                      struct btrfs_super_block *super,
1977                                      int *num_backups_tried, int *backup_index)
1978 {
1979         struct btrfs_root_backup *root_backup;
1980         int newest = *backup_index;
1981
1982         if (*num_backups_tried == 0) {
1983                 u64 gen = btrfs_super_generation(super);
1984
1985                 newest = find_newest_super_backup(info, gen);
1986                 if (newest == -1)
1987                         return -1;
1988
1989                 *backup_index = newest;
1990                 *num_backups_tried = 1;
1991         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1992                 /* we've tried all the backups, all done */
1993                 return -1;
1994         } else {
1995                 /* jump to the next oldest backup */
1996                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1997                         BTRFS_NUM_BACKUP_ROOTS;
1998                 *backup_index = newest;
1999                 *num_backups_tried += 1;
2000         }
2001         root_backup = super->super_roots + newest;
2002
2003         btrfs_set_super_generation(super,
2004                                    btrfs_backup_tree_root_gen(root_backup));
2005         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2006         btrfs_set_super_root_level(super,
2007                                    btrfs_backup_tree_root_level(root_backup));
2008         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2009
2010         /*
2011          * fixme: the total bytes and num_devices need to match or we should
2012          * need a fsck
2013          */
2014         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2015         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2016         return 0;
2017 }
2018
2019 /* helper to cleanup workers */
2020 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2021 {
2022         btrfs_destroy_workqueue(fs_info->fixup_workers);
2023         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2024         btrfs_destroy_workqueue(fs_info->workers);
2025         btrfs_destroy_workqueue(fs_info->endio_workers);
2026         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2027         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2028         btrfs_destroy_workqueue(fs_info->rmw_workers);
2029         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2030         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2031         btrfs_destroy_workqueue(fs_info->submit_workers);
2032         btrfs_destroy_workqueue(fs_info->delayed_workers);
2033         btrfs_destroy_workqueue(fs_info->caching_workers);
2034         btrfs_destroy_workqueue(fs_info->readahead_workers);
2035         btrfs_destroy_workqueue(fs_info->flush_workers);
2036         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2037         btrfs_destroy_workqueue(fs_info->extent_workers);
2038         /*
2039          * Now that all other work queues are destroyed, we can safely destroy
2040          * the queues used for metadata I/O, since tasks from those other work
2041          * queues can do metadata I/O operations.
2042          */
2043         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2044         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2045 }
2046
2047 static void free_root_extent_buffers(struct btrfs_root *root)
2048 {
2049         if (root) {
2050                 free_extent_buffer(root->node);
2051                 free_extent_buffer(root->commit_root);
2052                 root->node = NULL;
2053                 root->commit_root = NULL;
2054         }
2055 }
2056
2057 /* helper to cleanup tree roots */
2058 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2059 {
2060         free_root_extent_buffers(info->tree_root);
2061
2062         free_root_extent_buffers(info->dev_root);
2063         free_root_extent_buffers(info->extent_root);
2064         free_root_extent_buffers(info->csum_root);
2065         free_root_extent_buffers(info->quota_root);
2066         free_root_extent_buffers(info->uuid_root);
2067         if (chunk_root)
2068                 free_root_extent_buffers(info->chunk_root);
2069         free_root_extent_buffers(info->free_space_root);
2070 }
2071
2072 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2073 {
2074         int ret;
2075         struct btrfs_root *gang[8];
2076         int i;
2077
2078         while (!list_empty(&fs_info->dead_roots)) {
2079                 gang[0] = list_entry(fs_info->dead_roots.next,
2080                                      struct btrfs_root, root_list);
2081                 list_del(&gang[0]->root_list);
2082
2083                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2084                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2085                 } else {
2086                         free_extent_buffer(gang[0]->node);
2087                         free_extent_buffer(gang[0]->commit_root);
2088                         btrfs_put_fs_root(gang[0]);
2089                 }
2090         }
2091
2092         while (1) {
2093                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2094                                              (void **)gang, 0,
2095                                              ARRAY_SIZE(gang));
2096                 if (!ret)
2097                         break;
2098                 for (i = 0; i < ret; i++)
2099                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2100         }
2101
2102         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2103                 btrfs_free_log_root_tree(NULL, fs_info);
2104                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2105         }
2106 }
2107
2108 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2109 {
2110         mutex_init(&fs_info->scrub_lock);
2111         atomic_set(&fs_info->scrubs_running, 0);
2112         atomic_set(&fs_info->scrub_pause_req, 0);
2113         atomic_set(&fs_info->scrubs_paused, 0);
2114         atomic_set(&fs_info->scrub_cancel_req, 0);
2115         init_waitqueue_head(&fs_info->scrub_pause_wait);
2116         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2117 }
2118
2119 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2120 {
2121         spin_lock_init(&fs_info->balance_lock);
2122         mutex_init(&fs_info->balance_mutex);
2123         atomic_set(&fs_info->balance_pause_req, 0);
2124         atomic_set(&fs_info->balance_cancel_req, 0);
2125         fs_info->balance_ctl = NULL;
2126         init_waitqueue_head(&fs_info->balance_wait_q);
2127 }
2128
2129 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2130 {
2131         struct inode *inode = fs_info->btree_inode;
2132
2133         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2134         set_nlink(inode, 1);
2135         /*
2136          * we set the i_size on the btree inode to the max possible int.
2137          * the real end of the address space is determined by all of
2138          * the devices in the system
2139          */
2140         inode->i_size = OFFSET_MAX;
2141         inode->i_mapping->a_ops = &btree_aops;
2142
2143         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2144         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2145                             IO_TREE_INODE_IO, inode);
2146         BTRFS_I(inode)->io_tree.track_uptodate = false;
2147         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2148
2149         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2150
2151         BTRFS_I(inode)->root = fs_info->tree_root;
2152         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2153         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2154         btrfs_insert_inode_hash(inode);
2155 }
2156
2157 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2158 {
2159         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2160         init_rwsem(&fs_info->dev_replace.rwsem);
2161         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2162 }
2163
2164 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2165 {
2166         spin_lock_init(&fs_info->qgroup_lock);
2167         mutex_init(&fs_info->qgroup_ioctl_lock);
2168         fs_info->qgroup_tree = RB_ROOT;
2169         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2170         fs_info->qgroup_seq = 1;
2171         fs_info->qgroup_ulist = NULL;
2172         fs_info->qgroup_rescan_running = false;
2173         mutex_init(&fs_info->qgroup_rescan_lock);
2174 }
2175
2176 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2177                 struct btrfs_fs_devices *fs_devices)
2178 {
2179         u32 max_active = fs_info->thread_pool_size;
2180         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2181
2182         fs_info->workers =
2183                 btrfs_alloc_workqueue(fs_info, "worker",
2184                                       flags | WQ_HIGHPRI, max_active, 16);
2185
2186         fs_info->delalloc_workers =
2187                 btrfs_alloc_workqueue(fs_info, "delalloc",
2188                                       flags, max_active, 2);
2189
2190         fs_info->flush_workers =
2191                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2192                                       flags, max_active, 0);
2193
2194         fs_info->caching_workers =
2195                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2196
2197         /*
2198          * a higher idle thresh on the submit workers makes it much more
2199          * likely that bios will be send down in a sane order to the
2200          * devices
2201          */
2202         fs_info->submit_workers =
2203                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2204                                       min_t(u64, fs_devices->num_devices,
2205                                             max_active), 64);
2206
2207         fs_info->fixup_workers =
2208                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2209
2210         /*
2211          * endios are largely parallel and should have a very
2212          * low idle thresh
2213          */
2214         fs_info->endio_workers =
2215                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2216         fs_info->endio_meta_workers =
2217                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2218                                       max_active, 4);
2219         fs_info->endio_meta_write_workers =
2220                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2221                                       max_active, 2);
2222         fs_info->endio_raid56_workers =
2223                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2224                                       max_active, 4);
2225         fs_info->endio_repair_workers =
2226                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2227         fs_info->rmw_workers =
2228                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2229         fs_info->endio_write_workers =
2230                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2231                                       max_active, 2);
2232         fs_info->endio_freespace_worker =
2233                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2234                                       max_active, 0);
2235         fs_info->delayed_workers =
2236                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2237                                       max_active, 0);
2238         fs_info->readahead_workers =
2239                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2240                                       max_active, 2);
2241         fs_info->qgroup_rescan_workers =
2242                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2243         fs_info->extent_workers =
2244                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2245                                       min_t(u64, fs_devices->num_devices,
2246                                             max_active), 8);
2247
2248         if (!(fs_info->workers && fs_info->delalloc_workers &&
2249               fs_info->submit_workers && fs_info->flush_workers &&
2250               fs_info->endio_workers && fs_info->endio_meta_workers &&
2251               fs_info->endio_meta_write_workers &&
2252               fs_info->endio_repair_workers &&
2253               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2254               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2255               fs_info->caching_workers && fs_info->readahead_workers &&
2256               fs_info->fixup_workers && fs_info->delayed_workers &&
2257               fs_info->extent_workers &&
2258               fs_info->qgroup_rescan_workers)) {
2259                 return -ENOMEM;
2260         }
2261
2262         return 0;
2263 }
2264
2265 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2266                             struct btrfs_fs_devices *fs_devices)
2267 {
2268         int ret;
2269         struct btrfs_root *log_tree_root;
2270         struct btrfs_super_block *disk_super = fs_info->super_copy;
2271         u64 bytenr = btrfs_super_log_root(disk_super);
2272         int level = btrfs_super_log_root_level(disk_super);
2273
2274         if (fs_devices->rw_devices == 0) {
2275                 btrfs_warn(fs_info, "log replay required on RO media");
2276                 return -EIO;
2277         }
2278
2279         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2280         if (!log_tree_root)
2281                 return -ENOMEM;
2282
2283         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2284
2285         log_tree_root->node = read_tree_block(fs_info, bytenr,
2286                                               fs_info->generation + 1,
2287                                               level, NULL);
2288         if (IS_ERR(log_tree_root->node)) {
2289                 btrfs_warn(fs_info, "failed to read log tree");
2290                 ret = PTR_ERR(log_tree_root->node);
2291                 kfree(log_tree_root);
2292                 return ret;
2293         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2294                 btrfs_err(fs_info, "failed to read log tree");
2295                 free_extent_buffer(log_tree_root->node);
2296                 kfree(log_tree_root);
2297                 return -EIO;
2298         }
2299         /* returns with log_tree_root freed on success */
2300         ret = btrfs_recover_log_trees(log_tree_root);
2301         if (ret) {
2302                 btrfs_handle_fs_error(fs_info, ret,
2303                                       "Failed to recover log tree");
2304                 free_extent_buffer(log_tree_root->node);
2305                 kfree(log_tree_root);
2306                 return ret;
2307         }
2308
2309         if (sb_rdonly(fs_info->sb)) {
2310                 ret = btrfs_commit_super(fs_info);
2311                 if (ret)
2312                         return ret;
2313         }
2314
2315         return 0;
2316 }
2317
2318 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2319 {
2320         struct btrfs_root *tree_root = fs_info->tree_root;
2321         struct btrfs_root *root;
2322         struct btrfs_key location;
2323         int ret;
2324
2325         BUG_ON(!fs_info->tree_root);
2326
2327         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2328         location.type = BTRFS_ROOT_ITEM_KEY;
2329         location.offset = 0;
2330
2331         root = btrfs_read_tree_root(tree_root, &location);
2332         if (IS_ERR(root)) {
2333                 ret = PTR_ERR(root);
2334                 goto out;
2335         }
2336         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2337         fs_info->extent_root = root;
2338
2339         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2340         root = btrfs_read_tree_root(tree_root, &location);
2341         if (IS_ERR(root)) {
2342                 ret = PTR_ERR(root);
2343                 goto out;
2344         }
2345         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2346         fs_info->dev_root = root;
2347         btrfs_init_devices_late(fs_info);
2348
2349         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2350         root = btrfs_read_tree_root(tree_root, &location);
2351         if (IS_ERR(root)) {
2352                 ret = PTR_ERR(root);
2353                 goto out;
2354         }
2355         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2356         fs_info->csum_root = root;
2357
2358         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2359         root = btrfs_read_tree_root(tree_root, &location);
2360         if (!IS_ERR(root)) {
2361                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2363                 fs_info->quota_root = root;
2364         }
2365
2366         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2367         root = btrfs_read_tree_root(tree_root, &location);
2368         if (IS_ERR(root)) {
2369                 ret = PTR_ERR(root);
2370                 if (ret != -ENOENT)
2371                         goto out;
2372         } else {
2373                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2374                 fs_info->uuid_root = root;
2375         }
2376
2377         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2378                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2379                 root = btrfs_read_tree_root(tree_root, &location);
2380                 if (IS_ERR(root)) {
2381                         ret = PTR_ERR(root);
2382                         goto out;
2383                 }
2384                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2385                 fs_info->free_space_root = root;
2386         }
2387
2388         return 0;
2389 out:
2390         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2391                    location.objectid, ret);
2392         return ret;
2393 }
2394
2395 /*
2396  * Real super block validation
2397  * NOTE: super csum type and incompat features will not be checked here.
2398  *
2399  * @sb:         super block to check
2400  * @mirror_num: the super block number to check its bytenr:
2401  *              0       the primary (1st) sb
2402  *              1, 2    2nd and 3rd backup copy
2403  *             -1       skip bytenr check
2404  */
2405 static int validate_super(struct btrfs_fs_info *fs_info,
2406                             struct btrfs_super_block *sb, int mirror_num)
2407 {
2408         u64 nodesize = btrfs_super_nodesize(sb);
2409         u64 sectorsize = btrfs_super_sectorsize(sb);
2410         int ret = 0;
2411
2412         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2413                 btrfs_err(fs_info, "no valid FS found");
2414                 ret = -EINVAL;
2415         }
2416         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2417                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2418                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2419                 ret = -EINVAL;
2420         }
2421         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2422                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2423                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2424                 ret = -EINVAL;
2425         }
2426         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2427                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2428                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2429                 ret = -EINVAL;
2430         }
2431         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2432                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2433                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2434                 ret = -EINVAL;
2435         }
2436
2437         /*
2438          * Check sectorsize and nodesize first, other check will need it.
2439          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2440          */
2441         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2442             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2443                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2444                 ret = -EINVAL;
2445         }
2446         /* Only PAGE SIZE is supported yet */
2447         if (sectorsize != PAGE_SIZE) {
2448                 btrfs_err(fs_info,
2449                         "sectorsize %llu not supported yet, only support %lu",
2450                         sectorsize, PAGE_SIZE);
2451                 ret = -EINVAL;
2452         }
2453         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2454             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2455                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2456                 ret = -EINVAL;
2457         }
2458         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2459                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2460                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2461                 ret = -EINVAL;
2462         }
2463
2464         /* Root alignment check */
2465         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2466                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2467                            btrfs_super_root(sb));
2468                 ret = -EINVAL;
2469         }
2470         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2471                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2472                            btrfs_super_chunk_root(sb));
2473                 ret = -EINVAL;
2474         }
2475         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2476                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2477                            btrfs_super_log_root(sb));
2478                 ret = -EINVAL;
2479         }
2480
2481         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2482                    BTRFS_FSID_SIZE) != 0) {
2483                 btrfs_err(fs_info,
2484                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2485                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2486                 ret = -EINVAL;
2487         }
2488
2489         /*
2490          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2491          * done later
2492          */
2493         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2494                 btrfs_err(fs_info, "bytes_used is too small %llu",
2495                           btrfs_super_bytes_used(sb));
2496                 ret = -EINVAL;
2497         }
2498         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2499                 btrfs_err(fs_info, "invalid stripesize %u",
2500                           btrfs_super_stripesize(sb));
2501                 ret = -EINVAL;
2502         }
2503         if (btrfs_super_num_devices(sb) > (1UL << 31))
2504                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2505                            btrfs_super_num_devices(sb));
2506         if (btrfs_super_num_devices(sb) == 0) {
2507                 btrfs_err(fs_info, "number of devices is 0");
2508                 ret = -EINVAL;
2509         }
2510
2511         if (mirror_num >= 0 &&
2512             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2513                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2514                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2515                 ret = -EINVAL;
2516         }
2517
2518         /*
2519          * Obvious sys_chunk_array corruptions, it must hold at least one key
2520          * and one chunk
2521          */
2522         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2523                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2524                           btrfs_super_sys_array_size(sb),
2525                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2526                 ret = -EINVAL;
2527         }
2528         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2529                         + sizeof(struct btrfs_chunk)) {
2530                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2531                           btrfs_super_sys_array_size(sb),
2532                           sizeof(struct btrfs_disk_key)
2533                           + sizeof(struct btrfs_chunk));
2534                 ret = -EINVAL;
2535         }
2536
2537         /*
2538          * The generation is a global counter, we'll trust it more than the others
2539          * but it's still possible that it's the one that's wrong.
2540          */
2541         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2542                 btrfs_warn(fs_info,
2543                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2544                         btrfs_super_generation(sb),
2545                         btrfs_super_chunk_root_generation(sb));
2546         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2547             && btrfs_super_cache_generation(sb) != (u64)-1)
2548                 btrfs_warn(fs_info,
2549                         "suspicious: generation < cache_generation: %llu < %llu",
2550                         btrfs_super_generation(sb),
2551                         btrfs_super_cache_generation(sb));
2552
2553         return ret;
2554 }
2555
2556 /*
2557  * Validation of super block at mount time.
2558  * Some checks already done early at mount time, like csum type and incompat
2559  * flags will be skipped.
2560  */
2561 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2562 {
2563         return validate_super(fs_info, fs_info->super_copy, 0);
2564 }
2565
2566 /*
2567  * Validation of super block at write time.
2568  * Some checks like bytenr check will be skipped as their values will be
2569  * overwritten soon.
2570  * Extra checks like csum type and incompat flags will be done here.
2571  */
2572 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2573                                       struct btrfs_super_block *sb)
2574 {
2575         int ret;
2576
2577         ret = validate_super(fs_info, sb, -1);
2578         if (ret < 0)
2579                 goto out;
2580         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2581                 ret = -EUCLEAN;
2582                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2583                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2584                 goto out;
2585         }
2586         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2587                 ret = -EUCLEAN;
2588                 btrfs_err(fs_info,
2589                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2590                           btrfs_super_incompat_flags(sb),
2591                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2592                 goto out;
2593         }
2594 out:
2595         if (ret < 0)
2596                 btrfs_err(fs_info,
2597                 "super block corruption detected before writing it to disk");
2598         return ret;
2599 }
2600
2601 int open_ctree(struct super_block *sb,
2602                struct btrfs_fs_devices *fs_devices,
2603                char *options)
2604 {
2605         u32 sectorsize;
2606         u32 nodesize;
2607         u32 stripesize;
2608         u64 generation;
2609         u64 features;
2610         struct btrfs_key location;
2611         struct buffer_head *bh;
2612         struct btrfs_super_block *disk_super;
2613         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2614         struct btrfs_root *tree_root;
2615         struct btrfs_root *chunk_root;
2616         int ret;
2617         int err = -EINVAL;
2618         int num_backups_tried = 0;
2619         int backup_index = 0;
2620         int clear_free_space_tree = 0;
2621         int level;
2622
2623         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2624         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2625         if (!tree_root || !chunk_root) {
2626                 err = -ENOMEM;
2627                 goto fail;
2628         }
2629
2630         ret = init_srcu_struct(&fs_info->subvol_srcu);
2631         if (ret) {
2632                 err = ret;
2633                 goto fail;
2634         }
2635
2636         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2637         if (ret) {
2638                 err = ret;
2639                 goto fail_srcu;
2640         }
2641
2642         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2643         if (ret) {
2644                 err = ret;
2645                 goto fail_dio_bytes;
2646         }
2647         fs_info->dirty_metadata_batch = PAGE_SIZE *
2648                                         (1 + ilog2(nr_cpu_ids));
2649
2650         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2651         if (ret) {
2652                 err = ret;
2653                 goto fail_dirty_metadata_bytes;
2654         }
2655
2656         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2657                         GFP_KERNEL);
2658         if (ret) {
2659                 err = ret;
2660                 goto fail_delalloc_bytes;
2661         }
2662
2663         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2664         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2665         INIT_LIST_HEAD(&fs_info->trans_list);
2666         INIT_LIST_HEAD(&fs_info->dead_roots);
2667         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2668         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2669         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2670         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2671         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2672         spin_lock_init(&fs_info->delalloc_root_lock);
2673         spin_lock_init(&fs_info->trans_lock);
2674         spin_lock_init(&fs_info->fs_roots_radix_lock);
2675         spin_lock_init(&fs_info->delayed_iput_lock);
2676         spin_lock_init(&fs_info->defrag_inodes_lock);
2677         spin_lock_init(&fs_info->tree_mod_seq_lock);
2678         spin_lock_init(&fs_info->super_lock);
2679         spin_lock_init(&fs_info->buffer_lock);
2680         spin_lock_init(&fs_info->unused_bgs_lock);
2681         rwlock_init(&fs_info->tree_mod_log_lock);
2682         mutex_init(&fs_info->unused_bg_unpin_mutex);
2683         mutex_init(&fs_info->delete_unused_bgs_mutex);
2684         mutex_init(&fs_info->reloc_mutex);
2685         mutex_init(&fs_info->delalloc_root_mutex);
2686         seqlock_init(&fs_info->profiles_lock);
2687
2688         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2689         INIT_LIST_HEAD(&fs_info->space_info);
2690         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2691         INIT_LIST_HEAD(&fs_info->unused_bgs);
2692         btrfs_mapping_init(&fs_info->mapping_tree);
2693         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2694                              BTRFS_BLOCK_RSV_GLOBAL);
2695         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2696         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2697         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2698         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2699                              BTRFS_BLOCK_RSV_DELOPS);
2700         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2701                              BTRFS_BLOCK_RSV_DELREFS);
2702
2703         atomic_set(&fs_info->async_delalloc_pages, 0);
2704         atomic_set(&fs_info->defrag_running, 0);
2705         atomic_set(&fs_info->reada_works_cnt, 0);
2706         atomic_set(&fs_info->nr_delayed_iputs, 0);
2707         atomic64_set(&fs_info->tree_mod_seq, 0);
2708         fs_info->sb = sb;
2709         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2710         fs_info->metadata_ratio = 0;
2711         fs_info->defrag_inodes = RB_ROOT;
2712         atomic64_set(&fs_info->free_chunk_space, 0);
2713         fs_info->tree_mod_log = RB_ROOT;
2714         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2715         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2716         /* readahead state */
2717         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2718         spin_lock_init(&fs_info->reada_lock);
2719         btrfs_init_ref_verify(fs_info);
2720
2721         fs_info->thread_pool_size = min_t(unsigned long,
2722                                           num_online_cpus() + 2, 8);
2723
2724         INIT_LIST_HEAD(&fs_info->ordered_roots);
2725         spin_lock_init(&fs_info->ordered_root_lock);
2726
2727         fs_info->btree_inode = new_inode(sb);
2728         if (!fs_info->btree_inode) {
2729                 err = -ENOMEM;
2730                 goto fail_bio_counter;
2731         }
2732         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2733
2734         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2735                                         GFP_KERNEL);
2736         if (!fs_info->delayed_root) {
2737                 err = -ENOMEM;
2738                 goto fail_iput;
2739         }
2740         btrfs_init_delayed_root(fs_info->delayed_root);
2741
2742         btrfs_init_scrub(fs_info);
2743 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2744         fs_info->check_integrity_print_mask = 0;
2745 #endif
2746         btrfs_init_balance(fs_info);
2747         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2748
2749         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2750         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2751
2752         btrfs_init_btree_inode(fs_info);
2753
2754         spin_lock_init(&fs_info->block_group_cache_lock);
2755         fs_info->block_group_cache_tree = RB_ROOT;
2756         fs_info->first_logical_byte = (u64)-1;
2757
2758         extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2759                             IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2760         extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2761                             IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2762         fs_info->pinned_extents = &fs_info->freed_extents[0];
2763         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2764
2765         mutex_init(&fs_info->ordered_operations_mutex);
2766         mutex_init(&fs_info->tree_log_mutex);
2767         mutex_init(&fs_info->chunk_mutex);
2768         mutex_init(&fs_info->transaction_kthread_mutex);
2769         mutex_init(&fs_info->cleaner_mutex);
2770         mutex_init(&fs_info->ro_block_group_mutex);
2771         init_rwsem(&fs_info->commit_root_sem);
2772         init_rwsem(&fs_info->cleanup_work_sem);
2773         init_rwsem(&fs_info->subvol_sem);
2774         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2775
2776         btrfs_init_dev_replace_locks(fs_info);
2777         btrfs_init_qgroup(fs_info);
2778
2779         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2780         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2781
2782         init_waitqueue_head(&fs_info->transaction_throttle);
2783         init_waitqueue_head(&fs_info->transaction_wait);
2784         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2785         init_waitqueue_head(&fs_info->async_submit_wait);
2786         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2787
2788         /* Usable values until the real ones are cached from the superblock */
2789         fs_info->nodesize = 4096;
2790         fs_info->sectorsize = 4096;
2791         fs_info->stripesize = 4096;
2792
2793         spin_lock_init(&fs_info->swapfile_pins_lock);
2794         fs_info->swapfile_pins = RB_ROOT;
2795
2796         ret = btrfs_alloc_stripe_hash_table(fs_info);
2797         if (ret) {
2798                 err = ret;
2799                 goto fail_alloc;
2800         }
2801
2802         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2803
2804         invalidate_bdev(fs_devices->latest_bdev);
2805
2806         /*
2807          * Read super block and check the signature bytes only
2808          */
2809         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2810         if (IS_ERR(bh)) {
2811                 err = PTR_ERR(bh);
2812                 goto fail_alloc;
2813         }
2814
2815         /*
2816          * We want to check superblock checksum, the type is stored inside.
2817          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2818          */
2819         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2820                 btrfs_err(fs_info, "superblock checksum mismatch");
2821                 err = -EINVAL;
2822                 brelse(bh);
2823                 goto fail_alloc;
2824         }
2825
2826         /*
2827          * super_copy is zeroed at allocation time and we never touch the
2828          * following bytes up to INFO_SIZE, the checksum is calculated from
2829          * the whole block of INFO_SIZE
2830          */
2831         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2832         brelse(bh);
2833
2834         disk_super = fs_info->super_copy;
2835
2836         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2837                        BTRFS_FSID_SIZE));
2838
2839         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2840                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2841                                 fs_info->super_copy->metadata_uuid,
2842                                 BTRFS_FSID_SIZE));
2843         }
2844
2845         features = btrfs_super_flags(disk_super);
2846         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2847                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2848                 btrfs_set_super_flags(disk_super, features);
2849                 btrfs_info(fs_info,
2850                         "found metadata UUID change in progress flag, clearing");
2851         }
2852
2853         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2854                sizeof(*fs_info->super_for_commit));
2855
2856         ret = btrfs_validate_mount_super(fs_info);
2857         if (ret) {
2858                 btrfs_err(fs_info, "superblock contains fatal errors");
2859                 err = -EINVAL;
2860                 goto fail_alloc;
2861         }
2862
2863         if (!btrfs_super_root(disk_super))
2864                 goto fail_alloc;
2865
2866         /* check FS state, whether FS is broken. */
2867         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2868                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2869
2870         /*
2871          * run through our array of backup supers and setup
2872          * our ring pointer to the oldest one
2873          */
2874         generation = btrfs_super_generation(disk_super);
2875         find_oldest_super_backup(fs_info, generation);
2876
2877         /*
2878          * In the long term, we'll store the compression type in the super
2879          * block, and it'll be used for per file compression control.
2880          */
2881         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2882
2883         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2884         if (ret) {
2885                 err = ret;
2886                 goto fail_alloc;
2887         }
2888
2889         features = btrfs_super_incompat_flags(disk_super) &
2890                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2891         if (features) {
2892                 btrfs_err(fs_info,
2893                     "cannot mount because of unsupported optional features (%llx)",
2894                     features);
2895                 err = -EINVAL;
2896                 goto fail_alloc;
2897         }
2898
2899         features = btrfs_super_incompat_flags(disk_super);
2900         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2901         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2902                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2903         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2904                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2905
2906         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2907                 btrfs_info(fs_info, "has skinny extents");
2908
2909         /*
2910          * flag our filesystem as having big metadata blocks if
2911          * they are bigger than the page size
2912          */
2913         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2914                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2915                         btrfs_info(fs_info,
2916                                 "flagging fs with big metadata feature");
2917                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2918         }
2919
2920         nodesize = btrfs_super_nodesize(disk_super);
2921         sectorsize = btrfs_super_sectorsize(disk_super);
2922         stripesize = sectorsize;
2923         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2924         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2925
2926         /* Cache block sizes */
2927         fs_info->nodesize = nodesize;
2928         fs_info->sectorsize = sectorsize;
2929         fs_info->stripesize = stripesize;
2930
2931         /*
2932          * mixed block groups end up with duplicate but slightly offset
2933          * extent buffers for the same range.  It leads to corruptions
2934          */
2935         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2936             (sectorsize != nodesize)) {
2937                 btrfs_err(fs_info,
2938 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2939                         nodesize, sectorsize);
2940                 goto fail_alloc;
2941         }
2942
2943         /*
2944          * Needn't use the lock because there is no other task which will
2945          * update the flag.
2946          */
2947         btrfs_set_super_incompat_flags(disk_super, features);
2948
2949         features = btrfs_super_compat_ro_flags(disk_super) &
2950                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2951         if (!sb_rdonly(sb) && features) {
2952                 btrfs_err(fs_info,
2953         "cannot mount read-write because of unsupported optional features (%llx)",
2954                        features);
2955                 err = -EINVAL;
2956                 goto fail_alloc;
2957         }
2958
2959         ret = btrfs_init_workqueues(fs_info, fs_devices);
2960         if (ret) {
2961                 err = ret;
2962                 goto fail_sb_buffer;
2963         }
2964
2965         sb->s_bdi->congested_fn = btrfs_congested_fn;
2966         sb->s_bdi->congested_data = fs_info;
2967         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2968         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2969         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2970         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2971
2972         sb->s_blocksize = sectorsize;
2973         sb->s_blocksize_bits = blksize_bits(sectorsize);
2974         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2975
2976         mutex_lock(&fs_info->chunk_mutex);
2977         ret = btrfs_read_sys_array(fs_info);
2978         mutex_unlock(&fs_info->chunk_mutex);
2979         if (ret) {
2980                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2981                 goto fail_sb_buffer;
2982         }
2983
2984         generation = btrfs_super_chunk_root_generation(disk_super);
2985         level = btrfs_super_chunk_root_level(disk_super);
2986
2987         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2988
2989         chunk_root->node = read_tree_block(fs_info,
2990                                            btrfs_super_chunk_root(disk_super),
2991                                            generation, level, NULL);
2992         if (IS_ERR(chunk_root->node) ||
2993             !extent_buffer_uptodate(chunk_root->node)) {
2994                 btrfs_err(fs_info, "failed to read chunk root");
2995                 if (!IS_ERR(chunk_root->node))
2996                         free_extent_buffer(chunk_root->node);
2997                 chunk_root->node = NULL;
2998                 goto fail_tree_roots;
2999         }
3000         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3001         chunk_root->commit_root = btrfs_root_node(chunk_root);
3002
3003         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3004            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3005
3006         ret = btrfs_read_chunk_tree(fs_info);
3007         if (ret) {
3008                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3009                 goto fail_tree_roots;
3010         }
3011
3012         /*
3013          * Keep the devid that is marked to be the target device for the
3014          * device replace procedure
3015          */
3016         btrfs_free_extra_devids(fs_devices, 0);
3017
3018         if (!fs_devices->latest_bdev) {
3019                 btrfs_err(fs_info, "failed to read devices");
3020                 goto fail_tree_roots;
3021         }
3022
3023 retry_root_backup:
3024         generation = btrfs_super_generation(disk_super);
3025         level = btrfs_super_root_level(disk_super);
3026
3027         tree_root->node = read_tree_block(fs_info,
3028                                           btrfs_super_root(disk_super),
3029                                           generation, level, NULL);
3030         if (IS_ERR(tree_root->node) ||
3031             !extent_buffer_uptodate(tree_root->node)) {
3032                 btrfs_warn(fs_info, "failed to read tree root");
3033                 if (!IS_ERR(tree_root->node))
3034                         free_extent_buffer(tree_root->node);
3035                 tree_root->node = NULL;
3036                 goto recovery_tree_root;
3037         }
3038
3039         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3040         tree_root->commit_root = btrfs_root_node(tree_root);
3041         btrfs_set_root_refs(&tree_root->root_item, 1);
3042
3043         mutex_lock(&tree_root->objectid_mutex);
3044         ret = btrfs_find_highest_objectid(tree_root,
3045                                         &tree_root->highest_objectid);
3046         if (ret) {
3047                 mutex_unlock(&tree_root->objectid_mutex);
3048                 goto recovery_tree_root;
3049         }
3050
3051         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3052
3053         mutex_unlock(&tree_root->objectid_mutex);
3054
3055         ret = btrfs_read_roots(fs_info);
3056         if (ret)
3057                 goto recovery_tree_root;
3058
3059         fs_info->generation = generation;
3060         fs_info->last_trans_committed = generation;
3061
3062         ret = btrfs_verify_dev_extents(fs_info);
3063         if (ret) {
3064                 btrfs_err(fs_info,
3065                           "failed to verify dev extents against chunks: %d",
3066                           ret);
3067                 goto fail_block_groups;
3068         }
3069         ret = btrfs_recover_balance(fs_info);
3070         if (ret) {
3071                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3072                 goto fail_block_groups;
3073         }
3074
3075         ret = btrfs_init_dev_stats(fs_info);
3076         if (ret) {
3077                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3078                 goto fail_block_groups;
3079         }
3080
3081         ret = btrfs_init_dev_replace(fs_info);
3082         if (ret) {
3083                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3084                 goto fail_block_groups;
3085         }
3086
3087         btrfs_free_extra_devids(fs_devices, 1);
3088
3089         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3090         if (ret) {
3091                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3092                                 ret);
3093                 goto fail_block_groups;
3094         }
3095
3096         ret = btrfs_sysfs_add_device(fs_devices);
3097         if (ret) {
3098                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3099                                 ret);
3100                 goto fail_fsdev_sysfs;
3101         }
3102
3103         ret = btrfs_sysfs_add_mounted(fs_info);
3104         if (ret) {
3105                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3106                 goto fail_fsdev_sysfs;
3107         }
3108
3109         ret = btrfs_init_space_info(fs_info);
3110         if (ret) {
3111                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3112                 goto fail_sysfs;
3113         }
3114
3115         ret = btrfs_read_block_groups(fs_info);
3116         if (ret) {
3117                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3118                 goto fail_sysfs;
3119         }
3120
3121         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3122                 btrfs_warn(fs_info,
3123                 "writable mount is not allowed due to too many missing devices");
3124                 goto fail_sysfs;
3125         }
3126
3127         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3128                                                "btrfs-cleaner");
3129         if (IS_ERR(fs_info->cleaner_kthread))
3130                 goto fail_sysfs;
3131
3132         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3133                                                    tree_root,
3134                                                    "btrfs-transaction");
3135         if (IS_ERR(fs_info->transaction_kthread))
3136                 goto fail_cleaner;
3137
3138         if (!btrfs_test_opt(fs_info, NOSSD) &&
3139             !fs_info->fs_devices->rotating) {
3140                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3141         }
3142
3143         /*
3144          * Mount does not set all options immediately, we can do it now and do
3145          * not have to wait for transaction commit
3146          */
3147         btrfs_apply_pending_changes(fs_info);
3148
3149 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3150         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3151                 ret = btrfsic_mount(fs_info, fs_devices,
3152                                     btrfs_test_opt(fs_info,
3153                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3154                                     1 : 0,
3155                                     fs_info->check_integrity_print_mask);
3156                 if (ret)
3157                         btrfs_warn(fs_info,
3158                                 "failed to initialize integrity check module: %d",
3159                                 ret);
3160         }
3161 #endif
3162         ret = btrfs_read_qgroup_config(fs_info);
3163         if (ret)
3164                 goto fail_trans_kthread;
3165
3166         if (btrfs_build_ref_tree(fs_info))
3167                 btrfs_err(fs_info, "couldn't build ref tree");
3168
3169         /* do not make disk changes in broken FS or nologreplay is given */
3170         if (btrfs_super_log_root(disk_super) != 0 &&
3171             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3172                 ret = btrfs_replay_log(fs_info, fs_devices);
3173                 if (ret) {
3174                         err = ret;
3175                         goto fail_qgroup;
3176                 }
3177         }
3178
3179         ret = btrfs_find_orphan_roots(fs_info);
3180         if (ret)
3181                 goto fail_qgroup;
3182
3183         if (!sb_rdonly(sb)) {
3184                 ret = btrfs_cleanup_fs_roots(fs_info);
3185                 if (ret)
3186                         goto fail_qgroup;
3187
3188                 mutex_lock(&fs_info->cleaner_mutex);
3189                 ret = btrfs_recover_relocation(tree_root);
3190                 mutex_unlock(&fs_info->cleaner_mutex);
3191                 if (ret < 0) {
3192                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3193                                         ret);
3194                         err = -EINVAL;
3195                         goto fail_qgroup;
3196                 }
3197         }
3198
3199         location.objectid = BTRFS_FS_TREE_OBJECTID;
3200         location.type = BTRFS_ROOT_ITEM_KEY;
3201         location.offset = 0;
3202
3203         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3204         if (IS_ERR(fs_info->fs_root)) {
3205                 err = PTR_ERR(fs_info->fs_root);
3206                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3207                 goto fail_qgroup;
3208         }
3209
3210         if (sb_rdonly(sb))
3211                 return 0;
3212
3213         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3214             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3215                 clear_free_space_tree = 1;
3216         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3217                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3218                 btrfs_warn(fs_info, "free space tree is invalid");
3219                 clear_free_space_tree = 1;
3220         }
3221
3222         if (clear_free_space_tree) {
3223                 btrfs_info(fs_info, "clearing free space tree");
3224                 ret = btrfs_clear_free_space_tree(fs_info);
3225                 if (ret) {
3226                         btrfs_warn(fs_info,
3227                                    "failed to clear free space tree: %d", ret);
3228                         close_ctree(fs_info);
3229                         return ret;
3230                 }
3231         }
3232
3233         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3234             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3235                 btrfs_info(fs_info, "creating free space tree");
3236                 ret = btrfs_create_free_space_tree(fs_info);
3237                 if (ret) {
3238                         btrfs_warn(fs_info,
3239                                 "failed to create free space tree: %d", ret);
3240                         close_ctree(fs_info);
3241                         return ret;
3242                 }
3243         }
3244
3245         down_read(&fs_info->cleanup_work_sem);
3246         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3247             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3248                 up_read(&fs_info->cleanup_work_sem);
3249                 close_ctree(fs_info);
3250                 return ret;
3251         }
3252         up_read(&fs_info->cleanup_work_sem);
3253
3254         ret = btrfs_resume_balance_async(fs_info);
3255         if (ret) {
3256                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3257                 close_ctree(fs_info);
3258                 return ret;
3259         }
3260
3261         ret = btrfs_resume_dev_replace_async(fs_info);
3262         if (ret) {
3263                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3264                 close_ctree(fs_info);
3265                 return ret;
3266         }
3267
3268         btrfs_qgroup_rescan_resume(fs_info);
3269
3270         if (!fs_info->uuid_root) {
3271                 btrfs_info(fs_info, "creating UUID tree");
3272                 ret = btrfs_create_uuid_tree(fs_info);
3273                 if (ret) {
3274                         btrfs_warn(fs_info,
3275                                 "failed to create the UUID tree: %d", ret);
3276                         close_ctree(fs_info);
3277                         return ret;
3278                 }
3279         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3280                    fs_info->generation !=
3281                                 btrfs_super_uuid_tree_generation(disk_super)) {
3282                 btrfs_info(fs_info, "checking UUID tree");
3283                 ret = btrfs_check_uuid_tree(fs_info);
3284                 if (ret) {
3285                         btrfs_warn(fs_info,
3286                                 "failed to check the UUID tree: %d", ret);
3287                         close_ctree(fs_info);
3288                         return ret;
3289                 }
3290         } else {
3291                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3292         }
3293         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3294
3295         /*
3296          * backuproot only affect mount behavior, and if open_ctree succeeded,
3297          * no need to keep the flag
3298          */
3299         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3300
3301         return 0;
3302
3303 fail_qgroup:
3304         btrfs_free_qgroup_config(fs_info);
3305 fail_trans_kthread:
3306         kthread_stop(fs_info->transaction_kthread);
3307         btrfs_cleanup_transaction(fs_info);
3308         btrfs_free_fs_roots(fs_info);
3309 fail_cleaner:
3310         kthread_stop(fs_info->cleaner_kthread);
3311
3312         /*
3313          * make sure we're done with the btree inode before we stop our
3314          * kthreads
3315          */
3316         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3317
3318 fail_sysfs:
3319         btrfs_sysfs_remove_mounted(fs_info);
3320
3321 fail_fsdev_sysfs:
3322         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3323
3324 fail_block_groups:
3325         btrfs_put_block_group_cache(fs_info);
3326
3327 fail_tree_roots:
3328         free_root_pointers(fs_info, 1);
3329         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3330
3331 fail_sb_buffer:
3332         btrfs_stop_all_workers(fs_info);
3333         btrfs_free_block_groups(fs_info);
3334 fail_alloc:
3335 fail_iput:
3336         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3337
3338         iput(fs_info->btree_inode);
3339 fail_bio_counter:
3340         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3341 fail_delalloc_bytes:
3342         percpu_counter_destroy(&fs_info->delalloc_bytes);
3343 fail_dirty_metadata_bytes:
3344         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3345 fail_dio_bytes:
3346         percpu_counter_destroy(&fs_info->dio_bytes);
3347 fail_srcu:
3348         cleanup_srcu_struct(&fs_info->subvol_srcu);
3349 fail:
3350         btrfs_free_stripe_hash_table(fs_info);
3351         btrfs_close_devices(fs_info->fs_devices);
3352         return err;
3353
3354 recovery_tree_root:
3355         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3356                 goto fail_tree_roots;
3357
3358         free_root_pointers(fs_info, 0);
3359
3360         /* don't use the log in recovery mode, it won't be valid */
3361         btrfs_set_super_log_root(disk_super, 0);
3362
3363         /* we can't trust the free space cache either */
3364         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3365
3366         ret = next_root_backup(fs_info, fs_info->super_copy,
3367                                &num_backups_tried, &backup_index);
3368         if (ret == -1)
3369                 goto fail_block_groups;
3370         goto retry_root_backup;
3371 }
3372 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3373
3374 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3375 {
3376         if (uptodate) {
3377                 set_buffer_uptodate(bh);
3378         } else {
3379                 struct btrfs_device *device = (struct btrfs_device *)
3380                         bh->b_private;
3381
3382                 btrfs_warn_rl_in_rcu(device->fs_info,
3383                                 "lost page write due to IO error on %s",
3384                                           rcu_str_deref(device->name));
3385                 /* note, we don't set_buffer_write_io_error because we have
3386                  * our own ways of dealing with the IO errors
3387                  */
3388                 clear_buffer_uptodate(bh);
3389                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3390         }
3391         unlock_buffer(bh);
3392         put_bh(bh);
3393 }
3394
3395 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3396                         struct buffer_head **bh_ret)
3397 {
3398         struct buffer_head *bh;
3399         struct btrfs_super_block *super;
3400         u64 bytenr;
3401
3402         bytenr = btrfs_sb_offset(copy_num);
3403         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3404                 return -EINVAL;
3405
3406         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3407         /*
3408          * If we fail to read from the underlying devices, as of now
3409          * the best option we have is to mark it EIO.
3410          */
3411         if (!bh)
3412                 return -EIO;
3413
3414         super = (struct btrfs_super_block *)bh->b_data;
3415         if (btrfs_super_bytenr(super) != bytenr ||
3416                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3417                 brelse(bh);
3418                 return -EINVAL;
3419         }
3420
3421         *bh_ret = bh;
3422         return 0;
3423 }
3424
3425
3426 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3427 {
3428         struct buffer_head *bh;
3429         struct buffer_head *latest = NULL;
3430         struct btrfs_super_block *super;
3431         int i;
3432         u64 transid = 0;
3433         int ret = -EINVAL;
3434
3435         /* we would like to check all the supers, but that would make
3436          * a btrfs mount succeed after a mkfs from a different FS.
3437          * So, we need to add a special mount option to scan for
3438          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3439          */
3440         for (i = 0; i < 1; i++) {
3441                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3442                 if (ret)
3443                         continue;
3444
3445                 super = (struct btrfs_super_block *)bh->b_data;
3446
3447                 if (!latest || btrfs_super_generation(super) > transid) {
3448                         brelse(latest);
3449                         latest = bh;
3450                         transid = btrfs_super_generation(super);
3451                 } else {
3452                         brelse(bh);
3453                 }
3454         }
3455
3456         if (!latest)
3457                 return ERR_PTR(ret);
3458
3459         return latest;
3460 }
3461
3462 /*
3463  * Write superblock @sb to the @device. Do not wait for completion, all the
3464  * buffer heads we write are pinned.
3465  *
3466  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3467  * the expected device size at commit time. Note that max_mirrors must be
3468  * same for write and wait phases.
3469  *
3470  * Return number of errors when buffer head is not found or submission fails.
3471  */
3472 static int write_dev_supers(struct btrfs_device *device,
3473                             struct btrfs_super_block *sb, int max_mirrors)
3474 {
3475         struct buffer_head *bh;
3476         int i;
3477         int ret;
3478         int errors = 0;
3479         u32 crc;
3480         u64 bytenr;
3481         int op_flags;
3482
3483         if (max_mirrors == 0)
3484                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3485
3486         for (i = 0; i < max_mirrors; i++) {
3487                 bytenr = btrfs_sb_offset(i);
3488                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3489                     device->commit_total_bytes)
3490                         break;
3491
3492                 btrfs_set_super_bytenr(sb, bytenr);
3493
3494                 crc = ~(u32)0;
3495                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3496                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3497                 btrfs_csum_final(crc, sb->csum);
3498
3499                 /* One reference for us, and we leave it for the caller */
3500                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3501                               BTRFS_SUPER_INFO_SIZE);
3502                 if (!bh) {
3503                         btrfs_err(device->fs_info,
3504                             "couldn't get super buffer head for bytenr %llu",
3505                             bytenr);
3506                         errors++;
3507                         continue;
3508                 }
3509
3510                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3511
3512                 /* one reference for submit_bh */
3513                 get_bh(bh);
3514
3515                 set_buffer_uptodate(bh);
3516                 lock_buffer(bh);
3517                 bh->b_end_io = btrfs_end_buffer_write_sync;
3518                 bh->b_private = device;
3519
3520                 /*
3521                  * we fua the first super.  The others we allow
3522                  * to go down lazy.
3523                  */
3524                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3525                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3526                         op_flags |= REQ_FUA;
3527                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3528                 if (ret)
3529                         errors++;
3530         }
3531         return errors < i ? 0 : -1;
3532 }
3533
3534 /*
3535  * Wait for write completion of superblocks done by write_dev_supers,
3536  * @max_mirrors same for write and wait phases.
3537  *
3538  * Return number of errors when buffer head is not found or not marked up to
3539  * date.
3540  */
3541 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3542 {
3543         struct buffer_head *bh;
3544         int i;
3545         int errors = 0;
3546         bool primary_failed = false;
3547         u64 bytenr;
3548
3549         if (max_mirrors == 0)
3550                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3551
3552         for (i = 0; i < max_mirrors; i++) {
3553                 bytenr = btrfs_sb_offset(i);
3554                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3555                     device->commit_total_bytes)
3556                         break;
3557
3558                 bh = __find_get_block(device->bdev,
3559                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3560                                       BTRFS_SUPER_INFO_SIZE);
3561                 if (!bh) {
3562                         errors++;
3563                         if (i == 0)
3564                                 primary_failed = true;
3565                         continue;
3566                 }
3567                 wait_on_buffer(bh);
3568                 if (!buffer_uptodate(bh)) {
3569                         errors++;
3570                         if (i == 0)
3571                                 primary_failed = true;
3572                 }
3573
3574                 /* drop our reference */
3575                 brelse(bh);
3576
3577                 /* drop the reference from the writing run */
3578                 brelse(bh);
3579         }
3580
3581         /* log error, force error return */
3582         if (primary_failed) {
3583                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3584                           device->devid);
3585                 return -1;
3586         }
3587
3588         return errors < i ? 0 : -1;
3589 }
3590
3591 /*
3592  * endio for the write_dev_flush, this will wake anyone waiting
3593  * for the barrier when it is done
3594  */
3595 static void btrfs_end_empty_barrier(struct bio *bio)
3596 {
3597         complete(bio->bi_private);
3598 }
3599
3600 /*
3601  * Submit a flush request to the device if it supports it. Error handling is
3602  * done in the waiting counterpart.
3603  */
3604 static void write_dev_flush(struct btrfs_device *device)
3605 {
3606         struct request_queue *q = bdev_get_queue(device->bdev);
3607         struct bio *bio = device->flush_bio;
3608
3609         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3610                 return;
3611
3612         bio_reset(bio);
3613         bio->bi_end_io = btrfs_end_empty_barrier;
3614         bio_set_dev(bio, device->bdev);
3615         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3616         init_completion(&device->flush_wait);
3617         bio->bi_private = &device->flush_wait;
3618
3619         btrfsic_submit_bio(bio);
3620         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3621 }
3622
3623 /*
3624  * If the flush bio has been submitted by write_dev_flush, wait for it.
3625  */
3626 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3627 {
3628         struct bio *bio = device->flush_bio;
3629
3630         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3631                 return BLK_STS_OK;
3632
3633         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3634         wait_for_completion_io(&device->flush_wait);
3635
3636         return bio->bi_status;
3637 }
3638
3639 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3640 {
3641         if (!btrfs_check_rw_degradable(fs_info, NULL))
3642                 return -EIO;
3643         return 0;
3644 }
3645
3646 /*
3647  * send an empty flush down to each device in parallel,
3648  * then wait for them
3649  */
3650 static int barrier_all_devices(struct btrfs_fs_info *info)
3651 {
3652         struct list_head *head;
3653         struct btrfs_device *dev;
3654         int errors_wait = 0;
3655         blk_status_t ret;
3656
3657         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3658         /* send down all the barriers */
3659         head = &info->fs_devices->devices;
3660         list_for_each_entry(dev, head, dev_list) {
3661                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3662                         continue;
3663                 if (!dev->bdev)
3664                         continue;
3665                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3666                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3667                         continue;
3668
3669                 write_dev_flush(dev);
3670                 dev->last_flush_error = BLK_STS_OK;
3671         }
3672
3673         /* wait for all the barriers */
3674         list_for_each_entry(dev, head, dev_list) {
3675                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3676                         continue;
3677                 if (!dev->bdev) {
3678                         errors_wait++;
3679                         continue;
3680                 }
3681                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3682                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3683                         continue;
3684
3685                 ret = wait_dev_flush(dev);
3686                 if (ret) {
3687                         dev->last_flush_error = ret;
3688                         btrfs_dev_stat_inc_and_print(dev,
3689                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3690                         errors_wait++;
3691                 }
3692         }
3693
3694         if (errors_wait) {
3695                 /*
3696                  * At some point we need the status of all disks
3697                  * to arrive at the volume status. So error checking
3698                  * is being pushed to a separate loop.
3699                  */
3700                 return check_barrier_error(info);
3701         }
3702         return 0;
3703 }
3704
3705 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3706 {
3707         int raid_type;
3708         int min_tolerated = INT_MAX;
3709
3710         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3711             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3712                 min_tolerated = min(min_tolerated,
3713                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3714                                     tolerated_failures);
3715
3716         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3717                 if (raid_type == BTRFS_RAID_SINGLE)
3718                         continue;
3719                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3720                         continue;
3721                 min_tolerated = min(min_tolerated,
3722                                     btrfs_raid_array[raid_type].
3723                                     tolerated_failures);
3724         }
3725
3726         if (min_tolerated == INT_MAX) {
3727                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3728                 min_tolerated = 0;
3729         }
3730
3731         return min_tolerated;
3732 }
3733
3734 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3735 {
3736         struct list_head *head;
3737         struct btrfs_device *dev;
3738         struct btrfs_super_block *sb;
3739         struct btrfs_dev_item *dev_item;
3740         int ret;
3741         int do_barriers;
3742         int max_errors;
3743         int total_errors = 0;
3744         u64 flags;
3745
3746         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3747
3748         /*
3749          * max_mirrors == 0 indicates we're from commit_transaction,
3750          * not from fsync where the tree roots in fs_info have not
3751          * been consistent on disk.
3752          */
3753         if (max_mirrors == 0)
3754                 backup_super_roots(fs_info);
3755
3756         sb = fs_info->super_for_commit;
3757         dev_item = &sb->dev_item;
3758
3759         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3760         head = &fs_info->fs_devices->devices;
3761         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3762
3763         if (do_barriers) {
3764                 ret = barrier_all_devices(fs_info);
3765                 if (ret) {
3766                         mutex_unlock(
3767                                 &fs_info->fs_devices->device_list_mutex);
3768                         btrfs_handle_fs_error(fs_info, ret,
3769                                               "errors while submitting device barriers.");
3770                         return ret;
3771                 }
3772         }
3773
3774         list_for_each_entry(dev, head, dev_list) {
3775                 if (!dev->bdev) {
3776                         total_errors++;
3777                         continue;
3778                 }
3779                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3780                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3781                         continue;
3782
3783                 btrfs_set_stack_device_generation(dev_item, 0);
3784                 btrfs_set_stack_device_type(dev_item, dev->type);
3785                 btrfs_set_stack_device_id(dev_item, dev->devid);
3786                 btrfs_set_stack_device_total_bytes(dev_item,
3787                                                    dev->commit_total_bytes);
3788                 btrfs_set_stack_device_bytes_used(dev_item,
3789                                                   dev->commit_bytes_used);
3790                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3791                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3792                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3793                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3794                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3795                        BTRFS_FSID_SIZE);
3796
3797                 flags = btrfs_super_flags(sb);
3798                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3799
3800                 ret = btrfs_validate_write_super(fs_info, sb);
3801                 if (ret < 0) {
3802                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3803                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3804                                 "unexpected superblock corruption detected");
3805                         return -EUCLEAN;
3806                 }
3807
3808                 ret = write_dev_supers(dev, sb, max_mirrors);
3809                 if (ret)
3810                         total_errors++;
3811         }
3812         if (total_errors > max_errors) {
3813                 btrfs_err(fs_info, "%d errors while writing supers",
3814                           total_errors);
3815                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3816
3817                 /* FUA is masked off if unsupported and can't be the reason */
3818                 btrfs_handle_fs_error(fs_info, -EIO,
3819                                       "%d errors while writing supers",
3820                                       total_errors);
3821                 return -EIO;
3822         }
3823
3824         total_errors = 0;
3825         list_for_each_entry(dev, head, dev_list) {
3826                 if (!dev->bdev)
3827                         continue;
3828                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3829                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3830                         continue;
3831
3832                 ret = wait_dev_supers(dev, max_mirrors);
3833                 if (ret)
3834                         total_errors++;
3835         }
3836         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3837         if (total_errors > max_errors) {
3838                 btrfs_handle_fs_error(fs_info, -EIO,
3839                                       "%d errors while writing supers",
3840                                       total_errors);
3841                 return -EIO;
3842         }
3843         return 0;
3844 }
3845
3846 /* Drop a fs root from the radix tree and free it. */
3847 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3848                                   struct btrfs_root *root)
3849 {
3850         spin_lock(&fs_info->fs_roots_radix_lock);
3851         radix_tree_delete(&fs_info->fs_roots_radix,
3852                           (unsigned long)root->root_key.objectid);
3853         spin_unlock(&fs_info->fs_roots_radix_lock);
3854
3855         if (btrfs_root_refs(&root->root_item) == 0)
3856                 synchronize_srcu(&fs_info->subvol_srcu);
3857
3858         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3859                 btrfs_free_log(NULL, root);
3860                 if (root->reloc_root) {
3861                         free_extent_buffer(root->reloc_root->node);
3862                         free_extent_buffer(root->reloc_root->commit_root);
3863                         btrfs_put_fs_root(root->reloc_root);
3864                         root->reloc_root = NULL;
3865                 }
3866         }
3867
3868         if (root->free_ino_pinned)
3869                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3870         if (root->free_ino_ctl)
3871                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3872         btrfs_free_fs_root(root);
3873 }
3874
3875 void btrfs_free_fs_root(struct btrfs_root *root)
3876 {
3877         iput(root->ino_cache_inode);
3878         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3879         if (root->anon_dev)
3880                 free_anon_bdev(root->anon_dev);
3881         if (root->subv_writers)
3882                 btrfs_free_subvolume_writers(root->subv_writers);
3883         free_extent_buffer(root->node);
3884         free_extent_buffer(root->commit_root);
3885         kfree(root->free_ino_ctl);
3886         kfree(root->free_ino_pinned);
3887         btrfs_put_fs_root(root);
3888 }
3889
3890 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3891 {
3892         u64 root_objectid = 0;
3893         struct btrfs_root *gang[8];
3894         int i = 0;
3895         int err = 0;
3896         unsigned int ret = 0;
3897         int index;
3898
3899         while (1) {
3900                 index = srcu_read_lock(&fs_info->subvol_srcu);
3901                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3902                                              (void **)gang, root_objectid,
3903                                              ARRAY_SIZE(gang));
3904                 if (!ret) {
3905                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3906                         break;
3907                 }
3908                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3909
3910                 for (i = 0; i < ret; i++) {
3911                         /* Avoid to grab roots in dead_roots */
3912                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3913                                 gang[i] = NULL;
3914                                 continue;
3915                         }
3916                         /* grab all the search result for later use */
3917                         gang[i] = btrfs_grab_fs_root(gang[i]);
3918                 }
3919                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3920
3921                 for (i = 0; i < ret; i++) {
3922                         if (!gang[i])
3923                                 continue;
3924                         root_objectid = gang[i]->root_key.objectid;
3925                         err = btrfs_orphan_cleanup(gang[i]);
3926                         if (err)
3927                                 break;
3928                         btrfs_put_fs_root(gang[i]);
3929                 }
3930                 root_objectid++;
3931         }
3932
3933         /* release the uncleaned roots due to error */
3934         for (; i < ret; i++) {
3935                 if (gang[i])
3936                         btrfs_put_fs_root(gang[i]);
3937         }
3938         return err;
3939 }
3940
3941 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3942 {
3943         struct btrfs_root *root = fs_info->tree_root;
3944         struct btrfs_trans_handle *trans;
3945
3946         mutex_lock(&fs_info->cleaner_mutex);
3947         btrfs_run_delayed_iputs(fs_info);
3948         mutex_unlock(&fs_info->cleaner_mutex);
3949         wake_up_process(fs_info->cleaner_kthread);
3950
3951         /* wait until ongoing cleanup work done */
3952         down_write(&fs_info->cleanup_work_sem);
3953         up_write(&fs_info->cleanup_work_sem);
3954
3955         trans = btrfs_join_transaction(root);
3956         if (IS_ERR(trans))
3957                 return PTR_ERR(trans);
3958         return btrfs_commit_transaction(trans);
3959 }
3960
3961 void close_ctree(struct btrfs_fs_info *fs_info)
3962 {
3963         int ret;
3964
3965         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3966         /*
3967          * We don't want the cleaner to start new transactions, add more delayed
3968          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3969          * because that frees the task_struct, and the transaction kthread might
3970          * still try to wake up the cleaner.
3971          */
3972         kthread_park(fs_info->cleaner_kthread);
3973
3974         /* wait for the qgroup rescan worker to stop */
3975         btrfs_qgroup_wait_for_completion(fs_info, false);
3976
3977         /* wait for the uuid_scan task to finish */
3978         down(&fs_info->uuid_tree_rescan_sem);
3979         /* avoid complains from lockdep et al., set sem back to initial state */
3980         up(&fs_info->uuid_tree_rescan_sem);
3981
3982         /* pause restriper - we want to resume on mount */
3983         btrfs_pause_balance(fs_info);
3984
3985         btrfs_dev_replace_suspend_for_unmount(fs_info);
3986
3987         btrfs_scrub_cancel(fs_info);
3988
3989         /* wait for any defraggers to finish */
3990         wait_event(fs_info->transaction_wait,
3991                    (atomic_read(&fs_info->defrag_running) == 0));
3992
3993         /* clear out the rbtree of defraggable inodes */
3994         btrfs_cleanup_defrag_inodes(fs_info);
3995
3996         cancel_work_sync(&fs_info->async_reclaim_work);
3997
3998         if (!sb_rdonly(fs_info->sb)) {
3999                 /*
4000                  * The cleaner kthread is stopped, so do one final pass over
4001                  * unused block groups.
4002                  */
4003                 btrfs_delete_unused_bgs(fs_info);
4004
4005                 ret = btrfs_commit_super(fs_info);
4006                 if (ret)
4007                         btrfs_err(fs_info, "commit super ret %d", ret);
4008         }
4009
4010         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4011             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4012                 btrfs_error_commit_super(fs_info);
4013
4014         kthread_stop(fs_info->transaction_kthread);
4015         kthread_stop(fs_info->cleaner_kthread);
4016
4017         ASSERT(list_empty(&fs_info->delayed_iputs));
4018         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4019
4020         btrfs_free_qgroup_config(fs_info);
4021         ASSERT(list_empty(&fs_info->delalloc_roots));
4022
4023         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4024                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4025                        percpu_counter_sum(&fs_info->delalloc_bytes));
4026         }
4027
4028         if (percpu_counter_sum(&fs_info->dio_bytes))
4029                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4030                            percpu_counter_sum(&fs_info->dio_bytes));
4031
4032         btrfs_sysfs_remove_mounted(fs_info);
4033         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4034
4035         btrfs_free_fs_roots(fs_info);
4036
4037         btrfs_put_block_group_cache(fs_info);
4038
4039         /*
4040          * we must make sure there is not any read request to
4041          * submit after we stopping all workers.
4042          */
4043         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4044         btrfs_stop_all_workers(fs_info);
4045
4046         btrfs_free_block_groups(fs_info);
4047
4048         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4049         free_root_pointers(fs_info, 1);
4050
4051         iput(fs_info->btree_inode);
4052
4053 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4054         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4055                 btrfsic_unmount(fs_info->fs_devices);
4056 #endif
4057
4058         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4059         btrfs_close_devices(fs_info->fs_devices);
4060
4061         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4062         percpu_counter_destroy(&fs_info->delalloc_bytes);
4063         percpu_counter_destroy(&fs_info->dio_bytes);
4064         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4065         cleanup_srcu_struct(&fs_info->subvol_srcu);
4066
4067         btrfs_free_stripe_hash_table(fs_info);
4068         btrfs_free_ref_cache(fs_info);
4069 }
4070
4071 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4072                           int atomic)
4073 {
4074         int ret;
4075         struct inode *btree_inode = buf->pages[0]->mapping->host;
4076
4077         ret = extent_buffer_uptodate(buf);
4078         if (!ret)
4079                 return ret;
4080
4081         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4082                                     parent_transid, atomic);
4083         if (ret == -EAGAIN)
4084                 return ret;
4085         return !ret;
4086 }
4087
4088 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4089 {
4090         struct btrfs_fs_info *fs_info;
4091         struct btrfs_root *root;
4092         u64 transid = btrfs_header_generation(buf);
4093         int was_dirty;
4094
4095 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4096         /*
4097          * This is a fast path so only do this check if we have sanity tests
4098          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4099          * outside of the sanity tests.
4100          */
4101         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4102                 return;
4103 #endif
4104         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4105         fs_info = root->fs_info;
4106         btrfs_assert_tree_locked(buf);
4107         if (transid != fs_info->generation)
4108                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4109                         buf->start, transid, fs_info->generation);
4110         was_dirty = set_extent_buffer_dirty(buf);
4111         if (!was_dirty)
4112                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4113                                          buf->len,
4114                                          fs_info->dirty_metadata_batch);
4115 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4116         /*
4117          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4118          * but item data not updated.
4119          * So here we should only check item pointers, not item data.
4120          */
4121         if (btrfs_header_level(buf) == 0 &&
4122             btrfs_check_leaf_relaxed(buf)) {
4123                 btrfs_print_leaf(buf);
4124                 ASSERT(0);
4125         }
4126 #endif
4127 }
4128
4129 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4130                                         int flush_delayed)
4131 {
4132         /*
4133          * looks as though older kernels can get into trouble with
4134          * this code, they end up stuck in balance_dirty_pages forever
4135          */
4136         int ret;
4137
4138         if (current->flags & PF_MEMALLOC)
4139                 return;
4140
4141         if (flush_delayed)
4142                 btrfs_balance_delayed_items(fs_info);
4143
4144         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4145                                      BTRFS_DIRTY_METADATA_THRESH,
4146                                      fs_info->dirty_metadata_batch);
4147         if (ret > 0) {
4148                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4149         }
4150 }
4151
4152 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4153 {
4154         __btrfs_btree_balance_dirty(fs_info, 1);
4155 }
4156
4157 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4158 {
4159         __btrfs_btree_balance_dirty(fs_info, 0);
4160 }
4161
4162 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4163                       struct btrfs_key *first_key)
4164 {
4165         return btree_read_extent_buffer_pages(buf, parent_transid,
4166                                               level, first_key);
4167 }
4168
4169 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4170 {
4171         /* cleanup FS via transaction */
4172         btrfs_cleanup_transaction(fs_info);
4173
4174         mutex_lock(&fs_info->cleaner_mutex);
4175         btrfs_run_delayed_iputs(fs_info);
4176         mutex_unlock(&fs_info->cleaner_mutex);
4177
4178         down_write(&fs_info->cleanup_work_sem);
4179         up_write(&fs_info->cleanup_work_sem);
4180 }
4181
4182 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4183 {
4184         struct btrfs_ordered_extent *ordered;
4185
4186         spin_lock(&root->ordered_extent_lock);
4187         /*
4188          * This will just short circuit the ordered completion stuff which will
4189          * make sure the ordered extent gets properly cleaned up.
4190          */
4191         list_for_each_entry(ordered, &root->ordered_extents,
4192                             root_extent_list)
4193                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4194         spin_unlock(&root->ordered_extent_lock);
4195 }
4196
4197 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4198 {
4199         struct btrfs_root *root;
4200         struct list_head splice;
4201
4202         INIT_LIST_HEAD(&splice);
4203
4204         spin_lock(&fs_info->ordered_root_lock);
4205         list_splice_init(&fs_info->ordered_roots, &splice);
4206         while (!list_empty(&splice)) {
4207                 root = list_first_entry(&splice, struct btrfs_root,
4208                                         ordered_root);
4209                 list_move_tail(&root->ordered_root,
4210                                &fs_info->ordered_roots);
4211
4212                 spin_unlock(&fs_info->ordered_root_lock);
4213                 btrfs_destroy_ordered_extents(root);
4214
4215                 cond_resched();
4216                 spin_lock(&fs_info->ordered_root_lock);
4217         }
4218         spin_unlock(&fs_info->ordered_root_lock);
4219
4220         /*
4221          * We need this here because if we've been flipped read-only we won't
4222          * get sync() from the umount, so we need to make sure any ordered
4223          * extents that haven't had their dirty pages IO start writeout yet
4224          * actually get run and error out properly.
4225          */
4226         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4227 }
4228
4229 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4230                                       struct btrfs_fs_info *fs_info)
4231 {
4232         struct rb_node *node;
4233         struct btrfs_delayed_ref_root *delayed_refs;
4234         struct btrfs_delayed_ref_node *ref;
4235         int ret = 0;
4236
4237         delayed_refs = &trans->delayed_refs;
4238
4239         spin_lock(&delayed_refs->lock);
4240         if (atomic_read(&delayed_refs->num_entries) == 0) {
4241                 spin_unlock(&delayed_refs->lock);
4242                 btrfs_info(fs_info, "delayed_refs has NO entry");
4243                 return ret;
4244         }
4245
4246         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4247                 struct btrfs_delayed_ref_head *head;
4248                 struct rb_node *n;
4249                 bool pin_bytes = false;
4250
4251                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4252                                 href_node);
4253                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4254                         continue;
4255
4256                 spin_lock(&head->lock);
4257                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4258                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4259                                        ref_node);
4260                         ref->in_tree = 0;
4261                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4262                         RB_CLEAR_NODE(&ref->ref_node);
4263                         if (!list_empty(&ref->add_list))
4264                                 list_del(&ref->add_list);
4265                         atomic_dec(&delayed_refs->num_entries);
4266                         btrfs_put_delayed_ref(ref);
4267                 }
4268                 if (head->must_insert_reserved)
4269                         pin_bytes = true;
4270                 btrfs_free_delayed_extent_op(head->extent_op);
4271                 btrfs_delete_ref_head(delayed_refs, head);
4272                 spin_unlock(&head->lock);
4273                 spin_unlock(&delayed_refs->lock);
4274                 mutex_unlock(&head->mutex);
4275
4276                 if (pin_bytes)
4277                         btrfs_pin_extent(fs_info, head->bytenr,
4278                                          head->num_bytes, 1);
4279                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4280                 btrfs_put_delayed_ref_head(head);
4281                 cond_resched();
4282                 spin_lock(&delayed_refs->lock);
4283         }
4284
4285         spin_unlock(&delayed_refs->lock);
4286
4287         return ret;
4288 }
4289
4290 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4291 {
4292         struct btrfs_inode *btrfs_inode;
4293         struct list_head splice;
4294
4295         INIT_LIST_HEAD(&splice);
4296
4297         spin_lock(&root->delalloc_lock);
4298         list_splice_init(&root->delalloc_inodes, &splice);
4299
4300         while (!list_empty(&splice)) {
4301                 struct inode *inode = NULL;
4302                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4303                                                delalloc_inodes);
4304                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4305                 spin_unlock(&root->delalloc_lock);
4306
4307                 /*
4308                  * Make sure we get a live inode and that it'll not disappear
4309                  * meanwhile.
4310                  */
4311                 inode = igrab(&btrfs_inode->vfs_inode);
4312                 if (inode) {
4313                         invalidate_inode_pages2(inode->i_mapping);
4314                         iput(inode);
4315                 }
4316                 spin_lock(&root->delalloc_lock);
4317         }
4318         spin_unlock(&root->delalloc_lock);
4319 }
4320
4321 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4322 {
4323         struct btrfs_root *root;
4324         struct list_head splice;
4325
4326         INIT_LIST_HEAD(&splice);
4327
4328         spin_lock(&fs_info->delalloc_root_lock);
4329         list_splice_init(&fs_info->delalloc_roots, &splice);
4330         while (!list_empty(&splice)) {
4331                 root = list_first_entry(&splice, struct btrfs_root,
4332                                          delalloc_root);
4333                 root = btrfs_grab_fs_root(root);
4334                 BUG_ON(!root);
4335                 spin_unlock(&fs_info->delalloc_root_lock);
4336
4337                 btrfs_destroy_delalloc_inodes(root);
4338                 btrfs_put_fs_root(root);
4339
4340                 spin_lock(&fs_info->delalloc_root_lock);
4341         }
4342         spin_unlock(&fs_info->delalloc_root_lock);
4343 }
4344
4345 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4346                                         struct extent_io_tree *dirty_pages,
4347                                         int mark)
4348 {
4349         int ret;
4350         struct extent_buffer *eb;
4351         u64 start = 0;
4352         u64 end;
4353
4354         while (1) {
4355                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4356                                             mark, NULL);
4357                 if (ret)
4358                         break;
4359
4360                 clear_extent_bits(dirty_pages, start, end, mark);
4361                 while (start <= end) {
4362                         eb = find_extent_buffer(fs_info, start);
4363                         start += fs_info->nodesize;
4364                         if (!eb)
4365                                 continue;
4366                         wait_on_extent_buffer_writeback(eb);
4367
4368                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4369                                                &eb->bflags))
4370                                 clear_extent_buffer_dirty(eb);
4371                         free_extent_buffer_stale(eb);
4372                 }
4373         }
4374
4375         return ret;
4376 }
4377
4378 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4379                                        struct extent_io_tree *pinned_extents)
4380 {
4381         struct extent_io_tree *unpin;
4382         u64 start;
4383         u64 end;
4384         int ret;
4385         bool loop = true;
4386
4387         unpin = pinned_extents;
4388 again:
4389         while (1) {
4390                 struct extent_state *cached_state = NULL;
4391
4392                 /*
4393                  * The btrfs_finish_extent_commit() may get the same range as
4394                  * ours between find_first_extent_bit and clear_extent_dirty.
4395                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4396                  * the same extent range.
4397                  */
4398                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4399                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4400                                             EXTENT_DIRTY, &cached_state);
4401                 if (ret) {
4402                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4403                         break;
4404                 }
4405
4406                 clear_extent_dirty(unpin, start, end, &cached_state);
4407                 free_extent_state(cached_state);
4408                 btrfs_error_unpin_extent_range(fs_info, start, end);
4409                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4410                 cond_resched();
4411         }
4412
4413         if (loop) {
4414                 if (unpin == &fs_info->freed_extents[0])
4415                         unpin = &fs_info->freed_extents[1];
4416                 else
4417                         unpin = &fs_info->freed_extents[0];
4418                 loop = false;
4419                 goto again;
4420         }
4421
4422         return 0;
4423 }
4424
4425 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4426 {
4427         struct inode *inode;
4428
4429         inode = cache->io_ctl.inode;
4430         if (inode) {
4431                 invalidate_inode_pages2(inode->i_mapping);
4432                 BTRFS_I(inode)->generation = 0;
4433                 cache->io_ctl.inode = NULL;
4434                 iput(inode);
4435         }
4436         btrfs_put_block_group(cache);
4437 }
4438
4439 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4440                              struct btrfs_fs_info *fs_info)
4441 {
4442         struct btrfs_block_group_cache *cache;
4443
4444         spin_lock(&cur_trans->dirty_bgs_lock);
4445         while (!list_empty(&cur_trans->dirty_bgs)) {
4446                 cache = list_first_entry(&cur_trans->dirty_bgs,
4447                                          struct btrfs_block_group_cache,
4448                                          dirty_list);
4449
4450                 if (!list_empty(&cache->io_list)) {
4451                         spin_unlock(&cur_trans->dirty_bgs_lock);
4452                         list_del_init(&cache->io_list);
4453                         btrfs_cleanup_bg_io(cache);
4454                         spin_lock(&cur_trans->dirty_bgs_lock);
4455                 }
4456
4457                 list_del_init(&cache->dirty_list);
4458                 spin_lock(&cache->lock);
4459                 cache->disk_cache_state = BTRFS_DC_ERROR;
4460                 spin_unlock(&cache->lock);
4461
4462                 spin_unlock(&cur_trans->dirty_bgs_lock);
4463                 btrfs_put_block_group(cache);
4464                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4465                 spin_lock(&cur_trans->dirty_bgs_lock);
4466         }
4467         spin_unlock(&cur_trans->dirty_bgs_lock);
4468
4469         /*
4470          * Refer to the definition of io_bgs member for details why it's safe
4471          * to use it without any locking
4472          */
4473         while (!list_empty(&cur_trans->io_bgs)) {
4474                 cache = list_first_entry(&cur_trans->io_bgs,
4475                                          struct btrfs_block_group_cache,
4476                                          io_list);
4477
4478                 list_del_init(&cache->io_list);
4479                 spin_lock(&cache->lock);
4480                 cache->disk_cache_state = BTRFS_DC_ERROR;
4481                 spin_unlock(&cache->lock);
4482                 btrfs_cleanup_bg_io(cache);
4483         }
4484 }
4485
4486 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4487                                    struct btrfs_fs_info *fs_info)
4488 {
4489         struct btrfs_device *dev, *tmp;
4490
4491         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4492         ASSERT(list_empty(&cur_trans->dirty_bgs));
4493         ASSERT(list_empty(&cur_trans->io_bgs));
4494
4495         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4496                                  post_commit_list) {
4497                 list_del_init(&dev->post_commit_list);
4498         }
4499
4500         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4501
4502         cur_trans->state = TRANS_STATE_COMMIT_START;
4503         wake_up(&fs_info->transaction_blocked_wait);
4504
4505         cur_trans->state = TRANS_STATE_UNBLOCKED;
4506         wake_up(&fs_info->transaction_wait);
4507
4508         btrfs_destroy_delayed_inodes(fs_info);
4509         btrfs_assert_delayed_root_empty(fs_info);
4510
4511         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4512                                      EXTENT_DIRTY);
4513         btrfs_destroy_pinned_extent(fs_info,
4514                                     fs_info->pinned_extents);
4515
4516         cur_trans->state =TRANS_STATE_COMPLETED;
4517         wake_up(&cur_trans->commit_wait);
4518 }
4519
4520 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4521 {
4522         struct btrfs_transaction *t;
4523
4524         mutex_lock(&fs_info->transaction_kthread_mutex);
4525
4526         spin_lock(&fs_info->trans_lock);
4527         while (!list_empty(&fs_info->trans_list)) {
4528                 t = list_first_entry(&fs_info->trans_list,
4529                                      struct btrfs_transaction, list);
4530                 if (t->state >= TRANS_STATE_COMMIT_START) {
4531                         refcount_inc(&t->use_count);
4532                         spin_unlock(&fs_info->trans_lock);
4533                         btrfs_wait_for_commit(fs_info, t->transid);
4534                         btrfs_put_transaction(t);
4535                         spin_lock(&fs_info->trans_lock);
4536                         continue;
4537                 }
4538                 if (t == fs_info->running_transaction) {
4539                         t->state = TRANS_STATE_COMMIT_DOING;
4540                         spin_unlock(&fs_info->trans_lock);
4541                         /*
4542                          * We wait for 0 num_writers since we don't hold a trans
4543                          * handle open currently for this transaction.
4544                          */
4545                         wait_event(t->writer_wait,
4546                                    atomic_read(&t->num_writers) == 0);
4547                 } else {
4548                         spin_unlock(&fs_info->trans_lock);
4549                 }
4550                 btrfs_cleanup_one_transaction(t, fs_info);
4551
4552                 spin_lock(&fs_info->trans_lock);
4553                 if (t == fs_info->running_transaction)
4554                         fs_info->running_transaction = NULL;
4555                 list_del_init(&t->list);
4556                 spin_unlock(&fs_info->trans_lock);
4557
4558                 btrfs_put_transaction(t);
4559                 trace_btrfs_transaction_commit(fs_info->tree_root);
4560                 spin_lock(&fs_info->trans_lock);
4561         }
4562         spin_unlock(&fs_info->trans_lock);
4563         btrfs_destroy_all_ordered_extents(fs_info);
4564         btrfs_destroy_delayed_inodes(fs_info);
4565         btrfs_assert_delayed_root_empty(fs_info);
4566         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4567         btrfs_destroy_all_delalloc_inodes(fs_info);
4568         mutex_unlock(&fs_info->transaction_kthread_mutex);
4569
4570         return 0;
4571 }
4572
4573 static const struct extent_io_ops btree_extent_io_ops = {
4574         /* mandatory callbacks */
4575         .submit_bio_hook = btree_submit_bio_hook,
4576         .readpage_end_io_hook = btree_readpage_end_io_hook,
4577 };