fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/kthread.h>
  12 #include <linux/slab.h>
  13 #include <linux/migrate.h>
  14 #include <linux/ratelimit.h>
  15 #include <linux/uuid.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/crc32c.h>
  19 #include <linux/sched/mm.h>
  20 #include <asm/unaligned.h>
  21 #include <crypto/hash.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "check-integrity.h"
  33 #include "rcu-string.h"
  34 #include "dev-replace.h"
  35 #include "raid56.h"
  36 #include "sysfs.h"
  37 #include "qgroup.h"
  38 #include "compression.h"
  39 #include "tree-checker.h"
  40 #include "ref-verify.h"
  41 #include "block-group.h"
  42 #include "discard.h"
  43 #include "space-info.h"
  44 #include "zoned.h"
  45 #include "subpage.h"
  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 void end_workqueue_fn(struct btrfs_work *work);
  55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  57                                       struct btrfs_fs_info *fs_info);
  58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  60                                         struct extent_io_tree *dirty_pages,
  61                                         int mark);
  62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  63                                        struct extent_io_tree *pinned_extents);
  64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  66
  67 /*
  68  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  69  * is complete.  This is used during reads to verify checksums, and it is used
  70  * by writes to insert metadata for new file extents after IO is complete.
  71  */
  72 struct btrfs_end_io_wq {
  73         struct bio *bio;
  74         bio_end_io_t *end_io;
  75         void *private;
  76         struct btrfs_fs_info *info;
  77         blk_status_t status;
  78         enum btrfs_wq_endio_type metadata;
  79         struct btrfs_work work;
  80 };
  81
  82 static struct kmem_cache *btrfs_end_io_wq_cache;
  83
  84 int __init btrfs_end_io_wq_init(void)
  85 {
  86         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  87                                         sizeof(struct btrfs_end_io_wq),
  88                                         0,
  89                                         SLAB_MEM_SPREAD,
  90                                         NULL);
  91         if (!btrfs_end_io_wq_cache)
  92                 return -ENOMEM;
  93         return 0;
  94 }
  95
  96 void __cold btrfs_end_io_wq_exit(void)
  97 {
  98         kmem_cache_destroy(btrfs_end_io_wq_cache);
  99 }
 100
 101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
 102 {
 103         if (fs_info->csum_shash)
 104                 crypto_free_shash(fs_info->csum_shash);
 105 }
 106
 107 /*
 108  * async submit bios are used to offload expensive checksumming
 109  * onto the worker threads.  They checksum file and metadata bios
 110  * just before they are sent down the IO stack.
 111  */
 112 struct async_submit_bio {
 113         struct inode *inode;
 114         struct bio *bio;
 115         extent_submit_bio_start_t *submit_bio_start;
 116         int mirror_num;
 117
 118         /* Optional parameter for submit_bio_start used by direct io */
 119         u64 dio_file_offset;
 120         struct btrfs_work work;
 121         blk_status_t status;
 122 };
 123
 124 /*
 125  * Compute the csum of a btree block and store the result to provided buffer.
 126  */
 127 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
 128 {
 129         struct btrfs_fs_info *fs_info = buf->fs_info;
 130         const int num_pages = num_extent_pages(buf);
 131         const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
 132         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 133         char *kaddr;
 134         int i;
 135
 136         shash->tfm = fs_info->csum_shash;
 137         crypto_shash_init(shash);
 138         kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
 139         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
 140                             first_page_part - BTRFS_CSUM_SIZE);
 141
 142         for (i = 1; i < num_pages; i++) {
 143                 kaddr = page_address(buf->pages[i]);
 144                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 145         }
 146         memset(result, 0, BTRFS_CSUM_SIZE);
 147         crypto_shash_final(shash, result);
 148 }
 149
 150 /*
 151  * we can't consider a given block up to date unless the transid of the
 152  * block matches the transid in the parent node's pointer.  This is how we
 153  * detect blocks that either didn't get written at all or got written
 154  * in the wrong place.
 155  */
 156 static int verify_parent_transid(struct extent_io_tree *io_tree,
 157                                  struct extent_buffer *eb, u64 parent_transid,
 158                                  int atomic)
 159 {
 160         struct extent_state *cached_state = NULL;
 161         int ret;
 162
 163         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 164                 return 0;
 165
 166         if (atomic)
 167                 return -EAGAIN;
 168
 169         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 170                          &cached_state);
 171         if (extent_buffer_uptodate(eb) &&
 172             btrfs_header_generation(eb) == parent_transid) {
 173                 ret = 0;
 174                 goto out;
 175         }
 176         btrfs_err_rl(eb->fs_info,
 177                 "parent transid verify failed on %llu wanted %llu found %llu",
 178                         eb->start,
 179                         parent_transid, btrfs_header_generation(eb));
 180         ret = 1;
 181         clear_extent_buffer_uptodate(eb);
 182 out:
 183         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 184                              &cached_state);
 185         return ret;
 186 }
 187
 188 static bool btrfs_supported_super_csum(u16 csum_type)
 189 {
 190         switch (csum_type) {
 191         case BTRFS_CSUM_TYPE_CRC32:
 192         case BTRFS_CSUM_TYPE_XXHASH:
 193         case BTRFS_CSUM_TYPE_SHA256:
 194         case BTRFS_CSUM_TYPE_BLAKE2:
 195                 return true;
 196         default:
 197                 return false;
 198         }
 199 }
 200
 201 /*
 202  * Return 0 if the superblock checksum type matches the checksum value of that
 203  * algorithm. Pass the raw disk superblock data.
 204  */
 205 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 206                            const struct btrfs_super_block *disk_sb)
 207 {
 208         char result[BTRFS_CSUM_SIZE];
 209         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 210
 211         shash->tfm = fs_info->csum_shash;
 212
 213         /*
 214          * The super_block structure does not span the whole
 215          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 216          * filled with zeros and is included in the checksum.
 217          */
 218         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 219                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 220
 221         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 222                 return 1;
 223
 224         return 0;
 225 }
 226
 227 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 228                            struct btrfs_key *first_key, u64 parent_transid)
 229 {
 230         struct btrfs_fs_info *fs_info = eb->fs_info;
 231         int found_level;
 232         struct btrfs_key found_key;
 233         int ret;
 234
 235         found_level = btrfs_header_level(eb);
 236         if (found_level != level) {
 237                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 238                      KERN_ERR "BTRFS: tree level check failed\n");
 239                 btrfs_err(fs_info,
 240 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 241                           eb->start, level, found_level);
 242                 return -EIO;
 243         }
 244
 245         if (!first_key)
 246                 return 0;
 247
 248         /*
 249          * For live tree block (new tree blocks in current transaction),
 250          * we need proper lock context to avoid race, which is impossible here.
 251          * So we only checks tree blocks which is read from disk, whose
 252          * generation <= fs_info->last_trans_committed.
 253          */
 254         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 255                 return 0;
 256
 257         /* We have @first_key, so this @eb must have at least one item */
 258         if (btrfs_header_nritems(eb) == 0) {
 259                 btrfs_err(fs_info,
 260                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 261                           eb->start);
 262                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 263                 return -EUCLEAN;
 264         }
 265
 266         if (found_level)
 267                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 268         else
 269                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 270         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 271
 272         if (ret) {
 273                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 274                      KERN_ERR "BTRFS: tree first key check failed\n");
 275                 btrfs_err(fs_info,
 276 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 277                           eb->start, parent_transid, first_key->objectid,
 278                           first_key->type, first_key->offset,
 279                           found_key.objectid, found_key.type,
 280                           found_key.offset);
 281         }
 282         return ret;
 283 }
 284
 285 /*
 286  * helper to read a given tree block, doing retries as required when
 287  * the checksums don't match and we have alternate mirrors to try.
 288  *
 289  * @parent_transid:     expected transid, skip check if 0
 290  * @level:              expected level, mandatory check
 291  * @first_key:          expected key of first slot, skip check if NULL
 292  */
 293 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 294                                           u64 parent_transid, int level,
 295                                           struct btrfs_key *first_key)
 296 {
 297         struct btrfs_fs_info *fs_info = eb->fs_info;
 298         struct extent_io_tree *io_tree;
 299         int failed = 0;
 300         int ret;
 301         int num_copies = 0;
 302         int mirror_num = 0;
 303         int failed_mirror = 0;
 304
 305         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 306         while (1) {
 307                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 308                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 309                 if (!ret) {
 310                         if (verify_parent_transid(io_tree, eb,
 311                                                    parent_transid, 0))
 312                                 ret = -EIO;
 313                         else if (btrfs_verify_level_key(eb, level,
 314                                                 first_key, parent_transid))
 315                                 ret = -EUCLEAN;
 316                         else
 317                                 break;
 318                 }
 319
 320                 num_copies = btrfs_num_copies(fs_info,
 321                                               eb->start, eb->len);
 322                 if (num_copies == 1)
 323                         break;
 324
 325                 if (!failed_mirror) {
 326                         failed = 1;
 327                         failed_mirror = eb->read_mirror;
 328                 }
 329
 330                 mirror_num++;
 331                 if (mirror_num == failed_mirror)
 332                         mirror_num++;
 333
 334                 if (mirror_num > num_copies)
 335                         break;
 336         }
 337
 338         if (failed && !ret && failed_mirror)
 339                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 340
 341         return ret;
 342 }
 343
 344 static int csum_one_extent_buffer(struct extent_buffer *eb)
 345 {
 346         struct btrfs_fs_info *fs_info = eb->fs_info;
 347         u8 result[BTRFS_CSUM_SIZE];
 348         int ret;
 349
 350         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 351                                     offsetof(struct btrfs_header, fsid),
 352                                     BTRFS_FSID_SIZE) == 0);
 353         csum_tree_block(eb, result);
 354
 355         if (btrfs_header_level(eb))
 356                 ret = btrfs_check_node(eb);
 357         else
 358                 ret = btrfs_check_leaf_full(eb);
 359
 360         if (ret < 0)
 361                 goto error;
 362
 363         /*
 364          * Also check the generation, the eb reached here must be newer than
 365          * last committed. Or something seriously wrong happened.
 366          */
 367         if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
 368                 ret = -EUCLEAN;
 369                 btrfs_err(fs_info,
 370                         "block=%llu bad generation, have %llu expect > %llu",
 371                           eb->start, btrfs_header_generation(eb),
 372                           fs_info->last_trans_committed);
 373                 goto error;
 374         }
 375         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 376
 377         return 0;
 378
 379 error:
 380         btrfs_print_tree(eb, 0);
 381         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 382                   eb->start);
 383         /*
 384          * Be noisy if this is an extent buffer from a log tree. We don't abort
 385          * a transaction in case there's a bad log tree extent buffer, we just
 386          * fallback to a transaction commit. Still we want to know when there is
 387          * a bad log tree extent buffer, as that may signal a bug somewhere.
 388          */
 389         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 390                 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 391         return ret;
 392 }
 393
 394 /* Checksum all dirty extent buffers in one bio_vec */
 395 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
 396                                       struct bio_vec *bvec)
 397 {
 398         struct page *page = bvec->bv_page;
 399         u64 bvec_start = page_offset(page) + bvec->bv_offset;
 400         u64 cur;
 401         int ret = 0;
 402
 403         for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
 404              cur += fs_info->nodesize) {
 405                 struct extent_buffer *eb;
 406                 bool uptodate;
 407
 408                 eb = find_extent_buffer(fs_info, cur);
 409                 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
 410                                                        fs_info->nodesize);
 411
 412                 /* A dirty eb shouldn't disappear from buffer_radix */
 413                 if (WARN_ON(!eb))
 414                         return -EUCLEAN;
 415
 416                 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
 417                         free_extent_buffer(eb);
 418                         return -EUCLEAN;
 419                 }
 420                 if (WARN_ON(!uptodate)) {
 421                         free_extent_buffer(eb);
 422                         return -EUCLEAN;
 423                 }
 424
 425                 ret = csum_one_extent_buffer(eb);
 426                 free_extent_buffer(eb);
 427                 if (ret < 0)
 428                         return ret;
 429         }
 430         return ret;
 431 }
 432
 433 /*
 434  * Checksum a dirty tree block before IO.  This has extra checks to make sure
 435  * we only fill in the checksum field in the first page of a multi-page block.
 436  * For subpage extent buffers we need bvec to also read the offset in the page.
 437  */
 438 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
 439 {
 440         struct page *page = bvec->bv_page;
 441         u64 start = page_offset(page);
 442         u64 found_start;
 443         struct extent_buffer *eb;
 444
 445         if (fs_info->sectorsize < PAGE_SIZE)
 446                 return csum_dirty_subpage_buffers(fs_info, bvec);
 447
 448         eb = (struct extent_buffer *)page->private;
 449         if (page != eb->pages[0])
 450                 return 0;
 451
 452         found_start = btrfs_header_bytenr(eb);
 453
 454         if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
 455                 WARN_ON(found_start != 0);
 456                 return 0;
 457         }
 458
 459         /*
 460          * Please do not consolidate these warnings into a single if.
 461          * It is useful to know what went wrong.
 462          */
 463         if (WARN_ON(found_start != start))
 464                 return -EUCLEAN;
 465         if (WARN_ON(!PageUptodate(page)))
 466                 return -EUCLEAN;
 467
 468         return csum_one_extent_buffer(eb);
 469 }
 470
 471 static int check_tree_block_fsid(struct extent_buffer *eb)
 472 {
 473         struct btrfs_fs_info *fs_info = eb->fs_info;
 474         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 475         u8 fsid[BTRFS_FSID_SIZE];
 476         u8 *metadata_uuid;
 477
 478         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 479                            BTRFS_FSID_SIZE);
 480         /*
 481          * Checking the incompat flag is only valid for the current fs. For
 482          * seed devices it's forbidden to have their uuid changed so reading
 483          * ->fsid in this case is fine
 484          */
 485         if (btrfs_fs_incompat(fs_info, METADATA_UUID))
 486                 metadata_uuid = fs_devices->metadata_uuid;
 487         else
 488                 metadata_uuid = fs_devices->fsid;
 489
 490         if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
 491                 return 0;
 492
 493         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 494                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 495                         return 0;
 496
 497         return 1;
 498 }
 499
 500 /* Do basic extent buffer checks at read time */
 501 static int validate_extent_buffer(struct extent_buffer *eb)
 502 {
 503         struct btrfs_fs_info *fs_info = eb->fs_info;
 504         u64 found_start;
 505         const u32 csum_size = fs_info->csum_size;
 506         u8 found_level;
 507         u8 result[BTRFS_CSUM_SIZE];
 508         const u8 *header_csum;
 509         int ret = 0;
 510
 511         found_start = btrfs_header_bytenr(eb);
 512         if (found_start != eb->start) {
 513                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 514                              eb->start, found_start);
 515                 ret = -EIO;
 516                 goto out;
 517         }
 518         if (check_tree_block_fsid(eb)) {
 519                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 520                              eb->start);
 521                 ret = -EIO;
 522                 goto out;
 523         }
 524         found_level = btrfs_header_level(eb);
 525         if (found_level >= BTRFS_MAX_LEVEL) {
 526                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 527                           (int)btrfs_header_level(eb), eb->start);
 528                 ret = -EIO;
 529                 goto out;
 530         }
 531
 532         csum_tree_block(eb, result);
 533         header_csum = page_address(eb->pages[0]) +
 534                 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
 535
 536         if (memcmp(result, header_csum, csum_size) != 0) {
 537                 btrfs_warn_rl(fs_info,
 538         "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 539                               eb->start,
 540                               CSUM_FMT_VALUE(csum_size, header_csum),
 541                               CSUM_FMT_VALUE(csum_size, result),
 542                               btrfs_header_level(eb));
 543                 ret = -EUCLEAN;
 544                 goto out;
 545         }
 546
 547         /*
 548          * If this is a leaf block and it is corrupt, set the corrupt bit so
 549          * that we don't try and read the other copies of this block, just
 550          * return -EIO.
 551          */
 552         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 553                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 554                 ret = -EIO;
 555         }
 556
 557         if (found_level > 0 && btrfs_check_node(eb))
 558                 ret = -EIO;
 559
 560         if (!ret)
 561                 set_extent_buffer_uptodate(eb);
 562         else
 563                 btrfs_err(fs_info,
 564                           "block=%llu read time tree block corruption detected",
 565                           eb->start);
 566 out:
 567         return ret;
 568 }
 569
 570 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
 571                                    int mirror)
 572 {
 573         struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
 574         struct extent_buffer *eb;
 575         bool reads_done;
 576         int ret = 0;
 577
 578         /*
 579          * We don't allow bio merge for subpage metadata read, so we should
 580          * only get one eb for each endio hook.
 581          */
 582         ASSERT(end == start + fs_info->nodesize - 1);
 583         ASSERT(PagePrivate(page));
 584
 585         eb = find_extent_buffer(fs_info, start);
 586         /*
 587          * When we are reading one tree block, eb must have been inserted into
 588          * the radix tree. If not, something is wrong.
 589          */
 590         ASSERT(eb);
 591
 592         reads_done = atomic_dec_and_test(&eb->io_pages);
 593         /* Subpage read must finish in page read */
 594         ASSERT(reads_done);
 595
 596         eb->read_mirror = mirror;
 597         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 598                 ret = -EIO;
 599                 goto err;
 600         }
 601         ret = validate_extent_buffer(eb);
 602         if (ret < 0)
 603                 goto err;
 604
 605         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 606                 btree_readahead_hook(eb, ret);
 607
 608         set_extent_buffer_uptodate(eb);
 609
 610         free_extent_buffer(eb);
 611         return ret;
 612 err:
 613         /*
 614          * end_bio_extent_readpage decrements io_pages in case of error,
 615          * make sure it has something to decrement.
 616          */
 617         atomic_inc(&eb->io_pages);
 618         clear_extent_buffer_uptodate(eb);
 619         free_extent_buffer(eb);
 620         return ret;
 621 }
 622
 623 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
 624                                    struct page *page, u64 start, u64 end,
 625                                    int mirror)
 626 {
 627         struct extent_buffer *eb;
 628         int ret = 0;
 629         int reads_done;
 630
 631         ASSERT(page->private);
 632
 633         if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
 634                 return validate_subpage_buffer(page, start, end, mirror);
 635
 636         eb = (struct extent_buffer *)page->private;
 637
 638         /*
 639          * The pending IO might have been the only thing that kept this buffer
 640          * in memory.  Make sure we have a ref for all this other checks
 641          */
 642         atomic_inc(&eb->refs);
 643
 644         reads_done = atomic_dec_and_test(&eb->io_pages);
 645         if (!reads_done)
 646                 goto err;
 647
 648         eb->read_mirror = mirror;
 649         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 650                 ret = -EIO;
 651                 goto err;
 652         }
 653         ret = validate_extent_buffer(eb);
 654 err:
 655         if (reads_done &&
 656             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 657                 btree_readahead_hook(eb, ret);
 658
 659         if (ret) {
 660                 /*
 661                  * our io error hook is going to dec the io pages
 662                  * again, we have to make sure it has something
 663                  * to decrement
 664                  */
 665                 atomic_inc(&eb->io_pages);
 666                 clear_extent_buffer_uptodate(eb);
 667         }
 668         free_extent_buffer(eb);
 669
 670         return ret;
 671 }
 672
 673 static void end_workqueue_bio(struct bio *bio)
 674 {
 675         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 676         struct btrfs_fs_info *fs_info;
 677         struct btrfs_workqueue *wq;
 678
 679         fs_info = end_io_wq->info;
 680         end_io_wq->status = bio->bi_status;
 681
 682         if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
 683                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 684                         wq = fs_info->endio_meta_write_workers;
 685                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 686                         wq = fs_info->endio_freespace_worker;
 687                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 688                         wq = fs_info->endio_raid56_workers;
 689                 else
 690                         wq = fs_info->endio_write_workers;
 691         } else {
 692                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 693                         wq = fs_info->endio_raid56_workers;
 694                 else if (end_io_wq->metadata)
 695                         wq = fs_info->endio_meta_workers;
 696                 else
 697                         wq = fs_info->endio_workers;
 698         }
 699
 700         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 701         btrfs_queue_work(wq, &end_io_wq->work);
 702 }
 703
 704 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 705                         enum btrfs_wq_endio_type metadata)
 706 {
 707         struct btrfs_end_io_wq *end_io_wq;
 708
 709         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 710         if (!end_io_wq)
 711                 return BLK_STS_RESOURCE;
 712
 713         end_io_wq->private = bio->bi_private;
 714         end_io_wq->end_io = bio->bi_end_io;
 715         end_io_wq->info = info;
 716         end_io_wq->status = 0;
 717         end_io_wq->bio = bio;
 718         end_io_wq->metadata = metadata;
 719
 720         bio->bi_private = end_io_wq;
 721         bio->bi_end_io = end_workqueue_bio;
 722         return 0;
 723 }
 724
 725 static void run_one_async_start(struct btrfs_work *work)
 726 {
 727         struct async_submit_bio *async;
 728         blk_status_t ret;
 729
 730         async = container_of(work, struct  async_submit_bio, work);
 731         ret = async->submit_bio_start(async->inode, async->bio,
 732                                       async->dio_file_offset);
 733         if (ret)
 734                 async->status = ret;
 735 }
 736
 737 /*
 738  * In order to insert checksums into the metadata in large chunks, we wait
 739  * until bio submission time.   All the pages in the bio are checksummed and
 740  * sums are attached onto the ordered extent record.
 741  *
 742  * At IO completion time the csums attached on the ordered extent record are
 743  * inserted into the tree.
 744  */
 745 static void run_one_async_done(struct btrfs_work *work)
 746 {
 747         struct async_submit_bio *async;
 748         struct inode *inode;
 749         blk_status_t ret;
 750
 751         async = container_of(work, struct  async_submit_bio, work);
 752         inode = async->inode;
 753
 754         /* If an error occurred we just want to clean up the bio and move on */
 755         if (async->status) {
 756                 async->bio->bi_status = async->status;
 757                 bio_endio(async->bio);
 758                 return;
 759         }
 760
 761         /*
 762          * All of the bios that pass through here are from async helpers.
 763          * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
 764          * This changes nothing when cgroups aren't in use.
 765          */
 766         async->bio->bi_opf |= REQ_CGROUP_PUNT;
 767         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
 768         if (ret) {
 769                 async->bio->bi_status = ret;
 770                 bio_endio(async->bio);
 771         }
 772 }
 773
 774 static void run_one_async_free(struct btrfs_work *work)
 775 {
 776         struct async_submit_bio *async;
 777
 778         async = container_of(work, struct  async_submit_bio, work);
 779         kfree(async);
 780 }
 781
 782 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
 783                                  int mirror_num, unsigned long bio_flags,
 784                                  u64 dio_file_offset,
 785                                  extent_submit_bio_start_t *submit_bio_start)
 786 {
 787         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
 788         struct async_submit_bio *async;
 789
 790         async = kmalloc(sizeof(*async), GFP_NOFS);
 791         if (!async)
 792                 return BLK_STS_RESOURCE;
 793
 794         async->inode = inode;
 795         async->bio = bio;
 796         async->mirror_num = mirror_num;
 797         async->submit_bio_start = submit_bio_start;
 798
 799         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 800                         run_one_async_free);
 801
 802         async->dio_file_offset = dio_file_offset;
 803
 804         async->status = 0;
 805
 806         if (op_is_sync(bio->bi_opf))
 807                 btrfs_set_work_high_priority(&async->work);
 808
 809         btrfs_queue_work(fs_info->workers, &async->work);
 810         return 0;
 811 }
 812
 813 static blk_status_t btree_csum_one_bio(struct bio *bio)
 814 {
 815         struct bio_vec *bvec;
 816         struct btrfs_root *root;
 817         int ret = 0;
 818         struct bvec_iter_all iter_all;
 819
 820         ASSERT(!bio_flagged(bio, BIO_CLONED));
 821         bio_for_each_segment_all(bvec, bio, iter_all) {
 822                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 823                 ret = csum_dirty_buffer(root->fs_info, bvec);
 824                 if (ret)
 825                         break;
 826         }
 827
 828         return errno_to_blk_status(ret);
 829 }
 830
 831 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
 832                                            u64 dio_file_offset)
 833 {
 834         /*
 835          * when we're called for a write, we're already in the async
 836          * submission context.  Just jump into btrfs_map_bio
 837          */
 838         return btree_csum_one_bio(bio);
 839 }
 840
 841 static bool should_async_write(struct btrfs_fs_info *fs_info,
 842                              struct btrfs_inode *bi)
 843 {
 844         if (btrfs_is_zoned(fs_info))
 845                 return false;
 846         if (atomic_read(&bi->sync_writers))
 847                 return false;
 848         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 849                 return false;
 850         return true;
 851 }
 852
 853 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
 854                                        int mirror_num, unsigned long bio_flags)
 855 {
 856         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 857         blk_status_t ret;
 858
 859         if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
 860                 /*
 861                  * called for a read, do the setup so that checksum validation
 862                  * can happen in the async kernel threads
 863                  */
 864                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 865                                           BTRFS_WQ_ENDIO_METADATA);
 866                 if (ret)
 867                         goto out_w_error;
 868                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 869         } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
 870                 ret = btree_csum_one_bio(bio);
 871                 if (ret)
 872                         goto out_w_error;
 873                 ret = btrfs_map_bio(fs_info, bio, mirror_num);
 874         } else {
 875                 /*
 876                  * kthread helpers are used to submit writes so that
 877                  * checksumming can happen in parallel across all CPUs
 878                  */
 879                 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
 880                                           0, btree_submit_bio_start);
 881         }
 882
 883         if (ret)
 884                 goto out_w_error;
 885         return 0;
 886
 887 out_w_error:
 888         bio->bi_status = ret;
 889         bio_endio(bio);
 890         return ret;
 891 }
 892
 893 #ifdef CONFIG_MIGRATION
 894 static int btree_migratepage(struct address_space *mapping,
 895                         struct page *newpage, struct page *page,
 896                         enum migrate_mode mode)
 897 {
 898         /*
 899          * we can't safely write a btree page from here,
 900          * we haven't done the locking hook
 901          */
 902         if (PageDirty(page))
 903                 return -EAGAIN;
 904         /*
 905          * Buffers may be managed in a filesystem specific way.
 906          * We must have no buffers or drop them.
 907          */
 908         if (page_has_private(page) &&
 909             !try_to_release_page(page, GFP_KERNEL))
 910                 return -EAGAIN;
 911         return migrate_page(mapping, newpage, page, mode);
 912 }
 913 #endif
 914
 915
 916 static int btree_writepages(struct address_space *mapping,
 917                             struct writeback_control *wbc)
 918 {
 919         struct btrfs_fs_info *fs_info;
 920         int ret;
 921
 922         if (wbc->sync_mode == WB_SYNC_NONE) {
 923
 924                 if (wbc->for_kupdate)
 925                         return 0;
 926
 927                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 928                 /* this is a bit racy, but that's ok */
 929                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 930                                              BTRFS_DIRTY_METADATA_THRESH,
 931                                              fs_info->dirty_metadata_batch);
 932                 if (ret < 0)
 933                         return 0;
 934         }
 935         return btree_write_cache_pages(mapping, wbc);
 936 }
 937
 938 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 939 {
 940         if (PageWriteback(page) || PageDirty(page))
 941                 return 0;
 942
 943         return try_release_extent_buffer(page);
 944 }
 945
 946 static void btree_invalidatepage(struct page *page, unsigned int offset,
 947                                  unsigned int length)
 948 {
 949         struct extent_io_tree *tree;
 950         tree = &BTRFS_I(page->mapping->host)->io_tree;
 951         extent_invalidatepage(tree, page, offset);
 952         btree_releasepage(page, GFP_NOFS);
 953         if (PagePrivate(page)) {
 954                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 955                            "page private not zero on page %llu",
 956                            (unsigned long long)page_offset(page));
 957                 detach_page_private(page);
 958         }
 959 }
 960
 961 static int btree_set_page_dirty(struct page *page)
 962 {
 963 #ifdef DEBUG
 964         struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
 965         struct btrfs_subpage *subpage;
 966         struct extent_buffer *eb;
 967         int cur_bit = 0;
 968         u64 page_start = page_offset(page);
 969
 970         if (fs_info->sectorsize == PAGE_SIZE) {
 971                 BUG_ON(!PagePrivate(page));
 972                 eb = (struct extent_buffer *)page->private;
 973                 BUG_ON(!eb);
 974                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 975                 BUG_ON(!atomic_read(&eb->refs));
 976                 btrfs_assert_tree_locked(eb);
 977                 return __set_page_dirty_nobuffers(page);
 978         }
 979         ASSERT(PagePrivate(page) && page->private);
 980         subpage = (struct btrfs_subpage *)page->private;
 981
 982         ASSERT(subpage->dirty_bitmap);
 983         while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
 984                 unsigned long flags;
 985                 u64 cur;
 986                 u16 tmp = (1 << cur_bit);
 987
 988                 spin_lock_irqsave(&subpage->lock, flags);
 989                 if (!(tmp & subpage->dirty_bitmap)) {
 990                         spin_unlock_irqrestore(&subpage->lock, flags);
 991                         cur_bit++;
 992                         continue;
 993                 }
 994                 spin_unlock_irqrestore(&subpage->lock, flags);
 995                 cur = page_start + cur_bit * fs_info->sectorsize;
 996
 997                 eb = find_extent_buffer(fs_info, cur);
 998                 ASSERT(eb);
 999                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1000                 ASSERT(atomic_read(&eb->refs));
1001                 btrfs_assert_tree_locked(eb);
1002                 free_extent_buffer(eb);
1003
1004                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1005         }
1006 #endif
1007         return __set_page_dirty_nobuffers(page);
1008 }
1009
1010 static const struct address_space_operations btree_aops = {
1011         .writepages     = btree_writepages,
1012         .releasepage    = btree_releasepage,
1013         .invalidatepage = btree_invalidatepage,
1014 #ifdef CONFIG_MIGRATION
1015         .migratepage    = btree_migratepage,
1016 #endif
1017         .set_page_dirty = btree_set_page_dirty,
1018 };
1019
1020 struct extent_buffer *btrfs_find_create_tree_block(
1021                                                 struct btrfs_fs_info *fs_info,
1022                                                 u64 bytenr, u64 owner_root,
1023                                                 int level)
1024 {
1025         if (btrfs_is_testing(fs_info))
1026                 return alloc_test_extent_buffer(fs_info, bytenr);
1027         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1028 }
1029
1030 /*
1031  * Read tree block at logical address @bytenr and do variant basic but critical
1032  * verification.
1033  *
1034  * @owner_root:         the objectid of the root owner for this block.
1035  * @parent_transid:     expected transid of this tree block, skip check if 0
1036  * @level:              expected level, mandatory check
1037  * @first_key:          expected key in slot 0, skip check if NULL
1038  */
1039 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1040                                       u64 owner_root, u64 parent_transid,
1041                                       int level, struct btrfs_key *first_key)
1042 {
1043         struct extent_buffer *buf = NULL;
1044         int ret;
1045
1046         buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1047         if (IS_ERR(buf))
1048                 return buf;
1049
1050         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1051                                              level, first_key);
1052         if (ret) {
1053                 free_extent_buffer_stale(buf);
1054                 return ERR_PTR(ret);
1055         }
1056         return buf;
1057
1058 }
1059
1060 void btrfs_clean_tree_block(struct extent_buffer *buf)
1061 {
1062         struct btrfs_fs_info *fs_info = buf->fs_info;
1063         if (btrfs_header_generation(buf) ==
1064             fs_info->running_transaction->transid) {
1065                 btrfs_assert_tree_locked(buf);
1066
1067                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1068                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1069                                                  -buf->len,
1070                                                  fs_info->dirty_metadata_batch);
1071                         clear_extent_buffer_dirty(buf);
1072                 }
1073         }
1074 }
1075
1076 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1077                          u64 objectid)
1078 {
1079         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1080         root->fs_info = fs_info;
1081         root->node = NULL;
1082         root->commit_root = NULL;
1083         root->state = 0;
1084         root->orphan_cleanup_state = 0;
1085
1086         root->last_trans = 0;
1087         root->free_objectid = 0;
1088         root->nr_delalloc_inodes = 0;
1089         root->nr_ordered_extents = 0;
1090         root->inode_tree = RB_ROOT;
1091         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1092         root->block_rsv = NULL;
1093
1094         INIT_LIST_HEAD(&root->dirty_list);
1095         INIT_LIST_HEAD(&root->root_list);
1096         INIT_LIST_HEAD(&root->delalloc_inodes);
1097         INIT_LIST_HEAD(&root->delalloc_root);
1098         INIT_LIST_HEAD(&root->ordered_extents);
1099         INIT_LIST_HEAD(&root->ordered_root);
1100         INIT_LIST_HEAD(&root->reloc_dirty_list);
1101         INIT_LIST_HEAD(&root->logged_list[0]);
1102         INIT_LIST_HEAD(&root->logged_list[1]);
1103         spin_lock_init(&root->inode_lock);
1104         spin_lock_init(&root->delalloc_lock);
1105         spin_lock_init(&root->ordered_extent_lock);
1106         spin_lock_init(&root->accounting_lock);
1107         spin_lock_init(&root->log_extents_lock[0]);
1108         spin_lock_init(&root->log_extents_lock[1]);
1109         spin_lock_init(&root->qgroup_meta_rsv_lock);
1110         mutex_init(&root->objectid_mutex);
1111         mutex_init(&root->log_mutex);
1112         mutex_init(&root->ordered_extent_mutex);
1113         mutex_init(&root->delalloc_mutex);
1114         init_waitqueue_head(&root->qgroup_flush_wait);
1115         init_waitqueue_head(&root->log_writer_wait);
1116         init_waitqueue_head(&root->log_commit_wait[0]);
1117         init_waitqueue_head(&root->log_commit_wait[1]);
1118         INIT_LIST_HEAD(&root->log_ctxs[0]);
1119         INIT_LIST_HEAD(&root->log_ctxs[1]);
1120         atomic_set(&root->log_commit[0], 0);
1121         atomic_set(&root->log_commit[1], 0);
1122         atomic_set(&root->log_writers, 0);
1123         atomic_set(&root->log_batch, 0);
1124         refcount_set(&root->refs, 1);
1125         atomic_set(&root->snapshot_force_cow, 0);
1126         atomic_set(&root->nr_swapfiles, 0);
1127         root->log_transid = 0;
1128         root->log_transid_committed = -1;
1129         root->last_log_commit = 0;
1130         if (!dummy) {
1131                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1132                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1133                 extent_io_tree_init(fs_info, &root->log_csum_range,
1134                                     IO_TREE_LOG_CSUM_RANGE, NULL);
1135         }
1136
1137         memset(&root->root_key, 0, sizeof(root->root_key));
1138         memset(&root->root_item, 0, sizeof(root->root_item));
1139         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1140         root->root_key.objectid = objectid;
1141         root->anon_dev = 0;
1142
1143         spin_lock_init(&root->root_item_lock);
1144         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1145 #ifdef CONFIG_BTRFS_DEBUG
1146         INIT_LIST_HEAD(&root->leak_list);
1147         spin_lock(&fs_info->fs_roots_radix_lock);
1148         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1149         spin_unlock(&fs_info->fs_roots_radix_lock);
1150 #endif
1151 }
1152
1153 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1154                                            u64 objectid, gfp_t flags)
1155 {
1156         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1157         if (root)
1158                 __setup_root(root, fs_info, objectid);
1159         return root;
1160 }
1161
1162 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1163 /* Should only be used by the testing infrastructure */
1164 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1165 {
1166         struct btrfs_root *root;
1167
1168         if (!fs_info)
1169                 return ERR_PTR(-EINVAL);
1170
1171         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1172         if (!root)
1173                 return ERR_PTR(-ENOMEM);
1174
1175         /* We don't use the stripesize in selftest, set it as sectorsize */
1176         root->alloc_bytenr = 0;
1177
1178         return root;
1179 }
1180 #endif
1181
1182 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1183                                      u64 objectid)
1184 {
1185         struct btrfs_fs_info *fs_info = trans->fs_info;
1186         struct extent_buffer *leaf;
1187         struct btrfs_root *tree_root = fs_info->tree_root;
1188         struct btrfs_root *root;
1189         struct btrfs_key key;
1190         unsigned int nofs_flag;
1191         int ret = 0;
1192
1193         /*
1194          * We're holding a transaction handle, so use a NOFS memory allocation
1195          * context to avoid deadlock if reclaim happens.
1196          */
1197         nofs_flag = memalloc_nofs_save();
1198         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1199         memalloc_nofs_restore(nofs_flag);
1200         if (!root)
1201                 return ERR_PTR(-ENOMEM);
1202
1203         root->root_key.objectid = objectid;
1204         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1205         root->root_key.offset = 0;
1206
1207         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1208                                       BTRFS_NESTING_NORMAL);
1209         if (IS_ERR(leaf)) {
1210                 ret = PTR_ERR(leaf);
1211                 leaf = NULL;
1212                 goto fail_unlock;
1213         }
1214
1215         root->node = leaf;
1216         btrfs_mark_buffer_dirty(leaf);
1217
1218         root->commit_root = btrfs_root_node(root);
1219         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1220
1221         btrfs_set_root_flags(&root->root_item, 0);
1222         btrfs_set_root_limit(&root->root_item, 0);
1223         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1224         btrfs_set_root_generation(&root->root_item, trans->transid);
1225         btrfs_set_root_level(&root->root_item, 0);
1226         btrfs_set_root_refs(&root->root_item, 1);
1227         btrfs_set_root_used(&root->root_item, leaf->len);
1228         btrfs_set_root_last_snapshot(&root->root_item, 0);
1229         btrfs_set_root_dirid(&root->root_item, 0);
1230         if (is_fstree(objectid))
1231                 generate_random_guid(root->root_item.uuid);
1232         else
1233                 export_guid(root->root_item.uuid, &guid_null);
1234         btrfs_set_root_drop_level(&root->root_item, 0);
1235
1236         btrfs_tree_unlock(leaf);
1237
1238         key.objectid = objectid;
1239         key.type = BTRFS_ROOT_ITEM_KEY;
1240         key.offset = 0;
1241         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1242         if (ret)
1243                 goto fail;
1244
1245         return root;
1246
1247 fail_unlock:
1248         if (leaf)
1249                 btrfs_tree_unlock(leaf);
1250 fail:
1251         btrfs_put_root(root);
1252
1253         return ERR_PTR(ret);
1254 }
1255
1256 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1257                                          struct btrfs_fs_info *fs_info)
1258 {
1259         struct btrfs_root *root;
1260
1261         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1262         if (!root)
1263                 return ERR_PTR(-ENOMEM);
1264
1265         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1266         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1267         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1268
1269         return root;
1270 }
1271
1272 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1273                               struct btrfs_root *root)
1274 {
1275         struct extent_buffer *leaf;
1276
1277         /*
1278          * DON'T set SHAREABLE bit for log trees.
1279          *
1280          * Log trees are not exposed to user space thus can't be snapshotted,
1281          * and they go away before a real commit is actually done.
1282          *
1283          * They do store pointers to file data extents, and those reference
1284          * counts still get updated (along with back refs to the log tree).
1285          */
1286
1287         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1288                         NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1289         if (IS_ERR(leaf))
1290                 return PTR_ERR(leaf);
1291
1292         root->node = leaf;
1293
1294         btrfs_mark_buffer_dirty(root->node);
1295         btrfs_tree_unlock(root->node);
1296
1297         return 0;
1298 }
1299
1300 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1301                              struct btrfs_fs_info *fs_info)
1302 {
1303         struct btrfs_root *log_root;
1304
1305         log_root = alloc_log_tree(trans, fs_info);
1306         if (IS_ERR(log_root))
1307                 return PTR_ERR(log_root);
1308
1309         if (!btrfs_is_zoned(fs_info)) {
1310                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1311
1312                 if (ret) {
1313                         btrfs_put_root(log_root);
1314                         return ret;
1315                 }
1316         }
1317
1318         WARN_ON(fs_info->log_root_tree);
1319         fs_info->log_root_tree = log_root;
1320         return 0;
1321 }
1322
1323 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1324                        struct btrfs_root *root)
1325 {
1326         struct btrfs_fs_info *fs_info = root->fs_info;
1327         struct btrfs_root *log_root;
1328         struct btrfs_inode_item *inode_item;
1329         int ret;
1330
1331         log_root = alloc_log_tree(trans, fs_info);
1332         if (IS_ERR(log_root))
1333                 return PTR_ERR(log_root);
1334
1335         ret = btrfs_alloc_log_tree_node(trans, log_root);
1336         if (ret) {
1337                 btrfs_put_root(log_root);
1338                 return ret;
1339         }
1340
1341         log_root->last_trans = trans->transid;
1342         log_root->root_key.offset = root->root_key.objectid;
1343
1344         inode_item = &log_root->root_item.inode;
1345         btrfs_set_stack_inode_generation(inode_item, 1);
1346         btrfs_set_stack_inode_size(inode_item, 3);
1347         btrfs_set_stack_inode_nlink(inode_item, 1);
1348         btrfs_set_stack_inode_nbytes(inode_item,
1349                                      fs_info->nodesize);
1350         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1351
1352         btrfs_set_root_node(&log_root->root_item, log_root->node);
1353
1354         WARN_ON(root->log_root);
1355         root->log_root = log_root;
1356         root->log_transid = 0;
1357         root->log_transid_committed = -1;
1358         root->last_log_commit = 0;
1359         return 0;
1360 }
1361
1362 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1363                                               struct btrfs_path *path,
1364                                               struct btrfs_key *key)
1365 {
1366         struct btrfs_root *root;
1367         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1368         u64 generation;
1369         int ret;
1370         int level;
1371
1372         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1373         if (!root)
1374                 return ERR_PTR(-ENOMEM);
1375
1376         ret = btrfs_find_root(tree_root, key, path,
1377                               &root->root_item, &root->root_key);
1378         if (ret) {
1379                 if (ret > 0)
1380                         ret = -ENOENT;
1381                 goto fail;
1382         }
1383
1384         generation = btrfs_root_generation(&root->root_item);
1385         level = btrfs_root_level(&root->root_item);
1386         root->node = read_tree_block(fs_info,
1387                                      btrfs_root_bytenr(&root->root_item),
1388                                      key->objectid, generation, level, NULL);
1389         if (IS_ERR(root->node)) {
1390                 ret = PTR_ERR(root->node);
1391                 root->node = NULL;
1392                 goto fail;
1393         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1394                 ret = -EIO;
1395                 goto fail;
1396         }
1397         root->commit_root = btrfs_root_node(root);
1398         return root;
1399 fail:
1400         btrfs_put_root(root);
1401         return ERR_PTR(ret);
1402 }
1403
1404 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1405                                         struct btrfs_key *key)
1406 {
1407         struct btrfs_root *root;
1408         struct btrfs_path *path;
1409
1410         path = btrfs_alloc_path();
1411         if (!path)
1412                 return ERR_PTR(-ENOMEM);
1413         root = read_tree_root_path(tree_root, path, key);
1414         btrfs_free_path(path);
1415
1416         return root;
1417 }
1418
1419 /*
1420  * Initialize subvolume root in-memory structure
1421  *
1422  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1423  */
1424 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1425 {
1426         int ret;
1427         unsigned int nofs_flag;
1428
1429         /*
1430          * We might be called under a transaction (e.g. indirect backref
1431          * resolution) which could deadlock if it triggers memory reclaim
1432          */
1433         nofs_flag = memalloc_nofs_save();
1434         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1435         memalloc_nofs_restore(nofs_flag);
1436         if (ret)
1437                 goto fail;
1438
1439         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1440             !btrfs_is_data_reloc_root(root)) {
1441                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1442                 btrfs_check_and_init_root_item(&root->root_item);
1443         }
1444
1445         /*
1446          * Don't assign anonymous block device to roots that are not exposed to
1447          * userspace, the id pool is limited to 1M
1448          */
1449         if (is_fstree(root->root_key.objectid) &&
1450             btrfs_root_refs(&root->root_item) > 0) {
1451                 if (!anon_dev) {
1452                         ret = get_anon_bdev(&root->anon_dev);
1453                         if (ret)
1454                                 goto fail;
1455                 } else {
1456                         root->anon_dev = anon_dev;
1457                 }
1458         }
1459
1460         mutex_lock(&root->objectid_mutex);
1461         ret = btrfs_init_root_free_objectid(root);
1462         if (ret) {
1463                 mutex_unlock(&root->objectid_mutex);
1464                 goto fail;
1465         }
1466
1467         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1468
1469         mutex_unlock(&root->objectid_mutex);
1470
1471         return 0;
1472 fail:
1473         /* The caller is responsible to call btrfs_free_fs_root */
1474         return ret;
1475 }
1476
1477 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1478                                                u64 root_id)
1479 {
1480         struct btrfs_root *root;
1481
1482         spin_lock(&fs_info->fs_roots_radix_lock);
1483         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1484                                  (unsigned long)root_id);
1485         if (root)
1486                 root = btrfs_grab_root(root);
1487         spin_unlock(&fs_info->fs_roots_radix_lock);
1488         return root;
1489 }
1490
1491 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1492                                                 u64 objectid)
1493 {
1494         if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1495                 return btrfs_grab_root(fs_info->tree_root);
1496         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1497                 return btrfs_grab_root(fs_info->extent_root);
1498         if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1499                 return btrfs_grab_root(fs_info->chunk_root);
1500         if (objectid == BTRFS_DEV_TREE_OBJECTID)
1501                 return btrfs_grab_root(fs_info->dev_root);
1502         if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1503                 return btrfs_grab_root(fs_info->csum_root);
1504         if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1505                 return btrfs_grab_root(fs_info->quota_root) ?
1506                         fs_info->quota_root : ERR_PTR(-ENOENT);
1507         if (objectid == BTRFS_UUID_TREE_OBJECTID)
1508                 return btrfs_grab_root(fs_info->uuid_root) ?
1509                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1510         if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1511                 return btrfs_grab_root(fs_info->free_space_root) ?
1512                         fs_info->free_space_root : ERR_PTR(-ENOENT);
1513         return NULL;
1514 }
1515
1516 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1517                          struct btrfs_root *root)
1518 {
1519         int ret;
1520
1521         ret = radix_tree_preload(GFP_NOFS);
1522         if (ret)
1523                 return ret;
1524
1525         spin_lock(&fs_info->fs_roots_radix_lock);
1526         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1527                                 (unsigned long)root->root_key.objectid,
1528                                 root);
1529         if (ret == 0) {
1530                 btrfs_grab_root(root);
1531                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1532         }
1533         spin_unlock(&fs_info->fs_roots_radix_lock);
1534         radix_tree_preload_end();
1535
1536         return ret;
1537 }
1538
1539 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1540 {
1541 #ifdef CONFIG_BTRFS_DEBUG
1542         struct btrfs_root *root;
1543
1544         while (!list_empty(&fs_info->allocated_roots)) {
1545                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1546
1547                 root = list_first_entry(&fs_info->allocated_roots,
1548                                         struct btrfs_root, leak_list);
1549                 btrfs_err(fs_info, "leaked root %s refcount %d",
1550                           btrfs_root_name(&root->root_key, buf),
1551                           refcount_read(&root->refs));
1552                 while (refcount_read(&root->refs) > 1)
1553                         btrfs_put_root(root);
1554                 btrfs_put_root(root);
1555         }
1556 #endif
1557 }
1558
1559 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1560 {
1561         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1562         percpu_counter_destroy(&fs_info->delalloc_bytes);
1563         percpu_counter_destroy(&fs_info->ordered_bytes);
1564         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1565         btrfs_free_csum_hash(fs_info);
1566         btrfs_free_stripe_hash_table(fs_info);
1567         btrfs_free_ref_cache(fs_info);
1568         kfree(fs_info->balance_ctl);
1569         kfree(fs_info->delayed_root);
1570         btrfs_put_root(fs_info->extent_root);
1571         btrfs_put_root(fs_info->tree_root);
1572         btrfs_put_root(fs_info->chunk_root);
1573         btrfs_put_root(fs_info->dev_root);
1574         btrfs_put_root(fs_info->csum_root);
1575         btrfs_put_root(fs_info->quota_root);
1576         btrfs_put_root(fs_info->uuid_root);
1577         btrfs_put_root(fs_info->free_space_root);
1578         btrfs_put_root(fs_info->fs_root);
1579         btrfs_put_root(fs_info->data_reloc_root);
1580         btrfs_check_leaked_roots(fs_info);
1581         btrfs_extent_buffer_leak_debug_check(fs_info);
1582         kfree(fs_info->super_copy);
1583         kfree(fs_info->super_for_commit);
1584         kvfree(fs_info);
1585 }
1586
1587
1588 /*
1589  * Get an in-memory reference of a root structure.
1590  *
1591  * For essential trees like root/extent tree, we grab it from fs_info directly.
1592  * For subvolume trees, we check the cached filesystem roots first. If not
1593  * found, then read it from disk and add it to cached fs roots.
1594  *
1595  * Caller should release the root by calling btrfs_put_root() after the usage.
1596  *
1597  * NOTE: Reloc and log trees can't be read by this function as they share the
1598  *       same root objectid.
1599  *
1600  * @objectid:   root id
1601  * @anon_dev:   preallocated anonymous block device number for new roots,
1602  *              pass 0 for new allocation.
1603  * @check_ref:  whether to check root item references, If true, return -ENOENT
1604  *              for orphan roots
1605  */
1606 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1607                                              u64 objectid, dev_t anon_dev,
1608                                              bool check_ref)
1609 {
1610         struct btrfs_root *root;
1611         struct btrfs_path *path;
1612         struct btrfs_key key;
1613         int ret;
1614
1615         root = btrfs_get_global_root(fs_info, objectid);
1616         if (root)
1617                 return root;
1618 again:
1619         root = btrfs_lookup_fs_root(fs_info, objectid);
1620         if (root) {
1621                 /* Shouldn't get preallocated anon_dev for cached roots */
1622                 ASSERT(!anon_dev);
1623                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1624                         btrfs_put_root(root);
1625                         return ERR_PTR(-ENOENT);
1626                 }
1627                 return root;
1628         }
1629
1630         key.objectid = objectid;
1631         key.type = BTRFS_ROOT_ITEM_KEY;
1632         key.offset = (u64)-1;
1633         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1634         if (IS_ERR(root))
1635                 return root;
1636
1637         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1638                 ret = -ENOENT;
1639                 goto fail;
1640         }
1641
1642         ret = btrfs_init_fs_root(root, anon_dev);
1643         if (ret)
1644                 goto fail;
1645
1646         path = btrfs_alloc_path();
1647         if (!path) {
1648                 ret = -ENOMEM;
1649                 goto fail;
1650         }
1651         key.objectid = BTRFS_ORPHAN_OBJECTID;
1652         key.type = BTRFS_ORPHAN_ITEM_KEY;
1653         key.offset = objectid;
1654
1655         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1656         btrfs_free_path(path);
1657         if (ret < 0)
1658                 goto fail;
1659         if (ret == 0)
1660                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1661
1662         ret = btrfs_insert_fs_root(fs_info, root);
1663         if (ret) {
1664                 if (ret == -EEXIST) {
1665                         btrfs_put_root(root);
1666                         goto again;
1667                 }
1668                 goto fail;
1669         }
1670         return root;
1671 fail:
1672         /*
1673          * If our caller provided us an anonymous device, then it's his
1674          * responsability to free it in case we fail. So we have to set our
1675          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1676          * and once again by our caller.
1677          */
1678         if (anon_dev)
1679                 root->anon_dev = 0;
1680         btrfs_put_root(root);
1681         return ERR_PTR(ret);
1682 }
1683
1684 /*
1685  * Get in-memory reference of a root structure
1686  *
1687  * @objectid:   tree objectid
1688  * @check_ref:  if set, verify that the tree exists and the item has at least
1689  *              one reference
1690  */
1691 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1692                                      u64 objectid, bool check_ref)
1693 {
1694         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1695 }
1696
1697 /*
1698  * Get in-memory reference of a root structure, created as new, optionally pass
1699  * the anonymous block device id
1700  *
1701  * @objectid:   tree objectid
1702  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1703  *              parameter value
1704  */
1705 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1706                                          u64 objectid, dev_t anon_dev)
1707 {
1708         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1709 }
1710
1711 /*
1712  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1713  * @fs_info:    the fs_info
1714  * @objectid:   the objectid we need to lookup
1715  *
1716  * This is exclusively used for backref walking, and exists specifically because
1717  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1718  * creation time, which means we may have to read the tree_root in order to look
1719  * up a fs root that is not in memory.  If the root is not in memory we will
1720  * read the tree root commit root and look up the fs root from there.  This is a
1721  * temporary root, it will not be inserted into the radix tree as it doesn't
1722  * have the most uptodate information, it'll simply be discarded once the
1723  * backref code is finished using the root.
1724  */
1725 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1726                                                  struct btrfs_path *path,
1727                                                  u64 objectid)
1728 {
1729         struct btrfs_root *root;
1730         struct btrfs_key key;
1731
1732         ASSERT(path->search_commit_root && path->skip_locking);
1733
1734         /*
1735          * This can return -ENOENT if we ask for a root that doesn't exist, but
1736          * since this is called via the backref walking code we won't be looking
1737          * up a root that doesn't exist, unless there's corruption.  So if root
1738          * != NULL just return it.
1739          */
1740         root = btrfs_get_global_root(fs_info, objectid);
1741         if (root)
1742                 return root;
1743
1744         root = btrfs_lookup_fs_root(fs_info, objectid);
1745         if (root)
1746                 return root;
1747
1748         key.objectid = objectid;
1749         key.type = BTRFS_ROOT_ITEM_KEY;
1750         key.offset = (u64)-1;
1751         root = read_tree_root_path(fs_info->tree_root, path, &key);
1752         btrfs_release_path(path);
1753
1754         return root;
1755 }
1756
1757 /*
1758  * called by the kthread helper functions to finally call the bio end_io
1759  * functions.  This is where read checksum verification actually happens
1760  */
1761 static void end_workqueue_fn(struct btrfs_work *work)
1762 {
1763         struct bio *bio;
1764         struct btrfs_end_io_wq *end_io_wq;
1765
1766         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1767         bio = end_io_wq->bio;
1768
1769         bio->bi_status = end_io_wq->status;
1770         bio->bi_private = end_io_wq->private;
1771         bio->bi_end_io = end_io_wq->end_io;
1772         bio_endio(bio);
1773         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1774 }
1775
1776 static int cleaner_kthread(void *arg)
1777 {
1778         struct btrfs_root *root = arg;
1779         struct btrfs_fs_info *fs_info = root->fs_info;
1780         int again;
1781
1782         while (1) {
1783                 again = 0;
1784
1785                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1786
1787                 /* Make the cleaner go to sleep early. */
1788                 if (btrfs_need_cleaner_sleep(fs_info))
1789                         goto sleep;
1790
1791                 /*
1792                  * Do not do anything if we might cause open_ctree() to block
1793                  * before we have finished mounting the filesystem.
1794                  */
1795                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1796                         goto sleep;
1797
1798                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1799                         goto sleep;
1800
1801                 /*
1802                  * Avoid the problem that we change the status of the fs
1803                  * during the above check and trylock.
1804                  */
1805                 if (btrfs_need_cleaner_sleep(fs_info)) {
1806                         mutex_unlock(&fs_info->cleaner_mutex);
1807                         goto sleep;
1808                 }
1809
1810                 btrfs_run_delayed_iputs(fs_info);
1811
1812                 again = btrfs_clean_one_deleted_snapshot(root);
1813                 mutex_unlock(&fs_info->cleaner_mutex);
1814
1815                 /*
1816                  * The defragger has dealt with the R/O remount and umount,
1817                  * needn't do anything special here.
1818                  */
1819                 btrfs_run_defrag_inodes(fs_info);
1820
1821                 /*
1822                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1823                  * with relocation (btrfs_relocate_chunk) and relocation
1824                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1825                  * after acquiring fs_info->reclaim_bgs_lock. So we
1826                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1827                  * unused block groups.
1828                  */
1829                 btrfs_delete_unused_bgs(fs_info);
1830
1831                 /*
1832                  * Reclaim block groups in the reclaim_bgs list after we deleted
1833                  * all unused block_groups. This possibly gives us some more free
1834                  * space.
1835                  */
1836                 btrfs_reclaim_bgs(fs_info);
1837 sleep:
1838                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1839                 if (kthread_should_park())
1840                         kthread_parkme();
1841                 if (kthread_should_stop())
1842                         return 0;
1843                 if (!again) {
1844                         set_current_state(TASK_INTERRUPTIBLE);
1845                         schedule();
1846                         __set_current_state(TASK_RUNNING);
1847                 }
1848         }
1849 }
1850
1851 static int transaction_kthread(void *arg)
1852 {
1853         struct btrfs_root *root = arg;
1854         struct btrfs_fs_info *fs_info = root->fs_info;
1855         struct btrfs_trans_handle *trans;
1856         struct btrfs_transaction *cur;
1857         u64 transid;
1858         time64_t delta;
1859         unsigned long delay;
1860         bool cannot_commit;
1861
1862         do {
1863                 cannot_commit = false;
1864                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1865                 mutex_lock(&fs_info->transaction_kthread_mutex);
1866
1867                 spin_lock(&fs_info->trans_lock);
1868                 cur = fs_info->running_transaction;
1869                 if (!cur) {
1870                         spin_unlock(&fs_info->trans_lock);
1871                         goto sleep;
1872                 }
1873
1874                 delta = ktime_get_seconds() - cur->start_time;
1875                 if (cur->state < TRANS_STATE_COMMIT_START &&
1876                     delta < fs_info->commit_interval) {
1877                         spin_unlock(&fs_info->trans_lock);
1878                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1879                         delay = min(delay,
1880                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1881                         goto sleep;
1882                 }
1883                 transid = cur->transid;
1884                 spin_unlock(&fs_info->trans_lock);
1885
1886                 /* If the file system is aborted, this will always fail. */
1887                 trans = btrfs_attach_transaction(root);
1888                 if (IS_ERR(trans)) {
1889                         if (PTR_ERR(trans) != -ENOENT)
1890                                 cannot_commit = true;
1891                         goto sleep;
1892                 }
1893                 if (transid == trans->transid) {
1894                         btrfs_commit_transaction(trans);
1895                 } else {
1896                         btrfs_end_transaction(trans);
1897                 }
1898 sleep:
1899                 wake_up_process(fs_info->cleaner_kthread);
1900                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1901
1902                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1903                                       &fs_info->fs_state)))
1904                         btrfs_cleanup_transaction(fs_info);
1905                 if (!kthread_should_stop() &&
1906                                 (!btrfs_transaction_blocked(fs_info) ||
1907                                  cannot_commit))
1908                         schedule_timeout_interruptible(delay);
1909         } while (!kthread_should_stop());
1910         return 0;
1911 }
1912
1913 /*
1914  * This will find the highest generation in the array of root backups.  The
1915  * index of the highest array is returned, or -EINVAL if we can't find
1916  * anything.
1917  *
1918  * We check to make sure the array is valid by comparing the
1919  * generation of the latest  root in the array with the generation
1920  * in the super block.  If they don't match we pitch it.
1921  */
1922 static int find_newest_super_backup(struct btrfs_fs_info *info)
1923 {
1924         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1925         u64 cur;
1926         struct btrfs_root_backup *root_backup;
1927         int i;
1928
1929         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1930                 root_backup = info->super_copy->super_roots + i;
1931                 cur = btrfs_backup_tree_root_gen(root_backup);
1932                 if (cur == newest_gen)
1933                         return i;
1934         }
1935
1936         return -EINVAL;
1937 }
1938
1939 /*
1940  * copy all the root pointers into the super backup array.
1941  * this will bump the backup pointer by one when it is
1942  * done
1943  */
1944 static void backup_super_roots(struct btrfs_fs_info *info)
1945 {
1946         const int next_backup = info->backup_root_index;
1947         struct btrfs_root_backup *root_backup;
1948
1949         root_backup = info->super_for_commit->super_roots + next_backup;
1950
1951         /*
1952          * make sure all of our padding and empty slots get zero filled
1953          * regardless of which ones we use today
1954          */
1955         memset(root_backup, 0, sizeof(*root_backup));
1956
1957         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1958
1959         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1960         btrfs_set_backup_tree_root_gen(root_backup,
1961                                btrfs_header_generation(info->tree_root->node));
1962
1963         btrfs_set_backup_tree_root_level(root_backup,
1964                                btrfs_header_level(info->tree_root->node));
1965
1966         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1967         btrfs_set_backup_chunk_root_gen(root_backup,
1968                                btrfs_header_generation(info->chunk_root->node));
1969         btrfs_set_backup_chunk_root_level(root_backup,
1970                                btrfs_header_level(info->chunk_root->node));
1971
1972         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1973         btrfs_set_backup_extent_root_gen(root_backup,
1974                                btrfs_header_generation(info->extent_root->node));
1975         btrfs_set_backup_extent_root_level(root_backup,
1976                                btrfs_header_level(info->extent_root->node));
1977
1978         /*
1979          * we might commit during log recovery, which happens before we set
1980          * the fs_root.  Make sure it is valid before we fill it in.
1981          */
1982         if (info->fs_root && info->fs_root->node) {
1983                 btrfs_set_backup_fs_root(root_backup,
1984                                          info->fs_root->node->start);
1985                 btrfs_set_backup_fs_root_gen(root_backup,
1986                                btrfs_header_generation(info->fs_root->node));
1987                 btrfs_set_backup_fs_root_level(root_backup,
1988                                btrfs_header_level(info->fs_root->node));
1989         }
1990
1991         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1992         btrfs_set_backup_dev_root_gen(root_backup,
1993                                btrfs_header_generation(info->dev_root->node));
1994         btrfs_set_backup_dev_root_level(root_backup,
1995                                        btrfs_header_level(info->dev_root->node));
1996
1997         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1998         btrfs_set_backup_csum_root_gen(root_backup,
1999                                btrfs_header_generation(info->csum_root->node));
2000         btrfs_set_backup_csum_root_level(root_backup,
2001                                btrfs_header_level(info->csum_root->node));
2002
2003         btrfs_set_backup_total_bytes(root_backup,
2004                              btrfs_super_total_bytes(info->super_copy));
2005         btrfs_set_backup_bytes_used(root_backup,
2006                              btrfs_super_bytes_used(info->super_copy));
2007         btrfs_set_backup_num_devices(root_backup,
2008                              btrfs_super_num_devices(info->super_copy));
2009
2010         /*
2011          * if we don't copy this out to the super_copy, it won't get remembered
2012          * for the next commit
2013          */
2014         memcpy(&info->super_copy->super_roots,
2015                &info->super_for_commit->super_roots,
2016                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2017 }
2018
2019 /*
2020  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2021  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2022  *
2023  * fs_info - filesystem whose backup roots need to be read
2024  * priority - priority of backup root required
2025  *
2026  * Returns backup root index on success and -EINVAL otherwise.
2027  */
2028 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2029 {
2030         int backup_index = find_newest_super_backup(fs_info);
2031         struct btrfs_super_block *super = fs_info->super_copy;
2032         struct btrfs_root_backup *root_backup;
2033
2034         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2035                 if (priority == 0)
2036                         return backup_index;
2037
2038                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2039                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2040         } else {
2041                 return -EINVAL;
2042         }
2043
2044         root_backup = super->super_roots + backup_index;
2045
2046         btrfs_set_super_generation(super,
2047                                    btrfs_backup_tree_root_gen(root_backup));
2048         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2049         btrfs_set_super_root_level(super,
2050                                    btrfs_backup_tree_root_level(root_backup));
2051         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2052
2053         /*
2054          * Fixme: the total bytes and num_devices need to match or we should
2055          * need a fsck
2056          */
2057         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2058         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2059
2060         return backup_index;
2061 }
2062
2063 /* helper to cleanup workers */
2064 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2065 {
2066         btrfs_destroy_workqueue(fs_info->fixup_workers);
2067         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2068         btrfs_destroy_workqueue(fs_info->workers);
2069         btrfs_destroy_workqueue(fs_info->endio_workers);
2070         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2071         btrfs_destroy_workqueue(fs_info->rmw_workers);
2072         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2074         btrfs_destroy_workqueue(fs_info->delayed_workers);
2075         btrfs_destroy_workqueue(fs_info->caching_workers);
2076         btrfs_destroy_workqueue(fs_info->readahead_workers);
2077         btrfs_destroy_workqueue(fs_info->flush_workers);
2078         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2079         if (fs_info->discard_ctl.discard_workers)
2080                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2081         /*
2082          * Now that all other work queues are destroyed, we can safely destroy
2083          * the queues used for metadata I/O, since tasks from those other work
2084          * queues can do metadata I/O operations.
2085          */
2086         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2087         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2088 }
2089
2090 static void free_root_extent_buffers(struct btrfs_root *root)
2091 {
2092         if (root) {
2093                 free_extent_buffer(root->node);
2094                 free_extent_buffer(root->commit_root);
2095                 root->node = NULL;
2096                 root->commit_root = NULL;
2097         }
2098 }
2099
2100 /* helper to cleanup tree roots */
2101 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2102 {
2103         free_root_extent_buffers(info->tree_root);
2104
2105         free_root_extent_buffers(info->dev_root);
2106         free_root_extent_buffers(info->extent_root);
2107         free_root_extent_buffers(info->csum_root);
2108         free_root_extent_buffers(info->quota_root);
2109         free_root_extent_buffers(info->uuid_root);
2110         free_root_extent_buffers(info->fs_root);
2111         free_root_extent_buffers(info->data_reloc_root);
2112         if (free_chunk_root)
2113                 free_root_extent_buffers(info->chunk_root);
2114         free_root_extent_buffers(info->free_space_root);
2115 }
2116
2117 void btrfs_put_root(struct btrfs_root *root)
2118 {
2119         if (!root)
2120                 return;
2121
2122         if (refcount_dec_and_test(&root->refs)) {
2123                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2124                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2125                 if (root->anon_dev)
2126                         free_anon_bdev(root->anon_dev);
2127                 btrfs_drew_lock_destroy(&root->snapshot_lock);
2128                 free_root_extent_buffers(root);
2129 #ifdef CONFIG_BTRFS_DEBUG
2130                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2131                 list_del_init(&root->leak_list);
2132                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2133 #endif
2134                 kfree(root);
2135         }
2136 }
2137
2138 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2139 {
2140         int ret;
2141         struct btrfs_root *gang[8];
2142         int i;
2143
2144         while (!list_empty(&fs_info->dead_roots)) {
2145                 gang[0] = list_entry(fs_info->dead_roots.next,
2146                                      struct btrfs_root, root_list);
2147                 list_del(&gang[0]->root_list);
2148
2149                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2150                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2151                 btrfs_put_root(gang[0]);
2152         }
2153
2154         while (1) {
2155                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2156                                              (void **)gang, 0,
2157                                              ARRAY_SIZE(gang));
2158                 if (!ret)
2159                         break;
2160                 for (i = 0; i < ret; i++)
2161                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2162         }
2163 }
2164
2165 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2166 {
2167         mutex_init(&fs_info->scrub_lock);
2168         atomic_set(&fs_info->scrubs_running, 0);
2169         atomic_set(&fs_info->scrub_pause_req, 0);
2170         atomic_set(&fs_info->scrubs_paused, 0);
2171         atomic_set(&fs_info->scrub_cancel_req, 0);
2172         init_waitqueue_head(&fs_info->scrub_pause_wait);
2173         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2174 }
2175
2176 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2177 {
2178         spin_lock_init(&fs_info->balance_lock);
2179         mutex_init(&fs_info->balance_mutex);
2180         atomic_set(&fs_info->balance_pause_req, 0);
2181         atomic_set(&fs_info->balance_cancel_req, 0);
2182         fs_info->balance_ctl = NULL;
2183         init_waitqueue_head(&fs_info->balance_wait_q);
2184         atomic_set(&fs_info->reloc_cancel_req, 0);
2185 }
2186
2187 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2188 {
2189         struct inode *inode = fs_info->btree_inode;
2190
2191         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2192         set_nlink(inode, 1);
2193         /*
2194          * we set the i_size on the btree inode to the max possible int.
2195          * the real end of the address space is determined by all of
2196          * the devices in the system
2197          */
2198         inode->i_size = OFFSET_MAX;
2199         inode->i_mapping->a_ops = &btree_aops;
2200
2201         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2202         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2203                             IO_TREE_BTREE_INODE_IO, inode);
2204         BTRFS_I(inode)->io_tree.track_uptodate = false;
2205         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2206
2207         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2208         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2209         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2210         btrfs_insert_inode_hash(inode);
2211 }
2212
2213 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2214 {
2215         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2216         init_rwsem(&fs_info->dev_replace.rwsem);
2217         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2218 }
2219
2220 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2221 {
2222         spin_lock_init(&fs_info->qgroup_lock);
2223         mutex_init(&fs_info->qgroup_ioctl_lock);
2224         fs_info->qgroup_tree = RB_ROOT;
2225         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2226         fs_info->qgroup_seq = 1;
2227         fs_info->qgroup_ulist = NULL;
2228         fs_info->qgroup_rescan_running = false;
2229         mutex_init(&fs_info->qgroup_rescan_lock);
2230 }
2231
2232 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2233                 struct btrfs_fs_devices *fs_devices)
2234 {
2235         u32 max_active = fs_info->thread_pool_size;
2236         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2237
2238         fs_info->workers =
2239                 btrfs_alloc_workqueue(fs_info, "worker",
2240                                       flags | WQ_HIGHPRI, max_active, 16);
2241
2242         fs_info->delalloc_workers =
2243                 btrfs_alloc_workqueue(fs_info, "delalloc",
2244                                       flags, max_active, 2);
2245
2246         fs_info->flush_workers =
2247                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2248                                       flags, max_active, 0);
2249
2250         fs_info->caching_workers =
2251                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2252
2253         fs_info->fixup_workers =
2254                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2255
2256         /*
2257          * endios are largely parallel and should have a very
2258          * low idle thresh
2259          */
2260         fs_info->endio_workers =
2261                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2262         fs_info->endio_meta_workers =
2263                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2264                                       max_active, 4);
2265         fs_info->endio_meta_write_workers =
2266                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2267                                       max_active, 2);
2268         fs_info->endio_raid56_workers =
2269                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2270                                       max_active, 4);
2271         fs_info->rmw_workers =
2272                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2273         fs_info->endio_write_workers =
2274                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2275                                       max_active, 2);
2276         fs_info->endio_freespace_worker =
2277                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2278                                       max_active, 0);
2279         fs_info->delayed_workers =
2280                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2281                                       max_active, 0);
2282         fs_info->readahead_workers =
2283                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2284                                       max_active, 2);
2285         fs_info->qgroup_rescan_workers =
2286                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2287         fs_info->discard_ctl.discard_workers =
2288                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2289
2290         if (!(fs_info->workers && fs_info->delalloc_workers &&
2291               fs_info->flush_workers &&
2292               fs_info->endio_workers && fs_info->endio_meta_workers &&
2293               fs_info->endio_meta_write_workers &&
2294               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2295               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2296               fs_info->caching_workers && fs_info->readahead_workers &&
2297               fs_info->fixup_workers && fs_info->delayed_workers &&
2298               fs_info->qgroup_rescan_workers &&
2299               fs_info->discard_ctl.discard_workers)) {
2300                 return -ENOMEM;
2301         }
2302
2303         return 0;
2304 }
2305
2306 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2307 {
2308         struct crypto_shash *csum_shash;
2309         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2310
2311         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2312
2313         if (IS_ERR(csum_shash)) {
2314                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2315                           csum_driver);
2316                 return PTR_ERR(csum_shash);
2317         }
2318
2319         fs_info->csum_shash = csum_shash;
2320
2321         return 0;
2322 }
2323
2324 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2325                             struct btrfs_fs_devices *fs_devices)
2326 {
2327         int ret;
2328         struct btrfs_root *log_tree_root;
2329         struct btrfs_super_block *disk_super = fs_info->super_copy;
2330         u64 bytenr = btrfs_super_log_root(disk_super);
2331         int level = btrfs_super_log_root_level(disk_super);
2332
2333         if (fs_devices->rw_devices == 0) {
2334                 btrfs_warn(fs_info, "log replay required on RO media");
2335                 return -EIO;
2336         }
2337
2338         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2339                                          GFP_KERNEL);
2340         if (!log_tree_root)
2341                 return -ENOMEM;
2342
2343         log_tree_root->node = read_tree_block(fs_info, bytenr,
2344                                               BTRFS_TREE_LOG_OBJECTID,
2345                                               fs_info->generation + 1, level,
2346                                               NULL);
2347         if (IS_ERR(log_tree_root->node)) {
2348                 btrfs_warn(fs_info, "failed to read log tree");
2349                 ret = PTR_ERR(log_tree_root->node);
2350                 log_tree_root->node = NULL;
2351                 btrfs_put_root(log_tree_root);
2352                 return ret;
2353         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2354                 btrfs_err(fs_info, "failed to read log tree");
2355                 btrfs_put_root(log_tree_root);
2356                 return -EIO;
2357         }
2358         /* returns with log_tree_root freed on success */
2359         ret = btrfs_recover_log_trees(log_tree_root);
2360         if (ret) {
2361                 btrfs_handle_fs_error(fs_info, ret,
2362                                       "Failed to recover log tree");
2363                 btrfs_put_root(log_tree_root);
2364                 return ret;
2365         }
2366
2367         if (sb_rdonly(fs_info->sb)) {
2368                 ret = btrfs_commit_super(fs_info);
2369                 if (ret)
2370                         return ret;
2371         }
2372
2373         return 0;
2374 }
2375
2376 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2377 {
2378         struct btrfs_root *tree_root = fs_info->tree_root;
2379         struct btrfs_root *root;
2380         struct btrfs_key location;
2381         int ret;
2382
2383         BUG_ON(!fs_info->tree_root);
2384
2385         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2386         location.type = BTRFS_ROOT_ITEM_KEY;
2387         location.offset = 0;
2388
2389         root = btrfs_read_tree_root(tree_root, &location);
2390         if (IS_ERR(root)) {
2391                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2392                         ret = PTR_ERR(root);
2393                         goto out;
2394                 }
2395         } else {
2396                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2397                 fs_info->extent_root = root;
2398         }
2399
2400         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2401         root = btrfs_read_tree_root(tree_root, &location);
2402         if (IS_ERR(root)) {
2403                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2404                         ret = PTR_ERR(root);
2405                         goto out;
2406                 }
2407         } else {
2408                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2409                 fs_info->dev_root = root;
2410         }
2411         /* Initialize fs_info for all devices in any case */
2412         ret = btrfs_init_devices_late(fs_info);
2413         if (ret)
2414                 goto out;
2415
2416         /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2417         if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2418                 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2419                 root = btrfs_read_tree_root(tree_root, &location);
2420                 if (IS_ERR(root)) {
2421                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2422                                 ret = PTR_ERR(root);
2423                                 goto out;
2424                         }
2425                 } else {
2426                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2427                         fs_info->csum_root = root;
2428                 }
2429         }
2430
2431         /*
2432          * This tree can share blocks with some other fs tree during relocation
2433          * and we need a proper setup by btrfs_get_fs_root
2434          */
2435         root = btrfs_get_fs_root(tree_root->fs_info,
2436                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2437         if (IS_ERR(root)) {
2438                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2439                         ret = PTR_ERR(root);
2440                         goto out;
2441                 }
2442         } else {
2443                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2444                 fs_info->data_reloc_root = root;
2445         }
2446
2447         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2448         root = btrfs_read_tree_root(tree_root, &location);
2449         if (!IS_ERR(root)) {
2450                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2452                 fs_info->quota_root = root;
2453         }
2454
2455         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2456         root = btrfs_read_tree_root(tree_root, &location);
2457         if (IS_ERR(root)) {
2458                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2459                         ret = PTR_ERR(root);
2460                         if (ret != -ENOENT)
2461                                 goto out;
2462                 }
2463         } else {
2464                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2465                 fs_info->uuid_root = root;
2466         }
2467
2468         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2469                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2470                 root = btrfs_read_tree_root(tree_root, &location);
2471                 if (IS_ERR(root)) {
2472                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2473                                 ret = PTR_ERR(root);
2474                                 goto out;
2475                         }
2476                 }  else {
2477                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2478                         fs_info->free_space_root = root;
2479                 }
2480         }
2481
2482         return 0;
2483 out:
2484         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2485                    location.objectid, ret);
2486         return ret;
2487 }
2488
2489 /*
2490  * Real super block validation
2491  * NOTE: super csum type and incompat features will not be checked here.
2492  *
2493  * @sb:         super block to check
2494  * @mirror_num: the super block number to check its bytenr:
2495  *              0       the primary (1st) sb
2496  *              1, 2    2nd and 3rd backup copy
2497  *             -1       skip bytenr check
2498  */
2499 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2500                          struct btrfs_super_block *sb, int mirror_num)
2501 {
2502         u64 nodesize = btrfs_super_nodesize(sb);
2503         u64 sectorsize = btrfs_super_sectorsize(sb);
2504         int ret = 0;
2505
2506         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2507                 btrfs_err(fs_info, "no valid FS found");
2508                 ret = -EINVAL;
2509         }
2510         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2511                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2512                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2513                 ret = -EINVAL;
2514         }
2515         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2516                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2517                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2518                 ret = -EINVAL;
2519         }
2520         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2521                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2522                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2523                 ret = -EINVAL;
2524         }
2525         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2526                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2527                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2528                 ret = -EINVAL;
2529         }
2530
2531         /*
2532          * Check sectorsize and nodesize first, other check will need it.
2533          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2534          */
2535         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2536             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2537                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2538                 ret = -EINVAL;
2539         }
2540
2541         /*
2542          * For 4K page size, we only support 4K sector size.
2543          * For 64K page size, we support read-write for 64K sector size, and
2544          * read-only for 4K sector size.
2545          */
2546         if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2547             (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2548                                      sectorsize != SZ_64K))) {
2549                 btrfs_err(fs_info,
2550                         "sectorsize %llu not yet supported for page size %lu",
2551                         sectorsize, PAGE_SIZE);
2552                 ret = -EINVAL;
2553         }
2554
2555         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2556             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2557                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2558                 ret = -EINVAL;
2559         }
2560         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2561                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2562                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2563                 ret = -EINVAL;
2564         }
2565
2566         /* Root alignment check */
2567         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2568                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2569                            btrfs_super_root(sb));
2570                 ret = -EINVAL;
2571         }
2572         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2573                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2574                            btrfs_super_chunk_root(sb));
2575                 ret = -EINVAL;
2576         }
2577         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2578                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2579                            btrfs_super_log_root(sb));
2580                 ret = -EINVAL;
2581         }
2582
2583         if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2584                    BTRFS_FSID_SIZE)) {
2585                 btrfs_err(fs_info,
2586                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2587                         fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2588                 ret = -EINVAL;
2589         }
2590
2591         if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2592             memcmp(fs_info->fs_devices->metadata_uuid,
2593                    fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2594                 btrfs_err(fs_info,
2595 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2596                         fs_info->super_copy->metadata_uuid,
2597                         fs_info->fs_devices->metadata_uuid);
2598                 ret = -EINVAL;
2599         }
2600
2601         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2602                    BTRFS_FSID_SIZE) != 0) {
2603                 btrfs_err(fs_info,
2604                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2605                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2606                 ret = -EINVAL;
2607         }
2608
2609         /*
2610          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2611          * done later
2612          */
2613         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2614                 btrfs_err(fs_info, "bytes_used is too small %llu",
2615                           btrfs_super_bytes_used(sb));
2616                 ret = -EINVAL;
2617         }
2618         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2619                 btrfs_err(fs_info, "invalid stripesize %u",
2620                           btrfs_super_stripesize(sb));
2621                 ret = -EINVAL;
2622         }
2623         if (btrfs_super_num_devices(sb) > (1UL << 31))
2624                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2625                            btrfs_super_num_devices(sb));
2626         if (btrfs_super_num_devices(sb) == 0) {
2627                 btrfs_err(fs_info, "number of devices is 0");
2628                 ret = -EINVAL;
2629         }
2630
2631         if (mirror_num >= 0 &&
2632             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2633                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2634                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2635                 ret = -EINVAL;
2636         }
2637
2638         /*
2639          * Obvious sys_chunk_array corruptions, it must hold at least one key
2640          * and one chunk
2641          */
2642         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2643                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2644                           btrfs_super_sys_array_size(sb),
2645                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2646                 ret = -EINVAL;
2647         }
2648         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2649                         + sizeof(struct btrfs_chunk)) {
2650                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2651                           btrfs_super_sys_array_size(sb),
2652                           sizeof(struct btrfs_disk_key)
2653                           + sizeof(struct btrfs_chunk));
2654                 ret = -EINVAL;
2655         }
2656
2657         /*
2658          * The generation is a global counter, we'll trust it more than the others
2659          * but it's still possible that it's the one that's wrong.
2660          */
2661         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2662                 btrfs_warn(fs_info,
2663                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2664                         btrfs_super_generation(sb),
2665                         btrfs_super_chunk_root_generation(sb));
2666         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2667             && btrfs_super_cache_generation(sb) != (u64)-1)
2668                 btrfs_warn(fs_info,
2669                         "suspicious: generation < cache_generation: %llu < %llu",
2670                         btrfs_super_generation(sb),
2671                         btrfs_super_cache_generation(sb));
2672
2673         return ret;
2674 }
2675
2676 /*
2677  * Validation of super block at mount time.
2678  * Some checks already done early at mount time, like csum type and incompat
2679  * flags will be skipped.
2680  */
2681 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2682 {
2683         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2684 }
2685
2686 /*
2687  * Validation of super block at write time.
2688  * Some checks like bytenr check will be skipped as their values will be
2689  * overwritten soon.
2690  * Extra checks like csum type and incompat flags will be done here.
2691  */
2692 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2693                                       struct btrfs_super_block *sb)
2694 {
2695         int ret;
2696
2697         ret = btrfs_validate_super(fs_info, sb, -1);
2698         if (ret < 0)
2699                 goto out;
2700         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2701                 ret = -EUCLEAN;
2702                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2703                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2704                 goto out;
2705         }
2706         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2707                 ret = -EUCLEAN;
2708                 btrfs_err(fs_info,
2709                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2710                           btrfs_super_incompat_flags(sb),
2711                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2712                 goto out;
2713         }
2714 out:
2715         if (ret < 0)
2716                 btrfs_err(fs_info,
2717                 "super block corruption detected before writing it to disk");
2718         return ret;
2719 }
2720
2721 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2722 {
2723         int backup_index = find_newest_super_backup(fs_info);
2724         struct btrfs_super_block *sb = fs_info->super_copy;
2725         struct btrfs_root *tree_root = fs_info->tree_root;
2726         bool handle_error = false;
2727         int ret = 0;
2728         int i;
2729
2730         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2731                 u64 generation;
2732                 int level;
2733
2734                 if (handle_error) {
2735                         if (!IS_ERR(tree_root->node))
2736                                 free_extent_buffer(tree_root->node);
2737                         tree_root->node = NULL;
2738
2739                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2740                                 break;
2741
2742                         free_root_pointers(fs_info, 0);
2743
2744                         /*
2745                          * Don't use the log in recovery mode, it won't be
2746                          * valid
2747                          */
2748                         btrfs_set_super_log_root(sb, 0);
2749
2750                         /* We can't trust the free space cache either */
2751                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2752
2753                         ret = read_backup_root(fs_info, i);
2754                         backup_index = ret;
2755                         if (ret < 0)
2756                                 return ret;
2757                 }
2758                 generation = btrfs_super_generation(sb);
2759                 level = btrfs_super_root_level(sb);
2760                 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2761                                                   BTRFS_ROOT_TREE_OBJECTID,
2762                                                   generation, level, NULL);
2763                 if (IS_ERR(tree_root->node)) {
2764                         handle_error = true;
2765                         ret = PTR_ERR(tree_root->node);
2766                         tree_root->node = NULL;
2767                         btrfs_warn(fs_info, "couldn't read tree root");
2768                         continue;
2769
2770                 } else if (!extent_buffer_uptodate(tree_root->node)) {
2771                         handle_error = true;
2772                         ret = -EIO;
2773                         btrfs_warn(fs_info, "error while reading tree root");
2774                         continue;
2775                 }
2776
2777                 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2778                 tree_root->commit_root = btrfs_root_node(tree_root);
2779                 btrfs_set_root_refs(&tree_root->root_item, 1);
2780
2781                 /*
2782                  * No need to hold btrfs_root::objectid_mutex since the fs
2783                  * hasn't been fully initialised and we are the only user
2784                  */
2785                 ret = btrfs_init_root_free_objectid(tree_root);
2786                 if (ret < 0) {
2787                         handle_error = true;
2788                         continue;
2789                 }
2790
2791                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2792
2793                 ret = btrfs_read_roots(fs_info);
2794                 if (ret < 0) {
2795                         handle_error = true;
2796                         continue;
2797                 }
2798
2799                 /* All successful */
2800                 fs_info->generation = generation;
2801                 fs_info->last_trans_committed = generation;
2802                 fs_info->last_reloc_trans = 0;
2803
2804                 /* Always begin writing backup roots after the one being used */
2805                 if (backup_index < 0) {
2806                         fs_info->backup_root_index = 0;
2807                 } else {
2808                         fs_info->backup_root_index = backup_index + 1;
2809                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2810                 }
2811                 break;
2812         }
2813
2814         return ret;
2815 }
2816
2817 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2818 {
2819         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2820         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2821         INIT_LIST_HEAD(&fs_info->trans_list);
2822         INIT_LIST_HEAD(&fs_info->dead_roots);
2823         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2824         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2825         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2826         spin_lock_init(&fs_info->delalloc_root_lock);
2827         spin_lock_init(&fs_info->trans_lock);
2828         spin_lock_init(&fs_info->fs_roots_radix_lock);
2829         spin_lock_init(&fs_info->delayed_iput_lock);
2830         spin_lock_init(&fs_info->defrag_inodes_lock);
2831         spin_lock_init(&fs_info->super_lock);
2832         spin_lock_init(&fs_info->buffer_lock);
2833         spin_lock_init(&fs_info->unused_bgs_lock);
2834         spin_lock_init(&fs_info->treelog_bg_lock);
2835         spin_lock_init(&fs_info->relocation_bg_lock);
2836         rwlock_init(&fs_info->tree_mod_log_lock);
2837         mutex_init(&fs_info->unused_bg_unpin_mutex);
2838         mutex_init(&fs_info->reclaim_bgs_lock);
2839         mutex_init(&fs_info->reloc_mutex);
2840         mutex_init(&fs_info->delalloc_root_mutex);
2841         mutex_init(&fs_info->zoned_meta_io_lock);
2842         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2843         seqlock_init(&fs_info->profiles_lock);
2844
2845         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2846         INIT_LIST_HEAD(&fs_info->space_info);
2847         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2848         INIT_LIST_HEAD(&fs_info->unused_bgs);
2849         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2850 #ifdef CONFIG_BTRFS_DEBUG
2851         INIT_LIST_HEAD(&fs_info->allocated_roots);
2852         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2853         spin_lock_init(&fs_info->eb_leak_lock);
2854 #endif
2855         extent_map_tree_init(&fs_info->mapping_tree);
2856         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2857                              BTRFS_BLOCK_RSV_GLOBAL);
2858         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2859         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2860         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2861         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2862                              BTRFS_BLOCK_RSV_DELOPS);
2863         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2864                              BTRFS_BLOCK_RSV_DELREFS);
2865
2866         atomic_set(&fs_info->async_delalloc_pages, 0);
2867         atomic_set(&fs_info->defrag_running, 0);
2868         atomic_set(&fs_info->reada_works_cnt, 0);
2869         atomic_set(&fs_info->nr_delayed_iputs, 0);
2870         atomic64_set(&fs_info->tree_mod_seq, 0);
2871         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2872         fs_info->metadata_ratio = 0;
2873         fs_info->defrag_inodes = RB_ROOT;
2874         atomic64_set(&fs_info->free_chunk_space, 0);
2875         fs_info->tree_mod_log = RB_ROOT;
2876         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2877         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2878         /* readahead state */
2879         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2880         spin_lock_init(&fs_info->reada_lock);
2881         btrfs_init_ref_verify(fs_info);
2882
2883         fs_info->thread_pool_size = min_t(unsigned long,
2884                                           num_online_cpus() + 2, 8);
2885
2886         INIT_LIST_HEAD(&fs_info->ordered_roots);
2887         spin_lock_init(&fs_info->ordered_root_lock);
2888
2889         btrfs_init_scrub(fs_info);
2890 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2891         fs_info->check_integrity_print_mask = 0;
2892 #endif
2893         btrfs_init_balance(fs_info);
2894         btrfs_init_async_reclaim_work(fs_info);
2895
2896         spin_lock_init(&fs_info->block_group_cache_lock);
2897         fs_info->block_group_cache_tree = RB_ROOT;
2898         fs_info->first_logical_byte = (u64)-1;
2899
2900         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2901                             IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2902         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2903
2904         mutex_init(&fs_info->ordered_operations_mutex);
2905         mutex_init(&fs_info->tree_log_mutex);
2906         mutex_init(&fs_info->chunk_mutex);
2907         mutex_init(&fs_info->transaction_kthread_mutex);
2908         mutex_init(&fs_info->cleaner_mutex);
2909         mutex_init(&fs_info->ro_block_group_mutex);
2910         init_rwsem(&fs_info->commit_root_sem);
2911         init_rwsem(&fs_info->cleanup_work_sem);
2912         init_rwsem(&fs_info->subvol_sem);
2913         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2914
2915         btrfs_init_dev_replace_locks(fs_info);
2916         btrfs_init_qgroup(fs_info);
2917         btrfs_discard_init(fs_info);
2918
2919         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2920         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2921
2922         init_waitqueue_head(&fs_info->transaction_throttle);
2923         init_waitqueue_head(&fs_info->transaction_wait);
2924         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2925         init_waitqueue_head(&fs_info->async_submit_wait);
2926         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2927
2928         /* Usable values until the real ones are cached from the superblock */
2929         fs_info->nodesize = 4096;
2930         fs_info->sectorsize = 4096;
2931         fs_info->sectorsize_bits = ilog2(4096);
2932         fs_info->stripesize = 4096;
2933
2934         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2935
2936         spin_lock_init(&fs_info->swapfile_pins_lock);
2937         fs_info->swapfile_pins = RB_ROOT;
2938
2939         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2940         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2941 }
2942
2943 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2944 {
2945         int ret;
2946
2947         fs_info->sb = sb;
2948         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2949         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2950
2951         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2952         if (ret)
2953                 return ret;
2954
2955         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2956         if (ret)
2957                 return ret;
2958
2959         fs_info->dirty_metadata_batch = PAGE_SIZE *
2960                                         (1 + ilog2(nr_cpu_ids));
2961
2962         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2963         if (ret)
2964                 return ret;
2965
2966         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2967                         GFP_KERNEL);
2968         if (ret)
2969                 return ret;
2970
2971         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2972                                         GFP_KERNEL);
2973         if (!fs_info->delayed_root)
2974                 return -ENOMEM;
2975         btrfs_init_delayed_root(fs_info->delayed_root);
2976
2977         if (sb_rdonly(sb))
2978                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2979
2980         return btrfs_alloc_stripe_hash_table(fs_info);
2981 }
2982
2983 static int btrfs_uuid_rescan_kthread(void *data)
2984 {
2985         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2986         int ret;
2987
2988         /*
2989          * 1st step is to iterate through the existing UUID tree and
2990          * to delete all entries that contain outdated data.
2991          * 2nd step is to add all missing entries to the UUID tree.
2992          */
2993         ret = btrfs_uuid_tree_iterate(fs_info);
2994         if (ret < 0) {
2995                 if (ret != -EINTR)
2996                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2997                                    ret);
2998                 up(&fs_info->uuid_tree_rescan_sem);
2999                 return ret;
3000         }
3001         return btrfs_uuid_scan_kthread(data);
3002 }
3003
3004 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3005 {
3006         struct task_struct *task;
3007
3008         down(&fs_info->uuid_tree_rescan_sem);
3009         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3010         if (IS_ERR(task)) {
3011                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3012                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3013                 up(&fs_info->uuid_tree_rescan_sem);
3014                 return PTR_ERR(task);
3015         }
3016
3017         return 0;
3018 }
3019
3020 /*
3021  * Some options only have meaning at mount time and shouldn't persist across
3022  * remounts, or be displayed. Clear these at the end of mount and remount
3023  * code paths.
3024  */
3025 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3026 {
3027         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3028         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3029 }
3030
3031 /*
3032  * Mounting logic specific to read-write file systems. Shared by open_ctree
3033  * and btrfs_remount when remounting from read-only to read-write.
3034  */
3035 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3036 {
3037         int ret;
3038         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3039         bool clear_free_space_tree = false;
3040
3041         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3042             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3043                 clear_free_space_tree = true;
3044         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3045                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3046                 btrfs_warn(fs_info, "free space tree is invalid");
3047                 clear_free_space_tree = true;
3048         }
3049
3050         if (clear_free_space_tree) {
3051                 btrfs_info(fs_info, "clearing free space tree");
3052                 ret = btrfs_clear_free_space_tree(fs_info);
3053                 if (ret) {
3054                         btrfs_warn(fs_info,
3055                                    "failed to clear free space tree: %d", ret);
3056                         goto out;
3057                 }
3058         }
3059
3060         /*
3061          * btrfs_find_orphan_roots() is responsible for finding all the dead
3062          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3063          * them into the fs_info->fs_roots_radix tree. This must be done before
3064          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3065          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3066          * item before the root's tree is deleted - this means that if we unmount
3067          * or crash before the deletion completes, on the next mount we will not
3068          * delete what remains of the tree because the orphan item does not
3069          * exists anymore, which is what tells us we have a pending deletion.
3070          */
3071         ret = btrfs_find_orphan_roots(fs_info);
3072         if (ret)
3073                 goto out;
3074
3075         ret = btrfs_cleanup_fs_roots(fs_info);
3076         if (ret)
3077                 goto out;
3078
3079         down_read(&fs_info->cleanup_work_sem);
3080         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3081             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3082                 up_read(&fs_info->cleanup_work_sem);
3083                 goto out;
3084         }
3085         up_read(&fs_info->cleanup_work_sem);
3086
3087         mutex_lock(&fs_info->cleaner_mutex);
3088         ret = btrfs_recover_relocation(fs_info->tree_root);
3089         mutex_unlock(&fs_info->cleaner_mutex);
3090         if (ret < 0) {
3091                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3092                 goto out;
3093         }
3094
3095         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3096             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3097                 btrfs_info(fs_info, "creating free space tree");
3098                 ret = btrfs_create_free_space_tree(fs_info);
3099                 if (ret) {
3100                         btrfs_warn(fs_info,
3101                                 "failed to create free space tree: %d", ret);
3102                         goto out;
3103                 }
3104         }
3105
3106         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3107                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3108                 if (ret)
3109                         goto out;
3110         }
3111
3112         ret = btrfs_resume_balance_async(fs_info);
3113         if (ret)
3114                 goto out;
3115
3116         ret = btrfs_resume_dev_replace_async(fs_info);
3117         if (ret) {
3118                 btrfs_warn(fs_info, "failed to resume dev_replace");
3119                 goto out;
3120         }
3121
3122         btrfs_qgroup_rescan_resume(fs_info);
3123
3124         if (!fs_info->uuid_root) {
3125                 btrfs_info(fs_info, "creating UUID tree");
3126                 ret = btrfs_create_uuid_tree(fs_info);
3127                 if (ret) {
3128                         btrfs_warn(fs_info,
3129                                    "failed to create the UUID tree %d", ret);
3130                         goto out;
3131                 }
3132         }
3133
3134 out:
3135         return ret;
3136 }
3137
3138 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3139                       char *options)
3140 {
3141         u32 sectorsize;
3142         u32 nodesize;
3143         u32 stripesize;
3144         u64 generation;
3145         u64 features;
3146         u16 csum_type;
3147         struct btrfs_super_block *disk_super;
3148         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3149         struct btrfs_root *tree_root;
3150         struct btrfs_root *chunk_root;
3151         int ret;
3152         int err = -EINVAL;
3153         int level;
3154
3155         ret = init_mount_fs_info(fs_info, sb);
3156         if (ret) {
3157                 err = ret;
3158                 goto fail;
3159         }
3160
3161         /* These need to be init'ed before we start creating inodes and such. */
3162         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3163                                      GFP_KERNEL);
3164         fs_info->tree_root = tree_root;
3165         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3166                                       GFP_KERNEL);
3167         fs_info->chunk_root = chunk_root;
3168         if (!tree_root || !chunk_root) {
3169                 err = -ENOMEM;
3170                 goto fail;
3171         }
3172
3173         fs_info->btree_inode = new_inode(sb);
3174         if (!fs_info->btree_inode) {
3175                 err = -ENOMEM;
3176                 goto fail;
3177         }
3178         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3179         btrfs_init_btree_inode(fs_info);
3180
3181         invalidate_bdev(fs_devices->latest_dev->bdev);
3182
3183         /*
3184          * Read super block and check the signature bytes only
3185          */
3186         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3187         if (IS_ERR(disk_super)) {
3188                 err = PTR_ERR(disk_super);
3189                 goto fail_alloc;
3190         }
3191
3192         /*
3193          * Verify the type first, if that or the checksum value are
3194          * corrupted, we'll find out
3195          */
3196         csum_type = btrfs_super_csum_type(disk_super);
3197         if (!btrfs_supported_super_csum(csum_type)) {
3198                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3199                           csum_type);
3200                 err = -EINVAL;
3201                 btrfs_release_disk_super(disk_super);
3202                 goto fail_alloc;
3203         }
3204
3205         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3206
3207         ret = btrfs_init_csum_hash(fs_info, csum_type);
3208         if (ret) {
3209                 err = ret;
3210                 btrfs_release_disk_super(disk_super);
3211                 goto fail_alloc;
3212         }
3213
3214         /*
3215          * We want to check superblock checksum, the type is stored inside.
3216          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3217          */
3218         if (btrfs_check_super_csum(fs_info, disk_super)) {
3219                 btrfs_err(fs_info, "superblock checksum mismatch");
3220                 err = -EINVAL;
3221                 btrfs_release_disk_super(disk_super);
3222                 goto fail_alloc;
3223         }
3224
3225         /*
3226          * super_copy is zeroed at allocation time and we never touch the
3227          * following bytes up to INFO_SIZE, the checksum is calculated from
3228          * the whole block of INFO_SIZE
3229          */
3230         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3231         btrfs_release_disk_super(disk_super);
3232
3233         disk_super = fs_info->super_copy;
3234
3235
3236         features = btrfs_super_flags(disk_super);
3237         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3238                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3239                 btrfs_set_super_flags(disk_super, features);
3240                 btrfs_info(fs_info,
3241                         "found metadata UUID change in progress flag, clearing");
3242         }
3243
3244         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3245                sizeof(*fs_info->super_for_commit));
3246
3247         ret = btrfs_validate_mount_super(fs_info);
3248         if (ret) {
3249                 btrfs_err(fs_info, "superblock contains fatal errors");
3250                 err = -EINVAL;
3251                 goto fail_alloc;
3252         }
3253
3254         if (!btrfs_super_root(disk_super))
3255                 goto fail_alloc;
3256
3257         /* check FS state, whether FS is broken. */
3258         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3259                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3260
3261         /*
3262          * In the long term, we'll store the compression type in the super
3263          * block, and it'll be used for per file compression control.
3264          */
3265         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3266
3267
3268         /* Set up fs_info before parsing mount options */
3269         nodesize = btrfs_super_nodesize(disk_super);
3270         sectorsize = btrfs_super_sectorsize(disk_super);
3271         stripesize = sectorsize;
3272         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3273         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3274
3275         fs_info->nodesize = nodesize;
3276         fs_info->sectorsize = sectorsize;
3277         fs_info->sectorsize_bits = ilog2(sectorsize);
3278         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3279         fs_info->stripesize = stripesize;
3280
3281         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3282         if (ret) {
3283                 err = ret;
3284                 goto fail_alloc;
3285         }
3286
3287         features = btrfs_super_incompat_flags(disk_super) &
3288                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3289         if (features) {
3290                 btrfs_err(fs_info,
3291                     "cannot mount because of unsupported optional features (0x%llx)",
3292                     features);
3293                 err = -EINVAL;
3294                 goto fail_alloc;
3295         }
3296
3297         features = btrfs_super_incompat_flags(disk_super);
3298         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3299         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3300                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3301         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3302                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3303
3304         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3305                 btrfs_info(fs_info, "has skinny extents");
3306
3307         /*
3308          * Flag our filesystem as having big metadata blocks if they are bigger
3309          * than the page size.
3310          */
3311         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3312                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3313                         btrfs_info(fs_info,
3314                                 "flagging fs with big metadata feature");
3315                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3316         }
3317
3318         /*
3319          * mixed block groups end up with duplicate but slightly offset
3320          * extent buffers for the same range.  It leads to corruptions
3321          */
3322         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3323             (sectorsize != nodesize)) {
3324                 btrfs_err(fs_info,
3325 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3326                         nodesize, sectorsize);
3327                 goto fail_alloc;
3328         }
3329
3330         /*
3331          * Needn't use the lock because there is no other task which will
3332          * update the flag.
3333          */
3334         btrfs_set_super_incompat_flags(disk_super, features);
3335
3336         features = btrfs_super_compat_ro_flags(disk_super) &
3337                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3338         if (!sb_rdonly(sb) && features) {
3339                 btrfs_err(fs_info,
3340         "cannot mount read-write because of unsupported optional features (0x%llx)",
3341                        features);
3342                 err = -EINVAL;
3343                 goto fail_alloc;
3344         }
3345
3346         if (sectorsize != PAGE_SIZE) {
3347                 /*
3348                  * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3349                  * going to be deprecated.
3350                  *
3351                  * Force to use v2 cache for subpage case.
3352                  */
3353                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3354                 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3355                         "forcing free space tree for sector size %u with page size %lu",
3356                         sectorsize, PAGE_SIZE);
3357
3358                 btrfs_warn(fs_info,
3359                 "read-write for sector size %u with page size %lu is experimental",
3360                            sectorsize, PAGE_SIZE);
3361         }
3362         if (sectorsize != PAGE_SIZE) {
3363                 if (btrfs_super_incompat_flags(fs_info->super_copy) &
3364                         BTRFS_FEATURE_INCOMPAT_RAID56) {
3365                         btrfs_err(fs_info,
3366                 "RAID56 is not yet supported for sector size %u with page size %lu",
3367                                 sectorsize, PAGE_SIZE);
3368                         err = -EINVAL;
3369                         goto fail_alloc;
3370                 }
3371         }
3372
3373         ret = btrfs_init_workqueues(fs_info, fs_devices);
3374         if (ret) {
3375                 err = ret;
3376                 goto fail_sb_buffer;
3377         }
3378
3379         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3380         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3381
3382         sb->s_blocksize = sectorsize;
3383         sb->s_blocksize_bits = blksize_bits(sectorsize);
3384         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3385
3386         mutex_lock(&fs_info->chunk_mutex);
3387         ret = btrfs_read_sys_array(fs_info);
3388         mutex_unlock(&fs_info->chunk_mutex);
3389         if (ret) {
3390                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3391                 goto fail_sb_buffer;
3392         }
3393
3394         generation = btrfs_super_chunk_root_generation(disk_super);
3395         level = btrfs_super_chunk_root_level(disk_super);
3396
3397         chunk_root->node = read_tree_block(fs_info,
3398                                            btrfs_super_chunk_root(disk_super),
3399                                            BTRFS_CHUNK_TREE_OBJECTID,
3400                                            generation, level, NULL);
3401         if (IS_ERR(chunk_root->node) ||
3402             !extent_buffer_uptodate(chunk_root->node)) {
3403                 btrfs_err(fs_info, "failed to read chunk root");
3404                 if (!IS_ERR(chunk_root->node))
3405                         free_extent_buffer(chunk_root->node);
3406                 chunk_root->node = NULL;
3407                 goto fail_tree_roots;
3408         }
3409         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3410         chunk_root->commit_root = btrfs_root_node(chunk_root);
3411
3412         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3413                            offsetof(struct btrfs_header, chunk_tree_uuid),
3414                            BTRFS_UUID_SIZE);
3415
3416         ret = btrfs_read_chunk_tree(fs_info);
3417         if (ret) {
3418                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3419                 goto fail_tree_roots;
3420         }
3421
3422         /*
3423          * At this point we know all the devices that make this filesystem,
3424          * including the seed devices but we don't know yet if the replace
3425          * target is required. So free devices that are not part of this
3426          * filesystem but skip the replace target device which is checked
3427          * below in btrfs_init_dev_replace().
3428          */
3429         btrfs_free_extra_devids(fs_devices);
3430         if (!fs_devices->latest_dev->bdev) {
3431                 btrfs_err(fs_info, "failed to read devices");
3432                 goto fail_tree_roots;
3433         }
3434
3435         ret = init_tree_roots(fs_info);
3436         if (ret)
3437                 goto fail_tree_roots;
3438
3439         /*
3440          * Get zone type information of zoned block devices. This will also
3441          * handle emulation of a zoned filesystem if a regular device has the
3442          * zoned incompat feature flag set.
3443          */
3444         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3445         if (ret) {
3446                 btrfs_err(fs_info,
3447                           "zoned: failed to read device zone info: %d",
3448                           ret);
3449                 goto fail_block_groups;
3450         }
3451
3452         /*
3453          * If we have a uuid root and we're not being told to rescan we need to
3454          * check the generation here so we can set the
3455          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3456          * transaction during a balance or the log replay without updating the
3457          * uuid generation, and then if we crash we would rescan the uuid tree,
3458          * even though it was perfectly fine.
3459          */
3460         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3461             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3462                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3463
3464         ret = btrfs_verify_dev_extents(fs_info);
3465         if (ret) {
3466                 btrfs_err(fs_info,
3467                           "failed to verify dev extents against chunks: %d",
3468                           ret);
3469                 goto fail_block_groups;
3470         }
3471         ret = btrfs_recover_balance(fs_info);
3472         if (ret) {
3473                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3474                 goto fail_block_groups;
3475         }
3476
3477         ret = btrfs_init_dev_stats(fs_info);
3478         if (ret) {
3479                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3480                 goto fail_block_groups;
3481         }
3482
3483         ret = btrfs_init_dev_replace(fs_info);
3484         if (ret) {
3485                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3486                 goto fail_block_groups;
3487         }
3488         /*
3489          * We have unsupported RO compat features, although RO mounted, we
3490          * should not cause any metadata write, including log replay.
3491          * Or we could screw up whatever the new feature requires.
3492          */
3493         if (unlikely(features && btrfs_super_log_root(disk_super) &&
3494                      !btrfs_test_opt(fs_info, NOLOGREPLAY))) {
3495                 btrfs_err(fs_info,
3496 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3497                           features);
3498                 err = -EINVAL;
3499                 goto fail_alloc;
3500         }
3501
3502
3503         ret = btrfs_check_zoned_mode(fs_info);
3504         if (ret) {
3505                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3506                           ret);
3507                 goto fail_block_groups;
3508         }
3509
3510         ret = btrfs_sysfs_add_fsid(fs_devices);
3511         if (ret) {
3512                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3513                                 ret);
3514                 goto fail_block_groups;
3515         }
3516
3517         ret = btrfs_sysfs_add_mounted(fs_info);
3518         if (ret) {
3519                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3520                 goto fail_fsdev_sysfs;
3521         }
3522
3523         ret = btrfs_init_space_info(fs_info);
3524         if (ret) {
3525                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3526                 goto fail_sysfs;
3527         }
3528
3529         ret = btrfs_read_block_groups(fs_info);
3530         if (ret) {
3531                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3532                 goto fail_sysfs;
3533         }
3534
3535         btrfs_free_zone_cache(fs_info);
3536
3537         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3538             !btrfs_check_rw_degradable(fs_info, NULL)) {
3539                 btrfs_warn(fs_info,
3540                 "writable mount is not allowed due to too many missing devices");
3541                 goto fail_sysfs;
3542         }
3543
3544         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3545                                                "btrfs-cleaner");
3546         if (IS_ERR(fs_info->cleaner_kthread))
3547                 goto fail_sysfs;
3548
3549         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3550                                                    tree_root,
3551                                                    "btrfs-transaction");
3552         if (IS_ERR(fs_info->transaction_kthread))
3553                 goto fail_cleaner;
3554
3555         if (!btrfs_test_opt(fs_info, NOSSD) &&
3556             !fs_info->fs_devices->rotating) {
3557                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3558         }
3559
3560         /*
3561          * Mount does not set all options immediately, we can do it now and do
3562          * not have to wait for transaction commit
3563          */
3564         btrfs_apply_pending_changes(fs_info);
3565
3566 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3567         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3568                 ret = btrfsic_mount(fs_info, fs_devices,
3569                                     btrfs_test_opt(fs_info,
3570                                         CHECK_INTEGRITY_DATA) ? 1 : 0,
3571                                     fs_info->check_integrity_print_mask);
3572                 if (ret)
3573                         btrfs_warn(fs_info,
3574                                 "failed to initialize integrity check module: %d",
3575                                 ret);
3576         }
3577 #endif
3578         ret = btrfs_read_qgroup_config(fs_info);
3579         if (ret)
3580                 goto fail_trans_kthread;
3581
3582         if (btrfs_build_ref_tree(fs_info))
3583                 btrfs_err(fs_info, "couldn't build ref tree");
3584
3585         /* do not make disk changes in broken FS or nologreplay is given */
3586         if (btrfs_super_log_root(disk_super) != 0 &&
3587             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3588                 btrfs_info(fs_info, "start tree-log replay");
3589                 ret = btrfs_replay_log(fs_info, fs_devices);
3590                 if (ret) {
3591                         err = ret;
3592                         goto fail_qgroup;
3593                 }
3594         }
3595
3596         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3597         if (IS_ERR(fs_info->fs_root)) {
3598                 err = PTR_ERR(fs_info->fs_root);
3599                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3600                 fs_info->fs_root = NULL;
3601                 goto fail_qgroup;
3602         }
3603
3604         if (sb_rdonly(sb))
3605                 goto clear_oneshot;
3606
3607         ret = btrfs_start_pre_rw_mount(fs_info);
3608         if (ret) {
3609                 close_ctree(fs_info);
3610                 return ret;
3611         }
3612         btrfs_discard_resume(fs_info);
3613
3614         if (fs_info->uuid_root &&
3615             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3616              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3617                 btrfs_info(fs_info, "checking UUID tree");
3618                 ret = btrfs_check_uuid_tree(fs_info);
3619                 if (ret) {
3620                         btrfs_warn(fs_info,
3621                                 "failed to check the UUID tree: %d", ret);
3622                         close_ctree(fs_info);
3623                         return ret;
3624                 }
3625         }
3626
3627         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3628
3629         /* Kick the cleaner thread so it'll start deleting snapshots. */
3630         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3631                 wake_up_process(fs_info->cleaner_kthread);
3632
3633 clear_oneshot:
3634         btrfs_clear_oneshot_options(fs_info);
3635         return 0;
3636
3637 fail_qgroup:
3638         btrfs_free_qgroup_config(fs_info);
3639 fail_trans_kthread:
3640         kthread_stop(fs_info->transaction_kthread);
3641         btrfs_cleanup_transaction(fs_info);
3642         btrfs_free_fs_roots(fs_info);
3643 fail_cleaner:
3644         kthread_stop(fs_info->cleaner_kthread);
3645
3646         /*
3647          * make sure we're done with the btree inode before we stop our
3648          * kthreads
3649          */
3650         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3651
3652 fail_sysfs:
3653         btrfs_sysfs_remove_mounted(fs_info);
3654
3655 fail_fsdev_sysfs:
3656         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3657
3658 fail_block_groups:
3659         btrfs_put_block_group_cache(fs_info);
3660
3661 fail_tree_roots:
3662         if (fs_info->data_reloc_root)
3663                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3664         free_root_pointers(fs_info, true);
3665         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3666
3667 fail_sb_buffer:
3668         btrfs_stop_all_workers(fs_info);
3669         btrfs_free_block_groups(fs_info);
3670 fail_alloc:
3671         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3672
3673         iput(fs_info->btree_inode);
3674 fail:
3675         btrfs_close_devices(fs_info->fs_devices);
3676         return err;
3677 }
3678 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3679
3680 static void btrfs_end_super_write(struct bio *bio)
3681 {
3682         struct btrfs_device *device = bio->bi_private;
3683         struct bio_vec *bvec;
3684         struct bvec_iter_all iter_all;
3685         struct page *page;
3686
3687         bio_for_each_segment_all(bvec, bio, iter_all) {
3688                 page = bvec->bv_page;
3689
3690                 if (bio->bi_status) {
3691                         btrfs_warn_rl_in_rcu(device->fs_info,
3692                                 "lost page write due to IO error on %s (%d)",
3693                                 rcu_str_deref(device->name),
3694                                 blk_status_to_errno(bio->bi_status));
3695                         ClearPageUptodate(page);
3696                         SetPageError(page);
3697                         btrfs_dev_stat_inc_and_print(device,
3698                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3699                 } else {
3700                         SetPageUptodate(page);
3701                 }
3702
3703                 put_page(page);
3704                 unlock_page(page);
3705         }
3706
3707         bio_put(bio);
3708 }
3709
3710 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3711                                                    int copy_num, bool drop_cache)
3712 {
3713         struct btrfs_super_block *super;
3714         struct page *page;
3715         u64 bytenr, bytenr_orig;
3716         struct address_space *mapping = bdev->bd_inode->i_mapping;
3717         int ret;
3718
3719         bytenr_orig = btrfs_sb_offset(copy_num);
3720         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3721         if (ret == -ENOENT)
3722                 return ERR_PTR(-EINVAL);
3723         else if (ret)
3724                 return ERR_PTR(ret);
3725
3726         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3727                 return ERR_PTR(-EINVAL);
3728
3729         if (drop_cache) {
3730                 /* This should only be called with the primary sb. */
3731                 ASSERT(copy_num == 0);
3732
3733                 /*
3734                  * Drop the page of the primary superblock, so later read will
3735                  * always read from the device.
3736                  */
3737                 invalidate_inode_pages2_range(mapping,
3738                                 bytenr >> PAGE_SHIFT,
3739                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3740         }
3741
3742         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3743         if (IS_ERR(page))
3744                 return ERR_CAST(page);
3745
3746         super = page_address(page);
3747         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3748                 btrfs_release_disk_super(super);
3749                 return ERR_PTR(-ENODATA);
3750         }
3751
3752         if (btrfs_super_bytenr(super) != bytenr_orig) {
3753                 btrfs_release_disk_super(super);
3754                 return ERR_PTR(-EINVAL);
3755         }
3756
3757         return super;
3758 }
3759
3760
3761 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3762 {
3763         struct btrfs_super_block *super, *latest = NULL;
3764         int i;
3765         u64 transid = 0;
3766
3767         /* we would like to check all the supers, but that would make
3768          * a btrfs mount succeed after a mkfs from a different FS.
3769          * So, we need to add a special mount option to scan for
3770          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3771          */
3772         for (i = 0; i < 1; i++) {
3773                 super = btrfs_read_dev_one_super(bdev, i, false);
3774                 if (IS_ERR(super))
3775                         continue;
3776
3777                 if (!latest || btrfs_super_generation(super) > transid) {
3778                         if (latest)
3779                                 btrfs_release_disk_super(super);
3780
3781                         latest = super;
3782                         transid = btrfs_super_generation(super);
3783                 }
3784         }
3785
3786         return super;
3787 }
3788
3789 /*
3790  * Write superblock @sb to the @device. Do not wait for completion, all the
3791  * pages we use for writing are locked.
3792  *
3793  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3794  * the expected device size at commit time. Note that max_mirrors must be
3795  * same for write and wait phases.
3796  *
3797  * Return number of errors when page is not found or submission fails.
3798  */
3799 static int write_dev_supers(struct btrfs_device *device,
3800                             struct btrfs_super_block *sb, int max_mirrors)
3801 {
3802         struct btrfs_fs_info *fs_info = device->fs_info;
3803         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3804         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3805         int i;
3806         int errors = 0;
3807         int ret;
3808         u64 bytenr, bytenr_orig;
3809
3810         if (max_mirrors == 0)
3811                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3812
3813         shash->tfm = fs_info->csum_shash;
3814
3815         for (i = 0; i < max_mirrors; i++) {
3816                 struct page *page;
3817                 struct bio *bio;
3818                 struct btrfs_super_block *disk_super;
3819
3820                 bytenr_orig = btrfs_sb_offset(i);
3821                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3822                 if (ret == -ENOENT) {
3823                         continue;
3824                 } else if (ret < 0) {
3825                         btrfs_err(device->fs_info,
3826                                 "couldn't get super block location for mirror %d",
3827                                 i);
3828                         errors++;
3829                         continue;
3830                 }
3831                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3832                     device->commit_total_bytes)
3833                         break;
3834
3835                 btrfs_set_super_bytenr(sb, bytenr_orig);
3836
3837                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3838                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3839                                     sb->csum);
3840
3841                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3842                                            GFP_NOFS);
3843                 if (!page) {
3844                         btrfs_err(device->fs_info,
3845                             "couldn't get super block page for bytenr %llu",
3846                             bytenr);
3847                         errors++;
3848                         continue;
3849                 }
3850
3851                 /* Bump the refcount for wait_dev_supers() */
3852                 get_page(page);
3853
3854                 disk_super = page_address(page);
3855                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3856
3857                 /*
3858                  * Directly use bios here instead of relying on the page cache
3859                  * to do I/O, so we don't lose the ability to do integrity
3860                  * checking.
3861                  */
3862                 bio = bio_alloc(GFP_NOFS, 1);
3863                 bio_set_dev(bio, device->bdev);
3864                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3865                 bio->bi_private = device;
3866                 bio->bi_end_io = btrfs_end_super_write;
3867                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3868                                offset_in_page(bytenr));
3869
3870                 /*
3871                  * We FUA only the first super block.  The others we allow to
3872                  * go down lazy and there's a short window where the on-disk
3873                  * copies might still contain the older version.
3874                  */
3875                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3876                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3877                         bio->bi_opf |= REQ_FUA;
3878
3879                 btrfsic_submit_bio(bio);
3880                 btrfs_advance_sb_log(device, i);
3881         }
3882         return errors < i ? 0 : -1;
3883 }
3884
3885 /*
3886  * Wait for write completion of superblocks done by write_dev_supers,
3887  * @max_mirrors same for write and wait phases.
3888  *
3889  * Return number of errors when page is not found or not marked up to
3890  * date.
3891  */
3892 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3893 {
3894         int i;
3895         int errors = 0;
3896         bool primary_failed = false;
3897         int ret;
3898         u64 bytenr;
3899
3900         if (max_mirrors == 0)
3901                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3902
3903         for (i = 0; i < max_mirrors; i++) {
3904                 struct page *page;
3905
3906                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3907                 if (ret == -ENOENT) {
3908                         break;
3909                 } else if (ret < 0) {
3910                         errors++;
3911                         if (i == 0)
3912                                 primary_failed = true;
3913                         continue;
3914                 }
3915                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3916                     device->commit_total_bytes)
3917                         break;
3918
3919                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3920                                      bytenr >> PAGE_SHIFT);
3921                 if (!page) {
3922                         errors++;
3923                         if (i == 0)
3924                                 primary_failed = true;
3925                         continue;
3926                 }
3927                 /* Page is submitted locked and unlocked once the IO completes */
3928                 wait_on_page_locked(page);
3929                 if (PageError(page)) {
3930                         errors++;
3931                         if (i == 0)
3932                                 primary_failed = true;
3933                 }
3934
3935                 /* Drop our reference */
3936                 put_page(page);
3937
3938                 /* Drop the reference from the writing run */
3939                 put_page(page);
3940         }
3941
3942         /* log error, force error return */
3943         if (primary_failed) {
3944                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3945                           device->devid);
3946                 return -1;
3947         }
3948
3949         return errors < i ? 0 : -1;
3950 }
3951
3952 /*
3953  * endio for the write_dev_flush, this will wake anyone waiting
3954  * for the barrier when it is done
3955  */
3956 static void btrfs_end_empty_barrier(struct bio *bio)
3957 {
3958         complete(bio->bi_private);
3959 }
3960
3961 /*
3962  * Submit a flush request to the device if it supports it. Error handling is
3963  * done in the waiting counterpart.
3964  */
3965 static void write_dev_flush(struct btrfs_device *device)
3966 {
3967         struct bio *bio = device->flush_bio;
3968
3969 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3970         /*
3971          * When a disk has write caching disabled, we skip submission of a bio
3972          * with flush and sync requests before writing the superblock, since
3973          * it's not needed. However when the integrity checker is enabled, this
3974          * results in reports that there are metadata blocks referred by a
3975          * superblock that were not properly flushed. So don't skip the bio
3976          * submission only when the integrity checker is enabled for the sake
3977          * of simplicity, since this is a debug tool and not meant for use in
3978          * non-debug builds.
3979          */
3980         struct request_queue *q = bdev_get_queue(device->bdev);
3981         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3982                 return;
3983 #endif
3984
3985         bio_reset(bio);
3986         bio->bi_end_io = btrfs_end_empty_barrier;
3987         bio_set_dev(bio, device->bdev);
3988         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3989         init_completion(&device->flush_wait);
3990         bio->bi_private = &device->flush_wait;
3991
3992         btrfsic_submit_bio(bio);
3993         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3994 }
3995
3996 /*
3997  * If the flush bio has been submitted by write_dev_flush, wait for it.
3998  */
3999 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4000 {
4001         struct bio *bio = device->flush_bio;
4002
4003         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4004                 return BLK_STS_OK;
4005
4006         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4007         wait_for_completion_io(&device->flush_wait);
4008
4009         return bio->bi_status;
4010 }
4011
4012 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4013 {
4014         if (!btrfs_check_rw_degradable(fs_info, NULL))
4015                 return -EIO;
4016         return 0;
4017 }
4018
4019 /*
4020  * send an empty flush down to each device in parallel,
4021  * then wait for them
4022  */
4023 static int barrier_all_devices(struct btrfs_fs_info *info)
4024 {
4025         struct list_head *head;
4026         struct btrfs_device *dev;
4027         int errors_wait = 0;
4028         blk_status_t ret;
4029
4030         lockdep_assert_held(&info->fs_devices->device_list_mutex);
4031         /* send down all the barriers */
4032         head = &info->fs_devices->devices;
4033         list_for_each_entry(dev, head, dev_list) {
4034                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4035                         continue;
4036                 if (!dev->bdev)
4037                         continue;
4038                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4039                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4040                         continue;
4041
4042                 write_dev_flush(dev);
4043                 dev->last_flush_error = BLK_STS_OK;
4044         }
4045
4046         /* wait for all the barriers */
4047         list_for_each_entry(dev, head, dev_list) {
4048                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4049                         continue;
4050                 if (!dev->bdev) {
4051                         errors_wait++;
4052                         continue;
4053                 }
4054                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4055                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4056                         continue;
4057
4058                 ret = wait_dev_flush(dev);
4059                 if (ret) {
4060                         dev->last_flush_error = ret;
4061                         btrfs_dev_stat_inc_and_print(dev,
4062                                         BTRFS_DEV_STAT_FLUSH_ERRS);
4063                         errors_wait++;
4064                 }
4065         }
4066
4067         if (errors_wait) {
4068                 /*
4069                  * At some point we need the status of all disks
4070                  * to arrive at the volume status. So error checking
4071                  * is being pushed to a separate loop.
4072                  */
4073                 return check_barrier_error(info);
4074         }
4075         return 0;
4076 }
4077
4078 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4079 {
4080         int raid_type;
4081         int min_tolerated = INT_MAX;
4082
4083         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4084             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4085                 min_tolerated = min_t(int, min_tolerated,
4086                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4087                                     tolerated_failures);
4088
4089         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4090                 if (raid_type == BTRFS_RAID_SINGLE)
4091                         continue;
4092                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4093                         continue;
4094                 min_tolerated = min_t(int, min_tolerated,
4095                                     btrfs_raid_array[raid_type].
4096                                     tolerated_failures);
4097         }
4098
4099         if (min_tolerated == INT_MAX) {
4100                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4101                 min_tolerated = 0;
4102         }
4103
4104         return min_tolerated;
4105 }
4106
4107 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4108 {
4109         struct list_head *head;
4110         struct btrfs_device *dev;
4111         struct btrfs_super_block *sb;
4112         struct btrfs_dev_item *dev_item;
4113         int ret;
4114         int do_barriers;
4115         int max_errors;
4116         int total_errors = 0;
4117         u64 flags;
4118
4119         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4120
4121         /*
4122          * max_mirrors == 0 indicates we're from commit_transaction,
4123          * not from fsync where the tree roots in fs_info have not
4124          * been consistent on disk.
4125          */
4126         if (max_mirrors == 0)
4127                 backup_super_roots(fs_info);
4128
4129         sb = fs_info->super_for_commit;
4130         dev_item = &sb->dev_item;
4131
4132         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4133         head = &fs_info->fs_devices->devices;
4134         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4135
4136         if (do_barriers) {
4137                 ret = barrier_all_devices(fs_info);
4138                 if (ret) {
4139                         mutex_unlock(
4140                                 &fs_info->fs_devices->device_list_mutex);
4141                         btrfs_handle_fs_error(fs_info, ret,
4142                                               "errors while submitting device barriers.");
4143                         return ret;
4144                 }
4145         }
4146
4147         list_for_each_entry(dev, head, dev_list) {
4148                 if (!dev->bdev) {
4149                         total_errors++;
4150                         continue;
4151                 }
4152                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4153                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4154                         continue;
4155
4156                 btrfs_set_stack_device_generation(dev_item, 0);
4157                 btrfs_set_stack_device_type(dev_item, dev->type);
4158                 btrfs_set_stack_device_id(dev_item, dev->devid);
4159                 btrfs_set_stack_device_total_bytes(dev_item,
4160                                                    dev->commit_total_bytes);
4161                 btrfs_set_stack_device_bytes_used(dev_item,
4162                                                   dev->commit_bytes_used);
4163                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4164                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4165                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4166                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4167                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4168                        BTRFS_FSID_SIZE);
4169
4170                 flags = btrfs_super_flags(sb);
4171                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4172
4173                 ret = btrfs_validate_write_super(fs_info, sb);
4174                 if (ret < 0) {
4175                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4176                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4177                                 "unexpected superblock corruption detected");
4178                         return -EUCLEAN;
4179                 }
4180
4181                 ret = write_dev_supers(dev, sb, max_mirrors);
4182                 if (ret)
4183                         total_errors++;
4184         }
4185         if (total_errors > max_errors) {
4186                 btrfs_err(fs_info, "%d errors while writing supers",
4187                           total_errors);
4188                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4189
4190                 /* FUA is masked off if unsupported and can't be the reason */
4191                 btrfs_handle_fs_error(fs_info, -EIO,
4192                                       "%d errors while writing supers",
4193                                       total_errors);
4194                 return -EIO;
4195         }
4196
4197         total_errors = 0;
4198         list_for_each_entry(dev, head, dev_list) {
4199                 if (!dev->bdev)
4200                         continue;
4201                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4202                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4203                         continue;
4204
4205                 ret = wait_dev_supers(dev, max_mirrors);
4206                 if (ret)
4207                         total_errors++;
4208         }
4209         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4210         if (total_errors > max_errors) {
4211                 btrfs_handle_fs_error(fs_info, -EIO,
4212                                       "%d errors while writing supers",
4213                                       total_errors);
4214                 return -EIO;
4215         }
4216         return 0;
4217 }
4218
4219 /* Drop a fs root from the radix tree and free it. */
4220 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4221                                   struct btrfs_root *root)
4222 {
4223         bool drop_ref = false;
4224
4225         spin_lock(&fs_info->fs_roots_radix_lock);
4226         radix_tree_delete(&fs_info->fs_roots_radix,
4227                           (unsigned long)root->root_key.objectid);
4228         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4229                 drop_ref = true;
4230         spin_unlock(&fs_info->fs_roots_radix_lock);
4231
4232         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4233                 ASSERT(root->log_root == NULL);
4234                 if (root->reloc_root) {
4235                         btrfs_put_root(root->reloc_root);
4236                         root->reloc_root = NULL;
4237                 }
4238         }
4239
4240         if (drop_ref)
4241                 btrfs_put_root(root);
4242 }
4243
4244 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4245 {
4246         u64 root_objectid = 0;
4247         struct btrfs_root *gang[8];
4248         int i = 0;
4249         int err = 0;
4250         unsigned int ret = 0;
4251
4252         while (1) {
4253                 spin_lock(&fs_info->fs_roots_radix_lock);
4254                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4255                                              (void **)gang, root_objectid,
4256                                              ARRAY_SIZE(gang));
4257                 if (!ret) {
4258                         spin_unlock(&fs_info->fs_roots_radix_lock);
4259                         break;
4260                 }
4261                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4262
4263                 for (i = 0; i < ret; i++) {
4264                         /* Avoid to grab roots in dead_roots */
4265                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4266                                 gang[i] = NULL;
4267                                 continue;
4268                         }
4269                         /* grab all the search result for later use */
4270                         gang[i] = btrfs_grab_root(gang[i]);
4271                 }
4272                 spin_unlock(&fs_info->fs_roots_radix_lock);
4273
4274                 for (i = 0; i < ret; i++) {
4275                         if (!gang[i])
4276                                 continue;
4277                         root_objectid = gang[i]->root_key.objectid;
4278                         err = btrfs_orphan_cleanup(gang[i]);
4279                         if (err)
4280                                 break;
4281                         btrfs_put_root(gang[i]);
4282                 }
4283                 root_objectid++;
4284         }
4285
4286         /* release the uncleaned roots due to error */
4287         for (; i < ret; i++) {
4288                 if (gang[i])
4289                         btrfs_put_root(gang[i]);
4290         }
4291         return err;
4292 }
4293
4294 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4295 {
4296         struct btrfs_root *root = fs_info->tree_root;
4297         struct btrfs_trans_handle *trans;
4298
4299         mutex_lock(&fs_info->cleaner_mutex);
4300         btrfs_run_delayed_iputs(fs_info);
4301         mutex_unlock(&fs_info->cleaner_mutex);
4302         wake_up_process(fs_info->cleaner_kthread);
4303
4304         /* wait until ongoing cleanup work done */
4305         down_write(&fs_info->cleanup_work_sem);
4306         up_write(&fs_info->cleanup_work_sem);
4307
4308         trans = btrfs_join_transaction(root);
4309         if (IS_ERR(trans))
4310                 return PTR_ERR(trans);
4311         return btrfs_commit_transaction(trans);
4312 }
4313
4314 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4315 {
4316         int ret;
4317
4318         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4319
4320         /*
4321          * If we had UNFINISHED_DROPS we could still be processing them, so
4322          * clear that bit and wake up relocation so it can stop.
4323          * We must do this before stopping the block group reclaim task, because
4324          * at btrfs_relocate_block_group() we wait for this bit, and after the
4325          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4326          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4327          * return 1.
4328          */
4329         btrfs_wake_unfinished_drop(fs_info);
4330
4331         /*
4332          * We may have the reclaim task running and relocating a data block group,
4333          * in which case it may create delayed iputs. So stop it before we park
4334          * the cleaner kthread otherwise we can get new delayed iputs after
4335          * parking the cleaner, and that can make the async reclaim task to hang
4336          * if it's waiting for delayed iputs to complete, since the cleaner is
4337          * parked and can not run delayed iputs - this will make us hang when
4338          * trying to stop the async reclaim task.
4339          */
4340         cancel_work_sync(&fs_info->reclaim_bgs_work);
4341         /*
4342          * We don't want the cleaner to start new transactions, add more delayed
4343          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4344          * because that frees the task_struct, and the transaction kthread might
4345          * still try to wake up the cleaner.
4346          */
4347         kthread_park(fs_info->cleaner_kthread);
4348
4349         /* wait for the qgroup rescan worker to stop */
4350         btrfs_qgroup_wait_for_completion(fs_info, false);
4351
4352         /* wait for the uuid_scan task to finish */
4353         down(&fs_info->uuid_tree_rescan_sem);
4354         /* avoid complains from lockdep et al., set sem back to initial state */
4355         up(&fs_info->uuid_tree_rescan_sem);
4356
4357         /* pause restriper - we want to resume on mount */
4358         btrfs_pause_balance(fs_info);
4359
4360         btrfs_dev_replace_suspend_for_unmount(fs_info);
4361
4362         btrfs_scrub_cancel(fs_info);
4363
4364         /* wait for any defraggers to finish */
4365         wait_event(fs_info->transaction_wait,
4366                    (atomic_read(&fs_info->defrag_running) == 0));
4367
4368         /* clear out the rbtree of defraggable inodes */
4369         btrfs_cleanup_defrag_inodes(fs_info);
4370
4371         /*
4372          * After we parked the cleaner kthread, ordered extents may have
4373          * completed and created new delayed iputs. If one of the async reclaim
4374          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4375          * can hang forever trying to stop it, because if a delayed iput is
4376          * added after it ran btrfs_run_delayed_iputs() and before it called
4377          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4378          * no one else to run iputs.
4379          *
4380          * So wait for all ongoing ordered extents to complete and then run
4381          * delayed iputs. This works because once we reach this point no one
4382          * can either create new ordered extents nor create delayed iputs
4383          * through some other means.
4384          *
4385          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4386          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4387          * but the delayed iput for the respective inode is made only when doing
4388          * the final btrfs_put_ordered_extent() (which must happen at
4389          * btrfs_finish_ordered_io() when we are unmounting).
4390          */
4391         btrfs_flush_workqueue(fs_info->endio_write_workers);
4392         /* Ordered extents for free space inodes. */
4393         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4394         btrfs_run_delayed_iputs(fs_info);
4395
4396         cancel_work_sync(&fs_info->async_reclaim_work);
4397         cancel_work_sync(&fs_info->async_data_reclaim_work);
4398         cancel_work_sync(&fs_info->preempt_reclaim_work);
4399
4400         /* Cancel or finish ongoing discard work */
4401         btrfs_discard_cleanup(fs_info);
4402
4403         if (!sb_rdonly(fs_info->sb)) {
4404                 /*
4405                  * The cleaner kthread is stopped, so do one final pass over
4406                  * unused block groups.
4407                  */
4408                 btrfs_delete_unused_bgs(fs_info);
4409
4410                 /*
4411                  * There might be existing delayed inode workers still running
4412                  * and holding an empty delayed inode item. We must wait for
4413                  * them to complete first because they can create a transaction.
4414                  * This happens when someone calls btrfs_balance_delayed_items()
4415                  * and then a transaction commit runs the same delayed nodes
4416                  * before any delayed worker has done something with the nodes.
4417                  * We must wait for any worker here and not at transaction
4418                  * commit time since that could cause a deadlock.
4419                  * This is a very rare case.
4420                  */
4421                 btrfs_flush_workqueue(fs_info->delayed_workers);
4422
4423                 ret = btrfs_commit_super(fs_info);
4424                 if (ret)
4425                         btrfs_err(fs_info, "commit super ret %d", ret);
4426         }
4427
4428         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4429             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4430                 btrfs_error_commit_super(fs_info);
4431
4432         kthread_stop(fs_info->transaction_kthread);
4433         kthread_stop(fs_info->cleaner_kthread);
4434
4435         ASSERT(list_empty(&fs_info->delayed_iputs));
4436         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4437
4438         if (btrfs_check_quota_leak(fs_info)) {
4439                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4440                 btrfs_err(fs_info, "qgroup reserved space leaked");
4441         }
4442
4443         btrfs_free_qgroup_config(fs_info);
4444         ASSERT(list_empty(&fs_info->delalloc_roots));
4445
4446         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4447                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4448                        percpu_counter_sum(&fs_info->delalloc_bytes));
4449         }
4450
4451         if (percpu_counter_sum(&fs_info->ordered_bytes))
4452                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4453                            percpu_counter_sum(&fs_info->ordered_bytes));
4454
4455         btrfs_sysfs_remove_mounted(fs_info);
4456         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4457
4458         btrfs_put_block_group_cache(fs_info);
4459
4460         /*
4461          * we must make sure there is not any read request to
4462          * submit after we stopping all workers.
4463          */
4464         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4465         btrfs_stop_all_workers(fs_info);
4466
4467         /* We shouldn't have any transaction open at this point */
4468         ASSERT(list_empty(&fs_info->trans_list));
4469
4470         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4471         free_root_pointers(fs_info, true);
4472         btrfs_free_fs_roots(fs_info);
4473
4474         /*
4475          * We must free the block groups after dropping the fs_roots as we could
4476          * have had an IO error and have left over tree log blocks that aren't
4477          * cleaned up until the fs roots are freed.  This makes the block group
4478          * accounting appear to be wrong because there's pending reserved bytes,
4479          * so make sure we do the block group cleanup afterwards.
4480          */
4481         btrfs_free_block_groups(fs_info);
4482
4483         iput(fs_info->btree_inode);
4484
4485 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4486         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4487                 btrfsic_unmount(fs_info->fs_devices);
4488 #endif
4489
4490         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4491         btrfs_close_devices(fs_info->fs_devices);
4492 }
4493
4494 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4495                           int atomic)
4496 {
4497         int ret;
4498         struct inode *btree_inode = buf->pages[0]->mapping->host;
4499
4500         ret = extent_buffer_uptodate(buf);
4501         if (!ret)
4502                 return ret;
4503
4504         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4505                                     parent_transid, atomic);
4506         if (ret == -EAGAIN)
4507                 return ret;
4508         return !ret;
4509 }
4510
4511 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4512 {
4513         struct btrfs_fs_info *fs_info = buf->fs_info;
4514         u64 transid = btrfs_header_generation(buf);
4515         int was_dirty;
4516
4517 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4518         /*
4519          * This is a fast path so only do this check if we have sanity tests
4520          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4521          * outside of the sanity tests.
4522          */
4523         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4524                 return;
4525 #endif
4526         btrfs_assert_tree_locked(buf);
4527         if (transid != fs_info->generation)
4528                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4529                         buf->start, transid, fs_info->generation);
4530         was_dirty = set_extent_buffer_dirty(buf);
4531         if (!was_dirty)
4532                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4533                                          buf->len,
4534                                          fs_info->dirty_metadata_batch);
4535 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4536         /*
4537          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4538          * but item data not updated.
4539          * So here we should only check item pointers, not item data.
4540          */
4541         if (btrfs_header_level(buf) == 0 &&
4542             btrfs_check_leaf_relaxed(buf)) {
4543                 btrfs_print_leaf(buf);
4544                 ASSERT(0);
4545         }
4546 #endif
4547 }
4548
4549 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4550                                         int flush_delayed)
4551 {
4552         /*
4553          * looks as though older kernels can get into trouble with
4554          * this code, they end up stuck in balance_dirty_pages forever
4555          */
4556         int ret;
4557
4558         if (current->flags & PF_MEMALLOC)
4559                 return;
4560
4561         if (flush_delayed)
4562                 btrfs_balance_delayed_items(fs_info);
4563
4564         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4565                                      BTRFS_DIRTY_METADATA_THRESH,
4566                                      fs_info->dirty_metadata_batch);
4567         if (ret > 0) {
4568                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4569         }
4570 }
4571
4572 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4573 {
4574         __btrfs_btree_balance_dirty(fs_info, 1);
4575 }
4576
4577 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4578 {
4579         __btrfs_btree_balance_dirty(fs_info, 0);
4580 }
4581
4582 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4583                       struct btrfs_key *first_key)
4584 {
4585         return btree_read_extent_buffer_pages(buf, parent_transid,
4586                                               level, first_key);
4587 }
4588
4589 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4590 {
4591         /* cleanup FS via transaction */
4592         btrfs_cleanup_transaction(fs_info);
4593
4594         mutex_lock(&fs_info->cleaner_mutex);
4595         btrfs_run_delayed_iputs(fs_info);
4596         mutex_unlock(&fs_info->cleaner_mutex);
4597
4598         down_write(&fs_info->cleanup_work_sem);
4599         up_write(&fs_info->cleanup_work_sem);
4600 }
4601
4602 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4603 {
4604         struct btrfs_root *gang[8];
4605         u64 root_objectid = 0;
4606         int ret;
4607
4608         spin_lock(&fs_info->fs_roots_radix_lock);
4609         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4610                                              (void **)gang, root_objectid,
4611                                              ARRAY_SIZE(gang))) != 0) {
4612                 int i;
4613
4614                 for (i = 0; i < ret; i++)
4615                         gang[i] = btrfs_grab_root(gang[i]);
4616                 spin_unlock(&fs_info->fs_roots_radix_lock);
4617
4618                 for (i = 0; i < ret; i++) {
4619                         if (!gang[i])
4620                                 continue;
4621                         root_objectid = gang[i]->root_key.objectid;
4622                         btrfs_free_log(NULL, gang[i]);
4623                         btrfs_put_root(gang[i]);
4624                 }
4625                 root_objectid++;
4626                 spin_lock(&fs_info->fs_roots_radix_lock);
4627         }
4628         spin_unlock(&fs_info->fs_roots_radix_lock);
4629         btrfs_free_log_root_tree(NULL, fs_info);
4630 }
4631
4632 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4633 {
4634         struct btrfs_ordered_extent *ordered;
4635
4636         spin_lock(&root->ordered_extent_lock);
4637         /*
4638          * This will just short circuit the ordered completion stuff which will
4639          * make sure the ordered extent gets properly cleaned up.
4640          */
4641         list_for_each_entry(ordered, &root->ordered_extents,
4642                             root_extent_list)
4643                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4644         spin_unlock(&root->ordered_extent_lock);
4645 }
4646
4647 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4648 {
4649         struct btrfs_root *root;
4650         struct list_head splice;
4651
4652         INIT_LIST_HEAD(&splice);
4653
4654         spin_lock(&fs_info->ordered_root_lock);
4655         list_splice_init(&fs_info->ordered_roots, &splice);
4656         while (!list_empty(&splice)) {
4657                 root = list_first_entry(&splice, struct btrfs_root,
4658                                         ordered_root);
4659                 list_move_tail(&root->ordered_root,
4660                                &fs_info->ordered_roots);
4661
4662                 spin_unlock(&fs_info->ordered_root_lock);
4663                 btrfs_destroy_ordered_extents(root);
4664
4665                 cond_resched();
4666                 spin_lock(&fs_info->ordered_root_lock);
4667         }
4668         spin_unlock(&fs_info->ordered_root_lock);
4669
4670         /*
4671          * We need this here because if we've been flipped read-only we won't
4672          * get sync() from the umount, so we need to make sure any ordered
4673          * extents that haven't had their dirty pages IO start writeout yet
4674          * actually get run and error out properly.
4675          */
4676         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4677 }
4678
4679 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4680                                       struct btrfs_fs_info *fs_info)
4681 {
4682         struct rb_node *node;
4683         struct btrfs_delayed_ref_root *delayed_refs;
4684         struct btrfs_delayed_ref_node *ref;
4685         int ret = 0;
4686
4687         delayed_refs = &trans->delayed_refs;
4688
4689         spin_lock(&delayed_refs->lock);
4690         if (atomic_read(&delayed_refs->num_entries) == 0) {
4691                 spin_unlock(&delayed_refs->lock);
4692                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4693                 return ret;
4694         }
4695
4696         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4697                 struct btrfs_delayed_ref_head *head;
4698                 struct rb_node *n;
4699                 bool pin_bytes = false;
4700
4701                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4702                                 href_node);
4703                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4704                         continue;
4705
4706                 spin_lock(&head->lock);
4707                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4708                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4709                                        ref_node);
4710                         ref->in_tree = 0;
4711                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4712                         RB_CLEAR_NODE(&ref->ref_node);
4713                         if (!list_empty(&ref->add_list))
4714                                 list_del(&ref->add_list);
4715                         atomic_dec(&delayed_refs->num_entries);
4716                         btrfs_put_delayed_ref(ref);
4717                 }
4718                 if (head->must_insert_reserved)
4719                         pin_bytes = true;
4720                 btrfs_free_delayed_extent_op(head->extent_op);
4721                 btrfs_delete_ref_head(delayed_refs, head);
4722                 spin_unlock(&head->lock);
4723                 spin_unlock(&delayed_refs->lock);
4724                 mutex_unlock(&head->mutex);
4725
4726                 if (pin_bytes) {
4727                         struct btrfs_block_group *cache;
4728
4729                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4730                         BUG_ON(!cache);
4731
4732                         spin_lock(&cache->space_info->lock);
4733                         spin_lock(&cache->lock);
4734                         cache->pinned += head->num_bytes;
4735                         btrfs_space_info_update_bytes_pinned(fs_info,
4736                                 cache->space_info, head->num_bytes);
4737                         cache->reserved -= head->num_bytes;
4738                         cache->space_info->bytes_reserved -= head->num_bytes;
4739                         spin_unlock(&cache->lock);
4740                         spin_unlock(&cache->space_info->lock);
4741
4742                         btrfs_put_block_group(cache);
4743
4744                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4745                                 head->bytenr + head->num_bytes - 1);
4746                 }
4747                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4748                 btrfs_put_delayed_ref_head(head);
4749                 cond_resched();
4750                 spin_lock(&delayed_refs->lock);
4751         }
4752         btrfs_qgroup_destroy_extent_records(trans);
4753
4754         spin_unlock(&delayed_refs->lock);
4755
4756         return ret;
4757 }
4758
4759 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4760 {
4761         struct btrfs_inode *btrfs_inode;
4762         struct list_head splice;
4763
4764         INIT_LIST_HEAD(&splice);
4765
4766         spin_lock(&root->delalloc_lock);
4767         list_splice_init(&root->delalloc_inodes, &splice);
4768
4769         while (!list_empty(&splice)) {
4770                 struct inode *inode = NULL;
4771                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4772                                                delalloc_inodes);
4773                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4774                 spin_unlock(&root->delalloc_lock);
4775
4776                 /*
4777                  * Make sure we get a live inode and that it'll not disappear
4778                  * meanwhile.
4779                  */
4780                 inode = igrab(&btrfs_inode->vfs_inode);
4781                 if (inode) {
4782                         invalidate_inode_pages2(inode->i_mapping);
4783                         iput(inode);
4784                 }
4785                 spin_lock(&root->delalloc_lock);
4786         }
4787         spin_unlock(&root->delalloc_lock);
4788 }
4789
4790 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4791 {
4792         struct btrfs_root *root;
4793         struct list_head splice;
4794
4795         INIT_LIST_HEAD(&splice);
4796
4797         spin_lock(&fs_info->delalloc_root_lock);
4798         list_splice_init(&fs_info->delalloc_roots, &splice);
4799         while (!list_empty(&splice)) {
4800                 root = list_first_entry(&splice, struct btrfs_root,
4801                                          delalloc_root);
4802                 root = btrfs_grab_root(root);
4803                 BUG_ON(!root);
4804                 spin_unlock(&fs_info->delalloc_root_lock);
4805
4806                 btrfs_destroy_delalloc_inodes(root);
4807                 btrfs_put_root(root);
4808
4809                 spin_lock(&fs_info->delalloc_root_lock);
4810         }
4811         spin_unlock(&fs_info->delalloc_root_lock);
4812 }
4813
4814 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4815                                         struct extent_io_tree *dirty_pages,
4816                                         int mark)
4817 {
4818         int ret;
4819         struct extent_buffer *eb;
4820         u64 start = 0;
4821         u64 end;
4822
4823         while (1) {
4824                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4825                                             mark, NULL);
4826                 if (ret)
4827                         break;
4828
4829                 clear_extent_bits(dirty_pages, start, end, mark);
4830                 while (start <= end) {
4831                         eb = find_extent_buffer(fs_info, start);
4832                         start += fs_info->nodesize;
4833                         if (!eb)
4834                                 continue;
4835                         wait_on_extent_buffer_writeback(eb);
4836
4837                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4838                                                &eb->bflags))
4839                                 clear_extent_buffer_dirty(eb);
4840                         free_extent_buffer_stale(eb);
4841                 }
4842         }
4843
4844         return ret;
4845 }
4846
4847 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4848                                        struct extent_io_tree *unpin)
4849 {
4850         u64 start;
4851         u64 end;
4852         int ret;
4853
4854         while (1) {
4855                 struct extent_state *cached_state = NULL;
4856
4857                 /*
4858                  * The btrfs_finish_extent_commit() may get the same range as
4859                  * ours between find_first_extent_bit and clear_extent_dirty.
4860                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4861                  * the same extent range.
4862                  */
4863                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4864                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4865                                             EXTENT_DIRTY, &cached_state);
4866                 if (ret) {
4867                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4868                         break;
4869                 }
4870
4871                 clear_extent_dirty(unpin, start, end, &cached_state);
4872                 free_extent_state(cached_state);
4873                 btrfs_error_unpin_extent_range(fs_info, start, end);
4874                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4875                 cond_resched();
4876         }
4877
4878         return 0;
4879 }
4880
4881 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4882 {
4883         struct inode *inode;
4884
4885         inode = cache->io_ctl.inode;
4886         if (inode) {
4887                 invalidate_inode_pages2(inode->i_mapping);
4888                 BTRFS_I(inode)->generation = 0;
4889                 cache->io_ctl.inode = NULL;
4890                 iput(inode);
4891         }
4892         ASSERT(cache->io_ctl.pages == NULL);
4893         btrfs_put_block_group(cache);
4894 }
4895
4896 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4897                              struct btrfs_fs_info *fs_info)
4898 {
4899         struct btrfs_block_group *cache;
4900
4901         spin_lock(&cur_trans->dirty_bgs_lock);
4902         while (!list_empty(&cur_trans->dirty_bgs)) {
4903                 cache = list_first_entry(&cur_trans->dirty_bgs,
4904                                          struct btrfs_block_group,
4905                                          dirty_list);
4906
4907                 if (!list_empty(&cache->io_list)) {
4908                         spin_unlock(&cur_trans->dirty_bgs_lock);
4909                         list_del_init(&cache->io_list);
4910                         btrfs_cleanup_bg_io(cache);
4911                         spin_lock(&cur_trans->dirty_bgs_lock);
4912                 }
4913
4914                 list_del_init(&cache->dirty_list);
4915                 spin_lock(&cache->lock);
4916                 cache->disk_cache_state = BTRFS_DC_ERROR;
4917                 spin_unlock(&cache->lock);
4918
4919                 spin_unlock(&cur_trans->dirty_bgs_lock);
4920                 btrfs_put_block_group(cache);
4921                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4922                 spin_lock(&cur_trans->dirty_bgs_lock);
4923         }
4924         spin_unlock(&cur_trans->dirty_bgs_lock);
4925
4926         /*
4927          * Refer to the definition of io_bgs member for details why it's safe
4928          * to use it without any locking
4929          */
4930         while (!list_empty(&cur_trans->io_bgs)) {
4931                 cache = list_first_entry(&cur_trans->io_bgs,
4932                                          struct btrfs_block_group,
4933                                          io_list);
4934
4935                 list_del_init(&cache->io_list);
4936                 spin_lock(&cache->lock);
4937                 cache->disk_cache_state = BTRFS_DC_ERROR;
4938                 spin_unlock(&cache->lock);
4939                 btrfs_cleanup_bg_io(cache);
4940         }
4941 }
4942
4943 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4944                                    struct btrfs_fs_info *fs_info)
4945 {
4946         struct btrfs_device *dev, *tmp;
4947
4948         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4949         ASSERT(list_empty(&cur_trans->dirty_bgs));
4950         ASSERT(list_empty(&cur_trans->io_bgs));
4951
4952         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4953                                  post_commit_list) {
4954                 list_del_init(&dev->post_commit_list);
4955         }
4956
4957         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4958
4959         cur_trans->state = TRANS_STATE_COMMIT_START;
4960         wake_up(&fs_info->transaction_blocked_wait);
4961
4962         cur_trans->state = TRANS_STATE_UNBLOCKED;
4963         wake_up(&fs_info->transaction_wait);
4964
4965         btrfs_destroy_delayed_inodes(fs_info);
4966
4967         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4968                                      EXTENT_DIRTY);
4969         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4970
4971         btrfs_free_redirty_list(cur_trans);
4972
4973         cur_trans->state =TRANS_STATE_COMPLETED;
4974         wake_up(&cur_trans->commit_wait);
4975 }
4976
4977 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4978 {
4979         struct btrfs_transaction *t;
4980
4981         mutex_lock(&fs_info->transaction_kthread_mutex);
4982
4983         spin_lock(&fs_info->trans_lock);
4984         while (!list_empty(&fs_info->trans_list)) {
4985                 t = list_first_entry(&fs_info->trans_list,
4986                                      struct btrfs_transaction, list);
4987                 if (t->state >= TRANS_STATE_COMMIT_START) {
4988                         refcount_inc(&t->use_count);
4989                         spin_unlock(&fs_info->trans_lock);
4990                         btrfs_wait_for_commit(fs_info, t->transid);
4991                         btrfs_put_transaction(t);
4992                         spin_lock(&fs_info->trans_lock);
4993                         continue;
4994                 }
4995                 if (t == fs_info->running_transaction) {
4996                         t->state = TRANS_STATE_COMMIT_DOING;
4997                         spin_unlock(&fs_info->trans_lock);
4998                         /*
4999                          * We wait for 0 num_writers since we don't hold a trans
5000                          * handle open currently for this transaction.
5001                          */
5002                         wait_event(t->writer_wait,
5003                                    atomic_read(&t->num_writers) == 0);
5004                 } else {
5005                         spin_unlock(&fs_info->trans_lock);
5006                 }
5007                 btrfs_cleanup_one_transaction(t, fs_info);
5008
5009                 spin_lock(&fs_info->trans_lock);
5010                 if (t == fs_info->running_transaction)
5011                         fs_info->running_transaction = NULL;
5012                 list_del_init(&t->list);
5013                 spin_unlock(&fs_info->trans_lock);
5014
5015                 btrfs_put_transaction(t);
5016                 trace_btrfs_transaction_commit(fs_info->tree_root);
5017                 spin_lock(&fs_info->trans_lock);
5018         }
5019         spin_unlock(&fs_info->trans_lock);
5020         btrfs_destroy_all_ordered_extents(fs_info);
5021         btrfs_destroy_delayed_inodes(fs_info);
5022         btrfs_assert_delayed_root_empty(fs_info);
5023         btrfs_destroy_all_delalloc_inodes(fs_info);
5024         btrfs_drop_all_logs(fs_info);
5025         mutex_unlock(&fs_info->transaction_kthread_mutex);
5026
5027         return 0;
5028 }
5029
5030 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5031 {
5032         struct btrfs_path *path;
5033         int ret;
5034         struct extent_buffer *l;
5035         struct btrfs_key search_key;
5036         struct btrfs_key found_key;
5037         int slot;
5038
5039         path = btrfs_alloc_path();
5040         if (!path)
5041                 return -ENOMEM;
5042
5043         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5044         search_key.type = -1;
5045         search_key.offset = (u64)-1;
5046         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5047         if (ret < 0)
5048                 goto error;
5049         BUG_ON(ret == 0); /* Corruption */
5050         if (path->slots[0] > 0) {
5051                 slot = path->slots[0] - 1;
5052                 l = path->nodes[0];
5053                 btrfs_item_key_to_cpu(l, &found_key, slot);
5054                 root->free_objectid = max_t(u64, found_key.objectid + 1,
5055                                             BTRFS_FIRST_FREE_OBJECTID);
5056         } else {
5057                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5058         }
5059         ret = 0;
5060 error:
5061         btrfs_free_path(path);
5062         return ret;
5063 }
5064
5065 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5066 {
5067         int ret;
5068         mutex_lock(&root->objectid_mutex);
5069
5070         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5071                 btrfs_warn(root->fs_info,
5072                            "the objectid of root %llu reaches its highest value",
5073                            root->root_key.objectid);
5074                 ret = -ENOSPC;
5075                 goto out;
5076         }
5077
5078         *objectid = root->free_objectid++;
5079         ret = 0;
5080 out:
5081         mutex_unlock(&root->objectid_mutex);
5082         return ret;
5083 }