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 btrfs_destroy_ordered_extents(struct btrfs_root *root);
  55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  56                                       struct btrfs_fs_info *fs_info);
  57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  65
  66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  67 {
  68         if (fs_info->csum_shash)
  69                 crypto_free_shash(fs_info->csum_shash);
  70 }
  71
  72 /*
  73  * async submit bios are used to offload expensive checksumming
  74  * onto the worker threads.  They checksum file and metadata bios
  75  * just before they are sent down the IO stack.
  76  */
  77 struct async_submit_bio {
  78         struct inode *inode;
  79         struct bio *bio;
  80         extent_submit_bio_start_t *submit_bio_start;
  81         int mirror_num;
  82
  83         /* Optional parameter for submit_bio_start used by direct io */
  84         u64 dio_file_offset;
  85         struct btrfs_work work;
  86         blk_status_t status;
  87 };
  88
  89 /*
  90  * Compute the csum of a btree block and store the result to provided buffer.
  91  */
  92 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  93 {
  94         struct btrfs_fs_info *fs_info = buf->fs_info;
  95         const int num_pages = num_extent_pages(buf);
  96         const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  97         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  98         char *kaddr;
  99         int i;
 100
 101         shash->tfm = fs_info->csum_shash;
 102         crypto_shash_init(shash);
 103         kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
 104         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
 105                             first_page_part - BTRFS_CSUM_SIZE);
 106
 107         for (i = 1; i < num_pages; i++) {
 108                 kaddr = page_address(buf->pages[i]);
 109                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 110         }
 111         memset(result, 0, BTRFS_CSUM_SIZE);
 112         crypto_shash_final(shash, result);
 113 }
 114
 115 /*
 116  * we can't consider a given block up to date unless the transid of the
 117  * block matches the transid in the parent node's pointer.  This is how we
 118  * detect blocks that either didn't get written at all or got written
 119  * in the wrong place.
 120  */
 121 static int verify_parent_transid(struct extent_io_tree *io_tree,
 122                                  struct extent_buffer *eb, u64 parent_transid,
 123                                  int atomic)
 124 {
 125         struct extent_state *cached_state = NULL;
 126         int ret;
 127
 128         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 129                 return 0;
 130
 131         if (atomic)
 132                 return -EAGAIN;
 133
 134         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 135                          &cached_state);
 136         if (extent_buffer_uptodate(eb) &&
 137             btrfs_header_generation(eb) == parent_transid) {
 138                 ret = 0;
 139                 goto out;
 140         }
 141         btrfs_err_rl(eb->fs_info,
 142 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 143                         eb->start, eb->read_mirror,
 144                         parent_transid, btrfs_header_generation(eb));
 145         ret = 1;
 146         clear_extent_buffer_uptodate(eb);
 147 out:
 148         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 149                              &cached_state);
 150         return ret;
 151 }
 152
 153 static bool btrfs_supported_super_csum(u16 csum_type)
 154 {
 155         switch (csum_type) {
 156         case BTRFS_CSUM_TYPE_CRC32:
 157         case BTRFS_CSUM_TYPE_XXHASH:
 158         case BTRFS_CSUM_TYPE_SHA256:
 159         case BTRFS_CSUM_TYPE_BLAKE2:
 160                 return true;
 161         default:
 162                 return false;
 163         }
 164 }
 165
 166 /*
 167  * Return 0 if the superblock checksum type matches the checksum value of that
 168  * algorithm. Pass the raw disk superblock data.
 169  */
 170 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 171                                   char *raw_disk_sb)
 172 {
 173         struct btrfs_super_block *disk_sb =
 174                 (struct btrfs_super_block *)raw_disk_sb;
 175         char result[BTRFS_CSUM_SIZE];
 176         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 177
 178         shash->tfm = fs_info->csum_shash;
 179
 180         /*
 181          * The super_block structure does not span the whole
 182          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 183          * filled with zeros and is included in the checksum.
 184          */
 185         crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
 186                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 187
 188         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 189                 return 1;
 190
 191         return 0;
 192 }
 193
 194 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 195                            struct btrfs_key *first_key, u64 parent_transid)
 196 {
 197         struct btrfs_fs_info *fs_info = eb->fs_info;
 198         int found_level;
 199         struct btrfs_key found_key;
 200         int ret;
 201
 202         found_level = btrfs_header_level(eb);
 203         if (found_level != level) {
 204                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 205                      KERN_ERR "BTRFS: tree level check failed\n");
 206                 btrfs_err(fs_info,
 207 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 208                           eb->start, level, found_level);
 209                 return -EIO;
 210         }
 211
 212         if (!first_key)
 213                 return 0;
 214
 215         /*
 216          * For live tree block (new tree blocks in current transaction),
 217          * we need proper lock context to avoid race, which is impossible here.
 218          * So we only checks tree blocks which is read from disk, whose
 219          * generation <= fs_info->last_trans_committed.
 220          */
 221         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 222                 return 0;
 223
 224         /* We have @first_key, so this @eb must have at least one item */
 225         if (btrfs_header_nritems(eb) == 0) {
 226                 btrfs_err(fs_info,
 227                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 228                           eb->start);
 229                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 230                 return -EUCLEAN;
 231         }
 232
 233         if (found_level)
 234                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 235         else
 236                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 237         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 238
 239         if (ret) {
 240                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 241                      KERN_ERR "BTRFS: tree first key check failed\n");
 242                 btrfs_err(fs_info,
 243 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 244                           eb->start, parent_transid, first_key->objectid,
 245                           first_key->type, first_key->offset,
 246                           found_key.objectid, found_key.type,
 247                           found_key.offset);
 248         }
 249         return ret;
 250 }
 251
 252 /*
 253  * helper to read a given tree block, doing retries as required when
 254  * the checksums don't match and we have alternate mirrors to try.
 255  *
 256  * @parent_transid:     expected transid, skip check if 0
 257  * @level:              expected level, mandatory check
 258  * @first_key:          expected key of first slot, skip check if NULL
 259  */
 260 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 261                              u64 parent_transid, int level,
 262                              struct btrfs_key *first_key)
 263 {
 264         struct btrfs_fs_info *fs_info = eb->fs_info;
 265         struct extent_io_tree *io_tree;
 266         int failed = 0;
 267         int ret;
 268         int num_copies = 0;
 269         int mirror_num = 0;
 270         int failed_mirror = 0;
 271
 272         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 273         while (1) {
 274                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 275                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 276                 if (!ret) {
 277                         if (verify_parent_transid(io_tree, eb,
 278                                                    parent_transid, 0))
 279                                 ret = -EIO;
 280                         else if (btrfs_verify_level_key(eb, level,
 281                                                 first_key, parent_transid))
 282                                 ret = -EUCLEAN;
 283                         else
 284                                 break;
 285                 }
 286
 287                 num_copies = btrfs_num_copies(fs_info,
 288                                               eb->start, eb->len);
 289                 if (num_copies == 1)
 290                         break;
 291
 292                 if (!failed_mirror) {
 293                         failed = 1;
 294                         failed_mirror = eb->read_mirror;
 295                 }
 296
 297                 mirror_num++;
 298                 if (mirror_num == failed_mirror)
 299                         mirror_num++;
 300
 301                 if (mirror_num > num_copies)
 302                         break;
 303         }
 304
 305         if (failed && !ret && failed_mirror)
 306                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 307
 308         return ret;
 309 }
 310
 311 static int csum_one_extent_buffer(struct extent_buffer *eb)
 312 {
 313         struct btrfs_fs_info *fs_info = eb->fs_info;
 314         u8 result[BTRFS_CSUM_SIZE];
 315         int ret;
 316
 317         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 318                                     offsetof(struct btrfs_header, fsid),
 319                                     BTRFS_FSID_SIZE) == 0);
 320         csum_tree_block(eb, result);
 321
 322         if (btrfs_header_level(eb))
 323                 ret = btrfs_check_node(eb);
 324         else
 325                 ret = btrfs_check_leaf_full(eb);
 326
 327         if (ret < 0)
 328                 goto error;
 329
 330         /*
 331          * Also check the generation, the eb reached here must be newer than
 332          * last committed. Or something seriously wrong happened.
 333          */
 334         if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
 335                 ret = -EUCLEAN;
 336                 btrfs_err(fs_info,
 337                         "block=%llu bad generation, have %llu expect > %llu",
 338                           eb->start, btrfs_header_generation(eb),
 339                           fs_info->last_trans_committed);
 340                 goto error;
 341         }
 342         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 343
 344         return 0;
 345
 346 error:
 347         btrfs_print_tree(eb, 0);
 348         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 349                   eb->start);
 350         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 351         return ret;
 352 }
 353
 354 /* Checksum all dirty extent buffers in one bio_vec */
 355 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
 356                                       struct bio_vec *bvec)
 357 {
 358         struct page *page = bvec->bv_page;
 359         u64 bvec_start = page_offset(page) + bvec->bv_offset;
 360         u64 cur;
 361         int ret = 0;
 362
 363         for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
 364              cur += fs_info->nodesize) {
 365                 struct extent_buffer *eb;
 366                 bool uptodate;
 367
 368                 eb = find_extent_buffer(fs_info, cur);
 369                 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
 370                                                        fs_info->nodesize);
 371
 372                 /* A dirty eb shouldn't disappear from buffer_radix */
 373                 if (WARN_ON(!eb))
 374                         return -EUCLEAN;
 375
 376                 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
 377                         free_extent_buffer(eb);
 378                         return -EUCLEAN;
 379                 }
 380                 if (WARN_ON(!uptodate)) {
 381                         free_extent_buffer(eb);
 382                         return -EUCLEAN;
 383                 }
 384
 385                 ret = csum_one_extent_buffer(eb);
 386                 free_extent_buffer(eb);
 387                 if (ret < 0)
 388                         return ret;
 389         }
 390         return ret;
 391 }
 392
 393 /*
 394  * Checksum a dirty tree block before IO.  This has extra checks to make sure
 395  * we only fill in the checksum field in the first page of a multi-page block.
 396  * For subpage extent buffers we need bvec to also read the offset in the page.
 397  */
 398 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
 399 {
 400         struct page *page = bvec->bv_page;
 401         u64 start = page_offset(page);
 402         u64 found_start;
 403         struct extent_buffer *eb;
 404
 405         if (fs_info->nodesize < PAGE_SIZE)
 406                 return csum_dirty_subpage_buffers(fs_info, bvec);
 407
 408         eb = (struct extent_buffer *)page->private;
 409         if (page != eb->pages[0])
 410                 return 0;
 411
 412         found_start = btrfs_header_bytenr(eb);
 413
 414         if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
 415                 WARN_ON(found_start != 0);
 416                 return 0;
 417         }
 418
 419         /*
 420          * Please do not consolidate these warnings into a single if.
 421          * It is useful to know what went wrong.
 422          */
 423         if (WARN_ON(found_start != start))
 424                 return -EUCLEAN;
 425         if (WARN_ON(!PageUptodate(page)))
 426                 return -EUCLEAN;
 427
 428         return csum_one_extent_buffer(eb);
 429 }
 430
 431 static int check_tree_block_fsid(struct extent_buffer *eb)
 432 {
 433         struct btrfs_fs_info *fs_info = eb->fs_info;
 434         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 435         u8 fsid[BTRFS_FSID_SIZE];
 436         u8 *metadata_uuid;
 437
 438         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 439                            BTRFS_FSID_SIZE);
 440         /*
 441          * Checking the incompat flag is only valid for the current fs. For
 442          * seed devices it's forbidden to have their uuid changed so reading
 443          * ->fsid in this case is fine
 444          */
 445         if (btrfs_fs_incompat(fs_info, METADATA_UUID))
 446                 metadata_uuid = fs_devices->metadata_uuid;
 447         else
 448                 metadata_uuid = fs_devices->fsid;
 449
 450         if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
 451                 return 0;
 452
 453         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 454                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 455                         return 0;
 456
 457         return 1;
 458 }
 459
 460 /* Do basic extent buffer checks at read time */
 461 static int validate_extent_buffer(struct extent_buffer *eb)
 462 {
 463         struct btrfs_fs_info *fs_info = eb->fs_info;
 464         u64 found_start;
 465         const u32 csum_size = fs_info->csum_size;
 466         u8 found_level;
 467         u8 result[BTRFS_CSUM_SIZE];
 468         const u8 *header_csum;
 469         int ret = 0;
 470
 471         found_start = btrfs_header_bytenr(eb);
 472         if (found_start != eb->start) {
 473                 btrfs_err_rl(fs_info,
 474                         "bad tree block start, mirror %u want %llu have %llu",
 475                              eb->read_mirror, eb->start, found_start);
 476                 ret = -EIO;
 477                 goto out;
 478         }
 479         if (check_tree_block_fsid(eb)) {
 480                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 481                              eb->start, eb->read_mirror);
 482                 ret = -EIO;
 483                 goto out;
 484         }
 485         found_level = btrfs_header_level(eb);
 486         if (found_level >= BTRFS_MAX_LEVEL) {
 487                 btrfs_err(fs_info,
 488                         "bad tree block level, mirror %u level %d on logical %llu",
 489                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 490                 ret = -EIO;
 491                 goto out;
 492         }
 493
 494         csum_tree_block(eb, result);
 495         header_csum = page_address(eb->pages[0]) +
 496                 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
 497
 498         if (memcmp(result, header_csum, csum_size) != 0) {
 499                 btrfs_warn_rl(fs_info,
 500 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 501                               eb->start, eb->read_mirror,
 502                               CSUM_FMT_VALUE(csum_size, header_csum),
 503                               CSUM_FMT_VALUE(csum_size, result),
 504                               btrfs_header_level(eb));
 505                 ret = -EUCLEAN;
 506                 goto out;
 507         }
 508
 509         /*
 510          * If this is a leaf block and it is corrupt, set the corrupt bit so
 511          * that we don't try and read the other copies of this block, just
 512          * return -EIO.
 513          */
 514         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 515                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 516                 ret = -EIO;
 517         }
 518
 519         if (found_level > 0 && btrfs_check_node(eb))
 520                 ret = -EIO;
 521
 522         if (!ret)
 523                 set_extent_buffer_uptodate(eb);
 524         else
 525                 btrfs_err(fs_info,
 526                 "read time tree block corruption detected on logical %llu mirror %u",
 527                           eb->start, eb->read_mirror);
 528 out:
 529         return ret;
 530 }
 531
 532 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
 533                                    int mirror)
 534 {
 535         struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
 536         struct extent_buffer *eb;
 537         bool reads_done;
 538         int ret = 0;
 539
 540         /*
 541          * We don't allow bio merge for subpage metadata read, so we should
 542          * only get one eb for each endio hook.
 543          */
 544         ASSERT(end == start + fs_info->nodesize - 1);
 545         ASSERT(PagePrivate(page));
 546
 547         eb = find_extent_buffer(fs_info, start);
 548         /*
 549          * When we are reading one tree block, eb must have been inserted into
 550          * the radix tree. If not, something is wrong.
 551          */
 552         ASSERT(eb);
 553
 554         reads_done = atomic_dec_and_test(&eb->io_pages);
 555         /* Subpage read must finish in page read */
 556         ASSERT(reads_done);
 557
 558         eb->read_mirror = mirror;
 559         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 560                 ret = -EIO;
 561                 goto err;
 562         }
 563         ret = validate_extent_buffer(eb);
 564         if (ret < 0)
 565                 goto err;
 566
 567         set_extent_buffer_uptodate(eb);
 568
 569         free_extent_buffer(eb);
 570         return ret;
 571 err:
 572         /*
 573          * end_bio_extent_readpage decrements io_pages in case of error,
 574          * make sure it has something to decrement.
 575          */
 576         atomic_inc(&eb->io_pages);
 577         clear_extent_buffer_uptodate(eb);
 578         free_extent_buffer(eb);
 579         return ret;
 580 }
 581
 582 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
 583                                    struct page *page, u64 start, u64 end,
 584                                    int mirror)
 585 {
 586         struct extent_buffer *eb;
 587         int ret = 0;
 588         int reads_done;
 589
 590         ASSERT(page->private);
 591
 592         if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
 593                 return validate_subpage_buffer(page, start, end, mirror);
 594
 595         eb = (struct extent_buffer *)page->private;
 596
 597         /*
 598          * The pending IO might have been the only thing that kept this buffer
 599          * in memory.  Make sure we have a ref for all this other checks
 600          */
 601         atomic_inc(&eb->refs);
 602
 603         reads_done = atomic_dec_and_test(&eb->io_pages);
 604         if (!reads_done)
 605                 goto err;
 606
 607         eb->read_mirror = mirror;
 608         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 609                 ret = -EIO;
 610                 goto err;
 611         }
 612         ret = validate_extent_buffer(eb);
 613 err:
 614         if (ret) {
 615                 /*
 616                  * our io error hook is going to dec the io pages
 617                  * again, we have to make sure it has something
 618                  * to decrement
 619                  */
 620                 atomic_inc(&eb->io_pages);
 621                 clear_extent_buffer_uptodate(eb);
 622         }
 623         free_extent_buffer(eb);
 624
 625         return ret;
 626 }
 627
 628 static void run_one_async_start(struct btrfs_work *work)
 629 {
 630         struct async_submit_bio *async;
 631         blk_status_t ret;
 632
 633         async = container_of(work, struct  async_submit_bio, work);
 634         ret = async->submit_bio_start(async->inode, async->bio,
 635                                       async->dio_file_offset);
 636         if (ret)
 637                 async->status = ret;
 638 }
 639
 640 /*
 641  * In order to insert checksums into the metadata in large chunks, we wait
 642  * until bio submission time.   All the pages in the bio are checksummed and
 643  * sums are attached onto the ordered extent record.
 644  *
 645  * At IO completion time the csums attached on the ordered extent record are
 646  * inserted into the tree.
 647  */
 648 static void run_one_async_done(struct btrfs_work *work)
 649 {
 650         struct async_submit_bio *async;
 651         struct inode *inode;
 652
 653         async = container_of(work, struct  async_submit_bio, work);
 654         inode = async->inode;
 655
 656         /* If an error occurred we just want to clean up the bio and move on */
 657         if (async->status) {
 658                 async->bio->bi_status = async->status;
 659                 bio_endio(async->bio);
 660                 return;
 661         }
 662
 663         /*
 664          * All of the bios that pass through here are from async helpers.
 665          * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
 666          * This changes nothing when cgroups aren't in use.
 667          */
 668         async->bio->bi_opf |= REQ_CGROUP_PUNT;
 669         btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
 670 }
 671
 672 static void run_one_async_free(struct btrfs_work *work)
 673 {
 674         struct async_submit_bio *async;
 675
 676         async = container_of(work, struct  async_submit_bio, work);
 677         kfree(async);
 678 }
 679
 680 /*
 681  * Submit bio to an async queue.
 682  *
 683  * Retrun:
 684  * - true if the work has been succesfuly submitted
 685  * - false in case of error
 686  */
 687 bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num,
 688                          u64 dio_file_offset,
 689                          extent_submit_bio_start_t *submit_bio_start)
 690 {
 691         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
 692         struct async_submit_bio *async;
 693
 694         async = kmalloc(sizeof(*async), GFP_NOFS);
 695         if (!async)
 696                 return false;
 697
 698         async->inode = inode;
 699         async->bio = bio;
 700         async->mirror_num = mirror_num;
 701         async->submit_bio_start = submit_bio_start;
 702
 703         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 704                         run_one_async_free);
 705
 706         async->dio_file_offset = dio_file_offset;
 707
 708         async->status = 0;
 709
 710         if (op_is_sync(bio->bi_opf))
 711                 btrfs_queue_work(fs_info->hipri_workers, &async->work);
 712         else
 713                 btrfs_queue_work(fs_info->workers, &async->work);
 714         return true;
 715 }
 716
 717 static blk_status_t btree_csum_one_bio(struct bio *bio)
 718 {
 719         struct bio_vec *bvec;
 720         struct btrfs_root *root;
 721         int ret = 0;
 722         struct bvec_iter_all iter_all;
 723
 724         ASSERT(!bio_flagged(bio, BIO_CLONED));
 725         bio_for_each_segment_all(bvec, bio, iter_all) {
 726                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 727                 ret = csum_dirty_buffer(root->fs_info, bvec);
 728                 if (ret)
 729                         break;
 730         }
 731
 732         return errno_to_blk_status(ret);
 733 }
 734
 735 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
 736                                            u64 dio_file_offset)
 737 {
 738         /*
 739          * when we're called for a write, we're already in the async
 740          * submission context.  Just jump into btrfs_submit_bio.
 741          */
 742         return btree_csum_one_bio(bio);
 743 }
 744
 745 static bool should_async_write(struct btrfs_fs_info *fs_info,
 746                              struct btrfs_inode *bi)
 747 {
 748         if (btrfs_is_zoned(fs_info))
 749                 return false;
 750         if (atomic_read(&bi->sync_writers))
 751                 return false;
 752         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 753                 return false;
 754         return true;
 755 }
 756
 757 void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num)
 758 {
 759         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 760         blk_status_t ret;
 761
 762         bio->bi_opf |= REQ_META;
 763
 764         if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
 765                 btrfs_submit_bio(fs_info, bio, mirror_num);
 766                 return;
 767         }
 768
 769         /*
 770          * Kthread helpers are used to submit writes so that checksumming can
 771          * happen in parallel across all CPUs.
 772          */
 773         if (should_async_write(fs_info, BTRFS_I(inode)) &&
 774             btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start))
 775                 return;
 776
 777         ret = btree_csum_one_bio(bio);
 778         if (ret) {
 779                 bio->bi_status = ret;
 780                 bio_endio(bio);
 781                 return;
 782         }
 783
 784         btrfs_submit_bio(fs_info, bio, mirror_num);
 785 }
 786
 787 #ifdef CONFIG_MIGRATION
 788 static int btree_migrate_folio(struct address_space *mapping,
 789                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 790 {
 791         /*
 792          * we can't safely write a btree page from here,
 793          * we haven't done the locking hook
 794          */
 795         if (folio_test_dirty(src))
 796                 return -EAGAIN;
 797         /*
 798          * Buffers may be managed in a filesystem specific way.
 799          * We must have no buffers or drop them.
 800          */
 801         if (folio_get_private(src) &&
 802             !filemap_release_folio(src, GFP_KERNEL))
 803                 return -EAGAIN;
 804         return migrate_folio(mapping, dst, src, mode);
 805 }
 806 #else
 807 #define btree_migrate_folio NULL
 808 #endif
 809
 810 static int btree_writepages(struct address_space *mapping,
 811                             struct writeback_control *wbc)
 812 {
 813         struct btrfs_fs_info *fs_info;
 814         int ret;
 815
 816         if (wbc->sync_mode == WB_SYNC_NONE) {
 817
 818                 if (wbc->for_kupdate)
 819                         return 0;
 820
 821                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 822                 /* this is a bit racy, but that's ok */
 823                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 824                                              BTRFS_DIRTY_METADATA_THRESH,
 825                                              fs_info->dirty_metadata_batch);
 826                 if (ret < 0)
 827                         return 0;
 828         }
 829         return btree_write_cache_pages(mapping, wbc);
 830 }
 831
 832 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 833 {
 834         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 835                 return false;
 836
 837         return try_release_extent_buffer(&folio->page);
 838 }
 839
 840 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 841                                  size_t length)
 842 {
 843         struct extent_io_tree *tree;
 844         tree = &BTRFS_I(folio->mapping->host)->io_tree;
 845         extent_invalidate_folio(tree, folio, offset);
 846         btree_release_folio(folio, GFP_NOFS);
 847         if (folio_get_private(folio)) {
 848                 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 849                            "folio private not zero on folio %llu",
 850                            (unsigned long long)folio_pos(folio));
 851                 folio_detach_private(folio);
 852         }
 853 }
 854
 855 #ifdef DEBUG
 856 static bool btree_dirty_folio(struct address_space *mapping,
 857                 struct folio *folio)
 858 {
 859         struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 860         struct btrfs_subpage *subpage;
 861         struct extent_buffer *eb;
 862         int cur_bit = 0;
 863         u64 page_start = folio_pos(folio);
 864
 865         if (fs_info->sectorsize == PAGE_SIZE) {
 866                 eb = folio_get_private(folio);
 867                 BUG_ON(!eb);
 868                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 869                 BUG_ON(!atomic_read(&eb->refs));
 870                 btrfs_assert_tree_write_locked(eb);
 871                 return filemap_dirty_folio(mapping, folio);
 872         }
 873         subpage = folio_get_private(folio);
 874
 875         ASSERT(subpage->dirty_bitmap);
 876         while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
 877                 unsigned long flags;
 878                 u64 cur;
 879                 u16 tmp = (1 << cur_bit);
 880
 881                 spin_lock_irqsave(&subpage->lock, flags);
 882                 if (!(tmp & subpage->dirty_bitmap)) {
 883                         spin_unlock_irqrestore(&subpage->lock, flags);
 884                         cur_bit++;
 885                         continue;
 886                 }
 887                 spin_unlock_irqrestore(&subpage->lock, flags);
 888                 cur = page_start + cur_bit * fs_info->sectorsize;
 889
 890                 eb = find_extent_buffer(fs_info, cur);
 891                 ASSERT(eb);
 892                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 893                 ASSERT(atomic_read(&eb->refs));
 894                 btrfs_assert_tree_write_locked(eb);
 895                 free_extent_buffer(eb);
 896
 897                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
 898         }
 899         return filemap_dirty_folio(mapping, folio);
 900 }
 901 #else
 902 #define btree_dirty_folio filemap_dirty_folio
 903 #endif
 904
 905 static const struct address_space_operations btree_aops = {
 906         .writepages     = btree_writepages,
 907         .release_folio  = btree_release_folio,
 908         .invalidate_folio = btree_invalidate_folio,
 909         .migrate_folio  = btree_migrate_folio,
 910         .dirty_folio    = btree_dirty_folio,
 911 };
 912
 913 struct extent_buffer *btrfs_find_create_tree_block(
 914                                                 struct btrfs_fs_info *fs_info,
 915                                                 u64 bytenr, u64 owner_root,
 916                                                 int level)
 917 {
 918         if (btrfs_is_testing(fs_info))
 919                 return alloc_test_extent_buffer(fs_info, bytenr);
 920         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 921 }
 922
 923 /*
 924  * Read tree block at logical address @bytenr and do variant basic but critical
 925  * verification.
 926  *
 927  * @owner_root:         the objectid of the root owner for this block.
 928  * @parent_transid:     expected transid of this tree block, skip check if 0
 929  * @level:              expected level, mandatory check
 930  * @first_key:          expected key in slot 0, skip check if NULL
 931  */
 932 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 933                                       u64 owner_root, u64 parent_transid,
 934                                       int level, struct btrfs_key *first_key)
 935 {
 936         struct extent_buffer *buf = NULL;
 937         int ret;
 938
 939         buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
 940         if (IS_ERR(buf))
 941                 return buf;
 942
 943         ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key);
 944         if (ret) {
 945                 free_extent_buffer_stale(buf);
 946                 return ERR_PTR(ret);
 947         }
 948         if (btrfs_check_eb_owner(buf, owner_root)) {
 949                 free_extent_buffer_stale(buf);
 950                 return ERR_PTR(-EUCLEAN);
 951         }
 952         return buf;
 953
 954 }
 955
 956 void btrfs_clean_tree_block(struct extent_buffer *buf)
 957 {
 958         struct btrfs_fs_info *fs_info = buf->fs_info;
 959         if (btrfs_header_generation(buf) ==
 960             fs_info->running_transaction->transid) {
 961                 btrfs_assert_tree_write_locked(buf);
 962
 963                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 964                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
 965                                                  -buf->len,
 966                                                  fs_info->dirty_metadata_batch);
 967                         clear_extent_buffer_dirty(buf);
 968                 }
 969         }
 970 }
 971
 972 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 973                          u64 objectid)
 974 {
 975         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 976
 977         memset(&root->root_key, 0, sizeof(root->root_key));
 978         memset(&root->root_item, 0, sizeof(root->root_item));
 979         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 980         root->fs_info = fs_info;
 981         root->root_key.objectid = objectid;
 982         root->node = NULL;
 983         root->commit_root = NULL;
 984         root->state = 0;
 985         RB_CLEAR_NODE(&root->rb_node);
 986
 987         root->last_trans = 0;
 988         root->free_objectid = 0;
 989         root->nr_delalloc_inodes = 0;
 990         root->nr_ordered_extents = 0;
 991         root->inode_tree = RB_ROOT;
 992         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
 993
 994         btrfs_init_root_block_rsv(root);
 995
 996         INIT_LIST_HEAD(&root->dirty_list);
 997         INIT_LIST_HEAD(&root->root_list);
 998         INIT_LIST_HEAD(&root->delalloc_inodes);
 999         INIT_LIST_HEAD(&root->delalloc_root);
1000         INIT_LIST_HEAD(&root->ordered_extents);
1001         INIT_LIST_HEAD(&root->ordered_root);
1002         INIT_LIST_HEAD(&root->reloc_dirty_list);
1003         INIT_LIST_HEAD(&root->logged_list[0]);
1004         INIT_LIST_HEAD(&root->logged_list[1]);
1005         spin_lock_init(&root->inode_lock);
1006         spin_lock_init(&root->delalloc_lock);
1007         spin_lock_init(&root->ordered_extent_lock);
1008         spin_lock_init(&root->accounting_lock);
1009         spin_lock_init(&root->log_extents_lock[0]);
1010         spin_lock_init(&root->log_extents_lock[1]);
1011         spin_lock_init(&root->qgroup_meta_rsv_lock);
1012         mutex_init(&root->objectid_mutex);
1013         mutex_init(&root->log_mutex);
1014         mutex_init(&root->ordered_extent_mutex);
1015         mutex_init(&root->delalloc_mutex);
1016         init_waitqueue_head(&root->qgroup_flush_wait);
1017         init_waitqueue_head(&root->log_writer_wait);
1018         init_waitqueue_head(&root->log_commit_wait[0]);
1019         init_waitqueue_head(&root->log_commit_wait[1]);
1020         INIT_LIST_HEAD(&root->log_ctxs[0]);
1021         INIT_LIST_HEAD(&root->log_ctxs[1]);
1022         atomic_set(&root->log_commit[0], 0);
1023         atomic_set(&root->log_commit[1], 0);
1024         atomic_set(&root->log_writers, 0);
1025         atomic_set(&root->log_batch, 0);
1026         refcount_set(&root->refs, 1);
1027         atomic_set(&root->snapshot_force_cow, 0);
1028         atomic_set(&root->nr_swapfiles, 0);
1029         root->log_transid = 0;
1030         root->log_transid_committed = -1;
1031         root->last_log_commit = 0;
1032         root->anon_dev = 0;
1033         if (!dummy) {
1034                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1035                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1036                 extent_io_tree_init(fs_info, &root->log_csum_range,
1037                                     IO_TREE_LOG_CSUM_RANGE, NULL);
1038         }
1039
1040         spin_lock_init(&root->root_item_lock);
1041         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1042 #ifdef CONFIG_BTRFS_DEBUG
1043         INIT_LIST_HEAD(&root->leak_list);
1044         spin_lock(&fs_info->fs_roots_radix_lock);
1045         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1046         spin_unlock(&fs_info->fs_roots_radix_lock);
1047 #endif
1048 }
1049
1050 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1051                                            u64 objectid, gfp_t flags)
1052 {
1053         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1054         if (root)
1055                 __setup_root(root, fs_info, objectid);
1056         return root;
1057 }
1058
1059 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1060 /* Should only be used by the testing infrastructure */
1061 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1062 {
1063         struct btrfs_root *root;
1064
1065         if (!fs_info)
1066                 return ERR_PTR(-EINVAL);
1067
1068         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1069         if (!root)
1070                 return ERR_PTR(-ENOMEM);
1071
1072         /* We don't use the stripesize in selftest, set it as sectorsize */
1073         root->alloc_bytenr = 0;
1074
1075         return root;
1076 }
1077 #endif
1078
1079 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1080 {
1081         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1082         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1083
1084         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1085 }
1086
1087 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1088 {
1089         const struct btrfs_key *key = k;
1090         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1091
1092         return btrfs_comp_cpu_keys(key, &root->root_key);
1093 }
1094
1095 int btrfs_global_root_insert(struct btrfs_root *root)
1096 {
1097         struct btrfs_fs_info *fs_info = root->fs_info;
1098         struct rb_node *tmp;
1099
1100         write_lock(&fs_info->global_root_lock);
1101         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1102         write_unlock(&fs_info->global_root_lock);
1103         ASSERT(!tmp);
1104
1105         return tmp ? -EEXIST : 0;
1106 }
1107
1108 void btrfs_global_root_delete(struct btrfs_root *root)
1109 {
1110         struct btrfs_fs_info *fs_info = root->fs_info;
1111
1112         write_lock(&fs_info->global_root_lock);
1113         rb_erase(&root->rb_node, &fs_info->global_root_tree);
1114         write_unlock(&fs_info->global_root_lock);
1115 }
1116
1117 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1118                                      struct btrfs_key *key)
1119 {
1120         struct rb_node *node;
1121         struct btrfs_root *root = NULL;
1122
1123         read_lock(&fs_info->global_root_lock);
1124         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1125         if (node)
1126                 root = container_of(node, struct btrfs_root, rb_node);
1127         read_unlock(&fs_info->global_root_lock);
1128
1129         return root;
1130 }
1131
1132 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1133 {
1134         struct btrfs_block_group *block_group;
1135         u64 ret;
1136
1137         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1138                 return 0;
1139
1140         if (bytenr)
1141                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1142         else
1143                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1144         ASSERT(block_group);
1145         if (!block_group)
1146                 return 0;
1147         ret = block_group->global_root_id;
1148         btrfs_put_block_group(block_group);
1149
1150         return ret;
1151 }
1152
1153 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1154 {
1155         struct btrfs_key key = {
1156                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1157                 .type = BTRFS_ROOT_ITEM_KEY,
1158                 .offset = btrfs_global_root_id(fs_info, bytenr),
1159         };
1160
1161         return btrfs_global_root(fs_info, &key);
1162 }
1163
1164 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1165 {
1166         struct btrfs_key key = {
1167                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1168                 .type = BTRFS_ROOT_ITEM_KEY,
1169                 .offset = btrfs_global_root_id(fs_info, bytenr),
1170         };
1171
1172         return btrfs_global_root(fs_info, &key);
1173 }
1174
1175 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1176                                      u64 objectid)
1177 {
1178         struct btrfs_fs_info *fs_info = trans->fs_info;
1179         struct extent_buffer *leaf;
1180         struct btrfs_root *tree_root = fs_info->tree_root;
1181         struct btrfs_root *root;
1182         struct btrfs_key key;
1183         unsigned int nofs_flag;
1184         int ret = 0;
1185
1186         /*
1187          * We're holding a transaction handle, so use a NOFS memory allocation
1188          * context to avoid deadlock if reclaim happens.
1189          */
1190         nofs_flag = memalloc_nofs_save();
1191         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1192         memalloc_nofs_restore(nofs_flag);
1193         if (!root)
1194                 return ERR_PTR(-ENOMEM);
1195
1196         root->root_key.objectid = objectid;
1197         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1198         root->root_key.offset = 0;
1199
1200         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1201                                       BTRFS_NESTING_NORMAL);
1202         if (IS_ERR(leaf)) {
1203                 ret = PTR_ERR(leaf);
1204                 leaf = NULL;
1205                 goto fail_unlock;
1206         }
1207
1208         root->node = leaf;
1209         btrfs_mark_buffer_dirty(leaf);
1210
1211         root->commit_root = btrfs_root_node(root);
1212         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1213
1214         btrfs_set_root_flags(&root->root_item, 0);
1215         btrfs_set_root_limit(&root->root_item, 0);
1216         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1217         btrfs_set_root_generation(&root->root_item, trans->transid);
1218         btrfs_set_root_level(&root->root_item, 0);
1219         btrfs_set_root_refs(&root->root_item, 1);
1220         btrfs_set_root_used(&root->root_item, leaf->len);
1221         btrfs_set_root_last_snapshot(&root->root_item, 0);
1222         btrfs_set_root_dirid(&root->root_item, 0);
1223         if (is_fstree(objectid))
1224                 generate_random_guid(root->root_item.uuid);
1225         else
1226                 export_guid(root->root_item.uuid, &guid_null);
1227         btrfs_set_root_drop_level(&root->root_item, 0);
1228
1229         btrfs_tree_unlock(leaf);
1230
1231         key.objectid = objectid;
1232         key.type = BTRFS_ROOT_ITEM_KEY;
1233         key.offset = 0;
1234         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1235         if (ret)
1236                 goto fail;
1237
1238         return root;
1239
1240 fail_unlock:
1241         if (leaf)
1242                 btrfs_tree_unlock(leaf);
1243 fail:
1244         btrfs_put_root(root);
1245
1246         return ERR_PTR(ret);
1247 }
1248
1249 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1250                                          struct btrfs_fs_info *fs_info)
1251 {
1252         struct btrfs_root *root;
1253
1254         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1255         if (!root)
1256                 return ERR_PTR(-ENOMEM);
1257
1258         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1259         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1260         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1261
1262         return root;
1263 }
1264
1265 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1266                               struct btrfs_root *root)
1267 {
1268         struct extent_buffer *leaf;
1269
1270         /*
1271          * DON'T set SHAREABLE bit for log trees.
1272          *
1273          * Log trees are not exposed to user space thus can't be snapshotted,
1274          * and they go away before a real commit is actually done.
1275          *
1276          * They do store pointers to file data extents, and those reference
1277          * counts still get updated (along with back refs to the log tree).
1278          */
1279
1280         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1281                         NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1282         if (IS_ERR(leaf))
1283                 return PTR_ERR(leaf);
1284
1285         root->node = leaf;
1286
1287         btrfs_mark_buffer_dirty(root->node);
1288         btrfs_tree_unlock(root->node);
1289
1290         return 0;
1291 }
1292
1293 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1294                              struct btrfs_fs_info *fs_info)
1295 {
1296         struct btrfs_root *log_root;
1297
1298         log_root = alloc_log_tree(trans, fs_info);
1299         if (IS_ERR(log_root))
1300                 return PTR_ERR(log_root);
1301
1302         if (!btrfs_is_zoned(fs_info)) {
1303                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1304
1305                 if (ret) {
1306                         btrfs_put_root(log_root);
1307                         return ret;
1308                 }
1309         }
1310
1311         WARN_ON(fs_info->log_root_tree);
1312         fs_info->log_root_tree = log_root;
1313         return 0;
1314 }
1315
1316 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1317                        struct btrfs_root *root)
1318 {
1319         struct btrfs_fs_info *fs_info = root->fs_info;
1320         struct btrfs_root *log_root;
1321         struct btrfs_inode_item *inode_item;
1322         int ret;
1323
1324         log_root = alloc_log_tree(trans, fs_info);
1325         if (IS_ERR(log_root))
1326                 return PTR_ERR(log_root);
1327
1328         ret = btrfs_alloc_log_tree_node(trans, log_root);
1329         if (ret) {
1330                 btrfs_put_root(log_root);
1331                 return ret;
1332         }
1333
1334         log_root->last_trans = trans->transid;
1335         log_root->root_key.offset = root->root_key.objectid;
1336
1337         inode_item = &log_root->root_item.inode;
1338         btrfs_set_stack_inode_generation(inode_item, 1);
1339         btrfs_set_stack_inode_size(inode_item, 3);
1340         btrfs_set_stack_inode_nlink(inode_item, 1);
1341         btrfs_set_stack_inode_nbytes(inode_item,
1342                                      fs_info->nodesize);
1343         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1344
1345         btrfs_set_root_node(&log_root->root_item, log_root->node);
1346
1347         WARN_ON(root->log_root);
1348         root->log_root = log_root;
1349         root->log_transid = 0;
1350         root->log_transid_committed = -1;
1351         root->last_log_commit = 0;
1352         return 0;
1353 }
1354
1355 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1356                                               struct btrfs_path *path,
1357                                               struct btrfs_key *key)
1358 {
1359         struct btrfs_root *root;
1360         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1361         u64 generation;
1362         int ret;
1363         int level;
1364
1365         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1366         if (!root)
1367                 return ERR_PTR(-ENOMEM);
1368
1369         ret = btrfs_find_root(tree_root, key, path,
1370                               &root->root_item, &root->root_key);
1371         if (ret) {
1372                 if (ret > 0)
1373                         ret = -ENOENT;
1374                 goto fail;
1375         }
1376
1377         generation = btrfs_root_generation(&root->root_item);
1378         level = btrfs_root_level(&root->root_item);
1379         root->node = read_tree_block(fs_info,
1380                                      btrfs_root_bytenr(&root->root_item),
1381                                      key->objectid, generation, level, NULL);
1382         if (IS_ERR(root->node)) {
1383                 ret = PTR_ERR(root->node);
1384                 root->node = NULL;
1385                 goto fail;
1386         }
1387         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1388                 ret = -EIO;
1389                 goto fail;
1390         }
1391
1392         /*
1393          * For real fs, and not log/reloc trees, root owner must
1394          * match its root node owner
1395          */
1396         if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1397             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1398             root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1399             root->root_key.objectid != btrfs_header_owner(root->node)) {
1400                 btrfs_crit(fs_info,
1401 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1402                            root->root_key.objectid, root->node->start,
1403                            btrfs_header_owner(root->node),
1404                            root->root_key.objectid);
1405                 ret = -EUCLEAN;
1406                 goto fail;
1407         }
1408         root->commit_root = btrfs_root_node(root);
1409         return root;
1410 fail:
1411         btrfs_put_root(root);
1412         return ERR_PTR(ret);
1413 }
1414
1415 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1416                                         struct btrfs_key *key)
1417 {
1418         struct btrfs_root *root;
1419         struct btrfs_path *path;
1420
1421         path = btrfs_alloc_path();
1422         if (!path)
1423                 return ERR_PTR(-ENOMEM);
1424         root = read_tree_root_path(tree_root, path, key);
1425         btrfs_free_path(path);
1426
1427         return root;
1428 }
1429
1430 /*
1431  * Initialize subvolume root in-memory structure
1432  *
1433  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1434  */
1435 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1436 {
1437         int ret;
1438         unsigned int nofs_flag;
1439
1440         /*
1441          * We might be called under a transaction (e.g. indirect backref
1442          * resolution) which could deadlock if it triggers memory reclaim
1443          */
1444         nofs_flag = memalloc_nofs_save();
1445         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1446         memalloc_nofs_restore(nofs_flag);
1447         if (ret)
1448                 goto fail;
1449
1450         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1451             !btrfs_is_data_reloc_root(root)) {
1452                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1453                 btrfs_check_and_init_root_item(&root->root_item);
1454         }
1455
1456         /*
1457          * Don't assign anonymous block device to roots that are not exposed to
1458          * userspace, the id pool is limited to 1M
1459          */
1460         if (is_fstree(root->root_key.objectid) &&
1461             btrfs_root_refs(&root->root_item) > 0) {
1462                 if (!anon_dev) {
1463                         ret = get_anon_bdev(&root->anon_dev);
1464                         if (ret)
1465                                 goto fail;
1466                 } else {
1467                         root->anon_dev = anon_dev;
1468                 }
1469         }
1470
1471         mutex_lock(&root->objectid_mutex);
1472         ret = btrfs_init_root_free_objectid(root);
1473         if (ret) {
1474                 mutex_unlock(&root->objectid_mutex);
1475                 goto fail;
1476         }
1477
1478         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1479
1480         mutex_unlock(&root->objectid_mutex);
1481
1482         return 0;
1483 fail:
1484         /* The caller is responsible to call btrfs_free_fs_root */
1485         return ret;
1486 }
1487
1488 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1489                                                u64 root_id)
1490 {
1491         struct btrfs_root *root;
1492
1493         spin_lock(&fs_info->fs_roots_radix_lock);
1494         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1495                                  (unsigned long)root_id);
1496         if (root)
1497                 root = btrfs_grab_root(root);
1498         spin_unlock(&fs_info->fs_roots_radix_lock);
1499         return root;
1500 }
1501
1502 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1503                                                 u64 objectid)
1504 {
1505         struct btrfs_key key = {
1506                 .objectid = objectid,
1507                 .type = BTRFS_ROOT_ITEM_KEY,
1508                 .offset = 0,
1509         };
1510
1511         if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1512                 return btrfs_grab_root(fs_info->tree_root);
1513         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1514                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1515         if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1516                 return btrfs_grab_root(fs_info->chunk_root);
1517         if (objectid == BTRFS_DEV_TREE_OBJECTID)
1518                 return btrfs_grab_root(fs_info->dev_root);
1519         if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1520                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1521         if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1522                 return btrfs_grab_root(fs_info->quota_root) ?
1523                         fs_info->quota_root : ERR_PTR(-ENOENT);
1524         if (objectid == BTRFS_UUID_TREE_OBJECTID)
1525                 return btrfs_grab_root(fs_info->uuid_root) ?
1526                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1527         if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1528                 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1529
1530                 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1531         }
1532         return NULL;
1533 }
1534
1535 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1536                          struct btrfs_root *root)
1537 {
1538         int ret;
1539
1540         ret = radix_tree_preload(GFP_NOFS);
1541         if (ret)
1542                 return ret;
1543
1544         spin_lock(&fs_info->fs_roots_radix_lock);
1545         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1546                                 (unsigned long)root->root_key.objectid,
1547                                 root);
1548         if (ret == 0) {
1549                 btrfs_grab_root(root);
1550                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1551         }
1552         spin_unlock(&fs_info->fs_roots_radix_lock);
1553         radix_tree_preload_end();
1554
1555         return ret;
1556 }
1557
1558 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1559 {
1560 #ifdef CONFIG_BTRFS_DEBUG
1561         struct btrfs_root *root;
1562
1563         while (!list_empty(&fs_info->allocated_roots)) {
1564                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1565
1566                 root = list_first_entry(&fs_info->allocated_roots,
1567                                         struct btrfs_root, leak_list);
1568                 btrfs_err(fs_info, "leaked root %s refcount %d",
1569                           btrfs_root_name(&root->root_key, buf),
1570                           refcount_read(&root->refs));
1571                 while (refcount_read(&root->refs) > 1)
1572                         btrfs_put_root(root);
1573                 btrfs_put_root(root);
1574         }
1575 #endif
1576 }
1577
1578 static void free_global_roots(struct btrfs_fs_info *fs_info)
1579 {
1580         struct btrfs_root *root;
1581         struct rb_node *node;
1582
1583         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1584                 root = rb_entry(node, struct btrfs_root, rb_node);
1585                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1586                 btrfs_put_root(root);
1587         }
1588 }
1589
1590 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1591 {
1592         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1593         percpu_counter_destroy(&fs_info->delalloc_bytes);
1594         percpu_counter_destroy(&fs_info->ordered_bytes);
1595         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1596         btrfs_free_csum_hash(fs_info);
1597         btrfs_free_stripe_hash_table(fs_info);
1598         btrfs_free_ref_cache(fs_info);
1599         kfree(fs_info->balance_ctl);
1600         kfree(fs_info->delayed_root);
1601         free_global_roots(fs_info);
1602         btrfs_put_root(fs_info->tree_root);
1603         btrfs_put_root(fs_info->chunk_root);
1604         btrfs_put_root(fs_info->dev_root);
1605         btrfs_put_root(fs_info->quota_root);
1606         btrfs_put_root(fs_info->uuid_root);
1607         btrfs_put_root(fs_info->fs_root);
1608         btrfs_put_root(fs_info->data_reloc_root);
1609         btrfs_put_root(fs_info->block_group_root);
1610         btrfs_check_leaked_roots(fs_info);
1611         btrfs_extent_buffer_leak_debug_check(fs_info);
1612         kfree(fs_info->super_copy);
1613         kfree(fs_info->super_for_commit);
1614         kfree(fs_info->subpage_info);
1615         kvfree(fs_info);
1616 }
1617
1618
1619 /*
1620  * Get an in-memory reference of a root structure.
1621  *
1622  * For essential trees like root/extent tree, we grab it from fs_info directly.
1623  * For subvolume trees, we check the cached filesystem roots first. If not
1624  * found, then read it from disk and add it to cached fs roots.
1625  *
1626  * Caller should release the root by calling btrfs_put_root() after the usage.
1627  *
1628  * NOTE: Reloc and log trees can't be read by this function as they share the
1629  *       same root objectid.
1630  *
1631  * @objectid:   root id
1632  * @anon_dev:   preallocated anonymous block device number for new roots,
1633  *              pass 0 for new allocation.
1634  * @check_ref:  whether to check root item references, If true, return -ENOENT
1635  *              for orphan roots
1636  */
1637 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1638                                              u64 objectid, dev_t anon_dev,
1639                                              bool check_ref)
1640 {
1641         struct btrfs_root *root;
1642         struct btrfs_path *path;
1643         struct btrfs_key key;
1644         int ret;
1645
1646         root = btrfs_get_global_root(fs_info, objectid);
1647         if (root)
1648                 return root;
1649 again:
1650         root = btrfs_lookup_fs_root(fs_info, objectid);
1651         if (root) {
1652                 /* Shouldn't get preallocated anon_dev for cached roots */
1653                 ASSERT(!anon_dev);
1654                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1655                         btrfs_put_root(root);
1656                         return ERR_PTR(-ENOENT);
1657                 }
1658                 return root;
1659         }
1660
1661         key.objectid = objectid;
1662         key.type = BTRFS_ROOT_ITEM_KEY;
1663         key.offset = (u64)-1;
1664         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1665         if (IS_ERR(root))
1666                 return root;
1667
1668         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1669                 ret = -ENOENT;
1670                 goto fail;
1671         }
1672
1673         ret = btrfs_init_fs_root(root, anon_dev);
1674         if (ret)
1675                 goto fail;
1676
1677         path = btrfs_alloc_path();
1678         if (!path) {
1679                 ret = -ENOMEM;
1680                 goto fail;
1681         }
1682         key.objectid = BTRFS_ORPHAN_OBJECTID;
1683         key.type = BTRFS_ORPHAN_ITEM_KEY;
1684         key.offset = objectid;
1685
1686         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1687         btrfs_free_path(path);
1688         if (ret < 0)
1689                 goto fail;
1690         if (ret == 0)
1691                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1692
1693         ret = btrfs_insert_fs_root(fs_info, root);
1694         if (ret) {
1695                 if (ret == -EEXIST) {
1696                         btrfs_put_root(root);
1697                         goto again;
1698                 }
1699                 goto fail;
1700         }
1701         return root;
1702 fail:
1703         /*
1704          * If our caller provided us an anonymous device, then it's his
1705          * responsibility to free it in case we fail. So we have to set our
1706          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1707          * and once again by our caller.
1708          */
1709         if (anon_dev)
1710                 root->anon_dev = 0;
1711         btrfs_put_root(root);
1712         return ERR_PTR(ret);
1713 }
1714
1715 /*
1716  * Get in-memory reference of a root structure
1717  *
1718  * @objectid:   tree objectid
1719  * @check_ref:  if set, verify that the tree exists and the item has at least
1720  *              one reference
1721  */
1722 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1723                                      u64 objectid, bool check_ref)
1724 {
1725         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1726 }
1727
1728 /*
1729  * Get in-memory reference of a root structure, created as new, optionally pass
1730  * the anonymous block device id
1731  *
1732  * @objectid:   tree objectid
1733  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1734  *              parameter value
1735  */
1736 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1737                                          u64 objectid, dev_t anon_dev)
1738 {
1739         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1740 }
1741
1742 /*
1743  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1744  * @fs_info:    the fs_info
1745  * @objectid:   the objectid we need to lookup
1746  *
1747  * This is exclusively used for backref walking, and exists specifically because
1748  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1749  * creation time, which means we may have to read the tree_root in order to look
1750  * up a fs root that is not in memory.  If the root is not in memory we will
1751  * read the tree root commit root and look up the fs root from there.  This is a
1752  * temporary root, it will not be inserted into the radix tree as it doesn't
1753  * have the most uptodate information, it'll simply be discarded once the
1754  * backref code is finished using the root.
1755  */
1756 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1757                                                  struct btrfs_path *path,
1758                                                  u64 objectid)
1759 {
1760         struct btrfs_root *root;
1761         struct btrfs_key key;
1762
1763         ASSERT(path->search_commit_root && path->skip_locking);
1764
1765         /*
1766          * This can return -ENOENT if we ask for a root that doesn't exist, but
1767          * since this is called via the backref walking code we won't be looking
1768          * up a root that doesn't exist, unless there's corruption.  So if root
1769          * != NULL just return it.
1770          */
1771         root = btrfs_get_global_root(fs_info, objectid);
1772         if (root)
1773                 return root;
1774
1775         root = btrfs_lookup_fs_root(fs_info, objectid);
1776         if (root)
1777                 return root;
1778
1779         key.objectid = objectid;
1780         key.type = BTRFS_ROOT_ITEM_KEY;
1781         key.offset = (u64)-1;
1782         root = read_tree_root_path(fs_info->tree_root, path, &key);
1783         btrfs_release_path(path);
1784
1785         return root;
1786 }
1787
1788 static int cleaner_kthread(void *arg)
1789 {
1790         struct btrfs_fs_info *fs_info = arg;
1791         int again;
1792
1793         while (1) {
1794                 again = 0;
1795
1796                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1797
1798                 /* Make the cleaner go to sleep early. */
1799                 if (btrfs_need_cleaner_sleep(fs_info))
1800                         goto sleep;
1801
1802                 /*
1803                  * Do not do anything if we might cause open_ctree() to block
1804                  * before we have finished mounting the filesystem.
1805                  */
1806                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1807                         goto sleep;
1808
1809                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1810                         goto sleep;
1811
1812                 /*
1813                  * Avoid the problem that we change the status of the fs
1814                  * during the above check and trylock.
1815                  */
1816                 if (btrfs_need_cleaner_sleep(fs_info)) {
1817                         mutex_unlock(&fs_info->cleaner_mutex);
1818                         goto sleep;
1819                 }
1820
1821                 btrfs_run_delayed_iputs(fs_info);
1822
1823                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1824                 mutex_unlock(&fs_info->cleaner_mutex);
1825
1826                 /*
1827                  * The defragger has dealt with the R/O remount and umount,
1828                  * needn't do anything special here.
1829                  */
1830                 btrfs_run_defrag_inodes(fs_info);
1831
1832                 /*
1833                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1834                  * with relocation (btrfs_relocate_chunk) and relocation
1835                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1836                  * after acquiring fs_info->reclaim_bgs_lock. So we
1837                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1838                  * unused block groups.
1839                  */
1840                 btrfs_delete_unused_bgs(fs_info);
1841
1842                 /*
1843                  * Reclaim block groups in the reclaim_bgs list after we deleted
1844                  * all unused block_groups. This possibly gives us some more free
1845                  * space.
1846                  */
1847                 btrfs_reclaim_bgs(fs_info);
1848 sleep:
1849                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1850                 if (kthread_should_park())
1851                         kthread_parkme();
1852                 if (kthread_should_stop())
1853                         return 0;
1854                 if (!again) {
1855                         set_current_state(TASK_INTERRUPTIBLE);
1856                         schedule();
1857                         __set_current_state(TASK_RUNNING);
1858                 }
1859         }
1860 }
1861
1862 static int transaction_kthread(void *arg)
1863 {
1864         struct btrfs_root *root = arg;
1865         struct btrfs_fs_info *fs_info = root->fs_info;
1866         struct btrfs_trans_handle *trans;
1867         struct btrfs_transaction *cur;
1868         u64 transid;
1869         time64_t delta;
1870         unsigned long delay;
1871         bool cannot_commit;
1872
1873         do {
1874                 cannot_commit = false;
1875                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1876                 mutex_lock(&fs_info->transaction_kthread_mutex);
1877
1878                 spin_lock(&fs_info->trans_lock);
1879                 cur = fs_info->running_transaction;
1880                 if (!cur) {
1881                         spin_unlock(&fs_info->trans_lock);
1882                         goto sleep;
1883                 }
1884
1885                 delta = ktime_get_seconds() - cur->start_time;
1886                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1887                     cur->state < TRANS_STATE_COMMIT_START &&
1888                     delta < fs_info->commit_interval) {
1889                         spin_unlock(&fs_info->trans_lock);
1890                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1891                         delay = min(delay,
1892                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1893                         goto sleep;
1894                 }
1895                 transid = cur->transid;
1896                 spin_unlock(&fs_info->trans_lock);
1897
1898                 /* If the file system is aborted, this will always fail. */
1899                 trans = btrfs_attach_transaction(root);
1900                 if (IS_ERR(trans)) {
1901                         if (PTR_ERR(trans) != -ENOENT)
1902                                 cannot_commit = true;
1903                         goto sleep;
1904                 }
1905                 if (transid == trans->transid) {
1906                         btrfs_commit_transaction(trans);
1907                 } else {
1908                         btrfs_end_transaction(trans);
1909                 }
1910 sleep:
1911                 wake_up_process(fs_info->cleaner_kthread);
1912                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1913
1914                 if (BTRFS_FS_ERROR(fs_info))
1915                         btrfs_cleanup_transaction(fs_info);
1916                 if (!kthread_should_stop() &&
1917                                 (!btrfs_transaction_blocked(fs_info) ||
1918                                  cannot_commit))
1919                         schedule_timeout_interruptible(delay);
1920         } while (!kthread_should_stop());
1921         return 0;
1922 }
1923
1924 /*
1925  * This will find the highest generation in the array of root backups.  The
1926  * index of the highest array is returned, or -EINVAL if we can't find
1927  * anything.
1928  *
1929  * We check to make sure the array is valid by comparing the
1930  * generation of the latest  root in the array with the generation
1931  * in the super block.  If they don't match we pitch it.
1932  */
1933 static int find_newest_super_backup(struct btrfs_fs_info *info)
1934 {
1935         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1936         u64 cur;
1937         struct btrfs_root_backup *root_backup;
1938         int i;
1939
1940         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1941                 root_backup = info->super_copy->super_roots + i;
1942                 cur = btrfs_backup_tree_root_gen(root_backup);
1943                 if (cur == newest_gen)
1944                         return i;
1945         }
1946
1947         return -EINVAL;
1948 }
1949
1950 /*
1951  * copy all the root pointers into the super backup array.
1952  * this will bump the backup pointer by one when it is
1953  * done
1954  */
1955 static void backup_super_roots(struct btrfs_fs_info *info)
1956 {
1957         const int next_backup = info->backup_root_index;
1958         struct btrfs_root_backup *root_backup;
1959
1960         root_backup = info->super_for_commit->super_roots + next_backup;
1961
1962         /*
1963          * make sure all of our padding and empty slots get zero filled
1964          * regardless of which ones we use today
1965          */
1966         memset(root_backup, 0, sizeof(*root_backup));
1967
1968         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1969
1970         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1971         btrfs_set_backup_tree_root_gen(root_backup,
1972                                btrfs_header_generation(info->tree_root->node));
1973
1974         btrfs_set_backup_tree_root_level(root_backup,
1975                                btrfs_header_level(info->tree_root->node));
1976
1977         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1978         btrfs_set_backup_chunk_root_gen(root_backup,
1979                                btrfs_header_generation(info->chunk_root->node));
1980         btrfs_set_backup_chunk_root_level(root_backup,
1981                                btrfs_header_level(info->chunk_root->node));
1982
1983         if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
1984                 btrfs_set_backup_block_group_root(root_backup,
1985                                         info->block_group_root->node->start);
1986                 btrfs_set_backup_block_group_root_gen(root_backup,
1987                         btrfs_header_generation(info->block_group_root->node));
1988                 btrfs_set_backup_block_group_root_level(root_backup,
1989                         btrfs_header_level(info->block_group_root->node));
1990         } else {
1991                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1992                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1993
1994                 btrfs_set_backup_extent_root(root_backup,
1995                                              extent_root->node->start);
1996                 btrfs_set_backup_extent_root_gen(root_backup,
1997                                 btrfs_header_generation(extent_root->node));
1998                 btrfs_set_backup_extent_root_level(root_backup,
1999                                         btrfs_header_level(extent_root->node));
2000
2001                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2002                 btrfs_set_backup_csum_root_gen(root_backup,
2003                                                btrfs_header_generation(csum_root->node));
2004                 btrfs_set_backup_csum_root_level(root_backup,
2005                                                  btrfs_header_level(csum_root->node));
2006         }
2007
2008         /*
2009          * we might commit during log recovery, which happens before we set
2010          * the fs_root.  Make sure it is valid before we fill it in.
2011          */
2012         if (info->fs_root && info->fs_root->node) {
2013                 btrfs_set_backup_fs_root(root_backup,
2014                                          info->fs_root->node->start);
2015                 btrfs_set_backup_fs_root_gen(root_backup,
2016                                btrfs_header_generation(info->fs_root->node));
2017                 btrfs_set_backup_fs_root_level(root_backup,
2018                                btrfs_header_level(info->fs_root->node));
2019         }
2020
2021         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2022         btrfs_set_backup_dev_root_gen(root_backup,
2023                                btrfs_header_generation(info->dev_root->node));
2024         btrfs_set_backup_dev_root_level(root_backup,
2025                                        btrfs_header_level(info->dev_root->node));
2026
2027         btrfs_set_backup_total_bytes(root_backup,
2028                              btrfs_super_total_bytes(info->super_copy));
2029         btrfs_set_backup_bytes_used(root_backup,
2030                              btrfs_super_bytes_used(info->super_copy));
2031         btrfs_set_backup_num_devices(root_backup,
2032                              btrfs_super_num_devices(info->super_copy));
2033
2034         /*
2035          * if we don't copy this out to the super_copy, it won't get remembered
2036          * for the next commit
2037          */
2038         memcpy(&info->super_copy->super_roots,
2039                &info->super_for_commit->super_roots,
2040                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2041 }
2042
2043 /*
2044  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2045  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2046  *
2047  * fs_info - filesystem whose backup roots need to be read
2048  * priority - priority of backup root required
2049  *
2050  * Returns backup root index on success and -EINVAL otherwise.
2051  */
2052 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2053 {
2054         int backup_index = find_newest_super_backup(fs_info);
2055         struct btrfs_super_block *super = fs_info->super_copy;
2056         struct btrfs_root_backup *root_backup;
2057
2058         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2059                 if (priority == 0)
2060                         return backup_index;
2061
2062                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2063                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2064         } else {
2065                 return -EINVAL;
2066         }
2067
2068         root_backup = super->super_roots + backup_index;
2069
2070         btrfs_set_super_generation(super,
2071                                    btrfs_backup_tree_root_gen(root_backup));
2072         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2073         btrfs_set_super_root_level(super,
2074                                    btrfs_backup_tree_root_level(root_backup));
2075         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2076
2077         /*
2078          * Fixme: the total bytes and num_devices need to match or we should
2079          * need a fsck
2080          */
2081         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2082         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2083
2084         return backup_index;
2085 }
2086
2087 /* helper to cleanup workers */
2088 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2089 {
2090         btrfs_destroy_workqueue(fs_info->fixup_workers);
2091         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2092         btrfs_destroy_workqueue(fs_info->hipri_workers);
2093         btrfs_destroy_workqueue(fs_info->workers);
2094         if (fs_info->endio_workers)
2095                 destroy_workqueue(fs_info->endio_workers);
2096         if (fs_info->endio_raid56_workers)
2097                 destroy_workqueue(fs_info->endio_raid56_workers);
2098         if (fs_info->rmw_workers)
2099                 destroy_workqueue(fs_info->rmw_workers);
2100         if (fs_info->compressed_write_workers)
2101                 destroy_workqueue(fs_info->compressed_write_workers);
2102         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2103         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2104         btrfs_destroy_workqueue(fs_info->delayed_workers);
2105         btrfs_destroy_workqueue(fs_info->caching_workers);
2106         btrfs_destroy_workqueue(fs_info->flush_workers);
2107         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2108         if (fs_info->discard_ctl.discard_workers)
2109                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2110         /*
2111          * Now that all other work queues are destroyed, we can safely destroy
2112          * the queues used for metadata I/O, since tasks from those other work
2113          * queues can do metadata I/O operations.
2114          */
2115         if (fs_info->endio_meta_workers)
2116                 destroy_workqueue(fs_info->endio_meta_workers);
2117 }
2118
2119 static void free_root_extent_buffers(struct btrfs_root *root)
2120 {
2121         if (root) {
2122                 free_extent_buffer(root->node);
2123                 free_extent_buffer(root->commit_root);
2124                 root->node = NULL;
2125                 root->commit_root = NULL;
2126         }
2127 }
2128
2129 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2130 {
2131         struct btrfs_root *root, *tmp;
2132
2133         rbtree_postorder_for_each_entry_safe(root, tmp,
2134                                              &fs_info->global_root_tree,
2135                                              rb_node)
2136                 free_root_extent_buffers(root);
2137 }
2138
2139 /* helper to cleanup tree roots */
2140 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2141 {
2142         free_root_extent_buffers(info->tree_root);
2143
2144         free_global_root_pointers(info);
2145         free_root_extent_buffers(info->dev_root);
2146         free_root_extent_buffers(info->quota_root);
2147         free_root_extent_buffers(info->uuid_root);
2148         free_root_extent_buffers(info->fs_root);
2149         free_root_extent_buffers(info->data_reloc_root);
2150         free_root_extent_buffers(info->block_group_root);
2151         if (free_chunk_root)
2152                 free_root_extent_buffers(info->chunk_root);
2153 }
2154
2155 void btrfs_put_root(struct btrfs_root *root)
2156 {
2157         if (!root)
2158                 return;
2159
2160         if (refcount_dec_and_test(&root->refs)) {
2161                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2162                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2163                 if (root->anon_dev)
2164                         free_anon_bdev(root->anon_dev);
2165                 btrfs_drew_lock_destroy(&root->snapshot_lock);
2166                 free_root_extent_buffers(root);
2167 #ifdef CONFIG_BTRFS_DEBUG
2168                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2169                 list_del_init(&root->leak_list);
2170                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2171 #endif
2172                 kfree(root);
2173         }
2174 }
2175
2176 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2177 {
2178         int ret;
2179         struct btrfs_root *gang[8];
2180         int i;
2181
2182         while (!list_empty(&fs_info->dead_roots)) {
2183                 gang[0] = list_entry(fs_info->dead_roots.next,
2184                                      struct btrfs_root, root_list);
2185                 list_del(&gang[0]->root_list);
2186
2187                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2188                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2189                 btrfs_put_root(gang[0]);
2190         }
2191
2192         while (1) {
2193                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2194                                              (void **)gang, 0,
2195                                              ARRAY_SIZE(gang));
2196                 if (!ret)
2197                         break;
2198                 for (i = 0; i < ret; i++)
2199                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2200         }
2201 }
2202
2203 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2204 {
2205         mutex_init(&fs_info->scrub_lock);
2206         atomic_set(&fs_info->scrubs_running, 0);
2207         atomic_set(&fs_info->scrub_pause_req, 0);
2208         atomic_set(&fs_info->scrubs_paused, 0);
2209         atomic_set(&fs_info->scrub_cancel_req, 0);
2210         init_waitqueue_head(&fs_info->scrub_pause_wait);
2211         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2212 }
2213
2214 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2215 {
2216         spin_lock_init(&fs_info->balance_lock);
2217         mutex_init(&fs_info->balance_mutex);
2218         atomic_set(&fs_info->balance_pause_req, 0);
2219         atomic_set(&fs_info->balance_cancel_req, 0);
2220         fs_info->balance_ctl = NULL;
2221         init_waitqueue_head(&fs_info->balance_wait_q);
2222         atomic_set(&fs_info->reloc_cancel_req, 0);
2223 }
2224
2225 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2226 {
2227         struct inode *inode = fs_info->btree_inode;
2228
2229         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2230         set_nlink(inode, 1);
2231         /*
2232          * we set the i_size on the btree inode to the max possible int.
2233          * the real end of the address space is determined by all of
2234          * the devices in the system
2235          */
2236         inode->i_size = OFFSET_MAX;
2237         inode->i_mapping->a_ops = &btree_aops;
2238
2239         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2240         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2241                             IO_TREE_BTREE_INODE_IO, inode);
2242         BTRFS_I(inode)->io_tree.track_uptodate = false;
2243         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2244
2245         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2246         BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2247         BTRFS_I(inode)->location.type = 0;
2248         BTRFS_I(inode)->location.offset = 0;
2249         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2250         btrfs_insert_inode_hash(inode);
2251 }
2252
2253 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2254 {
2255         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2256         init_rwsem(&fs_info->dev_replace.rwsem);
2257         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2258 }
2259
2260 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2261 {
2262         spin_lock_init(&fs_info->qgroup_lock);
2263         mutex_init(&fs_info->qgroup_ioctl_lock);
2264         fs_info->qgroup_tree = RB_ROOT;
2265         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2266         fs_info->qgroup_seq = 1;
2267         fs_info->qgroup_ulist = NULL;
2268         fs_info->qgroup_rescan_running = false;
2269         mutex_init(&fs_info->qgroup_rescan_lock);
2270 }
2271
2272 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2273 {
2274         u32 max_active = fs_info->thread_pool_size;
2275         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2276
2277         fs_info->workers =
2278                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2279         fs_info->hipri_workers =
2280                 btrfs_alloc_workqueue(fs_info, "worker-high",
2281                                       flags | WQ_HIGHPRI, max_active, 16);
2282
2283         fs_info->delalloc_workers =
2284                 btrfs_alloc_workqueue(fs_info, "delalloc",
2285                                       flags, max_active, 2);
2286
2287         fs_info->flush_workers =
2288                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2289                                       flags, max_active, 0);
2290
2291         fs_info->caching_workers =
2292                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2293
2294         fs_info->fixup_workers =
2295                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2296
2297         fs_info->endio_workers =
2298                 alloc_workqueue("btrfs-endio", flags, max_active);
2299         fs_info->endio_meta_workers =
2300                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2301         fs_info->endio_raid56_workers =
2302                 alloc_workqueue("btrfs-endio-raid56", flags, max_active);
2303         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2304         fs_info->endio_write_workers =
2305                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2306                                       max_active, 2);
2307         fs_info->compressed_write_workers =
2308                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2309         fs_info->endio_freespace_worker =
2310                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2311                                       max_active, 0);
2312         fs_info->delayed_workers =
2313                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2314                                       max_active, 0);
2315         fs_info->qgroup_rescan_workers =
2316                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2317         fs_info->discard_ctl.discard_workers =
2318                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2319
2320         if (!(fs_info->workers && fs_info->hipri_workers &&
2321               fs_info->delalloc_workers && fs_info->flush_workers &&
2322               fs_info->endio_workers && fs_info->endio_meta_workers &&
2323               fs_info->compressed_write_workers &&
2324               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2325               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2326               fs_info->caching_workers && fs_info->fixup_workers &&
2327               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2328               fs_info->discard_ctl.discard_workers)) {
2329                 return -ENOMEM;
2330         }
2331
2332         return 0;
2333 }
2334
2335 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2336 {
2337         struct crypto_shash *csum_shash;
2338         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2339
2340         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2341
2342         if (IS_ERR(csum_shash)) {
2343                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2344                           csum_driver);
2345                 return PTR_ERR(csum_shash);
2346         }
2347
2348         fs_info->csum_shash = csum_shash;
2349
2350         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2351                         btrfs_super_csum_name(csum_type),
2352                         crypto_shash_driver_name(csum_shash));
2353         return 0;
2354 }
2355
2356 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2357                             struct btrfs_fs_devices *fs_devices)
2358 {
2359         int ret;
2360         struct btrfs_root *log_tree_root;
2361         struct btrfs_super_block *disk_super = fs_info->super_copy;
2362         u64 bytenr = btrfs_super_log_root(disk_super);
2363         int level = btrfs_super_log_root_level(disk_super);
2364
2365         if (fs_devices->rw_devices == 0) {
2366                 btrfs_warn(fs_info, "log replay required on RO media");
2367                 return -EIO;
2368         }
2369
2370         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2371                                          GFP_KERNEL);
2372         if (!log_tree_root)
2373                 return -ENOMEM;
2374
2375         log_tree_root->node = read_tree_block(fs_info, bytenr,
2376                                               BTRFS_TREE_LOG_OBJECTID,
2377                                               fs_info->generation + 1, level,
2378                                               NULL);
2379         if (IS_ERR(log_tree_root->node)) {
2380                 btrfs_warn(fs_info, "failed to read log tree");
2381                 ret = PTR_ERR(log_tree_root->node);
2382                 log_tree_root->node = NULL;
2383                 btrfs_put_root(log_tree_root);
2384                 return ret;
2385         }
2386         if (!extent_buffer_uptodate(log_tree_root->node)) {
2387                 btrfs_err(fs_info, "failed to read log tree");
2388                 btrfs_put_root(log_tree_root);
2389                 return -EIO;
2390         }
2391
2392         /* returns with log_tree_root freed on success */
2393         ret = btrfs_recover_log_trees(log_tree_root);
2394         if (ret) {
2395                 btrfs_handle_fs_error(fs_info, ret,
2396                                       "Failed to recover log tree");
2397                 btrfs_put_root(log_tree_root);
2398                 return ret;
2399         }
2400
2401         if (sb_rdonly(fs_info->sb)) {
2402                 ret = btrfs_commit_super(fs_info);
2403                 if (ret)
2404                         return ret;
2405         }
2406
2407         return 0;
2408 }
2409
2410 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2411                                       struct btrfs_path *path, u64 objectid,
2412                                       const char *name)
2413 {
2414         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2415         struct btrfs_root *root;
2416         u64 max_global_id = 0;
2417         int ret;
2418         struct btrfs_key key = {
2419                 .objectid = objectid,
2420                 .type = BTRFS_ROOT_ITEM_KEY,
2421                 .offset = 0,
2422         };
2423         bool found = false;
2424
2425         /* If we have IGNOREDATACSUMS skip loading these roots. */
2426         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2427             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2428                 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2429                 return 0;
2430         }
2431
2432         while (1) {
2433                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2434                 if (ret < 0)
2435                         break;
2436
2437                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2438                         ret = btrfs_next_leaf(tree_root, path);
2439                         if (ret) {
2440                                 if (ret > 0)
2441                                         ret = 0;
2442                                 break;
2443                         }
2444                 }
2445                 ret = 0;
2446
2447                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2448                 if (key.objectid != objectid)
2449                         break;
2450                 btrfs_release_path(path);
2451
2452                 /*
2453                  * Just worry about this for extent tree, it'll be the same for
2454                  * everybody.
2455                  */
2456                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2457                         max_global_id = max(max_global_id, key.offset);
2458
2459                 found = true;
2460                 root = read_tree_root_path(tree_root, path, &key);
2461                 if (IS_ERR(root)) {
2462                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2463                                 ret = PTR_ERR(root);
2464                         break;
2465                 }
2466                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2467                 ret = btrfs_global_root_insert(root);
2468                 if (ret) {
2469                         btrfs_put_root(root);
2470                         break;
2471                 }
2472                 key.offset++;
2473         }
2474         btrfs_release_path(path);
2475
2476         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2477                 fs_info->nr_global_roots = max_global_id + 1;
2478
2479         if (!found || ret) {
2480                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2481                         set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2482
2483                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2484                         ret = ret ? ret : -ENOENT;
2485                 else
2486                         ret = 0;
2487                 btrfs_err(fs_info, "failed to load root %s", name);
2488         }
2489         return ret;
2490 }
2491
2492 static int load_global_roots(struct btrfs_root *tree_root)
2493 {
2494         struct btrfs_path *path;
2495         int ret = 0;
2496
2497         path = btrfs_alloc_path();
2498         if (!path)
2499                 return -ENOMEM;
2500
2501         ret = load_global_roots_objectid(tree_root, path,
2502                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2503         if (ret)
2504                 goto out;
2505         ret = load_global_roots_objectid(tree_root, path,
2506                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2507         if (ret)
2508                 goto out;
2509         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2510                 goto out;
2511         ret = load_global_roots_objectid(tree_root, path,
2512                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2513                                          "free space");
2514 out:
2515         btrfs_free_path(path);
2516         return ret;
2517 }
2518
2519 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2520 {
2521         struct btrfs_root *tree_root = fs_info->tree_root;
2522         struct btrfs_root *root;
2523         struct btrfs_key location;
2524         int ret;
2525
2526         BUG_ON(!fs_info->tree_root);
2527
2528         ret = load_global_roots(tree_root);
2529         if (ret)
2530                 return ret;
2531
2532         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2533         location.type = BTRFS_ROOT_ITEM_KEY;
2534         location.offset = 0;
2535
2536         root = btrfs_read_tree_root(tree_root, &location);
2537         if (IS_ERR(root)) {
2538                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2539                         ret = PTR_ERR(root);
2540                         goto out;
2541                 }
2542         } else {
2543                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2544                 fs_info->dev_root = root;
2545         }
2546         /* Initialize fs_info for all devices in any case */
2547         btrfs_init_devices_late(fs_info);
2548
2549         /*
2550          * This tree can share blocks with some other fs tree during relocation
2551          * and we need a proper setup by btrfs_get_fs_root
2552          */
2553         root = btrfs_get_fs_root(tree_root->fs_info,
2554                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2555         if (IS_ERR(root)) {
2556                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2557                         ret = PTR_ERR(root);
2558                         goto out;
2559                 }
2560         } else {
2561                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2562                 fs_info->data_reloc_root = root;
2563         }
2564
2565         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2566         root = btrfs_read_tree_root(tree_root, &location);
2567         if (!IS_ERR(root)) {
2568                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2569                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2570                 fs_info->quota_root = root;
2571         }
2572
2573         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2574         root = btrfs_read_tree_root(tree_root, &location);
2575         if (IS_ERR(root)) {
2576                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2577                         ret = PTR_ERR(root);
2578                         if (ret != -ENOENT)
2579                                 goto out;
2580                 }
2581         } else {
2582                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2583                 fs_info->uuid_root = root;
2584         }
2585
2586         return 0;
2587 out:
2588         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2589                    location.objectid, ret);
2590         return ret;
2591 }
2592
2593 /*
2594  * Real super block validation
2595  * NOTE: super csum type and incompat features will not be checked here.
2596  *
2597  * @sb:         super block to check
2598  * @mirror_num: the super block number to check its bytenr:
2599  *              0       the primary (1st) sb
2600  *              1, 2    2nd and 3rd backup copy
2601  *             -1       skip bytenr check
2602  */
2603 static int validate_super(struct btrfs_fs_info *fs_info,
2604                             struct btrfs_super_block *sb, int mirror_num)
2605 {
2606         u64 nodesize = btrfs_super_nodesize(sb);
2607         u64 sectorsize = btrfs_super_sectorsize(sb);
2608         int ret = 0;
2609
2610         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2611                 btrfs_err(fs_info, "no valid FS found");
2612                 ret = -EINVAL;
2613         }
2614         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2615                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2616                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2617                 ret = -EINVAL;
2618         }
2619         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2620                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2621                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2622                 ret = -EINVAL;
2623         }
2624         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2625                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2626                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2627                 ret = -EINVAL;
2628         }
2629         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2630                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2631                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2632                 ret = -EINVAL;
2633         }
2634
2635         /*
2636          * Check sectorsize and nodesize first, other check will need it.
2637          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2638          */
2639         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2640             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2641                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2642                 ret = -EINVAL;
2643         }
2644
2645         /*
2646          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2647          *
2648          * We can support 16K sectorsize with 64K page size without problem,
2649          * but such sectorsize/pagesize combination doesn't make much sense.
2650          * 4K will be our future standard, PAGE_SIZE is supported from the very
2651          * beginning.
2652          */
2653         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2654                 btrfs_err(fs_info,
2655                         "sectorsize %llu not yet supported for page size %lu",
2656                         sectorsize, PAGE_SIZE);
2657                 ret = -EINVAL;
2658         }
2659
2660         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2661             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2662                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2663                 ret = -EINVAL;
2664         }
2665         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2666                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2667                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2668                 ret = -EINVAL;
2669         }
2670
2671         /* Root alignment check */
2672         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2673                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2674                            btrfs_super_root(sb));
2675                 ret = -EINVAL;
2676         }
2677         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2678                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2679                            btrfs_super_chunk_root(sb));
2680                 ret = -EINVAL;
2681         }
2682         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2683                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2684                            btrfs_super_log_root(sb));
2685                 ret = -EINVAL;
2686         }
2687
2688         if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2689                    BTRFS_FSID_SIZE)) {
2690                 btrfs_err(fs_info,
2691                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2692                         fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2693                 ret = -EINVAL;
2694         }
2695
2696         if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2697             memcmp(fs_info->fs_devices->metadata_uuid,
2698                    fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2699                 btrfs_err(fs_info,
2700 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2701                         fs_info->super_copy->metadata_uuid,
2702                         fs_info->fs_devices->metadata_uuid);
2703                 ret = -EINVAL;
2704         }
2705
2706         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2707                    BTRFS_FSID_SIZE) != 0) {
2708                 btrfs_err(fs_info,
2709                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2710                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2711                 ret = -EINVAL;
2712         }
2713
2714         /*
2715          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2716          * done later
2717          */
2718         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2719                 btrfs_err(fs_info, "bytes_used is too small %llu",
2720                           btrfs_super_bytes_used(sb));
2721                 ret = -EINVAL;
2722         }
2723         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2724                 btrfs_err(fs_info, "invalid stripesize %u",
2725                           btrfs_super_stripesize(sb));
2726                 ret = -EINVAL;
2727         }
2728         if (btrfs_super_num_devices(sb) > (1UL << 31))
2729                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2730                            btrfs_super_num_devices(sb));
2731         if (btrfs_super_num_devices(sb) == 0) {
2732                 btrfs_err(fs_info, "number of devices is 0");
2733                 ret = -EINVAL;
2734         }
2735
2736         if (mirror_num >= 0 &&
2737             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2738                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2739                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2740                 ret = -EINVAL;
2741         }
2742
2743         /*
2744          * Obvious sys_chunk_array corruptions, it must hold at least one key
2745          * and one chunk
2746          */
2747         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2748                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2749                           btrfs_super_sys_array_size(sb),
2750                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2751                 ret = -EINVAL;
2752         }
2753         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2754                         + sizeof(struct btrfs_chunk)) {
2755                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2756                           btrfs_super_sys_array_size(sb),
2757                           sizeof(struct btrfs_disk_key)
2758                           + sizeof(struct btrfs_chunk));
2759                 ret = -EINVAL;
2760         }
2761
2762         /*
2763          * The generation is a global counter, we'll trust it more than the others
2764          * but it's still possible that it's the one that's wrong.
2765          */
2766         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2767                 btrfs_warn(fs_info,
2768                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2769                         btrfs_super_generation(sb),
2770                         btrfs_super_chunk_root_generation(sb));
2771         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2772             && btrfs_super_cache_generation(sb) != (u64)-1)
2773                 btrfs_warn(fs_info,
2774                         "suspicious: generation < cache_generation: %llu < %llu",
2775                         btrfs_super_generation(sb),
2776                         btrfs_super_cache_generation(sb));
2777
2778         return ret;
2779 }
2780
2781 /*
2782  * Validation of super block at mount time.
2783  * Some checks already done early at mount time, like csum type and incompat
2784  * flags will be skipped.
2785  */
2786 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2787 {
2788         return validate_super(fs_info, fs_info->super_copy, 0);
2789 }
2790
2791 /*
2792  * Validation of super block at write time.
2793  * Some checks like bytenr check will be skipped as their values will be
2794  * overwritten soon.
2795  * Extra checks like csum type and incompat flags will be done here.
2796  */
2797 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2798                                       struct btrfs_super_block *sb)
2799 {
2800         int ret;
2801
2802         ret = validate_super(fs_info, sb, -1);
2803         if (ret < 0)
2804                 goto out;
2805         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2806                 ret = -EUCLEAN;
2807                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2808                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2809                 goto out;
2810         }
2811         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2812                 ret = -EUCLEAN;
2813                 btrfs_err(fs_info,
2814                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2815                           btrfs_super_incompat_flags(sb),
2816                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2817                 goto out;
2818         }
2819 out:
2820         if (ret < 0)
2821                 btrfs_err(fs_info,
2822                 "super block corruption detected before writing it to disk");
2823         return ret;
2824 }
2825
2826 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2827 {
2828         int ret = 0;
2829
2830         root->node = read_tree_block(root->fs_info, bytenr,
2831                                      root->root_key.objectid, gen, level, NULL);
2832         if (IS_ERR(root->node)) {
2833                 ret = PTR_ERR(root->node);
2834                 root->node = NULL;
2835                 return ret;
2836         }
2837         if (!extent_buffer_uptodate(root->node)) {
2838                 free_extent_buffer(root->node);
2839                 root->node = NULL;
2840                 return -EIO;
2841         }
2842
2843         btrfs_set_root_node(&root->root_item, root->node);
2844         root->commit_root = btrfs_root_node(root);
2845         btrfs_set_root_refs(&root->root_item, 1);
2846         return ret;
2847 }
2848
2849 static int load_important_roots(struct btrfs_fs_info *fs_info)
2850 {
2851         struct btrfs_super_block *sb = fs_info->super_copy;
2852         u64 gen, bytenr;
2853         int level, ret;
2854
2855         bytenr = btrfs_super_root(sb);
2856         gen = btrfs_super_generation(sb);
2857         level = btrfs_super_root_level(sb);
2858         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2859         if (ret) {
2860                 btrfs_warn(fs_info, "couldn't read tree root");
2861                 return ret;
2862         }
2863
2864         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2865                 return 0;
2866
2867         bytenr = btrfs_super_block_group_root(sb);
2868         gen = btrfs_super_block_group_root_generation(sb);
2869         level = btrfs_super_block_group_root_level(sb);
2870         ret = load_super_root(fs_info->block_group_root, bytenr, gen, level);
2871         if (ret)
2872                 btrfs_warn(fs_info, "couldn't read block group root");
2873         return ret;
2874 }
2875
2876 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2877 {
2878         int backup_index = find_newest_super_backup(fs_info);
2879         struct btrfs_super_block *sb = fs_info->super_copy;
2880         struct btrfs_root *tree_root = fs_info->tree_root;
2881         bool handle_error = false;
2882         int ret = 0;
2883         int i;
2884
2885         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2886                 struct btrfs_root *root;
2887
2888                 root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID,
2889                                         GFP_KERNEL);
2890                 if (!root)
2891                         return -ENOMEM;
2892                 fs_info->block_group_root = root;
2893         }
2894
2895         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2896                 if (handle_error) {
2897                         if (!IS_ERR(tree_root->node))
2898                                 free_extent_buffer(tree_root->node);
2899                         tree_root->node = NULL;
2900
2901                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2902                                 break;
2903
2904                         free_root_pointers(fs_info, 0);
2905
2906                         /*
2907                          * Don't use the log in recovery mode, it won't be
2908                          * valid
2909                          */
2910                         btrfs_set_super_log_root(sb, 0);
2911
2912                         /* We can't trust the free space cache either */
2913                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2914
2915                         ret = read_backup_root(fs_info, i);
2916                         backup_index = ret;
2917                         if (ret < 0)
2918                                 return ret;
2919                 }
2920
2921                 ret = load_important_roots(fs_info);
2922                 if (ret) {
2923                         handle_error = true;
2924                         continue;
2925                 }
2926
2927                 /*
2928                  * No need to hold btrfs_root::objectid_mutex since the fs
2929                  * hasn't been fully initialised and we are the only user
2930                  */
2931                 ret = btrfs_init_root_free_objectid(tree_root);
2932                 if (ret < 0) {
2933                         handle_error = true;
2934                         continue;
2935                 }
2936
2937                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2938
2939                 ret = btrfs_read_roots(fs_info);
2940                 if (ret < 0) {
2941                         handle_error = true;
2942                         continue;
2943                 }
2944
2945                 /* All successful */
2946                 fs_info->generation = btrfs_header_generation(tree_root->node);
2947                 fs_info->last_trans_committed = fs_info->generation;
2948                 fs_info->last_reloc_trans = 0;
2949
2950                 /* Always begin writing backup roots after the one being used */
2951                 if (backup_index < 0) {
2952                         fs_info->backup_root_index = 0;
2953                 } else {
2954                         fs_info->backup_root_index = backup_index + 1;
2955                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2956                 }
2957                 break;
2958         }
2959
2960         return ret;
2961 }
2962
2963 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2964 {
2965         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2966         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2967         INIT_LIST_HEAD(&fs_info->trans_list);
2968         INIT_LIST_HEAD(&fs_info->dead_roots);
2969         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2970         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2971         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2972         spin_lock_init(&fs_info->delalloc_root_lock);
2973         spin_lock_init(&fs_info->trans_lock);
2974         spin_lock_init(&fs_info->fs_roots_radix_lock);
2975         spin_lock_init(&fs_info->delayed_iput_lock);
2976         spin_lock_init(&fs_info->defrag_inodes_lock);
2977         spin_lock_init(&fs_info->super_lock);
2978         spin_lock_init(&fs_info->buffer_lock);
2979         spin_lock_init(&fs_info->unused_bgs_lock);
2980         spin_lock_init(&fs_info->treelog_bg_lock);
2981         spin_lock_init(&fs_info->zone_active_bgs_lock);
2982         spin_lock_init(&fs_info->relocation_bg_lock);
2983         rwlock_init(&fs_info->tree_mod_log_lock);
2984         rwlock_init(&fs_info->global_root_lock);
2985         mutex_init(&fs_info->unused_bg_unpin_mutex);
2986         mutex_init(&fs_info->reclaim_bgs_lock);
2987         mutex_init(&fs_info->reloc_mutex);
2988         mutex_init(&fs_info->delalloc_root_mutex);
2989         mutex_init(&fs_info->zoned_meta_io_lock);
2990         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2991         seqlock_init(&fs_info->profiles_lock);
2992
2993         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2994         INIT_LIST_HEAD(&fs_info->space_info);
2995         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2996         INIT_LIST_HEAD(&fs_info->unused_bgs);
2997         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2998         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2999 #ifdef CONFIG_BTRFS_DEBUG
3000         INIT_LIST_HEAD(&fs_info->allocated_roots);
3001         INIT_LIST_HEAD(&fs_info->allocated_ebs);
3002         spin_lock_init(&fs_info->eb_leak_lock);
3003 #endif
3004         extent_map_tree_init(&fs_info->mapping_tree);
3005         btrfs_init_block_rsv(&fs_info->global_block_rsv,
3006                              BTRFS_BLOCK_RSV_GLOBAL);
3007         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3008         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3009         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3010         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3011                              BTRFS_BLOCK_RSV_DELOPS);
3012         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3013                              BTRFS_BLOCK_RSV_DELREFS);
3014
3015         atomic_set(&fs_info->async_delalloc_pages, 0);
3016         atomic_set(&fs_info->defrag_running, 0);
3017         atomic_set(&fs_info->nr_delayed_iputs, 0);
3018         atomic64_set(&fs_info->tree_mod_seq, 0);
3019         fs_info->global_root_tree = RB_ROOT;
3020         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3021         fs_info->metadata_ratio = 0;
3022         fs_info->defrag_inodes = RB_ROOT;
3023         atomic64_set(&fs_info->free_chunk_space, 0);
3024         fs_info->tree_mod_log = RB_ROOT;
3025         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3026         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3027         btrfs_init_ref_verify(fs_info);
3028
3029         fs_info->thread_pool_size = min_t(unsigned long,
3030                                           num_online_cpus() + 2, 8);
3031
3032         INIT_LIST_HEAD(&fs_info->ordered_roots);
3033         spin_lock_init(&fs_info->ordered_root_lock);
3034
3035         btrfs_init_scrub(fs_info);
3036 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3037         fs_info->check_integrity_print_mask = 0;
3038 #endif
3039         btrfs_init_balance(fs_info);
3040         btrfs_init_async_reclaim_work(fs_info);
3041
3042         rwlock_init(&fs_info->block_group_cache_lock);
3043         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3044
3045         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3046                             IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3047
3048         mutex_init(&fs_info->ordered_operations_mutex);
3049         mutex_init(&fs_info->tree_log_mutex);
3050         mutex_init(&fs_info->chunk_mutex);
3051         mutex_init(&fs_info->transaction_kthread_mutex);
3052         mutex_init(&fs_info->cleaner_mutex);
3053         mutex_init(&fs_info->ro_block_group_mutex);
3054         init_rwsem(&fs_info->commit_root_sem);
3055         init_rwsem(&fs_info->cleanup_work_sem);
3056         init_rwsem(&fs_info->subvol_sem);
3057         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3058
3059         btrfs_init_dev_replace_locks(fs_info);
3060         btrfs_init_qgroup(fs_info);
3061         btrfs_discard_init(fs_info);
3062
3063         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3064         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3065
3066         init_waitqueue_head(&fs_info->transaction_throttle);
3067         init_waitqueue_head(&fs_info->transaction_wait);
3068         init_waitqueue_head(&fs_info->transaction_blocked_wait);
3069         init_waitqueue_head(&fs_info->async_submit_wait);
3070         init_waitqueue_head(&fs_info->delayed_iputs_wait);
3071
3072         /* Usable values until the real ones are cached from the superblock */
3073         fs_info->nodesize = 4096;
3074         fs_info->sectorsize = 4096;
3075         fs_info->sectorsize_bits = ilog2(4096);
3076         fs_info->stripesize = 4096;
3077
3078         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3079
3080         spin_lock_init(&fs_info->swapfile_pins_lock);
3081         fs_info->swapfile_pins = RB_ROOT;
3082
3083         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3084         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3085 }
3086
3087 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3088 {
3089         int ret;
3090
3091         fs_info->sb = sb;
3092         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3093         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3094
3095         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3096         if (ret)
3097                 return ret;
3098
3099         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3100         if (ret)
3101                 return ret;
3102
3103         fs_info->dirty_metadata_batch = PAGE_SIZE *
3104                                         (1 + ilog2(nr_cpu_ids));
3105
3106         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3107         if (ret)
3108                 return ret;
3109
3110         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3111                         GFP_KERNEL);
3112         if (ret)
3113                 return ret;
3114
3115         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3116                                         GFP_KERNEL);
3117         if (!fs_info->delayed_root)
3118                 return -ENOMEM;
3119         btrfs_init_delayed_root(fs_info->delayed_root);
3120
3121         if (sb_rdonly(sb))
3122                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3123
3124         return btrfs_alloc_stripe_hash_table(fs_info);
3125 }
3126
3127 static int btrfs_uuid_rescan_kthread(void *data)
3128 {
3129         struct btrfs_fs_info *fs_info = data;
3130         int ret;
3131
3132         /*
3133          * 1st step is to iterate through the existing UUID tree and
3134          * to delete all entries that contain outdated data.
3135          * 2nd step is to add all missing entries to the UUID tree.
3136          */
3137         ret = btrfs_uuid_tree_iterate(fs_info);
3138         if (ret < 0) {
3139                 if (ret != -EINTR)
3140                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3141                                    ret);
3142                 up(&fs_info->uuid_tree_rescan_sem);
3143                 return ret;
3144         }
3145         return btrfs_uuid_scan_kthread(data);
3146 }
3147
3148 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3149 {
3150         struct task_struct *task;
3151
3152         down(&fs_info->uuid_tree_rescan_sem);
3153         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3154         if (IS_ERR(task)) {
3155                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3156                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3157                 up(&fs_info->uuid_tree_rescan_sem);
3158                 return PTR_ERR(task);
3159         }
3160
3161         return 0;
3162 }
3163
3164 /*
3165  * Some options only have meaning at mount time and shouldn't persist across
3166  * remounts, or be displayed. Clear these at the end of mount and remount
3167  * code paths.
3168  */
3169 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3170 {
3171         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3172         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3173 }
3174
3175 /*
3176  * Mounting logic specific to read-write file systems. Shared by open_ctree
3177  * and btrfs_remount when remounting from read-only to read-write.
3178  */
3179 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3180 {
3181         int ret;
3182         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3183         bool clear_free_space_tree = false;
3184
3185         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3186             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3187                 clear_free_space_tree = true;
3188         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3189                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3190                 btrfs_warn(fs_info, "free space tree is invalid");
3191                 clear_free_space_tree = true;
3192         }
3193
3194         if (clear_free_space_tree) {
3195                 btrfs_info(fs_info, "clearing free space tree");
3196                 ret = btrfs_clear_free_space_tree(fs_info);
3197                 if (ret) {
3198                         btrfs_warn(fs_info,
3199                                    "failed to clear free space tree: %d", ret);
3200                         goto out;
3201                 }
3202         }
3203
3204         /*
3205          * btrfs_find_orphan_roots() is responsible for finding all the dead
3206          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3207          * them into the fs_info->fs_roots_radix tree. This must be done before
3208          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3209          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3210          * item before the root's tree is deleted - this means that if we unmount
3211          * or crash before the deletion completes, on the next mount we will not
3212          * delete what remains of the tree because the orphan item does not
3213          * exists anymore, which is what tells us we have a pending deletion.
3214          */
3215         ret = btrfs_find_orphan_roots(fs_info);
3216         if (ret)
3217                 goto out;
3218
3219         ret = btrfs_cleanup_fs_roots(fs_info);
3220         if (ret)
3221                 goto out;
3222
3223         down_read(&fs_info->cleanup_work_sem);
3224         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3225             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3226                 up_read(&fs_info->cleanup_work_sem);
3227                 goto out;
3228         }
3229         up_read(&fs_info->cleanup_work_sem);
3230
3231         mutex_lock(&fs_info->cleaner_mutex);
3232         ret = btrfs_recover_relocation(fs_info);
3233         mutex_unlock(&fs_info->cleaner_mutex);
3234         if (ret < 0) {
3235                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3236                 goto out;
3237         }
3238
3239         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3240             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3241                 btrfs_info(fs_info, "creating free space tree");
3242                 ret = btrfs_create_free_space_tree(fs_info);
3243                 if (ret) {
3244                         btrfs_warn(fs_info,
3245                                 "failed to create free space tree: %d", ret);
3246                         goto out;
3247                 }
3248         }
3249
3250         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3251                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3252                 if (ret)
3253                         goto out;
3254         }
3255
3256         ret = btrfs_resume_balance_async(fs_info);
3257         if (ret)
3258                 goto out;
3259
3260         ret = btrfs_resume_dev_replace_async(fs_info);
3261         if (ret) {
3262                 btrfs_warn(fs_info, "failed to resume dev_replace");
3263                 goto out;
3264         }
3265
3266         btrfs_qgroup_rescan_resume(fs_info);
3267
3268         if (!fs_info->uuid_root) {
3269                 btrfs_info(fs_info, "creating UUID tree");
3270                 ret = btrfs_create_uuid_tree(fs_info);
3271                 if (ret) {
3272                         btrfs_warn(fs_info,
3273                                    "failed to create the UUID tree %d", ret);
3274                         goto out;
3275                 }
3276         }
3277
3278 out:
3279         return ret;
3280 }
3281
3282 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3283                       char *options)
3284 {
3285         u32 sectorsize;
3286         u32 nodesize;
3287         u32 stripesize;
3288         u64 generation;
3289         u64 features;
3290         u16 csum_type;
3291         struct btrfs_super_block *disk_super;
3292         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3293         struct btrfs_root *tree_root;
3294         struct btrfs_root *chunk_root;
3295         int ret;
3296         int err = -EINVAL;
3297         int level;
3298
3299         ret = init_mount_fs_info(fs_info, sb);
3300         if (ret) {
3301                 err = ret;
3302                 goto fail;
3303         }
3304
3305         /* These need to be init'ed before we start creating inodes and such. */
3306         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3307                                      GFP_KERNEL);
3308         fs_info->tree_root = tree_root;
3309         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3310                                       GFP_KERNEL);
3311         fs_info->chunk_root = chunk_root;
3312         if (!tree_root || !chunk_root) {
3313                 err = -ENOMEM;
3314                 goto fail;
3315         }
3316
3317         fs_info->btree_inode = new_inode(sb);
3318         if (!fs_info->btree_inode) {
3319                 err = -ENOMEM;
3320                 goto fail;
3321         }
3322         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3323         btrfs_init_btree_inode(fs_info);
3324
3325         invalidate_bdev(fs_devices->latest_dev->bdev);
3326
3327         /*
3328          * Read super block and check the signature bytes only
3329          */
3330         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3331         if (IS_ERR(disk_super)) {
3332                 err = PTR_ERR(disk_super);
3333                 goto fail_alloc;
3334         }
3335
3336         /*
3337          * Verify the type first, if that or the checksum value are
3338          * corrupted, we'll find out
3339          */
3340         csum_type = btrfs_super_csum_type(disk_super);
3341         if (!btrfs_supported_super_csum(csum_type)) {
3342                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3343                           csum_type);
3344                 err = -EINVAL;
3345                 btrfs_release_disk_super(disk_super);
3346                 goto fail_alloc;
3347         }
3348
3349         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3350
3351         ret = btrfs_init_csum_hash(fs_info, csum_type);
3352         if (ret) {
3353                 err = ret;
3354                 btrfs_release_disk_super(disk_super);
3355                 goto fail_alloc;
3356         }
3357
3358         /*
3359          * We want to check superblock checksum, the type is stored inside.
3360          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3361          */
3362         if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3363                 btrfs_err(fs_info, "superblock checksum mismatch");
3364                 err = -EINVAL;
3365                 btrfs_release_disk_super(disk_super);
3366                 goto fail_alloc;
3367         }
3368
3369         /*
3370          * super_copy is zeroed at allocation time and we never touch the
3371          * following bytes up to INFO_SIZE, the checksum is calculated from
3372          * the whole block of INFO_SIZE
3373          */
3374         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3375         btrfs_release_disk_super(disk_super);
3376
3377         disk_super = fs_info->super_copy;
3378
3379
3380         features = btrfs_super_flags(disk_super);
3381         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3382                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3383                 btrfs_set_super_flags(disk_super, features);
3384                 btrfs_info(fs_info,
3385                         "found metadata UUID change in progress flag, clearing");
3386         }
3387
3388         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3389                sizeof(*fs_info->super_for_commit));
3390
3391         ret = btrfs_validate_mount_super(fs_info);
3392         if (ret) {
3393                 btrfs_err(fs_info, "superblock contains fatal errors");
3394                 err = -EINVAL;
3395                 goto fail_alloc;
3396         }
3397
3398         if (!btrfs_super_root(disk_super))
3399                 goto fail_alloc;
3400
3401         /* check FS state, whether FS is broken. */
3402         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3403                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3404
3405         /*
3406          * In the long term, we'll store the compression type in the super
3407          * block, and it'll be used for per file compression control.
3408          */
3409         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3410
3411
3412         /* Set up fs_info before parsing mount options */
3413         nodesize = btrfs_super_nodesize(disk_super);
3414         sectorsize = btrfs_super_sectorsize(disk_super);
3415         stripesize = sectorsize;
3416         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3417         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3418
3419         fs_info->nodesize = nodesize;
3420         fs_info->sectorsize = sectorsize;
3421         fs_info->sectorsize_bits = ilog2(sectorsize);
3422         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3423         fs_info->stripesize = stripesize;
3424
3425         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3426         if (ret) {
3427                 err = ret;
3428                 goto fail_alloc;
3429         }
3430
3431         features = btrfs_super_incompat_flags(disk_super) &
3432                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3433         if (features) {
3434                 btrfs_err(fs_info,
3435                     "cannot mount because of unsupported optional features (0x%llx)",
3436                     features);
3437                 err = -EINVAL;
3438                 goto fail_alloc;
3439         }
3440
3441         features = btrfs_super_incompat_flags(disk_super);
3442         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3443         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3444                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3445         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3446                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3447
3448         /*
3449          * Flag our filesystem as having big metadata blocks if they are bigger
3450          * than the page size.
3451          */
3452         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3453                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3454
3455         /*
3456          * mixed block groups end up with duplicate but slightly offset
3457          * extent buffers for the same range.  It leads to corruptions
3458          */
3459         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3460             (sectorsize != nodesize)) {
3461                 btrfs_err(fs_info,
3462 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3463                         nodesize, sectorsize);
3464                 goto fail_alloc;
3465         }
3466
3467         /*
3468          * Needn't use the lock because there is no other task which will
3469          * update the flag.
3470          */
3471         btrfs_set_super_incompat_flags(disk_super, features);
3472
3473         features = btrfs_super_compat_ro_flags(disk_super) &
3474                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3475         if (!sb_rdonly(sb) && features) {
3476                 btrfs_err(fs_info,
3477         "cannot mount read-write because of unsupported optional features (0x%llx)",
3478                        features);
3479                 err = -EINVAL;
3480                 goto fail_alloc;
3481         }
3482         /*
3483          * We have unsupported RO compat features, although RO mounted, we
3484          * should not cause any metadata write, including log replay.
3485          * Or we could screw up whatever the new feature requires.
3486          */
3487         if (unlikely(features && btrfs_super_log_root(disk_super) &&
3488                      !btrfs_test_opt(fs_info, NOLOGREPLAY))) {
3489                 btrfs_err(fs_info,
3490 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3491                           features);
3492                 err = -EINVAL;
3493                 goto fail_alloc;
3494         }
3495
3496
3497         if (sectorsize < PAGE_SIZE) {
3498                 struct btrfs_subpage_info *subpage_info;
3499
3500                 /*
3501                  * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3502                  * going to be deprecated.
3503                  *
3504                  * Force to use v2 cache for subpage case.
3505                  */
3506                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3507                 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3508                         "forcing free space tree for sector size %u with page size %lu",
3509                         sectorsize, PAGE_SIZE);
3510
3511                 btrfs_warn(fs_info,
3512                 "read-write for sector size %u with page size %lu is experimental",
3513                            sectorsize, PAGE_SIZE);
3514                 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3515                 if (!subpage_info)
3516                         goto fail_alloc;
3517                 btrfs_init_subpage_info(subpage_info, sectorsize);
3518                 fs_info->subpage_info = subpage_info;
3519         }
3520
3521         ret = btrfs_init_workqueues(fs_info);
3522         if (ret) {
3523                 err = ret;
3524                 goto fail_sb_buffer;
3525         }
3526
3527         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3528         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3529
3530         sb->s_blocksize = sectorsize;
3531         sb->s_blocksize_bits = blksize_bits(sectorsize);
3532         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3533
3534         mutex_lock(&fs_info->chunk_mutex);
3535         ret = btrfs_read_sys_array(fs_info);
3536         mutex_unlock(&fs_info->chunk_mutex);
3537         if (ret) {
3538                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3539                 goto fail_sb_buffer;
3540         }
3541
3542         generation = btrfs_super_chunk_root_generation(disk_super);
3543         level = btrfs_super_chunk_root_level(disk_super);
3544         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3545                               generation, level);
3546         if (ret) {
3547                 btrfs_err(fs_info, "failed to read chunk root");
3548                 goto fail_tree_roots;
3549         }
3550
3551         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3552                            offsetof(struct btrfs_header, chunk_tree_uuid),
3553                            BTRFS_UUID_SIZE);
3554
3555         ret = btrfs_read_chunk_tree(fs_info);
3556         if (ret) {
3557                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3558                 goto fail_tree_roots;
3559         }
3560
3561         /*
3562          * At this point we know all the devices that make this filesystem,
3563          * including the seed devices but we don't know yet if the replace
3564          * target is required. So free devices that are not part of this
3565          * filesystem but skip the replace target device which is checked
3566          * below in btrfs_init_dev_replace().
3567          */
3568         btrfs_free_extra_devids(fs_devices);
3569         if (!fs_devices->latest_dev->bdev) {
3570                 btrfs_err(fs_info, "failed to read devices");
3571                 goto fail_tree_roots;
3572         }
3573
3574         ret = init_tree_roots(fs_info);
3575         if (ret)
3576                 goto fail_tree_roots;
3577
3578         /*
3579          * Get zone type information of zoned block devices. This will also
3580          * handle emulation of a zoned filesystem if a regular device has the
3581          * zoned incompat feature flag set.
3582          */
3583         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3584         if (ret) {
3585                 btrfs_err(fs_info,
3586                           "zoned: failed to read device zone info: %d",
3587                           ret);
3588                 goto fail_block_groups;
3589         }
3590
3591         /*
3592          * If we have a uuid root and we're not being told to rescan we need to
3593          * check the generation here so we can set the
3594          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3595          * transaction during a balance or the log replay without updating the
3596          * uuid generation, and then if we crash we would rescan the uuid tree,
3597          * even though it was perfectly fine.
3598          */
3599         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3600             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3601                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3602
3603         ret = btrfs_verify_dev_extents(fs_info);
3604         if (ret) {
3605                 btrfs_err(fs_info,
3606                           "failed to verify dev extents against chunks: %d",
3607                           ret);
3608                 goto fail_block_groups;
3609         }
3610         ret = btrfs_recover_balance(fs_info);
3611         if (ret) {
3612                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3613                 goto fail_block_groups;
3614         }
3615
3616         ret = btrfs_init_dev_stats(fs_info);
3617         if (ret) {
3618                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3619                 goto fail_block_groups;
3620         }
3621
3622         ret = btrfs_init_dev_replace(fs_info);
3623         if (ret) {
3624                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3625                 goto fail_block_groups;
3626         }
3627
3628         ret = btrfs_check_zoned_mode(fs_info);
3629         if (ret) {
3630                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3631                           ret);
3632                 goto fail_block_groups;
3633         }
3634
3635         ret = btrfs_sysfs_add_fsid(fs_devices);
3636         if (ret) {
3637                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3638                                 ret);
3639                 goto fail_block_groups;
3640         }
3641
3642         ret = btrfs_sysfs_add_mounted(fs_info);
3643         if (ret) {
3644                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3645                 goto fail_fsdev_sysfs;
3646         }
3647
3648         ret = btrfs_init_space_info(fs_info);
3649         if (ret) {
3650                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3651                 goto fail_sysfs;
3652         }
3653
3654         ret = btrfs_read_block_groups(fs_info);
3655         if (ret) {
3656                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3657                 goto fail_sysfs;
3658         }
3659
3660         btrfs_free_zone_cache(fs_info);
3661
3662         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3663             !btrfs_check_rw_degradable(fs_info, NULL)) {
3664                 btrfs_warn(fs_info,
3665                 "writable mount is not allowed due to too many missing devices");
3666                 goto fail_sysfs;
3667         }
3668
3669         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3670                                                "btrfs-cleaner");
3671         if (IS_ERR(fs_info->cleaner_kthread))
3672                 goto fail_sysfs;
3673
3674         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3675                                                    tree_root,
3676                                                    "btrfs-transaction");
3677         if (IS_ERR(fs_info->transaction_kthread))
3678                 goto fail_cleaner;
3679
3680         if (!btrfs_test_opt(fs_info, NOSSD) &&
3681             !fs_info->fs_devices->rotating) {
3682                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3683         }
3684
3685         /*
3686          * Mount does not set all options immediately, we can do it now and do
3687          * not have to wait for transaction commit
3688          */
3689         btrfs_apply_pending_changes(fs_info);
3690
3691 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3692         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3693                 ret = btrfsic_mount(fs_info, fs_devices,
3694                                     btrfs_test_opt(fs_info,
3695                                         CHECK_INTEGRITY_DATA) ? 1 : 0,
3696                                     fs_info->check_integrity_print_mask);
3697                 if (ret)
3698                         btrfs_warn(fs_info,
3699                                 "failed to initialize integrity check module: %d",
3700                                 ret);
3701         }
3702 #endif
3703         ret = btrfs_read_qgroup_config(fs_info);
3704         if (ret)
3705                 goto fail_trans_kthread;
3706
3707         if (btrfs_build_ref_tree(fs_info))
3708                 btrfs_err(fs_info, "couldn't build ref tree");
3709
3710         /* do not make disk changes in broken FS or nologreplay is given */
3711         if (btrfs_super_log_root(disk_super) != 0 &&
3712             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3713                 btrfs_info(fs_info, "start tree-log replay");
3714                 ret = btrfs_replay_log(fs_info, fs_devices);
3715                 if (ret) {
3716                         err = ret;
3717                         goto fail_qgroup;
3718                 }
3719         }
3720
3721         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3722         if (IS_ERR(fs_info->fs_root)) {
3723                 err = PTR_ERR(fs_info->fs_root);
3724                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3725                 fs_info->fs_root = NULL;
3726                 goto fail_qgroup;
3727         }
3728
3729         if (sb_rdonly(sb))
3730                 goto clear_oneshot;
3731
3732         ret = btrfs_start_pre_rw_mount(fs_info);
3733         if (ret) {
3734                 close_ctree(fs_info);
3735                 return ret;
3736         }
3737         btrfs_discard_resume(fs_info);
3738
3739         if (fs_info->uuid_root &&
3740             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3741              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3742                 btrfs_info(fs_info, "checking UUID tree");
3743                 ret = btrfs_check_uuid_tree(fs_info);
3744                 if (ret) {
3745                         btrfs_warn(fs_info,
3746                                 "failed to check the UUID tree: %d", ret);
3747                         close_ctree(fs_info);
3748                         return ret;
3749                 }
3750         }
3751
3752         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3753
3754         /* Kick the cleaner thread so it'll start deleting snapshots. */
3755         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3756                 wake_up_process(fs_info->cleaner_kthread);
3757
3758 clear_oneshot:
3759         btrfs_clear_oneshot_options(fs_info);
3760         return 0;
3761
3762 fail_qgroup:
3763         btrfs_free_qgroup_config(fs_info);
3764 fail_trans_kthread:
3765         kthread_stop(fs_info->transaction_kthread);
3766         btrfs_cleanup_transaction(fs_info);
3767         btrfs_free_fs_roots(fs_info);
3768 fail_cleaner:
3769         kthread_stop(fs_info->cleaner_kthread);
3770
3771         /*
3772          * make sure we're done with the btree inode before we stop our
3773          * kthreads
3774          */
3775         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3776
3777 fail_sysfs:
3778         btrfs_sysfs_remove_mounted(fs_info);
3779
3780 fail_fsdev_sysfs:
3781         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3782
3783 fail_block_groups:
3784         btrfs_put_block_group_cache(fs_info);
3785
3786 fail_tree_roots:
3787         if (fs_info->data_reloc_root)
3788                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3789         free_root_pointers(fs_info, true);
3790         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3791
3792 fail_sb_buffer:
3793         btrfs_stop_all_workers(fs_info);
3794         btrfs_free_block_groups(fs_info);
3795 fail_alloc:
3796         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3797
3798         iput(fs_info->btree_inode);
3799 fail:
3800         btrfs_close_devices(fs_info->fs_devices);
3801         return err;
3802 }
3803 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3804
3805 static void btrfs_end_super_write(struct bio *bio)
3806 {
3807         struct btrfs_device *device = bio->bi_private;
3808         struct bio_vec *bvec;
3809         struct bvec_iter_all iter_all;
3810         struct page *page;
3811
3812         bio_for_each_segment_all(bvec, bio, iter_all) {
3813                 page = bvec->bv_page;
3814
3815                 if (bio->bi_status) {
3816                         btrfs_warn_rl_in_rcu(device->fs_info,
3817                                 "lost page write due to IO error on %s (%d)",
3818                                 rcu_str_deref(device->name),
3819                                 blk_status_to_errno(bio->bi_status));
3820                         ClearPageUptodate(page);
3821                         SetPageError(page);
3822                         btrfs_dev_stat_inc_and_print(device,
3823                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3824                 } else {
3825                         SetPageUptodate(page);
3826                 }
3827
3828                 put_page(page);
3829                 unlock_page(page);
3830         }
3831
3832         bio_put(bio);
3833 }
3834
3835 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3836                                                    int copy_num)
3837 {
3838         struct btrfs_super_block *super;
3839         struct page *page;
3840         u64 bytenr, bytenr_orig;
3841         struct address_space *mapping = bdev->bd_inode->i_mapping;
3842         int ret;
3843
3844         bytenr_orig = btrfs_sb_offset(copy_num);
3845         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3846         if (ret == -ENOENT)
3847                 return ERR_PTR(-EINVAL);
3848         else if (ret)
3849                 return ERR_PTR(ret);
3850
3851         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3852                 return ERR_PTR(-EINVAL);
3853
3854         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3855         if (IS_ERR(page))
3856                 return ERR_CAST(page);
3857
3858         super = page_address(page);
3859         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3860                 btrfs_release_disk_super(super);
3861                 return ERR_PTR(-ENODATA);
3862         }
3863
3864         if (btrfs_super_bytenr(super) != bytenr_orig) {
3865                 btrfs_release_disk_super(super);
3866                 return ERR_PTR(-EINVAL);
3867         }
3868
3869         return super;
3870 }
3871
3872
3873 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3874 {
3875         struct btrfs_super_block *super, *latest = NULL;
3876         int i;
3877         u64 transid = 0;
3878
3879         /* we would like to check all the supers, but that would make
3880          * a btrfs mount succeed after a mkfs from a different FS.
3881          * So, we need to add a special mount option to scan for
3882          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3883          */
3884         for (i = 0; i < 1; i++) {
3885                 super = btrfs_read_dev_one_super(bdev, i);
3886                 if (IS_ERR(super))
3887                         continue;
3888
3889                 if (!latest || btrfs_super_generation(super) > transid) {
3890                         if (latest)
3891                                 btrfs_release_disk_super(super);
3892
3893                         latest = super;
3894                         transid = btrfs_super_generation(super);
3895                 }
3896         }
3897
3898         return super;
3899 }
3900
3901 /*
3902  * Write superblock @sb to the @device. Do not wait for completion, all the
3903  * pages we use for writing are locked.
3904  *
3905  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3906  * the expected device size at commit time. Note that max_mirrors must be
3907  * same for write and wait phases.
3908  *
3909  * Return number of errors when page is not found or submission fails.
3910  */
3911 static int write_dev_supers(struct btrfs_device *device,
3912                             struct btrfs_super_block *sb, int max_mirrors)
3913 {
3914         struct btrfs_fs_info *fs_info = device->fs_info;
3915         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3916         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3917         int i;
3918         int errors = 0;
3919         int ret;
3920         u64 bytenr, bytenr_orig;
3921
3922         if (max_mirrors == 0)
3923                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3924
3925         shash->tfm = fs_info->csum_shash;
3926
3927         for (i = 0; i < max_mirrors; i++) {
3928                 struct page *page;
3929                 struct bio *bio;
3930                 struct btrfs_super_block *disk_super;
3931
3932                 bytenr_orig = btrfs_sb_offset(i);
3933                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3934                 if (ret == -ENOENT) {
3935                         continue;
3936                 } else if (ret < 0) {
3937                         btrfs_err(device->fs_info,
3938                                 "couldn't get super block location for mirror %d",
3939                                 i);
3940                         errors++;
3941                         continue;
3942                 }
3943                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3944                     device->commit_total_bytes)
3945                         break;
3946
3947                 btrfs_set_super_bytenr(sb, bytenr_orig);
3948
3949                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3950                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3951                                     sb->csum);
3952
3953                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3954                                            GFP_NOFS);
3955                 if (!page) {
3956                         btrfs_err(device->fs_info,
3957                             "couldn't get super block page for bytenr %llu",
3958                             bytenr);
3959                         errors++;
3960                         continue;
3961                 }
3962
3963                 /* Bump the refcount for wait_dev_supers() */
3964                 get_page(page);
3965
3966                 disk_super = page_address(page);
3967                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3968
3969                 /*
3970                  * Directly use bios here instead of relying on the page cache
3971                  * to do I/O, so we don't lose the ability to do integrity
3972                  * checking.
3973                  */
3974                 bio = bio_alloc(device->bdev, 1,
3975                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3976                                 GFP_NOFS);
3977                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3978                 bio->bi_private = device;
3979                 bio->bi_end_io = btrfs_end_super_write;
3980                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3981                                offset_in_page(bytenr));
3982
3983                 /*
3984                  * We FUA only the first super block.  The others we allow to
3985                  * go down lazy and there's a short window where the on-disk
3986                  * copies might still contain the older version.
3987                  */
3988                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3989                         bio->bi_opf |= REQ_FUA;
3990
3991                 btrfsic_check_bio(bio);
3992                 submit_bio(bio);
3993
3994                 if (btrfs_advance_sb_log(device, i))
3995                         errors++;
3996         }
3997         return errors < i ? 0 : -1;
3998 }
3999
4000 /*
4001  * Wait for write completion of superblocks done by write_dev_supers,
4002  * @max_mirrors same for write and wait phases.
4003  *
4004  * Return number of errors when page is not found or not marked up to
4005  * date.
4006  */
4007 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4008 {
4009         int i;
4010         int errors = 0;
4011         bool primary_failed = false;
4012         int ret;
4013         u64 bytenr;
4014
4015         if (max_mirrors == 0)
4016                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4017
4018         for (i = 0; i < max_mirrors; i++) {
4019                 struct page *page;
4020
4021                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4022                 if (ret == -ENOENT) {
4023                         break;
4024                 } else if (ret < 0) {
4025                         errors++;
4026                         if (i == 0)
4027                                 primary_failed = true;
4028                         continue;
4029                 }
4030                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4031                     device->commit_total_bytes)
4032                         break;
4033
4034                 page = find_get_page(device->bdev->bd_inode->i_mapping,
4035                                      bytenr >> PAGE_SHIFT);
4036                 if (!page) {
4037                         errors++;
4038                         if (i == 0)
4039                                 primary_failed = true;
4040                         continue;
4041                 }
4042                 /* Page is submitted locked and unlocked once the IO completes */
4043                 wait_on_page_locked(page);
4044                 if (PageError(page)) {
4045                         errors++;
4046                         if (i == 0)
4047                                 primary_failed = true;
4048                 }
4049
4050                 /* Drop our reference */
4051                 put_page(page);
4052
4053                 /* Drop the reference from the writing run */
4054                 put_page(page);
4055         }
4056
4057         /* log error, force error return */
4058         if (primary_failed) {
4059                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4060                           device->devid);
4061                 return -1;
4062         }
4063
4064         return errors < i ? 0 : -1;
4065 }
4066
4067 /*
4068  * endio for the write_dev_flush, this will wake anyone waiting
4069  * for the barrier when it is done
4070  */
4071 static void btrfs_end_empty_barrier(struct bio *bio)
4072 {
4073         bio_uninit(bio);
4074         complete(bio->bi_private);
4075 }
4076
4077 /*
4078  * Submit a flush request to the device if it supports it. Error handling is
4079  * done in the waiting counterpart.
4080  */
4081 static void write_dev_flush(struct btrfs_device *device)
4082 {
4083         struct bio *bio = &device->flush_bio;
4084
4085 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4086         /*
4087          * When a disk has write caching disabled, we skip submission of a bio
4088          * with flush and sync requests before writing the superblock, since
4089          * it's not needed. However when the integrity checker is enabled, this
4090          * results in reports that there are metadata blocks referred by a
4091          * superblock that were not properly flushed. So don't skip the bio
4092          * submission only when the integrity checker is enabled for the sake
4093          * of simplicity, since this is a debug tool and not meant for use in
4094          * non-debug builds.
4095          */
4096         if (!bdev_write_cache(device->bdev))
4097                 return;
4098 #endif
4099
4100         bio_init(bio, device->bdev, NULL, 0,
4101                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4102         bio->bi_end_io = btrfs_end_empty_barrier;
4103         init_completion(&device->flush_wait);
4104         bio->bi_private = &device->flush_wait;
4105
4106         btrfsic_check_bio(bio);
4107         submit_bio(bio);
4108         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4109 }
4110
4111 /*
4112  * If the flush bio has been submitted by write_dev_flush, wait for it.
4113  */
4114 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4115 {
4116         struct bio *bio = &device->flush_bio;
4117
4118         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4119                 return BLK_STS_OK;
4120
4121         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4122         wait_for_completion_io(&device->flush_wait);
4123
4124         return bio->bi_status;
4125 }
4126
4127 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4128 {
4129         if (!btrfs_check_rw_degradable(fs_info, NULL))
4130                 return -EIO;
4131         return 0;
4132 }
4133
4134 /*
4135  * send an empty flush down to each device in parallel,
4136  * then wait for them
4137  */
4138 static int barrier_all_devices(struct btrfs_fs_info *info)
4139 {
4140         struct list_head *head;
4141         struct btrfs_device *dev;
4142         int errors_wait = 0;
4143         blk_status_t ret;
4144
4145         lockdep_assert_held(&info->fs_devices->device_list_mutex);
4146         /* send down all the barriers */
4147         head = &info->fs_devices->devices;
4148         list_for_each_entry(dev, head, dev_list) {
4149                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4150                         continue;
4151                 if (!dev->bdev)
4152                         continue;
4153                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4154                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4155                         continue;
4156
4157                 write_dev_flush(dev);
4158                 dev->last_flush_error = BLK_STS_OK;
4159         }
4160
4161         /* wait for all the barriers */
4162         list_for_each_entry(dev, head, dev_list) {
4163                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4164                         continue;
4165                 if (!dev->bdev) {
4166                         errors_wait++;
4167                         continue;
4168                 }
4169                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4170                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4171                         continue;
4172
4173                 ret = wait_dev_flush(dev);
4174                 if (ret) {
4175                         dev->last_flush_error = ret;
4176                         btrfs_dev_stat_inc_and_print(dev,
4177                                         BTRFS_DEV_STAT_FLUSH_ERRS);
4178                         errors_wait++;
4179                 }
4180         }
4181
4182         if (errors_wait) {
4183                 /*
4184                  * At some point we need the status of all disks
4185                  * to arrive at the volume status. So error checking
4186                  * is being pushed to a separate loop.
4187                  */
4188                 return check_barrier_error(info);
4189         }
4190         return 0;
4191 }
4192
4193 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4194 {
4195         int raid_type;
4196         int min_tolerated = INT_MAX;
4197
4198         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4199             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4200                 min_tolerated = min_t(int, min_tolerated,
4201                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4202                                     tolerated_failures);
4203
4204         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4205                 if (raid_type == BTRFS_RAID_SINGLE)
4206                         continue;
4207                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4208                         continue;
4209                 min_tolerated = min_t(int, min_tolerated,
4210                                     btrfs_raid_array[raid_type].
4211                                     tolerated_failures);
4212         }
4213
4214         if (min_tolerated == INT_MAX) {
4215                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4216                 min_tolerated = 0;
4217         }
4218
4219         return min_tolerated;
4220 }
4221
4222 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4223 {
4224         struct list_head *head;
4225         struct btrfs_device *dev;
4226         struct btrfs_super_block *sb;
4227         struct btrfs_dev_item *dev_item;
4228         int ret;
4229         int do_barriers;
4230         int max_errors;
4231         int total_errors = 0;
4232         u64 flags;
4233
4234         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4235
4236         /*
4237          * max_mirrors == 0 indicates we're from commit_transaction,
4238          * not from fsync where the tree roots in fs_info have not
4239          * been consistent on disk.
4240          */
4241         if (max_mirrors == 0)
4242                 backup_super_roots(fs_info);
4243
4244         sb = fs_info->super_for_commit;
4245         dev_item = &sb->dev_item;
4246
4247         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4248         head = &fs_info->fs_devices->devices;
4249         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4250
4251         if (do_barriers) {
4252                 ret = barrier_all_devices(fs_info);
4253                 if (ret) {
4254                         mutex_unlock(
4255                                 &fs_info->fs_devices->device_list_mutex);
4256                         btrfs_handle_fs_error(fs_info, ret,
4257                                               "errors while submitting device barriers.");
4258                         return ret;
4259                 }
4260         }
4261
4262         list_for_each_entry(dev, head, dev_list) {
4263                 if (!dev->bdev) {
4264                         total_errors++;
4265                         continue;
4266                 }
4267                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4268                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4269                         continue;
4270
4271                 btrfs_set_stack_device_generation(dev_item, 0);
4272                 btrfs_set_stack_device_type(dev_item, dev->type);
4273                 btrfs_set_stack_device_id(dev_item, dev->devid);
4274                 btrfs_set_stack_device_total_bytes(dev_item,
4275                                                    dev->commit_total_bytes);
4276                 btrfs_set_stack_device_bytes_used(dev_item,
4277                                                   dev->commit_bytes_used);
4278                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4279                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4280                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4281                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4282                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4283                        BTRFS_FSID_SIZE);
4284
4285                 flags = btrfs_super_flags(sb);
4286                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4287
4288                 ret = btrfs_validate_write_super(fs_info, sb);
4289                 if (ret < 0) {
4290                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4291                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4292                                 "unexpected superblock corruption detected");
4293                         return -EUCLEAN;
4294                 }
4295
4296                 ret = write_dev_supers(dev, sb, max_mirrors);
4297                 if (ret)
4298                         total_errors++;
4299         }
4300         if (total_errors > max_errors) {
4301                 btrfs_err(fs_info, "%d errors while writing supers",
4302                           total_errors);
4303                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4304
4305                 /* FUA is masked off if unsupported and can't be the reason */
4306                 btrfs_handle_fs_error(fs_info, -EIO,
4307                                       "%d errors while writing supers",
4308                                       total_errors);
4309                 return -EIO;
4310         }
4311
4312         total_errors = 0;
4313         list_for_each_entry(dev, head, dev_list) {
4314                 if (!dev->bdev)
4315                         continue;
4316                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4317                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4318                         continue;
4319
4320                 ret = wait_dev_supers(dev, max_mirrors);
4321                 if (ret)
4322                         total_errors++;
4323         }
4324         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4325         if (total_errors > max_errors) {
4326                 btrfs_handle_fs_error(fs_info, -EIO,
4327                                       "%d errors while writing supers",
4328                                       total_errors);
4329                 return -EIO;
4330         }
4331         return 0;
4332 }
4333
4334 /* Drop a fs root from the radix tree and free it. */
4335 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4336                                   struct btrfs_root *root)
4337 {
4338         bool drop_ref = false;
4339
4340         spin_lock(&fs_info->fs_roots_radix_lock);
4341         radix_tree_delete(&fs_info->fs_roots_radix,
4342                           (unsigned long)root->root_key.objectid);
4343         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4344                 drop_ref = true;
4345         spin_unlock(&fs_info->fs_roots_radix_lock);
4346
4347         if (BTRFS_FS_ERROR(fs_info)) {
4348                 ASSERT(root->log_root == NULL);
4349                 if (root->reloc_root) {
4350                         btrfs_put_root(root->reloc_root);
4351                         root->reloc_root = NULL;
4352                 }
4353         }
4354
4355         if (drop_ref)
4356                 btrfs_put_root(root);
4357 }
4358
4359 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4360 {
4361         u64 root_objectid = 0;
4362         struct btrfs_root *gang[8];
4363         int i = 0;
4364         int err = 0;
4365         unsigned int ret = 0;
4366
4367         while (1) {
4368                 spin_lock(&fs_info->fs_roots_radix_lock);
4369                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4370                                              (void **)gang, root_objectid,
4371                                              ARRAY_SIZE(gang));
4372                 if (!ret) {
4373                         spin_unlock(&fs_info->fs_roots_radix_lock);
4374                         break;
4375                 }
4376                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4377
4378                 for (i = 0; i < ret; i++) {
4379                         /* Avoid to grab roots in dead_roots */
4380                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4381                                 gang[i] = NULL;
4382                                 continue;
4383                         }
4384                         /* grab all the search result for later use */
4385                         gang[i] = btrfs_grab_root(gang[i]);
4386                 }
4387                 spin_unlock(&fs_info->fs_roots_radix_lock);
4388
4389                 for (i = 0; i < ret; i++) {
4390                         if (!gang[i])
4391                                 continue;
4392                         root_objectid = gang[i]->root_key.objectid;
4393                         err = btrfs_orphan_cleanup(gang[i]);
4394                         if (err)
4395                                 break;
4396                         btrfs_put_root(gang[i]);
4397                 }
4398                 root_objectid++;
4399         }
4400
4401         /* release the uncleaned roots due to error */
4402         for (; i < ret; i++) {
4403                 if (gang[i])
4404                         btrfs_put_root(gang[i]);
4405         }
4406         return err;
4407 }
4408
4409 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4410 {
4411         struct btrfs_root *root = fs_info->tree_root;
4412         struct btrfs_trans_handle *trans;
4413
4414         mutex_lock(&fs_info->cleaner_mutex);
4415         btrfs_run_delayed_iputs(fs_info);
4416         mutex_unlock(&fs_info->cleaner_mutex);
4417         wake_up_process(fs_info->cleaner_kthread);
4418
4419         /* wait until ongoing cleanup work done */
4420         down_write(&fs_info->cleanup_work_sem);
4421         up_write(&fs_info->cleanup_work_sem);
4422
4423         trans = btrfs_join_transaction(root);
4424         if (IS_ERR(trans))
4425                 return PTR_ERR(trans);
4426         return btrfs_commit_transaction(trans);
4427 }
4428
4429 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4430 {
4431         struct btrfs_transaction *trans;
4432         struct btrfs_transaction *tmp;
4433         bool found = false;
4434
4435         if (list_empty(&fs_info->trans_list))
4436                 return;
4437
4438         /*
4439          * This function is only called at the very end of close_ctree(),
4440          * thus no other running transaction, no need to take trans_lock.
4441          */
4442         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4443         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4444                 struct extent_state *cached = NULL;
4445                 u64 dirty_bytes = 0;
4446                 u64 cur = 0;
4447                 u64 found_start;
4448                 u64 found_end;
4449
4450                 found = true;
4451                 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4452                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4453                         dirty_bytes += found_end + 1 - found_start;
4454                         cur = found_end + 1;
4455                 }
4456                 btrfs_warn(fs_info,
4457         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4458                            trans->transid, dirty_bytes);
4459                 btrfs_cleanup_one_transaction(trans, fs_info);
4460
4461                 if (trans == fs_info->running_transaction)
4462                         fs_info->running_transaction = NULL;
4463                 list_del_init(&trans->list);
4464
4465                 btrfs_put_transaction(trans);
4466                 trace_btrfs_transaction_commit(fs_info);
4467         }
4468         ASSERT(!found);
4469 }
4470
4471 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4472 {
4473         int ret;
4474
4475         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4476
4477         /*
4478          * If we had UNFINISHED_DROPS we could still be processing them, so
4479          * clear that bit and wake up relocation so it can stop.
4480          * We must do this before stopping the block group reclaim task, because
4481          * at btrfs_relocate_block_group() we wait for this bit, and after the
4482          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4483          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4484          * return 1.
4485          */
4486         btrfs_wake_unfinished_drop(fs_info);
4487
4488         /*
4489          * We may have the reclaim task running and relocating a data block group,
4490          * in which case it may create delayed iputs. So stop it before we park
4491          * the cleaner kthread otherwise we can get new delayed iputs after
4492          * parking the cleaner, and that can make the async reclaim task to hang
4493          * if it's waiting for delayed iputs to complete, since the cleaner is
4494          * parked and can not run delayed iputs - this will make us hang when
4495          * trying to stop the async reclaim task.
4496          */
4497         cancel_work_sync(&fs_info->reclaim_bgs_work);
4498         /*
4499          * We don't want the cleaner to start new transactions, add more delayed
4500          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4501          * because that frees the task_struct, and the transaction kthread might
4502          * still try to wake up the cleaner.
4503          */
4504         kthread_park(fs_info->cleaner_kthread);
4505
4506         /* wait for the qgroup rescan worker to stop */
4507         btrfs_qgroup_wait_for_completion(fs_info, false);
4508
4509         /* wait for the uuid_scan task to finish */
4510         down(&fs_info->uuid_tree_rescan_sem);
4511         /* avoid complains from lockdep et al., set sem back to initial state */
4512         up(&fs_info->uuid_tree_rescan_sem);
4513
4514         /* pause restriper - we want to resume on mount */
4515         btrfs_pause_balance(fs_info);
4516
4517         btrfs_dev_replace_suspend_for_unmount(fs_info);
4518
4519         btrfs_scrub_cancel(fs_info);
4520
4521         /* wait for any defraggers to finish */
4522         wait_event(fs_info->transaction_wait,
4523                    (atomic_read(&fs_info->defrag_running) == 0));
4524
4525         /* clear out the rbtree of defraggable inodes */
4526         btrfs_cleanup_defrag_inodes(fs_info);
4527
4528         /*
4529          * After we parked the cleaner kthread, ordered extents may have
4530          * completed and created new delayed iputs. If one of the async reclaim
4531          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4532          * can hang forever trying to stop it, because if a delayed iput is
4533          * added after it ran btrfs_run_delayed_iputs() and before it called
4534          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4535          * no one else to run iputs.
4536          *
4537          * So wait for all ongoing ordered extents to complete and then run
4538          * delayed iputs. This works because once we reach this point no one
4539          * can either create new ordered extents nor create delayed iputs
4540          * through some other means.
4541          *
4542          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4543          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4544          * but the delayed iput for the respective inode is made only when doing
4545          * the final btrfs_put_ordered_extent() (which must happen at
4546          * btrfs_finish_ordered_io() when we are unmounting).
4547          */
4548         btrfs_flush_workqueue(fs_info->endio_write_workers);
4549         /* Ordered extents for free space inodes. */
4550         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4551         btrfs_run_delayed_iputs(fs_info);
4552
4553         cancel_work_sync(&fs_info->async_reclaim_work);
4554         cancel_work_sync(&fs_info->async_data_reclaim_work);
4555         cancel_work_sync(&fs_info->preempt_reclaim_work);
4556
4557         /* Cancel or finish ongoing discard work */
4558         btrfs_discard_cleanup(fs_info);
4559
4560         if (!sb_rdonly(fs_info->sb)) {
4561                 /*
4562                  * The cleaner kthread is stopped, so do one final pass over
4563                  * unused block groups.
4564                  */
4565                 btrfs_delete_unused_bgs(fs_info);
4566
4567                 /*
4568                  * There might be existing delayed inode workers still running
4569                  * and holding an empty delayed inode item. We must wait for
4570                  * them to complete first because they can create a transaction.
4571                  * This happens when someone calls btrfs_balance_delayed_items()
4572                  * and then a transaction commit runs the same delayed nodes
4573                  * before any delayed worker has done something with the nodes.
4574                  * We must wait for any worker here and not at transaction
4575                  * commit time since that could cause a deadlock.
4576                  * This is a very rare case.
4577                  */
4578                 btrfs_flush_workqueue(fs_info->delayed_workers);
4579
4580                 ret = btrfs_commit_super(fs_info);
4581                 if (ret)
4582                         btrfs_err(fs_info, "commit super ret %d", ret);
4583         }
4584
4585         if (BTRFS_FS_ERROR(fs_info))
4586                 btrfs_error_commit_super(fs_info);
4587
4588         kthread_stop(fs_info->transaction_kthread);
4589         kthread_stop(fs_info->cleaner_kthread);
4590
4591         ASSERT(list_empty(&fs_info->delayed_iputs));
4592         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4593
4594         if (btrfs_check_quota_leak(fs_info)) {
4595                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4596                 btrfs_err(fs_info, "qgroup reserved space leaked");
4597         }
4598
4599         btrfs_free_qgroup_config(fs_info);
4600         ASSERT(list_empty(&fs_info->delalloc_roots));
4601
4602         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4603                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4604                        percpu_counter_sum(&fs_info->delalloc_bytes));
4605         }
4606
4607         if (percpu_counter_sum(&fs_info->ordered_bytes))
4608                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4609                            percpu_counter_sum(&fs_info->ordered_bytes));
4610
4611         btrfs_sysfs_remove_mounted(fs_info);
4612         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4613
4614         btrfs_put_block_group_cache(fs_info);
4615
4616         /*
4617          * we must make sure there is not any read request to
4618          * submit after we stopping all workers.
4619          */
4620         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4621         btrfs_stop_all_workers(fs_info);
4622
4623         /* We shouldn't have any transaction open at this point */
4624         warn_about_uncommitted_trans(fs_info);
4625
4626         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4627         free_root_pointers(fs_info, true);
4628         btrfs_free_fs_roots(fs_info);
4629
4630         /*
4631          * We must free the block groups after dropping the fs_roots as we could
4632          * have had an IO error and have left over tree log blocks that aren't
4633          * cleaned up until the fs roots are freed.  This makes the block group
4634          * accounting appear to be wrong because there's pending reserved bytes,
4635          * so make sure we do the block group cleanup afterwards.
4636          */
4637         btrfs_free_block_groups(fs_info);
4638
4639         iput(fs_info->btree_inode);
4640
4641 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4642         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4643                 btrfsic_unmount(fs_info->fs_devices);
4644 #endif
4645
4646         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4647         btrfs_close_devices(fs_info->fs_devices);
4648 }
4649
4650 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4651                           int atomic)
4652 {
4653         int ret;
4654         struct inode *btree_inode = buf->pages[0]->mapping->host;
4655
4656         ret = extent_buffer_uptodate(buf);
4657         if (!ret)
4658                 return ret;
4659
4660         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4661                                     parent_transid, atomic);
4662         if (ret == -EAGAIN)
4663                 return ret;
4664         return !ret;
4665 }
4666
4667 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4668 {
4669         struct btrfs_fs_info *fs_info = buf->fs_info;
4670         u64 transid = btrfs_header_generation(buf);
4671         int was_dirty;
4672
4673 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4674         /*
4675          * This is a fast path so only do this check if we have sanity tests
4676          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4677          * outside of the sanity tests.
4678          */
4679         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4680                 return;
4681 #endif
4682         btrfs_assert_tree_write_locked(buf);
4683         if (transid != fs_info->generation)
4684                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4685                         buf->start, transid, fs_info->generation);
4686         was_dirty = set_extent_buffer_dirty(buf);
4687         if (!was_dirty)
4688                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4689                                          buf->len,
4690                                          fs_info->dirty_metadata_batch);
4691 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4692         /*
4693          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4694          * but item data not updated.
4695          * So here we should only check item pointers, not item data.
4696          */
4697         if (btrfs_header_level(buf) == 0 &&
4698             btrfs_check_leaf_relaxed(buf)) {
4699                 btrfs_print_leaf(buf);
4700                 ASSERT(0);
4701         }
4702 #endif
4703 }
4704
4705 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4706                                         int flush_delayed)
4707 {
4708         /*
4709          * looks as though older kernels can get into trouble with
4710          * this code, they end up stuck in balance_dirty_pages forever
4711          */
4712         int ret;
4713
4714         if (current->flags & PF_MEMALLOC)
4715                 return;
4716
4717         if (flush_delayed)
4718                 btrfs_balance_delayed_items(fs_info);
4719
4720         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4721                                      BTRFS_DIRTY_METADATA_THRESH,
4722                                      fs_info->dirty_metadata_batch);
4723         if (ret > 0) {
4724                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4725         }
4726 }
4727
4728 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4729 {
4730         __btrfs_btree_balance_dirty(fs_info, 1);
4731 }
4732
4733 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4734 {
4735         __btrfs_btree_balance_dirty(fs_info, 0);
4736 }
4737
4738 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4739 {
4740         /* cleanup FS via transaction */
4741         btrfs_cleanup_transaction(fs_info);
4742
4743         mutex_lock(&fs_info->cleaner_mutex);
4744         btrfs_run_delayed_iputs(fs_info);
4745         mutex_unlock(&fs_info->cleaner_mutex);
4746
4747         down_write(&fs_info->cleanup_work_sem);
4748         up_write(&fs_info->cleanup_work_sem);
4749 }
4750
4751 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4752 {
4753         struct btrfs_root *gang[8];
4754         u64 root_objectid = 0;
4755         int ret;
4756
4757         spin_lock(&fs_info->fs_roots_radix_lock);
4758         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4759                                              (void **)gang, root_objectid,
4760                                              ARRAY_SIZE(gang))) != 0) {
4761                 int i;
4762
4763                 for (i = 0; i < ret; i++)
4764                         gang[i] = btrfs_grab_root(gang[i]);
4765                 spin_unlock(&fs_info->fs_roots_radix_lock);
4766
4767                 for (i = 0; i < ret; i++) {
4768                         if (!gang[i])
4769                                 continue;
4770                         root_objectid = gang[i]->root_key.objectid;
4771                         btrfs_free_log(NULL, gang[i]);
4772                         btrfs_put_root(gang[i]);
4773                 }
4774                 root_objectid++;
4775                 spin_lock(&fs_info->fs_roots_radix_lock);
4776         }
4777         spin_unlock(&fs_info->fs_roots_radix_lock);
4778         btrfs_free_log_root_tree(NULL, fs_info);
4779 }
4780
4781 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4782 {
4783         struct btrfs_ordered_extent *ordered;
4784
4785         spin_lock(&root->ordered_extent_lock);
4786         /*
4787          * This will just short circuit the ordered completion stuff which will
4788          * make sure the ordered extent gets properly cleaned up.
4789          */
4790         list_for_each_entry(ordered, &root->ordered_extents,
4791                             root_extent_list)
4792                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4793         spin_unlock(&root->ordered_extent_lock);
4794 }
4795
4796 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4797 {
4798         struct btrfs_root *root;
4799         struct list_head splice;
4800
4801         INIT_LIST_HEAD(&splice);
4802
4803         spin_lock(&fs_info->ordered_root_lock);
4804         list_splice_init(&fs_info->ordered_roots, &splice);
4805         while (!list_empty(&splice)) {
4806                 root = list_first_entry(&splice, struct btrfs_root,
4807                                         ordered_root);
4808                 list_move_tail(&root->ordered_root,
4809                                &fs_info->ordered_roots);
4810
4811                 spin_unlock(&fs_info->ordered_root_lock);
4812                 btrfs_destroy_ordered_extents(root);
4813
4814                 cond_resched();
4815                 spin_lock(&fs_info->ordered_root_lock);
4816         }
4817         spin_unlock(&fs_info->ordered_root_lock);
4818
4819         /*
4820          * We need this here because if we've been flipped read-only we won't
4821          * get sync() from the umount, so we need to make sure any ordered
4822          * extents that haven't had their dirty pages IO start writeout yet
4823          * actually get run and error out properly.
4824          */
4825         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4826 }
4827
4828 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4829                                       struct btrfs_fs_info *fs_info)
4830 {
4831         struct rb_node *node;
4832         struct btrfs_delayed_ref_root *delayed_refs;
4833         struct btrfs_delayed_ref_node *ref;
4834         int ret = 0;
4835
4836         delayed_refs = &trans->delayed_refs;
4837
4838         spin_lock(&delayed_refs->lock);
4839         if (atomic_read(&delayed_refs->num_entries) == 0) {
4840                 spin_unlock(&delayed_refs->lock);
4841                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4842                 return ret;
4843         }
4844
4845         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4846                 struct btrfs_delayed_ref_head *head;
4847                 struct rb_node *n;
4848                 bool pin_bytes = false;
4849
4850                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4851                                 href_node);
4852                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4853                         continue;
4854
4855                 spin_lock(&head->lock);
4856                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4857                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4858                                        ref_node);
4859                         ref->in_tree = 0;
4860                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4861                         RB_CLEAR_NODE(&ref->ref_node);
4862                         if (!list_empty(&ref->add_list))
4863                                 list_del(&ref->add_list);
4864                         atomic_dec(&delayed_refs->num_entries);
4865                         btrfs_put_delayed_ref(ref);
4866                 }
4867                 if (head->must_insert_reserved)
4868                         pin_bytes = true;
4869                 btrfs_free_delayed_extent_op(head->extent_op);
4870                 btrfs_delete_ref_head(delayed_refs, head);
4871                 spin_unlock(&head->lock);
4872                 spin_unlock(&delayed_refs->lock);
4873                 mutex_unlock(&head->mutex);
4874
4875                 if (pin_bytes) {
4876                         struct btrfs_block_group *cache;
4877
4878                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4879                         BUG_ON(!cache);
4880
4881                         spin_lock(&cache->space_info->lock);
4882                         spin_lock(&cache->lock);
4883                         cache->pinned += head->num_bytes;
4884                         btrfs_space_info_update_bytes_pinned(fs_info,
4885                                 cache->space_info, head->num_bytes);
4886                         cache->reserved -= head->num_bytes;
4887                         cache->space_info->bytes_reserved -= head->num_bytes;
4888                         spin_unlock(&cache->lock);
4889                         spin_unlock(&cache->space_info->lock);
4890
4891                         btrfs_put_block_group(cache);
4892
4893                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4894                                 head->bytenr + head->num_bytes - 1);
4895                 }
4896                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4897                 btrfs_put_delayed_ref_head(head);
4898                 cond_resched();
4899                 spin_lock(&delayed_refs->lock);
4900         }
4901         btrfs_qgroup_destroy_extent_records(trans);
4902
4903         spin_unlock(&delayed_refs->lock);
4904
4905         return ret;
4906 }
4907
4908 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4909 {
4910         struct btrfs_inode *btrfs_inode;
4911         struct list_head splice;
4912
4913         INIT_LIST_HEAD(&splice);
4914
4915         spin_lock(&root->delalloc_lock);
4916         list_splice_init(&root->delalloc_inodes, &splice);
4917
4918         while (!list_empty(&splice)) {
4919                 struct inode *inode = NULL;
4920                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4921                                                delalloc_inodes);
4922                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4923                 spin_unlock(&root->delalloc_lock);
4924
4925                 /*
4926                  * Make sure we get a live inode and that it'll not disappear
4927                  * meanwhile.
4928                  */
4929                 inode = igrab(&btrfs_inode->vfs_inode);
4930                 if (inode) {
4931                         invalidate_inode_pages2(inode->i_mapping);
4932                         iput(inode);
4933                 }
4934                 spin_lock(&root->delalloc_lock);
4935         }
4936         spin_unlock(&root->delalloc_lock);
4937 }
4938
4939 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4940 {
4941         struct btrfs_root *root;
4942         struct list_head splice;
4943
4944         INIT_LIST_HEAD(&splice);
4945
4946         spin_lock(&fs_info->delalloc_root_lock);
4947         list_splice_init(&fs_info->delalloc_roots, &splice);
4948         while (!list_empty(&splice)) {
4949                 root = list_first_entry(&splice, struct btrfs_root,
4950                                          delalloc_root);
4951                 root = btrfs_grab_root(root);
4952                 BUG_ON(!root);
4953                 spin_unlock(&fs_info->delalloc_root_lock);
4954
4955                 btrfs_destroy_delalloc_inodes(root);
4956                 btrfs_put_root(root);
4957
4958                 spin_lock(&fs_info->delalloc_root_lock);
4959         }
4960         spin_unlock(&fs_info->delalloc_root_lock);
4961 }
4962
4963 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4964                                         struct extent_io_tree *dirty_pages,
4965                                         int mark)
4966 {
4967         int ret;
4968         struct extent_buffer *eb;
4969         u64 start = 0;
4970         u64 end;
4971
4972         while (1) {
4973                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4974                                             mark, NULL);
4975                 if (ret)
4976                         break;
4977
4978                 clear_extent_bits(dirty_pages, start, end, mark);
4979                 while (start <= end) {
4980                         eb = find_extent_buffer(fs_info, start);
4981                         start += fs_info->nodesize;
4982                         if (!eb)
4983                                 continue;
4984                         wait_on_extent_buffer_writeback(eb);
4985
4986                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4987                                                &eb->bflags))
4988                                 clear_extent_buffer_dirty(eb);
4989                         free_extent_buffer_stale(eb);
4990                 }
4991         }
4992
4993         return ret;
4994 }
4995
4996 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4997                                        struct extent_io_tree *unpin)
4998 {
4999         u64 start;
5000         u64 end;
5001         int ret;
5002
5003         while (1) {
5004                 struct extent_state *cached_state = NULL;
5005
5006                 /*
5007                  * The btrfs_finish_extent_commit() may get the same range as
5008                  * ours between find_first_extent_bit and clear_extent_dirty.
5009                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5010                  * the same extent range.
5011                  */
5012                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5013                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5014                                             EXTENT_DIRTY, &cached_state);
5015                 if (ret) {
5016                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5017                         break;
5018                 }
5019
5020                 clear_extent_dirty(unpin, start, end, &cached_state);
5021                 free_extent_state(cached_state);
5022                 btrfs_error_unpin_extent_range(fs_info, start, end);
5023                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5024                 cond_resched();
5025         }
5026
5027         return 0;
5028 }
5029
5030 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5031 {
5032         struct inode *inode;
5033
5034         inode = cache->io_ctl.inode;
5035         if (inode) {
5036                 invalidate_inode_pages2(inode->i_mapping);
5037                 BTRFS_I(inode)->generation = 0;
5038                 cache->io_ctl.inode = NULL;
5039                 iput(inode);
5040         }
5041         ASSERT(cache->io_ctl.pages == NULL);
5042         btrfs_put_block_group(cache);
5043 }
5044
5045 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5046                              struct btrfs_fs_info *fs_info)
5047 {
5048         struct btrfs_block_group *cache;
5049
5050         spin_lock(&cur_trans->dirty_bgs_lock);
5051         while (!list_empty(&cur_trans->dirty_bgs)) {
5052                 cache = list_first_entry(&cur_trans->dirty_bgs,
5053                                          struct btrfs_block_group,
5054                                          dirty_list);
5055
5056                 if (!list_empty(&cache->io_list)) {
5057                         spin_unlock(&cur_trans->dirty_bgs_lock);
5058                         list_del_init(&cache->io_list);
5059                         btrfs_cleanup_bg_io(cache);
5060                         spin_lock(&cur_trans->dirty_bgs_lock);
5061                 }
5062
5063                 list_del_init(&cache->dirty_list);
5064                 spin_lock(&cache->lock);
5065                 cache->disk_cache_state = BTRFS_DC_ERROR;
5066                 spin_unlock(&cache->lock);
5067
5068                 spin_unlock(&cur_trans->dirty_bgs_lock);
5069                 btrfs_put_block_group(cache);
5070                 btrfs_delayed_refs_rsv_release(fs_info, 1);
5071                 spin_lock(&cur_trans->dirty_bgs_lock);
5072         }
5073         spin_unlock(&cur_trans->dirty_bgs_lock);
5074
5075         /*
5076          * Refer to the definition of io_bgs member for details why it's safe
5077          * to use it without any locking
5078          */
5079         while (!list_empty(&cur_trans->io_bgs)) {
5080                 cache = list_first_entry(&cur_trans->io_bgs,
5081                                          struct btrfs_block_group,
5082                                          io_list);
5083
5084                 list_del_init(&cache->io_list);
5085                 spin_lock(&cache->lock);
5086                 cache->disk_cache_state = BTRFS_DC_ERROR;
5087                 spin_unlock(&cache->lock);
5088                 btrfs_cleanup_bg_io(cache);
5089         }
5090 }
5091
5092 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5093                                    struct btrfs_fs_info *fs_info)
5094 {
5095         struct btrfs_device *dev, *tmp;
5096
5097         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5098         ASSERT(list_empty(&cur_trans->dirty_bgs));
5099         ASSERT(list_empty(&cur_trans->io_bgs));
5100
5101         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5102                                  post_commit_list) {
5103                 list_del_init(&dev->post_commit_list);
5104         }
5105
5106         btrfs_destroy_delayed_refs(cur_trans, fs_info);
5107
5108         cur_trans->state = TRANS_STATE_COMMIT_START;
5109         wake_up(&fs_info->transaction_blocked_wait);
5110
5111         cur_trans->state = TRANS_STATE_UNBLOCKED;
5112         wake_up(&fs_info->transaction_wait);
5113
5114         btrfs_destroy_delayed_inodes(fs_info);
5115
5116         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5117                                      EXTENT_DIRTY);
5118         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5119
5120         btrfs_free_redirty_list(cur_trans);
5121
5122         cur_trans->state =TRANS_STATE_COMPLETED;
5123         wake_up(&cur_trans->commit_wait);
5124 }
5125
5126 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5127 {
5128         struct btrfs_transaction *t;
5129
5130         mutex_lock(&fs_info->transaction_kthread_mutex);
5131
5132         spin_lock(&fs_info->trans_lock);
5133         while (!list_empty(&fs_info->trans_list)) {
5134                 t = list_first_entry(&fs_info->trans_list,
5135                                      struct btrfs_transaction, list);
5136                 if (t->state >= TRANS_STATE_COMMIT_START) {
5137                         refcount_inc(&t->use_count);
5138                         spin_unlock(&fs_info->trans_lock);
5139                         btrfs_wait_for_commit(fs_info, t->transid);
5140                         btrfs_put_transaction(t);
5141                         spin_lock(&fs_info->trans_lock);
5142                         continue;
5143                 }
5144                 if (t == fs_info->running_transaction) {
5145                         t->state = TRANS_STATE_COMMIT_DOING;
5146                         spin_unlock(&fs_info->trans_lock);
5147                         /*
5148                          * We wait for 0 num_writers since we don't hold a trans
5149                          * handle open currently for this transaction.
5150                          */
5151                         wait_event(t->writer_wait,
5152                                    atomic_read(&t->num_writers) == 0);
5153                 } else {
5154                         spin_unlock(&fs_info->trans_lock);
5155                 }
5156                 btrfs_cleanup_one_transaction(t, fs_info);
5157
5158                 spin_lock(&fs_info->trans_lock);
5159                 if (t == fs_info->running_transaction)
5160                         fs_info->running_transaction = NULL;
5161                 list_del_init(&t->list);
5162                 spin_unlock(&fs_info->trans_lock);
5163
5164                 btrfs_put_transaction(t);
5165                 trace_btrfs_transaction_commit(fs_info);
5166                 spin_lock(&fs_info->trans_lock);
5167         }
5168         spin_unlock(&fs_info->trans_lock);
5169         btrfs_destroy_all_ordered_extents(fs_info);
5170         btrfs_destroy_delayed_inodes(fs_info);
5171         btrfs_assert_delayed_root_empty(fs_info);
5172         btrfs_destroy_all_delalloc_inodes(fs_info);
5173         btrfs_drop_all_logs(fs_info);
5174         mutex_unlock(&fs_info->transaction_kthread_mutex);
5175
5176         return 0;
5177 }
5178
5179 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5180 {
5181         struct btrfs_path *path;
5182         int ret;
5183         struct extent_buffer *l;
5184         struct btrfs_key search_key;
5185         struct btrfs_key found_key;
5186         int slot;
5187
5188         path = btrfs_alloc_path();
5189         if (!path)
5190                 return -ENOMEM;
5191
5192         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5193         search_key.type = -1;
5194         search_key.offset = (u64)-1;
5195         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5196         if (ret < 0)
5197                 goto error;
5198         BUG_ON(ret == 0); /* Corruption */
5199         if (path->slots[0] > 0) {
5200                 slot = path->slots[0] - 1;
5201                 l = path->nodes[0];
5202                 btrfs_item_key_to_cpu(l, &found_key, slot);
5203                 root->free_objectid = max_t(u64, found_key.objectid + 1,
5204                                             BTRFS_FIRST_FREE_OBJECTID);
5205         } else {
5206                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5207         }
5208         ret = 0;
5209 error:
5210         btrfs_free_path(path);
5211         return ret;
5212 }
5213
5214 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5215 {
5216         int ret;
5217         mutex_lock(&root->objectid_mutex);
5218
5219         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5220                 btrfs_warn(root->fs_info,
5221                            "the objectid of root %llu reaches its highest value",
5222                            root->root_key.objectid);
5223                 ret = -ENOSPC;
5224                 goto out;
5225         }
5226
5227         *objectid = root->free_objectid++;
5228         ret = 0;
5229 out:
5230         mutex_unlock(&root->objectid_mutex);
5231         return ret;
5232 }