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 "bio.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 #include "fs.h"
  47 #include "accessors.h"
  48 #include "extent-tree.h"
  49 #include "root-tree.h"
  50 #include "defrag.h"
  51 #include "uuid-tree.h"
  52 #include "relocation.h"
  53 #include "scrub.h"
  54 #include "super.h"
  55
  56 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  57                                  BTRFS_HEADER_FLAG_RELOC |\
  58                                  BTRFS_SUPER_FLAG_ERROR |\
  59                                  BTRFS_SUPER_FLAG_SEEDING |\
  60                                  BTRFS_SUPER_FLAG_METADUMP |\
  61                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  62
  63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  65                                       struct btrfs_fs_info *fs_info);
  66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  67 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  68                                         struct extent_io_tree *dirty_pages,
  69                                         int mark);
  70 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  71                                        struct extent_io_tree *pinned_extents);
  72 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  73 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  74
  75 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  76 {
  77         if (fs_info->csum_shash)
  78                 crypto_free_shash(fs_info->csum_shash);
  79 }
  80
  81 /*
  82  * async submit bios are used to offload expensive checksumming
  83  * onto the worker threads.  They checksum file and metadata bios
  84  * just before they are sent down the IO stack.
  85  */
  86 struct async_submit_bio {
  87         struct btrfs_inode *inode;
  88         struct bio *bio;
  89         enum btrfs_wq_submit_cmd submit_cmd;
  90         int mirror_num;
  91
  92         /* Optional parameter for used by direct io */
  93         u64 dio_file_offset;
  94         struct btrfs_work work;
  95         blk_status_t status;
  96 };
  97
  98 /*
  99  * Compute the csum of a btree block and store the result to provided buffer.
 100  */
 101 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
 102 {
 103         struct btrfs_fs_info *fs_info = buf->fs_info;
 104         const int num_pages = num_extent_pages(buf);
 105         const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
 106         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 107         char *kaddr;
 108         int i;
 109
 110         shash->tfm = fs_info->csum_shash;
 111         crypto_shash_init(shash);
 112         kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
 113         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
 114                             first_page_part - BTRFS_CSUM_SIZE);
 115
 116         for (i = 1; i < num_pages; i++) {
 117                 kaddr = page_address(buf->pages[i]);
 118                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 119         }
 120         memset(result, 0, BTRFS_CSUM_SIZE);
 121         crypto_shash_final(shash, result);
 122 }
 123
 124 /*
 125  * we can't consider a given block up to date unless the transid of the
 126  * block matches the transid in the parent node's pointer.  This is how we
 127  * detect blocks that either didn't get written at all or got written
 128  * in the wrong place.
 129  */
 130 static int verify_parent_transid(struct extent_io_tree *io_tree,
 131                                  struct extent_buffer *eb, u64 parent_transid,
 132                                  int atomic)
 133 {
 134         struct extent_state *cached_state = NULL;
 135         int ret;
 136
 137         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 138                 return 0;
 139
 140         if (atomic)
 141                 return -EAGAIN;
 142
 143         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
 144         if (extent_buffer_uptodate(eb) &&
 145             btrfs_header_generation(eb) == parent_transid) {
 146                 ret = 0;
 147                 goto out;
 148         }
 149         btrfs_err_rl(eb->fs_info,
 150 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 151                         eb->start, eb->read_mirror,
 152                         parent_transid, btrfs_header_generation(eb));
 153         ret = 1;
 154         clear_extent_buffer_uptodate(eb);
 155 out:
 156         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 157                       &cached_state);
 158         return ret;
 159 }
 160
 161 static bool btrfs_supported_super_csum(u16 csum_type)
 162 {
 163         switch (csum_type) {
 164         case BTRFS_CSUM_TYPE_CRC32:
 165         case BTRFS_CSUM_TYPE_XXHASH:
 166         case BTRFS_CSUM_TYPE_SHA256:
 167         case BTRFS_CSUM_TYPE_BLAKE2:
 168                 return true;
 169         default:
 170                 return false;
 171         }
 172 }
 173
 174 /*
 175  * Return 0 if the superblock checksum type matches the checksum value of that
 176  * algorithm. Pass the raw disk superblock data.
 177  */
 178 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 179                            const struct btrfs_super_block *disk_sb)
 180 {
 181         char result[BTRFS_CSUM_SIZE];
 182         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 183
 184         shash->tfm = fs_info->csum_shash;
 185
 186         /*
 187          * The super_block structure does not span the whole
 188          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 189          * filled with zeros and is included in the checksum.
 190          */
 191         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 192                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 193
 194         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 195                 return 1;
 196
 197         return 0;
 198 }
 199
 200 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 201                            struct btrfs_key *first_key, u64 parent_transid)
 202 {
 203         struct btrfs_fs_info *fs_info = eb->fs_info;
 204         int found_level;
 205         struct btrfs_key found_key;
 206         int ret;
 207
 208         found_level = btrfs_header_level(eb);
 209         if (found_level != level) {
 210                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 211                      KERN_ERR "BTRFS: tree level check failed\n");
 212                 btrfs_err(fs_info,
 213 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 214                           eb->start, level, found_level);
 215                 return -EIO;
 216         }
 217
 218         if (!first_key)
 219                 return 0;
 220
 221         /*
 222          * For live tree block (new tree blocks in current transaction),
 223          * we need proper lock context to avoid race, which is impossible here.
 224          * So we only checks tree blocks which is read from disk, whose
 225          * generation <= fs_info->last_trans_committed.
 226          */
 227         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 228                 return 0;
 229
 230         /* We have @first_key, so this @eb must have at least one item */
 231         if (btrfs_header_nritems(eb) == 0) {
 232                 btrfs_err(fs_info,
 233                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 234                           eb->start);
 235                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 236                 return -EUCLEAN;
 237         }
 238
 239         if (found_level)
 240                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 241         else
 242                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 243         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 244
 245         if (ret) {
 246                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 247                      KERN_ERR "BTRFS: tree first key check failed\n");
 248                 btrfs_err(fs_info,
 249 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 250                           eb->start, parent_transid, first_key->objectid,
 251                           first_key->type, first_key->offset,
 252                           found_key.objectid, found_key.type,
 253                           found_key.offset);
 254         }
 255         return ret;
 256 }
 257
 258 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 259                                       int mirror_num)
 260 {
 261         struct btrfs_fs_info *fs_info = eb->fs_info;
 262         u64 start = eb->start;
 263         int i, num_pages = num_extent_pages(eb);
 264         int ret = 0;
 265
 266         if (sb_rdonly(fs_info->sb))
 267                 return -EROFS;
 268
 269         for (i = 0; i < num_pages; i++) {
 270                 struct page *p = eb->pages[i];
 271
 272                 ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE,
 273                                 start, p, start - page_offset(p), mirror_num);
 274                 if (ret)
 275                         break;
 276                 start += PAGE_SIZE;
 277         }
 278
 279         return ret;
 280 }
 281
 282 /*
 283  * helper to read a given tree block, doing retries as required when
 284  * the checksums don't match and we have alternate mirrors to try.
 285  *
 286  * @check:              expected tree parentness check, see the comments of the
 287  *                      structure for details.
 288  */
 289 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 290                              struct btrfs_tree_parent_check *check)
 291 {
 292         struct btrfs_fs_info *fs_info = eb->fs_info;
 293         int failed = 0;
 294         int ret;
 295         int num_copies = 0;
 296         int mirror_num = 0;
 297         int failed_mirror = 0;
 298
 299         ASSERT(check);
 300
 301         while (1) {
 302                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 303                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
 304                 if (!ret)
 305                         break;
 306
 307                 num_copies = btrfs_num_copies(fs_info,
 308                                               eb->start, eb->len);
 309                 if (num_copies == 1)
 310                         break;
 311
 312                 if (!failed_mirror) {
 313                         failed = 1;
 314                         failed_mirror = eb->read_mirror;
 315                 }
 316
 317                 mirror_num++;
 318                 if (mirror_num == failed_mirror)
 319                         mirror_num++;
 320
 321                 if (mirror_num > num_copies)
 322                         break;
 323         }
 324
 325         if (failed && !ret && failed_mirror)
 326                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 327
 328         return ret;
 329 }
 330
 331 static int csum_one_extent_buffer(struct extent_buffer *eb)
 332 {
 333         struct btrfs_fs_info *fs_info = eb->fs_info;
 334         u8 result[BTRFS_CSUM_SIZE];
 335         int ret;
 336
 337         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 338                                     offsetof(struct btrfs_header, fsid),
 339                                     BTRFS_FSID_SIZE) == 0);
 340         csum_tree_block(eb, result);
 341
 342         if (btrfs_header_level(eb))
 343                 ret = btrfs_check_node(eb);
 344         else
 345                 ret = btrfs_check_leaf_full(eb);
 346
 347         if (ret < 0)
 348                 goto error;
 349
 350         /*
 351          * Also check the generation, the eb reached here must be newer than
 352          * last committed. Or something seriously wrong happened.
 353          */
 354         if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
 355                 ret = -EUCLEAN;
 356                 btrfs_err(fs_info,
 357                         "block=%llu bad generation, have %llu expect > %llu",
 358                           eb->start, btrfs_header_generation(eb),
 359                           fs_info->last_trans_committed);
 360                 goto error;
 361         }
 362         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 363
 364         return 0;
 365
 366 error:
 367         btrfs_print_tree(eb, 0);
 368         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 369                   eb->start);
 370         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 371         return ret;
 372 }
 373
 374 /* Checksum all dirty extent buffers in one bio_vec */
 375 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
 376                                       struct bio_vec *bvec)
 377 {
 378         struct page *page = bvec->bv_page;
 379         u64 bvec_start = page_offset(page) + bvec->bv_offset;
 380         u64 cur;
 381         int ret = 0;
 382
 383         for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
 384              cur += fs_info->nodesize) {
 385                 struct extent_buffer *eb;
 386                 bool uptodate;
 387
 388                 eb = find_extent_buffer(fs_info, cur);
 389                 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
 390                                                        fs_info->nodesize);
 391
 392                 /* A dirty eb shouldn't disappear from buffer_radix */
 393                 if (WARN_ON(!eb))
 394                         return -EUCLEAN;
 395
 396                 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
 397                         free_extent_buffer(eb);
 398                         return -EUCLEAN;
 399                 }
 400                 if (WARN_ON(!uptodate)) {
 401                         free_extent_buffer(eb);
 402                         return -EUCLEAN;
 403                 }
 404
 405                 ret = csum_one_extent_buffer(eb);
 406                 free_extent_buffer(eb);
 407                 if (ret < 0)
 408                         return ret;
 409         }
 410         return ret;
 411 }
 412
 413 /*
 414  * Checksum a dirty tree block before IO.  This has extra checks to make sure
 415  * we only fill in the checksum field in the first page of a multi-page block.
 416  * For subpage extent buffers we need bvec to also read the offset in the page.
 417  */
 418 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
 419 {
 420         struct page *page = bvec->bv_page;
 421         u64 start = page_offset(page);
 422         u64 found_start;
 423         struct extent_buffer *eb;
 424
 425         if (fs_info->nodesize < PAGE_SIZE)
 426                 return csum_dirty_subpage_buffers(fs_info, bvec);
 427
 428         eb = (struct extent_buffer *)page->private;
 429         if (page != eb->pages[0])
 430                 return 0;
 431
 432         found_start = btrfs_header_bytenr(eb);
 433
 434         if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
 435                 WARN_ON(found_start != 0);
 436                 return 0;
 437         }
 438
 439         /*
 440          * Please do not consolidate these warnings into a single if.
 441          * It is useful to know what went wrong.
 442          */
 443         if (WARN_ON(found_start != start))
 444                 return -EUCLEAN;
 445         if (WARN_ON(!PageUptodate(page)))
 446                 return -EUCLEAN;
 447
 448         return csum_one_extent_buffer(eb);
 449 }
 450
 451 static int check_tree_block_fsid(struct extent_buffer *eb)
 452 {
 453         struct btrfs_fs_info *fs_info = eb->fs_info;
 454         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 455         u8 fsid[BTRFS_FSID_SIZE];
 456         u8 *metadata_uuid;
 457
 458         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 459                            BTRFS_FSID_SIZE);
 460         /*
 461          * Checking the incompat flag is only valid for the current fs. For
 462          * seed devices it's forbidden to have their uuid changed so reading
 463          * ->fsid in this case is fine
 464          */
 465         if (btrfs_fs_incompat(fs_info, METADATA_UUID))
 466                 metadata_uuid = fs_devices->metadata_uuid;
 467         else
 468                 metadata_uuid = fs_devices->fsid;
 469
 470         if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
 471                 return 0;
 472
 473         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 474                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 475                         return 0;
 476
 477         return 1;
 478 }
 479
 480 /* Do basic extent buffer checks at read time */
 481 static int validate_extent_buffer(struct extent_buffer *eb,
 482                                   struct btrfs_tree_parent_check *check)
 483 {
 484         struct btrfs_fs_info *fs_info = eb->fs_info;
 485         u64 found_start;
 486         const u32 csum_size = fs_info->csum_size;
 487         u8 found_level;
 488         u8 result[BTRFS_CSUM_SIZE];
 489         const u8 *header_csum;
 490         int ret = 0;
 491
 492         ASSERT(check);
 493
 494         found_start = btrfs_header_bytenr(eb);
 495         if (found_start != eb->start) {
 496                 btrfs_err_rl(fs_info,
 497                         "bad tree block start, mirror %u want %llu have %llu",
 498                              eb->read_mirror, eb->start, found_start);
 499                 ret = -EIO;
 500                 goto out;
 501         }
 502         if (check_tree_block_fsid(eb)) {
 503                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 504                              eb->start, eb->read_mirror);
 505                 ret = -EIO;
 506                 goto out;
 507         }
 508         found_level = btrfs_header_level(eb);
 509         if (found_level >= BTRFS_MAX_LEVEL) {
 510                 btrfs_err(fs_info,
 511                         "bad tree block level, mirror %u level %d on logical %llu",
 512                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 513                 ret = -EIO;
 514                 goto out;
 515         }
 516
 517         csum_tree_block(eb, result);
 518         header_csum = page_address(eb->pages[0]) +
 519                 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
 520
 521         if (memcmp(result, header_csum, csum_size) != 0) {
 522                 btrfs_warn_rl(fs_info,
 523 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 524                               eb->start, eb->read_mirror,
 525                               CSUM_FMT_VALUE(csum_size, header_csum),
 526                               CSUM_FMT_VALUE(csum_size, result),
 527                               btrfs_header_level(eb));
 528                 ret = -EUCLEAN;
 529                 goto out;
 530         }
 531
 532         if (found_level != check->level) {
 533                 btrfs_err(fs_info,
 534                 "level verify failed on logical %llu mirror %u wanted %u found %u",
 535                           eb->start, eb->read_mirror, check->level, found_level);
 536                 ret = -EIO;
 537                 goto out;
 538         }
 539         if (unlikely(check->transid &&
 540                      btrfs_header_generation(eb) != check->transid)) {
 541                 btrfs_err_rl(eb->fs_info,
 542 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 543                                 eb->start, eb->read_mirror, check->transid,
 544                                 btrfs_header_generation(eb));
 545                 ret = -EIO;
 546                 goto out;
 547         }
 548         if (check->has_first_key) {
 549                 struct btrfs_key *expect_key = &check->first_key;
 550                 struct btrfs_key found_key;
 551
 552                 if (found_level)
 553                         btrfs_node_key_to_cpu(eb, &found_key, 0);
 554                 else
 555                         btrfs_item_key_to_cpu(eb, &found_key, 0);
 556                 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 557                         btrfs_err(fs_info,
 558 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 559                                   eb->start, check->transid,
 560                                   expect_key->objectid,
 561                                   expect_key->type, expect_key->offset,
 562                                   found_key.objectid, found_key.type,
 563                                   found_key.offset);
 564                         ret = -EUCLEAN;
 565                         goto out;
 566                 }
 567         }
 568         if (check->owner_root) {
 569                 ret = btrfs_check_eb_owner(eb, check->owner_root);
 570                 if (ret < 0)
 571                         goto out;
 572         }
 573
 574         /*
 575          * If this is a leaf block and it is corrupt, set the corrupt bit so
 576          * that we don't try and read the other copies of this block, just
 577          * return -EIO.
 578          */
 579         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 580                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 581                 ret = -EIO;
 582         }
 583
 584         if (found_level > 0 && btrfs_check_node(eb))
 585                 ret = -EIO;
 586
 587         if (!ret)
 588                 set_extent_buffer_uptodate(eb);
 589         else
 590                 btrfs_err(fs_info,
 591                 "read time tree block corruption detected on logical %llu mirror %u",
 592                           eb->start, eb->read_mirror);
 593 out:
 594         return ret;
 595 }
 596
 597 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
 598                                    int mirror, struct btrfs_tree_parent_check *check)
 599 {
 600         struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
 601         struct extent_buffer *eb;
 602         bool reads_done;
 603         int ret = 0;
 604
 605         ASSERT(check);
 606
 607         /*
 608          * We don't allow bio merge for subpage metadata read, so we should
 609          * only get one eb for each endio hook.
 610          */
 611         ASSERT(end == start + fs_info->nodesize - 1);
 612         ASSERT(PagePrivate(page));
 613
 614         eb = find_extent_buffer(fs_info, start);
 615         /*
 616          * When we are reading one tree block, eb must have been inserted into
 617          * the radix tree. If not, something is wrong.
 618          */
 619         ASSERT(eb);
 620
 621         reads_done = atomic_dec_and_test(&eb->io_pages);
 622         /* Subpage read must finish in page read */
 623         ASSERT(reads_done);
 624
 625         eb->read_mirror = mirror;
 626         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 627                 ret = -EIO;
 628                 goto err;
 629         }
 630         ret = validate_extent_buffer(eb, check);
 631         if (ret < 0)
 632                 goto err;
 633
 634         set_extent_buffer_uptodate(eb);
 635
 636         free_extent_buffer(eb);
 637         return ret;
 638 err:
 639         /*
 640          * end_bio_extent_readpage decrements io_pages in case of error,
 641          * make sure it has something to decrement.
 642          */
 643         atomic_inc(&eb->io_pages);
 644         clear_extent_buffer_uptodate(eb);
 645         free_extent_buffer(eb);
 646         return ret;
 647 }
 648
 649 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
 650                                    struct page *page, u64 start, u64 end,
 651                                    int mirror)
 652 {
 653         struct extent_buffer *eb;
 654         int ret = 0;
 655         int reads_done;
 656
 657         ASSERT(page->private);
 658
 659         if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
 660                 return validate_subpage_buffer(page, start, end, mirror,
 661                                                &bbio->parent_check);
 662
 663         eb = (struct extent_buffer *)page->private;
 664
 665         /*
 666          * The pending IO might have been the only thing that kept this buffer
 667          * in memory.  Make sure we have a ref for all this other checks
 668          */
 669         atomic_inc(&eb->refs);
 670
 671         reads_done = atomic_dec_and_test(&eb->io_pages);
 672         if (!reads_done)
 673                 goto err;
 674
 675         eb->read_mirror = mirror;
 676         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 677                 ret = -EIO;
 678                 goto err;
 679         }
 680         ret = validate_extent_buffer(eb, &bbio->parent_check);
 681 err:
 682         if (ret) {
 683                 /*
 684                  * our io error hook is going to dec the io pages
 685                  * again, we have to make sure it has something
 686                  * to decrement
 687                  */
 688                 atomic_inc(&eb->io_pages);
 689                 clear_extent_buffer_uptodate(eb);
 690         }
 691         free_extent_buffer(eb);
 692
 693         return ret;
 694 }
 695
 696 static void run_one_async_start(struct btrfs_work *work)
 697 {
 698         struct async_submit_bio *async;
 699         blk_status_t ret;
 700
 701         async = container_of(work, struct  async_submit_bio, work);
 702         switch (async->submit_cmd) {
 703         case WQ_SUBMIT_METADATA:
 704                 ret = btree_submit_bio_start(async->bio);
 705                 break;
 706         case WQ_SUBMIT_DATA:
 707                 ret = btrfs_submit_bio_start(async->inode, async->bio);
 708                 break;
 709         case WQ_SUBMIT_DATA_DIO:
 710                 ret = btrfs_submit_bio_start_direct_io(async->inode,
 711                                 async->bio, async->dio_file_offset);
 712                 break;
 713         }
 714         if (ret)
 715                 async->status = ret;
 716 }
 717
 718 /*
 719  * In order to insert checksums into the metadata in large chunks, we wait
 720  * until bio submission time.   All the pages in the bio are checksummed and
 721  * sums are attached onto the ordered extent record.
 722  *
 723  * At IO completion time the csums attached on the ordered extent record are
 724  * inserted into the tree.
 725  */
 726 static void run_one_async_done(struct btrfs_work *work)
 727 {
 728         struct async_submit_bio *async =
 729                 container_of(work, struct  async_submit_bio, work);
 730         struct btrfs_inode *inode = async->inode;
 731         struct btrfs_bio *bbio = btrfs_bio(async->bio);
 732
 733         /* If an error occurred we just want to clean up the bio and move on */
 734         if (async->status) {
 735                 btrfs_bio_end_io(bbio, async->status);
 736                 return;
 737         }
 738
 739         /*
 740          * All of the bios that pass through here are from async helpers.
 741          * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
 742          * This changes nothing when cgroups aren't in use.
 743          */
 744         async->bio->bi_opf |= REQ_CGROUP_PUNT;
 745         btrfs_submit_bio(inode->root->fs_info, async->bio, async->mirror_num);
 746 }
 747
 748 static void run_one_async_free(struct btrfs_work *work)
 749 {
 750         struct async_submit_bio *async;
 751
 752         async = container_of(work, struct  async_submit_bio, work);
 753         kfree(async);
 754 }
 755
 756 /*
 757  * Submit bio to an async queue.
 758  *
 759  * Retrun:
 760  * - true if the work has been succesfuly submitted
 761  * - false in case of error
 762  */
 763 bool btrfs_wq_submit_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num,
 764                          u64 dio_file_offset, enum btrfs_wq_submit_cmd cmd)
 765 {
 766         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 767         struct async_submit_bio *async;
 768
 769         async = kmalloc(sizeof(*async), GFP_NOFS);
 770         if (!async)
 771                 return false;
 772
 773         async->inode = inode;
 774         async->bio = bio;
 775         async->mirror_num = mirror_num;
 776         async->submit_cmd = cmd;
 777
 778         btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
 779                         run_one_async_free);
 780
 781         async->dio_file_offset = dio_file_offset;
 782
 783         async->status = 0;
 784
 785         if (op_is_sync(bio->bi_opf))
 786                 btrfs_queue_work(fs_info->hipri_workers, &async->work);
 787         else
 788                 btrfs_queue_work(fs_info->workers, &async->work);
 789         return true;
 790 }
 791
 792 static blk_status_t btree_csum_one_bio(struct bio *bio)
 793 {
 794         struct bio_vec *bvec;
 795         struct btrfs_root *root;
 796         int ret = 0;
 797         struct bvec_iter_all iter_all;
 798
 799         ASSERT(!bio_flagged(bio, BIO_CLONED));
 800         bio_for_each_segment_all(bvec, bio, iter_all) {
 801                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 802                 ret = csum_dirty_buffer(root->fs_info, bvec);
 803                 if (ret)
 804                         break;
 805         }
 806
 807         return errno_to_blk_status(ret);
 808 }
 809
 810 blk_status_t btree_submit_bio_start(struct bio *bio)
 811 {
 812         /*
 813          * when we're called for a write, we're already in the async
 814          * submission context.  Just jump into btrfs_submit_bio.
 815          */
 816         return btree_csum_one_bio(bio);
 817 }
 818
 819 static bool should_async_write(struct btrfs_fs_info *fs_info,
 820                              struct btrfs_inode *bi)
 821 {
 822         if (btrfs_is_zoned(fs_info))
 823                 return false;
 824         if (atomic_read(&bi->sync_writers))
 825                 return false;
 826         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 827                 return false;
 828         return true;
 829 }
 830
 831 void btrfs_submit_metadata_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num)
 832 {
 833         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 834         struct btrfs_bio *bbio = btrfs_bio(bio);
 835         blk_status_t ret;
 836
 837         bio->bi_opf |= REQ_META;
 838         bbio->is_metadata = 1;
 839
 840         if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
 841                 btrfs_submit_bio(fs_info, bio, mirror_num);
 842                 return;
 843         }
 844
 845         /*
 846          * Kthread helpers are used to submit writes so that checksumming can
 847          * happen in parallel across all CPUs.
 848          */
 849         if (should_async_write(fs_info, inode) &&
 850             btrfs_wq_submit_bio(inode, bio, mirror_num, 0, WQ_SUBMIT_METADATA))
 851                 return;
 852
 853         ret = btree_csum_one_bio(bio);
 854         if (ret) {
 855                 btrfs_bio_end_io(bbio, ret);
 856                 return;
 857         }
 858
 859         btrfs_submit_bio(fs_info, bio, mirror_num);
 860 }
 861
 862 #ifdef CONFIG_MIGRATION
 863 static int btree_migrate_folio(struct address_space *mapping,
 864                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 865 {
 866         /*
 867          * we can't safely write a btree page from here,
 868          * we haven't done the locking hook
 869          */
 870         if (folio_test_dirty(src))
 871                 return -EAGAIN;
 872         /*
 873          * Buffers may be managed in a filesystem specific way.
 874          * We must have no buffers or drop them.
 875          */
 876         if (folio_get_private(src) &&
 877             !filemap_release_folio(src, GFP_KERNEL))
 878                 return -EAGAIN;
 879         return migrate_folio(mapping, dst, src, mode);
 880 }
 881 #else
 882 #define btree_migrate_folio NULL
 883 #endif
 884
 885 static int btree_writepages(struct address_space *mapping,
 886                             struct writeback_control *wbc)
 887 {
 888         struct btrfs_fs_info *fs_info;
 889         int ret;
 890
 891         if (wbc->sync_mode == WB_SYNC_NONE) {
 892
 893                 if (wbc->for_kupdate)
 894                         return 0;
 895
 896                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 897                 /* this is a bit racy, but that's ok */
 898                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 899                                              BTRFS_DIRTY_METADATA_THRESH,
 900                                              fs_info->dirty_metadata_batch);
 901                 if (ret < 0)
 902                         return 0;
 903         }
 904         return btree_write_cache_pages(mapping, wbc);
 905 }
 906
 907 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 908 {
 909         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 910                 return false;
 911
 912         return try_release_extent_buffer(&folio->page);
 913 }
 914
 915 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 916                                  size_t length)
 917 {
 918         struct extent_io_tree *tree;
 919         tree = &BTRFS_I(folio->mapping->host)->io_tree;
 920         extent_invalidate_folio(tree, folio, offset);
 921         btree_release_folio(folio, GFP_NOFS);
 922         if (folio_get_private(folio)) {
 923                 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 924                            "folio private not zero on folio %llu",
 925                            (unsigned long long)folio_pos(folio));
 926                 folio_detach_private(folio);
 927         }
 928 }
 929
 930 #ifdef DEBUG
 931 static bool btree_dirty_folio(struct address_space *mapping,
 932                 struct folio *folio)
 933 {
 934         struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 935         struct btrfs_subpage *subpage;
 936         struct extent_buffer *eb;
 937         int cur_bit = 0;
 938         u64 page_start = folio_pos(folio);
 939
 940         if (fs_info->sectorsize == PAGE_SIZE) {
 941                 eb = folio_get_private(folio);
 942                 BUG_ON(!eb);
 943                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 944                 BUG_ON(!atomic_read(&eb->refs));
 945                 btrfs_assert_tree_write_locked(eb);
 946                 return filemap_dirty_folio(mapping, folio);
 947         }
 948         subpage = folio_get_private(folio);
 949
 950         ASSERT(subpage->dirty_bitmap);
 951         while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
 952                 unsigned long flags;
 953                 u64 cur;
 954                 u16 tmp = (1 << cur_bit);
 955
 956                 spin_lock_irqsave(&subpage->lock, flags);
 957                 if (!(tmp & subpage->dirty_bitmap)) {
 958                         spin_unlock_irqrestore(&subpage->lock, flags);
 959                         cur_bit++;
 960                         continue;
 961                 }
 962                 spin_unlock_irqrestore(&subpage->lock, flags);
 963                 cur = page_start + cur_bit * fs_info->sectorsize;
 964
 965                 eb = find_extent_buffer(fs_info, cur);
 966                 ASSERT(eb);
 967                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 968                 ASSERT(atomic_read(&eb->refs));
 969                 btrfs_assert_tree_write_locked(eb);
 970                 free_extent_buffer(eb);
 971
 972                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
 973         }
 974         return filemap_dirty_folio(mapping, folio);
 975 }
 976 #else
 977 #define btree_dirty_folio filemap_dirty_folio
 978 #endif
 979
 980 static const struct address_space_operations btree_aops = {
 981         .writepages     = btree_writepages,
 982         .release_folio  = btree_release_folio,
 983         .invalidate_folio = btree_invalidate_folio,
 984         .migrate_folio  = btree_migrate_folio,
 985         .dirty_folio    = btree_dirty_folio,
 986 };
 987
 988 struct extent_buffer *btrfs_find_create_tree_block(
 989                                                 struct btrfs_fs_info *fs_info,
 990                                                 u64 bytenr, u64 owner_root,
 991                                                 int level)
 992 {
 993         if (btrfs_is_testing(fs_info))
 994                 return alloc_test_extent_buffer(fs_info, bytenr);
 995         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 996 }
 997
 998 /*
 999  * Read tree block at logical address @bytenr and do variant basic but critical
1000  * verification.
1001  *
1002  * @check:              expected tree parentness check, see comments of the
1003  *                      structure for details.
1004  */
1005 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1006                                       struct btrfs_tree_parent_check *check)
1007 {
1008         struct extent_buffer *buf = NULL;
1009         int ret;
1010
1011         ASSERT(check);
1012
1013         buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
1014                                            check->level);
1015         if (IS_ERR(buf))
1016                 return buf;
1017
1018         ret = btrfs_read_extent_buffer(buf, check);
1019         if (ret) {
1020                 free_extent_buffer_stale(buf);
1021                 return ERR_PTR(ret);
1022         }
1023         if (btrfs_check_eb_owner(buf, check->owner_root)) {
1024                 free_extent_buffer_stale(buf);
1025                 return ERR_PTR(-EUCLEAN);
1026         }
1027         return buf;
1028
1029 }
1030
1031 void btrfs_clean_tree_block(struct extent_buffer *buf)
1032 {
1033         struct btrfs_fs_info *fs_info = buf->fs_info;
1034         if (btrfs_header_generation(buf) ==
1035             fs_info->running_transaction->transid) {
1036                 btrfs_assert_tree_write_locked(buf);
1037
1038                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1039                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1040                                                  -buf->len,
1041                                                  fs_info->dirty_metadata_batch);
1042                         clear_extent_buffer_dirty(buf);
1043                 }
1044         }
1045 }
1046
1047 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1048                          u64 objectid)
1049 {
1050         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1051
1052         memset(&root->root_key, 0, sizeof(root->root_key));
1053         memset(&root->root_item, 0, sizeof(root->root_item));
1054         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1055         root->fs_info = fs_info;
1056         root->root_key.objectid = objectid;
1057         root->node = NULL;
1058         root->commit_root = NULL;
1059         root->state = 0;
1060         RB_CLEAR_NODE(&root->rb_node);
1061
1062         root->last_trans = 0;
1063         root->free_objectid = 0;
1064         root->nr_delalloc_inodes = 0;
1065         root->nr_ordered_extents = 0;
1066         root->inode_tree = RB_ROOT;
1067         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1068
1069         btrfs_init_root_block_rsv(root);
1070
1071         INIT_LIST_HEAD(&root->dirty_list);
1072         INIT_LIST_HEAD(&root->root_list);
1073         INIT_LIST_HEAD(&root->delalloc_inodes);
1074         INIT_LIST_HEAD(&root->delalloc_root);
1075         INIT_LIST_HEAD(&root->ordered_extents);
1076         INIT_LIST_HEAD(&root->ordered_root);
1077         INIT_LIST_HEAD(&root->reloc_dirty_list);
1078         INIT_LIST_HEAD(&root->logged_list[0]);
1079         INIT_LIST_HEAD(&root->logged_list[1]);
1080         spin_lock_init(&root->inode_lock);
1081         spin_lock_init(&root->delalloc_lock);
1082         spin_lock_init(&root->ordered_extent_lock);
1083         spin_lock_init(&root->accounting_lock);
1084         spin_lock_init(&root->log_extents_lock[0]);
1085         spin_lock_init(&root->log_extents_lock[1]);
1086         spin_lock_init(&root->qgroup_meta_rsv_lock);
1087         mutex_init(&root->objectid_mutex);
1088         mutex_init(&root->log_mutex);
1089         mutex_init(&root->ordered_extent_mutex);
1090         mutex_init(&root->delalloc_mutex);
1091         init_waitqueue_head(&root->qgroup_flush_wait);
1092         init_waitqueue_head(&root->log_writer_wait);
1093         init_waitqueue_head(&root->log_commit_wait[0]);
1094         init_waitqueue_head(&root->log_commit_wait[1]);
1095         INIT_LIST_HEAD(&root->log_ctxs[0]);
1096         INIT_LIST_HEAD(&root->log_ctxs[1]);
1097         atomic_set(&root->log_commit[0], 0);
1098         atomic_set(&root->log_commit[1], 0);
1099         atomic_set(&root->log_writers, 0);
1100         atomic_set(&root->log_batch, 0);
1101         refcount_set(&root->refs, 1);
1102         atomic_set(&root->snapshot_force_cow, 0);
1103         atomic_set(&root->nr_swapfiles, 0);
1104         root->log_transid = 0;
1105         root->log_transid_committed = -1;
1106         root->last_log_commit = 0;
1107         root->anon_dev = 0;
1108         if (!dummy) {
1109                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1110                                     IO_TREE_ROOT_DIRTY_LOG_PAGES);
1111                 extent_io_tree_init(fs_info, &root->log_csum_range,
1112                                     IO_TREE_LOG_CSUM_RANGE);
1113         }
1114
1115         spin_lock_init(&root->root_item_lock);
1116         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1117 #ifdef CONFIG_BTRFS_DEBUG
1118         INIT_LIST_HEAD(&root->leak_list);
1119         spin_lock(&fs_info->fs_roots_radix_lock);
1120         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1121         spin_unlock(&fs_info->fs_roots_radix_lock);
1122 #endif
1123 }
1124
1125 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1126                                            u64 objectid, gfp_t flags)
1127 {
1128         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1129         if (root)
1130                 __setup_root(root, fs_info, objectid);
1131         return root;
1132 }
1133
1134 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1135 /* Should only be used by the testing infrastructure */
1136 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1137 {
1138         struct btrfs_root *root;
1139
1140         if (!fs_info)
1141                 return ERR_PTR(-EINVAL);
1142
1143         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1144         if (!root)
1145                 return ERR_PTR(-ENOMEM);
1146
1147         /* We don't use the stripesize in selftest, set it as sectorsize */
1148         root->alloc_bytenr = 0;
1149
1150         return root;
1151 }
1152 #endif
1153
1154 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1155 {
1156         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1157         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1158
1159         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1160 }
1161
1162 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1163 {
1164         const struct btrfs_key *key = k;
1165         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1166
1167         return btrfs_comp_cpu_keys(key, &root->root_key);
1168 }
1169
1170 int btrfs_global_root_insert(struct btrfs_root *root)
1171 {
1172         struct btrfs_fs_info *fs_info = root->fs_info;
1173         struct rb_node *tmp;
1174
1175         write_lock(&fs_info->global_root_lock);
1176         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1177         write_unlock(&fs_info->global_root_lock);
1178         ASSERT(!tmp);
1179
1180         return tmp ? -EEXIST : 0;
1181 }
1182
1183 void btrfs_global_root_delete(struct btrfs_root *root)
1184 {
1185         struct btrfs_fs_info *fs_info = root->fs_info;
1186
1187         write_lock(&fs_info->global_root_lock);
1188         rb_erase(&root->rb_node, &fs_info->global_root_tree);
1189         write_unlock(&fs_info->global_root_lock);
1190 }
1191
1192 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1193                                      struct btrfs_key *key)
1194 {
1195         struct rb_node *node;
1196         struct btrfs_root *root = NULL;
1197
1198         read_lock(&fs_info->global_root_lock);
1199         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1200         if (node)
1201                 root = container_of(node, struct btrfs_root, rb_node);
1202         read_unlock(&fs_info->global_root_lock);
1203
1204         return root;
1205 }
1206
1207 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1208 {
1209         struct btrfs_block_group *block_group;
1210         u64 ret;
1211
1212         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1213                 return 0;
1214
1215         if (bytenr)
1216                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1217         else
1218                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1219         ASSERT(block_group);
1220         if (!block_group)
1221                 return 0;
1222         ret = block_group->global_root_id;
1223         btrfs_put_block_group(block_group);
1224
1225         return ret;
1226 }
1227
1228 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1229 {
1230         struct btrfs_key key = {
1231                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1232                 .type = BTRFS_ROOT_ITEM_KEY,
1233                 .offset = btrfs_global_root_id(fs_info, bytenr),
1234         };
1235
1236         return btrfs_global_root(fs_info, &key);
1237 }
1238
1239 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1240 {
1241         struct btrfs_key key = {
1242                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1243                 .type = BTRFS_ROOT_ITEM_KEY,
1244                 .offset = btrfs_global_root_id(fs_info, bytenr),
1245         };
1246
1247         return btrfs_global_root(fs_info, &key);
1248 }
1249
1250 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
1251 {
1252         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
1253                 return fs_info->block_group_root;
1254         return btrfs_extent_root(fs_info, 0);
1255 }
1256
1257 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1258                                      u64 objectid)
1259 {
1260         struct btrfs_fs_info *fs_info = trans->fs_info;
1261         struct extent_buffer *leaf;
1262         struct btrfs_root *tree_root = fs_info->tree_root;
1263         struct btrfs_root *root;
1264         struct btrfs_key key;
1265         unsigned int nofs_flag;
1266         int ret = 0;
1267
1268         /*
1269          * We're holding a transaction handle, so use a NOFS memory allocation
1270          * context to avoid deadlock if reclaim happens.
1271          */
1272         nofs_flag = memalloc_nofs_save();
1273         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1274         memalloc_nofs_restore(nofs_flag);
1275         if (!root)
1276                 return ERR_PTR(-ENOMEM);
1277
1278         root->root_key.objectid = objectid;
1279         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1280         root->root_key.offset = 0;
1281
1282         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1283                                       BTRFS_NESTING_NORMAL);
1284         if (IS_ERR(leaf)) {
1285                 ret = PTR_ERR(leaf);
1286                 leaf = NULL;
1287                 goto fail;
1288         }
1289
1290         root->node = leaf;
1291         btrfs_mark_buffer_dirty(leaf);
1292
1293         root->commit_root = btrfs_root_node(root);
1294         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1295
1296         btrfs_set_root_flags(&root->root_item, 0);
1297         btrfs_set_root_limit(&root->root_item, 0);
1298         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1299         btrfs_set_root_generation(&root->root_item, trans->transid);
1300         btrfs_set_root_level(&root->root_item, 0);
1301         btrfs_set_root_refs(&root->root_item, 1);
1302         btrfs_set_root_used(&root->root_item, leaf->len);
1303         btrfs_set_root_last_snapshot(&root->root_item, 0);
1304         btrfs_set_root_dirid(&root->root_item, 0);
1305         if (is_fstree(objectid))
1306                 generate_random_guid(root->root_item.uuid);
1307         else
1308                 export_guid(root->root_item.uuid, &guid_null);
1309         btrfs_set_root_drop_level(&root->root_item, 0);
1310
1311         btrfs_tree_unlock(leaf);
1312
1313         key.objectid = objectid;
1314         key.type = BTRFS_ROOT_ITEM_KEY;
1315         key.offset = 0;
1316         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1317         if (ret)
1318                 goto fail;
1319
1320         return root;
1321
1322 fail:
1323         btrfs_put_root(root);
1324
1325         return ERR_PTR(ret);
1326 }
1327
1328 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1329                                          struct btrfs_fs_info *fs_info)
1330 {
1331         struct btrfs_root *root;
1332
1333         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1334         if (!root)
1335                 return ERR_PTR(-ENOMEM);
1336
1337         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1338         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1339         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1340
1341         return root;
1342 }
1343
1344 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1345                               struct btrfs_root *root)
1346 {
1347         struct extent_buffer *leaf;
1348
1349         /*
1350          * DON'T set SHAREABLE bit for log trees.
1351          *
1352          * Log trees are not exposed to user space thus can't be snapshotted,
1353          * and they go away before a real commit is actually done.
1354          *
1355          * They do store pointers to file data extents, and those reference
1356          * counts still get updated (along with back refs to the log tree).
1357          */
1358
1359         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1360                         NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1361         if (IS_ERR(leaf))
1362                 return PTR_ERR(leaf);
1363
1364         root->node = leaf;
1365
1366         btrfs_mark_buffer_dirty(root->node);
1367         btrfs_tree_unlock(root->node);
1368
1369         return 0;
1370 }
1371
1372 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1373                              struct btrfs_fs_info *fs_info)
1374 {
1375         struct btrfs_root *log_root;
1376
1377         log_root = alloc_log_tree(trans, fs_info);
1378         if (IS_ERR(log_root))
1379                 return PTR_ERR(log_root);
1380
1381         if (!btrfs_is_zoned(fs_info)) {
1382                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1383
1384                 if (ret) {
1385                         btrfs_put_root(log_root);
1386                         return ret;
1387                 }
1388         }
1389
1390         WARN_ON(fs_info->log_root_tree);
1391         fs_info->log_root_tree = log_root;
1392         return 0;
1393 }
1394
1395 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1396                        struct btrfs_root *root)
1397 {
1398         struct btrfs_fs_info *fs_info = root->fs_info;
1399         struct btrfs_root *log_root;
1400         struct btrfs_inode_item *inode_item;
1401         int ret;
1402
1403         log_root = alloc_log_tree(trans, fs_info);
1404         if (IS_ERR(log_root))
1405                 return PTR_ERR(log_root);
1406
1407         ret = btrfs_alloc_log_tree_node(trans, log_root);
1408         if (ret) {
1409                 btrfs_put_root(log_root);
1410                 return ret;
1411         }
1412
1413         log_root->last_trans = trans->transid;
1414         log_root->root_key.offset = root->root_key.objectid;
1415
1416         inode_item = &log_root->root_item.inode;
1417         btrfs_set_stack_inode_generation(inode_item, 1);
1418         btrfs_set_stack_inode_size(inode_item, 3);
1419         btrfs_set_stack_inode_nlink(inode_item, 1);
1420         btrfs_set_stack_inode_nbytes(inode_item,
1421                                      fs_info->nodesize);
1422         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1423
1424         btrfs_set_root_node(&log_root->root_item, log_root->node);
1425
1426         WARN_ON(root->log_root);
1427         root->log_root = log_root;
1428         root->log_transid = 0;
1429         root->log_transid_committed = -1;
1430         root->last_log_commit = 0;
1431         return 0;
1432 }
1433
1434 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1435                                               struct btrfs_path *path,
1436                                               struct btrfs_key *key)
1437 {
1438         struct btrfs_root *root;
1439         struct btrfs_tree_parent_check check = { 0 };
1440         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1441         u64 generation;
1442         int ret;
1443         int level;
1444
1445         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1446         if (!root)
1447                 return ERR_PTR(-ENOMEM);
1448
1449         ret = btrfs_find_root(tree_root, key, path,
1450                               &root->root_item, &root->root_key);
1451         if (ret) {
1452                 if (ret > 0)
1453                         ret = -ENOENT;
1454                 goto fail;
1455         }
1456
1457         generation = btrfs_root_generation(&root->root_item);
1458         level = btrfs_root_level(&root->root_item);
1459         check.level = level;
1460         check.transid = generation;
1461         check.owner_root = key->objectid;
1462         root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1463                                      &check);
1464         if (IS_ERR(root->node)) {
1465                 ret = PTR_ERR(root->node);
1466                 root->node = NULL;
1467                 goto fail;
1468         }
1469         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1470                 ret = -EIO;
1471                 goto fail;
1472         }
1473
1474         /*
1475          * For real fs, and not log/reloc trees, root owner must
1476          * match its root node owner
1477          */
1478         if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1479             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1480             root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1481             root->root_key.objectid != btrfs_header_owner(root->node)) {
1482                 btrfs_crit(fs_info,
1483 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1484                            root->root_key.objectid, root->node->start,
1485                            btrfs_header_owner(root->node),
1486                            root->root_key.objectid);
1487                 ret = -EUCLEAN;
1488                 goto fail;
1489         }
1490         root->commit_root = btrfs_root_node(root);
1491         return root;
1492 fail:
1493         btrfs_put_root(root);
1494         return ERR_PTR(ret);
1495 }
1496
1497 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1498                                         struct btrfs_key *key)
1499 {
1500         struct btrfs_root *root;
1501         struct btrfs_path *path;
1502
1503         path = btrfs_alloc_path();
1504         if (!path)
1505                 return ERR_PTR(-ENOMEM);
1506         root = read_tree_root_path(tree_root, path, key);
1507         btrfs_free_path(path);
1508
1509         return root;
1510 }
1511
1512 /*
1513  * Initialize subvolume root in-memory structure
1514  *
1515  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1516  */
1517 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1518 {
1519         int ret;
1520         unsigned int nofs_flag;
1521
1522         /*
1523          * We might be called under a transaction (e.g. indirect backref
1524          * resolution) which could deadlock if it triggers memory reclaim
1525          */
1526         nofs_flag = memalloc_nofs_save();
1527         ret = btrfs_drew_lock_init(&root->snapshot_lock);
1528         memalloc_nofs_restore(nofs_flag);
1529         if (ret)
1530                 goto fail;
1531
1532         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1533             !btrfs_is_data_reloc_root(root)) {
1534                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1535                 btrfs_check_and_init_root_item(&root->root_item);
1536         }
1537
1538         /*
1539          * Don't assign anonymous block device to roots that are not exposed to
1540          * userspace, the id pool is limited to 1M
1541          */
1542         if (is_fstree(root->root_key.objectid) &&
1543             btrfs_root_refs(&root->root_item) > 0) {
1544                 if (!anon_dev) {
1545                         ret = get_anon_bdev(&root->anon_dev);
1546                         if (ret)
1547                                 goto fail;
1548                 } else {
1549                         root->anon_dev = anon_dev;
1550                 }
1551         }
1552
1553         mutex_lock(&root->objectid_mutex);
1554         ret = btrfs_init_root_free_objectid(root);
1555         if (ret) {
1556                 mutex_unlock(&root->objectid_mutex);
1557                 goto fail;
1558         }
1559
1560         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1561
1562         mutex_unlock(&root->objectid_mutex);
1563
1564         return 0;
1565 fail:
1566         /* The caller is responsible to call btrfs_free_fs_root */
1567         return ret;
1568 }
1569
1570 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1571                                                u64 root_id)
1572 {
1573         struct btrfs_root *root;
1574
1575         spin_lock(&fs_info->fs_roots_radix_lock);
1576         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1577                                  (unsigned long)root_id);
1578         if (root)
1579                 root = btrfs_grab_root(root);
1580         spin_unlock(&fs_info->fs_roots_radix_lock);
1581         return root;
1582 }
1583
1584 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1585                                                 u64 objectid)
1586 {
1587         struct btrfs_key key = {
1588                 .objectid = objectid,
1589                 .type = BTRFS_ROOT_ITEM_KEY,
1590                 .offset = 0,
1591         };
1592
1593         if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1594                 return btrfs_grab_root(fs_info->tree_root);
1595         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1596                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1597         if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1598                 return btrfs_grab_root(fs_info->chunk_root);
1599         if (objectid == BTRFS_DEV_TREE_OBJECTID)
1600                 return btrfs_grab_root(fs_info->dev_root);
1601         if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1602                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1603         if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1604                 return btrfs_grab_root(fs_info->quota_root) ?
1605                         fs_info->quota_root : ERR_PTR(-ENOENT);
1606         if (objectid == BTRFS_UUID_TREE_OBJECTID)
1607                 return btrfs_grab_root(fs_info->uuid_root) ?
1608                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1609         if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1610                 return btrfs_grab_root(fs_info->block_group_root) ?
1611                         fs_info->block_group_root : ERR_PTR(-ENOENT);
1612         if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1613                 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1614
1615                 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1616         }
1617         return NULL;
1618 }
1619
1620 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1621                          struct btrfs_root *root)
1622 {
1623         int ret;
1624
1625         ret = radix_tree_preload(GFP_NOFS);
1626         if (ret)
1627                 return ret;
1628
1629         spin_lock(&fs_info->fs_roots_radix_lock);
1630         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1631                                 (unsigned long)root->root_key.objectid,
1632                                 root);
1633         if (ret == 0) {
1634                 btrfs_grab_root(root);
1635                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1636         }
1637         spin_unlock(&fs_info->fs_roots_radix_lock);
1638         radix_tree_preload_end();
1639
1640         return ret;
1641 }
1642
1643 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1644 {
1645 #ifdef CONFIG_BTRFS_DEBUG
1646         struct btrfs_root *root;
1647
1648         while (!list_empty(&fs_info->allocated_roots)) {
1649                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1650
1651                 root = list_first_entry(&fs_info->allocated_roots,
1652                                         struct btrfs_root, leak_list);
1653                 btrfs_err(fs_info, "leaked root %s refcount %d",
1654                           btrfs_root_name(&root->root_key, buf),
1655                           refcount_read(&root->refs));
1656                 while (refcount_read(&root->refs) > 1)
1657                         btrfs_put_root(root);
1658                 btrfs_put_root(root);
1659         }
1660 #endif
1661 }
1662
1663 static void free_global_roots(struct btrfs_fs_info *fs_info)
1664 {
1665         struct btrfs_root *root;
1666         struct rb_node *node;
1667
1668         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1669                 root = rb_entry(node, struct btrfs_root, rb_node);
1670                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1671                 btrfs_put_root(root);
1672         }
1673 }
1674
1675 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1676 {
1677         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1678         percpu_counter_destroy(&fs_info->delalloc_bytes);
1679         percpu_counter_destroy(&fs_info->ordered_bytes);
1680         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1681         btrfs_free_csum_hash(fs_info);
1682         btrfs_free_stripe_hash_table(fs_info);
1683         btrfs_free_ref_cache(fs_info);
1684         kfree(fs_info->balance_ctl);
1685         kfree(fs_info->delayed_root);
1686         free_global_roots(fs_info);
1687         btrfs_put_root(fs_info->tree_root);
1688         btrfs_put_root(fs_info->chunk_root);
1689         btrfs_put_root(fs_info->dev_root);
1690         btrfs_put_root(fs_info->quota_root);
1691         btrfs_put_root(fs_info->uuid_root);
1692         btrfs_put_root(fs_info->fs_root);
1693         btrfs_put_root(fs_info->data_reloc_root);
1694         btrfs_put_root(fs_info->block_group_root);
1695         btrfs_check_leaked_roots(fs_info);
1696         btrfs_extent_buffer_leak_debug_check(fs_info);
1697         kfree(fs_info->super_copy);
1698         kfree(fs_info->super_for_commit);
1699         kfree(fs_info->subpage_info);
1700         kvfree(fs_info);
1701 }
1702
1703
1704 /*
1705  * Get an in-memory reference of a root structure.
1706  *
1707  * For essential trees like root/extent tree, we grab it from fs_info directly.
1708  * For subvolume trees, we check the cached filesystem roots first. If not
1709  * found, then read it from disk and add it to cached fs roots.
1710  *
1711  * Caller should release the root by calling btrfs_put_root() after the usage.
1712  *
1713  * NOTE: Reloc and log trees can't be read by this function as they share the
1714  *       same root objectid.
1715  *
1716  * @objectid:   root id
1717  * @anon_dev:   preallocated anonymous block device number for new roots,
1718  *              pass 0 for new allocation.
1719  * @check_ref:  whether to check root item references, If true, return -ENOENT
1720  *              for orphan roots
1721  */
1722 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1723                                              u64 objectid, dev_t anon_dev,
1724                                              bool check_ref)
1725 {
1726         struct btrfs_root *root;
1727         struct btrfs_path *path;
1728         struct btrfs_key key;
1729         int ret;
1730
1731         root = btrfs_get_global_root(fs_info, objectid);
1732         if (root)
1733                 return root;
1734 again:
1735         root = btrfs_lookup_fs_root(fs_info, objectid);
1736         if (root) {
1737                 /* Shouldn't get preallocated anon_dev for cached roots */
1738                 ASSERT(!anon_dev);
1739                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1740                         btrfs_put_root(root);
1741                         return ERR_PTR(-ENOENT);
1742                 }
1743                 return root;
1744         }
1745
1746         key.objectid = objectid;
1747         key.type = BTRFS_ROOT_ITEM_KEY;
1748         key.offset = (u64)-1;
1749         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1750         if (IS_ERR(root))
1751                 return root;
1752
1753         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1754                 ret = -ENOENT;
1755                 goto fail;
1756         }
1757
1758         ret = btrfs_init_fs_root(root, anon_dev);
1759         if (ret)
1760                 goto fail;
1761
1762         path = btrfs_alloc_path();
1763         if (!path) {
1764                 ret = -ENOMEM;
1765                 goto fail;
1766         }
1767         key.objectid = BTRFS_ORPHAN_OBJECTID;
1768         key.type = BTRFS_ORPHAN_ITEM_KEY;
1769         key.offset = objectid;
1770
1771         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772         btrfs_free_path(path);
1773         if (ret < 0)
1774                 goto fail;
1775         if (ret == 0)
1776                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1777
1778         ret = btrfs_insert_fs_root(fs_info, root);
1779         if (ret) {
1780                 if (ret == -EEXIST) {
1781                         btrfs_put_root(root);
1782                         goto again;
1783                 }
1784                 goto fail;
1785         }
1786         return root;
1787 fail:
1788         /*
1789          * If our caller provided us an anonymous device, then it's his
1790          * responsibility to free it in case we fail. So we have to set our
1791          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1792          * and once again by our caller.
1793          */
1794         if (anon_dev)
1795                 root->anon_dev = 0;
1796         btrfs_put_root(root);
1797         return ERR_PTR(ret);
1798 }
1799
1800 /*
1801  * Get in-memory reference of a root structure
1802  *
1803  * @objectid:   tree objectid
1804  * @check_ref:  if set, verify that the tree exists and the item has at least
1805  *              one reference
1806  */
1807 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1808                                      u64 objectid, bool check_ref)
1809 {
1810         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1811 }
1812
1813 /*
1814  * Get in-memory reference of a root structure, created as new, optionally pass
1815  * the anonymous block device id
1816  *
1817  * @objectid:   tree objectid
1818  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1819  *              parameter value
1820  */
1821 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1822                                          u64 objectid, dev_t anon_dev)
1823 {
1824         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1825 }
1826
1827 /*
1828  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1829  * @fs_info:    the fs_info
1830  * @objectid:   the objectid we need to lookup
1831  *
1832  * This is exclusively used for backref walking, and exists specifically because
1833  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1834  * creation time, which means we may have to read the tree_root in order to look
1835  * up a fs root that is not in memory.  If the root is not in memory we will
1836  * read the tree root commit root and look up the fs root from there.  This is a
1837  * temporary root, it will not be inserted into the radix tree as it doesn't
1838  * have the most uptodate information, it'll simply be discarded once the
1839  * backref code is finished using the root.
1840  */
1841 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1842                                                  struct btrfs_path *path,
1843                                                  u64 objectid)
1844 {
1845         struct btrfs_root *root;
1846         struct btrfs_key key;
1847
1848         ASSERT(path->search_commit_root && path->skip_locking);
1849
1850         /*
1851          * This can return -ENOENT if we ask for a root that doesn't exist, but
1852          * since this is called via the backref walking code we won't be looking
1853          * up a root that doesn't exist, unless there's corruption.  So if root
1854          * != NULL just return it.
1855          */
1856         root = btrfs_get_global_root(fs_info, objectid);
1857         if (root)
1858                 return root;
1859
1860         root = btrfs_lookup_fs_root(fs_info, objectid);
1861         if (root)
1862                 return root;
1863
1864         key.objectid = objectid;
1865         key.type = BTRFS_ROOT_ITEM_KEY;
1866         key.offset = (u64)-1;
1867         root = read_tree_root_path(fs_info->tree_root, path, &key);
1868         btrfs_release_path(path);
1869
1870         return root;
1871 }
1872
1873 static int cleaner_kthread(void *arg)
1874 {
1875         struct btrfs_fs_info *fs_info = arg;
1876         int again;
1877
1878         while (1) {
1879                 again = 0;
1880
1881                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1882
1883                 /* Make the cleaner go to sleep early. */
1884                 if (btrfs_need_cleaner_sleep(fs_info))
1885                         goto sleep;
1886
1887                 /*
1888                  * Do not do anything if we might cause open_ctree() to block
1889                  * before we have finished mounting the filesystem.
1890                  */
1891                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1892                         goto sleep;
1893
1894                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1895                         goto sleep;
1896
1897                 /*
1898                  * Avoid the problem that we change the status of the fs
1899                  * during the above check and trylock.
1900                  */
1901                 if (btrfs_need_cleaner_sleep(fs_info)) {
1902                         mutex_unlock(&fs_info->cleaner_mutex);
1903                         goto sleep;
1904                 }
1905
1906                 btrfs_run_delayed_iputs(fs_info);
1907
1908                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1909                 mutex_unlock(&fs_info->cleaner_mutex);
1910
1911                 /*
1912                  * The defragger has dealt with the R/O remount and umount,
1913                  * needn't do anything special here.
1914                  */
1915                 btrfs_run_defrag_inodes(fs_info);
1916
1917                 /*
1918                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1919                  * with relocation (btrfs_relocate_chunk) and relocation
1920                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1921                  * after acquiring fs_info->reclaim_bgs_lock. So we
1922                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1923                  * unused block groups.
1924                  */
1925                 btrfs_delete_unused_bgs(fs_info);
1926
1927                 /*
1928                  * Reclaim block groups in the reclaim_bgs list after we deleted
1929                  * all unused block_groups. This possibly gives us some more free
1930                  * space.
1931                  */
1932                 btrfs_reclaim_bgs(fs_info);
1933 sleep:
1934                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1935                 if (kthread_should_park())
1936                         kthread_parkme();
1937                 if (kthread_should_stop())
1938                         return 0;
1939                 if (!again) {
1940                         set_current_state(TASK_INTERRUPTIBLE);
1941                         schedule();
1942                         __set_current_state(TASK_RUNNING);
1943                 }
1944         }
1945 }
1946
1947 static int transaction_kthread(void *arg)
1948 {
1949         struct btrfs_root *root = arg;
1950         struct btrfs_fs_info *fs_info = root->fs_info;
1951         struct btrfs_trans_handle *trans;
1952         struct btrfs_transaction *cur;
1953         u64 transid;
1954         time64_t delta;
1955         unsigned long delay;
1956         bool cannot_commit;
1957
1958         do {
1959                 cannot_commit = false;
1960                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1961                 mutex_lock(&fs_info->transaction_kthread_mutex);
1962
1963                 spin_lock(&fs_info->trans_lock);
1964                 cur = fs_info->running_transaction;
1965                 if (!cur) {
1966                         spin_unlock(&fs_info->trans_lock);
1967                         goto sleep;
1968                 }
1969
1970                 delta = ktime_get_seconds() - cur->start_time;
1971                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1972                     cur->state < TRANS_STATE_COMMIT_START &&
1973                     delta < fs_info->commit_interval) {
1974                         spin_unlock(&fs_info->trans_lock);
1975                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1976                         delay = min(delay,
1977                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1978                         goto sleep;
1979                 }
1980                 transid = cur->transid;
1981                 spin_unlock(&fs_info->trans_lock);
1982
1983                 /* If the file system is aborted, this will always fail. */
1984                 trans = btrfs_attach_transaction(root);
1985                 if (IS_ERR(trans)) {
1986                         if (PTR_ERR(trans) != -ENOENT)
1987                                 cannot_commit = true;
1988                         goto sleep;
1989                 }
1990                 if (transid == trans->transid) {
1991                         btrfs_commit_transaction(trans);
1992                 } else {
1993                         btrfs_end_transaction(trans);
1994                 }
1995 sleep:
1996                 wake_up_process(fs_info->cleaner_kthread);
1997                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1998
1999                 if (BTRFS_FS_ERROR(fs_info))
2000                         btrfs_cleanup_transaction(fs_info);
2001                 if (!kthread_should_stop() &&
2002                                 (!btrfs_transaction_blocked(fs_info) ||
2003                                  cannot_commit))
2004                         schedule_timeout_interruptible(delay);
2005         } while (!kthread_should_stop());
2006         return 0;
2007 }
2008
2009 /*
2010  * This will find the highest generation in the array of root backups.  The
2011  * index of the highest array is returned, or -EINVAL if we can't find
2012  * anything.
2013  *
2014  * We check to make sure the array is valid by comparing the
2015  * generation of the latest  root in the array with the generation
2016  * in the super block.  If they don't match we pitch it.
2017  */
2018 static int find_newest_super_backup(struct btrfs_fs_info *info)
2019 {
2020         const u64 newest_gen = btrfs_super_generation(info->super_copy);
2021         u64 cur;
2022         struct btrfs_root_backup *root_backup;
2023         int i;
2024
2025         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2026                 root_backup = info->super_copy->super_roots + i;
2027                 cur = btrfs_backup_tree_root_gen(root_backup);
2028                 if (cur == newest_gen)
2029                         return i;
2030         }
2031
2032         return -EINVAL;
2033 }
2034
2035 /*
2036  * copy all the root pointers into the super backup array.
2037  * this will bump the backup pointer by one when it is
2038  * done
2039  */
2040 static void backup_super_roots(struct btrfs_fs_info *info)
2041 {
2042         const int next_backup = info->backup_root_index;
2043         struct btrfs_root_backup *root_backup;
2044
2045         root_backup = info->super_for_commit->super_roots + next_backup;
2046
2047         /*
2048          * make sure all of our padding and empty slots get zero filled
2049          * regardless of which ones we use today
2050          */
2051         memset(root_backup, 0, sizeof(*root_backup));
2052
2053         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2054
2055         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2056         btrfs_set_backup_tree_root_gen(root_backup,
2057                                btrfs_header_generation(info->tree_root->node));
2058
2059         btrfs_set_backup_tree_root_level(root_backup,
2060                                btrfs_header_level(info->tree_root->node));
2061
2062         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2063         btrfs_set_backup_chunk_root_gen(root_backup,
2064                                btrfs_header_generation(info->chunk_root->node));
2065         btrfs_set_backup_chunk_root_level(root_backup,
2066                                btrfs_header_level(info->chunk_root->node));
2067
2068         if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
2069                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2070                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2071
2072                 btrfs_set_backup_extent_root(root_backup,
2073                                              extent_root->node->start);
2074                 btrfs_set_backup_extent_root_gen(root_backup,
2075                                 btrfs_header_generation(extent_root->node));
2076                 btrfs_set_backup_extent_root_level(root_backup,
2077                                         btrfs_header_level(extent_root->node));
2078
2079                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2080                 btrfs_set_backup_csum_root_gen(root_backup,
2081                                                btrfs_header_generation(csum_root->node));
2082                 btrfs_set_backup_csum_root_level(root_backup,
2083                                                  btrfs_header_level(csum_root->node));
2084         }
2085
2086         /*
2087          * we might commit during log recovery, which happens before we set
2088          * the fs_root.  Make sure it is valid before we fill it in.
2089          */
2090         if (info->fs_root && info->fs_root->node) {
2091                 btrfs_set_backup_fs_root(root_backup,
2092                                          info->fs_root->node->start);
2093                 btrfs_set_backup_fs_root_gen(root_backup,
2094                                btrfs_header_generation(info->fs_root->node));
2095                 btrfs_set_backup_fs_root_level(root_backup,
2096                                btrfs_header_level(info->fs_root->node));
2097         }
2098
2099         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2100         btrfs_set_backup_dev_root_gen(root_backup,
2101                                btrfs_header_generation(info->dev_root->node));
2102         btrfs_set_backup_dev_root_level(root_backup,
2103                                        btrfs_header_level(info->dev_root->node));
2104
2105         btrfs_set_backup_total_bytes(root_backup,
2106                              btrfs_super_total_bytes(info->super_copy));
2107         btrfs_set_backup_bytes_used(root_backup,
2108                              btrfs_super_bytes_used(info->super_copy));
2109         btrfs_set_backup_num_devices(root_backup,
2110                              btrfs_super_num_devices(info->super_copy));
2111
2112         /*
2113          * if we don't copy this out to the super_copy, it won't get remembered
2114          * for the next commit
2115          */
2116         memcpy(&info->super_copy->super_roots,
2117                &info->super_for_commit->super_roots,
2118                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2119 }
2120
2121 /*
2122  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2123  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2124  *
2125  * fs_info - filesystem whose backup roots need to be read
2126  * priority - priority of backup root required
2127  *
2128  * Returns backup root index on success and -EINVAL otherwise.
2129  */
2130 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2131 {
2132         int backup_index = find_newest_super_backup(fs_info);
2133         struct btrfs_super_block *super = fs_info->super_copy;
2134         struct btrfs_root_backup *root_backup;
2135
2136         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2137                 if (priority == 0)
2138                         return backup_index;
2139
2140                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2141                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2142         } else {
2143                 return -EINVAL;
2144         }
2145
2146         root_backup = super->super_roots + backup_index;
2147
2148         btrfs_set_super_generation(super,
2149                                    btrfs_backup_tree_root_gen(root_backup));
2150         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2151         btrfs_set_super_root_level(super,
2152                                    btrfs_backup_tree_root_level(root_backup));
2153         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2154
2155         /*
2156          * Fixme: the total bytes and num_devices need to match or we should
2157          * need a fsck
2158          */
2159         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2160         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2161
2162         return backup_index;
2163 }
2164
2165 /* helper to cleanup workers */
2166 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2167 {
2168         btrfs_destroy_workqueue(fs_info->fixup_workers);
2169         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2170         btrfs_destroy_workqueue(fs_info->hipri_workers);
2171         btrfs_destroy_workqueue(fs_info->workers);
2172         if (fs_info->endio_workers)
2173                 destroy_workqueue(fs_info->endio_workers);
2174         if (fs_info->rmw_workers)
2175                 destroy_workqueue(fs_info->rmw_workers);
2176         if (fs_info->compressed_write_workers)
2177                 destroy_workqueue(fs_info->compressed_write_workers);
2178         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2179         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2180         btrfs_destroy_workqueue(fs_info->delayed_workers);
2181         btrfs_destroy_workqueue(fs_info->caching_workers);
2182         btrfs_destroy_workqueue(fs_info->flush_workers);
2183         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2184         if (fs_info->discard_ctl.discard_workers)
2185                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2186         /*
2187          * Now that all other work queues are destroyed, we can safely destroy
2188          * the queues used for metadata I/O, since tasks from those other work
2189          * queues can do metadata I/O operations.
2190          */
2191         if (fs_info->endio_meta_workers)
2192                 destroy_workqueue(fs_info->endio_meta_workers);
2193 }
2194
2195 static void free_root_extent_buffers(struct btrfs_root *root)
2196 {
2197         if (root) {
2198                 free_extent_buffer(root->node);
2199                 free_extent_buffer(root->commit_root);
2200                 root->node = NULL;
2201                 root->commit_root = NULL;
2202         }
2203 }
2204
2205 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2206 {
2207         struct btrfs_root *root, *tmp;
2208
2209         rbtree_postorder_for_each_entry_safe(root, tmp,
2210                                              &fs_info->global_root_tree,
2211                                              rb_node)
2212                 free_root_extent_buffers(root);
2213 }
2214
2215 /* helper to cleanup tree roots */
2216 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2217 {
2218         free_root_extent_buffers(info->tree_root);
2219
2220         free_global_root_pointers(info);
2221         free_root_extent_buffers(info->dev_root);
2222         free_root_extent_buffers(info->quota_root);
2223         free_root_extent_buffers(info->uuid_root);
2224         free_root_extent_buffers(info->fs_root);
2225         free_root_extent_buffers(info->data_reloc_root);
2226         free_root_extent_buffers(info->block_group_root);
2227         if (free_chunk_root)
2228                 free_root_extent_buffers(info->chunk_root);
2229 }
2230
2231 void btrfs_put_root(struct btrfs_root *root)
2232 {
2233         if (!root)
2234                 return;
2235
2236         if (refcount_dec_and_test(&root->refs)) {
2237                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2238                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2239                 if (root->anon_dev)
2240                         free_anon_bdev(root->anon_dev);
2241                 btrfs_drew_lock_destroy(&root->snapshot_lock);
2242                 free_root_extent_buffers(root);
2243 #ifdef CONFIG_BTRFS_DEBUG
2244                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2245                 list_del_init(&root->leak_list);
2246                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2247 #endif
2248                 kfree(root);
2249         }
2250 }
2251
2252 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2253 {
2254         int ret;
2255         struct btrfs_root *gang[8];
2256         int i;
2257
2258         while (!list_empty(&fs_info->dead_roots)) {
2259                 gang[0] = list_entry(fs_info->dead_roots.next,
2260                                      struct btrfs_root, root_list);
2261                 list_del(&gang[0]->root_list);
2262
2263                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2264                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2265                 btrfs_put_root(gang[0]);
2266         }
2267
2268         while (1) {
2269                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2270                                              (void **)gang, 0,
2271                                              ARRAY_SIZE(gang));
2272                 if (!ret)
2273                         break;
2274                 for (i = 0; i < ret; i++)
2275                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2276         }
2277 }
2278
2279 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2280 {
2281         mutex_init(&fs_info->scrub_lock);
2282         atomic_set(&fs_info->scrubs_running, 0);
2283         atomic_set(&fs_info->scrub_pause_req, 0);
2284         atomic_set(&fs_info->scrubs_paused, 0);
2285         atomic_set(&fs_info->scrub_cancel_req, 0);
2286         init_waitqueue_head(&fs_info->scrub_pause_wait);
2287         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2288 }
2289
2290 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2291 {
2292         spin_lock_init(&fs_info->balance_lock);
2293         mutex_init(&fs_info->balance_mutex);
2294         atomic_set(&fs_info->balance_pause_req, 0);
2295         atomic_set(&fs_info->balance_cancel_req, 0);
2296         fs_info->balance_ctl = NULL;
2297         init_waitqueue_head(&fs_info->balance_wait_q);
2298         atomic_set(&fs_info->reloc_cancel_req, 0);
2299 }
2300
2301 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2302 {
2303         struct inode *inode = fs_info->btree_inode;
2304         unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2305                                               fs_info->tree_root);
2306
2307         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2308         set_nlink(inode, 1);
2309         /*
2310          * we set the i_size on the btree inode to the max possible int.
2311          * the real end of the address space is determined by all of
2312          * the devices in the system
2313          */
2314         inode->i_size = OFFSET_MAX;
2315         inode->i_mapping->a_ops = &btree_aops;
2316
2317         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2318         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2319                             IO_TREE_BTREE_INODE_IO);
2320         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2321
2322         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2323         BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2324         BTRFS_I(inode)->location.type = 0;
2325         BTRFS_I(inode)->location.offset = 0;
2326         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2327         __insert_inode_hash(inode, hash);
2328 }
2329
2330 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2331 {
2332         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2333         init_rwsem(&fs_info->dev_replace.rwsem);
2334         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2335 }
2336
2337 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2338 {
2339         spin_lock_init(&fs_info->qgroup_lock);
2340         mutex_init(&fs_info->qgroup_ioctl_lock);
2341         fs_info->qgroup_tree = RB_ROOT;
2342         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2343         fs_info->qgroup_seq = 1;
2344         fs_info->qgroup_ulist = NULL;
2345         fs_info->qgroup_rescan_running = false;
2346         fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2347         mutex_init(&fs_info->qgroup_rescan_lock);
2348 }
2349
2350 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2351 {
2352         u32 max_active = fs_info->thread_pool_size;
2353         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2354
2355         fs_info->workers =
2356                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2357         fs_info->hipri_workers =
2358                 btrfs_alloc_workqueue(fs_info, "worker-high",
2359                                       flags | WQ_HIGHPRI, max_active, 16);
2360
2361         fs_info->delalloc_workers =
2362                 btrfs_alloc_workqueue(fs_info, "delalloc",
2363                                       flags, max_active, 2);
2364
2365         fs_info->flush_workers =
2366                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2367                                       flags, max_active, 0);
2368
2369         fs_info->caching_workers =
2370                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2371
2372         fs_info->fixup_workers =
2373                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2374
2375         fs_info->endio_workers =
2376                 alloc_workqueue("btrfs-endio", flags, max_active);
2377         fs_info->endio_meta_workers =
2378                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2379         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2380         fs_info->endio_write_workers =
2381                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2382                                       max_active, 2);
2383         fs_info->compressed_write_workers =
2384                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2385         fs_info->endio_freespace_worker =
2386                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2387                                       max_active, 0);
2388         fs_info->delayed_workers =
2389                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2390                                       max_active, 0);
2391         fs_info->qgroup_rescan_workers =
2392                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2393         fs_info->discard_ctl.discard_workers =
2394                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2395
2396         if (!(fs_info->workers && fs_info->hipri_workers &&
2397               fs_info->delalloc_workers && fs_info->flush_workers &&
2398               fs_info->endio_workers && fs_info->endio_meta_workers &&
2399               fs_info->compressed_write_workers &&
2400               fs_info->endio_write_workers &&
2401               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2402               fs_info->caching_workers && fs_info->fixup_workers &&
2403               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2404               fs_info->discard_ctl.discard_workers)) {
2405                 return -ENOMEM;
2406         }
2407
2408         return 0;
2409 }
2410
2411 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2412 {
2413         struct crypto_shash *csum_shash;
2414         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2415
2416         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2417
2418         if (IS_ERR(csum_shash)) {
2419                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2420                           csum_driver);
2421                 return PTR_ERR(csum_shash);
2422         }
2423
2424         fs_info->csum_shash = csum_shash;
2425
2426         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2427                         btrfs_super_csum_name(csum_type),
2428                         crypto_shash_driver_name(csum_shash));
2429         return 0;
2430 }
2431
2432 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2433                             struct btrfs_fs_devices *fs_devices)
2434 {
2435         int ret;
2436         struct btrfs_tree_parent_check check = { 0 };
2437         struct btrfs_root *log_tree_root;
2438         struct btrfs_super_block *disk_super = fs_info->super_copy;
2439         u64 bytenr = btrfs_super_log_root(disk_super);
2440         int level = btrfs_super_log_root_level(disk_super);
2441
2442         if (fs_devices->rw_devices == 0) {
2443                 btrfs_warn(fs_info, "log replay required on RO media");
2444                 return -EIO;
2445         }
2446
2447         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2448                                          GFP_KERNEL);
2449         if (!log_tree_root)
2450                 return -ENOMEM;
2451
2452         check.level = level;
2453         check.transid = fs_info->generation + 1;
2454         check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2455         log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2456         if (IS_ERR(log_tree_root->node)) {
2457                 btrfs_warn(fs_info, "failed to read log tree");
2458                 ret = PTR_ERR(log_tree_root->node);
2459                 log_tree_root->node = NULL;
2460                 btrfs_put_root(log_tree_root);
2461                 return ret;
2462         }
2463         if (!extent_buffer_uptodate(log_tree_root->node)) {
2464                 btrfs_err(fs_info, "failed to read log tree");
2465                 btrfs_put_root(log_tree_root);
2466                 return -EIO;
2467         }
2468
2469         /* returns with log_tree_root freed on success */
2470         ret = btrfs_recover_log_trees(log_tree_root);
2471         if (ret) {
2472                 btrfs_handle_fs_error(fs_info, ret,
2473                                       "Failed to recover log tree");
2474                 btrfs_put_root(log_tree_root);
2475                 return ret;
2476         }
2477
2478         if (sb_rdonly(fs_info->sb)) {
2479                 ret = btrfs_commit_super(fs_info);
2480                 if (ret)
2481                         return ret;
2482         }
2483
2484         return 0;
2485 }
2486
2487 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2488                                       struct btrfs_path *path, u64 objectid,
2489                                       const char *name)
2490 {
2491         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2492         struct btrfs_root *root;
2493         u64 max_global_id = 0;
2494         int ret;
2495         struct btrfs_key key = {
2496                 .objectid = objectid,
2497                 .type = BTRFS_ROOT_ITEM_KEY,
2498                 .offset = 0,
2499         };
2500         bool found = false;
2501
2502         /* If we have IGNOREDATACSUMS skip loading these roots. */
2503         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2504             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2505                 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2506                 return 0;
2507         }
2508
2509         while (1) {
2510                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2511                 if (ret < 0)
2512                         break;
2513
2514                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2515                         ret = btrfs_next_leaf(tree_root, path);
2516                         if (ret) {
2517                                 if (ret > 0)
2518                                         ret = 0;
2519                                 break;
2520                         }
2521                 }
2522                 ret = 0;
2523
2524                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2525                 if (key.objectid != objectid)
2526                         break;
2527                 btrfs_release_path(path);
2528
2529                 /*
2530                  * Just worry about this for extent tree, it'll be the same for
2531                  * everybody.
2532                  */
2533                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2534                         max_global_id = max(max_global_id, key.offset);
2535
2536                 found = true;
2537                 root = read_tree_root_path(tree_root, path, &key);
2538                 if (IS_ERR(root)) {
2539                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2540                                 ret = PTR_ERR(root);
2541                         break;
2542                 }
2543                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2544                 ret = btrfs_global_root_insert(root);
2545                 if (ret) {
2546                         btrfs_put_root(root);
2547                         break;
2548                 }
2549                 key.offset++;
2550         }
2551         btrfs_release_path(path);
2552
2553         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2554                 fs_info->nr_global_roots = max_global_id + 1;
2555
2556         if (!found || ret) {
2557                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2558                         set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2559
2560                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2561                         ret = ret ? ret : -ENOENT;
2562                 else
2563                         ret = 0;
2564                 btrfs_err(fs_info, "failed to load root %s", name);
2565         }
2566         return ret;
2567 }
2568
2569 static int load_global_roots(struct btrfs_root *tree_root)
2570 {
2571         struct btrfs_path *path;
2572         int ret = 0;
2573
2574         path = btrfs_alloc_path();
2575         if (!path)
2576                 return -ENOMEM;
2577
2578         ret = load_global_roots_objectid(tree_root, path,
2579                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2580         if (ret)
2581                 goto out;
2582         ret = load_global_roots_objectid(tree_root, path,
2583                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2584         if (ret)
2585                 goto out;
2586         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2587                 goto out;
2588         ret = load_global_roots_objectid(tree_root, path,
2589                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2590                                          "free space");
2591 out:
2592         btrfs_free_path(path);
2593         return ret;
2594 }
2595
2596 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2597 {
2598         struct btrfs_root *tree_root = fs_info->tree_root;
2599         struct btrfs_root *root;
2600         struct btrfs_key location;
2601         int ret;
2602
2603         BUG_ON(!fs_info->tree_root);
2604
2605         ret = load_global_roots(tree_root);
2606         if (ret)
2607                 return ret;
2608
2609         location.type = BTRFS_ROOT_ITEM_KEY;
2610         location.offset = 0;
2611
2612         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2613                 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2614                 root = btrfs_read_tree_root(tree_root, &location);
2615                 if (IS_ERR(root)) {
2616                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2617                                 ret = PTR_ERR(root);
2618                                 goto out;
2619                         }
2620                 } else {
2621                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2622                         fs_info->block_group_root = root;
2623                 }
2624         }
2625
2626         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2627         root = btrfs_read_tree_root(tree_root, &location);
2628         if (IS_ERR(root)) {
2629                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2630                         ret = PTR_ERR(root);
2631                         goto out;
2632                 }
2633         } else {
2634                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2635                 fs_info->dev_root = root;
2636         }
2637         /* Initialize fs_info for all devices in any case */
2638         ret = btrfs_init_devices_late(fs_info);
2639         if (ret)
2640                 goto out;
2641
2642         /*
2643          * This tree can share blocks with some other fs tree during relocation
2644          * and we need a proper setup by btrfs_get_fs_root
2645          */
2646         root = btrfs_get_fs_root(tree_root->fs_info,
2647                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2648         if (IS_ERR(root)) {
2649                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2650                         ret = PTR_ERR(root);
2651                         goto out;
2652                 }
2653         } else {
2654                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2655                 fs_info->data_reloc_root = root;
2656         }
2657
2658         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2659         root = btrfs_read_tree_root(tree_root, &location);
2660         if (!IS_ERR(root)) {
2661                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2662                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2663                 fs_info->quota_root = root;
2664         }
2665
2666         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2667         root = btrfs_read_tree_root(tree_root, &location);
2668         if (IS_ERR(root)) {
2669                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2670                         ret = PTR_ERR(root);
2671                         if (ret != -ENOENT)
2672                                 goto out;
2673                 }
2674         } else {
2675                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2676                 fs_info->uuid_root = root;
2677         }
2678
2679         return 0;
2680 out:
2681         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2682                    location.objectid, ret);
2683         return ret;
2684 }
2685
2686 /*
2687  * Real super block validation
2688  * NOTE: super csum type and incompat features will not be checked here.
2689  *
2690  * @sb:         super block to check
2691  * @mirror_num: the super block number to check its bytenr:
2692  *              0       the primary (1st) sb
2693  *              1, 2    2nd and 3rd backup copy
2694  *             -1       skip bytenr check
2695  */
2696 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2697                          struct btrfs_super_block *sb, int mirror_num)
2698 {
2699         u64 nodesize = btrfs_super_nodesize(sb);
2700         u64 sectorsize = btrfs_super_sectorsize(sb);
2701         int ret = 0;
2702
2703         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2704                 btrfs_err(fs_info, "no valid FS found");
2705                 ret = -EINVAL;
2706         }
2707         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2708                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2709                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2710                 ret = -EINVAL;
2711         }
2712         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2713                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2714                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2715                 ret = -EINVAL;
2716         }
2717         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2718                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2719                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2720                 ret = -EINVAL;
2721         }
2722         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2723                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2724                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2725                 ret = -EINVAL;
2726         }
2727
2728         /*
2729          * Check sectorsize and nodesize first, other check will need it.
2730          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2731          */
2732         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2733             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2734                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2735                 ret = -EINVAL;
2736         }
2737
2738         /*
2739          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2740          *
2741          * We can support 16K sectorsize with 64K page size without problem,
2742          * but such sectorsize/pagesize combination doesn't make much sense.
2743          * 4K will be our future standard, PAGE_SIZE is supported from the very
2744          * beginning.
2745          */
2746         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2747                 btrfs_err(fs_info,
2748                         "sectorsize %llu not yet supported for page size %lu",
2749                         sectorsize, PAGE_SIZE);
2750                 ret = -EINVAL;
2751         }
2752
2753         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2754             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2755                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2756                 ret = -EINVAL;
2757         }
2758         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2759                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2760                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2761                 ret = -EINVAL;
2762         }
2763
2764         /* Root alignment check */
2765         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2766                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2767                            btrfs_super_root(sb));
2768                 ret = -EINVAL;
2769         }
2770         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2771                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2772                            btrfs_super_chunk_root(sb));
2773                 ret = -EINVAL;
2774         }
2775         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2776                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2777                            btrfs_super_log_root(sb));
2778                 ret = -EINVAL;
2779         }
2780
2781         if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2782                    BTRFS_FSID_SIZE)) {
2783                 btrfs_err(fs_info,
2784                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2785                         fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2786                 ret = -EINVAL;
2787         }
2788
2789         if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2790             memcmp(fs_info->fs_devices->metadata_uuid,
2791                    fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2792                 btrfs_err(fs_info,
2793 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2794                         fs_info->super_copy->metadata_uuid,
2795                         fs_info->fs_devices->metadata_uuid);
2796                 ret = -EINVAL;
2797         }
2798
2799         /*
2800          * Artificial requirement for block-group-tree to force newer features
2801          * (free-space-tree, no-holes) so the test matrix is smaller.
2802          */
2803         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2804             (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2805              !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2806                 btrfs_err(fs_info,
2807                 "block-group-tree feature requires fres-space-tree and no-holes");
2808                 ret = -EINVAL;
2809         }
2810
2811         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2812                    BTRFS_FSID_SIZE) != 0) {
2813                 btrfs_err(fs_info,
2814                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2815                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2816                 ret = -EINVAL;
2817         }
2818
2819         /*
2820          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2821          * done later
2822          */
2823         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2824                 btrfs_err(fs_info, "bytes_used is too small %llu",
2825                           btrfs_super_bytes_used(sb));
2826                 ret = -EINVAL;
2827         }
2828         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2829                 btrfs_err(fs_info, "invalid stripesize %u",
2830                           btrfs_super_stripesize(sb));
2831                 ret = -EINVAL;
2832         }
2833         if (btrfs_super_num_devices(sb) > (1UL << 31))
2834                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2835                            btrfs_super_num_devices(sb));
2836         if (btrfs_super_num_devices(sb) == 0) {
2837                 btrfs_err(fs_info, "number of devices is 0");
2838                 ret = -EINVAL;
2839         }
2840
2841         if (mirror_num >= 0 &&
2842             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2843                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2844                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2845                 ret = -EINVAL;
2846         }
2847
2848         /*
2849          * Obvious sys_chunk_array corruptions, it must hold at least one key
2850          * and one chunk
2851          */
2852         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2853                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2854                           btrfs_super_sys_array_size(sb),
2855                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2856                 ret = -EINVAL;
2857         }
2858         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2859                         + sizeof(struct btrfs_chunk)) {
2860                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2861                           btrfs_super_sys_array_size(sb),
2862                           sizeof(struct btrfs_disk_key)
2863                           + sizeof(struct btrfs_chunk));
2864                 ret = -EINVAL;
2865         }
2866
2867         /*
2868          * The generation is a global counter, we'll trust it more than the others
2869          * but it's still possible that it's the one that's wrong.
2870          */
2871         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2872                 btrfs_warn(fs_info,
2873                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2874                         btrfs_super_generation(sb),
2875                         btrfs_super_chunk_root_generation(sb));
2876         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2877             && btrfs_super_cache_generation(sb) != (u64)-1)
2878                 btrfs_warn(fs_info,
2879                         "suspicious: generation < cache_generation: %llu < %llu",
2880                         btrfs_super_generation(sb),
2881                         btrfs_super_cache_generation(sb));
2882
2883         return ret;
2884 }
2885
2886 /*
2887  * Validation of super block at mount time.
2888  * Some checks already done early at mount time, like csum type and incompat
2889  * flags will be skipped.
2890  */
2891 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2892 {
2893         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2894 }
2895
2896 /*
2897  * Validation of super block at write time.
2898  * Some checks like bytenr check will be skipped as their values will be
2899  * overwritten soon.
2900  * Extra checks like csum type and incompat flags will be done here.
2901  */
2902 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2903                                       struct btrfs_super_block *sb)
2904 {
2905         int ret;
2906
2907         ret = btrfs_validate_super(fs_info, sb, -1);
2908         if (ret < 0)
2909                 goto out;
2910         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2911                 ret = -EUCLEAN;
2912                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2913                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2914                 goto out;
2915         }
2916         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2917                 ret = -EUCLEAN;
2918                 btrfs_err(fs_info,
2919                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2920                           btrfs_super_incompat_flags(sb),
2921                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2922                 goto out;
2923         }
2924 out:
2925         if (ret < 0)
2926                 btrfs_err(fs_info,
2927                 "super block corruption detected before writing it to disk");
2928         return ret;
2929 }
2930
2931 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2932 {
2933         struct btrfs_tree_parent_check check = {
2934                 .level = level,
2935                 .transid = gen,
2936                 .owner_root = root->root_key.objectid
2937         };
2938         int ret = 0;
2939
2940         root->node = read_tree_block(root->fs_info, bytenr, &check);
2941         if (IS_ERR(root->node)) {
2942                 ret = PTR_ERR(root->node);
2943                 root->node = NULL;
2944                 return ret;
2945         }
2946         if (!extent_buffer_uptodate(root->node)) {
2947                 free_extent_buffer(root->node);
2948                 root->node = NULL;
2949                 return -EIO;
2950         }
2951
2952         btrfs_set_root_node(&root->root_item, root->node);
2953         root->commit_root = btrfs_root_node(root);
2954         btrfs_set_root_refs(&root->root_item, 1);
2955         return ret;
2956 }
2957
2958 static int load_important_roots(struct btrfs_fs_info *fs_info)
2959 {
2960         struct btrfs_super_block *sb = fs_info->super_copy;
2961         u64 gen, bytenr;
2962         int level, ret;
2963
2964         bytenr = btrfs_super_root(sb);
2965         gen = btrfs_super_generation(sb);
2966         level = btrfs_super_root_level(sb);
2967         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2968         if (ret) {
2969                 btrfs_warn(fs_info, "couldn't read tree root");
2970                 return ret;
2971         }
2972         return 0;
2973 }
2974
2975 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2976 {
2977         int backup_index = find_newest_super_backup(fs_info);
2978         struct btrfs_super_block *sb = fs_info->super_copy;
2979         struct btrfs_root *tree_root = fs_info->tree_root;
2980         bool handle_error = false;
2981         int ret = 0;
2982         int i;
2983
2984         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2985                 if (handle_error) {
2986                         if (!IS_ERR(tree_root->node))
2987                                 free_extent_buffer(tree_root->node);
2988                         tree_root->node = NULL;
2989
2990                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2991                                 break;
2992
2993                         free_root_pointers(fs_info, 0);
2994
2995                         /*
2996                          * Don't use the log in recovery mode, it won't be
2997                          * valid
2998                          */
2999                         btrfs_set_super_log_root(sb, 0);
3000
3001                         /* We can't trust the free space cache either */
3002                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3003
3004                         ret = read_backup_root(fs_info, i);
3005                         backup_index = ret;
3006                         if (ret < 0)
3007                                 return ret;
3008                 }
3009
3010                 ret = load_important_roots(fs_info);
3011                 if (ret) {
3012                         handle_error = true;
3013                         continue;
3014                 }
3015
3016                 /*
3017                  * No need to hold btrfs_root::objectid_mutex since the fs
3018                  * hasn't been fully initialised and we are the only user
3019                  */
3020                 ret = btrfs_init_root_free_objectid(tree_root);
3021                 if (ret < 0) {
3022                         handle_error = true;
3023                         continue;
3024                 }
3025
3026                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3027
3028                 ret = btrfs_read_roots(fs_info);
3029                 if (ret < 0) {
3030                         handle_error = true;
3031                         continue;
3032                 }
3033
3034                 /* All successful */
3035                 fs_info->generation = btrfs_header_generation(tree_root->node);
3036                 fs_info->last_trans_committed = fs_info->generation;
3037                 fs_info->last_reloc_trans = 0;
3038
3039                 /* Always begin writing backup roots after the one being used */
3040                 if (backup_index < 0) {
3041                         fs_info->backup_root_index = 0;
3042                 } else {
3043                         fs_info->backup_root_index = backup_index + 1;
3044                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3045                 }
3046                 break;
3047         }
3048
3049         return ret;
3050 }
3051
3052 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3053 {
3054         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3055         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3056         INIT_LIST_HEAD(&fs_info->trans_list);
3057         INIT_LIST_HEAD(&fs_info->dead_roots);
3058         INIT_LIST_HEAD(&fs_info->delayed_iputs);
3059         INIT_LIST_HEAD(&fs_info->delalloc_roots);
3060         INIT_LIST_HEAD(&fs_info->caching_block_groups);
3061         spin_lock_init(&fs_info->delalloc_root_lock);
3062         spin_lock_init(&fs_info->trans_lock);
3063         spin_lock_init(&fs_info->fs_roots_radix_lock);
3064         spin_lock_init(&fs_info->delayed_iput_lock);
3065         spin_lock_init(&fs_info->defrag_inodes_lock);
3066         spin_lock_init(&fs_info->super_lock);
3067         spin_lock_init(&fs_info->buffer_lock);
3068         spin_lock_init(&fs_info->unused_bgs_lock);
3069         spin_lock_init(&fs_info->treelog_bg_lock);
3070         spin_lock_init(&fs_info->zone_active_bgs_lock);
3071         spin_lock_init(&fs_info->relocation_bg_lock);
3072         rwlock_init(&fs_info->tree_mod_log_lock);
3073         rwlock_init(&fs_info->global_root_lock);
3074         mutex_init(&fs_info->unused_bg_unpin_mutex);
3075         mutex_init(&fs_info->reclaim_bgs_lock);
3076         mutex_init(&fs_info->reloc_mutex);
3077         mutex_init(&fs_info->delalloc_root_mutex);
3078         mutex_init(&fs_info->zoned_meta_io_lock);
3079         mutex_init(&fs_info->zoned_data_reloc_io_lock);
3080         seqlock_init(&fs_info->profiles_lock);
3081
3082         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3083         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3084         btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3085         btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3086         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3087                                      BTRFS_LOCKDEP_TRANS_COMMIT_START);
3088         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3089                                      BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3090         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3091                                      BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3092         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3093                                      BTRFS_LOCKDEP_TRANS_COMPLETED);
3094
3095         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3096         INIT_LIST_HEAD(&fs_info->space_info);
3097         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3098         INIT_LIST_HEAD(&fs_info->unused_bgs);
3099         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3100         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3101 #ifdef CONFIG_BTRFS_DEBUG
3102         INIT_LIST_HEAD(&fs_info->allocated_roots);
3103         INIT_LIST_HEAD(&fs_info->allocated_ebs);
3104         spin_lock_init(&fs_info->eb_leak_lock);
3105 #endif
3106         extent_map_tree_init(&fs_info->mapping_tree);
3107         btrfs_init_block_rsv(&fs_info->global_block_rsv,
3108                              BTRFS_BLOCK_RSV_GLOBAL);
3109         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3110         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3111         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3112         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3113                              BTRFS_BLOCK_RSV_DELOPS);
3114         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3115                              BTRFS_BLOCK_RSV_DELREFS);
3116
3117         atomic_set(&fs_info->async_delalloc_pages, 0);
3118         atomic_set(&fs_info->defrag_running, 0);
3119         atomic_set(&fs_info->nr_delayed_iputs, 0);
3120         atomic64_set(&fs_info->tree_mod_seq, 0);
3121         fs_info->global_root_tree = RB_ROOT;
3122         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3123         fs_info->metadata_ratio = 0;
3124         fs_info->defrag_inodes = RB_ROOT;
3125         atomic64_set(&fs_info->free_chunk_space, 0);
3126         fs_info->tree_mod_log = RB_ROOT;
3127         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3128         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3129         btrfs_init_ref_verify(fs_info);
3130
3131         fs_info->thread_pool_size = min_t(unsigned long,
3132                                           num_online_cpus() + 2, 8);
3133
3134         INIT_LIST_HEAD(&fs_info->ordered_roots);
3135         spin_lock_init(&fs_info->ordered_root_lock);
3136
3137         btrfs_init_scrub(fs_info);
3138 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3139         fs_info->check_integrity_print_mask = 0;
3140 #endif
3141         btrfs_init_balance(fs_info);
3142         btrfs_init_async_reclaim_work(fs_info);
3143
3144         rwlock_init(&fs_info->block_group_cache_lock);
3145         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3146
3147         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3148                             IO_TREE_FS_EXCLUDED_EXTENTS);
3149
3150         mutex_init(&fs_info->ordered_operations_mutex);
3151         mutex_init(&fs_info->tree_log_mutex);
3152         mutex_init(&fs_info->chunk_mutex);
3153         mutex_init(&fs_info->transaction_kthread_mutex);
3154         mutex_init(&fs_info->cleaner_mutex);
3155         mutex_init(&fs_info->ro_block_group_mutex);
3156         init_rwsem(&fs_info->commit_root_sem);
3157         init_rwsem(&fs_info->cleanup_work_sem);
3158         init_rwsem(&fs_info->subvol_sem);
3159         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3160
3161         btrfs_init_dev_replace_locks(fs_info);
3162         btrfs_init_qgroup(fs_info);
3163         btrfs_discard_init(fs_info);
3164
3165         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3166         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3167
3168         init_waitqueue_head(&fs_info->transaction_throttle);
3169         init_waitqueue_head(&fs_info->transaction_wait);
3170         init_waitqueue_head(&fs_info->transaction_blocked_wait);
3171         init_waitqueue_head(&fs_info->async_submit_wait);
3172         init_waitqueue_head(&fs_info->delayed_iputs_wait);
3173
3174         /* Usable values until the real ones are cached from the superblock */
3175         fs_info->nodesize = 4096;
3176         fs_info->sectorsize = 4096;
3177         fs_info->sectorsize_bits = ilog2(4096);
3178         fs_info->stripesize = 4096;
3179
3180         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3181
3182         spin_lock_init(&fs_info->swapfile_pins_lock);
3183         fs_info->swapfile_pins = RB_ROOT;
3184
3185         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3186         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3187 }
3188
3189 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3190 {
3191         int ret;
3192
3193         fs_info->sb = sb;
3194         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3195         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3196
3197         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3198         if (ret)
3199                 return ret;
3200
3201         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3202         if (ret)
3203                 return ret;
3204
3205         fs_info->dirty_metadata_batch = PAGE_SIZE *
3206                                         (1 + ilog2(nr_cpu_ids));
3207
3208         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3209         if (ret)
3210                 return ret;
3211
3212         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3213                         GFP_KERNEL);
3214         if (ret)
3215                 return ret;
3216
3217         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3218                                         GFP_KERNEL);
3219         if (!fs_info->delayed_root)
3220                 return -ENOMEM;
3221         btrfs_init_delayed_root(fs_info->delayed_root);
3222
3223         if (sb_rdonly(sb))
3224                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3225
3226         return btrfs_alloc_stripe_hash_table(fs_info);
3227 }
3228
3229 static int btrfs_uuid_rescan_kthread(void *data)
3230 {
3231         struct btrfs_fs_info *fs_info = data;
3232         int ret;
3233
3234         /*
3235          * 1st step is to iterate through the existing UUID tree and
3236          * to delete all entries that contain outdated data.
3237          * 2nd step is to add all missing entries to the UUID tree.
3238          */
3239         ret = btrfs_uuid_tree_iterate(fs_info);
3240         if (ret < 0) {
3241                 if (ret != -EINTR)
3242                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3243                                    ret);
3244                 up(&fs_info->uuid_tree_rescan_sem);
3245                 return ret;
3246         }
3247         return btrfs_uuid_scan_kthread(data);
3248 }
3249
3250 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3251 {
3252         struct task_struct *task;
3253
3254         down(&fs_info->uuid_tree_rescan_sem);
3255         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3256         if (IS_ERR(task)) {
3257                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3258                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3259                 up(&fs_info->uuid_tree_rescan_sem);
3260                 return PTR_ERR(task);
3261         }
3262
3263         return 0;
3264 }
3265
3266 /*
3267  * Some options only have meaning at mount time and shouldn't persist across
3268  * remounts, or be displayed. Clear these at the end of mount and remount
3269  * code paths.
3270  */
3271 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3272 {
3273         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3274         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3275 }
3276
3277 /*
3278  * Mounting logic specific to read-write file systems. Shared by open_ctree
3279  * and btrfs_remount when remounting from read-only to read-write.
3280  */
3281 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3282 {
3283         int ret;
3284         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3285         bool clear_free_space_tree = false;
3286
3287         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3288             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3289                 clear_free_space_tree = true;
3290         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3291                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3292                 btrfs_warn(fs_info, "free space tree is invalid");
3293                 clear_free_space_tree = true;
3294         }
3295
3296         if (clear_free_space_tree) {
3297                 btrfs_info(fs_info, "clearing free space tree");
3298                 ret = btrfs_clear_free_space_tree(fs_info);
3299                 if (ret) {
3300                         btrfs_warn(fs_info,
3301                                    "failed to clear free space tree: %d", ret);
3302                         goto out;
3303                 }
3304         }
3305
3306         /*
3307          * btrfs_find_orphan_roots() is responsible for finding all the dead
3308          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3309          * them into the fs_info->fs_roots_radix tree. This must be done before
3310          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3311          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3312          * item before the root's tree is deleted - this means that if we unmount
3313          * or crash before the deletion completes, on the next mount we will not
3314          * delete what remains of the tree because the orphan item does not
3315          * exists anymore, which is what tells us we have a pending deletion.
3316          */
3317         ret = btrfs_find_orphan_roots(fs_info);
3318         if (ret)
3319                 goto out;
3320
3321         ret = btrfs_cleanup_fs_roots(fs_info);
3322         if (ret)
3323                 goto out;
3324
3325         down_read(&fs_info->cleanup_work_sem);
3326         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3327             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3328                 up_read(&fs_info->cleanup_work_sem);
3329                 goto out;
3330         }
3331         up_read(&fs_info->cleanup_work_sem);
3332
3333         mutex_lock(&fs_info->cleaner_mutex);
3334         ret = btrfs_recover_relocation(fs_info);
3335         mutex_unlock(&fs_info->cleaner_mutex);
3336         if (ret < 0) {
3337                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3338                 goto out;
3339         }
3340
3341         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3342             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3343                 btrfs_info(fs_info, "creating free space tree");
3344                 ret = btrfs_create_free_space_tree(fs_info);
3345                 if (ret) {
3346                         btrfs_warn(fs_info,
3347                                 "failed to create free space tree: %d", ret);
3348                         goto out;
3349                 }
3350         }
3351
3352         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3353                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3354                 if (ret)
3355                         goto out;
3356         }
3357
3358         ret = btrfs_resume_balance_async(fs_info);
3359         if (ret)
3360                 goto out;
3361
3362         ret = btrfs_resume_dev_replace_async(fs_info);
3363         if (ret) {
3364                 btrfs_warn(fs_info, "failed to resume dev_replace");
3365                 goto out;
3366         }
3367
3368         btrfs_qgroup_rescan_resume(fs_info);
3369
3370         if (!fs_info->uuid_root) {
3371                 btrfs_info(fs_info, "creating UUID tree");
3372                 ret = btrfs_create_uuid_tree(fs_info);
3373                 if (ret) {
3374                         btrfs_warn(fs_info,
3375                                    "failed to create the UUID tree %d", ret);
3376                         goto out;
3377                 }
3378         }
3379
3380 out:
3381         return ret;
3382 }
3383
3384 /*
3385  * Do various sanity and dependency checks of different features.
3386  *
3387  * This is the place for less strict checks (like for subpage or artificial
3388  * feature dependencies).
3389  *
3390  * For strict checks or possible corruption detection, see
3391  * btrfs_validate_super().
3392  *
3393  * This should be called after btrfs_parse_options(), as some mount options
3394  * (space cache related) can modify on-disk format like free space tree and
3395  * screw up certain feature dependencies.
3396  */
3397 int btrfs_check_features(struct btrfs_fs_info *fs_info, struct super_block *sb)
3398 {
3399         struct btrfs_super_block *disk_super = fs_info->super_copy;
3400         u64 incompat = btrfs_super_incompat_flags(disk_super);
3401         const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3402         const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3403
3404         if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3405                 btrfs_err(fs_info,
3406                 "cannot mount because of unknown incompat features (0x%llx)",
3407                     incompat);
3408                 return -EINVAL;
3409         }
3410
3411         /* Runtime limitation for mixed block groups. */
3412         if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3413             (fs_info->sectorsize != fs_info->nodesize)) {
3414                 btrfs_err(fs_info,
3415 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3416                         fs_info->nodesize, fs_info->sectorsize);
3417                 return -EINVAL;
3418         }
3419
3420         /* Mixed backref is an always-enabled feature. */
3421         incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3422
3423         /* Set compression related flags just in case. */
3424         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3425                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3426         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3427                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3428
3429         /*
3430          * An ancient flag, which should really be marked deprecated.
3431          * Such runtime limitation doesn't really need a incompat flag.
3432          */
3433         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3434                 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3435
3436         if (compat_ro_unsupp && !sb_rdonly(sb)) {
3437                 btrfs_err(fs_info,
3438         "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3439                        compat_ro);
3440                 return -EINVAL;
3441         }
3442
3443         /*
3444          * We have unsupported RO compat features, although RO mounted, we
3445          * should not cause any metadata writes, including log replay.
3446          * Or we could screw up whatever the new feature requires.
3447          */
3448         if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3449             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3450                 btrfs_err(fs_info,
3451 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3452                           compat_ro);
3453                 return -EINVAL;
3454         }
3455
3456         /*
3457          * Artificial limitations for block group tree, to force
3458          * block-group-tree to rely on no-holes and free-space-tree.
3459          */
3460         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3461             (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3462              !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3463                 btrfs_err(fs_info,
3464 "block-group-tree feature requires no-holes and free-space-tree features");
3465                 return -EINVAL;
3466         }
3467
3468         /*
3469          * Subpage runtime limitation on v1 cache.
3470          *
3471          * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3472          * we're already defaulting to v2 cache, no need to bother v1 as it's
3473          * going to be deprecated anyway.
3474          */
3475         if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3476                 btrfs_warn(fs_info,
3477         "v1 space cache is not supported for page size %lu with sectorsize %u",
3478                            PAGE_SIZE, fs_info->sectorsize);
3479                 return -EINVAL;
3480         }
3481
3482         /* This can be called by remount, we need to protect the super block. */
3483         spin_lock(&fs_info->super_lock);
3484         btrfs_set_super_incompat_flags(disk_super, incompat);
3485         spin_unlock(&fs_info->super_lock);
3486
3487         return 0;
3488 }
3489
3490 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3491                       char *options)
3492 {
3493         u32 sectorsize;
3494         u32 nodesize;
3495         u32 stripesize;
3496         u64 generation;
3497         u64 features;
3498         u16 csum_type;
3499         struct btrfs_super_block *disk_super;
3500         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3501         struct btrfs_root *tree_root;
3502         struct btrfs_root *chunk_root;
3503         int ret;
3504         int err = -EINVAL;
3505         int level;
3506
3507         ret = init_mount_fs_info(fs_info, sb);
3508         if (ret) {
3509                 err = ret;
3510                 goto fail;
3511         }
3512
3513         /* These need to be init'ed before we start creating inodes and such. */
3514         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3515                                      GFP_KERNEL);
3516         fs_info->tree_root = tree_root;
3517         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3518                                       GFP_KERNEL);
3519         fs_info->chunk_root = chunk_root;
3520         if (!tree_root || !chunk_root) {
3521                 err = -ENOMEM;
3522                 goto fail;
3523         }
3524
3525         fs_info->btree_inode = new_inode(sb);
3526         if (!fs_info->btree_inode) {
3527                 err = -ENOMEM;
3528                 goto fail;
3529         }
3530         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3531         btrfs_init_btree_inode(fs_info);
3532
3533         invalidate_bdev(fs_devices->latest_dev->bdev);
3534
3535         /*
3536          * Read super block and check the signature bytes only
3537          */
3538         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3539         if (IS_ERR(disk_super)) {
3540                 err = PTR_ERR(disk_super);
3541                 goto fail_alloc;
3542         }
3543
3544         /*
3545          * Verify the type first, if that or the checksum value are
3546          * corrupted, we'll find out
3547          */
3548         csum_type = btrfs_super_csum_type(disk_super);
3549         if (!btrfs_supported_super_csum(csum_type)) {
3550                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3551                           csum_type);
3552                 err = -EINVAL;
3553                 btrfs_release_disk_super(disk_super);
3554                 goto fail_alloc;
3555         }
3556
3557         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3558
3559         ret = btrfs_init_csum_hash(fs_info, csum_type);
3560         if (ret) {
3561                 err = ret;
3562                 btrfs_release_disk_super(disk_super);
3563                 goto fail_alloc;
3564         }
3565
3566         /*
3567          * We want to check superblock checksum, the type is stored inside.
3568          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3569          */
3570         if (btrfs_check_super_csum(fs_info, disk_super)) {
3571                 btrfs_err(fs_info, "superblock checksum mismatch");
3572                 err = -EINVAL;
3573                 btrfs_release_disk_super(disk_super);
3574                 goto fail_alloc;
3575         }
3576
3577         /*
3578          * super_copy is zeroed at allocation time and we never touch the
3579          * following bytes up to INFO_SIZE, the checksum is calculated from
3580          * the whole block of INFO_SIZE
3581          */
3582         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3583         btrfs_release_disk_super(disk_super);
3584
3585         disk_super = fs_info->super_copy;
3586
3587
3588         features = btrfs_super_flags(disk_super);
3589         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3590                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3591                 btrfs_set_super_flags(disk_super, features);
3592                 btrfs_info(fs_info,
3593                         "found metadata UUID change in progress flag, clearing");
3594         }
3595
3596         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3597                sizeof(*fs_info->super_for_commit));
3598
3599         ret = btrfs_validate_mount_super(fs_info);
3600         if (ret) {
3601                 btrfs_err(fs_info, "superblock contains fatal errors");
3602                 err = -EINVAL;
3603                 goto fail_alloc;
3604         }
3605
3606         if (!btrfs_super_root(disk_super))
3607                 goto fail_alloc;
3608
3609         /* check FS state, whether FS is broken. */
3610         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3611                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3612
3613         /*
3614          * In the long term, we'll store the compression type in the super
3615          * block, and it'll be used for per file compression control.
3616          */
3617         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3618
3619
3620         /* Set up fs_info before parsing mount options */
3621         nodesize = btrfs_super_nodesize(disk_super);
3622         sectorsize = btrfs_super_sectorsize(disk_super);
3623         stripesize = sectorsize;
3624         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3625         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3626
3627         fs_info->nodesize = nodesize;
3628         fs_info->sectorsize = sectorsize;
3629         fs_info->sectorsize_bits = ilog2(sectorsize);
3630         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3631         fs_info->stripesize = stripesize;
3632
3633         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3634         if (ret) {
3635                 err = ret;
3636                 goto fail_alloc;
3637         }
3638
3639         ret = btrfs_check_features(fs_info, sb);
3640         if (ret < 0) {
3641                 err = ret;
3642                 goto fail_alloc;
3643         }
3644
3645         if (sectorsize < PAGE_SIZE) {
3646                 struct btrfs_subpage_info *subpage_info;
3647
3648                 /*
3649                  * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3650                  * going to be deprecated.
3651                  *
3652                  * Force to use v2 cache for subpage case.
3653                  */
3654                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3655                 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3656                         "forcing free space tree for sector size %u with page size %lu",
3657                         sectorsize, PAGE_SIZE);
3658
3659                 btrfs_warn(fs_info,
3660                 "read-write for sector size %u with page size %lu is experimental",
3661                            sectorsize, PAGE_SIZE);
3662                 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3663                 if (!subpage_info)
3664                         goto fail_alloc;
3665                 btrfs_init_subpage_info(subpage_info, sectorsize);
3666                 fs_info->subpage_info = subpage_info;
3667         }
3668
3669         ret = btrfs_init_workqueues(fs_info);
3670         if (ret) {
3671                 err = ret;
3672                 goto fail_sb_buffer;
3673         }
3674
3675         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3676         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3677
3678         sb->s_blocksize = sectorsize;
3679         sb->s_blocksize_bits = blksize_bits(sectorsize);
3680         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3681
3682         mutex_lock(&fs_info->chunk_mutex);
3683         ret = btrfs_read_sys_array(fs_info);
3684         mutex_unlock(&fs_info->chunk_mutex);
3685         if (ret) {
3686                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3687                 goto fail_sb_buffer;
3688         }
3689
3690         generation = btrfs_super_chunk_root_generation(disk_super);
3691         level = btrfs_super_chunk_root_level(disk_super);
3692         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3693                               generation, level);
3694         if (ret) {
3695                 btrfs_err(fs_info, "failed to read chunk root");
3696                 goto fail_tree_roots;
3697         }
3698
3699         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3700                            offsetof(struct btrfs_header, chunk_tree_uuid),
3701                            BTRFS_UUID_SIZE);
3702
3703         ret = btrfs_read_chunk_tree(fs_info);
3704         if (ret) {
3705                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3706                 goto fail_tree_roots;
3707         }
3708
3709         /*
3710          * At this point we know all the devices that make this filesystem,
3711          * including the seed devices but we don't know yet if the replace
3712          * target is required. So free devices that are not part of this
3713          * filesystem but skip the replace target device which is checked
3714          * below in btrfs_init_dev_replace().
3715          */
3716         btrfs_free_extra_devids(fs_devices);
3717         if (!fs_devices->latest_dev->bdev) {
3718                 btrfs_err(fs_info, "failed to read devices");
3719                 goto fail_tree_roots;
3720         }
3721
3722         ret = init_tree_roots(fs_info);
3723         if (ret)
3724                 goto fail_tree_roots;
3725
3726         /*
3727          * Get zone type information of zoned block devices. This will also
3728          * handle emulation of a zoned filesystem if a regular device has the
3729          * zoned incompat feature flag set.
3730          */
3731         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3732         if (ret) {
3733                 btrfs_err(fs_info,
3734                           "zoned: failed to read device zone info: %d",
3735                           ret);
3736                 goto fail_block_groups;
3737         }
3738
3739         /*
3740          * If we have a uuid root and we're not being told to rescan we need to
3741          * check the generation here so we can set the
3742          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3743          * transaction during a balance or the log replay without updating the
3744          * uuid generation, and then if we crash we would rescan the uuid tree,
3745          * even though it was perfectly fine.
3746          */
3747         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3748             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3749                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3750
3751         ret = btrfs_verify_dev_extents(fs_info);
3752         if (ret) {
3753                 btrfs_err(fs_info,
3754                           "failed to verify dev extents against chunks: %d",
3755                           ret);
3756                 goto fail_block_groups;
3757         }
3758         ret = btrfs_recover_balance(fs_info);
3759         if (ret) {
3760                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3761                 goto fail_block_groups;
3762         }
3763
3764         ret = btrfs_init_dev_stats(fs_info);
3765         if (ret) {
3766                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3767                 goto fail_block_groups;
3768         }
3769
3770         ret = btrfs_init_dev_replace(fs_info);
3771         if (ret) {
3772                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3773                 goto fail_block_groups;
3774         }
3775
3776         ret = btrfs_check_zoned_mode(fs_info);
3777         if (ret) {
3778                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3779                           ret);
3780                 goto fail_block_groups;
3781         }
3782
3783         ret = btrfs_sysfs_add_fsid(fs_devices);
3784         if (ret) {
3785                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3786                                 ret);
3787                 goto fail_block_groups;
3788         }
3789
3790         ret = btrfs_sysfs_add_mounted(fs_info);
3791         if (ret) {
3792                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3793                 goto fail_fsdev_sysfs;
3794         }
3795
3796         ret = btrfs_init_space_info(fs_info);
3797         if (ret) {
3798                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3799                 goto fail_sysfs;
3800         }
3801
3802         ret = btrfs_read_block_groups(fs_info);
3803         if (ret) {
3804                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3805                 goto fail_sysfs;
3806         }
3807
3808         btrfs_free_zone_cache(fs_info);
3809
3810         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3811             !btrfs_check_rw_degradable(fs_info, NULL)) {
3812                 btrfs_warn(fs_info,
3813                 "writable mount is not allowed due to too many missing devices");
3814                 goto fail_sysfs;
3815         }
3816
3817         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3818                                                "btrfs-cleaner");
3819         if (IS_ERR(fs_info->cleaner_kthread))
3820                 goto fail_sysfs;
3821
3822         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3823                                                    tree_root,
3824                                                    "btrfs-transaction");
3825         if (IS_ERR(fs_info->transaction_kthread))
3826                 goto fail_cleaner;
3827
3828         if (!btrfs_test_opt(fs_info, NOSSD) &&
3829             !fs_info->fs_devices->rotating) {
3830                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3831         }
3832
3833         /*
3834          * For devices supporting discard turn on discard=async automatically,
3835          * unless it's already set or disabled. This could be turned off by
3836          * nodiscard for the same mount.
3837          */
3838         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3839               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3840               btrfs_test_opt(fs_info, NODISCARD)) &&
3841             fs_info->fs_devices->discardable) {
3842                 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3843                                    "auto enabling async discard");
3844                 btrfs_clear_opt(fs_info->mount_opt, NODISCARD);
3845         }
3846
3847 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3848         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3849                 ret = btrfsic_mount(fs_info, fs_devices,
3850                                     btrfs_test_opt(fs_info,
3851                                         CHECK_INTEGRITY_DATA) ? 1 : 0,
3852                                     fs_info->check_integrity_print_mask);
3853                 if (ret)
3854                         btrfs_warn(fs_info,
3855                                 "failed to initialize integrity check module: %d",
3856                                 ret);
3857         }
3858 #endif
3859         ret = btrfs_read_qgroup_config(fs_info);
3860         if (ret)
3861                 goto fail_trans_kthread;
3862
3863         if (btrfs_build_ref_tree(fs_info))
3864                 btrfs_err(fs_info, "couldn't build ref tree");
3865
3866         /* do not make disk changes in broken FS or nologreplay is given */
3867         if (btrfs_super_log_root(disk_super) != 0 &&
3868             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3869                 btrfs_info(fs_info, "start tree-log replay");
3870                 ret = btrfs_replay_log(fs_info, fs_devices);
3871                 if (ret) {
3872                         err = ret;
3873                         goto fail_qgroup;
3874                 }
3875         }
3876
3877         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3878         if (IS_ERR(fs_info->fs_root)) {
3879                 err = PTR_ERR(fs_info->fs_root);
3880                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3881                 fs_info->fs_root = NULL;
3882                 goto fail_qgroup;
3883         }
3884
3885         if (sb_rdonly(sb))
3886                 goto clear_oneshot;
3887
3888         ret = btrfs_start_pre_rw_mount(fs_info);
3889         if (ret) {
3890                 close_ctree(fs_info);
3891                 return ret;
3892         }
3893         btrfs_discard_resume(fs_info);
3894
3895         if (fs_info->uuid_root &&
3896             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3897              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3898                 btrfs_info(fs_info, "checking UUID tree");
3899                 ret = btrfs_check_uuid_tree(fs_info);
3900                 if (ret) {
3901                         btrfs_warn(fs_info,
3902                                 "failed to check the UUID tree: %d", ret);
3903                         close_ctree(fs_info);
3904                         return ret;
3905                 }
3906         }
3907
3908         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3909
3910         /* Kick the cleaner thread so it'll start deleting snapshots. */
3911         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3912                 wake_up_process(fs_info->cleaner_kthread);
3913
3914 clear_oneshot:
3915         btrfs_clear_oneshot_options(fs_info);
3916         return 0;
3917
3918 fail_qgroup:
3919         btrfs_free_qgroup_config(fs_info);
3920 fail_trans_kthread:
3921         kthread_stop(fs_info->transaction_kthread);
3922         btrfs_cleanup_transaction(fs_info);
3923         btrfs_free_fs_roots(fs_info);
3924 fail_cleaner:
3925         kthread_stop(fs_info->cleaner_kthread);
3926
3927         /*
3928          * make sure we're done with the btree inode before we stop our
3929          * kthreads
3930          */
3931         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3932
3933 fail_sysfs:
3934         btrfs_sysfs_remove_mounted(fs_info);
3935
3936 fail_fsdev_sysfs:
3937         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3938
3939 fail_block_groups:
3940         btrfs_put_block_group_cache(fs_info);
3941
3942 fail_tree_roots:
3943         if (fs_info->data_reloc_root)
3944                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3945         free_root_pointers(fs_info, true);
3946         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3947
3948 fail_sb_buffer:
3949         btrfs_stop_all_workers(fs_info);
3950         btrfs_free_block_groups(fs_info);
3951 fail_alloc:
3952         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3953
3954         iput(fs_info->btree_inode);
3955 fail:
3956         btrfs_close_devices(fs_info->fs_devices);
3957         return err;
3958 }
3959 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3960
3961 static void btrfs_end_super_write(struct bio *bio)
3962 {
3963         struct btrfs_device *device = bio->bi_private;
3964         struct bio_vec *bvec;
3965         struct bvec_iter_all iter_all;
3966         struct page *page;
3967
3968         bio_for_each_segment_all(bvec, bio, iter_all) {
3969                 page = bvec->bv_page;
3970
3971                 if (bio->bi_status) {
3972                         btrfs_warn_rl_in_rcu(device->fs_info,
3973                                 "lost page write due to IO error on %s (%d)",
3974                                 btrfs_dev_name(device),
3975                                 blk_status_to_errno(bio->bi_status));
3976                         ClearPageUptodate(page);
3977                         SetPageError(page);
3978                         btrfs_dev_stat_inc_and_print(device,
3979                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3980                 } else {
3981                         SetPageUptodate(page);
3982                 }
3983
3984                 put_page(page);
3985                 unlock_page(page);
3986         }
3987
3988         bio_put(bio);
3989 }
3990
3991 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3992                                                    int copy_num, bool drop_cache)
3993 {
3994         struct btrfs_super_block *super;
3995         struct page *page;
3996         u64 bytenr, bytenr_orig;
3997         struct address_space *mapping = bdev->bd_inode->i_mapping;
3998         int ret;
3999
4000         bytenr_orig = btrfs_sb_offset(copy_num);
4001         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4002         if (ret == -ENOENT)
4003                 return ERR_PTR(-EINVAL);
4004         else if (ret)
4005                 return ERR_PTR(ret);
4006
4007         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4008                 return ERR_PTR(-EINVAL);
4009
4010         if (drop_cache) {
4011                 /* This should only be called with the primary sb. */
4012                 ASSERT(copy_num == 0);
4013
4014                 /*
4015                  * Drop the page of the primary superblock, so later read will
4016                  * always read from the device.
4017                  */
4018                 invalidate_inode_pages2_range(mapping,
4019                                 bytenr >> PAGE_SHIFT,
4020                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
4021         }
4022
4023         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4024         if (IS_ERR(page))
4025                 return ERR_CAST(page);
4026
4027         super = page_address(page);
4028         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4029                 btrfs_release_disk_super(super);
4030                 return ERR_PTR(-ENODATA);
4031         }
4032
4033         if (btrfs_super_bytenr(super) != bytenr_orig) {
4034                 btrfs_release_disk_super(super);
4035                 return ERR_PTR(-EINVAL);
4036         }
4037
4038         return super;
4039 }
4040
4041
4042 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4043 {
4044         struct btrfs_super_block *super, *latest = NULL;
4045         int i;
4046         u64 transid = 0;
4047
4048         /* we would like to check all the supers, but that would make
4049          * a btrfs mount succeed after a mkfs from a different FS.
4050          * So, we need to add a special mount option to scan for
4051          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4052          */
4053         for (i = 0; i < 1; i++) {
4054                 super = btrfs_read_dev_one_super(bdev, i, false);
4055                 if (IS_ERR(super))
4056                         continue;
4057
4058                 if (!latest || btrfs_super_generation(super) > transid) {
4059                         if (latest)
4060                                 btrfs_release_disk_super(super);
4061
4062                         latest = super;
4063                         transid = btrfs_super_generation(super);
4064                 }
4065         }
4066
4067         return super;
4068 }
4069
4070 /*
4071  * Write superblock @sb to the @device. Do not wait for completion, all the
4072  * pages we use for writing are locked.
4073  *
4074  * Write @max_mirrors copies of the superblock, where 0 means default that fit
4075  * the expected device size at commit time. Note that max_mirrors must be
4076  * same for write and wait phases.
4077  *
4078  * Return number of errors when page is not found or submission fails.
4079  */
4080 static int write_dev_supers(struct btrfs_device *device,
4081                             struct btrfs_super_block *sb, int max_mirrors)
4082 {
4083         struct btrfs_fs_info *fs_info = device->fs_info;
4084         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4085         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4086         int i;
4087         int errors = 0;
4088         int ret;
4089         u64 bytenr, bytenr_orig;
4090
4091         if (max_mirrors == 0)
4092                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4093
4094         shash->tfm = fs_info->csum_shash;
4095
4096         for (i = 0; i < max_mirrors; i++) {
4097                 struct page *page;
4098                 struct bio *bio;
4099                 struct btrfs_super_block *disk_super;
4100
4101                 bytenr_orig = btrfs_sb_offset(i);
4102                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4103                 if (ret == -ENOENT) {
4104                         continue;
4105                 } else if (ret < 0) {
4106                         btrfs_err(device->fs_info,
4107                                 "couldn't get super block location for mirror %d",
4108                                 i);
4109                         errors++;
4110                         continue;
4111                 }
4112                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4113                     device->commit_total_bytes)
4114                         break;
4115
4116                 btrfs_set_super_bytenr(sb, bytenr_orig);
4117
4118                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4119                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4120                                     sb->csum);
4121
4122                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4123                                            GFP_NOFS);
4124                 if (!page) {
4125                         btrfs_err(device->fs_info,
4126                             "couldn't get super block page for bytenr %llu",
4127                             bytenr);
4128                         errors++;
4129                         continue;
4130                 }
4131
4132                 /* Bump the refcount for wait_dev_supers() */
4133                 get_page(page);
4134
4135                 disk_super = page_address(page);
4136                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4137
4138                 /*
4139                  * Directly use bios here instead of relying on the page cache
4140                  * to do I/O, so we don't lose the ability to do integrity
4141                  * checking.
4142                  */
4143                 bio = bio_alloc(device->bdev, 1,
4144                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4145                                 GFP_NOFS);
4146                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4147                 bio->bi_private = device;
4148                 bio->bi_end_io = btrfs_end_super_write;
4149                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4150                                offset_in_page(bytenr));
4151
4152                 /*
4153                  * We FUA only the first super block.  The others we allow to
4154                  * go down lazy and there's a short window where the on-disk
4155                  * copies might still contain the older version.
4156                  */
4157                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4158                         bio->bi_opf |= REQ_FUA;
4159
4160                 btrfsic_check_bio(bio);
4161                 submit_bio(bio);
4162
4163                 if (btrfs_advance_sb_log(device, i))
4164                         errors++;
4165         }
4166         return errors < i ? 0 : -1;
4167 }
4168
4169 /*
4170  * Wait for write completion of superblocks done by write_dev_supers,
4171  * @max_mirrors same for write and wait phases.
4172  *
4173  * Return number of errors when page is not found or not marked up to
4174  * date.
4175  */
4176 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4177 {
4178         int i;
4179         int errors = 0;
4180         bool primary_failed = false;
4181         int ret;
4182         u64 bytenr;
4183
4184         if (max_mirrors == 0)
4185                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4186
4187         for (i = 0; i < max_mirrors; i++) {
4188                 struct page *page;
4189
4190                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4191                 if (ret == -ENOENT) {
4192                         break;
4193                 } else if (ret < 0) {
4194                         errors++;
4195                         if (i == 0)
4196                                 primary_failed = true;
4197                         continue;
4198                 }
4199                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4200                     device->commit_total_bytes)
4201                         break;
4202
4203                 page = find_get_page(device->bdev->bd_inode->i_mapping,
4204                                      bytenr >> PAGE_SHIFT);
4205                 if (!page) {
4206                         errors++;
4207                         if (i == 0)
4208                                 primary_failed = true;
4209                         continue;
4210                 }
4211                 /* Page is submitted locked and unlocked once the IO completes */
4212                 wait_on_page_locked(page);
4213                 if (PageError(page)) {
4214                         errors++;
4215                         if (i == 0)
4216                                 primary_failed = true;
4217                 }
4218
4219                 /* Drop our reference */
4220                 put_page(page);
4221
4222                 /* Drop the reference from the writing run */
4223                 put_page(page);
4224         }
4225
4226         /* log error, force error return */
4227         if (primary_failed) {
4228                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4229                           device->devid);
4230                 return -1;
4231         }
4232
4233         return errors < i ? 0 : -1;
4234 }
4235
4236 /*
4237  * endio for the write_dev_flush, this will wake anyone waiting
4238  * for the barrier when it is done
4239  */
4240 static void btrfs_end_empty_barrier(struct bio *bio)
4241 {
4242         bio_uninit(bio);
4243         complete(bio->bi_private);
4244 }
4245
4246 /*
4247  * Submit a flush request to the device if it supports it. Error handling is
4248  * done in the waiting counterpart.
4249  */
4250 static void write_dev_flush(struct btrfs_device *device)
4251 {
4252         struct bio *bio = &device->flush_bio;
4253
4254 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4255         /*
4256          * When a disk has write caching disabled, we skip submission of a bio
4257          * with flush and sync requests before writing the superblock, since
4258          * it's not needed. However when the integrity checker is enabled, this
4259          * results in reports that there are metadata blocks referred by a
4260          * superblock that were not properly flushed. So don't skip the bio
4261          * submission only when the integrity checker is enabled for the sake
4262          * of simplicity, since this is a debug tool and not meant for use in
4263          * non-debug builds.
4264          */
4265         if (!bdev_write_cache(device->bdev))
4266                 return;
4267 #endif
4268
4269         bio_init(bio, device->bdev, NULL, 0,
4270                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4271         bio->bi_end_io = btrfs_end_empty_barrier;
4272         init_completion(&device->flush_wait);
4273         bio->bi_private = &device->flush_wait;
4274
4275         btrfsic_check_bio(bio);
4276         submit_bio(bio);
4277         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4278 }
4279
4280 /*
4281  * If the flush bio has been submitted by write_dev_flush, wait for it.
4282  */
4283 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4284 {
4285         struct bio *bio = &device->flush_bio;
4286
4287         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4288                 return BLK_STS_OK;
4289
4290         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4291         wait_for_completion_io(&device->flush_wait);
4292
4293         return bio->bi_status;
4294 }
4295
4296 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4297 {
4298         if (!btrfs_check_rw_degradable(fs_info, NULL))
4299                 return -EIO;
4300         return 0;
4301 }
4302
4303 /*
4304  * send an empty flush down to each device in parallel,
4305  * then wait for them
4306  */
4307 static int barrier_all_devices(struct btrfs_fs_info *info)
4308 {
4309         struct list_head *head;
4310         struct btrfs_device *dev;
4311         int errors_wait = 0;
4312         blk_status_t ret;
4313
4314         lockdep_assert_held(&info->fs_devices->device_list_mutex);
4315         /* send down all the barriers */
4316         head = &info->fs_devices->devices;
4317         list_for_each_entry(dev, head, dev_list) {
4318                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4319                         continue;
4320                 if (!dev->bdev)
4321                         continue;
4322                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4323                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4324                         continue;
4325
4326                 write_dev_flush(dev);
4327                 dev->last_flush_error = BLK_STS_OK;
4328         }
4329
4330         /* wait for all the barriers */
4331         list_for_each_entry(dev, head, dev_list) {
4332                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4333                         continue;
4334                 if (!dev->bdev) {
4335                         errors_wait++;
4336                         continue;
4337                 }
4338                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4339                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4340                         continue;
4341
4342                 ret = wait_dev_flush(dev);
4343                 if (ret) {
4344                         dev->last_flush_error = ret;
4345                         btrfs_dev_stat_inc_and_print(dev,
4346                                         BTRFS_DEV_STAT_FLUSH_ERRS);
4347                         errors_wait++;
4348                 }
4349         }
4350
4351         if (errors_wait) {
4352                 /*
4353                  * At some point we need the status of all disks
4354                  * to arrive at the volume status. So error checking
4355                  * is being pushed to a separate loop.
4356                  */
4357                 return check_barrier_error(info);
4358         }
4359         return 0;
4360 }
4361
4362 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4363 {
4364         int raid_type;
4365         int min_tolerated = INT_MAX;
4366
4367         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4368             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4369                 min_tolerated = min_t(int, min_tolerated,
4370                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4371                                     tolerated_failures);
4372
4373         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4374                 if (raid_type == BTRFS_RAID_SINGLE)
4375                         continue;
4376                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4377                         continue;
4378                 min_tolerated = min_t(int, min_tolerated,
4379                                     btrfs_raid_array[raid_type].
4380                                     tolerated_failures);
4381         }
4382
4383         if (min_tolerated == INT_MAX) {
4384                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4385                 min_tolerated = 0;
4386         }
4387
4388         return min_tolerated;
4389 }
4390
4391 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4392 {
4393         struct list_head *head;
4394         struct btrfs_device *dev;
4395         struct btrfs_super_block *sb;
4396         struct btrfs_dev_item *dev_item;
4397         int ret;
4398         int do_barriers;
4399         int max_errors;
4400         int total_errors = 0;
4401         u64 flags;
4402
4403         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4404
4405         /*
4406          * max_mirrors == 0 indicates we're from commit_transaction,
4407          * not from fsync where the tree roots in fs_info have not
4408          * been consistent on disk.
4409          */
4410         if (max_mirrors == 0)
4411                 backup_super_roots(fs_info);
4412
4413         sb = fs_info->super_for_commit;
4414         dev_item = &sb->dev_item;
4415
4416         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4417         head = &fs_info->fs_devices->devices;
4418         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4419
4420         if (do_barriers) {
4421                 ret = barrier_all_devices(fs_info);
4422                 if (ret) {
4423                         mutex_unlock(
4424                                 &fs_info->fs_devices->device_list_mutex);
4425                         btrfs_handle_fs_error(fs_info, ret,
4426                                               "errors while submitting device barriers.");
4427                         return ret;
4428                 }
4429         }
4430
4431         list_for_each_entry(dev, head, dev_list) {
4432                 if (!dev->bdev) {
4433                         total_errors++;
4434                         continue;
4435                 }
4436                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4437                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4438                         continue;
4439
4440                 btrfs_set_stack_device_generation(dev_item, 0);
4441                 btrfs_set_stack_device_type(dev_item, dev->type);
4442                 btrfs_set_stack_device_id(dev_item, dev->devid);
4443                 btrfs_set_stack_device_total_bytes(dev_item,
4444                                                    dev->commit_total_bytes);
4445                 btrfs_set_stack_device_bytes_used(dev_item,
4446                                                   dev->commit_bytes_used);
4447                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4448                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4449                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4450                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4451                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4452                        BTRFS_FSID_SIZE);
4453
4454                 flags = btrfs_super_flags(sb);
4455                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4456
4457                 ret = btrfs_validate_write_super(fs_info, sb);
4458                 if (ret < 0) {
4459                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4460                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4461                                 "unexpected superblock corruption detected");
4462                         return -EUCLEAN;
4463                 }
4464
4465                 ret = write_dev_supers(dev, sb, max_mirrors);
4466                 if (ret)
4467                         total_errors++;
4468         }
4469         if (total_errors > max_errors) {
4470                 btrfs_err(fs_info, "%d errors while writing supers",
4471                           total_errors);
4472                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4473
4474                 /* FUA is masked off if unsupported and can't be the reason */
4475                 btrfs_handle_fs_error(fs_info, -EIO,
4476                                       "%d errors while writing supers",
4477                                       total_errors);
4478                 return -EIO;
4479         }
4480
4481         total_errors = 0;
4482         list_for_each_entry(dev, head, dev_list) {
4483                 if (!dev->bdev)
4484                         continue;
4485                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4486                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4487                         continue;
4488
4489                 ret = wait_dev_supers(dev, max_mirrors);
4490                 if (ret)
4491                         total_errors++;
4492         }
4493         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4494         if (total_errors > max_errors) {
4495                 btrfs_handle_fs_error(fs_info, -EIO,
4496                                       "%d errors while writing supers",
4497                                       total_errors);
4498                 return -EIO;
4499         }
4500         return 0;
4501 }
4502
4503 /* Drop a fs root from the radix tree and free it. */
4504 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4505                                   struct btrfs_root *root)
4506 {
4507         bool drop_ref = false;
4508
4509         spin_lock(&fs_info->fs_roots_radix_lock);
4510         radix_tree_delete(&fs_info->fs_roots_radix,
4511                           (unsigned long)root->root_key.objectid);
4512         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4513                 drop_ref = true;
4514         spin_unlock(&fs_info->fs_roots_radix_lock);
4515
4516         if (BTRFS_FS_ERROR(fs_info)) {
4517                 ASSERT(root->log_root == NULL);
4518                 if (root->reloc_root) {
4519                         btrfs_put_root(root->reloc_root);
4520                         root->reloc_root = NULL;
4521                 }
4522         }
4523
4524         if (drop_ref)
4525                 btrfs_put_root(root);
4526 }
4527
4528 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4529 {
4530         u64 root_objectid = 0;
4531         struct btrfs_root *gang[8];
4532         int i = 0;
4533         int err = 0;
4534         unsigned int ret = 0;
4535
4536         while (1) {
4537                 spin_lock(&fs_info->fs_roots_radix_lock);
4538                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4539                                              (void **)gang, root_objectid,
4540                                              ARRAY_SIZE(gang));
4541                 if (!ret) {
4542                         spin_unlock(&fs_info->fs_roots_radix_lock);
4543                         break;
4544                 }
4545                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4546
4547                 for (i = 0; i < ret; i++) {
4548                         /* Avoid to grab roots in dead_roots */
4549                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4550                                 gang[i] = NULL;
4551                                 continue;
4552                         }
4553                         /* grab all the search result for later use */
4554                         gang[i] = btrfs_grab_root(gang[i]);
4555                 }
4556                 spin_unlock(&fs_info->fs_roots_radix_lock);
4557
4558                 for (i = 0; i < ret; i++) {
4559                         if (!gang[i])
4560                                 continue;
4561                         root_objectid = gang[i]->root_key.objectid;
4562                         err = btrfs_orphan_cleanup(gang[i]);
4563                         if (err)
4564                                 break;
4565                         btrfs_put_root(gang[i]);
4566                 }
4567                 root_objectid++;
4568         }
4569
4570         /* release the uncleaned roots due to error */
4571         for (; i < ret; i++) {
4572                 if (gang[i])
4573                         btrfs_put_root(gang[i]);
4574         }
4575         return err;
4576 }
4577
4578 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4579 {
4580         struct btrfs_root *root = fs_info->tree_root;
4581         struct btrfs_trans_handle *trans;
4582
4583         mutex_lock(&fs_info->cleaner_mutex);
4584         btrfs_run_delayed_iputs(fs_info);
4585         mutex_unlock(&fs_info->cleaner_mutex);
4586         wake_up_process(fs_info->cleaner_kthread);
4587
4588         /* wait until ongoing cleanup work done */
4589         down_write(&fs_info->cleanup_work_sem);
4590         up_write(&fs_info->cleanup_work_sem);
4591
4592         trans = btrfs_join_transaction(root);
4593         if (IS_ERR(trans))
4594                 return PTR_ERR(trans);
4595         return btrfs_commit_transaction(trans);
4596 }
4597
4598 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4599 {
4600         struct btrfs_transaction *trans;
4601         struct btrfs_transaction *tmp;
4602         bool found = false;
4603
4604         if (list_empty(&fs_info->trans_list))
4605                 return;
4606
4607         /*
4608          * This function is only called at the very end of close_ctree(),
4609          * thus no other running transaction, no need to take trans_lock.
4610          */
4611         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4612         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4613                 struct extent_state *cached = NULL;
4614                 u64 dirty_bytes = 0;
4615                 u64 cur = 0;
4616                 u64 found_start;
4617                 u64 found_end;
4618
4619                 found = true;
4620                 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4621                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4622                         dirty_bytes += found_end + 1 - found_start;
4623                         cur = found_end + 1;
4624                 }
4625                 btrfs_warn(fs_info,
4626         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4627                            trans->transid, dirty_bytes);
4628                 btrfs_cleanup_one_transaction(trans, fs_info);
4629
4630                 if (trans == fs_info->running_transaction)
4631                         fs_info->running_transaction = NULL;
4632                 list_del_init(&trans->list);
4633
4634                 btrfs_put_transaction(trans);
4635                 trace_btrfs_transaction_commit(fs_info);
4636         }
4637         ASSERT(!found);
4638 }
4639
4640 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4641 {
4642         int ret;
4643
4644         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4645
4646         /*
4647          * If we had UNFINISHED_DROPS we could still be processing them, so
4648          * clear that bit and wake up relocation so it can stop.
4649          * We must do this before stopping the block group reclaim task, because
4650          * at btrfs_relocate_block_group() we wait for this bit, and after the
4651          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4652          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4653          * return 1.
4654          */
4655         btrfs_wake_unfinished_drop(fs_info);
4656
4657         /*
4658          * We may have the reclaim task running and relocating a data block group,
4659          * in which case it may create delayed iputs. So stop it before we park
4660          * the cleaner kthread otherwise we can get new delayed iputs after
4661          * parking the cleaner, and that can make the async reclaim task to hang
4662          * if it's waiting for delayed iputs to complete, since the cleaner is
4663          * parked and can not run delayed iputs - this will make us hang when
4664          * trying to stop the async reclaim task.
4665          */
4666         cancel_work_sync(&fs_info->reclaim_bgs_work);
4667         /*
4668          * We don't want the cleaner to start new transactions, add more delayed
4669          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4670          * because that frees the task_struct, and the transaction kthread might
4671          * still try to wake up the cleaner.
4672          */
4673         kthread_park(fs_info->cleaner_kthread);
4674
4675         /* wait for the qgroup rescan worker to stop */
4676         btrfs_qgroup_wait_for_completion(fs_info, false);
4677
4678         /* wait for the uuid_scan task to finish */
4679         down(&fs_info->uuid_tree_rescan_sem);
4680         /* avoid complains from lockdep et al., set sem back to initial state */
4681         up(&fs_info->uuid_tree_rescan_sem);
4682
4683         /* pause restriper - we want to resume on mount */
4684         btrfs_pause_balance(fs_info);
4685
4686         btrfs_dev_replace_suspend_for_unmount(fs_info);
4687
4688         btrfs_scrub_cancel(fs_info);
4689
4690         /* wait for any defraggers to finish */
4691         wait_event(fs_info->transaction_wait,
4692                    (atomic_read(&fs_info->defrag_running) == 0));
4693
4694         /* clear out the rbtree of defraggable inodes */
4695         btrfs_cleanup_defrag_inodes(fs_info);
4696
4697         /*
4698          * After we parked the cleaner kthread, ordered extents may have
4699          * completed and created new delayed iputs. If one of the async reclaim
4700          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4701          * can hang forever trying to stop it, because if a delayed iput is
4702          * added after it ran btrfs_run_delayed_iputs() and before it called
4703          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4704          * no one else to run iputs.
4705          *
4706          * So wait for all ongoing ordered extents to complete and then run
4707          * delayed iputs. This works because once we reach this point no one
4708          * can either create new ordered extents nor create delayed iputs
4709          * through some other means.
4710          *
4711          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4712          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4713          * but the delayed iput for the respective inode is made only when doing
4714          * the final btrfs_put_ordered_extent() (which must happen at
4715          * btrfs_finish_ordered_io() when we are unmounting).
4716          */
4717         btrfs_flush_workqueue(fs_info->endio_write_workers);
4718         /* Ordered extents for free space inodes. */
4719         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4720         btrfs_run_delayed_iputs(fs_info);
4721
4722         cancel_work_sync(&fs_info->async_reclaim_work);
4723         cancel_work_sync(&fs_info->async_data_reclaim_work);
4724         cancel_work_sync(&fs_info->preempt_reclaim_work);
4725
4726         /* Cancel or finish ongoing discard work */
4727         btrfs_discard_cleanup(fs_info);
4728
4729         if (!sb_rdonly(fs_info->sb)) {
4730                 /*
4731                  * The cleaner kthread is stopped, so do one final pass over
4732                  * unused block groups.
4733                  */
4734                 btrfs_delete_unused_bgs(fs_info);
4735
4736                 /*
4737                  * There might be existing delayed inode workers still running
4738                  * and holding an empty delayed inode item. We must wait for
4739                  * them to complete first because they can create a transaction.
4740                  * This happens when someone calls btrfs_balance_delayed_items()
4741                  * and then a transaction commit runs the same delayed nodes
4742                  * before any delayed worker has done something with the nodes.
4743                  * We must wait for any worker here and not at transaction
4744                  * commit time since that could cause a deadlock.
4745                  * This is a very rare case.
4746                  */
4747                 btrfs_flush_workqueue(fs_info->delayed_workers);
4748
4749                 ret = btrfs_commit_super(fs_info);
4750                 if (ret)
4751                         btrfs_err(fs_info, "commit super ret %d", ret);
4752         }
4753
4754         if (BTRFS_FS_ERROR(fs_info))
4755                 btrfs_error_commit_super(fs_info);
4756
4757         kthread_stop(fs_info->transaction_kthread);
4758         kthread_stop(fs_info->cleaner_kthread);
4759
4760         ASSERT(list_empty(&fs_info->delayed_iputs));
4761         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4762
4763         if (btrfs_check_quota_leak(fs_info)) {
4764                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4765                 btrfs_err(fs_info, "qgroup reserved space leaked");
4766         }
4767
4768         btrfs_free_qgroup_config(fs_info);
4769         ASSERT(list_empty(&fs_info->delalloc_roots));
4770
4771         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4772                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4773                        percpu_counter_sum(&fs_info->delalloc_bytes));
4774         }
4775
4776         if (percpu_counter_sum(&fs_info->ordered_bytes))
4777                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4778                            percpu_counter_sum(&fs_info->ordered_bytes));
4779
4780         btrfs_sysfs_remove_mounted(fs_info);
4781         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4782
4783         btrfs_put_block_group_cache(fs_info);
4784
4785         /*
4786          * we must make sure there is not any read request to
4787          * submit after we stopping all workers.
4788          */
4789         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4790         btrfs_stop_all_workers(fs_info);
4791
4792         /* We shouldn't have any transaction open at this point */
4793         warn_about_uncommitted_trans(fs_info);
4794
4795         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4796         free_root_pointers(fs_info, true);
4797         btrfs_free_fs_roots(fs_info);
4798
4799         /*
4800          * We must free the block groups after dropping the fs_roots as we could
4801          * have had an IO error and have left over tree log blocks that aren't
4802          * cleaned up until the fs roots are freed.  This makes the block group
4803          * accounting appear to be wrong because there's pending reserved bytes,
4804          * so make sure we do the block group cleanup afterwards.
4805          */
4806         btrfs_free_block_groups(fs_info);
4807
4808         iput(fs_info->btree_inode);
4809
4810 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4811         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4812                 btrfsic_unmount(fs_info->fs_devices);
4813 #endif
4814
4815         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4816         btrfs_close_devices(fs_info->fs_devices);
4817 }
4818
4819 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4820                           int atomic)
4821 {
4822         int ret;
4823         struct inode *btree_inode = buf->pages[0]->mapping->host;
4824
4825         ret = extent_buffer_uptodate(buf);
4826         if (!ret)
4827                 return ret;
4828
4829         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4830                                     parent_transid, atomic);
4831         if (ret == -EAGAIN)
4832                 return ret;
4833         return !ret;
4834 }
4835
4836 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4837 {
4838         struct btrfs_fs_info *fs_info = buf->fs_info;
4839         u64 transid = btrfs_header_generation(buf);
4840         int was_dirty;
4841
4842 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4843         /*
4844          * This is a fast path so only do this check if we have sanity tests
4845          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4846          * outside of the sanity tests.
4847          */
4848         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4849                 return;
4850 #endif
4851         btrfs_assert_tree_write_locked(buf);
4852         if (transid != fs_info->generation)
4853                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4854                         buf->start, transid, fs_info->generation);
4855         was_dirty = set_extent_buffer_dirty(buf);
4856         if (!was_dirty)
4857                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4858                                          buf->len,
4859                                          fs_info->dirty_metadata_batch);
4860 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4861         /*
4862          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4863          * but item data not updated.
4864          * So here we should only check item pointers, not item data.
4865          */
4866         if (btrfs_header_level(buf) == 0 &&
4867             btrfs_check_leaf_relaxed(buf)) {
4868                 btrfs_print_leaf(buf);
4869                 ASSERT(0);
4870         }
4871 #endif
4872 }
4873
4874 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4875                                         int flush_delayed)
4876 {
4877         /*
4878          * looks as though older kernels can get into trouble with
4879          * this code, they end up stuck in balance_dirty_pages forever
4880          */
4881         int ret;
4882
4883         if (current->flags & PF_MEMALLOC)
4884                 return;
4885
4886         if (flush_delayed)
4887                 btrfs_balance_delayed_items(fs_info);
4888
4889         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4890                                      BTRFS_DIRTY_METADATA_THRESH,
4891                                      fs_info->dirty_metadata_batch);
4892         if (ret > 0) {
4893                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4894         }
4895 }
4896
4897 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4898 {
4899         __btrfs_btree_balance_dirty(fs_info, 1);
4900 }
4901
4902 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4903 {
4904         __btrfs_btree_balance_dirty(fs_info, 0);
4905 }
4906
4907 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4908 {
4909         /* cleanup FS via transaction */
4910         btrfs_cleanup_transaction(fs_info);
4911
4912         mutex_lock(&fs_info->cleaner_mutex);
4913         btrfs_run_delayed_iputs(fs_info);
4914         mutex_unlock(&fs_info->cleaner_mutex);
4915
4916         down_write(&fs_info->cleanup_work_sem);
4917         up_write(&fs_info->cleanup_work_sem);
4918 }
4919
4920 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4921 {
4922         struct btrfs_root *gang[8];
4923         u64 root_objectid = 0;
4924         int ret;
4925
4926         spin_lock(&fs_info->fs_roots_radix_lock);
4927         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4928                                              (void **)gang, root_objectid,
4929                                              ARRAY_SIZE(gang))) != 0) {
4930                 int i;
4931
4932                 for (i = 0; i < ret; i++)
4933                         gang[i] = btrfs_grab_root(gang[i]);
4934                 spin_unlock(&fs_info->fs_roots_radix_lock);
4935
4936                 for (i = 0; i < ret; i++) {
4937                         if (!gang[i])
4938                                 continue;
4939                         root_objectid = gang[i]->root_key.objectid;
4940                         btrfs_free_log(NULL, gang[i]);
4941                         btrfs_put_root(gang[i]);
4942                 }
4943                 root_objectid++;
4944                 spin_lock(&fs_info->fs_roots_radix_lock);
4945         }
4946         spin_unlock(&fs_info->fs_roots_radix_lock);
4947         btrfs_free_log_root_tree(NULL, fs_info);
4948 }
4949
4950 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4951 {
4952         struct btrfs_ordered_extent *ordered;
4953
4954         spin_lock(&root->ordered_extent_lock);
4955         /*
4956          * This will just short circuit the ordered completion stuff which will
4957          * make sure the ordered extent gets properly cleaned up.
4958          */
4959         list_for_each_entry(ordered, &root->ordered_extents,
4960                             root_extent_list)
4961                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4962         spin_unlock(&root->ordered_extent_lock);
4963 }
4964
4965 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4966 {
4967         struct btrfs_root *root;
4968         struct list_head splice;
4969
4970         INIT_LIST_HEAD(&splice);
4971
4972         spin_lock(&fs_info->ordered_root_lock);
4973         list_splice_init(&fs_info->ordered_roots, &splice);
4974         while (!list_empty(&splice)) {
4975                 root = list_first_entry(&splice, struct btrfs_root,
4976                                         ordered_root);
4977                 list_move_tail(&root->ordered_root,
4978                                &fs_info->ordered_roots);
4979
4980                 spin_unlock(&fs_info->ordered_root_lock);
4981                 btrfs_destroy_ordered_extents(root);
4982
4983                 cond_resched();
4984                 spin_lock(&fs_info->ordered_root_lock);
4985         }
4986         spin_unlock(&fs_info->ordered_root_lock);
4987
4988         /*
4989          * We need this here because if we've been flipped read-only we won't
4990          * get sync() from the umount, so we need to make sure any ordered
4991          * extents that haven't had their dirty pages IO start writeout yet
4992          * actually get run and error out properly.
4993          */
4994         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4995 }
4996
4997 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4998                                       struct btrfs_fs_info *fs_info)
4999 {
5000         struct rb_node *node;
5001         struct btrfs_delayed_ref_root *delayed_refs;
5002         struct btrfs_delayed_ref_node *ref;
5003         int ret = 0;
5004
5005         delayed_refs = &trans->delayed_refs;
5006
5007         spin_lock(&delayed_refs->lock);
5008         if (atomic_read(&delayed_refs->num_entries) == 0) {
5009                 spin_unlock(&delayed_refs->lock);
5010                 btrfs_debug(fs_info, "delayed_refs has NO entry");
5011                 return ret;
5012         }
5013
5014         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
5015                 struct btrfs_delayed_ref_head *head;
5016                 struct rb_node *n;
5017                 bool pin_bytes = false;
5018
5019                 head = rb_entry(node, struct btrfs_delayed_ref_head,
5020                                 href_node);
5021                 if (btrfs_delayed_ref_lock(delayed_refs, head))
5022                         continue;
5023
5024                 spin_lock(&head->lock);
5025                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
5026                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
5027                                        ref_node);
5028                         ref->in_tree = 0;
5029                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
5030                         RB_CLEAR_NODE(&ref->ref_node);
5031                         if (!list_empty(&ref->add_list))
5032                                 list_del(&ref->add_list);
5033                         atomic_dec(&delayed_refs->num_entries);
5034                         btrfs_put_delayed_ref(ref);
5035                 }
5036                 if (head->must_insert_reserved)
5037                         pin_bytes = true;
5038                 btrfs_free_delayed_extent_op(head->extent_op);
5039                 btrfs_delete_ref_head(delayed_refs, head);
5040                 spin_unlock(&head->lock);
5041                 spin_unlock(&delayed_refs->lock);
5042                 mutex_unlock(&head->mutex);
5043
5044                 if (pin_bytes) {
5045                         struct btrfs_block_group *cache;
5046
5047                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5048                         BUG_ON(!cache);
5049
5050                         spin_lock(&cache->space_info->lock);
5051                         spin_lock(&cache->lock);
5052                         cache->pinned += head->num_bytes;
5053                         btrfs_space_info_update_bytes_pinned(fs_info,
5054                                 cache->space_info, head->num_bytes);
5055                         cache->reserved -= head->num_bytes;
5056                         cache->space_info->bytes_reserved -= head->num_bytes;
5057                         spin_unlock(&cache->lock);
5058                         spin_unlock(&cache->space_info->lock);
5059
5060                         btrfs_put_block_group(cache);
5061
5062                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5063                                 head->bytenr + head->num_bytes - 1);
5064                 }
5065                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5066                 btrfs_put_delayed_ref_head(head);
5067                 cond_resched();
5068                 spin_lock(&delayed_refs->lock);
5069         }
5070         btrfs_qgroup_destroy_extent_records(trans);
5071
5072         spin_unlock(&delayed_refs->lock);
5073
5074         return ret;
5075 }
5076
5077 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5078 {
5079         struct btrfs_inode *btrfs_inode;
5080         struct list_head splice;
5081
5082         INIT_LIST_HEAD(&splice);
5083
5084         spin_lock(&root->delalloc_lock);
5085         list_splice_init(&root->delalloc_inodes, &splice);
5086
5087         while (!list_empty(&splice)) {
5088                 struct inode *inode = NULL;
5089                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5090                                                delalloc_inodes);
5091                 __btrfs_del_delalloc_inode(root, btrfs_inode);
5092                 spin_unlock(&root->delalloc_lock);
5093
5094                 /*
5095                  * Make sure we get a live inode and that it'll not disappear
5096                  * meanwhile.
5097                  */
5098                 inode = igrab(&btrfs_inode->vfs_inode);
5099                 if (inode) {
5100                         invalidate_inode_pages2(inode->i_mapping);
5101                         iput(inode);
5102                 }
5103                 spin_lock(&root->delalloc_lock);
5104         }
5105         spin_unlock(&root->delalloc_lock);
5106 }
5107
5108 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5109 {
5110         struct btrfs_root *root;
5111         struct list_head splice;
5112
5113         INIT_LIST_HEAD(&splice);
5114
5115         spin_lock(&fs_info->delalloc_root_lock);
5116         list_splice_init(&fs_info->delalloc_roots, &splice);
5117         while (!list_empty(&splice)) {
5118                 root = list_first_entry(&splice, struct btrfs_root,
5119                                          delalloc_root);
5120                 root = btrfs_grab_root(root);
5121                 BUG_ON(!root);
5122                 spin_unlock(&fs_info->delalloc_root_lock);
5123
5124                 btrfs_destroy_delalloc_inodes(root);
5125                 btrfs_put_root(root);
5126
5127                 spin_lock(&fs_info->delalloc_root_lock);
5128         }
5129         spin_unlock(&fs_info->delalloc_root_lock);
5130 }
5131
5132 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5133                                         struct extent_io_tree *dirty_pages,
5134                                         int mark)
5135 {
5136         int ret;
5137         struct extent_buffer *eb;
5138         u64 start = 0;
5139         u64 end;
5140
5141         while (1) {
5142                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5143                                             mark, NULL);
5144                 if (ret)
5145                         break;
5146
5147                 clear_extent_bits(dirty_pages, start, end, mark);
5148                 while (start <= end) {
5149                         eb = find_extent_buffer(fs_info, start);
5150                         start += fs_info->nodesize;
5151                         if (!eb)
5152                                 continue;
5153                         wait_on_extent_buffer_writeback(eb);
5154
5155                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5156                                                &eb->bflags))
5157                                 clear_extent_buffer_dirty(eb);
5158                         free_extent_buffer_stale(eb);
5159                 }
5160         }
5161
5162         return ret;
5163 }
5164
5165 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5166                                        struct extent_io_tree *unpin)
5167 {
5168         u64 start;
5169         u64 end;
5170         int ret;
5171
5172         while (1) {
5173                 struct extent_state *cached_state = NULL;
5174
5175                 /*
5176                  * The btrfs_finish_extent_commit() may get the same range as
5177                  * ours between find_first_extent_bit and clear_extent_dirty.
5178                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5179                  * the same extent range.
5180                  */
5181                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5182                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5183                                             EXTENT_DIRTY, &cached_state);
5184                 if (ret) {
5185                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5186                         break;
5187                 }
5188
5189                 clear_extent_dirty(unpin, start, end, &cached_state);
5190                 free_extent_state(cached_state);
5191                 btrfs_error_unpin_extent_range(fs_info, start, end);
5192                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5193                 cond_resched();
5194         }
5195
5196         return 0;
5197 }
5198
5199 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5200 {
5201         struct inode *inode;
5202
5203         inode = cache->io_ctl.inode;
5204         if (inode) {
5205                 invalidate_inode_pages2(inode->i_mapping);
5206                 BTRFS_I(inode)->generation = 0;
5207                 cache->io_ctl.inode = NULL;
5208                 iput(inode);
5209         }
5210         ASSERT(cache->io_ctl.pages == NULL);
5211         btrfs_put_block_group(cache);
5212 }
5213
5214 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5215                              struct btrfs_fs_info *fs_info)
5216 {
5217         struct btrfs_block_group *cache;
5218
5219         spin_lock(&cur_trans->dirty_bgs_lock);
5220         while (!list_empty(&cur_trans->dirty_bgs)) {
5221                 cache = list_first_entry(&cur_trans->dirty_bgs,
5222                                          struct btrfs_block_group,
5223                                          dirty_list);
5224
5225                 if (!list_empty(&cache->io_list)) {
5226                         spin_unlock(&cur_trans->dirty_bgs_lock);
5227                         list_del_init(&cache->io_list);
5228                         btrfs_cleanup_bg_io(cache);
5229                         spin_lock(&cur_trans->dirty_bgs_lock);
5230                 }
5231
5232                 list_del_init(&cache->dirty_list);
5233                 spin_lock(&cache->lock);
5234                 cache->disk_cache_state = BTRFS_DC_ERROR;
5235                 spin_unlock(&cache->lock);
5236
5237                 spin_unlock(&cur_trans->dirty_bgs_lock);
5238                 btrfs_put_block_group(cache);
5239                 btrfs_delayed_refs_rsv_release(fs_info, 1);
5240                 spin_lock(&cur_trans->dirty_bgs_lock);
5241         }
5242         spin_unlock(&cur_trans->dirty_bgs_lock);
5243
5244         /*
5245          * Refer to the definition of io_bgs member for details why it's safe
5246          * to use it without any locking
5247          */
5248         while (!list_empty(&cur_trans->io_bgs)) {
5249                 cache = list_first_entry(&cur_trans->io_bgs,
5250                                          struct btrfs_block_group,
5251                                          io_list);
5252
5253                 list_del_init(&cache->io_list);
5254                 spin_lock(&cache->lock);
5255                 cache->disk_cache_state = BTRFS_DC_ERROR;
5256                 spin_unlock(&cache->lock);
5257                 btrfs_cleanup_bg_io(cache);
5258         }
5259 }
5260
5261 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5262                                    struct btrfs_fs_info *fs_info)
5263 {
5264         struct btrfs_device *dev, *tmp;
5265
5266         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5267         ASSERT(list_empty(&cur_trans->dirty_bgs));
5268         ASSERT(list_empty(&cur_trans->io_bgs));
5269
5270         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5271                                  post_commit_list) {
5272                 list_del_init(&dev->post_commit_list);
5273         }
5274
5275         btrfs_destroy_delayed_refs(cur_trans, fs_info);
5276
5277         cur_trans->state = TRANS_STATE_COMMIT_START;
5278         wake_up(&fs_info->transaction_blocked_wait);
5279
5280         cur_trans->state = TRANS_STATE_UNBLOCKED;
5281         wake_up(&fs_info->transaction_wait);
5282
5283         btrfs_destroy_delayed_inodes(fs_info);
5284
5285         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5286                                      EXTENT_DIRTY);
5287         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5288
5289         btrfs_free_redirty_list(cur_trans);
5290
5291         cur_trans->state =TRANS_STATE_COMPLETED;
5292         wake_up(&cur_trans->commit_wait);
5293 }
5294
5295 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5296 {
5297         struct btrfs_transaction *t;
5298
5299         mutex_lock(&fs_info->transaction_kthread_mutex);
5300
5301         spin_lock(&fs_info->trans_lock);
5302         while (!list_empty(&fs_info->trans_list)) {
5303                 t = list_first_entry(&fs_info->trans_list,
5304                                      struct btrfs_transaction, list);
5305                 if (t->state >= TRANS_STATE_COMMIT_START) {
5306                         refcount_inc(&t->use_count);
5307                         spin_unlock(&fs_info->trans_lock);
5308                         btrfs_wait_for_commit(fs_info, t->transid);
5309                         btrfs_put_transaction(t);
5310                         spin_lock(&fs_info->trans_lock);
5311                         continue;
5312                 }
5313                 if (t == fs_info->running_transaction) {
5314                         t->state = TRANS_STATE_COMMIT_DOING;
5315                         spin_unlock(&fs_info->trans_lock);
5316                         /*
5317                          * We wait for 0 num_writers since we don't hold a trans
5318                          * handle open currently for this transaction.
5319                          */
5320                         wait_event(t->writer_wait,
5321                                    atomic_read(&t->num_writers) == 0);
5322                 } else {
5323                         spin_unlock(&fs_info->trans_lock);
5324                 }
5325                 btrfs_cleanup_one_transaction(t, fs_info);
5326
5327                 spin_lock(&fs_info->trans_lock);
5328                 if (t == fs_info->running_transaction)
5329                         fs_info->running_transaction = NULL;
5330                 list_del_init(&t->list);
5331                 spin_unlock(&fs_info->trans_lock);
5332
5333                 btrfs_put_transaction(t);
5334                 trace_btrfs_transaction_commit(fs_info);
5335                 spin_lock(&fs_info->trans_lock);
5336         }
5337         spin_unlock(&fs_info->trans_lock);
5338         btrfs_destroy_all_ordered_extents(fs_info);
5339         btrfs_destroy_delayed_inodes(fs_info);
5340         btrfs_assert_delayed_root_empty(fs_info);
5341         btrfs_destroy_all_delalloc_inodes(fs_info);
5342         btrfs_drop_all_logs(fs_info);
5343         mutex_unlock(&fs_info->transaction_kthread_mutex);
5344
5345         return 0;
5346 }
5347
5348 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5349 {
5350         struct btrfs_path *path;
5351         int ret;
5352         struct extent_buffer *l;
5353         struct btrfs_key search_key;
5354         struct btrfs_key found_key;
5355         int slot;
5356
5357         path = btrfs_alloc_path();
5358         if (!path)
5359                 return -ENOMEM;
5360
5361         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5362         search_key.type = -1;
5363         search_key.offset = (u64)-1;
5364         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5365         if (ret < 0)
5366                 goto error;
5367         BUG_ON(ret == 0); /* Corruption */
5368         if (path->slots[0] > 0) {
5369                 slot = path->slots[0] - 1;
5370                 l = path->nodes[0];
5371                 btrfs_item_key_to_cpu(l, &found_key, slot);
5372                 root->free_objectid = max_t(u64, found_key.objectid + 1,
5373                                             BTRFS_FIRST_FREE_OBJECTID);
5374         } else {
5375                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5376         }
5377         ret = 0;
5378 error:
5379         btrfs_free_path(path);
5380         return ret;
5381 }
5382
5383 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5384 {
5385         int ret;
5386         mutex_lock(&root->objectid_mutex);
5387
5388         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5389                 btrfs_warn(root->fs_info,
5390                            "the objectid of root %llu reaches its highest value",
5391                            root->root_key.objectid);
5392                 ret = -ENOSPC;
5393                 goto out;
5394         }
5395
5396         *objectid = root->free_objectid++;
5397         ret = 0;
5398 out:
5399         mutex_unlock(&root->objectid_mutex);
5400         return ret;
5401 }