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 <linux/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 "dev-replace.h"
  33 #include "raid56.h"
  34 #include "sysfs.h"
  35 #include "qgroup.h"
  36 #include "compression.h"
  37 #include "tree-checker.h"
  38 #include "ref-verify.h"
  39 #include "block-group.h"
  40 #include "discard.h"
  41 #include "space-info.h"
  42 #include "zoned.h"
  43 #include "subpage.h"
  44 #include "fs.h"
  45 #include "accessors.h"
  46 #include "extent-tree.h"
  47 #include "root-tree.h"
  48 #include "defrag.h"
  49 #include "uuid-tree.h"
  50 #include "relocation.h"
  51 #include "scrub.h"
  52 #include "super.h"
  53
  54 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  55                                  BTRFS_HEADER_FLAG_RELOC |\
  56                                  BTRFS_SUPER_FLAG_ERROR |\
  57                                  BTRFS_SUPER_FLAG_SEEDING |\
  58                                  BTRFS_SUPER_FLAG_METADUMP |\
  59                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  60
  61 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  62 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  63
  64 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  65 {
  66         if (fs_info->csum_shash)
  67                 crypto_free_shash(fs_info->csum_shash);
  68 }
  69
  70 /*
  71  * Compute the csum of a btree block and store the result to provided buffer.
  72  */
  73 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  74 {
  75         struct btrfs_fs_info *fs_info = buf->fs_info;
  76         int num_pages;
  77         u32 first_page_part;
  78         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  79         char *kaddr;
  80         int i;
  81
  82         shash->tfm = fs_info->csum_shash;
  83         crypto_shash_init(shash);
  84
  85         if (buf->addr) {
  86                 /* Pages are contiguous, handle them as a big one. */
  87                 kaddr = buf->addr;
  88                 first_page_part = fs_info->nodesize;
  89                 num_pages = 1;
  90         } else {
  91                 kaddr = folio_address(buf->folios[0]);
  92                 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  93                 num_pages = num_extent_pages(buf);
  94         }
  95
  96         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
  97                             first_page_part - BTRFS_CSUM_SIZE);
  98
  99         /*
 100          * Multiple single-page folios case would reach here.
 101          *
 102          * nodesize <= PAGE_SIZE and large folio all handled by above
 103          * crypto_shash_update() already.
 104          */
 105         for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
 106                 kaddr = folio_address(buf->folios[i]);
 107                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 108         }
 109         memset(result, 0, BTRFS_CSUM_SIZE);
 110         crypto_shash_final(shash, result);
 111 }
 112
 113 /*
 114  * we can't consider a given block up to date unless the transid of the
 115  * block matches the transid in the parent node's pointer.  This is how we
 116  * detect blocks that either didn't get written at all or got written
 117  * in the wrong place.
 118  */
 119 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
 120 {
 121         if (!extent_buffer_uptodate(eb))
 122                 return 0;
 123
 124         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 125                 return 1;
 126
 127         if (atomic)
 128                 return -EAGAIN;
 129
 130         if (!extent_buffer_uptodate(eb) ||
 131             btrfs_header_generation(eb) != parent_transid) {
 132                 btrfs_err_rl(eb->fs_info,
 133 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 134                         eb->start, eb->read_mirror,
 135                         parent_transid, btrfs_header_generation(eb));
 136                 clear_extent_buffer_uptodate(eb);
 137                 return 0;
 138         }
 139         return 1;
 140 }
 141
 142 static bool btrfs_supported_super_csum(u16 csum_type)
 143 {
 144         switch (csum_type) {
 145         case BTRFS_CSUM_TYPE_CRC32:
 146         case BTRFS_CSUM_TYPE_XXHASH:
 147         case BTRFS_CSUM_TYPE_SHA256:
 148         case BTRFS_CSUM_TYPE_BLAKE2:
 149                 return true;
 150         default:
 151                 return false;
 152         }
 153 }
 154
 155 /*
 156  * Return 0 if the superblock checksum type matches the checksum value of that
 157  * algorithm. Pass the raw disk superblock data.
 158  */
 159 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 160                            const struct btrfs_super_block *disk_sb)
 161 {
 162         char result[BTRFS_CSUM_SIZE];
 163         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 164
 165         shash->tfm = fs_info->csum_shash;
 166
 167         /*
 168          * The super_block structure does not span the whole
 169          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 170          * filled with zeros and is included in the checksum.
 171          */
 172         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 173                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 174
 175         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 176                 return 1;
 177
 178         return 0;
 179 }
 180
 181 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 182                                       int mirror_num)
 183 {
 184         struct btrfs_fs_info *fs_info = eb->fs_info;
 185         int num_folios = num_extent_folios(eb);
 186         int ret = 0;
 187
 188         if (sb_rdonly(fs_info->sb))
 189                 return -EROFS;
 190
 191         for (int i = 0; i < num_folios; i++) {
 192                 struct folio *folio = eb->folios[i];
 193                 u64 start = max_t(u64, eb->start, folio_pos(folio));
 194                 u64 end = min_t(u64, eb->start + eb->len,
 195                                 folio_pos(folio) + eb->folio_size);
 196                 u32 len = end - start;
 197
 198                 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
 199                                               start, folio, offset_in_folio(folio, start),
 200                                               mirror_num);
 201                 if (ret)
 202                         break;
 203         }
 204
 205         return ret;
 206 }
 207
 208 /*
 209  * helper to read a given tree block, doing retries as required when
 210  * the checksums don't match and we have alternate mirrors to try.
 211  *
 212  * @check:              expected tree parentness check, see the comments of the
 213  *                      structure for details.
 214  */
 215 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 216                              const struct btrfs_tree_parent_check *check)
 217 {
 218         struct btrfs_fs_info *fs_info = eb->fs_info;
 219         int failed = 0;
 220         int ret;
 221         int num_copies = 0;
 222         int mirror_num = 0;
 223         int failed_mirror = 0;
 224
 225         ASSERT(check);
 226
 227         while (1) {
 228                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 229                 ret = read_extent_buffer_pages(eb, mirror_num, check);
 230                 if (!ret)
 231                         break;
 232
 233                 num_copies = btrfs_num_copies(fs_info,
 234                                               eb->start, eb->len);
 235                 if (num_copies == 1)
 236                         break;
 237
 238                 if (!failed_mirror) {
 239                         failed = 1;
 240                         failed_mirror = eb->read_mirror;
 241                 }
 242
 243                 mirror_num++;
 244                 if (mirror_num == failed_mirror)
 245                         mirror_num++;
 246
 247                 if (mirror_num > num_copies)
 248                         break;
 249         }
 250
 251         if (failed && !ret && failed_mirror)
 252                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 253
 254         return ret;
 255 }
 256
 257 /*
 258  * Checksum a dirty tree block before IO.
 259  */
 260 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 261 {
 262         struct extent_buffer *eb = bbio->private;
 263         struct btrfs_fs_info *fs_info = eb->fs_info;
 264         u64 found_start = btrfs_header_bytenr(eb);
 265         u64 last_trans;
 266         u8 result[BTRFS_CSUM_SIZE];
 267         int ret;
 268
 269         /* Btree blocks are always contiguous on disk. */
 270         if (WARN_ON_ONCE(bbio->file_offset != eb->start))
 271                 return BLK_STS_IOERR;
 272         if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 273                 return BLK_STS_IOERR;
 274
 275         /*
 276          * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
 277          * checksum it but zero-out its content. This is done to preserve
 278          * ordering of I/O without unnecessarily writing out data.
 279          */
 280         if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
 281                 memzero_extent_buffer(eb, 0, eb->len);
 282                 return BLK_STS_OK;
 283         }
 284
 285         if (WARN_ON_ONCE(found_start != eb->start))
 286                 return BLK_STS_IOERR;
 287         if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
 288                                                eb->start, eb->len)))
 289                 return BLK_STS_IOERR;
 290
 291         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 292                                     offsetof(struct btrfs_header, fsid),
 293                                     BTRFS_FSID_SIZE) == 0);
 294         csum_tree_block(eb, result);
 295
 296         if (btrfs_header_level(eb))
 297                 ret = btrfs_check_node(eb);
 298         else
 299                 ret = btrfs_check_leaf(eb);
 300
 301         if (ret < 0)
 302                 goto error;
 303
 304         /*
 305          * Also check the generation, the eb reached here must be newer than
 306          * last committed. Or something seriously wrong happened.
 307          */
 308         last_trans = btrfs_get_last_trans_committed(fs_info);
 309         if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
 310                 ret = -EUCLEAN;
 311                 btrfs_err(fs_info,
 312                         "block=%llu bad generation, have %llu expect > %llu",
 313                           eb->start, btrfs_header_generation(eb), last_trans);
 314                 goto error;
 315         }
 316         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 317         return BLK_STS_OK;
 318
 319 error:
 320         btrfs_print_tree(eb, 0);
 321         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 322                   eb->start);
 323         /*
 324          * Be noisy if this is an extent buffer from a log tree. We don't abort
 325          * a transaction in case there's a bad log tree extent buffer, we just
 326          * fallback to a transaction commit. Still we want to know when there is
 327          * a bad log tree extent buffer, as that may signal a bug somewhere.
 328          */
 329         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 330                 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 331         return errno_to_blk_status(ret);
 332 }
 333
 334 static bool check_tree_block_fsid(struct extent_buffer *eb)
 335 {
 336         struct btrfs_fs_info *fs_info = eb->fs_info;
 337         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 338         u8 fsid[BTRFS_FSID_SIZE];
 339
 340         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 341                            BTRFS_FSID_SIZE);
 342
 343         /*
 344          * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
 345          * This is then overwritten by metadata_uuid if it is present in the
 346          * device_list_add(). The same true for a seed device as well. So use of
 347          * fs_devices::metadata_uuid is appropriate here.
 348          */
 349         if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
 350                 return false;
 351
 352         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 353                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 354                         return false;
 355
 356         return true;
 357 }
 358
 359 /* Do basic extent buffer checks at read time */
 360 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
 361                                  const struct btrfs_tree_parent_check *check)
 362 {
 363         struct btrfs_fs_info *fs_info = eb->fs_info;
 364         u64 found_start;
 365         const u32 csum_size = fs_info->csum_size;
 366         u8 found_level;
 367         u8 result[BTRFS_CSUM_SIZE];
 368         const u8 *header_csum;
 369         int ret = 0;
 370         const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
 371
 372         ASSERT(check);
 373
 374         found_start = btrfs_header_bytenr(eb);
 375         if (found_start != eb->start) {
 376                 btrfs_err_rl(fs_info,
 377                         "bad tree block start, mirror %u want %llu have %llu",
 378                              eb->read_mirror, eb->start, found_start);
 379                 ret = -EIO;
 380                 goto out;
 381         }
 382         if (check_tree_block_fsid(eb)) {
 383                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 384                              eb->start, eb->read_mirror);
 385                 ret = -EIO;
 386                 goto out;
 387         }
 388         found_level = btrfs_header_level(eb);
 389         if (found_level >= BTRFS_MAX_LEVEL) {
 390                 btrfs_err(fs_info,
 391                         "bad tree block level, mirror %u level %d on logical %llu",
 392                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 393                 ret = -EIO;
 394                 goto out;
 395         }
 396
 397         csum_tree_block(eb, result);
 398         header_csum = folio_address(eb->folios[0]) +
 399                 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
 400
 401         if (memcmp(result, header_csum, csum_size) != 0) {
 402                 btrfs_warn_rl(fs_info,
 403 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
 404                               eb->start, eb->read_mirror,
 405                               CSUM_FMT_VALUE(csum_size, header_csum),
 406                               CSUM_FMT_VALUE(csum_size, result),
 407                               btrfs_header_level(eb),
 408                               ignore_csum ? ", ignored" : "");
 409                 if (!ignore_csum) {
 410                         ret = -EUCLEAN;
 411                         goto out;
 412                 }
 413         }
 414
 415         if (found_level != check->level) {
 416                 btrfs_err(fs_info,
 417                 "level verify failed on logical %llu mirror %u wanted %u found %u",
 418                           eb->start, eb->read_mirror, check->level, found_level);
 419                 ret = -EIO;
 420                 goto out;
 421         }
 422         if (unlikely(check->transid &&
 423                      btrfs_header_generation(eb) != check->transid)) {
 424                 btrfs_err_rl(eb->fs_info,
 425 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 426                                 eb->start, eb->read_mirror, check->transid,
 427                                 btrfs_header_generation(eb));
 428                 ret = -EIO;
 429                 goto out;
 430         }
 431         if (check->has_first_key) {
 432                 const struct btrfs_key *expect_key = &check->first_key;
 433                 struct btrfs_key found_key;
 434
 435                 if (found_level)
 436                         btrfs_node_key_to_cpu(eb, &found_key, 0);
 437                 else
 438                         btrfs_item_key_to_cpu(eb, &found_key, 0);
 439                 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 440                         btrfs_err(fs_info,
 441 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 442                                   eb->start, check->transid,
 443                                   expect_key->objectid,
 444                                   expect_key->type, expect_key->offset,
 445                                   found_key.objectid, found_key.type,
 446                                   found_key.offset);
 447                         ret = -EUCLEAN;
 448                         goto out;
 449                 }
 450         }
 451         if (check->owner_root) {
 452                 ret = btrfs_check_eb_owner(eb, check->owner_root);
 453                 if (ret < 0)
 454                         goto out;
 455         }
 456
 457         /*
 458          * If this is a leaf block and it is corrupt, set the corrupt bit so
 459          * that we don't try and read the other copies of this block, just
 460          * return -EIO.
 461          */
 462         if (found_level == 0 && btrfs_check_leaf(eb)) {
 463                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 464                 ret = -EIO;
 465         }
 466
 467         if (found_level > 0 && btrfs_check_node(eb))
 468                 ret = -EIO;
 469
 470         if (ret)
 471                 btrfs_err(fs_info,
 472                 "read time tree block corruption detected on logical %llu mirror %u",
 473                           eb->start, eb->read_mirror);
 474 out:
 475         return ret;
 476 }
 477
 478 #ifdef CONFIG_MIGRATION
 479 static int btree_migrate_folio(struct address_space *mapping,
 480                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 481 {
 482         /*
 483          * we can't safely write a btree page from here,
 484          * we haven't done the locking hook
 485          */
 486         if (folio_test_dirty(src))
 487                 return -EAGAIN;
 488         /*
 489          * Buffers may be managed in a filesystem specific way.
 490          * We must have no buffers or drop them.
 491          */
 492         if (folio_get_private(src) &&
 493             !filemap_release_folio(src, GFP_KERNEL))
 494                 return -EAGAIN;
 495         return migrate_folio(mapping, dst, src, mode);
 496 }
 497 #else
 498 #define btree_migrate_folio NULL
 499 #endif
 500
 501 static int btree_writepages(struct address_space *mapping,
 502                             struct writeback_control *wbc)
 503 {
 504         int ret;
 505
 506         if (wbc->sync_mode == WB_SYNC_NONE) {
 507                 struct btrfs_fs_info *fs_info;
 508
 509                 if (wbc->for_kupdate)
 510                         return 0;
 511
 512                 fs_info = inode_to_fs_info(mapping->host);
 513                 /* this is a bit racy, but that's ok */
 514                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 515                                              BTRFS_DIRTY_METADATA_THRESH,
 516                                              fs_info->dirty_metadata_batch);
 517                 if (ret < 0)
 518                         return 0;
 519         }
 520         return btree_write_cache_pages(mapping, wbc);
 521 }
 522
 523 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 524 {
 525         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 526                 return false;
 527
 528         return try_release_extent_buffer(folio);
 529 }
 530
 531 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 532                                  size_t length)
 533 {
 534         struct extent_io_tree *tree;
 535
 536         tree = &folio_to_inode(folio)->io_tree;
 537         extent_invalidate_folio(tree, folio, offset);
 538         btree_release_folio(folio, GFP_NOFS);
 539         if (folio_get_private(folio)) {
 540                 btrfs_warn(folio_to_fs_info(folio),
 541                            "folio private not zero on folio %llu",
 542                            (unsigned long long)folio_pos(folio));
 543                 folio_detach_private(folio);
 544         }
 545 }
 546
 547 #ifdef DEBUG
 548 static bool btree_dirty_folio(struct address_space *mapping,
 549                 struct folio *folio)
 550 {
 551         struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
 552         struct btrfs_subpage_info *spi = fs_info->subpage_info;
 553         struct btrfs_subpage *subpage;
 554         struct extent_buffer *eb;
 555         int cur_bit = 0;
 556         u64 page_start = folio_pos(folio);
 557
 558         if (fs_info->sectorsize == PAGE_SIZE) {
 559                 eb = folio_get_private(folio);
 560                 BUG_ON(!eb);
 561                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 562                 BUG_ON(!atomic_read(&eb->refs));
 563                 btrfs_assert_tree_write_locked(eb);
 564                 return filemap_dirty_folio(mapping, folio);
 565         }
 566
 567         ASSERT(spi);
 568         subpage = folio_get_private(folio);
 569
 570         for (cur_bit = spi->dirty_offset;
 571              cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
 572              cur_bit++) {
 573                 unsigned long flags;
 574                 u64 cur;
 575
 576                 spin_lock_irqsave(&subpage->lock, flags);
 577                 if (!test_bit(cur_bit, subpage->bitmaps)) {
 578                         spin_unlock_irqrestore(&subpage->lock, flags);
 579                         continue;
 580                 }
 581                 spin_unlock_irqrestore(&subpage->lock, flags);
 582                 cur = page_start + cur_bit * fs_info->sectorsize;
 583
 584                 eb = find_extent_buffer(fs_info, cur);
 585                 ASSERT(eb);
 586                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 587                 ASSERT(atomic_read(&eb->refs));
 588                 btrfs_assert_tree_write_locked(eb);
 589                 free_extent_buffer(eb);
 590
 591                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
 592         }
 593         return filemap_dirty_folio(mapping, folio);
 594 }
 595 #else
 596 #define btree_dirty_folio filemap_dirty_folio
 597 #endif
 598
 599 static const struct address_space_operations btree_aops = {
 600         .writepages     = btree_writepages,
 601         .release_folio  = btree_release_folio,
 602         .invalidate_folio = btree_invalidate_folio,
 603         .migrate_folio  = btree_migrate_folio,
 604         .dirty_folio    = btree_dirty_folio,
 605 };
 606
 607 struct extent_buffer *btrfs_find_create_tree_block(
 608                                                 struct btrfs_fs_info *fs_info,
 609                                                 u64 bytenr, u64 owner_root,
 610                                                 int level)
 611 {
 612         if (btrfs_is_testing(fs_info))
 613                 return alloc_test_extent_buffer(fs_info, bytenr);
 614         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 615 }
 616
 617 /*
 618  * Read tree block at logical address @bytenr and do variant basic but critical
 619  * verification.
 620  *
 621  * @check:              expected tree parentness check, see comments of the
 622  *                      structure for details.
 623  */
 624 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 625                                       struct btrfs_tree_parent_check *check)
 626 {
 627         struct extent_buffer *buf = NULL;
 628         int ret;
 629
 630         ASSERT(check);
 631
 632         buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 633                                            check->level);
 634         if (IS_ERR(buf))
 635                 return buf;
 636
 637         ret = btrfs_read_extent_buffer(buf, check);
 638         if (ret) {
 639                 free_extent_buffer_stale(buf);
 640                 return ERR_PTR(ret);
 641         }
 642         return buf;
 643
 644 }
 645
 646 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 647                          u64 objectid)
 648 {
 649         bool dummy = btrfs_is_testing(fs_info);
 650
 651         memset(&root->root_key, 0, sizeof(root->root_key));
 652         memset(&root->root_item, 0, sizeof(root->root_item));
 653         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 654         root->fs_info = fs_info;
 655         root->root_key.objectid = objectid;
 656         root->node = NULL;
 657         root->commit_root = NULL;
 658         root->state = 0;
 659         RB_CLEAR_NODE(&root->rb_node);
 660
 661         btrfs_set_root_last_trans(root, 0);
 662         root->free_objectid = 0;
 663         root->nr_delalloc_inodes = 0;
 664         root->nr_ordered_extents = 0;
 665         xa_init(&root->inodes);
 666         xa_init(&root->delayed_nodes);
 667
 668         btrfs_init_root_block_rsv(root);
 669
 670         INIT_LIST_HEAD(&root->dirty_list);
 671         INIT_LIST_HEAD(&root->root_list);
 672         INIT_LIST_HEAD(&root->delalloc_inodes);
 673         INIT_LIST_HEAD(&root->delalloc_root);
 674         INIT_LIST_HEAD(&root->ordered_extents);
 675         INIT_LIST_HEAD(&root->ordered_root);
 676         INIT_LIST_HEAD(&root->reloc_dirty_list);
 677         spin_lock_init(&root->delalloc_lock);
 678         spin_lock_init(&root->ordered_extent_lock);
 679         spin_lock_init(&root->accounting_lock);
 680         spin_lock_init(&root->qgroup_meta_rsv_lock);
 681         mutex_init(&root->objectid_mutex);
 682         mutex_init(&root->log_mutex);
 683         mutex_init(&root->ordered_extent_mutex);
 684         mutex_init(&root->delalloc_mutex);
 685         init_waitqueue_head(&root->qgroup_flush_wait);
 686         init_waitqueue_head(&root->log_writer_wait);
 687         init_waitqueue_head(&root->log_commit_wait[0]);
 688         init_waitqueue_head(&root->log_commit_wait[1]);
 689         INIT_LIST_HEAD(&root->log_ctxs[0]);
 690         INIT_LIST_HEAD(&root->log_ctxs[1]);
 691         atomic_set(&root->log_commit[0], 0);
 692         atomic_set(&root->log_commit[1], 0);
 693         atomic_set(&root->log_writers, 0);
 694         atomic_set(&root->log_batch, 0);
 695         refcount_set(&root->refs, 1);
 696         atomic_set(&root->snapshot_force_cow, 0);
 697         atomic_set(&root->nr_swapfiles, 0);
 698         btrfs_set_root_log_transid(root, 0);
 699         root->log_transid_committed = -1;
 700         btrfs_set_root_last_log_commit(root, 0);
 701         root->anon_dev = 0;
 702         if (!dummy) {
 703                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
 704                                     IO_TREE_ROOT_DIRTY_LOG_PAGES);
 705                 extent_io_tree_init(fs_info, &root->log_csum_range,
 706                                     IO_TREE_LOG_CSUM_RANGE);
 707         }
 708
 709         spin_lock_init(&root->root_item_lock);
 710         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 711 #ifdef CONFIG_BTRFS_DEBUG
 712         INIT_LIST_HEAD(&root->leak_list);
 713         spin_lock(&fs_info->fs_roots_radix_lock);
 714         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 715         spin_unlock(&fs_info->fs_roots_radix_lock);
 716 #endif
 717 }
 718
 719 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 720                                            u64 objectid, gfp_t flags)
 721 {
 722         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
 723         if (root)
 724                 __setup_root(root, fs_info, objectid);
 725         return root;
 726 }
 727
 728 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 729 /* Should only be used by the testing infrastructure */
 730 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 731 {
 732         struct btrfs_root *root;
 733
 734         if (!fs_info)
 735                 return ERR_PTR(-EINVAL);
 736
 737         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 738         if (!root)
 739                 return ERR_PTR(-ENOMEM);
 740
 741         /* We don't use the stripesize in selftest, set it as sectorsize */
 742         root->alloc_bytenr = 0;
 743
 744         return root;
 745 }
 746 #endif
 747
 748 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 749 {
 750         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 751         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 752
 753         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 754 }
 755
 756 static int global_root_key_cmp(const void *k, const struct rb_node *node)
 757 {
 758         const struct btrfs_key *key = k;
 759         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 760
 761         return btrfs_comp_cpu_keys(key, &root->root_key);
 762 }
 763
 764 int btrfs_global_root_insert(struct btrfs_root *root)
 765 {
 766         struct btrfs_fs_info *fs_info = root->fs_info;
 767         struct rb_node *tmp;
 768         int ret = 0;
 769
 770         write_lock(&fs_info->global_root_lock);
 771         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
 772         write_unlock(&fs_info->global_root_lock);
 773
 774         if (tmp) {
 775                 ret = -EEXIST;
 776                 btrfs_warn(fs_info, "global root %llu %llu already exists",
 777                            btrfs_root_id(root), root->root_key.offset);
 778         }
 779         return ret;
 780 }
 781
 782 void btrfs_global_root_delete(struct btrfs_root *root)
 783 {
 784         struct btrfs_fs_info *fs_info = root->fs_info;
 785
 786         write_lock(&fs_info->global_root_lock);
 787         rb_erase(&root->rb_node, &fs_info->global_root_tree);
 788         write_unlock(&fs_info->global_root_lock);
 789 }
 790
 791 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 792                                      struct btrfs_key *key)
 793 {
 794         struct rb_node *node;
 795         struct btrfs_root *root = NULL;
 796
 797         read_lock(&fs_info->global_root_lock);
 798         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
 799         if (node)
 800                 root = container_of(node, struct btrfs_root, rb_node);
 801         read_unlock(&fs_info->global_root_lock);
 802
 803         return root;
 804 }
 805
 806 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
 807 {
 808         struct btrfs_block_group *block_group;
 809         u64 ret;
 810
 811         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
 812                 return 0;
 813
 814         if (bytenr)
 815                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
 816         else
 817                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
 818         ASSERT(block_group);
 819         if (!block_group)
 820                 return 0;
 821         ret = block_group->global_root_id;
 822         btrfs_put_block_group(block_group);
 823
 824         return ret;
 825 }
 826
 827 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 828 {
 829         struct btrfs_key key = {
 830                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
 831                 .type = BTRFS_ROOT_ITEM_KEY,
 832                 .offset = btrfs_global_root_id(fs_info, bytenr),
 833         };
 834
 835         return btrfs_global_root(fs_info, &key);
 836 }
 837
 838 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 839 {
 840         struct btrfs_key key = {
 841                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
 842                 .type = BTRFS_ROOT_ITEM_KEY,
 843                 .offset = btrfs_global_root_id(fs_info, bytenr),
 844         };
 845
 846         return btrfs_global_root(fs_info, &key);
 847 }
 848
 849 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 850                                      u64 objectid)
 851 {
 852         struct btrfs_fs_info *fs_info = trans->fs_info;
 853         struct extent_buffer *leaf;
 854         struct btrfs_root *tree_root = fs_info->tree_root;
 855         struct btrfs_root *root;
 856         struct btrfs_key key;
 857         unsigned int nofs_flag;
 858         int ret = 0;
 859
 860         /*
 861          * We're holding a transaction handle, so use a NOFS memory allocation
 862          * context to avoid deadlock if reclaim happens.
 863          */
 864         nofs_flag = memalloc_nofs_save();
 865         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
 866         memalloc_nofs_restore(nofs_flag);
 867         if (!root)
 868                 return ERR_PTR(-ENOMEM);
 869
 870         root->root_key.objectid = objectid;
 871         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 872         root->root_key.offset = 0;
 873
 874         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 875                                       0, BTRFS_NESTING_NORMAL);
 876         if (IS_ERR(leaf)) {
 877                 ret = PTR_ERR(leaf);
 878                 leaf = NULL;
 879                 goto fail;
 880         }
 881
 882         root->node = leaf;
 883         btrfs_mark_buffer_dirty(trans, leaf);
 884
 885         root->commit_root = btrfs_root_node(root);
 886         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 887
 888         btrfs_set_root_flags(&root->root_item, 0);
 889         btrfs_set_root_limit(&root->root_item, 0);
 890         btrfs_set_root_bytenr(&root->root_item, leaf->start);
 891         btrfs_set_root_generation(&root->root_item, trans->transid);
 892         btrfs_set_root_level(&root->root_item, 0);
 893         btrfs_set_root_refs(&root->root_item, 1);
 894         btrfs_set_root_used(&root->root_item, leaf->len);
 895         btrfs_set_root_last_snapshot(&root->root_item, 0);
 896         btrfs_set_root_dirid(&root->root_item, 0);
 897         if (is_fstree(objectid))
 898                 generate_random_guid(root->root_item.uuid);
 899         else
 900                 export_guid(root->root_item.uuid, &guid_null);
 901         btrfs_set_root_drop_level(&root->root_item, 0);
 902
 903         btrfs_tree_unlock(leaf);
 904
 905         key.objectid = objectid;
 906         key.type = BTRFS_ROOT_ITEM_KEY;
 907         key.offset = 0;
 908         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
 909         if (ret)
 910                 goto fail;
 911
 912         return root;
 913
 914 fail:
 915         btrfs_put_root(root);
 916
 917         return ERR_PTR(ret);
 918 }
 919
 920 static struct btrfs_root *alloc_log_tree(struct btrfs_fs_info *fs_info)
 921 {
 922         struct btrfs_root *root;
 923
 924         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
 925         if (!root)
 926                 return ERR_PTR(-ENOMEM);
 927
 928         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 929         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 930         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 931
 932         return root;
 933 }
 934
 935 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 936                               struct btrfs_root *root)
 937 {
 938         struct extent_buffer *leaf;
 939
 940         /*
 941          * DON'T set SHAREABLE bit for log trees.
 942          *
 943          * Log trees are not exposed to user space thus can't be snapshotted,
 944          * and they go away before a real commit is actually done.
 945          *
 946          * They do store pointers to file data extents, and those reference
 947          * counts still get updated (along with back refs to the log tree).
 948          */
 949
 950         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
 951                         NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
 952         if (IS_ERR(leaf))
 953                 return PTR_ERR(leaf);
 954
 955         root->node = leaf;
 956
 957         btrfs_mark_buffer_dirty(trans, root->node);
 958         btrfs_tree_unlock(root->node);
 959
 960         return 0;
 961 }
 962
 963 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 964                              struct btrfs_fs_info *fs_info)
 965 {
 966         struct btrfs_root *log_root;
 967
 968         log_root = alloc_log_tree(fs_info);
 969         if (IS_ERR(log_root))
 970                 return PTR_ERR(log_root);
 971
 972         if (!btrfs_is_zoned(fs_info)) {
 973                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
 974
 975                 if (ret) {
 976                         btrfs_put_root(log_root);
 977                         return ret;
 978                 }
 979         }
 980
 981         WARN_ON(fs_info->log_root_tree);
 982         fs_info->log_root_tree = log_root;
 983         return 0;
 984 }
 985
 986 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 987                        struct btrfs_root *root)
 988 {
 989         struct btrfs_fs_info *fs_info = root->fs_info;
 990         struct btrfs_root *log_root;
 991         struct btrfs_inode_item *inode_item;
 992         int ret;
 993
 994         log_root = alloc_log_tree(fs_info);
 995         if (IS_ERR(log_root))
 996                 return PTR_ERR(log_root);
 997
 998         ret = btrfs_alloc_log_tree_node(trans, log_root);
 999         if (ret) {
1000                 btrfs_put_root(log_root);
1001                 return ret;
1002         }
1003
1004         btrfs_set_root_last_trans(log_root, trans->transid);
1005         log_root->root_key.offset = btrfs_root_id(root);
1006
1007         inode_item = &log_root->root_item.inode;
1008         btrfs_set_stack_inode_generation(inode_item, 1);
1009         btrfs_set_stack_inode_size(inode_item, 3);
1010         btrfs_set_stack_inode_nlink(inode_item, 1);
1011         btrfs_set_stack_inode_nbytes(inode_item,
1012                                      fs_info->nodesize);
1013         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1014
1015         btrfs_set_root_node(&log_root->root_item, log_root->node);
1016
1017         WARN_ON(root->log_root);
1018         root->log_root = log_root;
1019         btrfs_set_root_log_transid(root, 0);
1020         root->log_transid_committed = -1;
1021         btrfs_set_root_last_log_commit(root, 0);
1022         return 0;
1023 }
1024
1025 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1026                                               struct btrfs_path *path,
1027                                               const struct btrfs_key *key)
1028 {
1029         struct btrfs_root *root;
1030         struct btrfs_tree_parent_check check = { 0 };
1031         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1032         u64 generation;
1033         int ret;
1034         int level;
1035
1036         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1037         if (!root)
1038                 return ERR_PTR(-ENOMEM);
1039
1040         ret = btrfs_find_root(tree_root, key, path,
1041                               &root->root_item, &root->root_key);
1042         if (ret) {
1043                 if (ret > 0)
1044                         ret = -ENOENT;
1045                 goto fail;
1046         }
1047
1048         generation = btrfs_root_generation(&root->root_item);
1049         level = btrfs_root_level(&root->root_item);
1050         check.level = level;
1051         check.transid = generation;
1052         check.owner_root = key->objectid;
1053         root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1054                                      &check);
1055         if (IS_ERR(root->node)) {
1056                 ret = PTR_ERR(root->node);
1057                 root->node = NULL;
1058                 goto fail;
1059         }
1060         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1061                 ret = -EIO;
1062                 goto fail;
1063         }
1064
1065         /*
1066          * For real fs, and not log/reloc trees, root owner must
1067          * match its root node owner
1068          */
1069         if (!btrfs_is_testing(fs_info) &&
1070             btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1071             btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1072             btrfs_root_id(root) != btrfs_header_owner(root->node)) {
1073                 btrfs_crit(fs_info,
1074 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1075                            btrfs_root_id(root), root->node->start,
1076                            btrfs_header_owner(root->node),
1077                            btrfs_root_id(root));
1078                 ret = -EUCLEAN;
1079                 goto fail;
1080         }
1081         root->commit_root = btrfs_root_node(root);
1082         return root;
1083 fail:
1084         btrfs_put_root(root);
1085         return ERR_PTR(ret);
1086 }
1087
1088 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1089                                         const struct btrfs_key *key)
1090 {
1091         struct btrfs_root *root;
1092         struct btrfs_path *path;
1093
1094         path = btrfs_alloc_path();
1095         if (!path)
1096                 return ERR_PTR(-ENOMEM);
1097         root = read_tree_root_path(tree_root, path, key);
1098         btrfs_free_path(path);
1099
1100         return root;
1101 }
1102
1103 /*
1104  * Initialize subvolume root in-memory structure
1105  *
1106  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1107  */
1108 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1109 {
1110         int ret;
1111
1112         btrfs_drew_lock_init(&root->snapshot_lock);
1113
1114         if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1115             !btrfs_is_data_reloc_root(root) &&
1116             is_fstree(btrfs_root_id(root))) {
1117                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1118                 btrfs_check_and_init_root_item(&root->root_item);
1119         }
1120
1121         /*
1122          * Don't assign anonymous block device to roots that are not exposed to
1123          * userspace, the id pool is limited to 1M
1124          */
1125         if (is_fstree(btrfs_root_id(root)) &&
1126             btrfs_root_refs(&root->root_item) > 0) {
1127                 if (!anon_dev) {
1128                         ret = get_anon_bdev(&root->anon_dev);
1129                         if (ret)
1130                                 goto fail;
1131                 } else {
1132                         root->anon_dev = anon_dev;
1133                 }
1134         }
1135
1136         mutex_lock(&root->objectid_mutex);
1137         ret = btrfs_init_root_free_objectid(root);
1138         if (ret) {
1139                 mutex_unlock(&root->objectid_mutex);
1140                 goto fail;
1141         }
1142
1143         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1144
1145         mutex_unlock(&root->objectid_mutex);
1146
1147         return 0;
1148 fail:
1149         /* The caller is responsible to call btrfs_free_fs_root */
1150         return ret;
1151 }
1152
1153 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1154                                                u64 root_id)
1155 {
1156         struct btrfs_root *root;
1157
1158         spin_lock(&fs_info->fs_roots_radix_lock);
1159         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1160                                  (unsigned long)root_id);
1161         root = btrfs_grab_root(root);
1162         spin_unlock(&fs_info->fs_roots_radix_lock);
1163         return root;
1164 }
1165
1166 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1167                                                 u64 objectid)
1168 {
1169         struct btrfs_key key = {
1170                 .objectid = objectid,
1171                 .type = BTRFS_ROOT_ITEM_KEY,
1172                 .offset = 0,
1173         };
1174
1175         switch (objectid) {
1176         case BTRFS_ROOT_TREE_OBJECTID:
1177                 return btrfs_grab_root(fs_info->tree_root);
1178         case BTRFS_EXTENT_TREE_OBJECTID:
1179                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1180         case BTRFS_CHUNK_TREE_OBJECTID:
1181                 return btrfs_grab_root(fs_info->chunk_root);
1182         case BTRFS_DEV_TREE_OBJECTID:
1183                 return btrfs_grab_root(fs_info->dev_root);
1184         case BTRFS_CSUM_TREE_OBJECTID:
1185                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1186         case BTRFS_QUOTA_TREE_OBJECTID:
1187                 return btrfs_grab_root(fs_info->quota_root);
1188         case BTRFS_UUID_TREE_OBJECTID:
1189                 return btrfs_grab_root(fs_info->uuid_root);
1190         case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1191                 return btrfs_grab_root(fs_info->block_group_root);
1192         case BTRFS_FREE_SPACE_TREE_OBJECTID:
1193                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1194         case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1195                 return btrfs_grab_root(fs_info->stripe_root);
1196         default:
1197                 return NULL;
1198         }
1199 }
1200
1201 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1202                          struct btrfs_root *root)
1203 {
1204         int ret;
1205
1206         ret = radix_tree_preload(GFP_NOFS);
1207         if (ret)
1208                 return ret;
1209
1210         spin_lock(&fs_info->fs_roots_radix_lock);
1211         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1212                                 (unsigned long)btrfs_root_id(root),
1213                                 root);
1214         if (ret == 0) {
1215                 btrfs_grab_root(root);
1216                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1217         }
1218         spin_unlock(&fs_info->fs_roots_radix_lock);
1219         radix_tree_preload_end();
1220
1221         return ret;
1222 }
1223
1224 void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1225 {
1226 #ifdef CONFIG_BTRFS_DEBUG
1227         struct btrfs_root *root;
1228
1229         while (!list_empty(&fs_info->allocated_roots)) {
1230                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1231
1232                 root = list_first_entry(&fs_info->allocated_roots,
1233                                         struct btrfs_root, leak_list);
1234                 btrfs_err(fs_info, "leaked root %s refcount %d",
1235                           btrfs_root_name(&root->root_key, buf),
1236                           refcount_read(&root->refs));
1237                 WARN_ON_ONCE(1);
1238                 while (refcount_read(&root->refs) > 1)
1239                         btrfs_put_root(root);
1240                 btrfs_put_root(root);
1241         }
1242 #endif
1243 }
1244
1245 static void free_global_roots(struct btrfs_fs_info *fs_info)
1246 {
1247         struct btrfs_root *root;
1248         struct rb_node *node;
1249
1250         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1251                 root = rb_entry(node, struct btrfs_root, rb_node);
1252                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1253                 btrfs_put_root(root);
1254         }
1255 }
1256
1257 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1258 {
1259         struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1260
1261         percpu_counter_destroy(&fs_info->stats_read_blocks);
1262         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1263         percpu_counter_destroy(&fs_info->delalloc_bytes);
1264         percpu_counter_destroy(&fs_info->ordered_bytes);
1265         if (percpu_counter_initialized(em_counter))
1266                 ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1267         percpu_counter_destroy(em_counter);
1268         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1269         btrfs_free_csum_hash(fs_info);
1270         btrfs_free_stripe_hash_table(fs_info);
1271         btrfs_free_ref_cache(fs_info);
1272         kfree(fs_info->balance_ctl);
1273         kfree(fs_info->delayed_root);
1274         free_global_roots(fs_info);
1275         btrfs_put_root(fs_info->tree_root);
1276         btrfs_put_root(fs_info->chunk_root);
1277         btrfs_put_root(fs_info->dev_root);
1278         btrfs_put_root(fs_info->quota_root);
1279         btrfs_put_root(fs_info->uuid_root);
1280         btrfs_put_root(fs_info->fs_root);
1281         btrfs_put_root(fs_info->data_reloc_root);
1282         btrfs_put_root(fs_info->block_group_root);
1283         btrfs_put_root(fs_info->stripe_root);
1284         btrfs_check_leaked_roots(fs_info);
1285         btrfs_extent_buffer_leak_debug_check(fs_info);
1286         kfree(fs_info->super_copy);
1287         kfree(fs_info->super_for_commit);
1288         kvfree(fs_info);
1289 }
1290
1291
1292 /*
1293  * Get an in-memory reference of a root structure.
1294  *
1295  * For essential trees like root/extent tree, we grab it from fs_info directly.
1296  * For subvolume trees, we check the cached filesystem roots first. If not
1297  * found, then read it from disk and add it to cached fs roots.
1298  *
1299  * Caller should release the root by calling btrfs_put_root() after the usage.
1300  *
1301  * NOTE: Reloc and log trees can't be read by this function as they share the
1302  *       same root objectid.
1303  *
1304  * @objectid:   root id
1305  * @anon_dev:   preallocated anonymous block device number for new roots,
1306  *              pass NULL for a new allocation.
1307  * @check_ref:  whether to check root item references, If true, return -ENOENT
1308  *              for orphan roots
1309  */
1310 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1311                                              u64 objectid, dev_t *anon_dev,
1312                                              bool check_ref)
1313 {
1314         struct btrfs_root *root;
1315         struct btrfs_path *path;
1316         struct btrfs_key key;
1317         int ret;
1318
1319         root = btrfs_get_global_root(fs_info, objectid);
1320         if (root)
1321                 return root;
1322
1323         /*
1324          * If we're called for non-subvolume trees, and above function didn't
1325          * find one, do not try to read it from disk.
1326          *
1327          * This is namely for free-space-tree and quota tree, which can change
1328          * at runtime and should only be grabbed from fs_info.
1329          */
1330         if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1331                 return ERR_PTR(-ENOENT);
1332 again:
1333         root = btrfs_lookup_fs_root(fs_info, objectid);
1334         if (root) {
1335                 /*
1336                  * Some other caller may have read out the newly inserted
1337                  * subvolume already (for things like backref walk etc).  Not
1338                  * that common but still possible.  In that case, we just need
1339                  * to free the anon_dev.
1340                  */
1341                 if (unlikely(anon_dev && *anon_dev)) {
1342                         free_anon_bdev(*anon_dev);
1343                         *anon_dev = 0;
1344                 }
1345
1346                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1347                         btrfs_put_root(root);
1348                         return ERR_PTR(-ENOENT);
1349                 }
1350                 return root;
1351         }
1352
1353         key.objectid = objectid;
1354         key.type = BTRFS_ROOT_ITEM_KEY;
1355         key.offset = (u64)-1;
1356         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1357         if (IS_ERR(root))
1358                 return root;
1359
1360         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1361                 ret = -ENOENT;
1362                 goto fail;
1363         }
1364
1365         ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1366         if (ret)
1367                 goto fail;
1368
1369         path = btrfs_alloc_path();
1370         if (!path) {
1371                 ret = -ENOMEM;
1372                 goto fail;
1373         }
1374         key.objectid = BTRFS_ORPHAN_OBJECTID;
1375         key.type = BTRFS_ORPHAN_ITEM_KEY;
1376         key.offset = objectid;
1377
1378         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1379         btrfs_free_path(path);
1380         if (ret < 0)
1381                 goto fail;
1382         if (ret == 0)
1383                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1384
1385         ret = btrfs_insert_fs_root(fs_info, root);
1386         if (ret) {
1387                 if (ret == -EEXIST) {
1388                         btrfs_put_root(root);
1389                         goto again;
1390                 }
1391                 goto fail;
1392         }
1393         return root;
1394 fail:
1395         /*
1396          * If our caller provided us an anonymous device, then it's his
1397          * responsibility to free it in case we fail. So we have to set our
1398          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1399          * and once again by our caller.
1400          */
1401         if (anon_dev && *anon_dev)
1402                 root->anon_dev = 0;
1403         btrfs_put_root(root);
1404         return ERR_PTR(ret);
1405 }
1406
1407 /*
1408  * Get in-memory reference of a root structure
1409  *
1410  * @objectid:   tree objectid
1411  * @check_ref:  if set, verify that the tree exists and the item has at least
1412  *              one reference
1413  */
1414 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1415                                      u64 objectid, bool check_ref)
1416 {
1417         return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1418 }
1419
1420 /*
1421  * Get in-memory reference of a root structure, created as new, optionally pass
1422  * the anonymous block device id
1423  *
1424  * @objectid:   tree objectid
1425  * @anon_dev:   if NULL, allocate a new anonymous block device or use the
1426  *              parameter value if not NULL
1427  */
1428 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1429                                          u64 objectid, dev_t *anon_dev)
1430 {
1431         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1432 }
1433
1434 /*
1435  * Return a root for the given objectid.
1436  *
1437  * @fs_info:    the fs_info
1438  * @objectid:   the objectid we need to lookup
1439  *
1440  * This is exclusively used for backref walking, and exists specifically because
1441  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1442  * creation time, which means we may have to read the tree_root in order to look
1443  * up a fs root that is not in memory.  If the root is not in memory we will
1444  * read the tree root commit root and look up the fs root from there.  This is a
1445  * temporary root, it will not be inserted into the radix tree as it doesn't
1446  * have the most uptodate information, it'll simply be discarded once the
1447  * backref code is finished using the root.
1448  */
1449 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1450                                                  struct btrfs_path *path,
1451                                                  u64 objectid)
1452 {
1453         struct btrfs_root *root;
1454         struct btrfs_key key;
1455
1456         ASSERT(path->search_commit_root && path->skip_locking);
1457
1458         /*
1459          * This can return -ENOENT if we ask for a root that doesn't exist, but
1460          * since this is called via the backref walking code we won't be looking
1461          * up a root that doesn't exist, unless there's corruption.  So if root
1462          * != NULL just return it.
1463          */
1464         root = btrfs_get_global_root(fs_info, objectid);
1465         if (root)
1466                 return root;
1467
1468         root = btrfs_lookup_fs_root(fs_info, objectid);
1469         if (root)
1470                 return root;
1471
1472         key.objectid = objectid;
1473         key.type = BTRFS_ROOT_ITEM_KEY;
1474         key.offset = (u64)-1;
1475         root = read_tree_root_path(fs_info->tree_root, path, &key);
1476         btrfs_release_path(path);
1477
1478         return root;
1479 }
1480
1481 static int cleaner_kthread(void *arg)
1482 {
1483         struct btrfs_fs_info *fs_info = arg;
1484         int again;
1485
1486         while (1) {
1487                 again = 0;
1488
1489                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1490
1491                 /* Make the cleaner go to sleep early. */
1492                 if (btrfs_need_cleaner_sleep(fs_info))
1493                         goto sleep;
1494
1495                 /*
1496                  * Do not do anything if we might cause open_ctree() to block
1497                  * before we have finished mounting the filesystem.
1498                  */
1499                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1500                         goto sleep;
1501
1502                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1503                         goto sleep;
1504
1505                 /*
1506                  * Avoid the problem that we change the status of the fs
1507                  * during the above check and trylock.
1508                  */
1509                 if (btrfs_need_cleaner_sleep(fs_info)) {
1510                         mutex_unlock(&fs_info->cleaner_mutex);
1511                         goto sleep;
1512                 }
1513
1514                 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1515                         btrfs_sysfs_feature_update(fs_info);
1516
1517                 btrfs_run_delayed_iputs(fs_info);
1518
1519                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1520                 mutex_unlock(&fs_info->cleaner_mutex);
1521
1522                 /*
1523                  * The defragger has dealt with the R/O remount and umount,
1524                  * needn't do anything special here.
1525                  */
1526                 btrfs_run_defrag_inodes(fs_info);
1527
1528                 /*
1529                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1530                  * with relocation (btrfs_relocate_chunk) and relocation
1531                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1532                  * after acquiring fs_info->reclaim_bgs_lock. So we
1533                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1534                  * unused block groups.
1535                  */
1536                 btrfs_delete_unused_bgs(fs_info);
1537
1538                 /*
1539                  * Reclaim block groups in the reclaim_bgs list after we deleted
1540                  * all unused block_groups. This possibly gives us some more free
1541                  * space.
1542                  */
1543                 btrfs_reclaim_bgs(fs_info);
1544 sleep:
1545                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1546                 if (kthread_should_park())
1547                         kthread_parkme();
1548                 if (kthread_should_stop())
1549                         return 0;
1550                 if (!again) {
1551                         set_current_state(TASK_INTERRUPTIBLE);
1552                         schedule();
1553                         __set_current_state(TASK_RUNNING);
1554                 }
1555         }
1556 }
1557
1558 static int transaction_kthread(void *arg)
1559 {
1560         struct btrfs_root *root = arg;
1561         struct btrfs_fs_info *fs_info = root->fs_info;
1562         struct btrfs_trans_handle *trans;
1563         struct btrfs_transaction *cur;
1564         u64 transid;
1565         time64_t delta;
1566         unsigned long delay;
1567         bool cannot_commit;
1568
1569         do {
1570                 cannot_commit = false;
1571                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1572                 mutex_lock(&fs_info->transaction_kthread_mutex);
1573
1574                 spin_lock(&fs_info->trans_lock);
1575                 cur = fs_info->running_transaction;
1576                 if (!cur) {
1577                         spin_unlock(&fs_info->trans_lock);
1578                         goto sleep;
1579                 }
1580
1581                 delta = ktime_get_seconds() - cur->start_time;
1582                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1583                     cur->state < TRANS_STATE_COMMIT_PREP &&
1584                     delta < fs_info->commit_interval) {
1585                         spin_unlock(&fs_info->trans_lock);
1586                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1587                         delay = min(delay,
1588                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1589                         goto sleep;
1590                 }
1591                 transid = cur->transid;
1592                 spin_unlock(&fs_info->trans_lock);
1593
1594                 /* If the file system is aborted, this will always fail. */
1595                 trans = btrfs_attach_transaction(root);
1596                 if (IS_ERR(trans)) {
1597                         if (PTR_ERR(trans) != -ENOENT)
1598                                 cannot_commit = true;
1599                         goto sleep;
1600                 }
1601                 if (transid == trans->transid) {
1602                         btrfs_commit_transaction(trans);
1603                 } else {
1604                         btrfs_end_transaction(trans);
1605                 }
1606 sleep:
1607                 wake_up_process(fs_info->cleaner_kthread);
1608                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1609
1610                 if (BTRFS_FS_ERROR(fs_info))
1611                         btrfs_cleanup_transaction(fs_info);
1612                 if (!kthread_should_stop() &&
1613                                 (!btrfs_transaction_blocked(fs_info) ||
1614                                  cannot_commit))
1615                         schedule_timeout_interruptible(delay);
1616         } while (!kthread_should_stop());
1617         return 0;
1618 }
1619
1620 /*
1621  * This will find the highest generation in the array of root backups.  The
1622  * index of the highest array is returned, or -EINVAL if we can't find
1623  * anything.
1624  *
1625  * We check to make sure the array is valid by comparing the
1626  * generation of the latest  root in the array with the generation
1627  * in the super block.  If they don't match we pitch it.
1628  */
1629 static int find_newest_super_backup(struct btrfs_fs_info *info)
1630 {
1631         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1632         u64 cur;
1633         struct btrfs_root_backup *root_backup;
1634         int i;
1635
1636         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1637                 root_backup = info->super_copy->super_roots + i;
1638                 cur = btrfs_backup_tree_root_gen(root_backup);
1639                 if (cur == newest_gen)
1640                         return i;
1641         }
1642
1643         return -EINVAL;
1644 }
1645
1646 /*
1647  * copy all the root pointers into the super backup array.
1648  * this will bump the backup pointer by one when it is
1649  * done
1650  */
1651 static void backup_super_roots(struct btrfs_fs_info *info)
1652 {
1653         const int next_backup = info->backup_root_index;
1654         struct btrfs_root_backup *root_backup;
1655
1656         root_backup = info->super_for_commit->super_roots + next_backup;
1657
1658         /*
1659          * make sure all of our padding and empty slots get zero filled
1660          * regardless of which ones we use today
1661          */
1662         memset(root_backup, 0, sizeof(*root_backup));
1663
1664         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1665
1666         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1667         btrfs_set_backup_tree_root_gen(root_backup,
1668                                btrfs_header_generation(info->tree_root->node));
1669
1670         btrfs_set_backup_tree_root_level(root_backup,
1671                                btrfs_header_level(info->tree_root->node));
1672
1673         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1674         btrfs_set_backup_chunk_root_gen(root_backup,
1675                                btrfs_header_generation(info->chunk_root->node));
1676         btrfs_set_backup_chunk_root_level(root_backup,
1677                                btrfs_header_level(info->chunk_root->node));
1678
1679         if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1680                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1681                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1682
1683                 btrfs_set_backup_extent_root(root_backup,
1684                                              extent_root->node->start);
1685                 btrfs_set_backup_extent_root_gen(root_backup,
1686                                 btrfs_header_generation(extent_root->node));
1687                 btrfs_set_backup_extent_root_level(root_backup,
1688                                         btrfs_header_level(extent_root->node));
1689
1690                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1691                 btrfs_set_backup_csum_root_gen(root_backup,
1692                                                btrfs_header_generation(csum_root->node));
1693                 btrfs_set_backup_csum_root_level(root_backup,
1694                                                  btrfs_header_level(csum_root->node));
1695         }
1696
1697         /*
1698          * we might commit during log recovery, which happens before we set
1699          * the fs_root.  Make sure it is valid before we fill it in.
1700          */
1701         if (info->fs_root && info->fs_root->node) {
1702                 btrfs_set_backup_fs_root(root_backup,
1703                                          info->fs_root->node->start);
1704                 btrfs_set_backup_fs_root_gen(root_backup,
1705                                btrfs_header_generation(info->fs_root->node));
1706                 btrfs_set_backup_fs_root_level(root_backup,
1707                                btrfs_header_level(info->fs_root->node));
1708         }
1709
1710         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1711         btrfs_set_backup_dev_root_gen(root_backup,
1712                                btrfs_header_generation(info->dev_root->node));
1713         btrfs_set_backup_dev_root_level(root_backup,
1714                                        btrfs_header_level(info->dev_root->node));
1715
1716         btrfs_set_backup_total_bytes(root_backup,
1717                              btrfs_super_total_bytes(info->super_copy));
1718         btrfs_set_backup_bytes_used(root_backup,
1719                              btrfs_super_bytes_used(info->super_copy));
1720         btrfs_set_backup_num_devices(root_backup,
1721                              btrfs_super_num_devices(info->super_copy));
1722
1723         /*
1724          * if we don't copy this out to the super_copy, it won't get remembered
1725          * for the next commit
1726          */
1727         memcpy(&info->super_copy->super_roots,
1728                &info->super_for_commit->super_roots,
1729                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1730 }
1731
1732 /*
1733  * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1734  * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1735  *
1736  * @fs_info:  filesystem whose backup roots need to be read
1737  * @priority: priority of backup root required
1738  *
1739  * Returns backup root index on success and -EINVAL otherwise.
1740  */
1741 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1742 {
1743         int backup_index = find_newest_super_backup(fs_info);
1744         struct btrfs_super_block *super = fs_info->super_copy;
1745         struct btrfs_root_backup *root_backup;
1746
1747         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1748                 if (priority == 0)
1749                         return backup_index;
1750
1751                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1752                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1753         } else {
1754                 return -EINVAL;
1755         }
1756
1757         root_backup = super->super_roots + backup_index;
1758
1759         btrfs_set_super_generation(super,
1760                                    btrfs_backup_tree_root_gen(root_backup));
1761         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1762         btrfs_set_super_root_level(super,
1763                                    btrfs_backup_tree_root_level(root_backup));
1764         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1765
1766         /*
1767          * Fixme: the total bytes and num_devices need to match or we should
1768          * need a fsck
1769          */
1770         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1771         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1772
1773         return backup_index;
1774 }
1775
1776 /* helper to cleanup workers */
1777 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1778 {
1779         btrfs_destroy_workqueue(fs_info->fixup_workers);
1780         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1781         btrfs_destroy_workqueue(fs_info->workers);
1782         if (fs_info->endio_workers)
1783                 destroy_workqueue(fs_info->endio_workers);
1784         if (fs_info->rmw_workers)
1785                 destroy_workqueue(fs_info->rmw_workers);
1786         if (fs_info->compressed_write_workers)
1787                 destroy_workqueue(fs_info->compressed_write_workers);
1788         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1789         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1790         btrfs_destroy_workqueue(fs_info->delayed_workers);
1791         btrfs_destroy_workqueue(fs_info->caching_workers);
1792         btrfs_destroy_workqueue(fs_info->flush_workers);
1793         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1794         if (fs_info->discard_ctl.discard_workers)
1795                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1796         /*
1797          * Now that all other work queues are destroyed, we can safely destroy
1798          * the queues used for metadata I/O, since tasks from those other work
1799          * queues can do metadata I/O operations.
1800          */
1801         if (fs_info->endio_meta_workers)
1802                 destroy_workqueue(fs_info->endio_meta_workers);
1803 }
1804
1805 static void free_root_extent_buffers(struct btrfs_root *root)
1806 {
1807         if (root) {
1808                 free_extent_buffer(root->node);
1809                 free_extent_buffer(root->commit_root);
1810                 root->node = NULL;
1811                 root->commit_root = NULL;
1812         }
1813 }
1814
1815 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1816 {
1817         struct btrfs_root *root, *tmp;
1818
1819         rbtree_postorder_for_each_entry_safe(root, tmp,
1820                                              &fs_info->global_root_tree,
1821                                              rb_node)
1822                 free_root_extent_buffers(root);
1823 }
1824
1825 /* helper to cleanup tree roots */
1826 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1827 {
1828         free_root_extent_buffers(info->tree_root);
1829
1830         free_global_root_pointers(info);
1831         free_root_extent_buffers(info->dev_root);
1832         free_root_extent_buffers(info->quota_root);
1833         free_root_extent_buffers(info->uuid_root);
1834         free_root_extent_buffers(info->fs_root);
1835         free_root_extent_buffers(info->data_reloc_root);
1836         free_root_extent_buffers(info->block_group_root);
1837         free_root_extent_buffers(info->stripe_root);
1838         if (free_chunk_root)
1839                 free_root_extent_buffers(info->chunk_root);
1840 }
1841
1842 void btrfs_put_root(struct btrfs_root *root)
1843 {
1844         if (!root)
1845                 return;
1846
1847         if (refcount_dec_and_test(&root->refs)) {
1848                 if (WARN_ON(!xa_empty(&root->inodes)))
1849                         xa_destroy(&root->inodes);
1850                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1851                 if (root->anon_dev)
1852                         free_anon_bdev(root->anon_dev);
1853                 free_root_extent_buffers(root);
1854 #ifdef CONFIG_BTRFS_DEBUG
1855                 spin_lock(&root->fs_info->fs_roots_radix_lock);
1856                 list_del_init(&root->leak_list);
1857                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1858 #endif
1859                 kfree(root);
1860         }
1861 }
1862
1863 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1864 {
1865         int ret;
1866         struct btrfs_root *gang[8];
1867         int i;
1868
1869         while (!list_empty(&fs_info->dead_roots)) {
1870                 gang[0] = list_entry(fs_info->dead_roots.next,
1871                                      struct btrfs_root, root_list);
1872                 list_del(&gang[0]->root_list);
1873
1874                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1875                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1876                 btrfs_put_root(gang[0]);
1877         }
1878
1879         while (1) {
1880                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1881                                              (void **)gang, 0,
1882                                              ARRAY_SIZE(gang));
1883                 if (!ret)
1884                         break;
1885                 for (i = 0; i < ret; i++)
1886                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1887         }
1888 }
1889
1890 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1891 {
1892         mutex_init(&fs_info->scrub_lock);
1893         atomic_set(&fs_info->scrubs_running, 0);
1894         atomic_set(&fs_info->scrub_pause_req, 0);
1895         atomic_set(&fs_info->scrubs_paused, 0);
1896         atomic_set(&fs_info->scrub_cancel_req, 0);
1897         init_waitqueue_head(&fs_info->scrub_pause_wait);
1898         refcount_set(&fs_info->scrub_workers_refcnt, 0);
1899 }
1900
1901 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1902 {
1903         spin_lock_init(&fs_info->balance_lock);
1904         mutex_init(&fs_info->balance_mutex);
1905         atomic_set(&fs_info->balance_pause_req, 0);
1906         atomic_set(&fs_info->balance_cancel_req, 0);
1907         fs_info->balance_ctl = NULL;
1908         init_waitqueue_head(&fs_info->balance_wait_q);
1909         atomic_set(&fs_info->reloc_cancel_req, 0);
1910 }
1911
1912 static int btrfs_init_btree_inode(struct super_block *sb)
1913 {
1914         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1915         unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1916                                               fs_info->tree_root);
1917         struct inode *inode;
1918
1919         inode = new_inode(sb);
1920         if (!inode)
1921                 return -ENOMEM;
1922
1923         btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1924         set_nlink(inode, 1);
1925         /*
1926          * we set the i_size on the btree inode to the max possible int.
1927          * the real end of the address space is determined by all of
1928          * the devices in the system
1929          */
1930         inode->i_size = OFFSET_MAX;
1931         inode->i_mapping->a_ops = &btree_aops;
1932         mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1933
1934         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1935                             IO_TREE_BTREE_INODE_IO);
1936         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1937
1938         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1939         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1940         __insert_inode_hash(inode, hash);
1941         fs_info->btree_inode = inode;
1942
1943         return 0;
1944 }
1945
1946 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1947 {
1948         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1949         init_rwsem(&fs_info->dev_replace.rwsem);
1950         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1951 }
1952
1953 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1954 {
1955         spin_lock_init(&fs_info->qgroup_lock);
1956         mutex_init(&fs_info->qgroup_ioctl_lock);
1957         fs_info->qgroup_tree = RB_ROOT;
1958         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1959         fs_info->qgroup_seq = 1;
1960         fs_info->qgroup_ulist = NULL;
1961         fs_info->qgroup_rescan_running = false;
1962         fs_info->qgroup_drop_subtree_thres = BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT;
1963         mutex_init(&fs_info->qgroup_rescan_lock);
1964 }
1965
1966 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1967 {
1968         u32 max_active = fs_info->thread_pool_size;
1969         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1970         unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1971
1972         fs_info->workers =
1973                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1974
1975         fs_info->delalloc_workers =
1976                 btrfs_alloc_workqueue(fs_info, "delalloc",
1977                                       flags, max_active, 2);
1978
1979         fs_info->flush_workers =
1980                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1981                                       flags, max_active, 0);
1982
1983         fs_info->caching_workers =
1984                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1985
1986         fs_info->fixup_workers =
1987                 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1988
1989         fs_info->endio_workers =
1990                 alloc_workqueue("btrfs-endio", flags, max_active);
1991         fs_info->endio_meta_workers =
1992                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1993         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1994         fs_info->endio_write_workers =
1995                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1996                                       max_active, 2);
1997         fs_info->compressed_write_workers =
1998                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1999         fs_info->endio_freespace_worker =
2000                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2001                                       max_active, 0);
2002         fs_info->delayed_workers =
2003                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2004                                       max_active, 0);
2005         fs_info->qgroup_rescan_workers =
2006                 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2007                                               ordered_flags);
2008         fs_info->discard_ctl.discard_workers =
2009                 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2010
2011         if (!(fs_info->workers &&
2012               fs_info->delalloc_workers && fs_info->flush_workers &&
2013               fs_info->endio_workers && fs_info->endio_meta_workers &&
2014               fs_info->compressed_write_workers &&
2015               fs_info->endio_write_workers &&
2016               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2017               fs_info->caching_workers && fs_info->fixup_workers &&
2018               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2019               fs_info->discard_ctl.discard_workers)) {
2020                 return -ENOMEM;
2021         }
2022
2023         return 0;
2024 }
2025
2026 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2027 {
2028         struct crypto_shash *csum_shash;
2029         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2030
2031         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2032
2033         if (IS_ERR(csum_shash)) {
2034                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2035                           csum_driver);
2036                 return PTR_ERR(csum_shash);
2037         }
2038
2039         fs_info->csum_shash = csum_shash;
2040
2041         /*
2042          * Check if the checksum implementation is a fast accelerated one.
2043          * As-is this is a bit of a hack and should be replaced once the csum
2044          * implementations provide that information themselves.
2045          */
2046         switch (csum_type) {
2047         case BTRFS_CSUM_TYPE_CRC32:
2048                 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2049                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2050                 break;
2051         case BTRFS_CSUM_TYPE_XXHASH:
2052                 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2053                 break;
2054         default:
2055                 break;
2056         }
2057
2058         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2059                         btrfs_super_csum_name(csum_type),
2060                         crypto_shash_driver_name(csum_shash));
2061         return 0;
2062 }
2063
2064 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2065                             struct btrfs_fs_devices *fs_devices)
2066 {
2067         int ret;
2068         struct btrfs_tree_parent_check check = { 0 };
2069         struct btrfs_root *log_tree_root;
2070         struct btrfs_super_block *disk_super = fs_info->super_copy;
2071         u64 bytenr = btrfs_super_log_root(disk_super);
2072         int level = btrfs_super_log_root_level(disk_super);
2073
2074         if (fs_devices->rw_devices == 0) {
2075                 btrfs_warn(fs_info, "log replay required on RO media");
2076                 return -EIO;
2077         }
2078
2079         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2080                                          GFP_KERNEL);
2081         if (!log_tree_root)
2082                 return -ENOMEM;
2083
2084         check.level = level;
2085         check.transid = fs_info->generation + 1;
2086         check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2087         log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2088         if (IS_ERR(log_tree_root->node)) {
2089                 btrfs_warn(fs_info, "failed to read log tree");
2090                 ret = PTR_ERR(log_tree_root->node);
2091                 log_tree_root->node = NULL;
2092                 btrfs_put_root(log_tree_root);
2093                 return ret;
2094         }
2095         if (!extent_buffer_uptodate(log_tree_root->node)) {
2096                 btrfs_err(fs_info, "failed to read log tree");
2097                 btrfs_put_root(log_tree_root);
2098                 return -EIO;
2099         }
2100
2101         /* returns with log_tree_root freed on success */
2102         ret = btrfs_recover_log_trees(log_tree_root);
2103         if (ret) {
2104                 btrfs_handle_fs_error(fs_info, ret,
2105                                       "Failed to recover log tree");
2106                 btrfs_put_root(log_tree_root);
2107                 return ret;
2108         }
2109
2110         if (sb_rdonly(fs_info->sb)) {
2111                 ret = btrfs_commit_super(fs_info);
2112                 if (ret)
2113                         return ret;
2114         }
2115
2116         return 0;
2117 }
2118
2119 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2120                                       struct btrfs_path *path, u64 objectid,
2121                                       const char *name)
2122 {
2123         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2124         struct btrfs_root *root;
2125         u64 max_global_id = 0;
2126         int ret;
2127         struct btrfs_key key = {
2128                 .objectid = objectid,
2129                 .type = BTRFS_ROOT_ITEM_KEY,
2130                 .offset = 0,
2131         };
2132         bool found = false;
2133
2134         /* If we have IGNOREDATACSUMS skip loading these roots. */
2135         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2136             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2137                 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2138                 return 0;
2139         }
2140
2141         while (1) {
2142                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2143                 if (ret < 0)
2144                         break;
2145
2146                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2147                         ret = btrfs_next_leaf(tree_root, path);
2148                         if (ret) {
2149                                 if (ret > 0)
2150                                         ret = 0;
2151                                 break;
2152                         }
2153                 }
2154                 ret = 0;
2155
2156                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2157                 if (key.objectid != objectid)
2158                         break;
2159                 btrfs_release_path(path);
2160
2161                 /*
2162                  * Just worry about this for extent tree, it'll be the same for
2163                  * everybody.
2164                  */
2165                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2166                         max_global_id = max(max_global_id, key.offset);
2167
2168                 found = true;
2169                 root = read_tree_root_path(tree_root, path, &key);
2170                 if (IS_ERR(root)) {
2171                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2172                                 ret = PTR_ERR(root);
2173                         break;
2174                 }
2175                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2176                 ret = btrfs_global_root_insert(root);
2177                 if (ret) {
2178                         btrfs_put_root(root);
2179                         break;
2180                 }
2181                 key.offset++;
2182         }
2183         btrfs_release_path(path);
2184
2185         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2186                 fs_info->nr_global_roots = max_global_id + 1;
2187
2188         if (!found || ret) {
2189                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2190                         set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2191
2192                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2193                         ret = ret ? ret : -ENOENT;
2194                 else
2195                         ret = 0;
2196                 btrfs_err(fs_info, "failed to load root %s", name);
2197         }
2198         return ret;
2199 }
2200
2201 static int load_global_roots(struct btrfs_root *tree_root)
2202 {
2203         struct btrfs_path *path;
2204         int ret = 0;
2205
2206         path = btrfs_alloc_path();
2207         if (!path)
2208                 return -ENOMEM;
2209
2210         ret = load_global_roots_objectid(tree_root, path,
2211                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2212         if (ret)
2213                 goto out;
2214         ret = load_global_roots_objectid(tree_root, path,
2215                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2216         if (ret)
2217                 goto out;
2218         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2219                 goto out;
2220         ret = load_global_roots_objectid(tree_root, path,
2221                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2222                                          "free space");
2223 out:
2224         btrfs_free_path(path);
2225         return ret;
2226 }
2227
2228 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2229 {
2230         struct btrfs_root *tree_root = fs_info->tree_root;
2231         struct btrfs_root *root;
2232         struct btrfs_key location;
2233         int ret;
2234
2235         ASSERT(fs_info->tree_root);
2236
2237         ret = load_global_roots(tree_root);
2238         if (ret)
2239                 return ret;
2240
2241         location.type = BTRFS_ROOT_ITEM_KEY;
2242         location.offset = 0;
2243
2244         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2245                 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2246                 root = btrfs_read_tree_root(tree_root, &location);
2247                 if (IS_ERR(root)) {
2248                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2249                                 ret = PTR_ERR(root);
2250                                 goto out;
2251                         }
2252                 } else {
2253                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2254                         fs_info->block_group_root = root;
2255                 }
2256         }
2257
2258         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2259         root = btrfs_read_tree_root(tree_root, &location);
2260         if (IS_ERR(root)) {
2261                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2262                         ret = PTR_ERR(root);
2263                         goto out;
2264                 }
2265         } else {
2266                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2267                 fs_info->dev_root = root;
2268         }
2269         /* Initialize fs_info for all devices in any case */
2270         ret = btrfs_init_devices_late(fs_info);
2271         if (ret)
2272                 goto out;
2273
2274         /*
2275          * This tree can share blocks with some other fs tree during relocation
2276          * and we need a proper setup by btrfs_get_fs_root
2277          */
2278         root = btrfs_get_fs_root(tree_root->fs_info,
2279                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2280         if (IS_ERR(root)) {
2281                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2282                         ret = PTR_ERR(root);
2283                         goto out;
2284                 }
2285         } else {
2286                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2287                 fs_info->data_reloc_root = root;
2288         }
2289
2290         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2291         root = btrfs_read_tree_root(tree_root, &location);
2292         if (!IS_ERR(root)) {
2293                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294                 fs_info->quota_root = root;
2295         }
2296
2297         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2298         root = btrfs_read_tree_root(tree_root, &location);
2299         if (IS_ERR(root)) {
2300                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2301                         ret = PTR_ERR(root);
2302                         if (ret != -ENOENT)
2303                                 goto out;
2304                 }
2305         } else {
2306                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2307                 fs_info->uuid_root = root;
2308         }
2309
2310         if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2311                 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2312                 root = btrfs_read_tree_root(tree_root, &location);
2313                 if (IS_ERR(root)) {
2314                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2315                                 ret = PTR_ERR(root);
2316                                 goto out;
2317                         }
2318                 } else {
2319                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320                         fs_info->stripe_root = root;
2321                 }
2322         }
2323
2324         return 0;
2325 out:
2326         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2327                    location.objectid, ret);
2328         return ret;
2329 }
2330
2331 static int validate_sys_chunk_array(const struct btrfs_fs_info *fs_info,
2332                                     const struct btrfs_super_block *sb)
2333 {
2334         unsigned int cur = 0; /* Offset inside the sys chunk array */
2335         /*
2336          * At sb read time, fs_info is not fully initialized. Thus we have
2337          * to use super block sectorsize, which should have been validated.
2338          */
2339         const u32 sectorsize = btrfs_super_sectorsize(sb);
2340         u32 sys_array_size = btrfs_super_sys_array_size(sb);
2341
2342         if (sys_array_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2343                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2344                           sys_array_size, BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2345                 return -EUCLEAN;
2346         }
2347
2348         while (cur < sys_array_size) {
2349                 struct btrfs_disk_key *disk_key;
2350                 struct btrfs_chunk *chunk;
2351                 struct btrfs_key key;
2352                 u64 type;
2353                 u16 num_stripes;
2354                 u32 len;
2355                 int ret;
2356
2357                 disk_key = (struct btrfs_disk_key *)(sb->sys_chunk_array + cur);
2358                 len = sizeof(*disk_key);
2359
2360                 if (cur + len > sys_array_size)
2361                         goto short_read;
2362                 cur += len;
2363
2364                 btrfs_disk_key_to_cpu(&key, disk_key);
2365                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
2366                         btrfs_err(fs_info,
2367                             "unexpected item type %u in sys_array at offset %u",
2368                                   key.type, cur);
2369                         return -EUCLEAN;
2370                 }
2371                 chunk = (struct btrfs_chunk *)(sb->sys_chunk_array + cur);
2372                 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2373                 if (cur + btrfs_chunk_item_size(num_stripes) > sys_array_size)
2374                         goto short_read;
2375                 type = btrfs_stack_chunk_type(chunk);
2376                 if (!(type & BTRFS_BLOCK_GROUP_SYSTEM)) {
2377                         btrfs_err(fs_info,
2378                         "invalid chunk type %llu in sys_array at offset %u",
2379                                   type, cur);
2380                         return -EUCLEAN;
2381                 }
2382                 ret = btrfs_check_chunk_valid(fs_info, NULL, chunk, key.offset,
2383                                               sectorsize);
2384                 if (ret < 0)
2385                         return ret;
2386                 cur += btrfs_chunk_item_size(num_stripes);
2387         }
2388         return 0;
2389 short_read:
2390         btrfs_err(fs_info,
2391         "super block sys chunk array short read, cur=%u sys_array_size=%u",
2392                   cur, sys_array_size);
2393         return -EUCLEAN;
2394 }
2395
2396 /*
2397  * Real super block validation
2398  * NOTE: super csum type and incompat features will not be checked here.
2399  *
2400  * @sb:         super block to check
2401  * @mirror_num: the super block number to check its bytenr:
2402  *              0       the primary (1st) sb
2403  *              1, 2    2nd and 3rd backup copy
2404  *             -1       skip bytenr check
2405  */
2406 int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2407                          const struct btrfs_super_block *sb, int mirror_num)
2408 {
2409         u64 nodesize = btrfs_super_nodesize(sb);
2410         u64 sectorsize = btrfs_super_sectorsize(sb);
2411         int ret = 0;
2412         const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2413
2414         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2415                 btrfs_err(fs_info, "no valid FS found");
2416                 ret = -EINVAL;
2417         }
2418         if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2419                 if (!ignore_flags) {
2420                         btrfs_err(fs_info,
2421                         "unrecognized or unsupported super flag 0x%llx",
2422                                   btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2423                         ret = -EINVAL;
2424                 } else {
2425                         btrfs_info(fs_info,
2426                         "unrecognized or unsupported super flags: 0x%llx, ignored",
2427                                    btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2428                 }
2429         }
2430         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2431                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2432                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2433                 ret = -EINVAL;
2434         }
2435         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2436                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2437                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2438                 ret = -EINVAL;
2439         }
2440         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2441                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2442                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2443                 ret = -EINVAL;
2444         }
2445
2446         /*
2447          * Check sectorsize and nodesize first, other check will need it.
2448          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2449          */
2450         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2451             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2452                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2453                 ret = -EINVAL;
2454         }
2455
2456         /*
2457          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2458          *
2459          * We can support 16K sectorsize with 64K page size without problem,
2460          * but such sectorsize/pagesize combination doesn't make much sense.
2461          * 4K will be our future standard, PAGE_SIZE is supported from the very
2462          * beginning.
2463          */
2464         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2465                 btrfs_err(fs_info,
2466                         "sectorsize %llu not yet supported for page size %lu",
2467                         sectorsize, PAGE_SIZE);
2468                 ret = -EINVAL;
2469         }
2470
2471         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2472             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2473                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2474                 ret = -EINVAL;
2475         }
2476         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2477                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2478                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2479                 ret = -EINVAL;
2480         }
2481
2482         /* Root alignment check */
2483         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2484                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2485                            btrfs_super_root(sb));
2486                 ret = -EINVAL;
2487         }
2488         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2489                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2490                            btrfs_super_chunk_root(sb));
2491                 ret = -EINVAL;
2492         }
2493         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2494                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2495                            btrfs_super_log_root(sb));
2496                 ret = -EINVAL;
2497         }
2498
2499         if (!fs_info->fs_devices->temp_fsid &&
2500             memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2501                 btrfs_err(fs_info,
2502                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2503                           sb->fsid, fs_info->fs_devices->fsid);
2504                 ret = -EINVAL;
2505         }
2506
2507         if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2508                    BTRFS_FSID_SIZE) != 0) {
2509                 btrfs_err(fs_info,
2510 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2511                           btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2512                 ret = -EINVAL;
2513         }
2514
2515         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2516                    BTRFS_FSID_SIZE) != 0) {
2517                 btrfs_err(fs_info,
2518                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2519                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2520                 ret = -EINVAL;
2521         }
2522
2523         /*
2524          * Artificial requirement for block-group-tree to force newer features
2525          * (free-space-tree, no-holes) so the test matrix is smaller.
2526          */
2527         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2528             (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2529              !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2530                 btrfs_err(fs_info,
2531                 "block-group-tree feature requires free-space-tree and no-holes");
2532                 ret = -EINVAL;
2533         }
2534
2535         /*
2536          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2537          * done later
2538          */
2539         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2540                 btrfs_err(fs_info, "bytes_used is too small %llu",
2541                           btrfs_super_bytes_used(sb));
2542                 ret = -EINVAL;
2543         }
2544         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2545                 btrfs_err(fs_info, "invalid stripesize %u",
2546                           btrfs_super_stripesize(sb));
2547                 ret = -EINVAL;
2548         }
2549         if (btrfs_super_num_devices(sb) > (1UL << 31))
2550                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2551                            btrfs_super_num_devices(sb));
2552         if (btrfs_super_num_devices(sb) == 0) {
2553                 btrfs_err(fs_info, "number of devices is 0");
2554                 ret = -EINVAL;
2555         }
2556
2557         if (mirror_num >= 0 &&
2558             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2559                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2560                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2561                 ret = -EINVAL;
2562         }
2563
2564         ret = validate_sys_chunk_array(fs_info, sb);
2565
2566         /*
2567          * Obvious sys_chunk_array corruptions, it must hold at least one key
2568          * and one chunk
2569          */
2570         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2571                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2572                           btrfs_super_sys_array_size(sb),
2573                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2574                 ret = -EINVAL;
2575         }
2576         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2577                         + sizeof(struct btrfs_chunk)) {
2578                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2579                           btrfs_super_sys_array_size(sb),
2580                           sizeof(struct btrfs_disk_key)
2581                           + sizeof(struct btrfs_chunk));
2582                 ret = -EINVAL;
2583         }
2584
2585         /*
2586          * The generation is a global counter, we'll trust it more than the others
2587          * but it's still possible that it's the one that's wrong.
2588          */
2589         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2590                 btrfs_warn(fs_info,
2591                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2592                         btrfs_super_generation(sb),
2593                         btrfs_super_chunk_root_generation(sb));
2594         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2595             && btrfs_super_cache_generation(sb) != (u64)-1)
2596                 btrfs_warn(fs_info,
2597                         "suspicious: generation < cache_generation: %llu < %llu",
2598                         btrfs_super_generation(sb),
2599                         btrfs_super_cache_generation(sb));
2600
2601         return ret;
2602 }
2603
2604 /*
2605  * Validation of super block at mount time.
2606  * Some checks already done early at mount time, like csum type and incompat
2607  * flags will be skipped.
2608  */
2609 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2610 {
2611         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2612 }
2613
2614 /*
2615  * Validation of super block at write time.
2616  * Some checks like bytenr check will be skipped as their values will be
2617  * overwritten soon.
2618  * Extra checks like csum type and incompat flags will be done here.
2619  */
2620 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2621                                       struct btrfs_super_block *sb)
2622 {
2623         int ret;
2624
2625         ret = btrfs_validate_super(fs_info, sb, -1);
2626         if (ret < 0)
2627                 goto out;
2628         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2629                 ret = -EUCLEAN;
2630                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2631                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2632                 goto out;
2633         }
2634         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2635                 ret = -EUCLEAN;
2636                 btrfs_err(fs_info,
2637                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2638                           btrfs_super_incompat_flags(sb),
2639                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2640                 goto out;
2641         }
2642 out:
2643         if (ret < 0)
2644                 btrfs_err(fs_info,
2645                 "super block corruption detected before writing it to disk");
2646         return ret;
2647 }
2648
2649 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2650 {
2651         struct btrfs_tree_parent_check check = {
2652                 .level = level,
2653                 .transid = gen,
2654                 .owner_root = btrfs_root_id(root)
2655         };
2656         int ret = 0;
2657
2658         root->node = read_tree_block(root->fs_info, bytenr, &check);
2659         if (IS_ERR(root->node)) {
2660                 ret = PTR_ERR(root->node);
2661                 root->node = NULL;
2662                 return ret;
2663         }
2664         if (!extent_buffer_uptodate(root->node)) {
2665                 free_extent_buffer(root->node);
2666                 root->node = NULL;
2667                 return -EIO;
2668         }
2669
2670         btrfs_set_root_node(&root->root_item, root->node);
2671         root->commit_root = btrfs_root_node(root);
2672         btrfs_set_root_refs(&root->root_item, 1);
2673         return ret;
2674 }
2675
2676 static int load_important_roots(struct btrfs_fs_info *fs_info)
2677 {
2678         struct btrfs_super_block *sb = fs_info->super_copy;
2679         u64 gen, bytenr;
2680         int level, ret;
2681
2682         bytenr = btrfs_super_root(sb);
2683         gen = btrfs_super_generation(sb);
2684         level = btrfs_super_root_level(sb);
2685         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2686         if (ret) {
2687                 btrfs_warn(fs_info, "couldn't read tree root");
2688                 return ret;
2689         }
2690         return 0;
2691 }
2692
2693 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2694 {
2695         int backup_index = find_newest_super_backup(fs_info);
2696         struct btrfs_super_block *sb = fs_info->super_copy;
2697         struct btrfs_root *tree_root = fs_info->tree_root;
2698         bool handle_error = false;
2699         int ret = 0;
2700         int i;
2701
2702         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2703                 if (handle_error) {
2704                         if (!IS_ERR(tree_root->node))
2705                                 free_extent_buffer(tree_root->node);
2706                         tree_root->node = NULL;
2707
2708                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2709                                 break;
2710
2711                         free_root_pointers(fs_info, 0);
2712
2713                         /*
2714                          * Don't use the log in recovery mode, it won't be
2715                          * valid
2716                          */
2717                         btrfs_set_super_log_root(sb, 0);
2718
2719                         btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2720                         ret = read_backup_root(fs_info, i);
2721                         backup_index = ret;
2722                         if (ret < 0)
2723                                 return ret;
2724                 }
2725
2726                 ret = load_important_roots(fs_info);
2727                 if (ret) {
2728                         handle_error = true;
2729                         continue;
2730                 }
2731
2732                 /*
2733                  * No need to hold btrfs_root::objectid_mutex since the fs
2734                  * hasn't been fully initialised and we are the only user
2735                  */
2736                 ret = btrfs_init_root_free_objectid(tree_root);
2737                 if (ret < 0) {
2738                         handle_error = true;
2739                         continue;
2740                 }
2741
2742                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2743
2744                 ret = btrfs_read_roots(fs_info);
2745                 if (ret < 0) {
2746                         handle_error = true;
2747                         continue;
2748                 }
2749
2750                 /* All successful */
2751                 fs_info->generation = btrfs_header_generation(tree_root->node);
2752                 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2753                 fs_info->last_reloc_trans = 0;
2754
2755                 /* Always begin writing backup roots after the one being used */
2756                 if (backup_index < 0) {
2757                         fs_info->backup_root_index = 0;
2758                 } else {
2759                         fs_info->backup_root_index = backup_index + 1;
2760                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2761                 }
2762                 break;
2763         }
2764
2765         return ret;
2766 }
2767
2768 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2769 {
2770         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2771         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2772         INIT_LIST_HEAD(&fs_info->trans_list);
2773         INIT_LIST_HEAD(&fs_info->dead_roots);
2774         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2775         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2776         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2777         spin_lock_init(&fs_info->delalloc_root_lock);
2778         spin_lock_init(&fs_info->trans_lock);
2779         spin_lock_init(&fs_info->fs_roots_radix_lock);
2780         spin_lock_init(&fs_info->delayed_iput_lock);
2781         spin_lock_init(&fs_info->defrag_inodes_lock);
2782         spin_lock_init(&fs_info->super_lock);
2783         spin_lock_init(&fs_info->buffer_lock);
2784         spin_lock_init(&fs_info->unused_bgs_lock);
2785         spin_lock_init(&fs_info->treelog_bg_lock);
2786         spin_lock_init(&fs_info->zone_active_bgs_lock);
2787         spin_lock_init(&fs_info->relocation_bg_lock);
2788         rwlock_init(&fs_info->tree_mod_log_lock);
2789         rwlock_init(&fs_info->global_root_lock);
2790         mutex_init(&fs_info->unused_bg_unpin_mutex);
2791         mutex_init(&fs_info->reclaim_bgs_lock);
2792         mutex_init(&fs_info->reloc_mutex);
2793         mutex_init(&fs_info->delalloc_root_mutex);
2794         mutex_init(&fs_info->zoned_meta_io_lock);
2795         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2796         seqlock_init(&fs_info->profiles_lock);
2797
2798         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2799         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2800         btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2801         btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2802         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2803                                      BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2804         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2805                                      BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2806         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2807                                      BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2808         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2809                                      BTRFS_LOCKDEP_TRANS_COMPLETED);
2810
2811         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2812         INIT_LIST_HEAD(&fs_info->space_info);
2813         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2814         INIT_LIST_HEAD(&fs_info->unused_bgs);
2815         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2816         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2817 #ifdef CONFIG_BTRFS_DEBUG
2818         INIT_LIST_HEAD(&fs_info->allocated_roots);
2819         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2820         spin_lock_init(&fs_info->eb_leak_lock);
2821 #endif
2822         fs_info->mapping_tree = RB_ROOT_CACHED;
2823         rwlock_init(&fs_info->mapping_tree_lock);
2824         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2825                              BTRFS_BLOCK_RSV_GLOBAL);
2826         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2827         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2828         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2829         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2830                              BTRFS_BLOCK_RSV_DELOPS);
2831         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2832                              BTRFS_BLOCK_RSV_DELREFS);
2833
2834         atomic_set(&fs_info->async_delalloc_pages, 0);
2835         atomic_set(&fs_info->defrag_running, 0);
2836         atomic_set(&fs_info->nr_delayed_iputs, 0);
2837         atomic64_set(&fs_info->tree_mod_seq, 0);
2838         fs_info->global_root_tree = RB_ROOT;
2839         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2840         fs_info->metadata_ratio = 0;
2841         fs_info->defrag_inodes = RB_ROOT;
2842         atomic64_set(&fs_info->free_chunk_space, 0);
2843         fs_info->tree_mod_log = RB_ROOT;
2844         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2845         btrfs_init_ref_verify(fs_info);
2846
2847         fs_info->thread_pool_size = min_t(unsigned long,
2848                                           num_online_cpus() + 2, 8);
2849
2850         INIT_LIST_HEAD(&fs_info->ordered_roots);
2851         spin_lock_init(&fs_info->ordered_root_lock);
2852
2853         btrfs_init_scrub(fs_info);
2854         btrfs_init_balance(fs_info);
2855         btrfs_init_async_reclaim_work(fs_info);
2856         btrfs_init_extent_map_shrinker_work(fs_info);
2857
2858         rwlock_init(&fs_info->block_group_cache_lock);
2859         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2860
2861         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2862                             IO_TREE_FS_EXCLUDED_EXTENTS);
2863
2864         mutex_init(&fs_info->ordered_operations_mutex);
2865         mutex_init(&fs_info->tree_log_mutex);
2866         mutex_init(&fs_info->chunk_mutex);
2867         mutex_init(&fs_info->transaction_kthread_mutex);
2868         mutex_init(&fs_info->cleaner_mutex);
2869         mutex_init(&fs_info->ro_block_group_mutex);
2870         init_rwsem(&fs_info->commit_root_sem);
2871         init_rwsem(&fs_info->cleanup_work_sem);
2872         init_rwsem(&fs_info->subvol_sem);
2873         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2874
2875         btrfs_init_dev_replace_locks(fs_info);
2876         btrfs_init_qgroup(fs_info);
2877         btrfs_discard_init(fs_info);
2878
2879         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2880         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2881
2882         init_waitqueue_head(&fs_info->transaction_throttle);
2883         init_waitqueue_head(&fs_info->transaction_wait);
2884         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2885         init_waitqueue_head(&fs_info->async_submit_wait);
2886         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2887
2888         /* Usable values until the real ones are cached from the superblock */
2889         fs_info->nodesize = 4096;
2890         fs_info->sectorsize = 4096;
2891         fs_info->sectorsize_bits = ilog2(4096);
2892         fs_info->stripesize = 4096;
2893
2894         /* Default compress algorithm when user does -o compress */
2895         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2896
2897         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2898
2899         spin_lock_init(&fs_info->swapfile_pins_lock);
2900         fs_info->swapfile_pins = RB_ROOT;
2901
2902         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2903         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2904 }
2905
2906 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2907 {
2908         int ret;
2909
2910         fs_info->sb = sb;
2911         /* Temporary fixed values for block size until we read the superblock. */
2912         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2913         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2914
2915         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2916         if (ret)
2917                 return ret;
2918
2919         ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2920         if (ret)
2921                 return ret;
2922
2923         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2924         if (ret)
2925                 return ret;
2926
2927         ret = percpu_counter_init(&fs_info->stats_read_blocks, 0, GFP_KERNEL);
2928         if (ret)
2929                 return ret;
2930
2931         fs_info->dirty_metadata_batch = PAGE_SIZE *
2932                                         (1 + ilog2(nr_cpu_ids));
2933
2934         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2935         if (ret)
2936                 return ret;
2937
2938         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2939                         GFP_KERNEL);
2940         if (ret)
2941                 return ret;
2942
2943         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2944                                         GFP_KERNEL);
2945         if (!fs_info->delayed_root)
2946                 return -ENOMEM;
2947         btrfs_init_delayed_root(fs_info->delayed_root);
2948
2949         if (sb_rdonly(sb))
2950                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2951         if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2952                 set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2953
2954         return btrfs_alloc_stripe_hash_table(fs_info);
2955 }
2956
2957 static int btrfs_uuid_rescan_kthread(void *data)
2958 {
2959         struct btrfs_fs_info *fs_info = data;
2960         int ret;
2961
2962         /*
2963          * 1st step is to iterate through the existing UUID tree and
2964          * to delete all entries that contain outdated data.
2965          * 2nd step is to add all missing entries to the UUID tree.
2966          */
2967         ret = btrfs_uuid_tree_iterate(fs_info);
2968         if (ret < 0) {
2969                 if (ret != -EINTR)
2970                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2971                                    ret);
2972                 up(&fs_info->uuid_tree_rescan_sem);
2973                 return ret;
2974         }
2975         return btrfs_uuid_scan_kthread(data);
2976 }
2977
2978 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2979 {
2980         struct task_struct *task;
2981
2982         down(&fs_info->uuid_tree_rescan_sem);
2983         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2984         if (IS_ERR(task)) {
2985                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2986                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2987                 up(&fs_info->uuid_tree_rescan_sem);
2988                 return PTR_ERR(task);
2989         }
2990
2991         return 0;
2992 }
2993
2994 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2995 {
2996         u64 root_objectid = 0;
2997         struct btrfs_root *gang[8];
2998         int ret = 0;
2999
3000         while (1) {
3001                 unsigned int found;
3002
3003                 spin_lock(&fs_info->fs_roots_radix_lock);
3004                 found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3005                                              (void **)gang, root_objectid,
3006                                              ARRAY_SIZE(gang));
3007                 if (!found) {
3008                         spin_unlock(&fs_info->fs_roots_radix_lock);
3009                         break;
3010                 }
3011                 root_objectid = btrfs_root_id(gang[found - 1]) + 1;
3012
3013                 for (int i = 0; i < found; i++) {
3014                         /* Avoid to grab roots in dead_roots. */
3015                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3016                                 gang[i] = NULL;
3017                                 continue;
3018                         }
3019                         /* Grab all the search result for later use. */
3020                         gang[i] = btrfs_grab_root(gang[i]);
3021                 }
3022                 spin_unlock(&fs_info->fs_roots_radix_lock);
3023
3024                 for (int i = 0; i < found; i++) {
3025                         if (!gang[i])
3026                                 continue;
3027                         root_objectid = btrfs_root_id(gang[i]);
3028                         /*
3029                          * Continue to release the remaining roots after the first
3030                          * error without cleanup and preserve the first error
3031                          * for the return.
3032                          */
3033                         if (!ret)
3034                                 ret = btrfs_orphan_cleanup(gang[i]);
3035                         btrfs_put_root(gang[i]);
3036                 }
3037                 if (ret)
3038                         break;
3039
3040                 root_objectid++;
3041         }
3042         return ret;
3043 }
3044
3045 /*
3046  * Mounting logic specific to read-write file systems. Shared by open_ctree
3047  * and btrfs_remount when remounting from read-only to read-write.
3048  */
3049 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3050 {
3051         int ret;
3052         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3053         bool rebuild_free_space_tree = false;
3054
3055         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3056             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3057                 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3058                         btrfs_warn(fs_info,
3059                                    "'clear_cache' option is ignored with extent tree v2");
3060                 else
3061                         rebuild_free_space_tree = true;
3062         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3063                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3064                 btrfs_warn(fs_info, "free space tree is invalid");
3065                 rebuild_free_space_tree = true;
3066         }
3067
3068         if (rebuild_free_space_tree) {
3069                 btrfs_info(fs_info, "rebuilding free space tree");
3070                 ret = btrfs_rebuild_free_space_tree(fs_info);
3071                 if (ret) {
3072                         btrfs_warn(fs_info,
3073                                    "failed to rebuild free space tree: %d", ret);
3074                         goto out;
3075                 }
3076         }
3077
3078         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3079             !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3080                 btrfs_info(fs_info, "disabling free space tree");
3081                 ret = btrfs_delete_free_space_tree(fs_info);
3082                 if (ret) {
3083                         btrfs_warn(fs_info,
3084                                    "failed to disable free space tree: %d", ret);
3085                         goto out;
3086                 }
3087         }
3088
3089         /*
3090          * btrfs_find_orphan_roots() is responsible for finding all the dead
3091          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3092          * them into the fs_info->fs_roots_radix tree. This must be done before
3093          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3094          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3095          * item before the root's tree is deleted - this means that if we unmount
3096          * or crash before the deletion completes, on the next mount we will not
3097          * delete what remains of the tree because the orphan item does not
3098          * exists anymore, which is what tells us we have a pending deletion.
3099          */
3100         ret = btrfs_find_orphan_roots(fs_info);
3101         if (ret)
3102                 goto out;
3103
3104         ret = btrfs_cleanup_fs_roots(fs_info);
3105         if (ret)
3106                 goto out;
3107
3108         down_read(&fs_info->cleanup_work_sem);
3109         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3110             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3111                 up_read(&fs_info->cleanup_work_sem);
3112                 goto out;
3113         }
3114         up_read(&fs_info->cleanup_work_sem);
3115
3116         mutex_lock(&fs_info->cleaner_mutex);
3117         ret = btrfs_recover_relocation(fs_info);
3118         mutex_unlock(&fs_info->cleaner_mutex);
3119         if (ret < 0) {
3120                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3121                 goto out;
3122         }
3123
3124         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3125             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3126                 btrfs_info(fs_info, "creating free space tree");
3127                 ret = btrfs_create_free_space_tree(fs_info);
3128                 if (ret) {
3129                         btrfs_warn(fs_info,
3130                                 "failed to create free space tree: %d", ret);
3131                         goto out;
3132                 }
3133         }
3134
3135         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3136                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3137                 if (ret)
3138                         goto out;
3139         }
3140
3141         ret = btrfs_resume_balance_async(fs_info);
3142         if (ret)
3143                 goto out;
3144
3145         ret = btrfs_resume_dev_replace_async(fs_info);
3146         if (ret) {
3147                 btrfs_warn(fs_info, "failed to resume dev_replace");
3148                 goto out;
3149         }
3150
3151         btrfs_qgroup_rescan_resume(fs_info);
3152
3153         if (!fs_info->uuid_root) {
3154                 btrfs_info(fs_info, "creating UUID tree");
3155                 ret = btrfs_create_uuid_tree(fs_info);
3156                 if (ret) {
3157                         btrfs_warn(fs_info,
3158                                    "failed to create the UUID tree %d", ret);
3159                         goto out;
3160                 }
3161         }
3162
3163 out:
3164         return ret;
3165 }
3166
3167 /*
3168  * Do various sanity and dependency checks of different features.
3169  *
3170  * @is_rw_mount:        If the mount is read-write.
3171  *
3172  * This is the place for less strict checks (like for subpage or artificial
3173  * feature dependencies).
3174  *
3175  * For strict checks or possible corruption detection, see
3176  * btrfs_validate_super().
3177  *
3178  * This should be called after btrfs_parse_options(), as some mount options
3179  * (space cache related) can modify on-disk format like free space tree and
3180  * screw up certain feature dependencies.
3181  */
3182 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3183 {
3184         struct btrfs_super_block *disk_super = fs_info->super_copy;
3185         u64 incompat = btrfs_super_incompat_flags(disk_super);
3186         const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3187         const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3188
3189         if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3190                 btrfs_err(fs_info,
3191                 "cannot mount because of unknown incompat features (0x%llx)",
3192                     incompat);
3193                 return -EINVAL;
3194         }
3195
3196         /* Runtime limitation for mixed block groups. */
3197         if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3198             (fs_info->sectorsize != fs_info->nodesize)) {
3199                 btrfs_err(fs_info,
3200 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3201                         fs_info->nodesize, fs_info->sectorsize);
3202                 return -EINVAL;
3203         }
3204
3205         /* Mixed backref is an always-enabled feature. */
3206         incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3207
3208         /* Set compression related flags just in case. */
3209         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3210                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3211         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3212                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3213
3214         /*
3215          * An ancient flag, which should really be marked deprecated.
3216          * Such runtime limitation doesn't really need a incompat flag.
3217          */
3218         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3219                 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3220
3221         if (compat_ro_unsupp && is_rw_mount) {
3222                 btrfs_err(fs_info,
3223         "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3224                        compat_ro);
3225                 return -EINVAL;
3226         }
3227
3228         /*
3229          * We have unsupported RO compat features, although RO mounted, we
3230          * should not cause any metadata writes, including log replay.
3231          * Or we could screw up whatever the new feature requires.
3232          */
3233         if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3234             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3235                 btrfs_err(fs_info,
3236 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3237                           compat_ro);
3238                 return -EINVAL;
3239         }
3240
3241         /*
3242          * Artificial limitations for block group tree, to force
3243          * block-group-tree to rely on no-holes and free-space-tree.
3244          */
3245         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3246             (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3247              !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3248                 btrfs_err(fs_info,
3249 "block-group-tree feature requires no-holes and free-space-tree features");
3250                 return -EINVAL;
3251         }
3252
3253         /*
3254          * Subpage runtime limitation on v1 cache.
3255          *
3256          * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3257          * we're already defaulting to v2 cache, no need to bother v1 as it's
3258          * going to be deprecated anyway.
3259          */
3260         if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3261                 btrfs_warn(fs_info,
3262         "v1 space cache is not supported for page size %lu with sectorsize %u",
3263                            PAGE_SIZE, fs_info->sectorsize);
3264                 return -EINVAL;
3265         }
3266
3267         /* This can be called by remount, we need to protect the super block. */
3268         spin_lock(&fs_info->super_lock);
3269         btrfs_set_super_incompat_flags(disk_super, incompat);
3270         spin_unlock(&fs_info->super_lock);
3271
3272         return 0;
3273 }
3274
3275 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices)
3276 {
3277         u32 sectorsize;
3278         u32 nodesize;
3279         u32 stripesize;
3280         u64 generation;
3281         u16 csum_type;
3282         struct btrfs_super_block *disk_super;
3283         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3284         struct btrfs_root *tree_root;
3285         struct btrfs_root *chunk_root;
3286         int ret;
3287         int level;
3288
3289         ret = init_mount_fs_info(fs_info, sb);
3290         if (ret)
3291                 goto fail;
3292
3293         /* These need to be init'ed before we start creating inodes and such. */
3294         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3295                                      GFP_KERNEL);
3296         fs_info->tree_root = tree_root;
3297         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3298                                       GFP_KERNEL);
3299         fs_info->chunk_root = chunk_root;
3300         if (!tree_root || !chunk_root) {
3301                 ret = -ENOMEM;
3302                 goto fail;
3303         }
3304
3305         ret = btrfs_init_btree_inode(sb);
3306         if (ret)
3307                 goto fail;
3308
3309         invalidate_bdev(fs_devices->latest_dev->bdev);
3310
3311         /*
3312          * Read super block and check the signature bytes only
3313          */
3314         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3315         if (IS_ERR(disk_super)) {
3316                 ret = PTR_ERR(disk_super);
3317                 goto fail_alloc;
3318         }
3319
3320         btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3321         /*
3322          * Verify the type first, if that or the checksum value are
3323          * corrupted, we'll find out
3324          */
3325         csum_type = btrfs_super_csum_type(disk_super);
3326         if (!btrfs_supported_super_csum(csum_type)) {
3327                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3328                           csum_type);
3329                 ret = -EINVAL;
3330                 btrfs_release_disk_super(disk_super);
3331                 goto fail_alloc;
3332         }
3333
3334         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3335
3336         ret = btrfs_init_csum_hash(fs_info, csum_type);
3337         if (ret) {
3338                 btrfs_release_disk_super(disk_super);
3339                 goto fail_alloc;
3340         }
3341
3342         /*
3343          * We want to check superblock checksum, the type is stored inside.
3344          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3345          */
3346         if (btrfs_check_super_csum(fs_info, disk_super)) {
3347                 btrfs_err(fs_info, "superblock checksum mismatch");
3348                 ret = -EINVAL;
3349                 btrfs_release_disk_super(disk_super);
3350                 goto fail_alloc;
3351         }
3352
3353         /*
3354          * super_copy is zeroed at allocation time and we never touch the
3355          * following bytes up to INFO_SIZE, the checksum is calculated from
3356          * the whole block of INFO_SIZE
3357          */
3358         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3359         btrfs_release_disk_super(disk_super);
3360
3361         disk_super = fs_info->super_copy;
3362
3363         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3364                sizeof(*fs_info->super_for_commit));
3365
3366         ret = btrfs_validate_mount_super(fs_info);
3367         if (ret) {
3368                 btrfs_err(fs_info, "superblock contains fatal errors");
3369                 ret = -EINVAL;
3370                 goto fail_alloc;
3371         }
3372
3373         if (!btrfs_super_root(disk_super)) {
3374                 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3375                 ret = -EINVAL;
3376                 goto fail_alloc;
3377         }
3378
3379         /* check FS state, whether FS is broken. */
3380         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3381                 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3382
3383         /* Set up fs_info before parsing mount options */
3384         nodesize = btrfs_super_nodesize(disk_super);
3385         sectorsize = btrfs_super_sectorsize(disk_super);
3386         stripesize = sectorsize;
3387         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3388         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3389
3390         fs_info->nodesize = nodesize;
3391         fs_info->sectorsize = sectorsize;
3392         fs_info->sectorsize_bits = ilog2(sectorsize);
3393         fs_info->sectors_per_page = (PAGE_SIZE >> fs_info->sectorsize_bits);
3394         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3395         fs_info->stripesize = stripesize;
3396         fs_info->fs_devices->fs_info = fs_info;
3397
3398         /*
3399          * Handle the space caching options appropriately now that we have the
3400          * super block loaded and validated.
3401          */
3402         btrfs_set_free_space_cache_settings(fs_info);
3403
3404         if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3405                 ret = -EINVAL;
3406                 goto fail_alloc;
3407         }
3408
3409         ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3410         if (ret < 0)
3411                 goto fail_alloc;
3412
3413         /*
3414          * At this point our mount options are validated, if we set ->max_inline
3415          * to something non-standard make sure we truncate it to sectorsize.
3416          */
3417         fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3418
3419         if (sectorsize < PAGE_SIZE)
3420                 btrfs_warn(fs_info,
3421                 "read-write for sector size %u with page size %lu is experimental",
3422                            sectorsize, PAGE_SIZE);
3423
3424         ret = btrfs_init_workqueues(fs_info);
3425         if (ret)
3426                 goto fail_sb_buffer;
3427
3428         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3429         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3430
3431         /* Update the values for the current filesystem. */
3432         sb->s_blocksize = sectorsize;
3433         sb->s_blocksize_bits = blksize_bits(sectorsize);
3434         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3435
3436         mutex_lock(&fs_info->chunk_mutex);
3437         ret = btrfs_read_sys_array(fs_info);
3438         mutex_unlock(&fs_info->chunk_mutex);
3439         if (ret) {
3440                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3441                 goto fail_sb_buffer;
3442         }
3443
3444         generation = btrfs_super_chunk_root_generation(disk_super);
3445         level = btrfs_super_chunk_root_level(disk_super);
3446         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3447                               generation, level);
3448         if (ret) {
3449                 btrfs_err(fs_info, "failed to read chunk root");
3450                 goto fail_tree_roots;
3451         }
3452
3453         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3454                            offsetof(struct btrfs_header, chunk_tree_uuid),
3455                            BTRFS_UUID_SIZE);
3456
3457         ret = btrfs_read_chunk_tree(fs_info);
3458         if (ret) {
3459                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3460                 goto fail_tree_roots;
3461         }
3462
3463         /*
3464          * At this point we know all the devices that make this filesystem,
3465          * including the seed devices but we don't know yet if the replace
3466          * target is required. So free devices that are not part of this
3467          * filesystem but skip the replace target device which is checked
3468          * below in btrfs_init_dev_replace().
3469          */
3470         btrfs_free_extra_devids(fs_devices);
3471         if (!fs_devices->latest_dev->bdev) {
3472                 btrfs_err(fs_info, "failed to read devices");
3473                 ret = -EIO;
3474                 goto fail_tree_roots;
3475         }
3476
3477         ret = init_tree_roots(fs_info);
3478         if (ret)
3479                 goto fail_tree_roots;
3480
3481         /*
3482          * Get zone type information of zoned block devices. This will also
3483          * handle emulation of a zoned filesystem if a regular device has the
3484          * zoned incompat feature flag set.
3485          */
3486         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3487         if (ret) {
3488                 btrfs_err(fs_info,
3489                           "zoned: failed to read device zone info: %d", ret);
3490                 goto fail_block_groups;
3491         }
3492
3493         /*
3494          * If we have a uuid root and we're not being told to rescan we need to
3495          * check the generation here so we can set the
3496          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3497          * transaction during a balance or the log replay without updating the
3498          * uuid generation, and then if we crash we would rescan the uuid tree,
3499          * even though it was perfectly fine.
3500          */
3501         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3502             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3503                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3504
3505         ret = btrfs_verify_dev_extents(fs_info);
3506         if (ret) {
3507                 btrfs_err(fs_info,
3508                           "failed to verify dev extents against chunks: %d",
3509                           ret);
3510                 goto fail_block_groups;
3511         }
3512         ret = btrfs_recover_balance(fs_info);
3513         if (ret) {
3514                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3515                 goto fail_block_groups;
3516         }
3517
3518         ret = btrfs_init_dev_stats(fs_info);
3519         if (ret) {
3520                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3521                 goto fail_block_groups;
3522         }
3523
3524         ret = btrfs_init_dev_replace(fs_info);
3525         if (ret) {
3526                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3527                 goto fail_block_groups;
3528         }
3529
3530         ret = btrfs_check_zoned_mode(fs_info);
3531         if (ret) {
3532                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3533                           ret);
3534                 goto fail_block_groups;
3535         }
3536
3537         ret = btrfs_sysfs_add_fsid(fs_devices);
3538         if (ret) {
3539                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3540                                 ret);
3541                 goto fail_block_groups;
3542         }
3543
3544         ret = btrfs_sysfs_add_mounted(fs_info);
3545         if (ret) {
3546                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3547                 goto fail_fsdev_sysfs;
3548         }
3549
3550         ret = btrfs_init_space_info(fs_info);
3551         if (ret) {
3552                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3553                 goto fail_sysfs;
3554         }
3555
3556         ret = btrfs_read_block_groups(fs_info);
3557         if (ret) {
3558                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3559                 goto fail_sysfs;
3560         }
3561
3562         btrfs_free_zone_cache(fs_info);
3563
3564         btrfs_check_active_zone_reservation(fs_info);
3565
3566         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3567             !btrfs_check_rw_degradable(fs_info, NULL)) {
3568                 btrfs_warn(fs_info,
3569                 "writable mount is not allowed due to too many missing devices");
3570                 ret = -EINVAL;
3571                 goto fail_sysfs;
3572         }
3573
3574         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3575                                                "btrfs-cleaner");
3576         if (IS_ERR(fs_info->cleaner_kthread)) {
3577                 ret = PTR_ERR(fs_info->cleaner_kthread);
3578                 goto fail_sysfs;
3579         }
3580
3581         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3582                                                    tree_root,
3583                                                    "btrfs-transaction");
3584         if (IS_ERR(fs_info->transaction_kthread)) {
3585                 ret = PTR_ERR(fs_info->transaction_kthread);
3586                 goto fail_cleaner;
3587         }
3588
3589         ret = btrfs_read_qgroup_config(fs_info);
3590         if (ret)
3591                 goto fail_trans_kthread;
3592
3593         if (btrfs_build_ref_tree(fs_info))
3594                 btrfs_err(fs_info, "couldn't build ref tree");
3595
3596         /* do not make disk changes in broken FS or nologreplay is given */
3597         if (btrfs_super_log_root(disk_super) != 0 &&
3598             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3599                 btrfs_info(fs_info, "start tree-log replay");
3600                 ret = btrfs_replay_log(fs_info, fs_devices);
3601                 if (ret)
3602                         goto fail_qgroup;
3603         }
3604
3605         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3606         if (IS_ERR(fs_info->fs_root)) {
3607                 ret = PTR_ERR(fs_info->fs_root);
3608                 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3609                 fs_info->fs_root = NULL;
3610                 goto fail_qgroup;
3611         }
3612
3613         if (sb_rdonly(sb))
3614                 return 0;
3615
3616         ret = btrfs_start_pre_rw_mount(fs_info);
3617         if (ret) {
3618                 close_ctree(fs_info);
3619                 return ret;
3620         }
3621         btrfs_discard_resume(fs_info);
3622
3623         if (fs_info->uuid_root &&
3624             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3625              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3626                 btrfs_info(fs_info, "checking UUID tree");
3627                 ret = btrfs_check_uuid_tree(fs_info);
3628                 if (ret) {
3629                         btrfs_warn(fs_info,
3630                                 "failed to check the UUID tree: %d", ret);
3631                         close_ctree(fs_info);
3632                         return ret;
3633                 }
3634         }
3635
3636         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3637
3638         /* Kick the cleaner thread so it'll start deleting snapshots. */
3639         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3640                 wake_up_process(fs_info->cleaner_kthread);
3641
3642         return 0;
3643
3644 fail_qgroup:
3645         btrfs_free_qgroup_config(fs_info);
3646 fail_trans_kthread:
3647         kthread_stop(fs_info->transaction_kthread);
3648         btrfs_cleanup_transaction(fs_info);
3649         btrfs_free_fs_roots(fs_info);
3650 fail_cleaner:
3651         kthread_stop(fs_info->cleaner_kthread);
3652
3653         /*
3654          * make sure we're done with the btree inode before we stop our
3655          * kthreads
3656          */
3657         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3658
3659 fail_sysfs:
3660         btrfs_sysfs_remove_mounted(fs_info);
3661
3662 fail_fsdev_sysfs:
3663         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3664
3665 fail_block_groups:
3666         btrfs_put_block_group_cache(fs_info);
3667
3668 fail_tree_roots:
3669         if (fs_info->data_reloc_root)
3670                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3671         free_root_pointers(fs_info, true);
3672         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3673
3674 fail_sb_buffer:
3675         btrfs_stop_all_workers(fs_info);
3676         btrfs_free_block_groups(fs_info);
3677 fail_alloc:
3678         btrfs_mapping_tree_free(fs_info);
3679
3680         iput(fs_info->btree_inode);
3681 fail:
3682         btrfs_close_devices(fs_info->fs_devices);
3683         ASSERT(ret < 0);
3684         return ret;
3685 }
3686 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3687
3688 static void btrfs_end_super_write(struct bio *bio)
3689 {
3690         struct btrfs_device *device = bio->bi_private;
3691         struct folio_iter fi;
3692
3693         bio_for_each_folio_all(fi, bio) {
3694                 if (bio->bi_status) {
3695                         btrfs_warn_rl_in_rcu(device->fs_info,
3696                                 "lost super block write due to IO error on %s (%d)",
3697                                 btrfs_dev_name(device),
3698                                 blk_status_to_errno(bio->bi_status));
3699                         btrfs_dev_stat_inc_and_print(device,
3700                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3701                         /* Ensure failure if the primary sb fails. */
3702                         if (bio->bi_opf & REQ_FUA)
3703                                 atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3704                                            &device->sb_write_errors);
3705                         else
3706                                 atomic_inc(&device->sb_write_errors);
3707                 }
3708                 folio_unlock(fi.folio);
3709                 folio_put(fi.folio);
3710         }
3711
3712         bio_put(bio);
3713 }
3714
3715 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3716                                                    int copy_num, bool drop_cache)
3717 {
3718         struct btrfs_super_block *super;
3719         struct page *page;
3720         u64 bytenr, bytenr_orig;
3721         struct address_space *mapping = bdev->bd_mapping;
3722         int ret;
3723
3724         bytenr_orig = btrfs_sb_offset(copy_num);
3725         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3726         if (ret == -ENOENT)
3727                 return ERR_PTR(-EINVAL);
3728         else if (ret)
3729                 return ERR_PTR(ret);
3730
3731         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3732                 return ERR_PTR(-EINVAL);
3733
3734         if (drop_cache) {
3735                 /* This should only be called with the primary sb. */
3736                 ASSERT(copy_num == 0);
3737
3738                 /*
3739                  * Drop the page of the primary superblock, so later read will
3740                  * always read from the device.
3741                  */
3742                 invalidate_inode_pages2_range(mapping,
3743                                 bytenr >> PAGE_SHIFT,
3744                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3745         }
3746
3747         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3748         if (IS_ERR(page))
3749                 return ERR_CAST(page);
3750
3751         super = page_address(page);
3752         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3753                 btrfs_release_disk_super(super);
3754                 return ERR_PTR(-ENODATA);
3755         }
3756
3757         if (btrfs_super_bytenr(super) != bytenr_orig) {
3758                 btrfs_release_disk_super(super);
3759                 return ERR_PTR(-EINVAL);
3760         }
3761
3762         return super;
3763 }
3764
3765
3766 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3767 {
3768         struct btrfs_super_block *super, *latest = NULL;
3769         int i;
3770         u64 transid = 0;
3771
3772         /* we would like to check all the supers, but that would make
3773          * a btrfs mount succeed after a mkfs from a different FS.
3774          * So, we need to add a special mount option to scan for
3775          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3776          */
3777         for (i = 0; i < 1; i++) {
3778                 super = btrfs_read_dev_one_super(bdev, i, false);
3779                 if (IS_ERR(super))
3780                         continue;
3781
3782                 if (!latest || btrfs_super_generation(super) > transid) {
3783                         if (latest)
3784                                 btrfs_release_disk_super(super);
3785
3786                         latest = super;
3787                         transid = btrfs_super_generation(super);
3788                 }
3789         }
3790
3791         return super;
3792 }
3793
3794 /*
3795  * Write superblock @sb to the @device. Do not wait for completion, all the
3796  * folios we use for writing are locked.
3797  *
3798  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3799  * the expected device size at commit time. Note that max_mirrors must be
3800  * same for write and wait phases.
3801  *
3802  * Return number of errors when folio is not found or submission fails.
3803  */
3804 static int write_dev_supers(struct btrfs_device *device,
3805                             struct btrfs_super_block *sb, int max_mirrors)
3806 {
3807         struct btrfs_fs_info *fs_info = device->fs_info;
3808         struct address_space *mapping = device->bdev->bd_mapping;
3809         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3810         int i;
3811         int ret;
3812         u64 bytenr, bytenr_orig;
3813
3814         atomic_set(&device->sb_write_errors, 0);
3815
3816         if (max_mirrors == 0)
3817                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3818
3819         shash->tfm = fs_info->csum_shash;
3820
3821         for (i = 0; i < max_mirrors; i++) {
3822                 struct folio *folio;
3823                 struct bio *bio;
3824                 struct btrfs_super_block *disk_super;
3825                 size_t offset;
3826
3827                 bytenr_orig = btrfs_sb_offset(i);
3828                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3829                 if (ret == -ENOENT) {
3830                         continue;
3831                 } else if (ret < 0) {
3832                         btrfs_err(device->fs_info,
3833                                 "couldn't get super block location for mirror %d",
3834                                 i);
3835                         atomic_inc(&device->sb_write_errors);
3836                         continue;
3837                 }
3838                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3839                     device->commit_total_bytes)
3840                         break;
3841
3842                 btrfs_set_super_bytenr(sb, bytenr_orig);
3843
3844                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3845                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3846                                     sb->csum);
3847
3848                 folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3849                                             FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3850                                             GFP_NOFS);
3851                 if (IS_ERR(folio)) {
3852                         btrfs_err(device->fs_info,
3853                             "couldn't get super block page for bytenr %llu",
3854                             bytenr);
3855                         atomic_inc(&device->sb_write_errors);
3856                         continue;
3857                 }
3858                 ASSERT(folio_order(folio) == 0);
3859
3860                 offset = offset_in_folio(folio, bytenr);
3861                 disk_super = folio_address(folio) + offset;
3862                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3863
3864                 /*
3865                  * Directly use bios here instead of relying on the page cache
3866                  * to do I/O, so we don't lose the ability to do integrity
3867                  * checking.
3868                  */
3869                 bio = bio_alloc(device->bdev, 1,
3870                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3871                                 GFP_NOFS);
3872                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3873                 bio->bi_private = device;
3874                 bio->bi_end_io = btrfs_end_super_write;
3875                 bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3876
3877                 /*
3878                  * We FUA only the first super block.  The others we allow to
3879                  * go down lazy and there's a short window where the on-disk
3880                  * copies might still contain the older version.
3881                  */
3882                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3883                         bio->bi_opf |= REQ_FUA;
3884                 submit_bio(bio);
3885
3886                 if (btrfs_advance_sb_log(device, i))
3887                         atomic_inc(&device->sb_write_errors);
3888         }
3889         return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3890 }
3891
3892 /*
3893  * Wait for write completion of superblocks done by write_dev_supers,
3894  * @max_mirrors same for write and wait phases.
3895  *
3896  * Return -1 if primary super block write failed or when there were no super block
3897  * copies written. Otherwise 0.
3898  */
3899 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3900 {
3901         int i;
3902         int errors = 0;
3903         bool primary_failed = false;
3904         int ret;
3905         u64 bytenr;
3906
3907         if (max_mirrors == 0)
3908                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3909
3910         for (i = 0; i < max_mirrors; i++) {
3911                 struct folio *folio;
3912
3913                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3914                 if (ret == -ENOENT) {
3915                         break;
3916                 } else if (ret < 0) {
3917                         errors++;
3918                         if (i == 0)
3919                                 primary_failed = true;
3920                         continue;
3921                 }
3922                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3923                     device->commit_total_bytes)
3924                         break;
3925
3926                 folio = filemap_get_folio(device->bdev->bd_mapping,
3927                                           bytenr >> PAGE_SHIFT);
3928                 /* If the folio has been removed, then we know it completed. */
3929                 if (IS_ERR(folio))
3930                         continue;
3931                 ASSERT(folio_order(folio) == 0);
3932
3933                 /* Folio will be unlocked once the write completes. */
3934                 folio_wait_locked(folio);
3935                 folio_put(folio);
3936         }
3937
3938         errors += atomic_read(&device->sb_write_errors);
3939         if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3940                 primary_failed = true;
3941         if (primary_failed) {
3942                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3943                           device->devid);
3944                 return -1;
3945         }
3946
3947         return errors < i ? 0 : -1;
3948 }
3949
3950 /*
3951  * endio for the write_dev_flush, this will wake anyone waiting
3952  * for the barrier when it is done
3953  */
3954 static void btrfs_end_empty_barrier(struct bio *bio)
3955 {
3956         bio_uninit(bio);
3957         complete(bio->bi_private);
3958 }
3959
3960 /*
3961  * Submit a flush request to the device if it supports it. Error handling is
3962  * done in the waiting counterpart.
3963  */
3964 static void write_dev_flush(struct btrfs_device *device)
3965 {
3966         struct bio *bio = &device->flush_bio;
3967
3968         device->last_flush_error = BLK_STS_OK;
3969
3970         bio_init(bio, device->bdev, NULL, 0,
3971                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3972         bio->bi_end_io = btrfs_end_empty_barrier;
3973         init_completion(&device->flush_wait);
3974         bio->bi_private = &device->flush_wait;
3975         submit_bio(bio);
3976         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3977 }
3978
3979 /*
3980  * If the flush bio has been submitted by write_dev_flush, wait for it.
3981  * Return true for any error, and false otherwise.
3982  */
3983 static bool wait_dev_flush(struct btrfs_device *device)
3984 {
3985         struct bio *bio = &device->flush_bio;
3986
3987         if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3988                 return false;
3989
3990         wait_for_completion_io(&device->flush_wait);
3991
3992         if (bio->bi_status) {
3993                 device->last_flush_error = bio->bi_status;
3994                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3995                 return true;
3996         }
3997
3998         return false;
3999 }
4000
4001 /*
4002  * send an empty flush down to each device in parallel,
4003  * then wait for them
4004  */
4005 static int barrier_all_devices(struct btrfs_fs_info *info)
4006 {
4007         struct list_head *head;
4008         struct btrfs_device *dev;
4009         int errors_wait = 0;
4010
4011         lockdep_assert_held(&info->fs_devices->device_list_mutex);
4012         /* send down all the barriers */
4013         head = &info->fs_devices->devices;
4014         list_for_each_entry(dev, head, dev_list) {
4015                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4016                         continue;
4017                 if (!dev->bdev)
4018                         continue;
4019                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4020                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4021                         continue;
4022
4023                 write_dev_flush(dev);
4024         }
4025
4026         /* wait for all the barriers */
4027         list_for_each_entry(dev, head, dev_list) {
4028                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4029                         continue;
4030                 if (!dev->bdev) {
4031                         errors_wait++;
4032                         continue;
4033                 }
4034                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4035                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4036                         continue;
4037
4038                 if (wait_dev_flush(dev))
4039                         errors_wait++;
4040         }
4041
4042         /*
4043          * Checks last_flush_error of disks in order to determine the device
4044          * state.
4045          */
4046         if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
4047                 return -EIO;
4048
4049         return 0;
4050 }
4051
4052 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4053 {
4054         int raid_type;
4055         int min_tolerated = INT_MAX;
4056
4057         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4058             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4059                 min_tolerated = min_t(int, min_tolerated,
4060                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4061                                     tolerated_failures);
4062
4063         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4064                 if (raid_type == BTRFS_RAID_SINGLE)
4065                         continue;
4066                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4067                         continue;
4068                 min_tolerated = min_t(int, min_tolerated,
4069                                     btrfs_raid_array[raid_type].
4070                                     tolerated_failures);
4071         }
4072
4073         if (min_tolerated == INT_MAX) {
4074                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4075                 min_tolerated = 0;
4076         }
4077
4078         return min_tolerated;
4079 }
4080
4081 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4082 {
4083         struct list_head *head;
4084         struct btrfs_device *dev;
4085         struct btrfs_super_block *sb;
4086         struct btrfs_dev_item *dev_item;
4087         int ret;
4088         int do_barriers;
4089         int max_errors;
4090         int total_errors = 0;
4091         u64 flags;
4092
4093         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4094
4095         /*
4096          * max_mirrors == 0 indicates we're from commit_transaction,
4097          * not from fsync where the tree roots in fs_info have not
4098          * been consistent on disk.
4099          */
4100         if (max_mirrors == 0)
4101                 backup_super_roots(fs_info);
4102
4103         sb = fs_info->super_for_commit;
4104         dev_item = &sb->dev_item;
4105
4106         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4107         head = &fs_info->fs_devices->devices;
4108         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4109
4110         if (do_barriers) {
4111                 ret = barrier_all_devices(fs_info);
4112                 if (ret) {
4113                         mutex_unlock(
4114                                 &fs_info->fs_devices->device_list_mutex);
4115                         btrfs_handle_fs_error(fs_info, ret,
4116                                               "errors while submitting device barriers.");
4117                         return ret;
4118                 }
4119         }
4120
4121         list_for_each_entry(dev, head, dev_list) {
4122                 if (!dev->bdev) {
4123                         total_errors++;
4124                         continue;
4125                 }
4126                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4127                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4128                         continue;
4129
4130                 btrfs_set_stack_device_generation(dev_item, 0);
4131                 btrfs_set_stack_device_type(dev_item, dev->type);
4132                 btrfs_set_stack_device_id(dev_item, dev->devid);
4133                 btrfs_set_stack_device_total_bytes(dev_item,
4134                                                    dev->commit_total_bytes);
4135                 btrfs_set_stack_device_bytes_used(dev_item,
4136                                                   dev->commit_bytes_used);
4137                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4138                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4139                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4140                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4141                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4142                        BTRFS_FSID_SIZE);
4143
4144                 flags = btrfs_super_flags(sb);
4145                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4146
4147                 ret = btrfs_validate_write_super(fs_info, sb);
4148                 if (ret < 0) {
4149                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4150                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4151                                 "unexpected superblock corruption detected");
4152                         return -EUCLEAN;
4153                 }
4154
4155                 ret = write_dev_supers(dev, sb, max_mirrors);
4156                 if (ret)
4157                         total_errors++;
4158         }
4159         if (total_errors > max_errors) {
4160                 btrfs_err(fs_info, "%d errors while writing supers",
4161                           total_errors);
4162                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4163
4164                 /* FUA is masked off if unsupported and can't be the reason */
4165                 btrfs_handle_fs_error(fs_info, -EIO,
4166                                       "%d errors while writing supers",
4167                                       total_errors);
4168                 return -EIO;
4169         }
4170
4171         total_errors = 0;
4172         list_for_each_entry(dev, head, dev_list) {
4173                 if (!dev->bdev)
4174                         continue;
4175                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4176                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4177                         continue;
4178
4179                 ret = wait_dev_supers(dev, max_mirrors);
4180                 if (ret)
4181                         total_errors++;
4182         }
4183         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4184         if (total_errors > max_errors) {
4185                 btrfs_handle_fs_error(fs_info, -EIO,
4186                                       "%d errors while writing supers",
4187                                       total_errors);
4188                 return -EIO;
4189         }
4190         return 0;
4191 }
4192
4193 /* Drop a fs root from the radix tree and free it. */
4194 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4195                                   struct btrfs_root *root)
4196 {
4197         bool drop_ref = false;
4198
4199         spin_lock(&fs_info->fs_roots_radix_lock);
4200         radix_tree_delete(&fs_info->fs_roots_radix,
4201                           (unsigned long)btrfs_root_id(root));
4202         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4203                 drop_ref = true;
4204         spin_unlock(&fs_info->fs_roots_radix_lock);
4205
4206         if (BTRFS_FS_ERROR(fs_info)) {
4207                 ASSERT(root->log_root == NULL);
4208                 if (root->reloc_root) {
4209                         btrfs_put_root(root->reloc_root);
4210                         root->reloc_root = NULL;
4211                 }
4212         }
4213
4214         if (drop_ref)
4215                 btrfs_put_root(root);
4216 }
4217
4218 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4219 {
4220         mutex_lock(&fs_info->cleaner_mutex);
4221         btrfs_run_delayed_iputs(fs_info);
4222         mutex_unlock(&fs_info->cleaner_mutex);
4223         wake_up_process(fs_info->cleaner_kthread);
4224
4225         /* wait until ongoing cleanup work done */
4226         down_write(&fs_info->cleanup_work_sem);
4227         up_write(&fs_info->cleanup_work_sem);
4228
4229         return btrfs_commit_current_transaction(fs_info->tree_root);
4230 }
4231
4232 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4233 {
4234         struct btrfs_transaction *trans;
4235         struct btrfs_transaction *tmp;
4236         bool found = false;
4237
4238         /*
4239          * This function is only called at the very end of close_ctree(),
4240          * thus no other running transaction, no need to take trans_lock.
4241          */
4242         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4243         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4244                 struct extent_state *cached = NULL;
4245                 u64 dirty_bytes = 0;
4246                 u64 cur = 0;
4247                 u64 found_start;
4248                 u64 found_end;
4249
4250                 found = true;
4251                 while (find_first_extent_bit(&trans->dirty_pages, cur,
4252                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4253                         dirty_bytes += found_end + 1 - found_start;
4254                         cur = found_end + 1;
4255                 }
4256                 btrfs_warn(fs_info,
4257         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4258                            trans->transid, dirty_bytes);
4259                 btrfs_cleanup_one_transaction(trans);
4260
4261                 if (trans == fs_info->running_transaction)
4262                         fs_info->running_transaction = NULL;
4263                 list_del_init(&trans->list);
4264
4265                 btrfs_put_transaction(trans);
4266                 trace_btrfs_transaction_commit(fs_info);
4267         }
4268         ASSERT(!found);
4269 }
4270
4271 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4272 {
4273         int ret;
4274
4275         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4276
4277         /*
4278          * If we had UNFINISHED_DROPS we could still be processing them, so
4279          * clear that bit and wake up relocation so it can stop.
4280          * We must do this before stopping the block group reclaim task, because
4281          * at btrfs_relocate_block_group() we wait for this bit, and after the
4282          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4283          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4284          * return 1.
4285          */
4286         btrfs_wake_unfinished_drop(fs_info);
4287
4288         /*
4289          * We may have the reclaim task running and relocating a data block group,
4290          * in which case it may create delayed iputs. So stop it before we park
4291          * the cleaner kthread otherwise we can get new delayed iputs after
4292          * parking the cleaner, and that can make the async reclaim task to hang
4293          * if it's waiting for delayed iputs to complete, since the cleaner is
4294          * parked and can not run delayed iputs - this will make us hang when
4295          * trying to stop the async reclaim task.
4296          */
4297         cancel_work_sync(&fs_info->reclaim_bgs_work);
4298         /*
4299          * We don't want the cleaner to start new transactions, add more delayed
4300          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4301          * because that frees the task_struct, and the transaction kthread might
4302          * still try to wake up the cleaner.
4303          */
4304         kthread_park(fs_info->cleaner_kthread);
4305
4306         /* wait for the qgroup rescan worker to stop */
4307         btrfs_qgroup_wait_for_completion(fs_info, false);
4308
4309         /* wait for the uuid_scan task to finish */
4310         down(&fs_info->uuid_tree_rescan_sem);
4311         /* avoid complains from lockdep et al., set sem back to initial state */
4312         up(&fs_info->uuid_tree_rescan_sem);
4313
4314         /* pause restriper - we want to resume on mount */
4315         btrfs_pause_balance(fs_info);
4316
4317         btrfs_dev_replace_suspend_for_unmount(fs_info);
4318
4319         btrfs_scrub_cancel(fs_info);
4320
4321         /* wait for any defraggers to finish */
4322         wait_event(fs_info->transaction_wait,
4323                    (atomic_read(&fs_info->defrag_running) == 0));
4324
4325         /* clear out the rbtree of defraggable inodes */
4326         btrfs_cleanup_defrag_inodes(fs_info);
4327
4328         /*
4329          * Wait for any fixup workers to complete.
4330          * If we don't wait for them here and they are still running by the time
4331          * we call kthread_stop() against the cleaner kthread further below, we
4332          * get an use-after-free on the cleaner because the fixup worker adds an
4333          * inode to the list of delayed iputs and then attempts to wakeup the
4334          * cleaner kthread, which was already stopped and destroyed. We parked
4335          * already the cleaner, but below we run all pending delayed iputs.
4336          */
4337         btrfs_flush_workqueue(fs_info->fixup_workers);
4338         /*
4339          * Similar case here, we have to wait for delalloc workers before we
4340          * proceed below and stop the cleaner kthread, otherwise we trigger a
4341          * use-after-tree on the cleaner kthread task_struct when a delalloc
4342          * worker running submit_compressed_extents() adds a delayed iput, which
4343          * does a wake up on the cleaner kthread, which was already freed below
4344          * when we call kthread_stop().
4345          */
4346         btrfs_flush_workqueue(fs_info->delalloc_workers);
4347
4348         /*
4349          * After we parked the cleaner kthread, ordered extents may have
4350          * completed and created new delayed iputs. If one of the async reclaim
4351          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4352          * can hang forever trying to stop it, because if a delayed iput is
4353          * added after it ran btrfs_run_delayed_iputs() and before it called
4354          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4355          * no one else to run iputs.
4356          *
4357          * So wait for all ongoing ordered extents to complete and then run
4358          * delayed iputs. This works because once we reach this point no one
4359          * can either create new ordered extents nor create delayed iputs
4360          * through some other means.
4361          *
4362          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4363          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4364          * but the delayed iput for the respective inode is made only when doing
4365          * the final btrfs_put_ordered_extent() (which must happen at
4366          * btrfs_finish_ordered_io() when we are unmounting).
4367          */
4368         btrfs_flush_workqueue(fs_info->endio_write_workers);
4369         /* Ordered extents for free space inodes. */
4370         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4371         btrfs_run_delayed_iputs(fs_info);
4372
4373         cancel_work_sync(&fs_info->async_reclaim_work);
4374         cancel_work_sync(&fs_info->async_data_reclaim_work);
4375         cancel_work_sync(&fs_info->preempt_reclaim_work);
4376         cancel_work_sync(&fs_info->em_shrinker_work);
4377
4378         /* Cancel or finish ongoing discard work */
4379         btrfs_discard_cleanup(fs_info);
4380
4381         if (!sb_rdonly(fs_info->sb)) {
4382                 /*
4383                  * The cleaner kthread is stopped, so do one final pass over
4384                  * unused block groups.
4385                  */
4386                 btrfs_delete_unused_bgs(fs_info);
4387
4388                 /*
4389                  * There might be existing delayed inode workers still running
4390                  * and holding an empty delayed inode item. We must wait for
4391                  * them to complete first because they can create a transaction.
4392                  * This happens when someone calls btrfs_balance_delayed_items()
4393                  * and then a transaction commit runs the same delayed nodes
4394                  * before any delayed worker has done something with the nodes.
4395                  * We must wait for any worker here and not at transaction
4396                  * commit time since that could cause a deadlock.
4397                  * This is a very rare case.
4398                  */
4399                 btrfs_flush_workqueue(fs_info->delayed_workers);
4400
4401                 ret = btrfs_commit_super(fs_info);
4402                 if (ret)
4403                         btrfs_err(fs_info, "commit super ret %d", ret);
4404         }
4405
4406         if (BTRFS_FS_ERROR(fs_info))
4407                 btrfs_error_commit_super(fs_info);
4408
4409         kthread_stop(fs_info->transaction_kthread);
4410         kthread_stop(fs_info->cleaner_kthread);
4411
4412         ASSERT(list_empty(&fs_info->delayed_iputs));
4413         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4414
4415         if (btrfs_check_quota_leak(fs_info)) {
4416                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4417                 btrfs_err(fs_info, "qgroup reserved space leaked");
4418         }
4419
4420         btrfs_free_qgroup_config(fs_info);
4421         ASSERT(list_empty(&fs_info->delalloc_roots));
4422
4423         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4424                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4425                        percpu_counter_sum(&fs_info->delalloc_bytes));
4426         }
4427
4428         if (percpu_counter_sum(&fs_info->ordered_bytes))
4429                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4430                            percpu_counter_sum(&fs_info->ordered_bytes));
4431
4432         btrfs_sysfs_remove_mounted(fs_info);
4433         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4434
4435         btrfs_put_block_group_cache(fs_info);
4436
4437         /*
4438          * we must make sure there is not any read request to
4439          * submit after we stopping all workers.
4440          */
4441         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4442         btrfs_stop_all_workers(fs_info);
4443
4444         /* We shouldn't have any transaction open at this point */
4445         warn_about_uncommitted_trans(fs_info);
4446
4447         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4448         free_root_pointers(fs_info, true);
4449         btrfs_free_fs_roots(fs_info);
4450
4451         /*
4452          * We must free the block groups after dropping the fs_roots as we could
4453          * have had an IO error and have left over tree log blocks that aren't
4454          * cleaned up until the fs roots are freed.  This makes the block group
4455          * accounting appear to be wrong because there's pending reserved bytes,
4456          * so make sure we do the block group cleanup afterwards.
4457          */
4458         btrfs_free_block_groups(fs_info);
4459
4460         iput(fs_info->btree_inode);
4461
4462         btrfs_mapping_tree_free(fs_info);
4463         btrfs_close_devices(fs_info->fs_devices);
4464 }
4465
4466 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4467                              struct extent_buffer *buf)
4468 {
4469         struct btrfs_fs_info *fs_info = buf->fs_info;
4470         u64 transid = btrfs_header_generation(buf);
4471
4472 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4473         /*
4474          * This is a fast path so only do this check if we have sanity tests
4475          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4476          * outside of the sanity tests.
4477          */
4478         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4479                 return;
4480 #endif
4481         /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4482         ASSERT(trans->transid == fs_info->generation);
4483         btrfs_assert_tree_write_locked(buf);
4484         if (unlikely(transid != fs_info->generation)) {
4485                 btrfs_abort_transaction(trans, -EUCLEAN);
4486                 btrfs_crit(fs_info,
4487 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4488                            buf->start, transid, fs_info->generation);
4489         }
4490         set_extent_buffer_dirty(buf);
4491 }
4492
4493 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4494                                         int flush_delayed)
4495 {
4496         /*
4497          * looks as though older kernels can get into trouble with
4498          * this code, they end up stuck in balance_dirty_pages forever
4499          */
4500         int ret;
4501
4502         if (current->flags & PF_MEMALLOC)
4503                 return;
4504
4505         if (flush_delayed)
4506                 btrfs_balance_delayed_items(fs_info);
4507
4508         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4509                                      BTRFS_DIRTY_METADATA_THRESH,
4510                                      fs_info->dirty_metadata_batch);
4511         if (ret > 0) {
4512                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4513         }
4514 }
4515
4516 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4517 {
4518         __btrfs_btree_balance_dirty(fs_info, 1);
4519 }
4520
4521 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4522 {
4523         __btrfs_btree_balance_dirty(fs_info, 0);
4524 }
4525
4526 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4527 {
4528         /* cleanup FS via transaction */
4529         btrfs_cleanup_transaction(fs_info);
4530
4531         mutex_lock(&fs_info->cleaner_mutex);
4532         btrfs_run_delayed_iputs(fs_info);
4533         mutex_unlock(&fs_info->cleaner_mutex);
4534
4535         down_write(&fs_info->cleanup_work_sem);
4536         up_write(&fs_info->cleanup_work_sem);
4537 }
4538
4539 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4540 {
4541         struct btrfs_root *gang[8];
4542         u64 root_objectid = 0;
4543         int ret;
4544
4545         spin_lock(&fs_info->fs_roots_radix_lock);
4546         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4547                                              (void **)gang, root_objectid,
4548                                              ARRAY_SIZE(gang))) != 0) {
4549                 int i;
4550
4551                 for (i = 0; i < ret; i++)
4552                         gang[i] = btrfs_grab_root(gang[i]);
4553                 spin_unlock(&fs_info->fs_roots_radix_lock);
4554
4555                 for (i = 0; i < ret; i++) {
4556                         if (!gang[i])
4557                                 continue;
4558                         root_objectid = btrfs_root_id(gang[i]);
4559                         btrfs_free_log(NULL, gang[i]);
4560                         btrfs_put_root(gang[i]);
4561                 }
4562                 root_objectid++;
4563                 spin_lock(&fs_info->fs_roots_radix_lock);
4564         }
4565         spin_unlock(&fs_info->fs_roots_radix_lock);
4566         btrfs_free_log_root_tree(NULL, fs_info);
4567 }
4568
4569 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4570 {
4571         struct btrfs_ordered_extent *ordered;
4572
4573         spin_lock(&root->ordered_extent_lock);
4574         /*
4575          * This will just short circuit the ordered completion stuff which will
4576          * make sure the ordered extent gets properly cleaned up.
4577          */
4578         list_for_each_entry(ordered, &root->ordered_extents,
4579                             root_extent_list)
4580                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4581         spin_unlock(&root->ordered_extent_lock);
4582 }
4583
4584 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4585 {
4586         struct btrfs_root *root;
4587         LIST_HEAD(splice);
4588
4589         spin_lock(&fs_info->ordered_root_lock);
4590         list_splice_init(&fs_info->ordered_roots, &splice);
4591         while (!list_empty(&splice)) {
4592                 root = list_first_entry(&splice, struct btrfs_root,
4593                                         ordered_root);
4594                 list_move_tail(&root->ordered_root,
4595                                &fs_info->ordered_roots);
4596
4597                 spin_unlock(&fs_info->ordered_root_lock);
4598                 btrfs_destroy_ordered_extents(root);
4599
4600                 cond_resched();
4601                 spin_lock(&fs_info->ordered_root_lock);
4602         }
4603         spin_unlock(&fs_info->ordered_root_lock);
4604
4605         /*
4606          * We need this here because if we've been flipped read-only we won't
4607          * get sync() from the umount, so we need to make sure any ordered
4608          * extents that haven't had their dirty pages IO start writeout yet
4609          * actually get run and error out properly.
4610          */
4611         btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4612 }
4613
4614 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4615 {
4616         struct btrfs_inode *btrfs_inode;
4617         LIST_HEAD(splice);
4618
4619         spin_lock(&root->delalloc_lock);
4620         list_splice_init(&root->delalloc_inodes, &splice);
4621
4622         while (!list_empty(&splice)) {
4623                 struct inode *inode = NULL;
4624                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4625                                                delalloc_inodes);
4626                 btrfs_del_delalloc_inode(btrfs_inode);
4627                 spin_unlock(&root->delalloc_lock);
4628
4629                 /*
4630                  * Make sure we get a live inode and that it'll not disappear
4631                  * meanwhile.
4632                  */
4633                 inode = igrab(&btrfs_inode->vfs_inode);
4634                 if (inode) {
4635                         unsigned int nofs_flag;
4636
4637                         nofs_flag = memalloc_nofs_save();
4638                         invalidate_inode_pages2(inode->i_mapping);
4639                         memalloc_nofs_restore(nofs_flag);
4640                         iput(inode);
4641                 }
4642                 spin_lock(&root->delalloc_lock);
4643         }
4644         spin_unlock(&root->delalloc_lock);
4645 }
4646
4647 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4648 {
4649         struct btrfs_root *root;
4650         LIST_HEAD(splice);
4651
4652         spin_lock(&fs_info->delalloc_root_lock);
4653         list_splice_init(&fs_info->delalloc_roots, &splice);
4654         while (!list_empty(&splice)) {
4655                 root = list_first_entry(&splice, struct btrfs_root,
4656                                          delalloc_root);
4657                 root = btrfs_grab_root(root);
4658                 BUG_ON(!root);
4659                 spin_unlock(&fs_info->delalloc_root_lock);
4660
4661                 btrfs_destroy_delalloc_inodes(root);
4662                 btrfs_put_root(root);
4663
4664                 spin_lock(&fs_info->delalloc_root_lock);
4665         }
4666         spin_unlock(&fs_info->delalloc_root_lock);
4667 }
4668
4669 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4670                                          struct extent_io_tree *dirty_pages,
4671                                          int mark)
4672 {
4673         struct extent_buffer *eb;
4674         u64 start = 0;
4675         u64 end;
4676
4677         while (find_first_extent_bit(dirty_pages, start, &start, &end,
4678                                      mark, NULL)) {
4679                 clear_extent_bits(dirty_pages, start, end, mark);
4680                 while (start <= end) {
4681                         eb = find_extent_buffer(fs_info, start);
4682                         start += fs_info->nodesize;
4683                         if (!eb)
4684                                 continue;
4685
4686                         btrfs_tree_lock(eb);
4687                         wait_on_extent_buffer_writeback(eb);
4688                         btrfs_clear_buffer_dirty(NULL, eb);
4689                         btrfs_tree_unlock(eb);
4690
4691                         free_extent_buffer_stale(eb);
4692                 }
4693         }
4694 }
4695
4696 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4697                                         struct extent_io_tree *unpin)
4698 {
4699         u64 start;
4700         u64 end;
4701
4702         while (1) {
4703                 struct extent_state *cached_state = NULL;
4704
4705                 /*
4706                  * The btrfs_finish_extent_commit() may get the same range as
4707                  * ours between find_first_extent_bit and clear_extent_dirty.
4708                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4709                  * the same extent range.
4710                  */
4711                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4712                 if (!find_first_extent_bit(unpin, 0, &start, &end,
4713                                            EXTENT_DIRTY, &cached_state)) {
4714                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4715                         break;
4716                 }
4717
4718                 clear_extent_dirty(unpin, start, end, &cached_state);
4719                 free_extent_state(cached_state);
4720                 btrfs_error_unpin_extent_range(fs_info, start, end);
4721                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4722                 cond_resched();
4723         }
4724 }
4725
4726 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4727 {
4728         struct inode *inode;
4729
4730         inode = cache->io_ctl.inode;
4731         if (inode) {
4732                 unsigned int nofs_flag;
4733
4734                 nofs_flag = memalloc_nofs_save();
4735                 invalidate_inode_pages2(inode->i_mapping);
4736                 memalloc_nofs_restore(nofs_flag);
4737
4738                 BTRFS_I(inode)->generation = 0;
4739                 cache->io_ctl.inode = NULL;
4740                 iput(inode);
4741         }
4742         ASSERT(cache->io_ctl.pages == NULL);
4743         btrfs_put_block_group(cache);
4744 }
4745
4746 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4747                              struct btrfs_fs_info *fs_info)
4748 {
4749         struct btrfs_block_group *cache;
4750
4751         spin_lock(&cur_trans->dirty_bgs_lock);
4752         while (!list_empty(&cur_trans->dirty_bgs)) {
4753                 cache = list_first_entry(&cur_trans->dirty_bgs,
4754                                          struct btrfs_block_group,
4755                                          dirty_list);
4756
4757                 if (!list_empty(&cache->io_list)) {
4758                         spin_unlock(&cur_trans->dirty_bgs_lock);
4759                         list_del_init(&cache->io_list);
4760                         btrfs_cleanup_bg_io(cache);
4761                         spin_lock(&cur_trans->dirty_bgs_lock);
4762                 }
4763
4764                 list_del_init(&cache->dirty_list);
4765                 spin_lock(&cache->lock);
4766                 cache->disk_cache_state = BTRFS_DC_ERROR;
4767                 spin_unlock(&cache->lock);
4768
4769                 spin_unlock(&cur_trans->dirty_bgs_lock);
4770                 btrfs_put_block_group(cache);
4771                 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4772                 spin_lock(&cur_trans->dirty_bgs_lock);
4773         }
4774         spin_unlock(&cur_trans->dirty_bgs_lock);
4775
4776         /*
4777          * Refer to the definition of io_bgs member for details why it's safe
4778          * to use it without any locking
4779          */
4780         while (!list_empty(&cur_trans->io_bgs)) {
4781                 cache = list_first_entry(&cur_trans->io_bgs,
4782                                          struct btrfs_block_group,
4783                                          io_list);
4784
4785                 list_del_init(&cache->io_list);
4786                 spin_lock(&cache->lock);
4787                 cache->disk_cache_state = BTRFS_DC_ERROR;
4788                 spin_unlock(&cache->lock);
4789                 btrfs_cleanup_bg_io(cache);
4790         }
4791 }
4792
4793 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4794 {
4795         struct btrfs_root *gang[8];
4796         int i;
4797         int ret;
4798
4799         spin_lock(&fs_info->fs_roots_radix_lock);
4800         while (1) {
4801                 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4802                                                  (void **)gang, 0,
4803                                                  ARRAY_SIZE(gang),
4804                                                  BTRFS_ROOT_TRANS_TAG);
4805                 if (ret == 0)
4806                         break;
4807                 for (i = 0; i < ret; i++) {
4808                         struct btrfs_root *root = gang[i];
4809
4810                         btrfs_qgroup_free_meta_all_pertrans(root);
4811                         radix_tree_tag_clear(&fs_info->fs_roots_radix,
4812                                         (unsigned long)btrfs_root_id(root),
4813                                         BTRFS_ROOT_TRANS_TAG);
4814                 }
4815         }
4816         spin_unlock(&fs_info->fs_roots_radix_lock);
4817 }
4818
4819 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans)
4820 {
4821         struct btrfs_fs_info *fs_info = cur_trans->fs_info;
4822         struct btrfs_device *dev, *tmp;
4823
4824         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4825         ASSERT(list_empty(&cur_trans->dirty_bgs));
4826         ASSERT(list_empty(&cur_trans->io_bgs));
4827
4828         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4829                                  post_commit_list) {
4830                 list_del_init(&dev->post_commit_list);
4831         }
4832
4833         btrfs_destroy_delayed_refs(cur_trans);
4834
4835         cur_trans->state = TRANS_STATE_COMMIT_START;
4836         wake_up(&fs_info->transaction_blocked_wait);
4837
4838         cur_trans->state = TRANS_STATE_UNBLOCKED;
4839         wake_up(&fs_info->transaction_wait);
4840
4841         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4842                                      EXTENT_DIRTY);
4843         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4844
4845         cur_trans->state =TRANS_STATE_COMPLETED;
4846         wake_up(&cur_trans->commit_wait);
4847 }
4848
4849 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4850 {
4851         struct btrfs_transaction *t;
4852
4853         mutex_lock(&fs_info->transaction_kthread_mutex);
4854
4855         spin_lock(&fs_info->trans_lock);
4856         while (!list_empty(&fs_info->trans_list)) {
4857                 t = list_first_entry(&fs_info->trans_list,
4858                                      struct btrfs_transaction, list);
4859                 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4860                         refcount_inc(&t->use_count);
4861                         spin_unlock(&fs_info->trans_lock);
4862                         btrfs_wait_for_commit(fs_info, t->transid);
4863                         btrfs_put_transaction(t);
4864                         spin_lock(&fs_info->trans_lock);
4865                         continue;
4866                 }
4867                 if (t == fs_info->running_transaction) {
4868                         t->state = TRANS_STATE_COMMIT_DOING;
4869                         spin_unlock(&fs_info->trans_lock);
4870                         /*
4871                          * We wait for 0 num_writers since we don't hold a trans
4872                          * handle open currently for this transaction.
4873                          */
4874                         wait_event(t->writer_wait,
4875                                    atomic_read(&t->num_writers) == 0);
4876                 } else {
4877                         spin_unlock(&fs_info->trans_lock);
4878                 }
4879                 btrfs_cleanup_one_transaction(t);
4880
4881                 spin_lock(&fs_info->trans_lock);
4882                 if (t == fs_info->running_transaction)
4883                         fs_info->running_transaction = NULL;
4884                 list_del_init(&t->list);
4885                 spin_unlock(&fs_info->trans_lock);
4886
4887                 btrfs_put_transaction(t);
4888                 trace_btrfs_transaction_commit(fs_info);
4889                 spin_lock(&fs_info->trans_lock);
4890         }
4891         spin_unlock(&fs_info->trans_lock);
4892         btrfs_destroy_all_ordered_extents(fs_info);
4893         btrfs_destroy_delayed_inodes(fs_info);
4894         btrfs_assert_delayed_root_empty(fs_info);
4895         btrfs_destroy_all_delalloc_inodes(fs_info);
4896         btrfs_drop_all_logs(fs_info);
4897         btrfs_free_all_qgroup_pertrans(fs_info);
4898         mutex_unlock(&fs_info->transaction_kthread_mutex);
4899
4900         return 0;
4901 }
4902
4903 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4904 {
4905         struct btrfs_path *path;
4906         int ret;
4907         struct extent_buffer *l;
4908         struct btrfs_key search_key;
4909         struct btrfs_key found_key;
4910         int slot;
4911
4912         path = btrfs_alloc_path();
4913         if (!path)
4914                 return -ENOMEM;
4915
4916         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4917         search_key.type = -1;
4918         search_key.offset = (u64)-1;
4919         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4920         if (ret < 0)
4921                 goto error;
4922         if (ret == 0) {
4923                 /*
4924                  * Key with offset -1 found, there would have to exist a root
4925                  * with such id, but this is out of valid range.
4926                  */
4927                 ret = -EUCLEAN;
4928                 goto error;
4929         }
4930         if (path->slots[0] > 0) {
4931                 slot = path->slots[0] - 1;
4932                 l = path->nodes[0];
4933                 btrfs_item_key_to_cpu(l, &found_key, slot);
4934                 root->free_objectid = max_t(u64, found_key.objectid + 1,
4935                                             BTRFS_FIRST_FREE_OBJECTID);
4936         } else {
4937                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4938         }
4939         ret = 0;
4940 error:
4941         btrfs_free_path(path);
4942         return ret;
4943 }
4944
4945 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4946 {
4947         int ret;
4948         mutex_lock(&root->objectid_mutex);
4949
4950         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4951                 btrfs_warn(root->fs_info,
4952                            "the objectid of root %llu reaches its highest value",
4953                            btrfs_root_id(root));
4954                 ret = -ENOSPC;
4955                 goto out;
4956         }
4957
4958         *objectid = root->free_objectid++;
4959         ret = 0;
4960 out:
4961         mutex_unlock(&root->objectid_mutex);
4962         return ret;
4963 }