fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <linux/sched/mm.h>
  21 #include <asm/unaligned.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42
  43 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  44                                  BTRFS_HEADER_FLAG_RELOC |\
  45                                  BTRFS_SUPER_FLAG_ERROR |\
  46                                  BTRFS_SUPER_FLAG_SEEDING |\
  47                                  BTRFS_SUPER_FLAG_METADUMP |\
  48                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  49
  50 static const struct extent_io_ops btree_extent_io_ops;
  51 static void end_workqueue_fn(struct btrfs_work *work);
  52 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  53 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  54                                       struct btrfs_fs_info *fs_info);
  55 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  56 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  57                                         struct extent_io_tree *dirty_pages,
  58                                         int mark);
  59 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  60                                        struct extent_io_tree *pinned_extents);
  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 /*
  65  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  66  * is complete.  This is used during reads to verify checksums, and it is used
  67  * by writes to insert metadata for new file extents after IO is complete.
  68  */
  69 struct btrfs_end_io_wq {
  70         struct bio *bio;
  71         bio_end_io_t *end_io;
  72         void *private;
  73         struct btrfs_fs_info *info;
  74         blk_status_t status;
  75         enum btrfs_wq_endio_type metadata;
  76         struct btrfs_work work;
  77 };
  78
  79 static struct kmem_cache *btrfs_end_io_wq_cache;
  80
  81 int __init btrfs_end_io_wq_init(void)
  82 {
  83         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  84                                         sizeof(struct btrfs_end_io_wq),
  85                                         0,
  86                                         SLAB_MEM_SPREAD,
  87                                         NULL);
  88         if (!btrfs_end_io_wq_cache)
  89                 return -ENOMEM;
  90         return 0;
  91 }
  92
  93 void __cold btrfs_end_io_wq_exit(void)
  94 {
  95         kmem_cache_destroy(btrfs_end_io_wq_cache);
  96 }
  97
  98 /*
  99  * async submit bios are used to offload expensive checksumming
 100  * onto the worker threads.  They checksum file and metadata bios
 101  * just before they are sent down the IO stack.
 102  */
 103 struct async_submit_bio {
 104         void *private_data;
 105         struct bio *bio;
 106         extent_submit_bio_start_t *submit_bio_start;
 107         int mirror_num;
 108         /*
 109          * bio_offset is optional, can be used if the pages in the bio
 110          * can't tell us where in the file the bio should go
 111          */
 112         u64 bio_offset;
 113         struct btrfs_work work;
 114         blk_status_t status;
 115 };
 116
 117 /*
 118  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 119  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 120  * the level the eb occupies in the tree.
 121  *
 122  * Different roots are used for different purposes and may nest inside each
 123  * other and they require separate keysets.  As lockdep keys should be
 124  * static, assign keysets according to the purpose of the root as indicated
 125  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 126  * roots have separate keysets.
 127  *
 128  * Lock-nesting across peer nodes is always done with the immediate parent
 129  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 130  * subclass to avoid triggering lockdep warning in such cases.
 131  *
 132  * The key is set by the readpage_end_io_hook after the buffer has passed
 133  * csum validation but before the pages are unlocked.  It is also set by
 134  * btrfs_init_new_buffer on freshly allocated blocks.
 135  *
 136  * We also add a check to make sure the highest level of the tree is the
 137  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 138  * needs update as well.
 139  */
 140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 141 # if BTRFS_MAX_LEVEL != 8
 142 #  error
 143 # endif
 144
 145 static struct btrfs_lockdep_keyset {
 146         u64                     id;             /* root objectid */
 147         const char              *name_stem;     /* lock name stem */
 148         char                    names[BTRFS_MAX_LEVEL + 1][20];
 149         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 150 } btrfs_lockdep_keysets[] = {
 151         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 152         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 153         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 154         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 155         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 156         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 157         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 158         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 159         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 160         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 161         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 162         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 163         { .id = 0,                              .name_stem = "tree"     },
 164 };
 165
 166 void __init btrfs_init_lockdep(void)
 167 {
 168         int i, j;
 169
 170         /* initialize lockdep class names */
 171         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 172                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 173
 174                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 175                         snprintf(ks->names[j], sizeof(ks->names[j]),
 176                                  "btrfs-%s-%02d", ks->name_stem, j);
 177         }
 178 }
 179
 180 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 181                                     int level)
 182 {
 183         struct btrfs_lockdep_keyset *ks;
 184
 185         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 186
 187         /* find the matching keyset, id 0 is the default entry */
 188         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 189                 if (ks->id == objectid)
 190                         break;
 191
 192         lockdep_set_class_and_name(&eb->lock,
 193                                    &ks->keys[level], ks->names[level]);
 194 }
 195
 196 #endif
 197
 198 /*
 199  * extents on the btree inode are pretty simple, there's one extent
 200  * that covers the entire device
 201  */
 202 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 203                 struct page *page, size_t pg_offset, u64 start, u64 len,
 204                 int create)
 205 {
 206         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 207         struct extent_map_tree *em_tree = &inode->extent_tree;
 208         struct extent_map *em;
 209         int ret;
 210
 211         read_lock(&em_tree->lock);
 212         em = lookup_extent_mapping(em_tree, start, len);
 213         if (em) {
 214                 em->bdev = fs_info->fs_devices->latest_bdev;
 215                 read_unlock(&em_tree->lock);
 216                 goto out;
 217         }
 218         read_unlock(&em_tree->lock);
 219
 220         em = alloc_extent_map();
 221         if (!em) {
 222                 em = ERR_PTR(-ENOMEM);
 223                 goto out;
 224         }
 225         em->start = 0;
 226         em->len = (u64)-1;
 227         em->block_len = (u64)-1;
 228         em->block_start = 0;
 229         em->bdev = fs_info->fs_devices->latest_bdev;
 230
 231         write_lock(&em_tree->lock);
 232         ret = add_extent_mapping(em_tree, em, 0);
 233         if (ret == -EEXIST) {
 234                 free_extent_map(em);
 235                 em = lookup_extent_mapping(em_tree, start, len);
 236                 if (!em)
 237                         em = ERR_PTR(-EIO);
 238         } else if (ret) {
 239                 free_extent_map(em);
 240                 em = ERR_PTR(ret);
 241         }
 242         write_unlock(&em_tree->lock);
 243
 244 out:
 245         return em;
 246 }
 247
 248 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 249 {
 250         return crc32c(seed, data, len);
 251 }
 252
 253 void btrfs_csum_final(u32 crc, u8 *result)
 254 {
 255         put_unaligned_le32(~crc, result);
 256 }
 257
 258 /*
 259  * Compute the csum of a btree block and store the result to provided buffer.
 260  *
 261  * Returns error if the extent buffer cannot be mapped.
 262  */
 263 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
 264 {
 265         unsigned long len;
 266         unsigned long cur_len;
 267         unsigned long offset = BTRFS_CSUM_SIZE;
 268         char *kaddr;
 269         unsigned long map_start;
 270         unsigned long map_len;
 271         int err;
 272         u32 crc = ~(u32)0;
 273
 274         len = buf->len - offset;
 275         while (len > 0) {
 276                 /*
 277                  * Note: we don't need to check for the err == 1 case here, as
 278                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 279                  * and 'min_len = 32' and the currently implemented mapping
 280                  * algorithm we cannot cross a page boundary.
 281                  */
 282                 err = map_private_extent_buffer(buf, offset, 32,
 283                                         &kaddr, &map_start, &map_len);
 284                 if (WARN_ON(err))
 285                         return err;
 286                 cur_len = min(len, map_len - (offset - map_start));
 287                 crc = btrfs_csum_data(kaddr + offset - map_start,
 288                                       crc, cur_len);
 289                 len -= cur_len;
 290                 offset += cur_len;
 291         }
 292         memset(result, 0, BTRFS_CSUM_SIZE);
 293
 294         btrfs_csum_final(crc, result);
 295
 296         return 0;
 297 }
 298
 299 /*
 300  * we can't consider a given block up to date unless the transid of the
 301  * block matches the transid in the parent node's pointer.  This is how we
 302  * detect blocks that either didn't get written at all or got written
 303  * in the wrong place.
 304  */
 305 static int verify_parent_transid(struct extent_io_tree *io_tree,
 306                                  struct extent_buffer *eb, u64 parent_transid,
 307                                  int atomic)
 308 {
 309         struct extent_state *cached_state = NULL;
 310         int ret;
 311         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 312
 313         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 314                 return 0;
 315
 316         if (atomic)
 317                 return -EAGAIN;
 318
 319         if (need_lock) {
 320                 btrfs_tree_read_lock(eb);
 321                 btrfs_set_lock_blocking_read(eb);
 322         }
 323
 324         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 325                          &cached_state);
 326         if (extent_buffer_uptodate(eb) &&
 327             btrfs_header_generation(eb) == parent_transid) {
 328                 ret = 0;
 329                 goto out;
 330         }
 331         btrfs_err_rl(eb->fs_info,
 332                 "parent transid verify failed on %llu wanted %llu found %llu",
 333                         eb->start,
 334                         parent_transid, btrfs_header_generation(eb));
 335         ret = 1;
 336
 337         /*
 338          * Things reading via commit roots that don't have normal protection,
 339          * like send, can have a really old block in cache that may point at a
 340          * block that has been freed and re-allocated.  So don't clear uptodate
 341          * if we find an eb that is under IO (dirty/writeback) because we could
 342          * end up reading in the stale data and then writing it back out and
 343          * making everybody very sad.
 344          */
 345         if (!extent_buffer_under_io(eb))
 346                 clear_extent_buffer_uptodate(eb);
 347 out:
 348         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 349                              &cached_state);
 350         if (need_lock)
 351                 btrfs_tree_read_unlock_blocking(eb);
 352         return ret;
 353 }
 354
 355 static bool btrfs_supported_super_csum(u16 csum_type)
 356 {
 357         switch (csum_type) {
 358         case BTRFS_CSUM_TYPE_CRC32:
 359                 return true;
 360         default:
 361                 return false;
 362         }
 363 }
 364
 365 /*
 366  * Return 0 if the superblock checksum type matches the checksum value of that
 367  * algorithm. Pass the raw disk superblock data.
 368  */
 369 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 370                                   char *raw_disk_sb)
 371 {
 372         struct btrfs_super_block *disk_sb =
 373                 (struct btrfs_super_block *)raw_disk_sb;
 374         u16 csum_type = btrfs_super_csum_type(disk_sb);
 375
 376         if (!btrfs_supported_super_csum(csum_type)) {
 377                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 378                           csum_type);
 379                 return 1;
 380         }
 381
 382         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 383                 u32 crc = ~(u32)0;
 384                 char result[sizeof(crc)];
 385
 386                 /*
 387                  * The super_block structure does not span the whole
 388                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 389                  * is filled with zeros and is included in the checksum.
 390                  */
 391                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 392                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 393                 btrfs_csum_final(crc, result);
 394
 395                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 396                         return 1;
 397         }
 398
 399         return 0;
 400 }
 401
 402 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 403                            struct btrfs_key *first_key, u64 parent_transid)
 404 {
 405         struct btrfs_fs_info *fs_info = eb->fs_info;
 406         int found_level;
 407         struct btrfs_key found_key;
 408         int ret;
 409
 410         found_level = btrfs_header_level(eb);
 411         if (found_level != level) {
 412                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 413                      KERN_ERR "BTRFS: tree level check failed\n");
 414                 btrfs_err(fs_info,
 415 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 416                           eb->start, level, found_level);
 417                 return -EIO;
 418         }
 419
 420         if (!first_key)
 421                 return 0;
 422
 423         /*
 424          * For live tree block (new tree blocks in current transaction),
 425          * we need proper lock context to avoid race, which is impossible here.
 426          * So we only checks tree blocks which is read from disk, whose
 427          * generation <= fs_info->last_trans_committed.
 428          */
 429         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 430                 return 0;
 431         if (found_level)
 432                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 433         else
 434                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 435         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 436
 437         if (ret) {
 438                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 439                      KERN_ERR "BTRFS: tree first key check failed\n");
 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, parent_transid, first_key->objectid,
 443                           first_key->type, first_key->offset,
 444                           found_key.objectid, found_key.type,
 445                           found_key.offset);
 446         }
 447         return ret;
 448 }
 449
 450 /*
 451  * helper to read a given tree block, doing retries as required when
 452  * the checksums don't match and we have alternate mirrors to try.
 453  *
 454  * @parent_transid:     expected transid, skip check if 0
 455  * @level:              expected level, mandatory check
 456  * @first_key:          expected key of first slot, skip check if NULL
 457  */
 458 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 459                                           u64 parent_transid, int level,
 460                                           struct btrfs_key *first_key)
 461 {
 462         struct btrfs_fs_info *fs_info = eb->fs_info;
 463         struct extent_io_tree *io_tree;
 464         int failed = 0;
 465         int ret;
 466         int num_copies = 0;
 467         int mirror_num = 0;
 468         int failed_mirror = 0;
 469
 470         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 471         while (1) {
 472                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 473                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 474                 if (!ret) {
 475                         if (verify_parent_transid(io_tree, eb,
 476                                                    parent_transid, 0))
 477                                 ret = -EIO;
 478                         else if (btrfs_verify_level_key(eb, level,
 479                                                 first_key, parent_transid))
 480                                 ret = -EUCLEAN;
 481                         else
 482                                 break;
 483                 }
 484
 485                 num_copies = btrfs_num_copies(fs_info,
 486                                               eb->start, eb->len);
 487                 if (num_copies == 1)
 488                         break;
 489
 490                 if (!failed_mirror) {
 491                         failed = 1;
 492                         failed_mirror = eb->read_mirror;
 493                 }
 494
 495                 mirror_num++;
 496                 if (mirror_num == failed_mirror)
 497                         mirror_num++;
 498
 499                 if (mirror_num > num_copies)
 500                         break;
 501         }
 502
 503         if (failed && !ret && failed_mirror)
 504                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 505
 506         return ret;
 507 }
 508
 509 /*
 510  * checksum a dirty tree block before IO.  This has extra checks to make sure
 511  * we only fill in the checksum field in the first page of a multi-page block
 512  */
 513
 514 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 515 {
 516         u64 start = page_offset(page);
 517         u64 found_start;
 518         u8 result[BTRFS_CSUM_SIZE];
 519         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 520         struct extent_buffer *eb;
 521         int ret;
 522
 523         eb = (struct extent_buffer *)page->private;
 524         if (page != eb->pages[0])
 525                 return 0;
 526
 527         found_start = btrfs_header_bytenr(eb);
 528         /*
 529          * Please do not consolidate these warnings into a single if.
 530          * It is useful to know what went wrong.
 531          */
 532         if (WARN_ON(found_start != start))
 533                 return -EUCLEAN;
 534         if (WARN_ON(!PageUptodate(page)))
 535                 return -EUCLEAN;
 536
 537         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 538                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 539
 540         if (csum_tree_block(eb, result))
 541                 return -EINVAL;
 542
 543         if (btrfs_header_level(eb))
 544                 ret = btrfs_check_node(eb);
 545         else
 546                 ret = btrfs_check_leaf_full(eb);
 547
 548         if (ret < 0) {
 549                 btrfs_err(fs_info,
 550                 "block=%llu write time tree block corruption detected",
 551                           eb->start);
 552                 return ret;
 553         }
 554         write_extent_buffer(eb, result, 0, csum_size);
 555
 556         return 0;
 557 }
 558
 559 static int check_tree_block_fsid(struct extent_buffer *eb)
 560 {
 561         struct btrfs_fs_info *fs_info = eb->fs_info;
 562         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 563         u8 fsid[BTRFS_FSID_SIZE];
 564         int ret = 1;
 565
 566         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 567         while (fs_devices) {
 568                 u8 *metadata_uuid;
 569
 570                 /*
 571                  * Checking the incompat flag is only valid for the current
 572                  * fs. For seed devices it's forbidden to have their uuid
 573                  * changed so reading ->fsid in this case is fine
 574                  */
 575                 if (fs_devices == fs_info->fs_devices &&
 576                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 577                         metadata_uuid = fs_devices->metadata_uuid;
 578                 else
 579                         metadata_uuid = fs_devices->fsid;
 580
 581                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 582                         ret = 0;
 583                         break;
 584                 }
 585                 fs_devices = fs_devices->seed;
 586         }
 587         return ret;
 588 }
 589
 590 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 591                                       u64 phy_offset, struct page *page,
 592                                       u64 start, u64 end, int mirror)
 593 {
 594         u64 found_start;
 595         int found_level;
 596         struct extent_buffer *eb;
 597         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 598         struct btrfs_fs_info *fs_info = root->fs_info;
 599         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 600         int ret = 0;
 601         u8 result[BTRFS_CSUM_SIZE];
 602         int reads_done;
 603
 604         if (!page->private)
 605                 goto out;
 606
 607         eb = (struct extent_buffer *)page->private;
 608
 609         /* the pending IO might have been the only thing that kept this buffer
 610          * in memory.  Make sure we have a ref for all this other checks
 611          */
 612         extent_buffer_get(eb);
 613
 614         reads_done = atomic_dec_and_test(&eb->io_pages);
 615         if (!reads_done)
 616                 goto err;
 617
 618         eb->read_mirror = mirror;
 619         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 620                 ret = -EIO;
 621                 goto err;
 622         }
 623
 624         found_start = btrfs_header_bytenr(eb);
 625         if (found_start != eb->start) {
 626                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 627                              eb->start, found_start);
 628                 ret = -EIO;
 629                 goto err;
 630         }
 631         if (check_tree_block_fsid(eb)) {
 632                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 633                              eb->start);
 634                 ret = -EIO;
 635                 goto err;
 636         }
 637         found_level = btrfs_header_level(eb);
 638         if (found_level >= BTRFS_MAX_LEVEL) {
 639                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 640                           (int)btrfs_header_level(eb), eb->start);
 641                 ret = -EIO;
 642                 goto err;
 643         }
 644
 645         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 646                                        eb, found_level);
 647
 648         ret = csum_tree_block(eb, result);
 649         if (ret)
 650                 goto err;
 651
 652         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 653                 u32 val;
 654                 u32 found = 0;
 655
 656                 memcpy(&found, result, csum_size);
 657
 658                 read_extent_buffer(eb, &val, 0, csum_size);
 659                 btrfs_warn_rl(fs_info,
 660                 "%s checksum verify failed on %llu wanted %x found %x level %d",
 661                               fs_info->sb->s_id, eb->start,
 662                               val, found, btrfs_header_level(eb));
 663                 ret = -EUCLEAN;
 664                 goto err;
 665         }
 666
 667         /*
 668          * If this is a leaf block and it is corrupt, set the corrupt bit so
 669          * that we don't try and read the other copies of this block, just
 670          * return -EIO.
 671          */
 672         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 673                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 674                 ret = -EIO;
 675         }
 676
 677         if (found_level > 0 && btrfs_check_node(eb))
 678                 ret = -EIO;
 679
 680         if (!ret)
 681                 set_extent_buffer_uptodate(eb);
 682         else
 683                 btrfs_err(fs_info,
 684                           "block=%llu read time tree block corruption detected",
 685                           eb->start);
 686 err:
 687         if (reads_done &&
 688             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 689                 btree_readahead_hook(eb, ret);
 690
 691         if (ret) {
 692                 /*
 693                  * our io error hook is going to dec the io pages
 694                  * again, we have to make sure it has something
 695                  * to decrement
 696                  */
 697                 atomic_inc(&eb->io_pages);
 698                 clear_extent_buffer_uptodate(eb);
 699         }
 700         free_extent_buffer(eb);
 701 out:
 702         return ret;
 703 }
 704
 705 static void end_workqueue_bio(struct bio *bio)
 706 {
 707         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 708         struct btrfs_fs_info *fs_info;
 709         struct btrfs_workqueue *wq;
 710         btrfs_work_func_t func;
 711
 712         fs_info = end_io_wq->info;
 713         end_io_wq->status = bio->bi_status;
 714
 715         if (bio_op(bio) == REQ_OP_WRITE) {
 716                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 717                         wq = fs_info->endio_meta_write_workers;
 718                         func = btrfs_endio_meta_write_helper;
 719                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 720                         wq = fs_info->endio_freespace_worker;
 721                         func = btrfs_freespace_write_helper;
 722                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 723                         wq = fs_info->endio_raid56_workers;
 724                         func = btrfs_endio_raid56_helper;
 725                 } else {
 726                         wq = fs_info->endio_write_workers;
 727                         func = btrfs_endio_write_helper;
 728                 }
 729         } else {
 730                 if (unlikely(end_io_wq->metadata ==
 731                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 732                         wq = fs_info->endio_repair_workers;
 733                         func = btrfs_endio_repair_helper;
 734                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 735                         wq = fs_info->endio_raid56_workers;
 736                         func = btrfs_endio_raid56_helper;
 737                 } else if (end_io_wq->metadata) {
 738                         wq = fs_info->endio_meta_workers;
 739                         func = btrfs_endio_meta_helper;
 740                 } else {
 741                         wq = fs_info->endio_workers;
 742                         func = btrfs_endio_helper;
 743                 }
 744         }
 745
 746         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 747         btrfs_queue_work(wq, &end_io_wq->work);
 748 }
 749
 750 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 751                         enum btrfs_wq_endio_type metadata)
 752 {
 753         struct btrfs_end_io_wq *end_io_wq;
 754
 755         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 756         if (!end_io_wq)
 757                 return BLK_STS_RESOURCE;
 758
 759         end_io_wq->private = bio->bi_private;
 760         end_io_wq->end_io = bio->bi_end_io;
 761         end_io_wq->info = info;
 762         end_io_wq->status = 0;
 763         end_io_wq->bio = bio;
 764         end_io_wq->metadata = metadata;
 765
 766         bio->bi_private = end_io_wq;
 767         bio->bi_end_io = end_workqueue_bio;
 768         return 0;
 769 }
 770
 771 static void run_one_async_start(struct btrfs_work *work)
 772 {
 773         struct async_submit_bio *async;
 774         blk_status_t ret;
 775
 776         async = container_of(work, struct  async_submit_bio, work);
 777         ret = async->submit_bio_start(async->private_data, async->bio,
 778                                       async->bio_offset);
 779         if (ret)
 780                 async->status = ret;
 781 }
 782
 783 /*
 784  * In order to insert checksums into the metadata in large chunks, we wait
 785  * until bio submission time.   All the pages in the bio are checksummed and
 786  * sums are attached onto the ordered extent record.
 787  *
 788  * At IO completion time the csums attached on the ordered extent record are
 789  * inserted into the tree.
 790  */
 791 static void run_one_async_done(struct btrfs_work *work)
 792 {
 793         struct async_submit_bio *async;
 794         struct inode *inode;
 795         blk_status_t ret;
 796
 797         async = container_of(work, struct  async_submit_bio, work);
 798         inode = async->private_data;
 799
 800         /* If an error occurred we just want to clean up the bio and move on */
 801         if (async->status) {
 802                 async->bio->bi_status = async->status;
 803                 bio_endio(async->bio);
 804                 return;
 805         }
 806
 807         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 808                         async->mirror_num, 1);
 809         if (ret) {
 810                 async->bio->bi_status = ret;
 811                 bio_endio(async->bio);
 812         }
 813 }
 814
 815 static void run_one_async_free(struct btrfs_work *work)
 816 {
 817         struct async_submit_bio *async;
 818
 819         async = container_of(work, struct  async_submit_bio, work);
 820         kfree(async);
 821 }
 822
 823 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 824                                  int mirror_num, unsigned long bio_flags,
 825                                  u64 bio_offset, void *private_data,
 826                                  extent_submit_bio_start_t *submit_bio_start)
 827 {
 828         struct async_submit_bio *async;
 829
 830         async = kmalloc(sizeof(*async), GFP_NOFS);
 831         if (!async)
 832                 return BLK_STS_RESOURCE;
 833
 834         async->private_data = private_data;
 835         async->bio = bio;
 836         async->mirror_num = mirror_num;
 837         async->submit_bio_start = submit_bio_start;
 838
 839         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 840                         run_one_async_done, run_one_async_free);
 841
 842         async->bio_offset = bio_offset;
 843
 844         async->status = 0;
 845
 846         if (op_is_sync(bio->bi_opf))
 847                 btrfs_set_work_high_priority(&async->work);
 848
 849         btrfs_queue_work(fs_info->workers, &async->work);
 850         return 0;
 851 }
 852
 853 static blk_status_t btree_csum_one_bio(struct bio *bio)
 854 {
 855         struct bio_vec *bvec;
 856         struct btrfs_root *root;
 857         int ret = 0;
 858         struct bvec_iter_all iter_all;
 859
 860         ASSERT(!bio_flagged(bio, BIO_CLONED));
 861         bio_for_each_segment_all(bvec, bio, iter_all) {
 862                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 863                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 864                 if (ret)
 865                         break;
 866         }
 867
 868         return errno_to_blk_status(ret);
 869 }
 870
 871 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 872                                              u64 bio_offset)
 873 {
 874         /*
 875          * when we're called for a write, we're already in the async
 876          * submission context.  Just jump into btrfs_map_bio
 877          */
 878         return btree_csum_one_bio(bio);
 879 }
 880
 881 static int check_async_write(struct btrfs_fs_info *fs_info,
 882                              struct btrfs_inode *bi)
 883 {
 884         if (atomic_read(&bi->sync_writers))
 885                 return 0;
 886         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 887                 return 0;
 888         return 1;
 889 }
 890
 891 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 892                                           int mirror_num,
 893                                           unsigned long bio_flags)
 894 {
 895         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 896         int async = check_async_write(fs_info, BTRFS_I(inode));
 897         blk_status_t ret;
 898
 899         if (bio_op(bio) != REQ_OP_WRITE) {
 900                 /*
 901                  * called for a read, do the setup so that checksum validation
 902                  * can happen in the async kernel threads
 903                  */
 904                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 905                                           BTRFS_WQ_ENDIO_METADATA);
 906                 if (ret)
 907                         goto out_w_error;
 908                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 909         } else if (!async) {
 910                 ret = btree_csum_one_bio(bio);
 911                 if (ret)
 912                         goto out_w_error;
 913                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 914         } else {
 915                 /*
 916                  * kthread helpers are used to submit writes so that
 917                  * checksumming can happen in parallel across all CPUs
 918                  */
 919                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 920                                           0, inode, btree_submit_bio_start);
 921         }
 922
 923         if (ret)
 924                 goto out_w_error;
 925         return 0;
 926
 927 out_w_error:
 928         bio->bi_status = ret;
 929         bio_endio(bio);
 930         return ret;
 931 }
 932
 933 #ifdef CONFIG_MIGRATION
 934 static int btree_migratepage(struct address_space *mapping,
 935                         struct page *newpage, struct page *page,
 936                         enum migrate_mode mode)
 937 {
 938         /*
 939          * we can't safely write a btree page from here,
 940          * we haven't done the locking hook
 941          */
 942         if (PageDirty(page))
 943                 return -EAGAIN;
 944         /*
 945          * Buffers may be managed in a filesystem specific way.
 946          * We must have no buffers or drop them.
 947          */
 948         if (page_has_private(page) &&
 949             !try_to_release_page(page, GFP_KERNEL))
 950                 return -EAGAIN;
 951         return migrate_page(mapping, newpage, page, mode);
 952 }
 953 #endif
 954
 955
 956 static int btree_writepages(struct address_space *mapping,
 957                             struct writeback_control *wbc)
 958 {
 959         struct btrfs_fs_info *fs_info;
 960         int ret;
 961
 962         if (wbc->sync_mode == WB_SYNC_NONE) {
 963
 964                 if (wbc->for_kupdate)
 965                         return 0;
 966
 967                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 968                 /* this is a bit racy, but that's ok */
 969                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 970                                              BTRFS_DIRTY_METADATA_THRESH,
 971                                              fs_info->dirty_metadata_batch);
 972                 if (ret < 0)
 973                         return 0;
 974         }
 975         return btree_write_cache_pages(mapping, wbc);
 976 }
 977
 978 static int btree_readpage(struct file *file, struct page *page)
 979 {
 980         struct extent_io_tree *tree;
 981         tree = &BTRFS_I(page->mapping->host)->io_tree;
 982         return extent_read_full_page(tree, page, btree_get_extent, 0);
 983 }
 984
 985 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 986 {
 987         if (PageWriteback(page) || PageDirty(page))
 988                 return 0;
 989
 990         return try_release_extent_buffer(page);
 991 }
 992
 993 static void btree_invalidatepage(struct page *page, unsigned int offset,
 994                                  unsigned int length)
 995 {
 996         struct extent_io_tree *tree;
 997         tree = &BTRFS_I(page->mapping->host)->io_tree;
 998         extent_invalidatepage(tree, page, offset);
 999         btree_releasepage(page, GFP_NOFS);
1000         if (PagePrivate(page)) {
1001                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1002                            "page private not zero on page %llu",
1003                            (unsigned long long)page_offset(page));
1004                 ClearPagePrivate(page);
1005                 set_page_private(page, 0);
1006                 put_page(page);
1007         }
1008 }
1009
1010 static int btree_set_page_dirty(struct page *page)
1011 {
1012 #ifdef DEBUG
1013         struct extent_buffer *eb;
1014
1015         BUG_ON(!PagePrivate(page));
1016         eb = (struct extent_buffer *)page->private;
1017         BUG_ON(!eb);
1018         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1019         BUG_ON(!atomic_read(&eb->refs));
1020         btrfs_assert_tree_locked(eb);
1021 #endif
1022         return __set_page_dirty_nobuffers(page);
1023 }
1024
1025 static const struct address_space_operations btree_aops = {
1026         .readpage       = btree_readpage,
1027         .writepages     = btree_writepages,
1028         .releasepage    = btree_releasepage,
1029         .invalidatepage = btree_invalidatepage,
1030 #ifdef CONFIG_MIGRATION
1031         .migratepage    = btree_migratepage,
1032 #endif
1033         .set_page_dirty = btree_set_page_dirty,
1034 };
1035
1036 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1037 {
1038         struct extent_buffer *buf = NULL;
1039         int ret;
1040
1041         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1042         if (IS_ERR(buf))
1043                 return;
1044
1045         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1046         if (ret < 0)
1047                 free_extent_buffer_stale(buf);
1048         else
1049                 free_extent_buffer(buf);
1050 }
1051
1052 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1053                          int mirror_num, struct extent_buffer **eb)
1054 {
1055         struct extent_buffer *buf = NULL;
1056         int ret;
1057
1058         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1059         if (IS_ERR(buf))
1060                 return 0;
1061
1062         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1063
1064         ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num);
1065         if (ret) {
1066                 free_extent_buffer_stale(buf);
1067                 return ret;
1068         }
1069
1070         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1071                 free_extent_buffer_stale(buf);
1072                 return -EIO;
1073         } else if (extent_buffer_uptodate(buf)) {
1074                 *eb = buf;
1075         } else {
1076                 free_extent_buffer(buf);
1077         }
1078         return 0;
1079 }
1080
1081 struct extent_buffer *btrfs_find_create_tree_block(
1082                                                 struct btrfs_fs_info *fs_info,
1083                                                 u64 bytenr)
1084 {
1085         if (btrfs_is_testing(fs_info))
1086                 return alloc_test_extent_buffer(fs_info, bytenr);
1087         return alloc_extent_buffer(fs_info, bytenr);
1088 }
1089
1090 /*
1091  * Read tree block at logical address @bytenr and do variant basic but critical
1092  * verification.
1093  *
1094  * @parent_transid:     expected transid of this tree block, skip check if 0
1095  * @level:              expected level, mandatory check
1096  * @first_key:          expected key in slot 0, skip check if NULL
1097  */
1098 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1099                                       u64 parent_transid, int level,
1100                                       struct btrfs_key *first_key)
1101 {
1102         struct extent_buffer *buf = NULL;
1103         int ret;
1104
1105         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1106         if (IS_ERR(buf))
1107                 return buf;
1108
1109         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1110                                              level, first_key);
1111         if (ret) {
1112                 free_extent_buffer_stale(buf);
1113                 return ERR_PTR(ret);
1114         }
1115         return buf;
1116
1117 }
1118
1119 void btrfs_clean_tree_block(struct extent_buffer *buf)
1120 {
1121         struct btrfs_fs_info *fs_info = buf->fs_info;
1122         if (btrfs_header_generation(buf) ==
1123             fs_info->running_transaction->transid) {
1124                 btrfs_assert_tree_locked(buf);
1125
1126                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1127                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1128                                                  -buf->len,
1129                                                  fs_info->dirty_metadata_batch);
1130                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1131                         btrfs_set_lock_blocking_write(buf);
1132                         clear_extent_buffer_dirty(buf);
1133                 }
1134         }
1135 }
1136
1137 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1138 {
1139         struct btrfs_subvolume_writers *writers;
1140         int ret;
1141
1142         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1143         if (!writers)
1144                 return ERR_PTR(-ENOMEM);
1145
1146         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1147         if (ret < 0) {
1148                 kfree(writers);
1149                 return ERR_PTR(ret);
1150         }
1151
1152         init_waitqueue_head(&writers->wait);
1153         return writers;
1154 }
1155
1156 static void
1157 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1158 {
1159         percpu_counter_destroy(&writers->counter);
1160         kfree(writers);
1161 }
1162
1163 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1164                          u64 objectid)
1165 {
1166         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1167         root->node = NULL;
1168         root->commit_root = NULL;
1169         root->state = 0;
1170         root->orphan_cleanup_state = 0;
1171
1172         root->last_trans = 0;
1173         root->highest_objectid = 0;
1174         root->nr_delalloc_inodes = 0;
1175         root->nr_ordered_extents = 0;
1176         root->inode_tree = RB_ROOT;
1177         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1178         root->block_rsv = NULL;
1179
1180         INIT_LIST_HEAD(&root->dirty_list);
1181         INIT_LIST_HEAD(&root->root_list);
1182         INIT_LIST_HEAD(&root->delalloc_inodes);
1183         INIT_LIST_HEAD(&root->delalloc_root);
1184         INIT_LIST_HEAD(&root->ordered_extents);
1185         INIT_LIST_HEAD(&root->ordered_root);
1186         INIT_LIST_HEAD(&root->reloc_dirty_list);
1187         INIT_LIST_HEAD(&root->logged_list[0]);
1188         INIT_LIST_HEAD(&root->logged_list[1]);
1189         spin_lock_init(&root->inode_lock);
1190         spin_lock_init(&root->delalloc_lock);
1191         spin_lock_init(&root->ordered_extent_lock);
1192         spin_lock_init(&root->accounting_lock);
1193         spin_lock_init(&root->log_extents_lock[0]);
1194         spin_lock_init(&root->log_extents_lock[1]);
1195         spin_lock_init(&root->qgroup_meta_rsv_lock);
1196         mutex_init(&root->objectid_mutex);
1197         mutex_init(&root->log_mutex);
1198         mutex_init(&root->ordered_extent_mutex);
1199         mutex_init(&root->delalloc_mutex);
1200         init_waitqueue_head(&root->log_writer_wait);
1201         init_waitqueue_head(&root->log_commit_wait[0]);
1202         init_waitqueue_head(&root->log_commit_wait[1]);
1203         INIT_LIST_HEAD(&root->log_ctxs[0]);
1204         INIT_LIST_HEAD(&root->log_ctxs[1]);
1205         atomic_set(&root->log_commit[0], 0);
1206         atomic_set(&root->log_commit[1], 0);
1207         atomic_set(&root->log_writers, 0);
1208         atomic_set(&root->log_batch, 0);
1209         refcount_set(&root->refs, 1);
1210         atomic_set(&root->will_be_snapshotted, 0);
1211         atomic_set(&root->snapshot_force_cow, 0);
1212         atomic_set(&root->nr_swapfiles, 0);
1213         root->log_transid = 0;
1214         root->log_transid_committed = -1;
1215         root->last_log_commit = 0;
1216         if (!dummy)
1217                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1218                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1219
1220         memset(&root->root_key, 0, sizeof(root->root_key));
1221         memset(&root->root_item, 0, sizeof(root->root_item));
1222         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1223         if (!dummy)
1224                 root->defrag_trans_start = fs_info->generation;
1225         else
1226                 root->defrag_trans_start = 0;
1227         root->root_key.objectid = objectid;
1228         root->anon_dev = 0;
1229
1230         spin_lock_init(&root->root_item_lock);
1231         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1232 }
1233
1234 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1235                 gfp_t flags)
1236 {
1237         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1238         if (root)
1239                 root->fs_info = fs_info;
1240         return root;
1241 }
1242
1243 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1244 /* Should only be used by the testing infrastructure */
1245 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1246 {
1247         struct btrfs_root *root;
1248
1249         if (!fs_info)
1250                 return ERR_PTR(-EINVAL);
1251
1252         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1253         if (!root)
1254                 return ERR_PTR(-ENOMEM);
1255
1256         /* We don't use the stripesize in selftest, set it as sectorsize */
1257         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1258         root->alloc_bytenr = 0;
1259
1260         return root;
1261 }
1262 #endif
1263
1264 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1265                                      u64 objectid)
1266 {
1267         struct btrfs_fs_info *fs_info = trans->fs_info;
1268         struct extent_buffer *leaf;
1269         struct btrfs_root *tree_root = fs_info->tree_root;
1270         struct btrfs_root *root;
1271         struct btrfs_key key;
1272         unsigned int nofs_flag;
1273         int ret = 0;
1274         uuid_le uuid = NULL_UUID_LE;
1275
1276         /*
1277          * We're holding a transaction handle, so use a NOFS memory allocation
1278          * context to avoid deadlock if reclaim happens.
1279          */
1280         nofs_flag = memalloc_nofs_save();
1281         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1282         memalloc_nofs_restore(nofs_flag);
1283         if (!root)
1284                 return ERR_PTR(-ENOMEM);
1285
1286         __setup_root(root, fs_info, objectid);
1287         root->root_key.objectid = objectid;
1288         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289         root->root_key.offset = 0;
1290
1291         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1292         if (IS_ERR(leaf)) {
1293                 ret = PTR_ERR(leaf);
1294                 leaf = NULL;
1295                 goto fail;
1296         }
1297
1298         root->node = leaf;
1299         btrfs_mark_buffer_dirty(leaf);
1300
1301         root->commit_root = btrfs_root_node(root);
1302         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1303
1304         root->root_item.flags = 0;
1305         root->root_item.byte_limit = 0;
1306         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1307         btrfs_set_root_generation(&root->root_item, trans->transid);
1308         btrfs_set_root_level(&root->root_item, 0);
1309         btrfs_set_root_refs(&root->root_item, 1);
1310         btrfs_set_root_used(&root->root_item, leaf->len);
1311         btrfs_set_root_last_snapshot(&root->root_item, 0);
1312         btrfs_set_root_dirid(&root->root_item, 0);
1313         if (is_fstree(objectid))
1314                 uuid_le_gen(&uuid);
1315         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1316         root->root_item.drop_level = 0;
1317
1318         key.objectid = objectid;
1319         key.type = BTRFS_ROOT_ITEM_KEY;
1320         key.offset = 0;
1321         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1322         if (ret)
1323                 goto fail;
1324
1325         btrfs_tree_unlock(leaf);
1326
1327         return root;
1328
1329 fail:
1330         if (leaf) {
1331                 btrfs_tree_unlock(leaf);
1332                 free_extent_buffer(root->commit_root);
1333                 free_extent_buffer(leaf);
1334         }
1335         kfree(root);
1336
1337         return ERR_PTR(ret);
1338 }
1339
1340 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1341                                          struct btrfs_fs_info *fs_info)
1342 {
1343         struct btrfs_root *root;
1344         struct extent_buffer *leaf;
1345
1346         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1347         if (!root)
1348                 return ERR_PTR(-ENOMEM);
1349
1350         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1351
1352         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1353         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1354         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1355
1356         /*
1357          * DON'T set REF_COWS for log trees
1358          *
1359          * log trees do not get reference counted because they go away
1360          * before a real commit is actually done.  They do store pointers
1361          * to file data extents, and those reference counts still get
1362          * updated (along with back refs to the log tree).
1363          */
1364
1365         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1366                         NULL, 0, 0, 0);
1367         if (IS_ERR(leaf)) {
1368                 kfree(root);
1369                 return ERR_CAST(leaf);
1370         }
1371
1372         root->node = leaf;
1373
1374         btrfs_mark_buffer_dirty(root->node);
1375         btrfs_tree_unlock(root->node);
1376         return root;
1377 }
1378
1379 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1380                              struct btrfs_fs_info *fs_info)
1381 {
1382         struct btrfs_root *log_root;
1383
1384         log_root = alloc_log_tree(trans, fs_info);
1385         if (IS_ERR(log_root))
1386                 return PTR_ERR(log_root);
1387         WARN_ON(fs_info->log_root_tree);
1388         fs_info->log_root_tree = log_root;
1389         return 0;
1390 }
1391
1392 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1393                        struct btrfs_root *root)
1394 {
1395         struct btrfs_fs_info *fs_info = root->fs_info;
1396         struct btrfs_root *log_root;
1397         struct btrfs_inode_item *inode_item;
1398
1399         log_root = alloc_log_tree(trans, fs_info);
1400         if (IS_ERR(log_root))
1401                 return PTR_ERR(log_root);
1402
1403         log_root->last_trans = trans->transid;
1404         log_root->root_key.offset = root->root_key.objectid;
1405
1406         inode_item = &log_root->root_item.inode;
1407         btrfs_set_stack_inode_generation(inode_item, 1);
1408         btrfs_set_stack_inode_size(inode_item, 3);
1409         btrfs_set_stack_inode_nlink(inode_item, 1);
1410         btrfs_set_stack_inode_nbytes(inode_item,
1411                                      fs_info->nodesize);
1412         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1413
1414         btrfs_set_root_node(&log_root->root_item, log_root->node);
1415
1416         WARN_ON(root->log_root);
1417         root->log_root = log_root;
1418         root->log_transid = 0;
1419         root->log_transid_committed = -1;
1420         root->last_log_commit = 0;
1421         return 0;
1422 }
1423
1424 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1425                                                struct btrfs_key *key)
1426 {
1427         struct btrfs_root *root;
1428         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1429         struct btrfs_path *path;
1430         u64 generation;
1431         int ret;
1432         int level;
1433
1434         path = btrfs_alloc_path();
1435         if (!path)
1436                 return ERR_PTR(-ENOMEM);
1437
1438         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1439         if (!root) {
1440                 ret = -ENOMEM;
1441                 goto alloc_fail;
1442         }
1443
1444         __setup_root(root, fs_info, key->objectid);
1445
1446         ret = btrfs_find_root(tree_root, key, path,
1447                               &root->root_item, &root->root_key);
1448         if (ret) {
1449                 if (ret > 0)
1450                         ret = -ENOENT;
1451                 goto find_fail;
1452         }
1453
1454         generation = btrfs_root_generation(&root->root_item);
1455         level = btrfs_root_level(&root->root_item);
1456         root->node = read_tree_block(fs_info,
1457                                      btrfs_root_bytenr(&root->root_item),
1458                                      generation, level, NULL);
1459         if (IS_ERR(root->node)) {
1460                 ret = PTR_ERR(root->node);
1461                 goto find_fail;
1462         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1463                 ret = -EIO;
1464                 free_extent_buffer(root->node);
1465                 goto find_fail;
1466         }
1467         root->commit_root = btrfs_root_node(root);
1468 out:
1469         btrfs_free_path(path);
1470         return root;
1471
1472 find_fail:
1473         kfree(root);
1474 alloc_fail:
1475         root = ERR_PTR(ret);
1476         goto out;
1477 }
1478
1479 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1480                                       struct btrfs_key *location)
1481 {
1482         struct btrfs_root *root;
1483
1484         root = btrfs_read_tree_root(tree_root, location);
1485         if (IS_ERR(root))
1486                 return root;
1487
1488         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1489                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1490                 btrfs_check_and_init_root_item(&root->root_item);
1491         }
1492
1493         return root;
1494 }
1495
1496 int btrfs_init_fs_root(struct btrfs_root *root)
1497 {
1498         int ret;
1499         struct btrfs_subvolume_writers *writers;
1500
1501         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1502         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1503                                         GFP_NOFS);
1504         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1505                 ret = -ENOMEM;
1506                 goto fail;
1507         }
1508
1509         writers = btrfs_alloc_subvolume_writers();
1510         if (IS_ERR(writers)) {
1511                 ret = PTR_ERR(writers);
1512                 goto fail;
1513         }
1514         root->subv_writers = writers;
1515
1516         btrfs_init_free_ino_ctl(root);
1517         spin_lock_init(&root->ino_cache_lock);
1518         init_waitqueue_head(&root->ino_cache_wait);
1519
1520         ret = get_anon_bdev(&root->anon_dev);
1521         if (ret)
1522                 goto fail;
1523
1524         mutex_lock(&root->objectid_mutex);
1525         ret = btrfs_find_highest_objectid(root,
1526                                         &root->highest_objectid);
1527         if (ret) {
1528                 mutex_unlock(&root->objectid_mutex);
1529                 goto fail;
1530         }
1531
1532         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1533
1534         mutex_unlock(&root->objectid_mutex);
1535
1536         return 0;
1537 fail:
1538         /* The caller is responsible to call btrfs_free_fs_root */
1539         return ret;
1540 }
1541
1542 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1543                                         u64 root_id)
1544 {
1545         struct btrfs_root *root;
1546
1547         spin_lock(&fs_info->fs_roots_radix_lock);
1548         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1549                                  (unsigned long)root_id);
1550         spin_unlock(&fs_info->fs_roots_radix_lock);
1551         return root;
1552 }
1553
1554 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1555                          struct btrfs_root *root)
1556 {
1557         int ret;
1558
1559         ret = radix_tree_preload(GFP_NOFS);
1560         if (ret)
1561                 return ret;
1562
1563         spin_lock(&fs_info->fs_roots_radix_lock);
1564         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1565                                 (unsigned long)root->root_key.objectid,
1566                                 root);
1567         if (ret == 0)
1568                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1569         spin_unlock(&fs_info->fs_roots_radix_lock);
1570         radix_tree_preload_end();
1571
1572         return ret;
1573 }
1574
1575 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1576                                      struct btrfs_key *location,
1577                                      bool check_ref)
1578 {
1579         struct btrfs_root *root;
1580         struct btrfs_path *path;
1581         struct btrfs_key key;
1582         int ret;
1583
1584         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1585                 return fs_info->tree_root;
1586         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1587                 return fs_info->extent_root;
1588         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1589                 return fs_info->chunk_root;
1590         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1591                 return fs_info->dev_root;
1592         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1593                 return fs_info->csum_root;
1594         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1595                 return fs_info->quota_root ? fs_info->quota_root :
1596                                              ERR_PTR(-ENOENT);
1597         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1598                 return fs_info->uuid_root ? fs_info->uuid_root :
1599                                             ERR_PTR(-ENOENT);
1600         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1601                 return fs_info->free_space_root ? fs_info->free_space_root :
1602                                                   ERR_PTR(-ENOENT);
1603 again:
1604         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1605         if (root) {
1606                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1607                         return ERR_PTR(-ENOENT);
1608                 return root;
1609         }
1610
1611         root = btrfs_read_fs_root(fs_info->tree_root, location);
1612         if (IS_ERR(root))
1613                 return root;
1614
1615         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1616                 ret = -ENOENT;
1617                 goto fail;
1618         }
1619
1620         ret = btrfs_init_fs_root(root);
1621         if (ret)
1622                 goto fail;
1623
1624         path = btrfs_alloc_path();
1625         if (!path) {
1626                 ret = -ENOMEM;
1627                 goto fail;
1628         }
1629         key.objectid = BTRFS_ORPHAN_OBJECTID;
1630         key.type = BTRFS_ORPHAN_ITEM_KEY;
1631         key.offset = location->objectid;
1632
1633         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1634         btrfs_free_path(path);
1635         if (ret < 0)
1636                 goto fail;
1637         if (ret == 0)
1638                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1639
1640         ret = btrfs_insert_fs_root(fs_info, root);
1641         if (ret) {
1642                 if (ret == -EEXIST) {
1643                         btrfs_free_fs_root(root);
1644                         goto again;
1645                 }
1646                 goto fail;
1647         }
1648         return root;
1649 fail:
1650         btrfs_free_fs_root(root);
1651         return ERR_PTR(ret);
1652 }
1653
1654 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1655 {
1656         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1657         int ret = 0;
1658         struct btrfs_device *device;
1659         struct backing_dev_info *bdi;
1660
1661         rcu_read_lock();
1662         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1663                 if (!device->bdev)
1664                         continue;
1665                 bdi = device->bdev->bd_bdi;
1666                 if (bdi_congested(bdi, bdi_bits)) {
1667                         ret = 1;
1668                         break;
1669                 }
1670         }
1671         rcu_read_unlock();
1672         return ret;
1673 }
1674
1675 /*
1676  * called by the kthread helper functions to finally call the bio end_io
1677  * functions.  This is where read checksum verification actually happens
1678  */
1679 static void end_workqueue_fn(struct btrfs_work *work)
1680 {
1681         struct bio *bio;
1682         struct btrfs_end_io_wq *end_io_wq;
1683
1684         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1685         bio = end_io_wq->bio;
1686
1687         bio->bi_status = end_io_wq->status;
1688         bio->bi_private = end_io_wq->private;
1689         bio->bi_end_io = end_io_wq->end_io;
1690         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1691         bio_endio(bio);
1692 }
1693
1694 static int cleaner_kthread(void *arg)
1695 {
1696         struct btrfs_root *root = arg;
1697         struct btrfs_fs_info *fs_info = root->fs_info;
1698         int again;
1699
1700         while (1) {
1701                 again = 0;
1702
1703                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1704
1705                 /* Make the cleaner go to sleep early. */
1706                 if (btrfs_need_cleaner_sleep(fs_info))
1707                         goto sleep;
1708
1709                 /*
1710                  * Do not do anything if we might cause open_ctree() to block
1711                  * before we have finished mounting the filesystem.
1712                  */
1713                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1714                         goto sleep;
1715
1716                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1717                         goto sleep;
1718
1719                 /*
1720                  * Avoid the problem that we change the status of the fs
1721                  * during the above check and trylock.
1722                  */
1723                 if (btrfs_need_cleaner_sleep(fs_info)) {
1724                         mutex_unlock(&fs_info->cleaner_mutex);
1725                         goto sleep;
1726                 }
1727
1728                 btrfs_run_delayed_iputs(fs_info);
1729
1730                 again = btrfs_clean_one_deleted_snapshot(root);
1731                 mutex_unlock(&fs_info->cleaner_mutex);
1732
1733                 /*
1734                  * The defragger has dealt with the R/O remount and umount,
1735                  * needn't do anything special here.
1736                  */
1737                 btrfs_run_defrag_inodes(fs_info);
1738
1739                 /*
1740                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1741                  * with relocation (btrfs_relocate_chunk) and relocation
1742                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1743                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1744                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1745                  * unused block groups.
1746                  */
1747                 btrfs_delete_unused_bgs(fs_info);
1748 sleep:
1749                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1750                 if (kthread_should_park())
1751                         kthread_parkme();
1752                 if (kthread_should_stop())
1753                         return 0;
1754                 if (!again) {
1755                         set_current_state(TASK_INTERRUPTIBLE);
1756                         schedule();
1757                         __set_current_state(TASK_RUNNING);
1758                 }
1759         }
1760 }
1761
1762 static int transaction_kthread(void *arg)
1763 {
1764         struct btrfs_root *root = arg;
1765         struct btrfs_fs_info *fs_info = root->fs_info;
1766         struct btrfs_trans_handle *trans;
1767         struct btrfs_transaction *cur;
1768         u64 transid;
1769         time64_t now;
1770         unsigned long delay;
1771         bool cannot_commit;
1772
1773         do {
1774                 cannot_commit = false;
1775                 delay = HZ * fs_info->commit_interval;
1776                 mutex_lock(&fs_info->transaction_kthread_mutex);
1777
1778                 spin_lock(&fs_info->trans_lock);
1779                 cur = fs_info->running_transaction;
1780                 if (!cur) {
1781                         spin_unlock(&fs_info->trans_lock);
1782                         goto sleep;
1783                 }
1784
1785                 now = ktime_get_seconds();
1786                 if (cur->state < TRANS_STATE_BLOCKED &&
1787                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1788                     (now < cur->start_time ||
1789                      now - cur->start_time < fs_info->commit_interval)) {
1790                         spin_unlock(&fs_info->trans_lock);
1791                         delay = HZ * 5;
1792                         goto sleep;
1793                 }
1794                 transid = cur->transid;
1795                 spin_unlock(&fs_info->trans_lock);
1796
1797                 /* If the file system is aborted, this will always fail. */
1798                 trans = btrfs_attach_transaction(root);
1799                 if (IS_ERR(trans)) {
1800                         if (PTR_ERR(trans) != -ENOENT)
1801                                 cannot_commit = true;
1802                         goto sleep;
1803                 }
1804                 if (transid == trans->transid) {
1805                         btrfs_commit_transaction(trans);
1806                 } else {
1807                         btrfs_end_transaction(trans);
1808                 }
1809 sleep:
1810                 wake_up_process(fs_info->cleaner_kthread);
1811                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1812
1813                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1814                                       &fs_info->fs_state)))
1815                         btrfs_cleanup_transaction(fs_info);
1816                 if (!kthread_should_stop() &&
1817                                 (!btrfs_transaction_blocked(fs_info) ||
1818                                  cannot_commit))
1819                         schedule_timeout_interruptible(delay);
1820         } while (!kthread_should_stop());
1821         return 0;
1822 }
1823
1824 /*
1825  * this will find the highest generation in the array of
1826  * root backups.  The index of the highest array is returned,
1827  * or -1 if we can't find anything.
1828  *
1829  * We check to make sure the array is valid by comparing the
1830  * generation of the latest  root in the array with the generation
1831  * in the super block.  If they don't match we pitch it.
1832  */
1833 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1834 {
1835         u64 cur;
1836         int newest_index = -1;
1837         struct btrfs_root_backup *root_backup;
1838         int i;
1839
1840         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1841                 root_backup = info->super_copy->super_roots + i;
1842                 cur = btrfs_backup_tree_root_gen(root_backup);
1843                 if (cur == newest_gen)
1844                         newest_index = i;
1845         }
1846
1847         /* check to see if we actually wrapped around */
1848         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1849                 root_backup = info->super_copy->super_roots;
1850                 cur = btrfs_backup_tree_root_gen(root_backup);
1851                 if (cur == newest_gen)
1852                         newest_index = 0;
1853         }
1854         return newest_index;
1855 }
1856
1857
1858 /*
1859  * find the oldest backup so we know where to store new entries
1860  * in the backup array.  This will set the backup_root_index
1861  * field in the fs_info struct
1862  */
1863 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1864                                      u64 newest_gen)
1865 {
1866         int newest_index = -1;
1867
1868         newest_index = find_newest_super_backup(info, newest_gen);
1869         /* if there was garbage in there, just move along */
1870         if (newest_index == -1) {
1871                 info->backup_root_index = 0;
1872         } else {
1873                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1874         }
1875 }
1876
1877 /*
1878  * copy all the root pointers into the super backup array.
1879  * this will bump the backup pointer by one when it is
1880  * done
1881  */
1882 static void backup_super_roots(struct btrfs_fs_info *info)
1883 {
1884         int next_backup;
1885         struct btrfs_root_backup *root_backup;
1886         int last_backup;
1887
1888         next_backup = info->backup_root_index;
1889         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1890                 BTRFS_NUM_BACKUP_ROOTS;
1891
1892         /*
1893          * just overwrite the last backup if we're at the same generation
1894          * this happens only at umount
1895          */
1896         root_backup = info->super_for_commit->super_roots + last_backup;
1897         if (btrfs_backup_tree_root_gen(root_backup) ==
1898             btrfs_header_generation(info->tree_root->node))
1899                 next_backup = last_backup;
1900
1901         root_backup = info->super_for_commit->super_roots + next_backup;
1902
1903         /*
1904          * make sure all of our padding and empty slots get zero filled
1905          * regardless of which ones we use today
1906          */
1907         memset(root_backup, 0, sizeof(*root_backup));
1908
1909         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1910
1911         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1912         btrfs_set_backup_tree_root_gen(root_backup,
1913                                btrfs_header_generation(info->tree_root->node));
1914
1915         btrfs_set_backup_tree_root_level(root_backup,
1916                                btrfs_header_level(info->tree_root->node));
1917
1918         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1919         btrfs_set_backup_chunk_root_gen(root_backup,
1920                                btrfs_header_generation(info->chunk_root->node));
1921         btrfs_set_backup_chunk_root_level(root_backup,
1922                                btrfs_header_level(info->chunk_root->node));
1923
1924         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1925         btrfs_set_backup_extent_root_gen(root_backup,
1926                                btrfs_header_generation(info->extent_root->node));
1927         btrfs_set_backup_extent_root_level(root_backup,
1928                                btrfs_header_level(info->extent_root->node));
1929
1930         /*
1931          * we might commit during log recovery, which happens before we set
1932          * the fs_root.  Make sure it is valid before we fill it in.
1933          */
1934         if (info->fs_root && info->fs_root->node) {
1935                 btrfs_set_backup_fs_root(root_backup,
1936                                          info->fs_root->node->start);
1937                 btrfs_set_backup_fs_root_gen(root_backup,
1938                                btrfs_header_generation(info->fs_root->node));
1939                 btrfs_set_backup_fs_root_level(root_backup,
1940                                btrfs_header_level(info->fs_root->node));
1941         }
1942
1943         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1944         btrfs_set_backup_dev_root_gen(root_backup,
1945                                btrfs_header_generation(info->dev_root->node));
1946         btrfs_set_backup_dev_root_level(root_backup,
1947                                        btrfs_header_level(info->dev_root->node));
1948
1949         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1950         btrfs_set_backup_csum_root_gen(root_backup,
1951                                btrfs_header_generation(info->csum_root->node));
1952         btrfs_set_backup_csum_root_level(root_backup,
1953                                btrfs_header_level(info->csum_root->node));
1954
1955         btrfs_set_backup_total_bytes(root_backup,
1956                              btrfs_super_total_bytes(info->super_copy));
1957         btrfs_set_backup_bytes_used(root_backup,
1958                              btrfs_super_bytes_used(info->super_copy));
1959         btrfs_set_backup_num_devices(root_backup,
1960                              btrfs_super_num_devices(info->super_copy));
1961
1962         /*
1963          * if we don't copy this out to the super_copy, it won't get remembered
1964          * for the next commit
1965          */
1966         memcpy(&info->super_copy->super_roots,
1967                &info->super_for_commit->super_roots,
1968                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1969 }
1970
1971 /*
1972  * this copies info out of the root backup array and back into
1973  * the in-memory super block.  It is meant to help iterate through
1974  * the array, so you send it the number of backups you've already
1975  * tried and the last backup index you used.
1976  *
1977  * this returns -1 when it has tried all the backups
1978  */
1979 static noinline int next_root_backup(struct btrfs_fs_info *info,
1980                                      struct btrfs_super_block *super,
1981                                      int *num_backups_tried, int *backup_index)
1982 {
1983         struct btrfs_root_backup *root_backup;
1984         int newest = *backup_index;
1985
1986         if (*num_backups_tried == 0) {
1987                 u64 gen = btrfs_super_generation(super);
1988
1989                 newest = find_newest_super_backup(info, gen);
1990                 if (newest == -1)
1991                         return -1;
1992
1993                 *backup_index = newest;
1994                 *num_backups_tried = 1;
1995         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1996                 /* we've tried all the backups, all done */
1997                 return -1;
1998         } else {
1999                 /* jump to the next oldest backup */
2000                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2001                         BTRFS_NUM_BACKUP_ROOTS;
2002                 *backup_index = newest;
2003                 *num_backups_tried += 1;
2004         }
2005         root_backup = super->super_roots + newest;
2006
2007         btrfs_set_super_generation(super,
2008                                    btrfs_backup_tree_root_gen(root_backup));
2009         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2010         btrfs_set_super_root_level(super,
2011                                    btrfs_backup_tree_root_level(root_backup));
2012         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2013
2014         /*
2015          * fixme: the total bytes and num_devices need to match or we should
2016          * need a fsck
2017          */
2018         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2019         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2020         return 0;
2021 }
2022
2023 /* helper to cleanup workers */
2024 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2025 {
2026         btrfs_destroy_workqueue(fs_info->fixup_workers);
2027         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2028         btrfs_destroy_workqueue(fs_info->workers);
2029         btrfs_destroy_workqueue(fs_info->endio_workers);
2030         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2031         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2032         btrfs_destroy_workqueue(fs_info->rmw_workers);
2033         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2034         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2035         btrfs_destroy_workqueue(fs_info->submit_workers);
2036         btrfs_destroy_workqueue(fs_info->delayed_workers);
2037         btrfs_destroy_workqueue(fs_info->caching_workers);
2038         btrfs_destroy_workqueue(fs_info->readahead_workers);
2039         btrfs_destroy_workqueue(fs_info->flush_workers);
2040         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2041         btrfs_destroy_workqueue(fs_info->extent_workers);
2042         /*
2043          * Now that all other work queues are destroyed, we can safely destroy
2044          * the queues used for metadata I/O, since tasks from those other work
2045          * queues can do metadata I/O operations.
2046          */
2047         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2048         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2049 }
2050
2051 static void free_root_extent_buffers(struct btrfs_root *root)
2052 {
2053         if (root) {
2054                 free_extent_buffer(root->node);
2055                 free_extent_buffer(root->commit_root);
2056                 root->node = NULL;
2057                 root->commit_root = NULL;
2058         }
2059 }
2060
2061 /* helper to cleanup tree roots */
2062 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2063 {
2064         free_root_extent_buffers(info->tree_root);
2065
2066         free_root_extent_buffers(info->dev_root);
2067         free_root_extent_buffers(info->extent_root);
2068         free_root_extent_buffers(info->csum_root);
2069         free_root_extent_buffers(info->quota_root);
2070         free_root_extent_buffers(info->uuid_root);
2071         if (chunk_root)
2072                 free_root_extent_buffers(info->chunk_root);
2073         free_root_extent_buffers(info->free_space_root);
2074 }
2075
2076 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2077 {
2078         int ret;
2079         struct btrfs_root *gang[8];
2080         int i;
2081
2082         while (!list_empty(&fs_info->dead_roots)) {
2083                 gang[0] = list_entry(fs_info->dead_roots.next,
2084                                      struct btrfs_root, root_list);
2085                 list_del(&gang[0]->root_list);
2086
2087                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2088                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2089                 } else {
2090                         free_extent_buffer(gang[0]->node);
2091                         free_extent_buffer(gang[0]->commit_root);
2092                         btrfs_put_fs_root(gang[0]);
2093                 }
2094         }
2095
2096         while (1) {
2097                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2098                                              (void **)gang, 0,
2099                                              ARRAY_SIZE(gang));
2100                 if (!ret)
2101                         break;
2102                 for (i = 0; i < ret; i++)
2103                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2104         }
2105
2106         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2107                 btrfs_free_log_root_tree(NULL, fs_info);
2108                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2109         }
2110 }
2111
2112 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2113 {
2114         mutex_init(&fs_info->scrub_lock);
2115         atomic_set(&fs_info->scrubs_running, 0);
2116         atomic_set(&fs_info->scrub_pause_req, 0);
2117         atomic_set(&fs_info->scrubs_paused, 0);
2118         atomic_set(&fs_info->scrub_cancel_req, 0);
2119         init_waitqueue_head(&fs_info->scrub_pause_wait);
2120         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2121 }
2122
2123 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2124 {
2125         spin_lock_init(&fs_info->balance_lock);
2126         mutex_init(&fs_info->balance_mutex);
2127         atomic_set(&fs_info->balance_pause_req, 0);
2128         atomic_set(&fs_info->balance_cancel_req, 0);
2129         fs_info->balance_ctl = NULL;
2130         init_waitqueue_head(&fs_info->balance_wait_q);
2131 }
2132
2133 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2134 {
2135         struct inode *inode = fs_info->btree_inode;
2136
2137         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2138         set_nlink(inode, 1);
2139         /*
2140          * we set the i_size on the btree inode to the max possible int.
2141          * the real end of the address space is determined by all of
2142          * the devices in the system
2143          */
2144         inode->i_size = OFFSET_MAX;
2145         inode->i_mapping->a_ops = &btree_aops;
2146
2147         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2148         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2149                             IO_TREE_INODE_IO, inode);
2150         BTRFS_I(inode)->io_tree.track_uptodate = false;
2151         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2152
2153         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2154
2155         BTRFS_I(inode)->root = fs_info->tree_root;
2156         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2157         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2158         btrfs_insert_inode_hash(inode);
2159 }
2160
2161 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2162 {
2163         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2164         init_rwsem(&fs_info->dev_replace.rwsem);
2165         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2166 }
2167
2168 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2169 {
2170         spin_lock_init(&fs_info->qgroup_lock);
2171         mutex_init(&fs_info->qgroup_ioctl_lock);
2172         fs_info->qgroup_tree = RB_ROOT;
2173         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2174         fs_info->qgroup_seq = 1;
2175         fs_info->qgroup_ulist = NULL;
2176         fs_info->qgroup_rescan_running = false;
2177         mutex_init(&fs_info->qgroup_rescan_lock);
2178 }
2179
2180 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2181                 struct btrfs_fs_devices *fs_devices)
2182 {
2183         u32 max_active = fs_info->thread_pool_size;
2184         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2185
2186         fs_info->workers =
2187                 btrfs_alloc_workqueue(fs_info, "worker",
2188                                       flags | WQ_HIGHPRI, max_active, 16);
2189
2190         fs_info->delalloc_workers =
2191                 btrfs_alloc_workqueue(fs_info, "delalloc",
2192                                       flags, max_active, 2);
2193
2194         fs_info->flush_workers =
2195                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2196                                       flags, max_active, 0);
2197
2198         fs_info->caching_workers =
2199                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2200
2201         /*
2202          * a higher idle thresh on the submit workers makes it much more
2203          * likely that bios will be send down in a sane order to the
2204          * devices
2205          */
2206         fs_info->submit_workers =
2207                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2208                                       min_t(u64, fs_devices->num_devices,
2209                                             max_active), 64);
2210
2211         fs_info->fixup_workers =
2212                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2213
2214         /*
2215          * endios are largely parallel and should have a very
2216          * low idle thresh
2217          */
2218         fs_info->endio_workers =
2219                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2220         fs_info->endio_meta_workers =
2221                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2222                                       max_active, 4);
2223         fs_info->endio_meta_write_workers =
2224                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2225                                       max_active, 2);
2226         fs_info->endio_raid56_workers =
2227                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2228                                       max_active, 4);
2229         fs_info->endio_repair_workers =
2230                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2231         fs_info->rmw_workers =
2232                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2233         fs_info->endio_write_workers =
2234                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2235                                       max_active, 2);
2236         fs_info->endio_freespace_worker =
2237                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2238                                       max_active, 0);
2239         fs_info->delayed_workers =
2240                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2241                                       max_active, 0);
2242         fs_info->readahead_workers =
2243                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2244                                       max_active, 2);
2245         fs_info->qgroup_rescan_workers =
2246                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2247         fs_info->extent_workers =
2248                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2249                                       min_t(u64, fs_devices->num_devices,
2250                                             max_active), 8);
2251
2252         if (!(fs_info->workers && fs_info->delalloc_workers &&
2253               fs_info->submit_workers && fs_info->flush_workers &&
2254               fs_info->endio_workers && fs_info->endio_meta_workers &&
2255               fs_info->endio_meta_write_workers &&
2256               fs_info->endio_repair_workers &&
2257               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2258               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2259               fs_info->caching_workers && fs_info->readahead_workers &&
2260               fs_info->fixup_workers && fs_info->delayed_workers &&
2261               fs_info->extent_workers &&
2262               fs_info->qgroup_rescan_workers)) {
2263                 return -ENOMEM;
2264         }
2265
2266         return 0;
2267 }
2268
2269 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2270                             struct btrfs_fs_devices *fs_devices)
2271 {
2272         int ret;
2273         struct btrfs_root *log_tree_root;
2274         struct btrfs_super_block *disk_super = fs_info->super_copy;
2275         u64 bytenr = btrfs_super_log_root(disk_super);
2276         int level = btrfs_super_log_root_level(disk_super);
2277
2278         if (fs_devices->rw_devices == 0) {
2279                 btrfs_warn(fs_info, "log replay required on RO media");
2280                 return -EIO;
2281         }
2282
2283         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2284         if (!log_tree_root)
2285                 return -ENOMEM;
2286
2287         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2288
2289         log_tree_root->node = read_tree_block(fs_info, bytenr,
2290                                               fs_info->generation + 1,
2291                                               level, NULL);
2292         if (IS_ERR(log_tree_root->node)) {
2293                 btrfs_warn(fs_info, "failed to read log tree");
2294                 ret = PTR_ERR(log_tree_root->node);
2295                 kfree(log_tree_root);
2296                 return ret;
2297         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2298                 btrfs_err(fs_info, "failed to read log tree");
2299                 free_extent_buffer(log_tree_root->node);
2300                 kfree(log_tree_root);
2301                 return -EIO;
2302         }
2303         /* returns with log_tree_root freed on success */
2304         ret = btrfs_recover_log_trees(log_tree_root);
2305         if (ret) {
2306                 btrfs_handle_fs_error(fs_info, ret,
2307                                       "Failed to recover log tree");
2308                 free_extent_buffer(log_tree_root->node);
2309                 kfree(log_tree_root);
2310                 return ret;
2311         }
2312
2313         if (sb_rdonly(fs_info->sb)) {
2314                 ret = btrfs_commit_super(fs_info);
2315                 if (ret)
2316                         return ret;
2317         }
2318
2319         return 0;
2320 }
2321
2322 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2323 {
2324         struct btrfs_root *tree_root = fs_info->tree_root;
2325         struct btrfs_root *root;
2326         struct btrfs_key location;
2327         int ret;
2328
2329         BUG_ON(!fs_info->tree_root);
2330
2331         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2332         location.type = BTRFS_ROOT_ITEM_KEY;
2333         location.offset = 0;
2334
2335         root = btrfs_read_tree_root(tree_root, &location);
2336         if (IS_ERR(root)) {
2337                 ret = PTR_ERR(root);
2338                 goto out;
2339         }
2340         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341         fs_info->extent_root = root;
2342
2343         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2344         root = btrfs_read_tree_root(tree_root, &location);
2345         if (IS_ERR(root)) {
2346                 ret = PTR_ERR(root);
2347                 goto out;
2348         }
2349         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2350         fs_info->dev_root = root;
2351         btrfs_init_devices_late(fs_info);
2352
2353         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2354         root = btrfs_read_tree_root(tree_root, &location);
2355         if (IS_ERR(root)) {
2356                 ret = PTR_ERR(root);
2357                 goto out;
2358         }
2359         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2360         fs_info->csum_root = root;
2361
2362         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2363         root = btrfs_read_tree_root(tree_root, &location);
2364         if (!IS_ERR(root)) {
2365                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2366                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2367                 fs_info->quota_root = root;
2368         }
2369
2370         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2371         root = btrfs_read_tree_root(tree_root, &location);
2372         if (IS_ERR(root)) {
2373                 ret = PTR_ERR(root);
2374                 if (ret != -ENOENT)
2375                         goto out;
2376         } else {
2377                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378                 fs_info->uuid_root = root;
2379         }
2380
2381         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2382                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2383                 root = btrfs_read_tree_root(tree_root, &location);
2384                 if (IS_ERR(root)) {
2385                         ret = PTR_ERR(root);
2386                         goto out;
2387                 }
2388                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389                 fs_info->free_space_root = root;
2390         }
2391
2392         return 0;
2393 out:
2394         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2395                    location.objectid, ret);
2396         return ret;
2397 }
2398
2399 /*
2400  * Real super block validation
2401  * NOTE: super csum type and incompat features will not be checked here.
2402  *
2403  * @sb:         super block to check
2404  * @mirror_num: the super block number to check its bytenr:
2405  *              0       the primary (1st) sb
2406  *              1, 2    2nd and 3rd backup copy
2407  *             -1       skip bytenr check
2408  */
2409 static int validate_super(struct btrfs_fs_info *fs_info,
2410                             struct btrfs_super_block *sb, int mirror_num)
2411 {
2412         u64 nodesize = btrfs_super_nodesize(sb);
2413         u64 sectorsize = btrfs_super_sectorsize(sb);
2414         int ret = 0;
2415
2416         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2417                 btrfs_err(fs_info, "no valid FS found");
2418                 ret = -EINVAL;
2419         }
2420         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2421                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2422                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2423                 ret = -EINVAL;
2424         }
2425         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2426                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2427                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2428                 ret = -EINVAL;
2429         }
2430         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2431                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2432                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2433                 ret = -EINVAL;
2434         }
2435         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2436                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2437                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2438                 ret = -EINVAL;
2439         }
2440
2441         /*
2442          * Check sectorsize and nodesize first, other check will need it.
2443          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2444          */
2445         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2446             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2447                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2448                 ret = -EINVAL;
2449         }
2450         /* Only PAGE SIZE is supported yet */
2451         if (sectorsize != PAGE_SIZE) {
2452                 btrfs_err(fs_info,
2453                         "sectorsize %llu not supported yet, only support %lu",
2454                         sectorsize, PAGE_SIZE);
2455                 ret = -EINVAL;
2456         }
2457         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2458             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2459                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2460                 ret = -EINVAL;
2461         }
2462         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2463                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2464                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2465                 ret = -EINVAL;
2466         }
2467
2468         /* Root alignment check */
2469         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2470                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2471                            btrfs_super_root(sb));
2472                 ret = -EINVAL;
2473         }
2474         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2475                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2476                            btrfs_super_chunk_root(sb));
2477                 ret = -EINVAL;
2478         }
2479         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2480                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2481                            btrfs_super_log_root(sb));
2482                 ret = -EINVAL;
2483         }
2484
2485         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2486                    BTRFS_FSID_SIZE) != 0) {
2487                 btrfs_err(fs_info,
2488                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2489                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2490                 ret = -EINVAL;
2491         }
2492
2493         /*
2494          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2495          * done later
2496          */
2497         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2498                 btrfs_err(fs_info, "bytes_used is too small %llu",
2499                           btrfs_super_bytes_used(sb));
2500                 ret = -EINVAL;
2501         }
2502         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2503                 btrfs_err(fs_info, "invalid stripesize %u",
2504                           btrfs_super_stripesize(sb));
2505                 ret = -EINVAL;
2506         }
2507         if (btrfs_super_num_devices(sb) > (1UL << 31))
2508                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2509                            btrfs_super_num_devices(sb));
2510         if (btrfs_super_num_devices(sb) == 0) {
2511                 btrfs_err(fs_info, "number of devices is 0");
2512                 ret = -EINVAL;
2513         }
2514
2515         if (mirror_num >= 0 &&
2516             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2517                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2518                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2519                 ret = -EINVAL;
2520         }
2521
2522         /*
2523          * Obvious sys_chunk_array corruptions, it must hold at least one key
2524          * and one chunk
2525          */
2526         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2527                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2528                           btrfs_super_sys_array_size(sb),
2529                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2530                 ret = -EINVAL;
2531         }
2532         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2533                         + sizeof(struct btrfs_chunk)) {
2534                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2535                           btrfs_super_sys_array_size(sb),
2536                           sizeof(struct btrfs_disk_key)
2537                           + sizeof(struct btrfs_chunk));
2538                 ret = -EINVAL;
2539         }
2540
2541         /*
2542          * The generation is a global counter, we'll trust it more than the others
2543          * but it's still possible that it's the one that's wrong.
2544          */
2545         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2546                 btrfs_warn(fs_info,
2547                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2548                         btrfs_super_generation(sb),
2549                         btrfs_super_chunk_root_generation(sb));
2550         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2551             && btrfs_super_cache_generation(sb) != (u64)-1)
2552                 btrfs_warn(fs_info,
2553                         "suspicious: generation < cache_generation: %llu < %llu",
2554                         btrfs_super_generation(sb),
2555                         btrfs_super_cache_generation(sb));
2556
2557         return ret;
2558 }
2559
2560 /*
2561  * Validation of super block at mount time.
2562  * Some checks already done early at mount time, like csum type and incompat
2563  * flags will be skipped.
2564  */
2565 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2566 {
2567         return validate_super(fs_info, fs_info->super_copy, 0);
2568 }
2569
2570 /*
2571  * Validation of super block at write time.
2572  * Some checks like bytenr check will be skipped as their values will be
2573  * overwritten soon.
2574  * Extra checks like csum type and incompat flags will be done here.
2575  */
2576 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2577                                       struct btrfs_super_block *sb)
2578 {
2579         int ret;
2580
2581         ret = validate_super(fs_info, sb, -1);
2582         if (ret < 0)
2583                 goto out;
2584         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2585                 ret = -EUCLEAN;
2586                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2587                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2588                 goto out;
2589         }
2590         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2591                 ret = -EUCLEAN;
2592                 btrfs_err(fs_info,
2593                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2594                           btrfs_super_incompat_flags(sb),
2595                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2596                 goto out;
2597         }
2598 out:
2599         if (ret < 0)
2600                 btrfs_err(fs_info,
2601                 "super block corruption detected before writing it to disk");
2602         return ret;
2603 }
2604
2605 int open_ctree(struct super_block *sb,
2606                struct btrfs_fs_devices *fs_devices,
2607                char *options)
2608 {
2609         u32 sectorsize;
2610         u32 nodesize;
2611         u32 stripesize;
2612         u64 generation;
2613         u64 features;
2614         struct btrfs_key location;
2615         struct buffer_head *bh;
2616         struct btrfs_super_block *disk_super;
2617         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2618         struct btrfs_root *tree_root;
2619         struct btrfs_root *chunk_root;
2620         int ret;
2621         int err = -EINVAL;
2622         int num_backups_tried = 0;
2623         int backup_index = 0;
2624         int clear_free_space_tree = 0;
2625         int level;
2626
2627         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2628         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2629         if (!tree_root || !chunk_root) {
2630                 err = -ENOMEM;
2631                 goto fail;
2632         }
2633
2634         ret = init_srcu_struct(&fs_info->subvol_srcu);
2635         if (ret) {
2636                 err = ret;
2637                 goto fail;
2638         }
2639
2640         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2641         if (ret) {
2642                 err = ret;
2643                 goto fail_srcu;
2644         }
2645
2646         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2647         if (ret) {
2648                 err = ret;
2649                 goto fail_dio_bytes;
2650         }
2651         fs_info->dirty_metadata_batch = PAGE_SIZE *
2652                                         (1 + ilog2(nr_cpu_ids));
2653
2654         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2655         if (ret) {
2656                 err = ret;
2657                 goto fail_dirty_metadata_bytes;
2658         }
2659
2660         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2661                         GFP_KERNEL);
2662         if (ret) {
2663                 err = ret;
2664                 goto fail_delalloc_bytes;
2665         }
2666
2667         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2668         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2669         INIT_LIST_HEAD(&fs_info->trans_list);
2670         INIT_LIST_HEAD(&fs_info->dead_roots);
2671         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2672         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2673         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2674         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2675         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2676         spin_lock_init(&fs_info->delalloc_root_lock);
2677         spin_lock_init(&fs_info->trans_lock);
2678         spin_lock_init(&fs_info->fs_roots_radix_lock);
2679         spin_lock_init(&fs_info->delayed_iput_lock);
2680         spin_lock_init(&fs_info->defrag_inodes_lock);
2681         spin_lock_init(&fs_info->tree_mod_seq_lock);
2682         spin_lock_init(&fs_info->super_lock);
2683         spin_lock_init(&fs_info->buffer_lock);
2684         spin_lock_init(&fs_info->unused_bgs_lock);
2685         rwlock_init(&fs_info->tree_mod_log_lock);
2686         mutex_init(&fs_info->unused_bg_unpin_mutex);
2687         mutex_init(&fs_info->delete_unused_bgs_mutex);
2688         mutex_init(&fs_info->reloc_mutex);
2689         mutex_init(&fs_info->delalloc_root_mutex);
2690         seqlock_init(&fs_info->profiles_lock);
2691
2692         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2693         INIT_LIST_HEAD(&fs_info->space_info);
2694         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2695         INIT_LIST_HEAD(&fs_info->unused_bgs);
2696         extent_map_tree_init(&fs_info->mapping_tree);
2697         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2698                              BTRFS_BLOCK_RSV_GLOBAL);
2699         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2700         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2701         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2702         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2703                              BTRFS_BLOCK_RSV_DELOPS);
2704         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2705                              BTRFS_BLOCK_RSV_DELREFS);
2706
2707         atomic_set(&fs_info->async_delalloc_pages, 0);
2708         atomic_set(&fs_info->defrag_running, 0);
2709         atomic_set(&fs_info->reada_works_cnt, 0);
2710         atomic_set(&fs_info->nr_delayed_iputs, 0);
2711         atomic64_set(&fs_info->tree_mod_seq, 0);
2712         fs_info->sb = sb;
2713         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2714         fs_info->metadata_ratio = 0;
2715         fs_info->defrag_inodes = RB_ROOT;
2716         atomic64_set(&fs_info->free_chunk_space, 0);
2717         fs_info->tree_mod_log = RB_ROOT;
2718         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2719         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2720         /* readahead state */
2721         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2722         spin_lock_init(&fs_info->reada_lock);
2723         btrfs_init_ref_verify(fs_info);
2724
2725         fs_info->thread_pool_size = min_t(unsigned long,
2726                                           num_online_cpus() + 2, 8);
2727
2728         INIT_LIST_HEAD(&fs_info->ordered_roots);
2729         spin_lock_init(&fs_info->ordered_root_lock);
2730
2731         fs_info->btree_inode = new_inode(sb);
2732         if (!fs_info->btree_inode) {
2733                 err = -ENOMEM;
2734                 goto fail_bio_counter;
2735         }
2736         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2737
2738         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2739                                         GFP_KERNEL);
2740         if (!fs_info->delayed_root) {
2741                 err = -ENOMEM;
2742                 goto fail_iput;
2743         }
2744         btrfs_init_delayed_root(fs_info->delayed_root);
2745
2746         btrfs_init_scrub(fs_info);
2747 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2748         fs_info->check_integrity_print_mask = 0;
2749 #endif
2750         btrfs_init_balance(fs_info);
2751         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2752
2753         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2754         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2755
2756         btrfs_init_btree_inode(fs_info);
2757
2758         spin_lock_init(&fs_info->block_group_cache_lock);
2759         fs_info->block_group_cache_tree = RB_ROOT;
2760         fs_info->first_logical_byte = (u64)-1;
2761
2762         extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2763                             IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2764         extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2765                             IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2766         fs_info->pinned_extents = &fs_info->freed_extents[0];
2767         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2768
2769         mutex_init(&fs_info->ordered_operations_mutex);
2770         mutex_init(&fs_info->tree_log_mutex);
2771         mutex_init(&fs_info->chunk_mutex);
2772         mutex_init(&fs_info->transaction_kthread_mutex);
2773         mutex_init(&fs_info->cleaner_mutex);
2774         mutex_init(&fs_info->ro_block_group_mutex);
2775         init_rwsem(&fs_info->commit_root_sem);
2776         init_rwsem(&fs_info->cleanup_work_sem);
2777         init_rwsem(&fs_info->subvol_sem);
2778         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2779
2780         btrfs_init_dev_replace_locks(fs_info);
2781         btrfs_init_qgroup(fs_info);
2782
2783         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2784         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2785
2786         init_waitqueue_head(&fs_info->transaction_throttle);
2787         init_waitqueue_head(&fs_info->transaction_wait);
2788         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2789         init_waitqueue_head(&fs_info->async_submit_wait);
2790         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2791
2792         /* Usable values until the real ones are cached from the superblock */
2793         fs_info->nodesize = 4096;
2794         fs_info->sectorsize = 4096;
2795         fs_info->stripesize = 4096;
2796
2797         spin_lock_init(&fs_info->swapfile_pins_lock);
2798         fs_info->swapfile_pins = RB_ROOT;
2799
2800         ret = btrfs_alloc_stripe_hash_table(fs_info);
2801         if (ret) {
2802                 err = ret;
2803                 goto fail_alloc;
2804         }
2805
2806         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2807
2808         invalidate_bdev(fs_devices->latest_bdev);
2809
2810         /*
2811          * Read super block and check the signature bytes only
2812          */
2813         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2814         if (IS_ERR(bh)) {
2815                 err = PTR_ERR(bh);
2816                 goto fail_alloc;
2817         }
2818
2819         /*
2820          * We want to check superblock checksum, the type is stored inside.
2821          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2822          */
2823         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2824                 btrfs_err(fs_info, "superblock checksum mismatch");
2825                 err = -EINVAL;
2826                 brelse(bh);
2827                 goto fail_alloc;
2828         }
2829
2830         /*
2831          * super_copy is zeroed at allocation time and we never touch the
2832          * following bytes up to INFO_SIZE, the checksum is calculated from
2833          * the whole block of INFO_SIZE
2834          */
2835         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2836         brelse(bh);
2837
2838         disk_super = fs_info->super_copy;
2839
2840         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2841                        BTRFS_FSID_SIZE));
2842
2843         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2844                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2845                                 fs_info->super_copy->metadata_uuid,
2846                                 BTRFS_FSID_SIZE));
2847         }
2848
2849         features = btrfs_super_flags(disk_super);
2850         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2851                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2852                 btrfs_set_super_flags(disk_super, features);
2853                 btrfs_info(fs_info,
2854                         "found metadata UUID change in progress flag, clearing");
2855         }
2856
2857         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2858                sizeof(*fs_info->super_for_commit));
2859
2860         ret = btrfs_validate_mount_super(fs_info);
2861         if (ret) {
2862                 btrfs_err(fs_info, "superblock contains fatal errors");
2863                 err = -EINVAL;
2864                 goto fail_alloc;
2865         }
2866
2867         if (!btrfs_super_root(disk_super))
2868                 goto fail_alloc;
2869
2870         /* check FS state, whether FS is broken. */
2871         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2872                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2873
2874         /*
2875          * run through our array of backup supers and setup
2876          * our ring pointer to the oldest one
2877          */
2878         generation = btrfs_super_generation(disk_super);
2879         find_oldest_super_backup(fs_info, generation);
2880
2881         /*
2882          * In the long term, we'll store the compression type in the super
2883          * block, and it'll be used for per file compression control.
2884          */
2885         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2886
2887         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2888         if (ret) {
2889                 err = ret;
2890                 goto fail_alloc;
2891         }
2892
2893         features = btrfs_super_incompat_flags(disk_super) &
2894                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2895         if (features) {
2896                 btrfs_err(fs_info,
2897                     "cannot mount because of unsupported optional features (%llx)",
2898                     features);
2899                 err = -EINVAL;
2900                 goto fail_alloc;
2901         }
2902
2903         features = btrfs_super_incompat_flags(disk_super);
2904         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2905         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2906                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2907         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2908                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2909
2910         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2911                 btrfs_info(fs_info, "has skinny extents");
2912
2913         /*
2914          * flag our filesystem as having big metadata blocks if
2915          * they are bigger than the page size
2916          */
2917         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2918                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2919                         btrfs_info(fs_info,
2920                                 "flagging fs with big metadata feature");
2921                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2922         }
2923
2924         nodesize = btrfs_super_nodesize(disk_super);
2925         sectorsize = btrfs_super_sectorsize(disk_super);
2926         stripesize = sectorsize;
2927         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2928         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2929
2930         /* Cache block sizes */
2931         fs_info->nodesize = nodesize;
2932         fs_info->sectorsize = sectorsize;
2933         fs_info->stripesize = stripesize;
2934
2935         /*
2936          * mixed block groups end up with duplicate but slightly offset
2937          * extent buffers for the same range.  It leads to corruptions
2938          */
2939         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2940             (sectorsize != nodesize)) {
2941                 btrfs_err(fs_info,
2942 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2943                         nodesize, sectorsize);
2944                 goto fail_alloc;
2945         }
2946
2947         /*
2948          * Needn't use the lock because there is no other task which will
2949          * update the flag.
2950          */
2951         btrfs_set_super_incompat_flags(disk_super, features);
2952
2953         features = btrfs_super_compat_ro_flags(disk_super) &
2954                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2955         if (!sb_rdonly(sb) && features) {
2956                 btrfs_err(fs_info,
2957         "cannot mount read-write because of unsupported optional features (%llx)",
2958                        features);
2959                 err = -EINVAL;
2960                 goto fail_alloc;
2961         }
2962
2963         ret = btrfs_init_workqueues(fs_info, fs_devices);
2964         if (ret) {
2965                 err = ret;
2966                 goto fail_sb_buffer;
2967         }
2968
2969         sb->s_bdi->congested_fn = btrfs_congested_fn;
2970         sb->s_bdi->congested_data = fs_info;
2971         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2972         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2973         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2974         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2975
2976         sb->s_blocksize = sectorsize;
2977         sb->s_blocksize_bits = blksize_bits(sectorsize);
2978         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2979
2980         mutex_lock(&fs_info->chunk_mutex);
2981         ret = btrfs_read_sys_array(fs_info);
2982         mutex_unlock(&fs_info->chunk_mutex);
2983         if (ret) {
2984                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2985                 goto fail_sb_buffer;
2986         }
2987
2988         generation = btrfs_super_chunk_root_generation(disk_super);
2989         level = btrfs_super_chunk_root_level(disk_super);
2990
2991         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2992
2993         chunk_root->node = read_tree_block(fs_info,
2994                                            btrfs_super_chunk_root(disk_super),
2995                                            generation, level, NULL);
2996         if (IS_ERR(chunk_root->node) ||
2997             !extent_buffer_uptodate(chunk_root->node)) {
2998                 btrfs_err(fs_info, "failed to read chunk root");
2999                 if (!IS_ERR(chunk_root->node))
3000                         free_extent_buffer(chunk_root->node);
3001                 chunk_root->node = NULL;
3002                 goto fail_tree_roots;
3003         }
3004         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3005         chunk_root->commit_root = btrfs_root_node(chunk_root);
3006
3007         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3008            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3009
3010         ret = btrfs_read_chunk_tree(fs_info);
3011         if (ret) {
3012                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3013                 goto fail_tree_roots;
3014         }
3015
3016         /*
3017          * Keep the devid that is marked to be the target device for the
3018          * device replace procedure
3019          */
3020         btrfs_free_extra_devids(fs_devices, 0);
3021
3022         if (!fs_devices->latest_bdev) {
3023                 btrfs_err(fs_info, "failed to read devices");
3024                 goto fail_tree_roots;
3025         }
3026
3027 retry_root_backup:
3028         generation = btrfs_super_generation(disk_super);
3029         level = btrfs_super_root_level(disk_super);
3030
3031         tree_root->node = read_tree_block(fs_info,
3032                                           btrfs_super_root(disk_super),
3033                                           generation, level, NULL);
3034         if (IS_ERR(tree_root->node) ||
3035             !extent_buffer_uptodate(tree_root->node)) {
3036                 btrfs_warn(fs_info, "failed to read tree root");
3037                 if (!IS_ERR(tree_root->node))
3038                         free_extent_buffer(tree_root->node);
3039                 tree_root->node = NULL;
3040                 goto recovery_tree_root;
3041         }
3042
3043         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3044         tree_root->commit_root = btrfs_root_node(tree_root);
3045         btrfs_set_root_refs(&tree_root->root_item, 1);
3046
3047         mutex_lock(&tree_root->objectid_mutex);
3048         ret = btrfs_find_highest_objectid(tree_root,
3049                                         &tree_root->highest_objectid);
3050         if (ret) {
3051                 mutex_unlock(&tree_root->objectid_mutex);
3052                 goto recovery_tree_root;
3053         }
3054
3055         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3056
3057         mutex_unlock(&tree_root->objectid_mutex);
3058
3059         ret = btrfs_read_roots(fs_info);
3060         if (ret)
3061                 goto recovery_tree_root;
3062
3063         fs_info->generation = generation;
3064         fs_info->last_trans_committed = generation;
3065
3066         ret = btrfs_verify_dev_extents(fs_info);
3067         if (ret) {
3068                 btrfs_err(fs_info,
3069                           "failed to verify dev extents against chunks: %d",
3070                           ret);
3071                 goto fail_block_groups;
3072         }
3073         ret = btrfs_recover_balance(fs_info);
3074         if (ret) {
3075                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3076                 goto fail_block_groups;
3077         }
3078
3079         ret = btrfs_init_dev_stats(fs_info);
3080         if (ret) {
3081                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3082                 goto fail_block_groups;
3083         }
3084
3085         ret = btrfs_init_dev_replace(fs_info);
3086         if (ret) {
3087                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3088                 goto fail_block_groups;
3089         }
3090
3091         btrfs_free_extra_devids(fs_devices, 1);
3092
3093         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3094         if (ret) {
3095                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3096                                 ret);
3097                 goto fail_block_groups;
3098         }
3099
3100         ret = btrfs_sysfs_add_device(fs_devices);
3101         if (ret) {
3102                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3103                                 ret);
3104                 goto fail_fsdev_sysfs;
3105         }
3106
3107         ret = btrfs_sysfs_add_mounted(fs_info);
3108         if (ret) {
3109                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3110                 goto fail_fsdev_sysfs;
3111         }
3112
3113         ret = btrfs_init_space_info(fs_info);
3114         if (ret) {
3115                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3116                 goto fail_sysfs;
3117         }
3118
3119         ret = btrfs_read_block_groups(fs_info);
3120         if (ret) {
3121                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3122                 goto fail_sysfs;
3123         }
3124
3125         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3126                 btrfs_warn(fs_info,
3127                 "writable mount is not allowed due to too many missing devices");
3128                 goto fail_sysfs;
3129         }
3130
3131         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3132                                                "btrfs-cleaner");
3133         if (IS_ERR(fs_info->cleaner_kthread))
3134                 goto fail_sysfs;
3135
3136         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3137                                                    tree_root,
3138                                                    "btrfs-transaction");
3139         if (IS_ERR(fs_info->transaction_kthread))
3140                 goto fail_cleaner;
3141
3142         if (!btrfs_test_opt(fs_info, NOSSD) &&
3143             !fs_info->fs_devices->rotating) {
3144                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3145         }
3146
3147         /*
3148          * Mount does not set all options immediately, we can do it now and do
3149          * not have to wait for transaction commit
3150          */
3151         btrfs_apply_pending_changes(fs_info);
3152
3153 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3154         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3155                 ret = btrfsic_mount(fs_info, fs_devices,
3156                                     btrfs_test_opt(fs_info,
3157                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3158                                     1 : 0,
3159                                     fs_info->check_integrity_print_mask);
3160                 if (ret)
3161                         btrfs_warn(fs_info,
3162                                 "failed to initialize integrity check module: %d",
3163                                 ret);
3164         }
3165 #endif
3166         ret = btrfs_read_qgroup_config(fs_info);
3167         if (ret)
3168                 goto fail_trans_kthread;
3169
3170         if (btrfs_build_ref_tree(fs_info))
3171                 btrfs_err(fs_info, "couldn't build ref tree");
3172
3173         /* do not make disk changes in broken FS or nologreplay is given */
3174         if (btrfs_super_log_root(disk_super) != 0 &&
3175             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3176                 ret = btrfs_replay_log(fs_info, fs_devices);
3177                 if (ret) {
3178                         err = ret;
3179                         goto fail_qgroup;
3180                 }
3181         }
3182
3183         ret = btrfs_find_orphan_roots(fs_info);
3184         if (ret)
3185                 goto fail_qgroup;
3186
3187         if (!sb_rdonly(sb)) {
3188                 ret = btrfs_cleanup_fs_roots(fs_info);
3189                 if (ret)
3190                         goto fail_qgroup;
3191
3192                 mutex_lock(&fs_info->cleaner_mutex);
3193                 ret = btrfs_recover_relocation(tree_root);
3194                 mutex_unlock(&fs_info->cleaner_mutex);
3195                 if (ret < 0) {
3196                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3197                                         ret);
3198                         err = -EINVAL;
3199                         goto fail_qgroup;
3200                 }
3201         }
3202
3203         location.objectid = BTRFS_FS_TREE_OBJECTID;
3204         location.type = BTRFS_ROOT_ITEM_KEY;
3205         location.offset = 0;
3206
3207         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3208         if (IS_ERR(fs_info->fs_root)) {
3209                 err = PTR_ERR(fs_info->fs_root);
3210                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3211                 goto fail_qgroup;
3212         }
3213
3214         if (sb_rdonly(sb))
3215                 return 0;
3216
3217         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3218             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3219                 clear_free_space_tree = 1;
3220         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3221                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3222                 btrfs_warn(fs_info, "free space tree is invalid");
3223                 clear_free_space_tree = 1;
3224         }
3225
3226         if (clear_free_space_tree) {
3227                 btrfs_info(fs_info, "clearing free space tree");
3228                 ret = btrfs_clear_free_space_tree(fs_info);
3229                 if (ret) {
3230                         btrfs_warn(fs_info,
3231                                    "failed to clear free space tree: %d", ret);
3232                         close_ctree(fs_info);
3233                         return ret;
3234                 }
3235         }
3236
3237         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3238             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3239                 btrfs_info(fs_info, "creating free space tree");
3240                 ret = btrfs_create_free_space_tree(fs_info);
3241                 if (ret) {
3242                         btrfs_warn(fs_info,
3243                                 "failed to create free space tree: %d", ret);
3244                         close_ctree(fs_info);
3245                         return ret;
3246                 }
3247         }
3248
3249         down_read(&fs_info->cleanup_work_sem);
3250         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3251             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3252                 up_read(&fs_info->cleanup_work_sem);
3253                 close_ctree(fs_info);
3254                 return ret;
3255         }
3256         up_read(&fs_info->cleanup_work_sem);
3257
3258         ret = btrfs_resume_balance_async(fs_info);
3259         if (ret) {
3260                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3261                 close_ctree(fs_info);
3262                 return ret;
3263         }
3264
3265         ret = btrfs_resume_dev_replace_async(fs_info);
3266         if (ret) {
3267                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3268                 close_ctree(fs_info);
3269                 return ret;
3270         }
3271
3272         btrfs_qgroup_rescan_resume(fs_info);
3273
3274         if (!fs_info->uuid_root) {
3275                 btrfs_info(fs_info, "creating UUID tree");
3276                 ret = btrfs_create_uuid_tree(fs_info);
3277                 if (ret) {
3278                         btrfs_warn(fs_info,
3279                                 "failed to create the UUID tree: %d", ret);
3280                         close_ctree(fs_info);
3281                         return ret;
3282                 }
3283         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3284                    fs_info->generation !=
3285                                 btrfs_super_uuid_tree_generation(disk_super)) {
3286                 btrfs_info(fs_info, "checking UUID tree");
3287                 ret = btrfs_check_uuid_tree(fs_info);
3288                 if (ret) {
3289                         btrfs_warn(fs_info,
3290                                 "failed to check the UUID tree: %d", ret);
3291                         close_ctree(fs_info);
3292                         return ret;
3293                 }
3294         } else {
3295                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3296         }
3297         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3298
3299         /*
3300          * backuproot only affect mount behavior, and if open_ctree succeeded,
3301          * no need to keep the flag
3302          */
3303         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3304
3305         return 0;
3306
3307 fail_qgroup:
3308         btrfs_free_qgroup_config(fs_info);
3309 fail_trans_kthread:
3310         kthread_stop(fs_info->transaction_kthread);
3311         btrfs_cleanup_transaction(fs_info);
3312         btrfs_free_fs_roots(fs_info);
3313 fail_cleaner:
3314         kthread_stop(fs_info->cleaner_kthread);
3315
3316         /*
3317          * make sure we're done with the btree inode before we stop our
3318          * kthreads
3319          */
3320         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3321
3322 fail_sysfs:
3323         btrfs_sysfs_remove_mounted(fs_info);
3324
3325 fail_fsdev_sysfs:
3326         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3327
3328 fail_block_groups:
3329         btrfs_put_block_group_cache(fs_info);
3330
3331 fail_tree_roots:
3332         free_root_pointers(fs_info, 1);
3333         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3334
3335 fail_sb_buffer:
3336         btrfs_stop_all_workers(fs_info);
3337         btrfs_free_block_groups(fs_info);
3338 fail_alloc:
3339 fail_iput:
3340         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3341
3342         iput(fs_info->btree_inode);
3343 fail_bio_counter:
3344         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3345 fail_delalloc_bytes:
3346         percpu_counter_destroy(&fs_info->delalloc_bytes);
3347 fail_dirty_metadata_bytes:
3348         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3349 fail_dio_bytes:
3350         percpu_counter_destroy(&fs_info->dio_bytes);
3351 fail_srcu:
3352         cleanup_srcu_struct(&fs_info->subvol_srcu);
3353 fail:
3354         btrfs_free_stripe_hash_table(fs_info);
3355         btrfs_close_devices(fs_info->fs_devices);
3356         return err;
3357
3358 recovery_tree_root:
3359         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3360                 goto fail_tree_roots;
3361
3362         free_root_pointers(fs_info, 0);
3363
3364         /* don't use the log in recovery mode, it won't be valid */
3365         btrfs_set_super_log_root(disk_super, 0);
3366
3367         /* we can't trust the free space cache either */
3368         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3369
3370         ret = next_root_backup(fs_info, fs_info->super_copy,
3371                                &num_backups_tried, &backup_index);
3372         if (ret == -1)
3373                 goto fail_block_groups;
3374         goto retry_root_backup;
3375 }
3376 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3377
3378 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3379 {
3380         if (uptodate) {
3381                 set_buffer_uptodate(bh);
3382         } else {
3383                 struct btrfs_device *device = (struct btrfs_device *)
3384                         bh->b_private;
3385
3386                 btrfs_warn_rl_in_rcu(device->fs_info,
3387                                 "lost page write due to IO error on %s",
3388                                           rcu_str_deref(device->name));
3389                 /* note, we don't set_buffer_write_io_error because we have
3390                  * our own ways of dealing with the IO errors
3391                  */
3392                 clear_buffer_uptodate(bh);
3393                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3394         }
3395         unlock_buffer(bh);
3396         put_bh(bh);
3397 }
3398
3399 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3400                         struct buffer_head **bh_ret)
3401 {
3402         struct buffer_head *bh;
3403         struct btrfs_super_block *super;
3404         u64 bytenr;
3405
3406         bytenr = btrfs_sb_offset(copy_num);
3407         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3408                 return -EINVAL;
3409
3410         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3411         /*
3412          * If we fail to read from the underlying devices, as of now
3413          * the best option we have is to mark it EIO.
3414          */
3415         if (!bh)
3416                 return -EIO;
3417
3418         super = (struct btrfs_super_block *)bh->b_data;
3419         if (btrfs_super_bytenr(super) != bytenr ||
3420                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3421                 brelse(bh);
3422                 return -EINVAL;
3423         }
3424
3425         *bh_ret = bh;
3426         return 0;
3427 }
3428
3429
3430 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3431 {
3432         struct buffer_head *bh;
3433         struct buffer_head *latest = NULL;
3434         struct btrfs_super_block *super;
3435         int i;
3436         u64 transid = 0;
3437         int ret = -EINVAL;
3438
3439         /* we would like to check all the supers, but that would make
3440          * a btrfs mount succeed after a mkfs from a different FS.
3441          * So, we need to add a special mount option to scan for
3442          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3443          */
3444         for (i = 0; i < 1; i++) {
3445                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3446                 if (ret)
3447                         continue;
3448
3449                 super = (struct btrfs_super_block *)bh->b_data;
3450
3451                 if (!latest || btrfs_super_generation(super) > transid) {
3452                         brelse(latest);
3453                         latest = bh;
3454                         transid = btrfs_super_generation(super);
3455                 } else {
3456                         brelse(bh);
3457                 }
3458         }
3459
3460         if (!latest)
3461                 return ERR_PTR(ret);
3462
3463         return latest;
3464 }
3465
3466 /*
3467  * Write superblock @sb to the @device. Do not wait for completion, all the
3468  * buffer heads we write are pinned.
3469  *
3470  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3471  * the expected device size at commit time. Note that max_mirrors must be
3472  * same for write and wait phases.
3473  *
3474  * Return number of errors when buffer head is not found or submission fails.
3475  */
3476 static int write_dev_supers(struct btrfs_device *device,
3477                             struct btrfs_super_block *sb, int max_mirrors)
3478 {
3479         struct buffer_head *bh;
3480         int i;
3481         int ret;
3482         int errors = 0;
3483         u32 crc;
3484         u64 bytenr;
3485         int op_flags;
3486
3487         if (max_mirrors == 0)
3488                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3489
3490         for (i = 0; i < max_mirrors; i++) {
3491                 bytenr = btrfs_sb_offset(i);
3492                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3493                     device->commit_total_bytes)
3494                         break;
3495
3496                 btrfs_set_super_bytenr(sb, bytenr);
3497
3498                 crc = ~(u32)0;
3499                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3500                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3501                 btrfs_csum_final(crc, sb->csum);
3502
3503                 /* One reference for us, and we leave it for the caller */
3504                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3505                               BTRFS_SUPER_INFO_SIZE);
3506                 if (!bh) {
3507                         btrfs_err(device->fs_info,
3508                             "couldn't get super buffer head for bytenr %llu",
3509                             bytenr);
3510                         errors++;
3511                         continue;
3512                 }
3513
3514                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3515
3516                 /* one reference for submit_bh */
3517                 get_bh(bh);
3518
3519                 set_buffer_uptodate(bh);
3520                 lock_buffer(bh);
3521                 bh->b_end_io = btrfs_end_buffer_write_sync;
3522                 bh->b_private = device;
3523
3524                 /*
3525                  * we fua the first super.  The others we allow
3526                  * to go down lazy.
3527                  */
3528                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3529                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3530                         op_flags |= REQ_FUA;
3531                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3532                 if (ret)
3533                         errors++;
3534         }
3535         return errors < i ? 0 : -1;
3536 }
3537
3538 /*
3539  * Wait for write completion of superblocks done by write_dev_supers,
3540  * @max_mirrors same for write and wait phases.
3541  *
3542  * Return number of errors when buffer head is not found or not marked up to
3543  * date.
3544  */
3545 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3546 {
3547         struct buffer_head *bh;
3548         int i;
3549         int errors = 0;
3550         bool primary_failed = false;
3551         u64 bytenr;
3552
3553         if (max_mirrors == 0)
3554                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3555
3556         for (i = 0; i < max_mirrors; i++) {
3557                 bytenr = btrfs_sb_offset(i);
3558                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3559                     device->commit_total_bytes)
3560                         break;
3561
3562                 bh = __find_get_block(device->bdev,
3563                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3564                                       BTRFS_SUPER_INFO_SIZE);
3565                 if (!bh) {
3566                         errors++;
3567                         if (i == 0)
3568                                 primary_failed = true;
3569                         continue;
3570                 }
3571                 wait_on_buffer(bh);
3572                 if (!buffer_uptodate(bh)) {
3573                         errors++;
3574                         if (i == 0)
3575                                 primary_failed = true;
3576                 }
3577
3578                 /* drop our reference */
3579                 brelse(bh);
3580
3581                 /* drop the reference from the writing run */
3582                 brelse(bh);
3583         }
3584
3585         /* log error, force error return */
3586         if (primary_failed) {
3587                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3588                           device->devid);
3589                 return -1;
3590         }
3591
3592         return errors < i ? 0 : -1;
3593 }
3594
3595 /*
3596  * endio for the write_dev_flush, this will wake anyone waiting
3597  * for the barrier when it is done
3598  */
3599 static void btrfs_end_empty_barrier(struct bio *bio)
3600 {
3601         complete(bio->bi_private);
3602 }
3603
3604 /*
3605  * Submit a flush request to the device if it supports it. Error handling is
3606  * done in the waiting counterpart.
3607  */
3608 static void write_dev_flush(struct btrfs_device *device)
3609 {
3610         struct request_queue *q = bdev_get_queue(device->bdev);
3611         struct bio *bio = device->flush_bio;
3612
3613         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3614                 return;
3615
3616         bio_reset(bio);
3617         bio->bi_end_io = btrfs_end_empty_barrier;
3618         bio_set_dev(bio, device->bdev);
3619         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3620         init_completion(&device->flush_wait);
3621         bio->bi_private = &device->flush_wait;
3622
3623         btrfsic_submit_bio(bio);
3624         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3625 }
3626
3627 /*
3628  * If the flush bio has been submitted by write_dev_flush, wait for it.
3629  */
3630 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3631 {
3632         struct bio *bio = device->flush_bio;
3633
3634         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3635                 return BLK_STS_OK;
3636
3637         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3638         wait_for_completion_io(&device->flush_wait);
3639
3640         return bio->bi_status;
3641 }
3642
3643 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3644 {
3645         if (!btrfs_check_rw_degradable(fs_info, NULL))
3646                 return -EIO;
3647         return 0;
3648 }
3649
3650 /*
3651  * send an empty flush down to each device in parallel,
3652  * then wait for them
3653  */
3654 static int barrier_all_devices(struct btrfs_fs_info *info)
3655 {
3656         struct list_head *head;
3657         struct btrfs_device *dev;
3658         int errors_wait = 0;
3659         blk_status_t ret;
3660
3661         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3662         /* send down all the barriers */
3663         head = &info->fs_devices->devices;
3664         list_for_each_entry(dev, head, dev_list) {
3665                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3666                         continue;
3667                 if (!dev->bdev)
3668                         continue;
3669                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3670                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3671                         continue;
3672
3673                 write_dev_flush(dev);
3674                 dev->last_flush_error = BLK_STS_OK;
3675         }
3676
3677         /* wait for all the barriers */
3678         list_for_each_entry(dev, head, dev_list) {
3679                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3680                         continue;
3681                 if (!dev->bdev) {
3682                         errors_wait++;
3683                         continue;
3684                 }
3685                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3686                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3687                         continue;
3688
3689                 ret = wait_dev_flush(dev);
3690                 if (ret) {
3691                         dev->last_flush_error = ret;
3692                         btrfs_dev_stat_inc_and_print(dev,
3693                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3694                         errors_wait++;
3695                 }
3696         }
3697
3698         if (errors_wait) {
3699                 /*
3700                  * At some point we need the status of all disks
3701                  * to arrive at the volume status. So error checking
3702                  * is being pushed to a separate loop.
3703                  */
3704                 return check_barrier_error(info);
3705         }
3706         return 0;
3707 }
3708
3709 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3710 {
3711         int raid_type;
3712         int min_tolerated = INT_MAX;
3713
3714         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3715             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3716                 min_tolerated = min_t(int, min_tolerated,
3717                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3718                                     tolerated_failures);
3719
3720         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3721                 if (raid_type == BTRFS_RAID_SINGLE)
3722                         continue;
3723                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3724                         continue;
3725                 min_tolerated = min_t(int, min_tolerated,
3726                                     btrfs_raid_array[raid_type].
3727                                     tolerated_failures);
3728         }
3729
3730         if (min_tolerated == INT_MAX) {
3731                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3732                 min_tolerated = 0;
3733         }
3734
3735         return min_tolerated;
3736 }
3737
3738 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3739 {
3740         struct list_head *head;
3741         struct btrfs_device *dev;
3742         struct btrfs_super_block *sb;
3743         struct btrfs_dev_item *dev_item;
3744         int ret;
3745         int do_barriers;
3746         int max_errors;
3747         int total_errors = 0;
3748         u64 flags;
3749
3750         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3751
3752         /*
3753          * max_mirrors == 0 indicates we're from commit_transaction,
3754          * not from fsync where the tree roots in fs_info have not
3755          * been consistent on disk.
3756          */
3757         if (max_mirrors == 0)
3758                 backup_super_roots(fs_info);
3759
3760         sb = fs_info->super_for_commit;
3761         dev_item = &sb->dev_item;
3762
3763         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3764         head = &fs_info->fs_devices->devices;
3765         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3766
3767         if (do_barriers) {
3768                 ret = barrier_all_devices(fs_info);
3769                 if (ret) {
3770                         mutex_unlock(
3771                                 &fs_info->fs_devices->device_list_mutex);
3772                         btrfs_handle_fs_error(fs_info, ret,
3773                                               "errors while submitting device barriers.");
3774                         return ret;
3775                 }
3776         }
3777
3778         list_for_each_entry(dev, head, dev_list) {
3779                 if (!dev->bdev) {
3780                         total_errors++;
3781                         continue;
3782                 }
3783                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3784                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3785                         continue;
3786
3787                 btrfs_set_stack_device_generation(dev_item, 0);
3788                 btrfs_set_stack_device_type(dev_item, dev->type);
3789                 btrfs_set_stack_device_id(dev_item, dev->devid);
3790                 btrfs_set_stack_device_total_bytes(dev_item,
3791                                                    dev->commit_total_bytes);
3792                 btrfs_set_stack_device_bytes_used(dev_item,
3793                                                   dev->commit_bytes_used);
3794                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3795                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3796                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3797                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3798                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3799                        BTRFS_FSID_SIZE);
3800
3801                 flags = btrfs_super_flags(sb);
3802                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3803
3804                 ret = btrfs_validate_write_super(fs_info, sb);
3805                 if (ret < 0) {
3806                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3807                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3808                                 "unexpected superblock corruption detected");
3809                         return -EUCLEAN;
3810                 }
3811
3812                 ret = write_dev_supers(dev, sb, max_mirrors);
3813                 if (ret)
3814                         total_errors++;
3815         }
3816         if (total_errors > max_errors) {
3817                 btrfs_err(fs_info, "%d errors while writing supers",
3818                           total_errors);
3819                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3820
3821                 /* FUA is masked off if unsupported and can't be the reason */
3822                 btrfs_handle_fs_error(fs_info, -EIO,
3823                                       "%d errors while writing supers",
3824                                       total_errors);
3825                 return -EIO;
3826         }
3827
3828         total_errors = 0;
3829         list_for_each_entry(dev, head, dev_list) {
3830                 if (!dev->bdev)
3831                         continue;
3832                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3833                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3834                         continue;
3835
3836                 ret = wait_dev_supers(dev, max_mirrors);
3837                 if (ret)
3838                         total_errors++;
3839         }
3840         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3841         if (total_errors > max_errors) {
3842                 btrfs_handle_fs_error(fs_info, -EIO,
3843                                       "%d errors while writing supers",
3844                                       total_errors);
3845                 return -EIO;
3846         }
3847         return 0;
3848 }
3849
3850 /* Drop a fs root from the radix tree and free it. */
3851 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3852                                   struct btrfs_root *root)
3853 {
3854         spin_lock(&fs_info->fs_roots_radix_lock);
3855         radix_tree_delete(&fs_info->fs_roots_radix,
3856                           (unsigned long)root->root_key.objectid);
3857         spin_unlock(&fs_info->fs_roots_radix_lock);
3858
3859         if (btrfs_root_refs(&root->root_item) == 0)
3860                 synchronize_srcu(&fs_info->subvol_srcu);
3861
3862         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3863                 btrfs_free_log(NULL, root);
3864                 if (root->reloc_root) {
3865                         free_extent_buffer(root->reloc_root->node);
3866                         free_extent_buffer(root->reloc_root->commit_root);
3867                         btrfs_put_fs_root(root->reloc_root);
3868                         root->reloc_root = NULL;
3869                 }
3870         }
3871
3872         if (root->free_ino_pinned)
3873                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3874         if (root->free_ino_ctl)
3875                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3876         btrfs_free_fs_root(root);
3877 }
3878
3879 void btrfs_free_fs_root(struct btrfs_root *root)
3880 {
3881         iput(root->ino_cache_inode);
3882         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3883         if (root->anon_dev)
3884                 free_anon_bdev(root->anon_dev);
3885         if (root->subv_writers)
3886                 btrfs_free_subvolume_writers(root->subv_writers);
3887         free_extent_buffer(root->node);
3888         free_extent_buffer(root->commit_root);
3889         kfree(root->free_ino_ctl);
3890         kfree(root->free_ino_pinned);
3891         btrfs_put_fs_root(root);
3892 }
3893
3894 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3895 {
3896         u64 root_objectid = 0;
3897         struct btrfs_root *gang[8];
3898         int i = 0;
3899         int err = 0;
3900         unsigned int ret = 0;
3901         int index;
3902
3903         while (1) {
3904                 index = srcu_read_lock(&fs_info->subvol_srcu);
3905                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3906                                              (void **)gang, root_objectid,
3907                                              ARRAY_SIZE(gang));
3908                 if (!ret) {
3909                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3910                         break;
3911                 }
3912                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3913
3914                 for (i = 0; i < ret; i++) {
3915                         /* Avoid to grab roots in dead_roots */
3916                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3917                                 gang[i] = NULL;
3918                                 continue;
3919                         }
3920                         /* grab all the search result for later use */
3921                         gang[i] = btrfs_grab_fs_root(gang[i]);
3922                 }
3923                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3924
3925                 for (i = 0; i < ret; i++) {
3926                         if (!gang[i])
3927                                 continue;
3928                         root_objectid = gang[i]->root_key.objectid;
3929                         err = btrfs_orphan_cleanup(gang[i]);
3930                         if (err)
3931                                 break;
3932                         btrfs_put_fs_root(gang[i]);
3933                 }
3934                 root_objectid++;
3935         }
3936
3937         /* release the uncleaned roots due to error */
3938         for (; i < ret; i++) {
3939                 if (gang[i])
3940                         btrfs_put_fs_root(gang[i]);
3941         }
3942         return err;
3943 }
3944
3945 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3946 {
3947         struct btrfs_root *root = fs_info->tree_root;
3948         struct btrfs_trans_handle *trans;
3949
3950         mutex_lock(&fs_info->cleaner_mutex);
3951         btrfs_run_delayed_iputs(fs_info);
3952         mutex_unlock(&fs_info->cleaner_mutex);
3953         wake_up_process(fs_info->cleaner_kthread);
3954
3955         /* wait until ongoing cleanup work done */
3956         down_write(&fs_info->cleanup_work_sem);
3957         up_write(&fs_info->cleanup_work_sem);
3958
3959         trans = btrfs_join_transaction(root);
3960         if (IS_ERR(trans))
3961                 return PTR_ERR(trans);
3962         return btrfs_commit_transaction(trans);
3963 }
3964
3965 void close_ctree(struct btrfs_fs_info *fs_info)
3966 {
3967         int ret;
3968
3969         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3970         /*
3971          * We don't want the cleaner to start new transactions, add more delayed
3972          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3973          * because that frees the task_struct, and the transaction kthread might
3974          * still try to wake up the cleaner.
3975          */
3976         kthread_park(fs_info->cleaner_kthread);
3977
3978         /* wait for the qgroup rescan worker to stop */
3979         btrfs_qgroup_wait_for_completion(fs_info, false);
3980
3981         /* wait for the uuid_scan task to finish */
3982         down(&fs_info->uuid_tree_rescan_sem);
3983         /* avoid complains from lockdep et al., set sem back to initial state */
3984         up(&fs_info->uuid_tree_rescan_sem);
3985
3986         /* pause restriper - we want to resume on mount */
3987         btrfs_pause_balance(fs_info);
3988
3989         btrfs_dev_replace_suspend_for_unmount(fs_info);
3990
3991         btrfs_scrub_cancel(fs_info);
3992
3993         /* wait for any defraggers to finish */
3994         wait_event(fs_info->transaction_wait,
3995                    (atomic_read(&fs_info->defrag_running) == 0));
3996
3997         /* clear out the rbtree of defraggable inodes */
3998         btrfs_cleanup_defrag_inodes(fs_info);
3999
4000         cancel_work_sync(&fs_info->async_reclaim_work);
4001
4002         if (!sb_rdonly(fs_info->sb)) {
4003                 /*
4004                  * The cleaner kthread is stopped, so do one final pass over
4005                  * unused block groups.
4006                  */
4007                 btrfs_delete_unused_bgs(fs_info);
4008
4009                 ret = btrfs_commit_super(fs_info);
4010                 if (ret)
4011                         btrfs_err(fs_info, "commit super ret %d", ret);
4012         }
4013
4014         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4015             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4016                 btrfs_error_commit_super(fs_info);
4017
4018         kthread_stop(fs_info->transaction_kthread);
4019         kthread_stop(fs_info->cleaner_kthread);
4020
4021         ASSERT(list_empty(&fs_info->delayed_iputs));
4022         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4023
4024         btrfs_free_qgroup_config(fs_info);
4025         ASSERT(list_empty(&fs_info->delalloc_roots));
4026
4027         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4028                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4029                        percpu_counter_sum(&fs_info->delalloc_bytes));
4030         }
4031
4032         if (percpu_counter_sum(&fs_info->dio_bytes))
4033                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4034                            percpu_counter_sum(&fs_info->dio_bytes));
4035
4036         btrfs_sysfs_remove_mounted(fs_info);
4037         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4038
4039         btrfs_free_fs_roots(fs_info);
4040
4041         btrfs_put_block_group_cache(fs_info);
4042
4043         /*
4044          * we must make sure there is not any read request to
4045          * submit after we stopping all workers.
4046          */
4047         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4048         btrfs_stop_all_workers(fs_info);
4049
4050         btrfs_free_block_groups(fs_info);
4051
4052         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4053         free_root_pointers(fs_info, 1);
4054
4055         iput(fs_info->btree_inode);
4056
4057 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4058         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4059                 btrfsic_unmount(fs_info->fs_devices);
4060 #endif
4061
4062         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4063         btrfs_close_devices(fs_info->fs_devices);
4064
4065         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4066         percpu_counter_destroy(&fs_info->delalloc_bytes);
4067         percpu_counter_destroy(&fs_info->dio_bytes);
4068         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4069         cleanup_srcu_struct(&fs_info->subvol_srcu);
4070
4071         btrfs_free_stripe_hash_table(fs_info);
4072         btrfs_free_ref_cache(fs_info);
4073 }
4074
4075 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4076                           int atomic)
4077 {
4078         int ret;
4079         struct inode *btree_inode = buf->pages[0]->mapping->host;
4080
4081         ret = extent_buffer_uptodate(buf);
4082         if (!ret)
4083                 return ret;
4084
4085         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4086                                     parent_transid, atomic);
4087         if (ret == -EAGAIN)
4088                 return ret;
4089         return !ret;
4090 }
4091
4092 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4093 {
4094         struct btrfs_fs_info *fs_info;
4095         struct btrfs_root *root;
4096         u64 transid = btrfs_header_generation(buf);
4097         int was_dirty;
4098
4099 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4100         /*
4101          * This is a fast path so only do this check if we have sanity tests
4102          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4103          * outside of the sanity tests.
4104          */
4105         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4106                 return;
4107 #endif
4108         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4109         fs_info = root->fs_info;
4110         btrfs_assert_tree_locked(buf);
4111         if (transid != fs_info->generation)
4112                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4113                         buf->start, transid, fs_info->generation);
4114         was_dirty = set_extent_buffer_dirty(buf);
4115         if (!was_dirty)
4116                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4117                                          buf->len,
4118                                          fs_info->dirty_metadata_batch);
4119 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4120         /*
4121          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4122          * but item data not updated.
4123          * So here we should only check item pointers, not item data.
4124          */
4125         if (btrfs_header_level(buf) == 0 &&
4126             btrfs_check_leaf_relaxed(buf)) {
4127                 btrfs_print_leaf(buf);
4128                 ASSERT(0);
4129         }
4130 #endif
4131 }
4132
4133 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4134                                         int flush_delayed)
4135 {
4136         /*
4137          * looks as though older kernels can get into trouble with
4138          * this code, they end up stuck in balance_dirty_pages forever
4139          */
4140         int ret;
4141
4142         if (current->flags & PF_MEMALLOC)
4143                 return;
4144
4145         if (flush_delayed)
4146                 btrfs_balance_delayed_items(fs_info);
4147
4148         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4149                                      BTRFS_DIRTY_METADATA_THRESH,
4150                                      fs_info->dirty_metadata_batch);
4151         if (ret > 0) {
4152                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4153         }
4154 }
4155
4156 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4157 {
4158         __btrfs_btree_balance_dirty(fs_info, 1);
4159 }
4160
4161 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4162 {
4163         __btrfs_btree_balance_dirty(fs_info, 0);
4164 }
4165
4166 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4167                       struct btrfs_key *first_key)
4168 {
4169         return btree_read_extent_buffer_pages(buf, parent_transid,
4170                                               level, first_key);
4171 }
4172
4173 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4174 {
4175         /* cleanup FS via transaction */
4176         btrfs_cleanup_transaction(fs_info);
4177
4178         mutex_lock(&fs_info->cleaner_mutex);
4179         btrfs_run_delayed_iputs(fs_info);
4180         mutex_unlock(&fs_info->cleaner_mutex);
4181
4182         down_write(&fs_info->cleanup_work_sem);
4183         up_write(&fs_info->cleanup_work_sem);
4184 }
4185
4186 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4187 {
4188         struct btrfs_ordered_extent *ordered;
4189
4190         spin_lock(&root->ordered_extent_lock);
4191         /*
4192          * This will just short circuit the ordered completion stuff which will
4193          * make sure the ordered extent gets properly cleaned up.
4194          */
4195         list_for_each_entry(ordered, &root->ordered_extents,
4196                             root_extent_list)
4197                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4198         spin_unlock(&root->ordered_extent_lock);
4199 }
4200
4201 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4202 {
4203         struct btrfs_root *root;
4204         struct list_head splice;
4205
4206         INIT_LIST_HEAD(&splice);
4207
4208         spin_lock(&fs_info->ordered_root_lock);
4209         list_splice_init(&fs_info->ordered_roots, &splice);
4210         while (!list_empty(&splice)) {
4211                 root = list_first_entry(&splice, struct btrfs_root,
4212                                         ordered_root);
4213                 list_move_tail(&root->ordered_root,
4214                                &fs_info->ordered_roots);
4215
4216                 spin_unlock(&fs_info->ordered_root_lock);
4217                 btrfs_destroy_ordered_extents(root);
4218
4219                 cond_resched();
4220                 spin_lock(&fs_info->ordered_root_lock);
4221         }
4222         spin_unlock(&fs_info->ordered_root_lock);
4223
4224         /*
4225          * We need this here because if we've been flipped read-only we won't
4226          * get sync() from the umount, so we need to make sure any ordered
4227          * extents that haven't had their dirty pages IO start writeout yet
4228          * actually get run and error out properly.
4229          */
4230         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4231 }
4232
4233 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4234                                       struct btrfs_fs_info *fs_info)
4235 {
4236         struct rb_node *node;
4237         struct btrfs_delayed_ref_root *delayed_refs;
4238         struct btrfs_delayed_ref_node *ref;
4239         int ret = 0;
4240
4241         delayed_refs = &trans->delayed_refs;
4242
4243         spin_lock(&delayed_refs->lock);
4244         if (atomic_read(&delayed_refs->num_entries) == 0) {
4245                 spin_unlock(&delayed_refs->lock);
4246                 btrfs_info(fs_info, "delayed_refs has NO entry");
4247                 return ret;
4248         }
4249
4250         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4251                 struct btrfs_delayed_ref_head *head;
4252                 struct rb_node *n;
4253                 bool pin_bytes = false;
4254
4255                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4256                                 href_node);
4257                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4258                         continue;
4259
4260                 spin_lock(&head->lock);
4261                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4262                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4263                                        ref_node);
4264                         ref->in_tree = 0;
4265                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4266                         RB_CLEAR_NODE(&ref->ref_node);
4267                         if (!list_empty(&ref->add_list))
4268                                 list_del(&ref->add_list);
4269                         atomic_dec(&delayed_refs->num_entries);
4270                         btrfs_put_delayed_ref(ref);
4271                 }
4272                 if (head->must_insert_reserved)
4273                         pin_bytes = true;
4274                 btrfs_free_delayed_extent_op(head->extent_op);
4275                 btrfs_delete_ref_head(delayed_refs, head);
4276                 spin_unlock(&head->lock);
4277                 spin_unlock(&delayed_refs->lock);
4278                 mutex_unlock(&head->mutex);
4279
4280                 if (pin_bytes)
4281                         btrfs_pin_extent(fs_info, head->bytenr,
4282                                          head->num_bytes, 1);
4283                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4284                 btrfs_put_delayed_ref_head(head);
4285                 cond_resched();
4286                 spin_lock(&delayed_refs->lock);
4287         }
4288
4289         spin_unlock(&delayed_refs->lock);
4290
4291         return ret;
4292 }
4293
4294 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4295 {
4296         struct btrfs_inode *btrfs_inode;
4297         struct list_head splice;
4298
4299         INIT_LIST_HEAD(&splice);
4300
4301         spin_lock(&root->delalloc_lock);
4302         list_splice_init(&root->delalloc_inodes, &splice);
4303
4304         while (!list_empty(&splice)) {
4305                 struct inode *inode = NULL;
4306                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4307                                                delalloc_inodes);
4308                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4309                 spin_unlock(&root->delalloc_lock);
4310
4311                 /*
4312                  * Make sure we get a live inode and that it'll not disappear
4313                  * meanwhile.
4314                  */
4315                 inode = igrab(&btrfs_inode->vfs_inode);
4316                 if (inode) {
4317                         invalidate_inode_pages2(inode->i_mapping);
4318                         iput(inode);
4319                 }
4320                 spin_lock(&root->delalloc_lock);
4321         }
4322         spin_unlock(&root->delalloc_lock);
4323 }
4324
4325 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4326 {
4327         struct btrfs_root *root;
4328         struct list_head splice;
4329
4330         INIT_LIST_HEAD(&splice);
4331
4332         spin_lock(&fs_info->delalloc_root_lock);
4333         list_splice_init(&fs_info->delalloc_roots, &splice);
4334         while (!list_empty(&splice)) {
4335                 root = list_first_entry(&splice, struct btrfs_root,
4336                                          delalloc_root);
4337                 root = btrfs_grab_fs_root(root);
4338                 BUG_ON(!root);
4339                 spin_unlock(&fs_info->delalloc_root_lock);
4340
4341                 btrfs_destroy_delalloc_inodes(root);
4342                 btrfs_put_fs_root(root);
4343
4344                 spin_lock(&fs_info->delalloc_root_lock);
4345         }
4346         spin_unlock(&fs_info->delalloc_root_lock);
4347 }
4348
4349 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4350                                         struct extent_io_tree *dirty_pages,
4351                                         int mark)
4352 {
4353         int ret;
4354         struct extent_buffer *eb;
4355         u64 start = 0;
4356         u64 end;
4357
4358         while (1) {
4359                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4360                                             mark, NULL);
4361                 if (ret)
4362                         break;
4363
4364                 clear_extent_bits(dirty_pages, start, end, mark);
4365                 while (start <= end) {
4366                         eb = find_extent_buffer(fs_info, start);
4367                         start += fs_info->nodesize;
4368                         if (!eb)
4369                                 continue;
4370                         wait_on_extent_buffer_writeback(eb);
4371
4372                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4373                                                &eb->bflags))
4374                                 clear_extent_buffer_dirty(eb);
4375                         free_extent_buffer_stale(eb);
4376                 }
4377         }
4378
4379         return ret;
4380 }
4381
4382 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4383                                        struct extent_io_tree *pinned_extents)
4384 {
4385         struct extent_io_tree *unpin;
4386         u64 start;
4387         u64 end;
4388         int ret;
4389         bool loop = true;
4390
4391         unpin = pinned_extents;
4392 again:
4393         while (1) {
4394                 struct extent_state *cached_state = NULL;
4395
4396                 /*
4397                  * The btrfs_finish_extent_commit() may get the same range as
4398                  * ours between find_first_extent_bit and clear_extent_dirty.
4399                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4400                  * the same extent range.
4401                  */
4402                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4403                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4404                                             EXTENT_DIRTY, &cached_state);
4405                 if (ret) {
4406                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4407                         break;
4408                 }
4409
4410                 clear_extent_dirty(unpin, start, end, &cached_state);
4411                 free_extent_state(cached_state);
4412                 btrfs_error_unpin_extent_range(fs_info, start, end);
4413                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4414                 cond_resched();
4415         }
4416
4417         if (loop) {
4418                 if (unpin == &fs_info->freed_extents[0])
4419                         unpin = &fs_info->freed_extents[1];
4420                 else
4421                         unpin = &fs_info->freed_extents[0];
4422                 loop = false;
4423                 goto again;
4424         }
4425
4426         return 0;
4427 }
4428
4429 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4430 {
4431         struct inode *inode;
4432
4433         inode = cache->io_ctl.inode;
4434         if (inode) {
4435                 invalidate_inode_pages2(inode->i_mapping);
4436                 BTRFS_I(inode)->generation = 0;
4437                 cache->io_ctl.inode = NULL;
4438                 iput(inode);
4439         }
4440         btrfs_put_block_group(cache);
4441 }
4442
4443 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4444                              struct btrfs_fs_info *fs_info)
4445 {
4446         struct btrfs_block_group_cache *cache;
4447
4448         spin_lock(&cur_trans->dirty_bgs_lock);
4449         while (!list_empty(&cur_trans->dirty_bgs)) {
4450                 cache = list_first_entry(&cur_trans->dirty_bgs,
4451                                          struct btrfs_block_group_cache,
4452                                          dirty_list);
4453
4454                 if (!list_empty(&cache->io_list)) {
4455                         spin_unlock(&cur_trans->dirty_bgs_lock);
4456                         list_del_init(&cache->io_list);
4457                         btrfs_cleanup_bg_io(cache);
4458                         spin_lock(&cur_trans->dirty_bgs_lock);
4459                 }
4460
4461                 list_del_init(&cache->dirty_list);
4462                 spin_lock(&cache->lock);
4463                 cache->disk_cache_state = BTRFS_DC_ERROR;
4464                 spin_unlock(&cache->lock);
4465
4466                 spin_unlock(&cur_trans->dirty_bgs_lock);
4467                 btrfs_put_block_group(cache);
4468                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4469                 spin_lock(&cur_trans->dirty_bgs_lock);
4470         }
4471         spin_unlock(&cur_trans->dirty_bgs_lock);
4472
4473         /*
4474          * Refer to the definition of io_bgs member for details why it's safe
4475          * to use it without any locking
4476          */
4477         while (!list_empty(&cur_trans->io_bgs)) {
4478                 cache = list_first_entry(&cur_trans->io_bgs,
4479                                          struct btrfs_block_group_cache,
4480                                          io_list);
4481
4482                 list_del_init(&cache->io_list);
4483                 spin_lock(&cache->lock);
4484                 cache->disk_cache_state = BTRFS_DC_ERROR;
4485                 spin_unlock(&cache->lock);
4486                 btrfs_cleanup_bg_io(cache);
4487         }
4488 }
4489
4490 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4491                                    struct btrfs_fs_info *fs_info)
4492 {
4493         struct btrfs_device *dev, *tmp;
4494
4495         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4496         ASSERT(list_empty(&cur_trans->dirty_bgs));
4497         ASSERT(list_empty(&cur_trans->io_bgs));
4498
4499         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4500                                  post_commit_list) {
4501                 list_del_init(&dev->post_commit_list);
4502         }
4503
4504         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4505
4506         cur_trans->state = TRANS_STATE_COMMIT_START;
4507         wake_up(&fs_info->transaction_blocked_wait);
4508
4509         cur_trans->state = TRANS_STATE_UNBLOCKED;
4510         wake_up(&fs_info->transaction_wait);
4511
4512         btrfs_destroy_delayed_inodes(fs_info);
4513         btrfs_assert_delayed_root_empty(fs_info);
4514
4515         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4516                                      EXTENT_DIRTY);
4517         btrfs_destroy_pinned_extent(fs_info,
4518                                     fs_info->pinned_extents);
4519
4520         cur_trans->state =TRANS_STATE_COMPLETED;
4521         wake_up(&cur_trans->commit_wait);
4522 }
4523
4524 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4525 {
4526         struct btrfs_transaction *t;
4527
4528         mutex_lock(&fs_info->transaction_kthread_mutex);
4529
4530         spin_lock(&fs_info->trans_lock);
4531         while (!list_empty(&fs_info->trans_list)) {
4532                 t = list_first_entry(&fs_info->trans_list,
4533                                      struct btrfs_transaction, list);
4534                 if (t->state >= TRANS_STATE_COMMIT_START) {
4535                         refcount_inc(&t->use_count);
4536                         spin_unlock(&fs_info->trans_lock);
4537                         btrfs_wait_for_commit(fs_info, t->transid);
4538                         btrfs_put_transaction(t);
4539                         spin_lock(&fs_info->trans_lock);
4540                         continue;
4541                 }
4542                 if (t == fs_info->running_transaction) {
4543                         t->state = TRANS_STATE_COMMIT_DOING;
4544                         spin_unlock(&fs_info->trans_lock);
4545                         /*
4546                          * We wait for 0 num_writers since we don't hold a trans
4547                          * handle open currently for this transaction.
4548                          */
4549                         wait_event(t->writer_wait,
4550                                    atomic_read(&t->num_writers) == 0);
4551                 } else {
4552                         spin_unlock(&fs_info->trans_lock);
4553                 }
4554                 btrfs_cleanup_one_transaction(t, fs_info);
4555
4556                 spin_lock(&fs_info->trans_lock);
4557                 if (t == fs_info->running_transaction)
4558                         fs_info->running_transaction = NULL;
4559                 list_del_init(&t->list);
4560                 spin_unlock(&fs_info->trans_lock);
4561
4562                 btrfs_put_transaction(t);
4563                 trace_btrfs_transaction_commit(fs_info->tree_root);
4564                 spin_lock(&fs_info->trans_lock);
4565         }
4566         spin_unlock(&fs_info->trans_lock);
4567         btrfs_destroy_all_ordered_extents(fs_info);
4568         btrfs_destroy_delayed_inodes(fs_info);
4569         btrfs_assert_delayed_root_empty(fs_info);
4570         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4571         btrfs_destroy_all_delalloc_inodes(fs_info);
4572         mutex_unlock(&fs_info->transaction_kthread_mutex);
4573
4574         return 0;
4575 }
4576
4577 static const struct extent_io_ops btree_extent_io_ops = {
4578         /* mandatory callbacks */
4579         .submit_bio_hook = btree_submit_bio_hook,
4580         .readpage_end_io_hook = btree_readpage_end_io_hook,
4581 };