fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/slab.h>
  30 #include <linux/migrate.h>
  31 #include <linux/ratelimit.h>
  32 #include <linux/uuid.h>
  33 #include <linux/semaphore.h>
  34 #include <asm/unaligned.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "hash.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "volumes.h"
  41 #include "print-tree.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  64                                       struct btrfs_root *root);
  65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  67                                         struct extent_io_tree *dirty_pages,
  68                                         int mark);
  69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  70                                        struct extent_io_tree *pinned_extents);
  71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  72 static void btrfs_error_commit_super(struct btrfs_root *root);
  73
  74 /*
  75  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  76  * is complete.  This is used during reads to verify checksums, and it is used
  77  * by writes to insert metadata for new file extents after IO is complete.
  78  */
  79 struct btrfs_end_io_wq {
  80         struct bio *bio;
  81         bio_end_io_t *end_io;
  82         void *private;
  83         struct btrfs_fs_info *info;
  84         int error;
  85         enum btrfs_wq_endio_type metadata;
  86         struct list_head list;
  87         struct btrfs_work work;
  88 };
  89
  90 static struct kmem_cache *btrfs_end_io_wq_cache;
  91
  92 int __init btrfs_end_io_wq_init(void)
  93 {
  94         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  95                                         sizeof(struct btrfs_end_io_wq),
  96                                         0,
  97                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  98                                         NULL);
  99         if (!btrfs_end_io_wq_cache)
 100                 return -ENOMEM;
 101         return 0;
 102 }
 103
 104 void btrfs_end_io_wq_exit(void)
 105 {
 106         if (btrfs_end_io_wq_cache)
 107                 kmem_cache_destroy(btrfs_end_io_wq_cache);
 108 }
 109
 110 /*
 111  * async submit bios are used to offload expensive checksumming
 112  * onto the worker threads.  They checksum file and metadata bios
 113  * just before they are sent down the IO stack.
 114  */
 115 struct async_submit_bio {
 116         struct inode *inode;
 117         struct bio *bio;
 118         struct list_head list;
 119         extent_submit_bio_hook_t *submit_bio_start;
 120         extent_submit_bio_hook_t *submit_bio_done;
 121         int rw;
 122         int mirror_num;
 123         unsigned long bio_flags;
 124         /*
 125          * bio_offset is optional, can be used if the pages in the bio
 126          * can't tell us where in the file the bio should go
 127          */
 128         u64 bio_offset;
 129         struct btrfs_work work;
 130         int error;
 131 };
 132
 133 /*
 134  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 135  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 136  * the level the eb occupies in the tree.
 137  *
 138  * Different roots are used for different purposes and may nest inside each
 139  * other and they require separate keysets.  As lockdep keys should be
 140  * static, assign keysets according to the purpose of the root as indicated
 141  * by btrfs_root->objectid.  This ensures that all special purpose roots
 142  * have separate keysets.
 143  *
 144  * Lock-nesting across peer nodes is always done with the immediate parent
 145  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 146  * subclass to avoid triggering lockdep warning in such cases.
 147  *
 148  * The key is set by the readpage_end_io_hook after the buffer has passed
 149  * csum validation but before the pages are unlocked.  It is also set by
 150  * btrfs_init_new_buffer on freshly allocated blocks.
 151  *
 152  * We also add a check to make sure the highest level of the tree is the
 153  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 154  * needs update as well.
 155  */
 156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 157 # if BTRFS_MAX_LEVEL != 8
 158 #  error
 159 # endif
 160
 161 static struct btrfs_lockdep_keyset {
 162         u64                     id;             /* root objectid */
 163         const char              *name_stem;     /* lock name stem */
 164         char                    names[BTRFS_MAX_LEVEL + 1][20];
 165         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 166 } btrfs_lockdep_keysets[] = {
 167         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 168         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 169         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 170         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 171         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 172         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 173         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 174         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 175         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 176         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 177         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 178         { .id = 0,                              .name_stem = "tree"     },
 179 };
 180
 181 void __init btrfs_init_lockdep(void)
 182 {
 183         int i, j;
 184
 185         /* initialize lockdep class names */
 186         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 187                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 188
 189                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 190                         snprintf(ks->names[j], sizeof(ks->names[j]),
 191                                  "btrfs-%s-%02d", ks->name_stem, j);
 192         }
 193 }
 194
 195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 196                                     int level)
 197 {
 198         struct btrfs_lockdep_keyset *ks;
 199
 200         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 201
 202         /* find the matching keyset, id 0 is the default entry */
 203         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 204                 if (ks->id == objectid)
 205                         break;
 206
 207         lockdep_set_class_and_name(&eb->lock,
 208                                    &ks->keys[level], ks->names[level]);
 209 }
 210
 211 #endif
 212
 213 /*
 214  * extents on the btree inode are pretty simple, there's one extent
 215  * that covers the entire device
 216  */
 217 static struct extent_map *btree_get_extent(struct inode *inode,
 218                 struct page *page, size_t pg_offset, u64 start, u64 len,
 219                 int create)
 220 {
 221         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 222         struct extent_map *em;
 223         int ret;
 224
 225         read_lock(&em_tree->lock);
 226         em = lookup_extent_mapping(em_tree, start, len);
 227         if (em) {
 228                 em->bdev =
 229                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 230                 read_unlock(&em_tree->lock);
 231                 goto out;
 232         }
 233         read_unlock(&em_tree->lock);
 234
 235         em = alloc_extent_map();
 236         if (!em) {
 237                 em = ERR_PTR(-ENOMEM);
 238                 goto out;
 239         }
 240         em->start = 0;
 241         em->len = (u64)-1;
 242         em->block_len = (u64)-1;
 243         em->block_start = 0;
 244         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 245
 246         write_lock(&em_tree->lock);
 247         ret = add_extent_mapping(em_tree, em, 0);
 248         if (ret == -EEXIST) {
 249                 free_extent_map(em);
 250                 em = lookup_extent_mapping(em_tree, start, len);
 251                 if (!em)
 252                         em = ERR_PTR(-EIO);
 253         } else if (ret) {
 254                 free_extent_map(em);
 255                 em = ERR_PTR(ret);
 256         }
 257         write_unlock(&em_tree->lock);
 258
 259 out:
 260         return em;
 261 }
 262
 263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 264 {
 265         return btrfs_crc32c(seed, data, len);
 266 }
 267
 268 void btrfs_csum_final(u32 crc, char *result)
 269 {
 270         put_unaligned_le32(~crc, result);
 271 }
 272
 273 /*
 274  * compute the csum for a btree block, and either verify it or write it
 275  * into the csum field of the block.
 276  */
 277 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 278                            int verify)
 279 {
 280         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 281         char *result = NULL;
 282         unsigned long len;
 283         unsigned long cur_len;
 284         unsigned long offset = BTRFS_CSUM_SIZE;
 285         char *kaddr;
 286         unsigned long map_start;
 287         unsigned long map_len;
 288         int err;
 289         u32 crc = ~(u32)0;
 290         unsigned long inline_result;
 291
 292         len = buf->len - offset;
 293         while (len > 0) {
 294                 err = map_private_extent_buffer(buf, offset, 32,
 295                                         &kaddr, &map_start, &map_len);
 296                 if (err)
 297                         return 1;
 298                 cur_len = min(len, map_len - (offset - map_start));
 299                 crc = btrfs_csum_data(kaddr + offset - map_start,
 300                                       crc, cur_len);
 301                 len -= cur_len;
 302                 offset += cur_len;
 303         }
 304         if (csum_size > sizeof(inline_result)) {
 305                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 306                 if (!result)
 307                         return 1;
 308         } else {
 309                 result = (char *)&inline_result;
 310         }
 311
 312         btrfs_csum_final(crc, result);
 313
 314         if (verify) {
 315                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 316                         u32 val;
 317                         u32 found = 0;
 318                         memcpy(&found, result, csum_size);
 319
 320                         read_extent_buffer(buf, &val, 0, csum_size);
 321                         printk_ratelimited(KERN_INFO
 322                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 323                                 "level %d\n",
 324                                 root->fs_info->sb->s_id, buf->start,
 325                                 val, found, btrfs_header_level(buf));
 326                         if (result != (char *)&inline_result)
 327                                 kfree(result);
 328                         return 1;
 329                 }
 330         } else {
 331                 write_extent_buffer(buf, result, 0, csum_size);
 332         }
 333         if (result != (char *)&inline_result)
 334                 kfree(result);
 335         return 0;
 336 }
 337
 338 /*
 339  * we can't consider a given block up to date unless the transid of the
 340  * block matches the transid in the parent node's pointer.  This is how we
 341  * detect blocks that either didn't get written at all or got written
 342  * in the wrong place.
 343  */
 344 static int verify_parent_transid(struct extent_io_tree *io_tree,
 345                                  struct extent_buffer *eb, u64 parent_transid,
 346                                  int atomic)
 347 {
 348         struct extent_state *cached_state = NULL;
 349         int ret;
 350         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 351
 352         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 353                 return 0;
 354
 355         if (atomic)
 356                 return -EAGAIN;
 357
 358         if (need_lock) {
 359                 btrfs_tree_read_lock(eb);
 360                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 361         }
 362
 363         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 364                          0, &cached_state);
 365         if (extent_buffer_uptodate(eb) &&
 366             btrfs_header_generation(eb) == parent_transid) {
 367                 ret = 0;
 368                 goto out;
 369         }
 370         printk_ratelimited(KERN_INFO "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 371                         eb->fs_info->sb->s_id, eb->start,
 372                         parent_transid, btrfs_header_generation(eb));
 373         ret = 1;
 374
 375         /*
 376          * Things reading via commit roots that don't have normal protection,
 377          * like send, can have a really old block in cache that may point at a
 378          * block that has been free'd and re-allocated.  So don't clear uptodate
 379          * if we find an eb that is under IO (dirty/writeback) because we could
 380          * end up reading in the stale data and then writing it back out and
 381          * making everybody very sad.
 382          */
 383         if (!extent_buffer_under_io(eb))
 384                 clear_extent_buffer_uptodate(eb);
 385 out:
 386         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 387                              &cached_state, GFP_NOFS);
 388         if (need_lock)
 389                 btrfs_tree_read_unlock_blocking(eb);
 390         return ret;
 391 }
 392
 393 /*
 394  * Return 0 if the superblock checksum type matches the checksum value of that
 395  * algorithm. Pass the raw disk superblock data.
 396  */
 397 static int btrfs_check_super_csum(char *raw_disk_sb)
 398 {
 399         struct btrfs_super_block *disk_sb =
 400                 (struct btrfs_super_block *)raw_disk_sb;
 401         u16 csum_type = btrfs_super_csum_type(disk_sb);
 402         int ret = 0;
 403
 404         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 405                 u32 crc = ~(u32)0;
 406                 const int csum_size = sizeof(crc);
 407                 char result[csum_size];
 408
 409                 /*
 410                  * The super_block structure does not span the whole
 411                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 412                  * is filled with zeros and is included in the checkum.
 413                  */
 414                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 415                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 416                 btrfs_csum_final(crc, result);
 417
 418                 if (memcmp(raw_disk_sb, result, csum_size))
 419                         ret = 1;
 420
 421                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 422                         printk(KERN_WARNING
 423                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 424                         ret = 0;
 425                 }
 426         }
 427
 428         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 429                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 430                                 csum_type);
 431                 ret = 1;
 432         }
 433
 434         return ret;
 435 }
 436
 437 /*
 438  * helper to read a given tree block, doing retries as required when
 439  * the checksums don't match and we have alternate mirrors to try.
 440  */
 441 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 442                                           struct extent_buffer *eb,
 443                                           u64 start, u64 parent_transid)
 444 {
 445         struct extent_io_tree *io_tree;
 446         int failed = 0;
 447         int ret;
 448         int num_copies = 0;
 449         int mirror_num = 0;
 450         int failed_mirror = 0;
 451
 452         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 453         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 454         while (1) {
 455                 ret = read_extent_buffer_pages(io_tree, eb, start,
 456                                                WAIT_COMPLETE,
 457                                                btree_get_extent, mirror_num);
 458                 if (!ret) {
 459                         if (!verify_parent_transid(io_tree, eb,
 460                                                    parent_transid, 0))
 461                                 break;
 462                         else
 463                                 ret = -EIO;
 464                 }
 465
 466                 /*
 467                  * This buffer's crc is fine, but its contents are corrupted, so
 468                  * there is no reason to read the other copies, they won't be
 469                  * any less wrong.
 470                  */
 471                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 472                         break;
 473
 474                 num_copies = btrfs_num_copies(root->fs_info,
 475                                               eb->start, eb->len);
 476                 if (num_copies == 1)
 477                         break;
 478
 479                 if (!failed_mirror) {
 480                         failed = 1;
 481                         failed_mirror = eb->read_mirror;
 482                 }
 483
 484                 mirror_num++;
 485                 if (mirror_num == failed_mirror)
 486                         mirror_num++;
 487
 488                 if (mirror_num > num_copies)
 489                         break;
 490         }
 491
 492         if (failed && !ret && failed_mirror)
 493                 repair_eb_io_failure(root, eb, failed_mirror);
 494
 495         return ret;
 496 }
 497
 498 /*
 499  * checksum a dirty tree block before IO.  This has extra checks to make sure
 500  * we only fill in the checksum field in the first page of a multi-page block
 501  */
 502
 503 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 504 {
 505         u64 start = page_offset(page);
 506         u64 found_start;
 507         struct extent_buffer *eb;
 508
 509         eb = (struct extent_buffer *)page->private;
 510         if (page != eb->pages[0])
 511                 return 0;
 512         found_start = btrfs_header_bytenr(eb);
 513         if (WARN_ON(found_start != start || !PageUptodate(page)))
 514                 return 0;
 515         csum_tree_block(root, eb, 0);
 516         return 0;
 517 }
 518
 519 static int check_tree_block_fsid(struct btrfs_root *root,
 520                                  struct extent_buffer *eb)
 521 {
 522         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 523         u8 fsid[BTRFS_UUID_SIZE];
 524         int ret = 1;
 525
 526         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 527         while (fs_devices) {
 528                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 529                         ret = 0;
 530                         break;
 531                 }
 532                 fs_devices = fs_devices->seed;
 533         }
 534         return ret;
 535 }
 536
 537 #define CORRUPT(reason, eb, root, slot)                         \
 538         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 539                    "root=%llu, slot=%d", reason,                        \
 540                btrfs_header_bytenr(eb), root->objectid, slot)
 541
 542 static noinline int check_leaf(struct btrfs_root *root,
 543                                struct extent_buffer *leaf)
 544 {
 545         struct btrfs_key key;
 546         struct btrfs_key leaf_key;
 547         u32 nritems = btrfs_header_nritems(leaf);
 548         int slot;
 549
 550         if (nritems == 0)
 551                 return 0;
 552
 553         /* Check the 0 item */
 554         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 555             BTRFS_LEAF_DATA_SIZE(root)) {
 556                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 557                 return -EIO;
 558         }
 559
 560         /*
 561          * Check to make sure each items keys are in the correct order and their
 562          * offsets make sense.  We only have to loop through nritems-1 because
 563          * we check the current slot against the next slot, which verifies the
 564          * next slot's offset+size makes sense and that the current's slot
 565          * offset is correct.
 566          */
 567         for (slot = 0; slot < nritems - 1; slot++) {
 568                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 569                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 570
 571                 /* Make sure the keys are in the right order */
 572                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 573                         CORRUPT("bad key order", leaf, root, slot);
 574                         return -EIO;
 575                 }
 576
 577                 /*
 578                  * Make sure the offset and ends are right, remember that the
 579                  * item data starts at the end of the leaf and grows towards the
 580                  * front.
 581                  */
 582                 if (btrfs_item_offset_nr(leaf, slot) !=
 583                         btrfs_item_end_nr(leaf, slot + 1)) {
 584                         CORRUPT("slot offset bad", leaf, root, slot);
 585                         return -EIO;
 586                 }
 587
 588                 /*
 589                  * Check to make sure that we don't point outside of the leaf,
 590                  * just incase all the items are consistent to eachother, but
 591                  * all point outside of the leaf.
 592                  */
 593                 if (btrfs_item_end_nr(leaf, slot) >
 594                     BTRFS_LEAF_DATA_SIZE(root)) {
 595                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 596                         return -EIO;
 597                 }
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 604                                       u64 phy_offset, struct page *page,
 605                                       u64 start, u64 end, int mirror)
 606 {
 607         u64 found_start;
 608         int found_level;
 609         struct extent_buffer *eb;
 610         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 611         int ret = 0;
 612         int reads_done;
 613
 614         if (!page->private)
 615                 goto out;
 616
 617         eb = (struct extent_buffer *)page->private;
 618
 619         /* the pending IO might have been the only thing that kept this buffer
 620          * in memory.  Make sure we have a ref for all this other checks
 621          */
 622         extent_buffer_get(eb);
 623
 624         reads_done = atomic_dec_and_test(&eb->io_pages);
 625         if (!reads_done)
 626                 goto err;
 627
 628         eb->read_mirror = mirror;
 629         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633
 634         found_start = btrfs_header_bytenr(eb);
 635         if (found_start != eb->start) {
 636                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad tree block start "
 637                                "%llu %llu\n",
 638                                eb->fs_info->sb->s_id, found_start, eb->start);
 639                 ret = -EIO;
 640                 goto err;
 641         }
 642         if (check_tree_block_fsid(root, eb)) {
 643                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad fsid on block %llu\n",
 644                                eb->fs_info->sb->s_id, eb->start);
 645                 ret = -EIO;
 646                 goto err;
 647         }
 648         found_level = btrfs_header_level(eb);
 649         if (found_level >= BTRFS_MAX_LEVEL) {
 650                 btrfs_info(root->fs_info, "bad tree block level %d",
 651                            (int)btrfs_header_level(eb));
 652                 ret = -EIO;
 653                 goto err;
 654         }
 655
 656         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 657                                        eb, found_level);
 658
 659         ret = csum_tree_block(root, eb, 1);
 660         if (ret) {
 661                 ret = -EIO;
 662                 goto err;
 663         }
 664
 665         /*
 666          * If this is a leaf block and it is corrupt, set the corrupt bit so
 667          * that we don't try and read the other copies of this block, just
 668          * return -EIO.
 669          */
 670         if (found_level == 0 && check_leaf(root, eb)) {
 671                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 672                 ret = -EIO;
 673         }
 674
 675         if (!ret)
 676                 set_extent_buffer_uptodate(eb);
 677 err:
 678         if (reads_done &&
 679             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 680                 btree_readahead_hook(root, eb, eb->start, ret);
 681
 682         if (ret) {
 683                 /*
 684                  * our io error hook is going to dec the io pages
 685                  * again, we have to make sure it has something
 686                  * to decrement
 687                  */
 688                 atomic_inc(&eb->io_pages);
 689                 clear_extent_buffer_uptodate(eb);
 690         }
 691         free_extent_buffer(eb);
 692 out:
 693         return ret;
 694 }
 695
 696 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 697 {
 698         struct extent_buffer *eb;
 699         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 700
 701         eb = (struct extent_buffer *)page->private;
 702         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 703         eb->read_mirror = failed_mirror;
 704         atomic_dec(&eb->io_pages);
 705         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 706                 btree_readahead_hook(root, eb, eb->start, -EIO);
 707         return -EIO;    /* we fixed nothing */
 708 }
 709
 710 static void end_workqueue_bio(struct bio *bio, int err)
 711 {
 712         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 713         struct btrfs_fs_info *fs_info;
 714         struct btrfs_workqueue *wq;
 715         btrfs_work_func_t func;
 716
 717         fs_info = end_io_wq->info;
 718         end_io_wq->error = err;
 719
 720         if (bio->bi_rw & REQ_WRITE) {
 721                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 722                         wq = fs_info->endio_meta_write_workers;
 723                         func = btrfs_endio_meta_write_helper;
 724                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 725                         wq = fs_info->endio_freespace_worker;
 726                         func = btrfs_freespace_write_helper;
 727                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 728                         wq = fs_info->endio_raid56_workers;
 729                         func = btrfs_endio_raid56_helper;
 730                 } else {
 731                         wq = fs_info->endio_write_workers;
 732                         func = btrfs_endio_write_helper;
 733                 }
 734         } else {
 735                 if (unlikely(end_io_wq->metadata ==
 736                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 737                         wq = fs_info->endio_repair_workers;
 738                         func = btrfs_endio_repair_helper;
 739                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 740                         wq = fs_info->endio_raid56_workers;
 741                         func = btrfs_endio_raid56_helper;
 742                 } else if (end_io_wq->metadata) {
 743                         wq = fs_info->endio_meta_workers;
 744                         func = btrfs_endio_meta_helper;
 745                 } else {
 746                         wq = fs_info->endio_workers;
 747                         func = btrfs_endio_helper;
 748                 }
 749         }
 750
 751         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 752         btrfs_queue_work(wq, &end_io_wq->work);
 753 }
 754
 755 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 756                         enum btrfs_wq_endio_type metadata)
 757 {
 758         struct btrfs_end_io_wq *end_io_wq;
 759
 760         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 761         if (!end_io_wq)
 762                 return -ENOMEM;
 763
 764         end_io_wq->private = bio->bi_private;
 765         end_io_wq->end_io = bio->bi_end_io;
 766         end_io_wq->info = info;
 767         end_io_wq->error = 0;
 768         end_io_wq->bio = bio;
 769         end_io_wq->metadata = metadata;
 770
 771         bio->bi_private = end_io_wq;
 772         bio->bi_end_io = end_workqueue_bio;
 773         return 0;
 774 }
 775
 776 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 777 {
 778         unsigned long limit = min_t(unsigned long,
 779                                     info->thread_pool_size,
 780                                     info->fs_devices->open_devices);
 781         return 256 * limit;
 782 }
 783
 784 static void run_one_async_start(struct btrfs_work *work)
 785 {
 786         struct async_submit_bio *async;
 787         int ret;
 788
 789         async = container_of(work, struct  async_submit_bio, work);
 790         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 791                                       async->mirror_num, async->bio_flags,
 792                                       async->bio_offset);
 793         if (ret)
 794                 async->error = ret;
 795 }
 796
 797 static void run_one_async_done(struct btrfs_work *work)
 798 {
 799         struct btrfs_fs_info *fs_info;
 800         struct async_submit_bio *async;
 801         int limit;
 802
 803         async = container_of(work, struct  async_submit_bio, work);
 804         fs_info = BTRFS_I(async->inode)->root->fs_info;
 805
 806         limit = btrfs_async_submit_limit(fs_info);
 807         limit = limit * 2 / 3;
 808
 809         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 810             waitqueue_active(&fs_info->async_submit_wait))
 811                 wake_up(&fs_info->async_submit_wait);
 812
 813         /* If an error occured we just want to clean up the bio and move on */
 814         if (async->error) {
 815                 bio_endio(async->bio, async->error);
 816                 return;
 817         }
 818
 819         async->submit_bio_done(async->inode, async->rw, async->bio,
 820                                async->mirror_num, async->bio_flags,
 821                                async->bio_offset);
 822 }
 823
 824 static void run_one_async_free(struct btrfs_work *work)
 825 {
 826         struct async_submit_bio *async;
 827
 828         async = container_of(work, struct  async_submit_bio, work);
 829         kfree(async);
 830 }
 831
 832 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 833                         int rw, struct bio *bio, int mirror_num,
 834                         unsigned long bio_flags,
 835                         u64 bio_offset,
 836                         extent_submit_bio_hook_t *submit_bio_start,
 837                         extent_submit_bio_hook_t *submit_bio_done)
 838 {
 839         struct async_submit_bio *async;
 840
 841         async = kmalloc(sizeof(*async), GFP_NOFS);
 842         if (!async)
 843                 return -ENOMEM;
 844
 845         async->inode = inode;
 846         async->rw = rw;
 847         async->bio = bio;
 848         async->mirror_num = mirror_num;
 849         async->submit_bio_start = submit_bio_start;
 850         async->submit_bio_done = submit_bio_done;
 851
 852         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 853                         run_one_async_done, run_one_async_free);
 854
 855         async->bio_flags = bio_flags;
 856         async->bio_offset = bio_offset;
 857
 858         async->error = 0;
 859
 860         atomic_inc(&fs_info->nr_async_submits);
 861
 862         if (rw & REQ_SYNC)
 863                 btrfs_set_work_high_priority(&async->work);
 864
 865         btrfs_queue_work(fs_info->workers, &async->work);
 866
 867         while (atomic_read(&fs_info->async_submit_draining) &&
 868               atomic_read(&fs_info->nr_async_submits)) {
 869                 wait_event(fs_info->async_submit_wait,
 870                            (atomic_read(&fs_info->nr_async_submits) == 0));
 871         }
 872
 873         return 0;
 874 }
 875
 876 static int btree_csum_one_bio(struct bio *bio)
 877 {
 878         struct bio_vec *bvec;
 879         struct btrfs_root *root;
 880         int i, ret = 0;
 881
 882         bio_for_each_segment_all(bvec, bio, i) {
 883                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 884                 ret = csum_dirty_buffer(root, bvec->bv_page);
 885                 if (ret)
 886                         break;
 887         }
 888
 889         return ret;
 890 }
 891
 892 static int __btree_submit_bio_start(struct inode *inode, int rw,
 893                                     struct bio *bio, int mirror_num,
 894                                     unsigned long bio_flags,
 895                                     u64 bio_offset)
 896 {
 897         /*
 898          * when we're called for a write, we're already in the async
 899          * submission context.  Just jump into btrfs_map_bio
 900          */
 901         return btree_csum_one_bio(bio);
 902 }
 903
 904 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 905                                  int mirror_num, unsigned long bio_flags,
 906                                  u64 bio_offset)
 907 {
 908         int ret;
 909
 910         /*
 911          * when we're called for a write, we're already in the async
 912          * submission context.  Just jump into btrfs_map_bio
 913          */
 914         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 915         if (ret)
 916                 bio_endio(bio, ret);
 917         return ret;
 918 }
 919
 920 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 921 {
 922         if (bio_flags & EXTENT_BIO_TREE_LOG)
 923                 return 0;
 924 #ifdef CONFIG_X86
 925         if (cpu_has_xmm4_2)
 926                 return 0;
 927 #endif
 928         return 1;
 929 }
 930
 931 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 932                                  int mirror_num, unsigned long bio_flags,
 933                                  u64 bio_offset)
 934 {
 935         int async = check_async_write(inode, bio_flags);
 936         int ret;
 937
 938         if (!(rw & REQ_WRITE)) {
 939                 /*
 940                  * called for a read, do the setup so that checksum validation
 941                  * can happen in the async kernel threads
 942                  */
 943                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 944                                           bio, BTRFS_WQ_ENDIO_METADATA);
 945                 if (ret)
 946                         goto out_w_error;
 947                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 948                                     mirror_num, 0);
 949         } else if (!async) {
 950                 ret = btree_csum_one_bio(bio);
 951                 if (ret)
 952                         goto out_w_error;
 953                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 954                                     mirror_num, 0);
 955         } else {
 956                 /*
 957                  * kthread helpers are used to submit writes so that
 958                  * checksumming can happen in parallel across all CPUs
 959                  */
 960                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 961                                           inode, rw, bio, mirror_num, 0,
 962                                           bio_offset,
 963                                           __btree_submit_bio_start,
 964                                           __btree_submit_bio_done);
 965         }
 966
 967         if (ret) {
 968 out_w_error:
 969                 bio_endio(bio, ret);
 970         }
 971         return ret;
 972 }
 973
 974 #ifdef CONFIG_MIGRATION
 975 static int btree_migratepage(struct address_space *mapping,
 976                         struct page *newpage, struct page *page,
 977                         enum migrate_mode mode)
 978 {
 979         /*
 980          * we can't safely write a btree page from here,
 981          * we haven't done the locking hook
 982          */
 983         if (PageDirty(page))
 984                 return -EAGAIN;
 985         /*
 986          * Buffers may be managed in a filesystem specific way.
 987          * We must have no buffers or drop them.
 988          */
 989         if (page_has_private(page) &&
 990             !try_to_release_page(page, GFP_KERNEL))
 991                 return -EAGAIN;
 992         return migrate_page(mapping, newpage, page, mode);
 993 }
 994 #endif
 995
 996
 997 static int btree_writepages(struct address_space *mapping,
 998                             struct writeback_control *wbc)
 999 {
1000         struct btrfs_fs_info *fs_info;
1001         int ret;
1002
1003         if (wbc->sync_mode == WB_SYNC_NONE) {
1004
1005                 if (wbc->for_kupdate)
1006                         return 0;
1007
1008                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1009                 /* this is a bit racy, but that's ok */
1010                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1011                                              BTRFS_DIRTY_METADATA_THRESH);
1012                 if (ret < 0)
1013                         return 0;
1014         }
1015         return btree_write_cache_pages(mapping, wbc);
1016 }
1017
1018 static int btree_readpage(struct file *file, struct page *page)
1019 {
1020         struct extent_io_tree *tree;
1021         tree = &BTRFS_I(page->mapping->host)->io_tree;
1022         return extent_read_full_page(tree, page, btree_get_extent, 0);
1023 }
1024
1025 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1026 {
1027         if (PageWriteback(page) || PageDirty(page))
1028                 return 0;
1029
1030         return try_release_extent_buffer(page);
1031 }
1032
1033 static void btree_invalidatepage(struct page *page, unsigned int offset,
1034                                  unsigned int length)
1035 {
1036         struct extent_io_tree *tree;
1037         tree = &BTRFS_I(page->mapping->host)->io_tree;
1038         extent_invalidatepage(tree, page, offset);
1039         btree_releasepage(page, GFP_NOFS);
1040         if (PagePrivate(page)) {
1041                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1042                            "page private not zero on page %llu",
1043                            (unsigned long long)page_offset(page));
1044                 ClearPagePrivate(page);
1045                 set_page_private(page, 0);
1046                 page_cache_release(page);
1047         }
1048 }
1049
1050 static int btree_set_page_dirty(struct page *page)
1051 {
1052 #ifdef DEBUG
1053         struct extent_buffer *eb;
1054
1055         BUG_ON(!PagePrivate(page));
1056         eb = (struct extent_buffer *)page->private;
1057         BUG_ON(!eb);
1058         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1059         BUG_ON(!atomic_read(&eb->refs));
1060         btrfs_assert_tree_locked(eb);
1061 #endif
1062         return __set_page_dirty_nobuffers(page);
1063 }
1064
1065 static const struct address_space_operations btree_aops = {
1066         .readpage       = btree_readpage,
1067         .writepages     = btree_writepages,
1068         .releasepage    = btree_releasepage,
1069         .invalidatepage = btree_invalidatepage,
1070 #ifdef CONFIG_MIGRATION
1071         .migratepage    = btree_migratepage,
1072 #endif
1073         .set_page_dirty = btree_set_page_dirty,
1074 };
1075
1076 void readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
1077 {
1078         struct extent_buffer *buf = NULL;
1079         struct inode *btree_inode = root->fs_info->btree_inode;
1080
1081         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082         if (!buf)
1083                 return;
1084         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1085                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1086         free_extent_buffer(buf);
1087 }
1088
1089 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1090                          int mirror_num, struct extent_buffer **eb)
1091 {
1092         struct extent_buffer *buf = NULL;
1093         struct inode *btree_inode = root->fs_info->btree_inode;
1094         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1095         int ret;
1096
1097         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1098         if (!buf)
1099                 return 0;
1100
1101         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1102
1103         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1104                                        btree_get_extent, mirror_num);
1105         if (ret) {
1106                 free_extent_buffer(buf);
1107                 return ret;
1108         }
1109
1110         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1111                 free_extent_buffer(buf);
1112                 return -EIO;
1113         } else if (extent_buffer_uptodate(buf)) {
1114                 *eb = buf;
1115         } else {
1116                 free_extent_buffer(buf);
1117         }
1118         return 0;
1119 }
1120
1121 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1122                                             u64 bytenr)
1123 {
1124         return find_extent_buffer(root->fs_info, bytenr);
1125 }
1126
1127 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1128                                                  u64 bytenr, u32 blocksize)
1129 {
1130         if (btrfs_test_is_dummy_root(root))
1131                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1132                                                 blocksize);
1133         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1134 }
1135
1136
1137 int btrfs_write_tree_block(struct extent_buffer *buf)
1138 {
1139         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1140                                         buf->start + buf->len - 1);
1141 }
1142
1143 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1144 {
1145         return filemap_fdatawait_range(buf->pages[0]->mapping,
1146                                        buf->start, buf->start + buf->len - 1);
1147 }
1148
1149 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1150                                       u64 parent_transid)
1151 {
1152         struct extent_buffer *buf = NULL;
1153         int ret;
1154
1155         buf = btrfs_find_create_tree_block(root, bytenr, root->nodesize);
1156         if (!buf)
1157                 return NULL;
1158
1159         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1160         if (ret) {
1161                 free_extent_buffer(buf);
1162                 return NULL;
1163         }
1164         return buf;
1165
1166 }
1167
1168 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1169                       struct extent_buffer *buf)
1170 {
1171         struct btrfs_fs_info *fs_info = root->fs_info;
1172
1173         if (btrfs_header_generation(buf) ==
1174             fs_info->running_transaction->transid) {
1175                 btrfs_assert_tree_locked(buf);
1176
1177                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1178                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1179                                              -buf->len,
1180                                              fs_info->dirty_metadata_batch);
1181                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182                         btrfs_set_lock_blocking(buf);
1183                         clear_extent_buffer_dirty(buf);
1184                 }
1185         }
1186 }
1187
1188 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1189 {
1190         struct btrfs_subvolume_writers *writers;
1191         int ret;
1192
1193         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1194         if (!writers)
1195                 return ERR_PTR(-ENOMEM);
1196
1197         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1198         if (ret < 0) {
1199                 kfree(writers);
1200                 return ERR_PTR(ret);
1201         }
1202
1203         init_waitqueue_head(&writers->wait);
1204         return writers;
1205 }
1206
1207 static void
1208 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1209 {
1210         percpu_counter_destroy(&writers->counter);
1211         kfree(writers);
1212 }
1213
1214 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1215                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1216                          u64 objectid)
1217 {
1218         root->node = NULL;
1219         root->commit_root = NULL;
1220         root->sectorsize = sectorsize;
1221         root->nodesize = nodesize;
1222         root->stripesize = stripesize;
1223         root->state = 0;
1224         root->orphan_cleanup_state = 0;
1225
1226         root->objectid = objectid;
1227         root->last_trans = 0;
1228         root->highest_objectid = 0;
1229         root->nr_delalloc_inodes = 0;
1230         root->nr_ordered_extents = 0;
1231         root->name = NULL;
1232         root->inode_tree = RB_ROOT;
1233         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1234         root->block_rsv = NULL;
1235         root->orphan_block_rsv = NULL;
1236
1237         INIT_LIST_HEAD(&root->dirty_list);
1238         INIT_LIST_HEAD(&root->root_list);
1239         INIT_LIST_HEAD(&root->delalloc_inodes);
1240         INIT_LIST_HEAD(&root->delalloc_root);
1241         INIT_LIST_HEAD(&root->ordered_extents);
1242         INIT_LIST_HEAD(&root->ordered_root);
1243         INIT_LIST_HEAD(&root->logged_list[0]);
1244         INIT_LIST_HEAD(&root->logged_list[1]);
1245         spin_lock_init(&root->orphan_lock);
1246         spin_lock_init(&root->inode_lock);
1247         spin_lock_init(&root->delalloc_lock);
1248         spin_lock_init(&root->ordered_extent_lock);
1249         spin_lock_init(&root->accounting_lock);
1250         spin_lock_init(&root->log_extents_lock[0]);
1251         spin_lock_init(&root->log_extents_lock[1]);
1252         mutex_init(&root->objectid_mutex);
1253         mutex_init(&root->log_mutex);
1254         mutex_init(&root->ordered_extent_mutex);
1255         mutex_init(&root->delalloc_mutex);
1256         init_waitqueue_head(&root->log_writer_wait);
1257         init_waitqueue_head(&root->log_commit_wait[0]);
1258         init_waitqueue_head(&root->log_commit_wait[1]);
1259         INIT_LIST_HEAD(&root->log_ctxs[0]);
1260         INIT_LIST_HEAD(&root->log_ctxs[1]);
1261         atomic_set(&root->log_commit[0], 0);
1262         atomic_set(&root->log_commit[1], 0);
1263         atomic_set(&root->log_writers, 0);
1264         atomic_set(&root->log_batch, 0);
1265         atomic_set(&root->orphan_inodes, 0);
1266         atomic_set(&root->refs, 1);
1267         atomic_set(&root->will_be_snapshoted, 0);
1268         root->log_transid = 0;
1269         root->log_transid_committed = -1;
1270         root->last_log_commit = 0;
1271         if (fs_info)
1272                 extent_io_tree_init(&root->dirty_log_pages,
1273                                      fs_info->btree_inode->i_mapping);
1274
1275         memset(&root->root_key, 0, sizeof(root->root_key));
1276         memset(&root->root_item, 0, sizeof(root->root_item));
1277         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1278         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1279         if (fs_info)
1280                 root->defrag_trans_start = fs_info->generation;
1281         else
1282                 root->defrag_trans_start = 0;
1283         init_completion(&root->kobj_unregister);
1284         root->root_key.objectid = objectid;
1285         root->anon_dev = 0;
1286
1287         spin_lock_init(&root->root_item_lock);
1288 }
1289
1290 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1291 {
1292         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1293         if (root)
1294                 root->fs_info = fs_info;
1295         return root;
1296 }
1297
1298 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1299 /* Should only be used by the testing infrastructure */
1300 struct btrfs_root *btrfs_alloc_dummy_root(void)
1301 {
1302         struct btrfs_root *root;
1303
1304         root = btrfs_alloc_root(NULL);
1305         if (!root)
1306                 return ERR_PTR(-ENOMEM);
1307         __setup_root(4096, 4096, 4096, root, NULL, 1);
1308         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1309         root->alloc_bytenr = 0;
1310
1311         return root;
1312 }
1313 #endif
1314
1315 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1316                                      struct btrfs_fs_info *fs_info,
1317                                      u64 objectid)
1318 {
1319         struct extent_buffer *leaf;
1320         struct btrfs_root *tree_root = fs_info->tree_root;
1321         struct btrfs_root *root;
1322         struct btrfs_key key;
1323         int ret = 0;
1324         uuid_le uuid;
1325
1326         root = btrfs_alloc_root(fs_info);
1327         if (!root)
1328                 return ERR_PTR(-ENOMEM);
1329
1330         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1331                 tree_root->stripesize, root, fs_info, objectid);
1332         root->root_key.objectid = objectid;
1333         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1334         root->root_key.offset = 0;
1335
1336         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1337         if (IS_ERR(leaf)) {
1338                 ret = PTR_ERR(leaf);
1339                 leaf = NULL;
1340                 goto fail;
1341         }
1342
1343         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1344         btrfs_set_header_bytenr(leaf, leaf->start);
1345         btrfs_set_header_generation(leaf, trans->transid);
1346         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1347         btrfs_set_header_owner(leaf, objectid);
1348         root->node = leaf;
1349
1350         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1351                             BTRFS_FSID_SIZE);
1352         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1353                             btrfs_header_chunk_tree_uuid(leaf),
1354                             BTRFS_UUID_SIZE);
1355         btrfs_mark_buffer_dirty(leaf);
1356
1357         root->commit_root = btrfs_root_node(root);
1358         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1359
1360         root->root_item.flags = 0;
1361         root->root_item.byte_limit = 0;
1362         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1363         btrfs_set_root_generation(&root->root_item, trans->transid);
1364         btrfs_set_root_level(&root->root_item, 0);
1365         btrfs_set_root_refs(&root->root_item, 1);
1366         btrfs_set_root_used(&root->root_item, leaf->len);
1367         btrfs_set_root_last_snapshot(&root->root_item, 0);
1368         btrfs_set_root_dirid(&root->root_item, 0);
1369         uuid_le_gen(&uuid);
1370         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1371         root->root_item.drop_level = 0;
1372
1373         key.objectid = objectid;
1374         key.type = BTRFS_ROOT_ITEM_KEY;
1375         key.offset = 0;
1376         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1377         if (ret)
1378                 goto fail;
1379
1380         btrfs_tree_unlock(leaf);
1381
1382         return root;
1383
1384 fail:
1385         if (leaf) {
1386                 btrfs_tree_unlock(leaf);
1387                 free_extent_buffer(root->commit_root);
1388                 free_extent_buffer(leaf);
1389         }
1390         kfree(root);
1391
1392         return ERR_PTR(ret);
1393 }
1394
1395 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1396                                          struct btrfs_fs_info *fs_info)
1397 {
1398         struct btrfs_root *root;
1399         struct btrfs_root *tree_root = fs_info->tree_root;
1400         struct extent_buffer *leaf;
1401
1402         root = btrfs_alloc_root(fs_info);
1403         if (!root)
1404                 return ERR_PTR(-ENOMEM);
1405
1406         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1407                      tree_root->stripesize, root, fs_info,
1408                      BTRFS_TREE_LOG_OBJECTID);
1409
1410         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1411         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1412         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1413
1414         /*
1415          * DON'T set REF_COWS for log trees
1416          *
1417          * log trees do not get reference counted because they go away
1418          * before a real commit is actually done.  They do store pointers
1419          * to file data extents, and those reference counts still get
1420          * updated (along with back refs to the log tree).
1421          */
1422
1423         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1424                         NULL, 0, 0, 0);
1425         if (IS_ERR(leaf)) {
1426                 kfree(root);
1427                 return ERR_CAST(leaf);
1428         }
1429
1430         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1431         btrfs_set_header_bytenr(leaf, leaf->start);
1432         btrfs_set_header_generation(leaf, trans->transid);
1433         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1434         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1435         root->node = leaf;
1436
1437         write_extent_buffer(root->node, root->fs_info->fsid,
1438                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1439         btrfs_mark_buffer_dirty(root->node);
1440         btrfs_tree_unlock(root->node);
1441         return root;
1442 }
1443
1444 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1445                              struct btrfs_fs_info *fs_info)
1446 {
1447         struct btrfs_root *log_root;
1448
1449         log_root = alloc_log_tree(trans, fs_info);
1450         if (IS_ERR(log_root))
1451                 return PTR_ERR(log_root);
1452         WARN_ON(fs_info->log_root_tree);
1453         fs_info->log_root_tree = log_root;
1454         return 0;
1455 }
1456
1457 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1458                        struct btrfs_root *root)
1459 {
1460         struct btrfs_root *log_root;
1461         struct btrfs_inode_item *inode_item;
1462
1463         log_root = alloc_log_tree(trans, root->fs_info);
1464         if (IS_ERR(log_root))
1465                 return PTR_ERR(log_root);
1466
1467         log_root->last_trans = trans->transid;
1468         log_root->root_key.offset = root->root_key.objectid;
1469
1470         inode_item = &log_root->root_item.inode;
1471         btrfs_set_stack_inode_generation(inode_item, 1);
1472         btrfs_set_stack_inode_size(inode_item, 3);
1473         btrfs_set_stack_inode_nlink(inode_item, 1);
1474         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1475         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1476
1477         btrfs_set_root_node(&log_root->root_item, log_root->node);
1478
1479         WARN_ON(root->log_root);
1480         root->log_root = log_root;
1481         root->log_transid = 0;
1482         root->log_transid_committed = -1;
1483         root->last_log_commit = 0;
1484         return 0;
1485 }
1486
1487 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1488                                                struct btrfs_key *key)
1489 {
1490         struct btrfs_root *root;
1491         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1492         struct btrfs_path *path;
1493         u64 generation;
1494         int ret;
1495
1496         path = btrfs_alloc_path();
1497         if (!path)
1498                 return ERR_PTR(-ENOMEM);
1499
1500         root = btrfs_alloc_root(fs_info);
1501         if (!root) {
1502                 ret = -ENOMEM;
1503                 goto alloc_fail;
1504         }
1505
1506         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1507                 tree_root->stripesize, root, fs_info, key->objectid);
1508
1509         ret = btrfs_find_root(tree_root, key, path,
1510                               &root->root_item, &root->root_key);
1511         if (ret) {
1512                 if (ret > 0)
1513                         ret = -ENOENT;
1514                 goto find_fail;
1515         }
1516
1517         generation = btrfs_root_generation(&root->root_item);
1518         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1519                                      generation);
1520         if (!root->node) {
1521                 ret = -ENOMEM;
1522                 goto find_fail;
1523         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1524                 ret = -EIO;
1525                 goto read_fail;
1526         }
1527         root->commit_root = btrfs_root_node(root);
1528 out:
1529         btrfs_free_path(path);
1530         return root;
1531
1532 read_fail:
1533         free_extent_buffer(root->node);
1534 find_fail:
1535         kfree(root);
1536 alloc_fail:
1537         root = ERR_PTR(ret);
1538         goto out;
1539 }
1540
1541 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1542                                       struct btrfs_key *location)
1543 {
1544         struct btrfs_root *root;
1545
1546         root = btrfs_read_tree_root(tree_root, location);
1547         if (IS_ERR(root))
1548                 return root;
1549
1550         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1551                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1552                 btrfs_check_and_init_root_item(&root->root_item);
1553         }
1554
1555         return root;
1556 }
1557
1558 int btrfs_init_fs_root(struct btrfs_root *root)
1559 {
1560         int ret;
1561         struct btrfs_subvolume_writers *writers;
1562
1563         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1564         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1565                                         GFP_NOFS);
1566         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1567                 ret = -ENOMEM;
1568                 goto fail;
1569         }
1570
1571         writers = btrfs_alloc_subvolume_writers();
1572         if (IS_ERR(writers)) {
1573                 ret = PTR_ERR(writers);
1574                 goto fail;
1575         }
1576         root->subv_writers = writers;
1577
1578         btrfs_init_free_ino_ctl(root);
1579         spin_lock_init(&root->ino_cache_lock);
1580         init_waitqueue_head(&root->ino_cache_wait);
1581
1582         ret = get_anon_bdev(&root->anon_dev);
1583         if (ret)
1584                 goto free_writers;
1585         return 0;
1586
1587 free_writers:
1588         btrfs_free_subvolume_writers(root->subv_writers);
1589 fail:
1590         kfree(root->free_ino_ctl);
1591         kfree(root->free_ino_pinned);
1592         return ret;
1593 }
1594
1595 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1596                                                u64 root_id)
1597 {
1598         struct btrfs_root *root;
1599
1600         spin_lock(&fs_info->fs_roots_radix_lock);
1601         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1602                                  (unsigned long)root_id);
1603         spin_unlock(&fs_info->fs_roots_radix_lock);
1604         return root;
1605 }
1606
1607 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1608                          struct btrfs_root *root)
1609 {
1610         int ret;
1611
1612         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1613         if (ret)
1614                 return ret;
1615
1616         spin_lock(&fs_info->fs_roots_radix_lock);
1617         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1618                                 (unsigned long)root->root_key.objectid,
1619                                 root);
1620         if (ret == 0)
1621                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1622         spin_unlock(&fs_info->fs_roots_radix_lock);
1623         radix_tree_preload_end();
1624
1625         return ret;
1626 }
1627
1628 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1629                                      struct btrfs_key *location,
1630                                      bool check_ref)
1631 {
1632         struct btrfs_root *root;
1633         struct btrfs_path *path;
1634         int ret;
1635
1636         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1637                 return fs_info->tree_root;
1638         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1639                 return fs_info->extent_root;
1640         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1641                 return fs_info->chunk_root;
1642         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1643                 return fs_info->dev_root;
1644         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1645                 return fs_info->csum_root;
1646         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1647                 return fs_info->quota_root ? fs_info->quota_root :
1648                                              ERR_PTR(-ENOENT);
1649         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1650                 return fs_info->uuid_root ? fs_info->uuid_root :
1651                                             ERR_PTR(-ENOENT);
1652 again:
1653         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1654         if (root) {
1655                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1656                         return ERR_PTR(-ENOENT);
1657                 return root;
1658         }
1659
1660         root = btrfs_read_fs_root(fs_info->tree_root, location);
1661         if (IS_ERR(root))
1662                 return root;
1663
1664         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1665                 ret = -ENOENT;
1666                 goto fail;
1667         }
1668
1669         ret = btrfs_init_fs_root(root);
1670         if (ret)
1671                 goto fail;
1672
1673         path = btrfs_alloc_path();
1674         if (!path) {
1675                 ret = -ENOMEM;
1676                 goto fail;
1677         }
1678         ret = btrfs_find_item(fs_info->tree_root, path, BTRFS_ORPHAN_OBJECTID,
1679                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1680         btrfs_free_path(path);
1681         if (ret < 0)
1682                 goto fail;
1683         if (ret == 0)
1684                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1685
1686         ret = btrfs_insert_fs_root(fs_info, root);
1687         if (ret) {
1688                 if (ret == -EEXIST) {
1689                         free_fs_root(root);
1690                         goto again;
1691                 }
1692                 goto fail;
1693         }
1694         return root;
1695 fail:
1696         free_fs_root(root);
1697         return ERR_PTR(ret);
1698 }
1699
1700 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1701 {
1702         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1703         int ret = 0;
1704         struct btrfs_device *device;
1705         struct backing_dev_info *bdi;
1706
1707         rcu_read_lock();
1708         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1709                 if (!device->bdev)
1710                         continue;
1711                 bdi = blk_get_backing_dev_info(device->bdev);
1712                 if (bdi_congested(bdi, bdi_bits)) {
1713                         ret = 1;
1714                         break;
1715                 }
1716         }
1717         rcu_read_unlock();
1718         return ret;
1719 }
1720
1721 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1722 {
1723         int err;
1724
1725         bdi->capabilities = BDI_CAP_MAP_COPY;
1726         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1727         if (err)
1728                 return err;
1729
1730         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1731         bdi->congested_fn       = btrfs_congested_fn;
1732         bdi->congested_data     = info;
1733         return 0;
1734 }
1735
1736 /*
1737  * called by the kthread helper functions to finally call the bio end_io
1738  * functions.  This is where read checksum verification actually happens
1739  */
1740 static void end_workqueue_fn(struct btrfs_work *work)
1741 {
1742         struct bio *bio;
1743         struct btrfs_end_io_wq *end_io_wq;
1744         int error;
1745
1746         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1747         bio = end_io_wq->bio;
1748
1749         error = end_io_wq->error;
1750         bio->bi_private = end_io_wq->private;
1751         bio->bi_end_io = end_io_wq->end_io;
1752         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1753         bio_endio_nodec(bio, error);
1754 }
1755
1756 static int cleaner_kthread(void *arg)
1757 {
1758         struct btrfs_root *root = arg;
1759         int again;
1760
1761         do {
1762                 again = 0;
1763
1764                 /* Make the cleaner go to sleep early. */
1765                 if (btrfs_need_cleaner_sleep(root))
1766                         goto sleep;
1767
1768                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1769                         goto sleep;
1770
1771                 /*
1772                  * Avoid the problem that we change the status of the fs
1773                  * during the above check and trylock.
1774                  */
1775                 if (btrfs_need_cleaner_sleep(root)) {
1776                         mutex_unlock(&root->fs_info->cleaner_mutex);
1777                         goto sleep;
1778                 }
1779
1780                 btrfs_run_delayed_iputs(root);
1781                 btrfs_delete_unused_bgs(root->fs_info);
1782                 again = btrfs_clean_one_deleted_snapshot(root);
1783                 mutex_unlock(&root->fs_info->cleaner_mutex);
1784
1785                 /*
1786                  * The defragger has dealt with the R/O remount and umount,
1787                  * needn't do anything special here.
1788                  */
1789                 btrfs_run_defrag_inodes(root->fs_info);
1790 sleep:
1791                 if (!try_to_freeze() && !again) {
1792                         set_current_state(TASK_INTERRUPTIBLE);
1793                         if (!kthread_should_stop())
1794                                 schedule();
1795                         __set_current_state(TASK_RUNNING);
1796                 }
1797         } while (!kthread_should_stop());
1798         return 0;
1799 }
1800
1801 static int transaction_kthread(void *arg)
1802 {
1803         struct btrfs_root *root = arg;
1804         struct btrfs_trans_handle *trans;
1805         struct btrfs_transaction *cur;
1806         u64 transid;
1807         unsigned long now;
1808         unsigned long delay;
1809         bool cannot_commit;
1810
1811         do {
1812                 cannot_commit = false;
1813                 delay = HZ * root->fs_info->commit_interval;
1814                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1815
1816                 spin_lock(&root->fs_info->trans_lock);
1817                 cur = root->fs_info->running_transaction;
1818                 if (!cur) {
1819                         spin_unlock(&root->fs_info->trans_lock);
1820                         goto sleep;
1821                 }
1822
1823                 now = get_seconds();
1824                 if (cur->state < TRANS_STATE_BLOCKED &&
1825                     (now < cur->start_time ||
1826                      now - cur->start_time < root->fs_info->commit_interval)) {
1827                         spin_unlock(&root->fs_info->trans_lock);
1828                         delay = HZ * 5;
1829                         goto sleep;
1830                 }
1831                 transid = cur->transid;
1832                 spin_unlock(&root->fs_info->trans_lock);
1833
1834                 /* If the file system is aborted, this will always fail. */
1835                 trans = btrfs_attach_transaction(root);
1836                 if (IS_ERR(trans)) {
1837                         if (PTR_ERR(trans) != -ENOENT)
1838                                 cannot_commit = true;
1839                         goto sleep;
1840                 }
1841                 if (transid == trans->transid) {
1842                         btrfs_commit_transaction(trans, root);
1843                 } else {
1844                         btrfs_end_transaction(trans, root);
1845                 }
1846 sleep:
1847                 wake_up_process(root->fs_info->cleaner_kthread);
1848                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1849
1850                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1851                                       &root->fs_info->fs_state)))
1852                         btrfs_cleanup_transaction(root);
1853                 if (!try_to_freeze()) {
1854                         set_current_state(TASK_INTERRUPTIBLE);
1855                         if (!kthread_should_stop() &&
1856                             (!btrfs_transaction_blocked(root->fs_info) ||
1857                              cannot_commit))
1858                                 schedule_timeout(delay);
1859                         __set_current_state(TASK_RUNNING);
1860                 }
1861         } while (!kthread_should_stop());
1862         return 0;
1863 }
1864
1865 /*
1866  * this will find the highest generation in the array of
1867  * root backups.  The index of the highest array is returned,
1868  * or -1 if we can't find anything.
1869  *
1870  * We check to make sure the array is valid by comparing the
1871  * generation of the latest  root in the array with the generation
1872  * in the super block.  If they don't match we pitch it.
1873  */
1874 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1875 {
1876         u64 cur;
1877         int newest_index = -1;
1878         struct btrfs_root_backup *root_backup;
1879         int i;
1880
1881         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1882                 root_backup = info->super_copy->super_roots + i;
1883                 cur = btrfs_backup_tree_root_gen(root_backup);
1884                 if (cur == newest_gen)
1885                         newest_index = i;
1886         }
1887
1888         /* check to see if we actually wrapped around */
1889         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1890                 root_backup = info->super_copy->super_roots;
1891                 cur = btrfs_backup_tree_root_gen(root_backup);
1892                 if (cur == newest_gen)
1893                         newest_index = 0;
1894         }
1895         return newest_index;
1896 }
1897
1898
1899 /*
1900  * find the oldest backup so we know where to store new entries
1901  * in the backup array.  This will set the backup_root_index
1902  * field in the fs_info struct
1903  */
1904 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1905                                      u64 newest_gen)
1906 {
1907         int newest_index = -1;
1908
1909         newest_index = find_newest_super_backup(info, newest_gen);
1910         /* if there was garbage in there, just move along */
1911         if (newest_index == -1) {
1912                 info->backup_root_index = 0;
1913         } else {
1914                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1915         }
1916 }
1917
1918 /*
1919  * copy all the root pointers into the super backup array.
1920  * this will bump the backup pointer by one when it is
1921  * done
1922  */
1923 static void backup_super_roots(struct btrfs_fs_info *info)
1924 {
1925         int next_backup;
1926         struct btrfs_root_backup *root_backup;
1927         int last_backup;
1928
1929         next_backup = info->backup_root_index;
1930         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1931                 BTRFS_NUM_BACKUP_ROOTS;
1932
1933         /*
1934          * just overwrite the last backup if we're at the same generation
1935          * this happens only at umount
1936          */
1937         root_backup = info->super_for_commit->super_roots + last_backup;
1938         if (btrfs_backup_tree_root_gen(root_backup) ==
1939             btrfs_header_generation(info->tree_root->node))
1940                 next_backup = last_backup;
1941
1942         root_backup = info->super_for_commit->super_roots + next_backup;
1943
1944         /*
1945          * make sure all of our padding and empty slots get zero filled
1946          * regardless of which ones we use today
1947          */
1948         memset(root_backup, 0, sizeof(*root_backup));
1949
1950         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1951
1952         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1953         btrfs_set_backup_tree_root_gen(root_backup,
1954                                btrfs_header_generation(info->tree_root->node));
1955
1956         btrfs_set_backup_tree_root_level(root_backup,
1957                                btrfs_header_level(info->tree_root->node));
1958
1959         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1960         btrfs_set_backup_chunk_root_gen(root_backup,
1961                                btrfs_header_generation(info->chunk_root->node));
1962         btrfs_set_backup_chunk_root_level(root_backup,
1963                                btrfs_header_level(info->chunk_root->node));
1964
1965         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1966         btrfs_set_backup_extent_root_gen(root_backup,
1967                                btrfs_header_generation(info->extent_root->node));
1968         btrfs_set_backup_extent_root_level(root_backup,
1969                                btrfs_header_level(info->extent_root->node));
1970
1971         /*
1972          * we might commit during log recovery, which happens before we set
1973          * the fs_root.  Make sure it is valid before we fill it in.
1974          */
1975         if (info->fs_root && info->fs_root->node) {
1976                 btrfs_set_backup_fs_root(root_backup,
1977                                          info->fs_root->node->start);
1978                 btrfs_set_backup_fs_root_gen(root_backup,
1979                                btrfs_header_generation(info->fs_root->node));
1980                 btrfs_set_backup_fs_root_level(root_backup,
1981                                btrfs_header_level(info->fs_root->node));
1982         }
1983
1984         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1985         btrfs_set_backup_dev_root_gen(root_backup,
1986                                btrfs_header_generation(info->dev_root->node));
1987         btrfs_set_backup_dev_root_level(root_backup,
1988                                        btrfs_header_level(info->dev_root->node));
1989
1990         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1991         btrfs_set_backup_csum_root_gen(root_backup,
1992                                btrfs_header_generation(info->csum_root->node));
1993         btrfs_set_backup_csum_root_level(root_backup,
1994                                btrfs_header_level(info->csum_root->node));
1995
1996         btrfs_set_backup_total_bytes(root_backup,
1997                              btrfs_super_total_bytes(info->super_copy));
1998         btrfs_set_backup_bytes_used(root_backup,
1999                              btrfs_super_bytes_used(info->super_copy));
2000         btrfs_set_backup_num_devices(root_backup,
2001                              btrfs_super_num_devices(info->super_copy));
2002
2003         /*
2004          * if we don't copy this out to the super_copy, it won't get remembered
2005          * for the next commit
2006          */
2007         memcpy(&info->super_copy->super_roots,
2008                &info->super_for_commit->super_roots,
2009                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2010 }
2011
2012 /*
2013  * this copies info out of the root backup array and back into
2014  * the in-memory super block.  It is meant to help iterate through
2015  * the array, so you send it the number of backups you've already
2016  * tried and the last backup index you used.
2017  *
2018  * this returns -1 when it has tried all the backups
2019  */
2020 static noinline int next_root_backup(struct btrfs_fs_info *info,
2021                                      struct btrfs_super_block *super,
2022                                      int *num_backups_tried, int *backup_index)
2023 {
2024         struct btrfs_root_backup *root_backup;
2025         int newest = *backup_index;
2026
2027         if (*num_backups_tried == 0) {
2028                 u64 gen = btrfs_super_generation(super);
2029
2030                 newest = find_newest_super_backup(info, gen);
2031                 if (newest == -1)
2032                         return -1;
2033
2034                 *backup_index = newest;
2035                 *num_backups_tried = 1;
2036         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2037                 /* we've tried all the backups, all done */
2038                 return -1;
2039         } else {
2040                 /* jump to the next oldest backup */
2041                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2042                         BTRFS_NUM_BACKUP_ROOTS;
2043                 *backup_index = newest;
2044                 *num_backups_tried += 1;
2045         }
2046         root_backup = super->super_roots + newest;
2047
2048         btrfs_set_super_generation(super,
2049                                    btrfs_backup_tree_root_gen(root_backup));
2050         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2051         btrfs_set_super_root_level(super,
2052                                    btrfs_backup_tree_root_level(root_backup));
2053         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2054
2055         /*
2056          * fixme: the total bytes and num_devices need to match or we should
2057          * need a fsck
2058          */
2059         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2060         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2061         return 0;
2062 }
2063
2064 /* helper to cleanup workers */
2065 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2066 {
2067         btrfs_destroy_workqueue(fs_info->fixup_workers);
2068         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2069         btrfs_destroy_workqueue(fs_info->workers);
2070         btrfs_destroy_workqueue(fs_info->endio_workers);
2071         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2072         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2074         btrfs_destroy_workqueue(fs_info->rmw_workers);
2075         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2076         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2077         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2078         btrfs_destroy_workqueue(fs_info->submit_workers);
2079         btrfs_destroy_workqueue(fs_info->delayed_workers);
2080         btrfs_destroy_workqueue(fs_info->caching_workers);
2081         btrfs_destroy_workqueue(fs_info->readahead_workers);
2082         btrfs_destroy_workqueue(fs_info->flush_workers);
2083         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2084         btrfs_destroy_workqueue(fs_info->extent_workers);
2085 }
2086
2087 static void free_root_extent_buffers(struct btrfs_root *root)
2088 {
2089         if (root) {
2090                 free_extent_buffer(root->node);
2091                 free_extent_buffer(root->commit_root);
2092                 root->node = NULL;
2093                 root->commit_root = NULL;
2094         }
2095 }
2096
2097 /* helper to cleanup tree roots */
2098 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2099 {
2100         free_root_extent_buffers(info->tree_root);
2101
2102         free_root_extent_buffers(info->dev_root);
2103         free_root_extent_buffers(info->extent_root);
2104         free_root_extent_buffers(info->csum_root);
2105         free_root_extent_buffers(info->quota_root);
2106         free_root_extent_buffers(info->uuid_root);
2107         if (chunk_root)
2108                 free_root_extent_buffers(info->chunk_root);
2109 }
2110
2111 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2112 {
2113         int ret;
2114         struct btrfs_root *gang[8];
2115         int i;
2116
2117         while (!list_empty(&fs_info->dead_roots)) {
2118                 gang[0] = list_entry(fs_info->dead_roots.next,
2119                                      struct btrfs_root, root_list);
2120                 list_del(&gang[0]->root_list);
2121
2122                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2123                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2124                 } else {
2125                         free_extent_buffer(gang[0]->node);
2126                         free_extent_buffer(gang[0]->commit_root);
2127                         btrfs_put_fs_root(gang[0]);
2128                 }
2129         }
2130
2131         while (1) {
2132                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2133                                              (void **)gang, 0,
2134                                              ARRAY_SIZE(gang));
2135                 if (!ret)
2136                         break;
2137                 for (i = 0; i < ret; i++)
2138                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2139         }
2140
2141         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2142                 btrfs_free_log_root_tree(NULL, fs_info);
2143                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2144                                             fs_info->pinned_extents);
2145         }
2146 }
2147
2148 int open_ctree(struct super_block *sb,
2149                struct btrfs_fs_devices *fs_devices,
2150                char *options)
2151 {
2152         u32 sectorsize;
2153         u32 nodesize;
2154         u32 stripesize;
2155         u64 generation;
2156         u64 features;
2157         struct btrfs_key location;
2158         struct buffer_head *bh;
2159         struct btrfs_super_block *disk_super;
2160         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2161         struct btrfs_root *tree_root;
2162         struct btrfs_root *extent_root;
2163         struct btrfs_root *csum_root;
2164         struct btrfs_root *chunk_root;
2165         struct btrfs_root *dev_root;
2166         struct btrfs_root *quota_root;
2167         struct btrfs_root *uuid_root;
2168         struct btrfs_root *log_tree_root;
2169         int ret;
2170         int err = -EINVAL;
2171         int num_backups_tried = 0;
2172         int backup_index = 0;
2173         int max_active;
2174         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2175         bool create_uuid_tree;
2176         bool check_uuid_tree;
2177
2178         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2179         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2180         if (!tree_root || !chunk_root) {
2181                 err = -ENOMEM;
2182                 goto fail;
2183         }
2184
2185         ret = init_srcu_struct(&fs_info->subvol_srcu);
2186         if (ret) {
2187                 err = ret;
2188                 goto fail;
2189         }
2190
2191         ret = setup_bdi(fs_info, &fs_info->bdi);
2192         if (ret) {
2193                 err = ret;
2194                 goto fail_srcu;
2195         }
2196
2197         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2198         if (ret) {
2199                 err = ret;
2200                 goto fail_bdi;
2201         }
2202         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2203                                         (1 + ilog2(nr_cpu_ids));
2204
2205         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2206         if (ret) {
2207                 err = ret;
2208                 goto fail_dirty_metadata_bytes;
2209         }
2210
2211         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2212         if (ret) {
2213                 err = ret;
2214                 goto fail_delalloc_bytes;
2215         }
2216
2217         fs_info->btree_inode = new_inode(sb);
2218         if (!fs_info->btree_inode) {
2219                 err = -ENOMEM;
2220                 goto fail_bio_counter;
2221         }
2222
2223         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2224
2225         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2226         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2227         INIT_LIST_HEAD(&fs_info->trans_list);
2228         INIT_LIST_HEAD(&fs_info->dead_roots);
2229         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2230         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2231         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2232         spin_lock_init(&fs_info->delalloc_root_lock);
2233         spin_lock_init(&fs_info->trans_lock);
2234         spin_lock_init(&fs_info->fs_roots_radix_lock);
2235         spin_lock_init(&fs_info->delayed_iput_lock);
2236         spin_lock_init(&fs_info->defrag_inodes_lock);
2237         spin_lock_init(&fs_info->free_chunk_lock);
2238         spin_lock_init(&fs_info->tree_mod_seq_lock);
2239         spin_lock_init(&fs_info->super_lock);
2240         spin_lock_init(&fs_info->qgroup_op_lock);
2241         spin_lock_init(&fs_info->buffer_lock);
2242         spin_lock_init(&fs_info->unused_bgs_lock);
2243         rwlock_init(&fs_info->tree_mod_log_lock);
2244         mutex_init(&fs_info->reloc_mutex);
2245         mutex_init(&fs_info->delalloc_root_mutex);
2246         seqlock_init(&fs_info->profiles_lock);
2247
2248         init_completion(&fs_info->kobj_unregister);
2249         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2250         INIT_LIST_HEAD(&fs_info->space_info);
2251         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2252         INIT_LIST_HEAD(&fs_info->unused_bgs);
2253         btrfs_mapping_init(&fs_info->mapping_tree);
2254         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2255                              BTRFS_BLOCK_RSV_GLOBAL);
2256         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2257                              BTRFS_BLOCK_RSV_DELALLOC);
2258         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2259         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2260         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2261         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2262                              BTRFS_BLOCK_RSV_DELOPS);
2263         atomic_set(&fs_info->nr_async_submits, 0);
2264         atomic_set(&fs_info->async_delalloc_pages, 0);
2265         atomic_set(&fs_info->async_submit_draining, 0);
2266         atomic_set(&fs_info->nr_async_bios, 0);
2267         atomic_set(&fs_info->defrag_running, 0);
2268         atomic_set(&fs_info->qgroup_op_seq, 0);
2269         atomic64_set(&fs_info->tree_mod_seq, 0);
2270         fs_info->sb = sb;
2271         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2272         fs_info->metadata_ratio = 0;
2273         fs_info->defrag_inodes = RB_ROOT;
2274         fs_info->free_chunk_space = 0;
2275         fs_info->tree_mod_log = RB_ROOT;
2276         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2277         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2278         /* readahead state */
2279         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2280         spin_lock_init(&fs_info->reada_lock);
2281
2282         fs_info->thread_pool_size = min_t(unsigned long,
2283                                           num_online_cpus() + 2, 8);
2284
2285         INIT_LIST_HEAD(&fs_info->ordered_roots);
2286         spin_lock_init(&fs_info->ordered_root_lock);
2287         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2288                                         GFP_NOFS);
2289         if (!fs_info->delayed_root) {
2290                 err = -ENOMEM;
2291                 goto fail_iput;
2292         }
2293         btrfs_init_delayed_root(fs_info->delayed_root);
2294
2295         mutex_init(&fs_info->scrub_lock);
2296         atomic_set(&fs_info->scrubs_running, 0);
2297         atomic_set(&fs_info->scrub_pause_req, 0);
2298         atomic_set(&fs_info->scrubs_paused, 0);
2299         atomic_set(&fs_info->scrub_cancel_req, 0);
2300         init_waitqueue_head(&fs_info->replace_wait);
2301         init_waitqueue_head(&fs_info->scrub_pause_wait);
2302         fs_info->scrub_workers_refcnt = 0;
2303 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2304         fs_info->check_integrity_print_mask = 0;
2305 #endif
2306
2307         spin_lock_init(&fs_info->balance_lock);
2308         mutex_init(&fs_info->balance_mutex);
2309         atomic_set(&fs_info->balance_running, 0);
2310         atomic_set(&fs_info->balance_pause_req, 0);
2311         atomic_set(&fs_info->balance_cancel_req, 0);
2312         fs_info->balance_ctl = NULL;
2313         init_waitqueue_head(&fs_info->balance_wait_q);
2314         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2315
2316         sb->s_blocksize = 4096;
2317         sb->s_blocksize_bits = blksize_bits(4096);
2318         sb->s_bdi = &fs_info->bdi;
2319
2320         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2321         set_nlink(fs_info->btree_inode, 1);
2322         /*
2323          * we set the i_size on the btree inode to the max possible int.
2324          * the real end of the address space is determined by all of
2325          * the devices in the system
2326          */
2327         fs_info->btree_inode->i_size = OFFSET_MAX;
2328         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2329         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2330
2331         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2332         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2333                              fs_info->btree_inode->i_mapping);
2334         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2335         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2336
2337         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2338
2339         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2340         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2341                sizeof(struct btrfs_key));
2342         set_bit(BTRFS_INODE_DUMMY,
2343                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2344         btrfs_insert_inode_hash(fs_info->btree_inode);
2345
2346         spin_lock_init(&fs_info->block_group_cache_lock);
2347         fs_info->block_group_cache_tree = RB_ROOT;
2348         fs_info->first_logical_byte = (u64)-1;
2349
2350         extent_io_tree_init(&fs_info->freed_extents[0],
2351                              fs_info->btree_inode->i_mapping);
2352         extent_io_tree_init(&fs_info->freed_extents[1],
2353                              fs_info->btree_inode->i_mapping);
2354         fs_info->pinned_extents = &fs_info->freed_extents[0];
2355         fs_info->do_barriers = 1;
2356
2357
2358         mutex_init(&fs_info->ordered_operations_mutex);
2359         mutex_init(&fs_info->ordered_extent_flush_mutex);
2360         mutex_init(&fs_info->tree_log_mutex);
2361         mutex_init(&fs_info->chunk_mutex);
2362         mutex_init(&fs_info->transaction_kthread_mutex);
2363         mutex_init(&fs_info->cleaner_mutex);
2364         mutex_init(&fs_info->volume_mutex);
2365         init_rwsem(&fs_info->commit_root_sem);
2366         init_rwsem(&fs_info->cleanup_work_sem);
2367         init_rwsem(&fs_info->subvol_sem);
2368         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2369         fs_info->dev_replace.lock_owner = 0;
2370         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2371         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2372         mutex_init(&fs_info->dev_replace.lock_management_lock);
2373         mutex_init(&fs_info->dev_replace.lock);
2374
2375         spin_lock_init(&fs_info->qgroup_lock);
2376         mutex_init(&fs_info->qgroup_ioctl_lock);
2377         fs_info->qgroup_tree = RB_ROOT;
2378         fs_info->qgroup_op_tree = RB_ROOT;
2379         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2380         fs_info->qgroup_seq = 1;
2381         fs_info->quota_enabled = 0;
2382         fs_info->pending_quota_state = 0;
2383         fs_info->qgroup_ulist = NULL;
2384         mutex_init(&fs_info->qgroup_rescan_lock);
2385
2386         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2387         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2388
2389         init_waitqueue_head(&fs_info->transaction_throttle);
2390         init_waitqueue_head(&fs_info->transaction_wait);
2391         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2392         init_waitqueue_head(&fs_info->async_submit_wait);
2393
2394         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2395
2396         ret = btrfs_alloc_stripe_hash_table(fs_info);
2397         if (ret) {
2398                 err = ret;
2399                 goto fail_alloc;
2400         }
2401
2402         __setup_root(4096, 4096, 4096, tree_root,
2403                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2404
2405         invalidate_bdev(fs_devices->latest_bdev);
2406
2407         /*
2408          * Read super block and check the signature bytes only
2409          */
2410         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2411         if (!bh) {
2412                 err = -EINVAL;
2413                 goto fail_alloc;
2414         }
2415
2416         /*
2417          * We want to check superblock checksum, the type is stored inside.
2418          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2419          */
2420         if (btrfs_check_super_csum(bh->b_data)) {
2421                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2422                 err = -EINVAL;
2423                 goto fail_alloc;
2424         }
2425
2426         /*
2427          * super_copy is zeroed at allocation time and we never touch the
2428          * following bytes up to INFO_SIZE, the checksum is calculated from
2429          * the whole block of INFO_SIZE
2430          */
2431         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2432         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2433                sizeof(*fs_info->super_for_commit));
2434         brelse(bh);
2435
2436         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2437
2438         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2439         if (ret) {
2440                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2441                 err = -EINVAL;
2442                 goto fail_alloc;
2443         }
2444
2445         disk_super = fs_info->super_copy;
2446         if (!btrfs_super_root(disk_super))
2447                 goto fail_alloc;
2448
2449         /* check FS state, whether FS is broken. */
2450         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2451                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2452
2453         /*
2454          * run through our array of backup supers and setup
2455          * our ring pointer to the oldest one
2456          */
2457         generation = btrfs_super_generation(disk_super);
2458         find_oldest_super_backup(fs_info, generation);
2459
2460         /*
2461          * In the long term, we'll store the compression type in the super
2462          * block, and it'll be used for per file compression control.
2463          */
2464         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2465
2466         ret = btrfs_parse_options(tree_root, options);
2467         if (ret) {
2468                 err = ret;
2469                 goto fail_alloc;
2470         }
2471
2472         features = btrfs_super_incompat_flags(disk_super) &
2473                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2474         if (features) {
2475                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2476                        "unsupported optional features (%Lx).\n",
2477                        features);
2478                 err = -EINVAL;
2479                 goto fail_alloc;
2480         }
2481
2482         /*
2483          * Leafsize and nodesize were always equal, this is only a sanity check.
2484          */
2485         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2486             btrfs_super_nodesize(disk_super)) {
2487                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2488                        "blocksizes don't match.  node %d leaf %d\n",
2489                        btrfs_super_nodesize(disk_super),
2490                        le32_to_cpu(disk_super->__unused_leafsize));
2491                 err = -EINVAL;
2492                 goto fail_alloc;
2493         }
2494         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2495                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2496                        "blocksize (%d) was too large\n",
2497                        btrfs_super_nodesize(disk_super));
2498                 err = -EINVAL;
2499                 goto fail_alloc;
2500         }
2501
2502         features = btrfs_super_incompat_flags(disk_super);
2503         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2504         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2505                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2506
2507         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2508                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2509
2510         /*
2511          * flag our filesystem as having big metadata blocks if
2512          * they are bigger than the page size
2513          */
2514         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2515                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2516                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2517                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2518         }
2519
2520         nodesize = btrfs_super_nodesize(disk_super);
2521         sectorsize = btrfs_super_sectorsize(disk_super);
2522         stripesize = btrfs_super_stripesize(disk_super);
2523         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2524         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2525
2526         /*
2527          * mixed block groups end up with duplicate but slightly offset
2528          * extent buffers for the same range.  It leads to corruptions
2529          */
2530         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2531             (sectorsize != nodesize)) {
2532                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2533                                 "are not allowed for mixed block groups on %s\n",
2534                                 sb->s_id);
2535                 goto fail_alloc;
2536         }
2537
2538         /*
2539          * Needn't use the lock because there is no other task which will
2540          * update the flag.
2541          */
2542         btrfs_set_super_incompat_flags(disk_super, features);
2543
2544         features = btrfs_super_compat_ro_flags(disk_super) &
2545                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2546         if (!(sb->s_flags & MS_RDONLY) && features) {
2547                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2548                        "unsupported option features (%Lx).\n",
2549                        features);
2550                 err = -EINVAL;
2551                 goto fail_alloc;
2552         }
2553
2554         max_active = fs_info->thread_pool_size;
2555
2556         fs_info->workers =
2557                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2558                                       max_active, 16);
2559
2560         fs_info->delalloc_workers =
2561                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2562
2563         fs_info->flush_workers =
2564                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2565
2566         fs_info->caching_workers =
2567                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2568
2569         /*
2570          * a higher idle thresh on the submit workers makes it much more
2571          * likely that bios will be send down in a sane order to the
2572          * devices
2573          */
2574         fs_info->submit_workers =
2575                 btrfs_alloc_workqueue("submit", flags,
2576                                       min_t(u64, fs_devices->num_devices,
2577                                             max_active), 64);
2578
2579         fs_info->fixup_workers =
2580                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2581
2582         /*
2583          * endios are largely parallel and should have a very
2584          * low idle thresh
2585          */
2586         fs_info->endio_workers =
2587                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2588         fs_info->endio_meta_workers =
2589                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2590         fs_info->endio_meta_write_workers =
2591                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2592         fs_info->endio_raid56_workers =
2593                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2594         fs_info->endio_repair_workers =
2595                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2596         fs_info->rmw_workers =
2597                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2598         fs_info->endio_write_workers =
2599                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2600         fs_info->endio_freespace_worker =
2601                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2602         fs_info->delayed_workers =
2603                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2604         fs_info->readahead_workers =
2605                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2606         fs_info->qgroup_rescan_workers =
2607                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2608         fs_info->extent_workers =
2609                 btrfs_alloc_workqueue("extent-refs", flags,
2610                                       min_t(u64, fs_devices->num_devices,
2611                                             max_active), 8);
2612
2613         if (!(fs_info->workers && fs_info->delalloc_workers &&
2614               fs_info->submit_workers && fs_info->flush_workers &&
2615               fs_info->endio_workers && fs_info->endio_meta_workers &&
2616               fs_info->endio_meta_write_workers &&
2617               fs_info->endio_repair_workers &&
2618               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2619               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2620               fs_info->caching_workers && fs_info->readahead_workers &&
2621               fs_info->fixup_workers && fs_info->delayed_workers &&
2622               fs_info->extent_workers &&
2623               fs_info->qgroup_rescan_workers)) {
2624                 err = -ENOMEM;
2625                 goto fail_sb_buffer;
2626         }
2627
2628         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2629         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2630                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2631
2632         tree_root->nodesize = nodesize;
2633         tree_root->sectorsize = sectorsize;
2634         tree_root->stripesize = stripesize;
2635
2636         sb->s_blocksize = sectorsize;
2637         sb->s_blocksize_bits = blksize_bits(sectorsize);
2638
2639         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2640                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2641                 goto fail_sb_buffer;
2642         }
2643
2644         if (sectorsize != PAGE_SIZE) {
2645                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2646                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2647                 goto fail_sb_buffer;
2648         }
2649
2650         mutex_lock(&fs_info->chunk_mutex);
2651         ret = btrfs_read_sys_array(tree_root);
2652         mutex_unlock(&fs_info->chunk_mutex);
2653         if (ret) {
2654                 printk(KERN_WARNING "BTRFS: failed to read the system "
2655                        "array on %s\n", sb->s_id);
2656                 goto fail_sb_buffer;
2657         }
2658
2659         generation = btrfs_super_chunk_root_generation(disk_super);
2660
2661         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2662                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2663
2664         chunk_root->node = read_tree_block(chunk_root,
2665                                            btrfs_super_chunk_root(disk_super),
2666                                            generation);
2667         if (!chunk_root->node ||
2668             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2669                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2670                        sb->s_id);
2671                 goto fail_tree_roots;
2672         }
2673         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2674         chunk_root->commit_root = btrfs_root_node(chunk_root);
2675
2676         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2677            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2678
2679         ret = btrfs_read_chunk_tree(chunk_root);
2680         if (ret) {
2681                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2682                        sb->s_id);
2683                 goto fail_tree_roots;
2684         }
2685
2686         /*
2687          * keep the device that is marked to be the target device for the
2688          * dev_replace procedure
2689          */
2690         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2691
2692         if (!fs_devices->latest_bdev) {
2693                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2694                        sb->s_id);
2695                 goto fail_tree_roots;
2696         }
2697
2698 retry_root_backup:
2699         generation = btrfs_super_generation(disk_super);
2700
2701         tree_root->node = read_tree_block(tree_root,
2702                                           btrfs_super_root(disk_super),
2703                                           generation);
2704         if (!tree_root->node ||
2705             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2706                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2707                        sb->s_id);
2708
2709                 goto recovery_tree_root;
2710         }
2711
2712         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2713         tree_root->commit_root = btrfs_root_node(tree_root);
2714         btrfs_set_root_refs(&tree_root->root_item, 1);
2715
2716         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2717         location.type = BTRFS_ROOT_ITEM_KEY;
2718         location.offset = 0;
2719
2720         extent_root = btrfs_read_tree_root(tree_root, &location);
2721         if (IS_ERR(extent_root)) {
2722                 ret = PTR_ERR(extent_root);
2723                 goto recovery_tree_root;
2724         }
2725         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2726         fs_info->extent_root = extent_root;
2727
2728         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2729         dev_root = btrfs_read_tree_root(tree_root, &location);
2730         if (IS_ERR(dev_root)) {
2731                 ret = PTR_ERR(dev_root);
2732                 goto recovery_tree_root;
2733         }
2734         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2735         fs_info->dev_root = dev_root;
2736         btrfs_init_devices_late(fs_info);
2737
2738         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2739         csum_root = btrfs_read_tree_root(tree_root, &location);
2740         if (IS_ERR(csum_root)) {
2741                 ret = PTR_ERR(csum_root);
2742                 goto recovery_tree_root;
2743         }
2744         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2745         fs_info->csum_root = csum_root;
2746
2747         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2748         quota_root = btrfs_read_tree_root(tree_root, &location);
2749         if (!IS_ERR(quota_root)) {
2750                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2751                 fs_info->quota_enabled = 1;
2752                 fs_info->pending_quota_state = 1;
2753                 fs_info->quota_root = quota_root;
2754         }
2755
2756         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2757         uuid_root = btrfs_read_tree_root(tree_root, &location);
2758         if (IS_ERR(uuid_root)) {
2759                 ret = PTR_ERR(uuid_root);
2760                 if (ret != -ENOENT)
2761                         goto recovery_tree_root;
2762                 create_uuid_tree = true;
2763                 check_uuid_tree = false;
2764         } else {
2765                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2766                 fs_info->uuid_root = uuid_root;
2767                 create_uuid_tree = false;
2768                 check_uuid_tree =
2769                     generation != btrfs_super_uuid_tree_generation(disk_super);
2770         }
2771
2772         fs_info->generation = generation;
2773         fs_info->last_trans_committed = generation;
2774
2775         ret = btrfs_recover_balance(fs_info);
2776         if (ret) {
2777                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2778                 goto fail_block_groups;
2779         }
2780
2781         ret = btrfs_init_dev_stats(fs_info);
2782         if (ret) {
2783                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2784                        ret);
2785                 goto fail_block_groups;
2786         }
2787
2788         ret = btrfs_init_dev_replace(fs_info);
2789         if (ret) {
2790                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2791                 goto fail_block_groups;
2792         }
2793
2794         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2795
2796         ret = btrfs_sysfs_add_one(fs_info);
2797         if (ret) {
2798                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2799                 goto fail_block_groups;
2800         }
2801
2802         ret = btrfs_init_space_info(fs_info);
2803         if (ret) {
2804                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2805                 goto fail_sysfs;
2806         }
2807
2808         ret = btrfs_read_block_groups(extent_root);
2809         if (ret) {
2810                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2811                 goto fail_sysfs;
2812         }
2813         fs_info->num_tolerated_disk_barrier_failures =
2814                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2815         if (fs_info->fs_devices->missing_devices >
2816              fs_info->num_tolerated_disk_barrier_failures &&
2817             !(sb->s_flags & MS_RDONLY)) {
2818                 printk(KERN_WARNING "BTRFS: "
2819                         "too many missing devices, writeable mount is not allowed\n");
2820                 goto fail_sysfs;
2821         }
2822
2823         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2824                                                "btrfs-cleaner");
2825         if (IS_ERR(fs_info->cleaner_kthread))
2826                 goto fail_sysfs;
2827
2828         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2829                                                    tree_root,
2830                                                    "btrfs-transaction");
2831         if (IS_ERR(fs_info->transaction_kthread))
2832                 goto fail_cleaner;
2833
2834         if (!btrfs_test_opt(tree_root, SSD) &&
2835             !btrfs_test_opt(tree_root, NOSSD) &&
2836             !fs_info->fs_devices->rotating) {
2837                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2838                        "mode\n");
2839                 btrfs_set_opt(fs_info->mount_opt, SSD);
2840         }
2841
2842         /*
2843          * Mount does not set all options immediatelly, we can do it now and do
2844          * not have to wait for transaction commit
2845          */
2846         btrfs_apply_pending_changes(fs_info);
2847
2848 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2849         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2850                 ret = btrfsic_mount(tree_root, fs_devices,
2851                                     btrfs_test_opt(tree_root,
2852                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2853                                     1 : 0,
2854                                     fs_info->check_integrity_print_mask);
2855                 if (ret)
2856                         printk(KERN_WARNING "BTRFS: failed to initialize"
2857                                " integrity check module %s\n", sb->s_id);
2858         }
2859 #endif
2860         ret = btrfs_read_qgroup_config(fs_info);
2861         if (ret)
2862                 goto fail_trans_kthread;
2863
2864         /* do not make disk changes in broken FS */
2865         if (btrfs_super_log_root(disk_super) != 0) {
2866                 u64 bytenr = btrfs_super_log_root(disk_super);
2867
2868                 if (fs_devices->rw_devices == 0) {
2869                         printk(KERN_WARNING "BTRFS: log replay required "
2870                                "on RO media\n");
2871                         err = -EIO;
2872                         goto fail_qgroup;
2873                 }
2874
2875                 log_tree_root = btrfs_alloc_root(fs_info);
2876                 if (!log_tree_root) {
2877                         err = -ENOMEM;
2878                         goto fail_qgroup;
2879                 }
2880
2881                 __setup_root(nodesize, sectorsize, stripesize,
2882                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2883
2884                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2885                                                       generation + 1);
2886                 if (!log_tree_root->node ||
2887                     !extent_buffer_uptodate(log_tree_root->node)) {
2888                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2889                         free_extent_buffer(log_tree_root->node);
2890                         kfree(log_tree_root);
2891                         goto fail_qgroup;
2892                 }
2893                 /* returns with log_tree_root freed on success */
2894                 ret = btrfs_recover_log_trees(log_tree_root);
2895                 if (ret) {
2896                         btrfs_error(tree_root->fs_info, ret,
2897                                     "Failed to recover log tree");
2898                         free_extent_buffer(log_tree_root->node);
2899                         kfree(log_tree_root);
2900                         goto fail_qgroup;
2901                 }
2902
2903                 if (sb->s_flags & MS_RDONLY) {
2904                         ret = btrfs_commit_super(tree_root);
2905                         if (ret)
2906                                 goto fail_qgroup;
2907                 }
2908         }
2909
2910         ret = btrfs_find_orphan_roots(tree_root);
2911         if (ret)
2912                 goto fail_qgroup;
2913
2914         if (!(sb->s_flags & MS_RDONLY)) {
2915                 ret = btrfs_cleanup_fs_roots(fs_info);
2916                 if (ret)
2917                         goto fail_qgroup;
2918
2919                 mutex_lock(&fs_info->cleaner_mutex);
2920                 ret = btrfs_recover_relocation(tree_root);
2921                 mutex_unlock(&fs_info->cleaner_mutex);
2922                 if (ret < 0) {
2923                         printk(KERN_WARNING
2924                                "BTRFS: failed to recover relocation\n");
2925                         err = -EINVAL;
2926                         goto fail_qgroup;
2927                 }
2928         }
2929
2930         location.objectid = BTRFS_FS_TREE_OBJECTID;
2931         location.type = BTRFS_ROOT_ITEM_KEY;
2932         location.offset = 0;
2933
2934         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2935         if (IS_ERR(fs_info->fs_root)) {
2936                 err = PTR_ERR(fs_info->fs_root);
2937                 goto fail_qgroup;
2938         }
2939
2940         if (sb->s_flags & MS_RDONLY)
2941                 return 0;
2942
2943         down_read(&fs_info->cleanup_work_sem);
2944         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2945             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2946                 up_read(&fs_info->cleanup_work_sem);
2947                 close_ctree(tree_root);
2948                 return ret;
2949         }
2950         up_read(&fs_info->cleanup_work_sem);
2951
2952         ret = btrfs_resume_balance_async(fs_info);
2953         if (ret) {
2954                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2955                 close_ctree(tree_root);
2956                 return ret;
2957         }
2958
2959         ret = btrfs_resume_dev_replace_async(fs_info);
2960         if (ret) {
2961                 pr_warn("BTRFS: failed to resume dev_replace\n");
2962                 close_ctree(tree_root);
2963                 return ret;
2964         }
2965
2966         btrfs_qgroup_rescan_resume(fs_info);
2967
2968         if (create_uuid_tree) {
2969                 pr_info("BTRFS: creating UUID tree\n");
2970                 ret = btrfs_create_uuid_tree(fs_info);
2971                 if (ret) {
2972                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2973                                 ret);
2974                         close_ctree(tree_root);
2975                         return ret;
2976                 }
2977         } else if (check_uuid_tree ||
2978                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2979                 pr_info("BTRFS: checking UUID tree\n");
2980                 ret = btrfs_check_uuid_tree(fs_info);
2981                 if (ret) {
2982                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2983                                 ret);
2984                         close_ctree(tree_root);
2985                         return ret;
2986                 }
2987         } else {
2988                 fs_info->update_uuid_tree_gen = 1;
2989         }
2990
2991         fs_info->open = 1;
2992
2993         return 0;
2994
2995 fail_qgroup:
2996         btrfs_free_qgroup_config(fs_info);
2997 fail_trans_kthread:
2998         kthread_stop(fs_info->transaction_kthread);
2999         btrfs_cleanup_transaction(fs_info->tree_root);
3000         btrfs_free_fs_roots(fs_info);
3001 fail_cleaner:
3002         kthread_stop(fs_info->cleaner_kthread);
3003
3004         /*
3005          * make sure we're done with the btree inode before we stop our
3006          * kthreads
3007          */
3008         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3009
3010 fail_sysfs:
3011         btrfs_sysfs_remove_one(fs_info);
3012
3013 fail_block_groups:
3014         btrfs_put_block_group_cache(fs_info);
3015         btrfs_free_block_groups(fs_info);
3016
3017 fail_tree_roots:
3018         free_root_pointers(fs_info, 1);
3019         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3020
3021 fail_sb_buffer:
3022         btrfs_stop_all_workers(fs_info);
3023 fail_alloc:
3024 fail_iput:
3025         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3026
3027         iput(fs_info->btree_inode);
3028 fail_bio_counter:
3029         percpu_counter_destroy(&fs_info->bio_counter);
3030 fail_delalloc_bytes:
3031         percpu_counter_destroy(&fs_info->delalloc_bytes);
3032 fail_dirty_metadata_bytes:
3033         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3034 fail_bdi:
3035         bdi_destroy(&fs_info->bdi);
3036 fail_srcu:
3037         cleanup_srcu_struct(&fs_info->subvol_srcu);
3038 fail:
3039         btrfs_free_stripe_hash_table(fs_info);
3040         btrfs_close_devices(fs_info->fs_devices);
3041         return err;
3042
3043 recovery_tree_root:
3044         if (!btrfs_test_opt(tree_root, RECOVERY))
3045                 goto fail_tree_roots;
3046
3047         free_root_pointers(fs_info, 0);
3048
3049         /* don't use the log in recovery mode, it won't be valid */
3050         btrfs_set_super_log_root(disk_super, 0);
3051
3052         /* we can't trust the free space cache either */
3053         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3054
3055         ret = next_root_backup(fs_info, fs_info->super_copy,
3056                                &num_backups_tried, &backup_index);
3057         if (ret == -1)
3058                 goto fail_block_groups;
3059         goto retry_root_backup;
3060 }
3061
3062 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3063 {
3064         if (uptodate) {
3065                 set_buffer_uptodate(bh);
3066         } else {
3067                 struct btrfs_device *device = (struct btrfs_device *)
3068                         bh->b_private;
3069
3070                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3071                                           "I/O error on %s\n",
3072                                           rcu_str_deref(device->name));
3073                 /* note, we dont' set_buffer_write_io_error because we have
3074                  * our own ways of dealing with the IO errors
3075                  */
3076                 clear_buffer_uptodate(bh);
3077                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3078         }
3079         unlock_buffer(bh);
3080         put_bh(bh);
3081 }
3082
3083 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3084 {
3085         struct buffer_head *bh;
3086         struct buffer_head *latest = NULL;
3087         struct btrfs_super_block *super;
3088         int i;
3089         u64 transid = 0;
3090         u64 bytenr;
3091
3092         /* we would like to check all the supers, but that would make
3093          * a btrfs mount succeed after a mkfs from a different FS.
3094          * So, we need to add a special mount option to scan for
3095          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3096          */
3097         for (i = 0; i < 1; i++) {
3098                 bytenr = btrfs_sb_offset(i);
3099                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3100                                         i_size_read(bdev->bd_inode))
3101                         break;
3102                 bh = __bread(bdev, bytenr / 4096,
3103                                         BTRFS_SUPER_INFO_SIZE);
3104                 if (!bh)
3105                         continue;
3106
3107                 super = (struct btrfs_super_block *)bh->b_data;
3108                 if (btrfs_super_bytenr(super) != bytenr ||
3109                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3110                         brelse(bh);
3111                         continue;
3112                 }
3113
3114                 if (!latest || btrfs_super_generation(super) > transid) {
3115                         brelse(latest);
3116                         latest = bh;
3117                         transid = btrfs_super_generation(super);
3118                 } else {
3119                         brelse(bh);
3120                 }
3121         }
3122         return latest;
3123 }
3124
3125 /*
3126  * this should be called twice, once with wait == 0 and
3127  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3128  * we write are pinned.
3129  *
3130  * They are released when wait == 1 is done.
3131  * max_mirrors must be the same for both runs, and it indicates how
3132  * many supers on this one device should be written.
3133  *
3134  * max_mirrors == 0 means to write them all.
3135  */
3136 static int write_dev_supers(struct btrfs_device *device,
3137                             struct btrfs_super_block *sb,
3138                             int do_barriers, int wait, int max_mirrors)
3139 {
3140         struct buffer_head *bh;
3141         int i;
3142         int ret;
3143         int errors = 0;
3144         u32 crc;
3145         u64 bytenr;
3146
3147         if (max_mirrors == 0)
3148                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3149
3150         for (i = 0; i < max_mirrors; i++) {
3151                 bytenr = btrfs_sb_offset(i);
3152                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3153                     device->commit_total_bytes)
3154                         break;
3155
3156                 if (wait) {
3157                         bh = __find_get_block(device->bdev, bytenr / 4096,
3158                                               BTRFS_SUPER_INFO_SIZE);
3159                         if (!bh) {
3160                                 errors++;
3161                                 continue;
3162                         }
3163                         wait_on_buffer(bh);
3164                         if (!buffer_uptodate(bh))
3165                                 errors++;
3166
3167                         /* drop our reference */
3168                         brelse(bh);
3169
3170                         /* drop the reference from the wait == 0 run */
3171                         brelse(bh);
3172                         continue;
3173                 } else {
3174                         btrfs_set_super_bytenr(sb, bytenr);
3175
3176                         crc = ~(u32)0;
3177                         crc = btrfs_csum_data((char *)sb +
3178                                               BTRFS_CSUM_SIZE, crc,
3179                                               BTRFS_SUPER_INFO_SIZE -
3180                                               BTRFS_CSUM_SIZE);
3181                         btrfs_csum_final(crc, sb->csum);
3182
3183                         /*
3184                          * one reference for us, and we leave it for the
3185                          * caller
3186                          */
3187                         bh = __getblk(device->bdev, bytenr / 4096,
3188                                       BTRFS_SUPER_INFO_SIZE);
3189                         if (!bh) {
3190                                 printk(KERN_ERR "BTRFS: couldn't get super "
3191                                        "buffer head for bytenr %Lu\n", bytenr);
3192                                 errors++;
3193                                 continue;
3194                         }
3195
3196                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3197
3198                         /* one reference for submit_bh */
3199                         get_bh(bh);
3200
3201                         set_buffer_uptodate(bh);
3202                         lock_buffer(bh);
3203                         bh->b_end_io = btrfs_end_buffer_write_sync;
3204                         bh->b_private = device;
3205                 }
3206
3207                 /*
3208                  * we fua the first super.  The others we allow
3209                  * to go down lazy.
3210                  */
3211                 if (i == 0)
3212                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3213                 else
3214                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3215                 if (ret)
3216                         errors++;
3217         }
3218         return errors < i ? 0 : -1;
3219 }
3220
3221 /*
3222  * endio for the write_dev_flush, this will wake anyone waiting
3223  * for the barrier when it is done
3224  */
3225 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3226 {
3227         if (err) {
3228                 if (err == -EOPNOTSUPP)
3229                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3230                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3231         }
3232         if (bio->bi_private)
3233                 complete(bio->bi_private);
3234         bio_put(bio);
3235 }
3236
3237 /*
3238  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3239  * sent down.  With wait == 1, it waits for the previous flush.
3240  *
3241  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3242  * capable
3243  */
3244 static int write_dev_flush(struct btrfs_device *device, int wait)
3245 {
3246         struct bio *bio;
3247         int ret = 0;
3248
3249         if (device->nobarriers)
3250                 return 0;
3251
3252         if (wait) {
3253                 bio = device->flush_bio;
3254                 if (!bio)
3255                         return 0;
3256
3257                 wait_for_completion(&device->flush_wait);
3258
3259                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3260                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3261                                       rcu_str_deref(device->name));
3262                         device->nobarriers = 1;
3263                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3264                         ret = -EIO;
3265                         btrfs_dev_stat_inc_and_print(device,
3266                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3267                 }
3268
3269                 /* drop the reference from the wait == 0 run */
3270                 bio_put(bio);
3271                 device->flush_bio = NULL;
3272
3273                 return ret;
3274         }
3275
3276         /*
3277          * one reference for us, and we leave it for the
3278          * caller
3279          */
3280         device->flush_bio = NULL;
3281         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3282         if (!bio)
3283                 return -ENOMEM;
3284
3285         bio->bi_end_io = btrfs_end_empty_barrier;
3286         bio->bi_bdev = device->bdev;
3287         init_completion(&device->flush_wait);
3288         bio->bi_private = &device->flush_wait;
3289         device->flush_bio = bio;
3290
3291         bio_get(bio);
3292         btrfsic_submit_bio(WRITE_FLUSH, bio);
3293
3294         return 0;
3295 }
3296
3297 /*
3298  * send an empty flush down to each device in parallel,
3299  * then wait for them
3300  */
3301 static int barrier_all_devices(struct btrfs_fs_info *info)
3302 {
3303         struct list_head *head;
3304         struct btrfs_device *dev;
3305         int errors_send = 0;
3306         int errors_wait = 0;
3307         int ret;
3308
3309         /* send down all the barriers */
3310         head = &info->fs_devices->devices;
3311         list_for_each_entry_rcu(dev, head, dev_list) {
3312                 if (dev->missing)
3313                         continue;
3314                 if (!dev->bdev) {
3315                         errors_send++;
3316                         continue;
3317                 }
3318                 if (!dev->in_fs_metadata || !dev->writeable)
3319                         continue;
3320
3321                 ret = write_dev_flush(dev, 0);
3322                 if (ret)
3323                         errors_send++;
3324         }
3325
3326         /* wait for all the barriers */
3327         list_for_each_entry_rcu(dev, head, dev_list) {
3328                 if (dev->missing)
3329                         continue;
3330                 if (!dev->bdev) {
3331                         errors_wait++;
3332                         continue;
3333                 }
3334                 if (!dev->in_fs_metadata || !dev->writeable)
3335                         continue;
3336
3337                 ret = write_dev_flush(dev, 1);
3338                 if (ret)
3339                         errors_wait++;
3340         }
3341         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3342             errors_wait > info->num_tolerated_disk_barrier_failures)
3343                 return -EIO;
3344         return 0;
3345 }
3346
3347 int btrfs_calc_num_tolerated_disk_barrier_failures(
3348         struct btrfs_fs_info *fs_info)
3349 {
3350         struct btrfs_ioctl_space_info space;
3351         struct btrfs_space_info *sinfo;
3352         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3353                        BTRFS_BLOCK_GROUP_SYSTEM,
3354                        BTRFS_BLOCK_GROUP_METADATA,
3355                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3356         int num_types = 4;
3357         int i;
3358         int c;
3359         int num_tolerated_disk_barrier_failures =
3360                 (int)fs_info->fs_devices->num_devices;
3361
3362         for (i = 0; i < num_types; i++) {
3363                 struct btrfs_space_info *tmp;
3364
3365                 sinfo = NULL;
3366                 rcu_read_lock();
3367                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3368                         if (tmp->flags == types[i]) {
3369                                 sinfo = tmp;
3370                                 break;
3371                         }
3372                 }
3373                 rcu_read_unlock();
3374
3375                 if (!sinfo)
3376                         continue;
3377
3378                 down_read(&sinfo->groups_sem);
3379                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3380                         if (!list_empty(&sinfo->block_groups[c])) {
3381                                 u64 flags;
3382
3383                                 btrfs_get_block_group_info(
3384                                         &sinfo->block_groups[c], &space);
3385                                 if (space.total_bytes == 0 ||
3386                                     space.used_bytes == 0)
3387                                         continue;
3388                                 flags = space.flags;
3389                                 /*
3390                                  * return
3391                                  * 0: if dup, single or RAID0 is configured for
3392                                  *    any of metadata, system or data, else
3393                                  * 1: if RAID5 is configured, or if RAID1 or
3394                                  *    RAID10 is configured and only two mirrors
3395                                  *    are used, else
3396                                  * 2: if RAID6 is configured, else
3397                                  * num_mirrors - 1: if RAID1 or RAID10 is
3398                                  *                  configured and more than
3399                                  *                  2 mirrors are used.
3400                                  */
3401                                 if (num_tolerated_disk_barrier_failures > 0 &&
3402                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3403                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3404                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3405                                       == 0)))
3406                                         num_tolerated_disk_barrier_failures = 0;
3407                                 else if (num_tolerated_disk_barrier_failures > 1) {
3408                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3409                                             BTRFS_BLOCK_GROUP_RAID5 |
3410                                             BTRFS_BLOCK_GROUP_RAID10)) {
3411                                                 num_tolerated_disk_barrier_failures = 1;
3412                                         } else if (flags &
3413                                                    BTRFS_BLOCK_GROUP_RAID6) {
3414                                                 num_tolerated_disk_barrier_failures = 2;
3415                                         }
3416                                 }
3417                         }
3418                 }
3419                 up_read(&sinfo->groups_sem);
3420         }
3421
3422         return num_tolerated_disk_barrier_failures;
3423 }
3424
3425 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3426 {
3427         struct list_head *head;
3428         struct btrfs_device *dev;
3429         struct btrfs_super_block *sb;
3430         struct btrfs_dev_item *dev_item;
3431         int ret;
3432         int do_barriers;
3433         int max_errors;
3434         int total_errors = 0;
3435         u64 flags;
3436
3437         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3438         backup_super_roots(root->fs_info);
3439
3440         sb = root->fs_info->super_for_commit;
3441         dev_item = &sb->dev_item;
3442
3443         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3444         head = &root->fs_info->fs_devices->devices;
3445         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3446
3447         if (do_barriers) {
3448                 ret = barrier_all_devices(root->fs_info);
3449                 if (ret) {
3450                         mutex_unlock(
3451                                 &root->fs_info->fs_devices->device_list_mutex);
3452                         btrfs_error(root->fs_info, ret,
3453                                     "errors while submitting device barriers.");
3454                         return ret;
3455                 }
3456         }
3457
3458         list_for_each_entry_rcu(dev, head, dev_list) {
3459                 if (!dev->bdev) {
3460                         total_errors++;
3461                         continue;
3462                 }
3463                 if (!dev->in_fs_metadata || !dev->writeable)
3464                         continue;
3465
3466                 btrfs_set_stack_device_generation(dev_item, 0);
3467                 btrfs_set_stack_device_type(dev_item, dev->type);
3468                 btrfs_set_stack_device_id(dev_item, dev->devid);
3469                 btrfs_set_stack_device_total_bytes(dev_item,
3470                                                    dev->commit_total_bytes);
3471                 btrfs_set_stack_device_bytes_used(dev_item,
3472                                                   dev->commit_bytes_used);
3473                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3474                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3475                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3476                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3477                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3478
3479                 flags = btrfs_super_flags(sb);
3480                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3481
3482                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3483                 if (ret)
3484                         total_errors++;
3485         }
3486         if (total_errors > max_errors) {
3487                 btrfs_err(root->fs_info, "%d errors while writing supers",
3488                        total_errors);
3489                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3490
3491                 /* FUA is masked off if unsupported and can't be the reason */
3492                 btrfs_error(root->fs_info, -EIO,
3493                             "%d errors while writing supers", total_errors);
3494                 return -EIO;
3495         }
3496
3497         total_errors = 0;
3498         list_for_each_entry_rcu(dev, head, dev_list) {
3499                 if (!dev->bdev)
3500                         continue;
3501                 if (!dev->in_fs_metadata || !dev->writeable)
3502                         continue;
3503
3504                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3505                 if (ret)
3506                         total_errors++;
3507         }
3508         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3509         if (total_errors > max_errors) {
3510                 btrfs_error(root->fs_info, -EIO,
3511                             "%d errors while writing supers", total_errors);
3512                 return -EIO;
3513         }
3514         return 0;
3515 }
3516
3517 int write_ctree_super(struct btrfs_trans_handle *trans,
3518                       struct btrfs_root *root, int max_mirrors)
3519 {
3520         return write_all_supers(root, max_mirrors);
3521 }
3522
3523 /* Drop a fs root from the radix tree and free it. */
3524 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3525                                   struct btrfs_root *root)
3526 {
3527         spin_lock(&fs_info->fs_roots_radix_lock);
3528         radix_tree_delete(&fs_info->fs_roots_radix,
3529                           (unsigned long)root->root_key.objectid);
3530         spin_unlock(&fs_info->fs_roots_radix_lock);
3531
3532         if (btrfs_root_refs(&root->root_item) == 0)
3533                 synchronize_srcu(&fs_info->subvol_srcu);
3534
3535         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3536                 btrfs_free_log(NULL, root);
3537
3538         if (root->free_ino_pinned)
3539                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3540         if (root->free_ino_ctl)
3541                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3542         free_fs_root(root);
3543 }
3544
3545 static void free_fs_root(struct btrfs_root *root)
3546 {
3547         iput(root->ino_cache_inode);
3548         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3549         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3550         root->orphan_block_rsv = NULL;
3551         if (root->anon_dev)
3552                 free_anon_bdev(root->anon_dev);
3553         if (root->subv_writers)
3554                 btrfs_free_subvolume_writers(root->subv_writers);
3555         free_extent_buffer(root->node);
3556         free_extent_buffer(root->commit_root);
3557         kfree(root->free_ino_ctl);
3558         kfree(root->free_ino_pinned);
3559         kfree(root->name);
3560         btrfs_put_fs_root(root);
3561 }
3562
3563 void btrfs_free_fs_root(struct btrfs_root *root)
3564 {
3565         free_fs_root(root);
3566 }
3567
3568 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3569 {
3570         u64 root_objectid = 0;
3571         struct btrfs_root *gang[8];
3572         int i = 0;
3573         int err = 0;
3574         unsigned int ret = 0;
3575         int index;
3576
3577         while (1) {
3578                 index = srcu_read_lock(&fs_info->subvol_srcu);
3579                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3580                                              (void **)gang, root_objectid,
3581                                              ARRAY_SIZE(gang));
3582                 if (!ret) {
3583                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3584                         break;
3585                 }
3586                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3587
3588                 for (i = 0; i < ret; i++) {
3589                         /* Avoid to grab roots in dead_roots */
3590                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3591                                 gang[i] = NULL;
3592                                 continue;
3593                         }
3594                         /* grab all the search result for later use */
3595                         gang[i] = btrfs_grab_fs_root(gang[i]);
3596                 }
3597                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3598
3599                 for (i = 0; i < ret; i++) {
3600                         if (!gang[i])
3601                                 continue;
3602                         root_objectid = gang[i]->root_key.objectid;
3603                         err = btrfs_orphan_cleanup(gang[i]);
3604                         if (err)
3605                                 break;
3606                         btrfs_put_fs_root(gang[i]);
3607                 }
3608                 root_objectid++;
3609         }
3610
3611         /* release the uncleaned roots due to error */
3612         for (; i < ret; i++) {
3613                 if (gang[i])
3614                         btrfs_put_fs_root(gang[i]);
3615         }
3616         return err;
3617 }
3618
3619 int btrfs_commit_super(struct btrfs_root *root)
3620 {
3621         struct btrfs_trans_handle *trans;
3622
3623         mutex_lock(&root->fs_info->cleaner_mutex);
3624         btrfs_run_delayed_iputs(root);
3625         mutex_unlock(&root->fs_info->cleaner_mutex);
3626         wake_up_process(root->fs_info->cleaner_kthread);
3627
3628         /* wait until ongoing cleanup work done */
3629         down_write(&root->fs_info->cleanup_work_sem);
3630         up_write(&root->fs_info->cleanup_work_sem);
3631
3632         trans = btrfs_join_transaction(root);
3633         if (IS_ERR(trans))
3634                 return PTR_ERR(trans);
3635         return btrfs_commit_transaction(trans, root);
3636 }
3637
3638 void close_ctree(struct btrfs_root *root)
3639 {
3640         struct btrfs_fs_info *fs_info = root->fs_info;
3641         int ret;
3642
3643         fs_info->closing = 1;
3644         smp_mb();
3645
3646         /* wait for the uuid_scan task to finish */
3647         down(&fs_info->uuid_tree_rescan_sem);
3648         /* avoid complains from lockdep et al., set sem back to initial state */
3649         up(&fs_info->uuid_tree_rescan_sem);
3650
3651         /* pause restriper - we want to resume on mount */
3652         btrfs_pause_balance(fs_info);
3653
3654         btrfs_dev_replace_suspend_for_unmount(fs_info);
3655
3656         btrfs_scrub_cancel(fs_info);
3657
3658         /* wait for any defraggers to finish */
3659         wait_event(fs_info->transaction_wait,
3660                    (atomic_read(&fs_info->defrag_running) == 0));
3661
3662         /* clear out the rbtree of defraggable inodes */
3663         btrfs_cleanup_defrag_inodes(fs_info);
3664
3665         cancel_work_sync(&fs_info->async_reclaim_work);
3666
3667         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3668                 ret = btrfs_commit_super(root);
3669                 if (ret)
3670                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3671         }
3672
3673         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3674                 btrfs_error_commit_super(root);
3675
3676         kthread_stop(fs_info->transaction_kthread);
3677         kthread_stop(fs_info->cleaner_kthread);
3678
3679         fs_info->closing = 2;
3680         smp_mb();
3681
3682         btrfs_free_qgroup_config(root->fs_info);
3683
3684         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3685                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3686                        percpu_counter_sum(&fs_info->delalloc_bytes));
3687         }
3688
3689         btrfs_sysfs_remove_one(fs_info);
3690
3691         btrfs_free_fs_roots(fs_info);
3692
3693         btrfs_put_block_group_cache(fs_info);
3694
3695         btrfs_free_block_groups(fs_info);
3696
3697         /*
3698          * we must make sure there is not any read request to
3699          * submit after we stopping all workers.
3700          */
3701         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3702         btrfs_stop_all_workers(fs_info);
3703
3704         fs_info->open = 0;
3705         free_root_pointers(fs_info, 1);
3706
3707         iput(fs_info->btree_inode);
3708
3709 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3710         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3711                 btrfsic_unmount(root, fs_info->fs_devices);
3712 #endif
3713
3714         btrfs_close_devices(fs_info->fs_devices);
3715         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3716
3717         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3718         percpu_counter_destroy(&fs_info->delalloc_bytes);
3719         percpu_counter_destroy(&fs_info->bio_counter);
3720         bdi_destroy(&fs_info->bdi);
3721         cleanup_srcu_struct(&fs_info->subvol_srcu);
3722
3723         btrfs_free_stripe_hash_table(fs_info);
3724
3725         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3726         root->orphan_block_rsv = NULL;
3727
3728         lock_chunks(root);
3729         while (!list_empty(&fs_info->pinned_chunks)) {
3730                 struct extent_map *em;
3731
3732                 em = list_first_entry(&fs_info->pinned_chunks,
3733                                       struct extent_map, list);
3734                 list_del_init(&em->list);
3735                 free_extent_map(em);
3736         }
3737         unlock_chunks(root);
3738 }
3739
3740 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3741                           int atomic)
3742 {
3743         int ret;
3744         struct inode *btree_inode = buf->pages[0]->mapping->host;
3745
3746         ret = extent_buffer_uptodate(buf);
3747         if (!ret)
3748                 return ret;
3749
3750         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3751                                     parent_transid, atomic);
3752         if (ret == -EAGAIN)
3753                 return ret;
3754         return !ret;
3755 }
3756
3757 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3758 {
3759         return set_extent_buffer_uptodate(buf);
3760 }
3761
3762 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3763 {
3764         struct btrfs_root *root;
3765         u64 transid = btrfs_header_generation(buf);
3766         int was_dirty;
3767
3768 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3769         /*
3770          * This is a fast path so only do this check if we have sanity tests
3771          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3772          * outside of the sanity tests.
3773          */
3774         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3775                 return;
3776 #endif
3777         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3778         btrfs_assert_tree_locked(buf);
3779         if (transid != root->fs_info->generation)
3780                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3781                        "found %llu running %llu\n",
3782                         buf->start, transid, root->fs_info->generation);
3783         was_dirty = set_extent_buffer_dirty(buf);
3784         if (!was_dirty)
3785                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3786                                      buf->len,
3787                                      root->fs_info->dirty_metadata_batch);
3788 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3789         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3790                 btrfs_print_leaf(root, buf);
3791                 ASSERT(0);
3792         }
3793 #endif
3794 }
3795
3796 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3797                                         int flush_delayed)
3798 {
3799         /*
3800          * looks as though older kernels can get into trouble with
3801          * this code, they end up stuck in balance_dirty_pages forever
3802          */
3803         int ret;
3804
3805         if (current->flags & PF_MEMALLOC)
3806                 return;
3807
3808         if (flush_delayed)
3809                 btrfs_balance_delayed_items(root);
3810
3811         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3812                                      BTRFS_DIRTY_METADATA_THRESH);
3813         if (ret > 0) {
3814                 balance_dirty_pages_ratelimited(
3815                                    root->fs_info->btree_inode->i_mapping);
3816         }
3817         return;
3818 }
3819
3820 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3821 {
3822         __btrfs_btree_balance_dirty(root, 1);
3823 }
3824
3825 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3826 {
3827         __btrfs_btree_balance_dirty(root, 0);
3828 }
3829
3830 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3831 {
3832         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3833         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3834 }
3835
3836 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3837                               int read_only)
3838 {
3839         struct btrfs_super_block *sb = fs_info->super_copy;
3840         int ret = 0;
3841
3842         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3843                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3844                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3845                 ret = -EINVAL;
3846         }
3847         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3848                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3849                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3850                 ret = -EINVAL;
3851         }
3852         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3853                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3854                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3855                 ret = -EINVAL;
3856         }
3857
3858         /*
3859          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3860          * items yet, they'll be verified later. Issue just a warning.
3861          */
3862         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3863                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3864                                 btrfs_super_root(sb));
3865         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3866                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3867                                 btrfs_super_chunk_root(sb));
3868         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3869                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3870                                 btrfs_super_log_root(sb));
3871
3872         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3873                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3874                                 fs_info->fsid, sb->dev_item.fsid);
3875                 ret = -EINVAL;
3876         }
3877
3878         /*
3879          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3880          * done later
3881          */
3882         if (btrfs_super_num_devices(sb) > (1UL << 31))
3883                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3884                                 btrfs_super_num_devices(sb));
3885
3886         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3887                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3888                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3889                 ret = -EINVAL;
3890         }
3891
3892         /*
3893          * The generation is a global counter, we'll trust it more than the others
3894          * but it's still possible that it's the one that's wrong.
3895          */
3896         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3897                 printk(KERN_WARNING
3898                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3899                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3900         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3901             && btrfs_super_cache_generation(sb) != (u64)-1)
3902                 printk(KERN_WARNING
3903                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3904                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3905
3906         return ret;
3907 }
3908
3909 static void btrfs_error_commit_super(struct btrfs_root *root)
3910 {
3911         mutex_lock(&root->fs_info->cleaner_mutex);
3912         btrfs_run_delayed_iputs(root);
3913         mutex_unlock(&root->fs_info->cleaner_mutex);
3914
3915         down_write(&root->fs_info->cleanup_work_sem);
3916         up_write(&root->fs_info->cleanup_work_sem);
3917
3918         /* cleanup FS via transaction */
3919         btrfs_cleanup_transaction(root);
3920 }
3921
3922 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3923 {
3924         struct btrfs_ordered_extent *ordered;
3925
3926         spin_lock(&root->ordered_extent_lock);
3927         /*
3928          * This will just short circuit the ordered completion stuff which will
3929          * make sure the ordered extent gets properly cleaned up.
3930          */
3931         list_for_each_entry(ordered, &root->ordered_extents,
3932                             root_extent_list)
3933                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3934         spin_unlock(&root->ordered_extent_lock);
3935 }
3936
3937 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3938 {
3939         struct btrfs_root *root;
3940         struct list_head splice;
3941
3942         INIT_LIST_HEAD(&splice);
3943
3944         spin_lock(&fs_info->ordered_root_lock);
3945         list_splice_init(&fs_info->ordered_roots, &splice);
3946         while (!list_empty(&splice)) {
3947                 root = list_first_entry(&splice, struct btrfs_root,
3948                                         ordered_root);
3949                 list_move_tail(&root->ordered_root,
3950                                &fs_info->ordered_roots);
3951
3952                 spin_unlock(&fs_info->ordered_root_lock);
3953                 btrfs_destroy_ordered_extents(root);
3954
3955                 cond_resched();
3956                 spin_lock(&fs_info->ordered_root_lock);
3957         }
3958         spin_unlock(&fs_info->ordered_root_lock);
3959 }
3960
3961 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3962                                       struct btrfs_root *root)
3963 {
3964         struct rb_node *node;
3965         struct btrfs_delayed_ref_root *delayed_refs;
3966         struct btrfs_delayed_ref_node *ref;
3967         int ret = 0;
3968
3969         delayed_refs = &trans->delayed_refs;
3970
3971         spin_lock(&delayed_refs->lock);
3972         if (atomic_read(&delayed_refs->num_entries) == 0) {
3973                 spin_unlock(&delayed_refs->lock);
3974                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3975                 return ret;
3976         }
3977
3978         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3979                 struct btrfs_delayed_ref_head *head;
3980                 bool pin_bytes = false;
3981
3982                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3983                                 href_node);
3984                 if (!mutex_trylock(&head->mutex)) {
3985                         atomic_inc(&head->node.refs);
3986                         spin_unlock(&delayed_refs->lock);
3987
3988                         mutex_lock(&head->mutex);
3989                         mutex_unlock(&head->mutex);
3990                         btrfs_put_delayed_ref(&head->node);
3991                         spin_lock(&delayed_refs->lock);
3992                         continue;
3993                 }
3994                 spin_lock(&head->lock);
3995                 while ((node = rb_first(&head->ref_root)) != NULL) {
3996                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3997                                        rb_node);
3998                         ref->in_tree = 0;
3999                         rb_erase(&ref->rb_node, &head->ref_root);
4000                         atomic_dec(&delayed_refs->num_entries);
4001                         btrfs_put_delayed_ref(ref);
4002                 }
4003                 if (head->must_insert_reserved)
4004                         pin_bytes = true;
4005                 btrfs_free_delayed_extent_op(head->extent_op);
4006                 delayed_refs->num_heads--;
4007                 if (head->processing == 0)
4008                         delayed_refs->num_heads_ready--;
4009                 atomic_dec(&delayed_refs->num_entries);
4010                 head->node.in_tree = 0;
4011                 rb_erase(&head->href_node, &delayed_refs->href_root);
4012                 spin_unlock(&head->lock);
4013                 spin_unlock(&delayed_refs->lock);
4014                 mutex_unlock(&head->mutex);
4015
4016                 if (pin_bytes)
4017                         btrfs_pin_extent(root, head->node.bytenr,
4018                                          head->node.num_bytes, 1);
4019                 btrfs_put_delayed_ref(&head->node);
4020                 cond_resched();
4021                 spin_lock(&delayed_refs->lock);
4022         }
4023
4024         spin_unlock(&delayed_refs->lock);
4025
4026         return ret;
4027 }
4028
4029 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4030 {
4031         struct btrfs_inode *btrfs_inode;
4032         struct list_head splice;
4033
4034         INIT_LIST_HEAD(&splice);
4035
4036         spin_lock(&root->delalloc_lock);
4037         list_splice_init(&root->delalloc_inodes, &splice);
4038
4039         while (!list_empty(&splice)) {
4040                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4041                                                delalloc_inodes);
4042
4043                 list_del_init(&btrfs_inode->delalloc_inodes);
4044                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4045                           &btrfs_inode->runtime_flags);
4046                 spin_unlock(&root->delalloc_lock);
4047
4048                 btrfs_invalidate_inodes(btrfs_inode->root);
4049
4050                 spin_lock(&root->delalloc_lock);
4051         }
4052
4053         spin_unlock(&root->delalloc_lock);
4054 }
4055
4056 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4057 {
4058         struct btrfs_root *root;
4059         struct list_head splice;
4060
4061         INIT_LIST_HEAD(&splice);
4062
4063         spin_lock(&fs_info->delalloc_root_lock);
4064         list_splice_init(&fs_info->delalloc_roots, &splice);
4065         while (!list_empty(&splice)) {
4066                 root = list_first_entry(&splice, struct btrfs_root,
4067                                          delalloc_root);
4068                 list_del_init(&root->delalloc_root);
4069                 root = btrfs_grab_fs_root(root);
4070                 BUG_ON(!root);
4071                 spin_unlock(&fs_info->delalloc_root_lock);
4072
4073                 btrfs_destroy_delalloc_inodes(root);
4074                 btrfs_put_fs_root(root);
4075
4076                 spin_lock(&fs_info->delalloc_root_lock);
4077         }
4078         spin_unlock(&fs_info->delalloc_root_lock);
4079 }
4080
4081 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4082                                         struct extent_io_tree *dirty_pages,
4083                                         int mark)
4084 {
4085         int ret;
4086         struct extent_buffer *eb;
4087         u64 start = 0;
4088         u64 end;
4089
4090         while (1) {
4091                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4092                                             mark, NULL);
4093                 if (ret)
4094                         break;
4095
4096                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4097                 while (start <= end) {
4098                         eb = btrfs_find_tree_block(root, start);
4099                         start += root->nodesize;
4100                         if (!eb)
4101                                 continue;
4102                         wait_on_extent_buffer_writeback(eb);
4103
4104                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4105                                                &eb->bflags))
4106                                 clear_extent_buffer_dirty(eb);
4107                         free_extent_buffer_stale(eb);
4108                 }
4109         }
4110
4111         return ret;
4112 }
4113
4114 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4115                                        struct extent_io_tree *pinned_extents)
4116 {
4117         struct extent_io_tree *unpin;
4118         u64 start;
4119         u64 end;
4120         int ret;
4121         bool loop = true;
4122
4123         unpin = pinned_extents;
4124 again:
4125         while (1) {
4126                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4127                                             EXTENT_DIRTY, NULL);
4128                 if (ret)
4129                         break;
4130
4131                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4132                 btrfs_error_unpin_extent_range(root, start, end);
4133                 cond_resched();
4134         }
4135
4136         if (loop) {
4137                 if (unpin == &root->fs_info->freed_extents[0])
4138                         unpin = &root->fs_info->freed_extents[1];
4139                 else
4140                         unpin = &root->fs_info->freed_extents[0];
4141                 loop = false;
4142                 goto again;
4143         }
4144
4145         return 0;
4146 }
4147
4148 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4149                                        struct btrfs_fs_info *fs_info)
4150 {
4151         struct btrfs_ordered_extent *ordered;
4152
4153         spin_lock(&fs_info->trans_lock);
4154         while (!list_empty(&cur_trans->pending_ordered)) {
4155                 ordered = list_first_entry(&cur_trans->pending_ordered,
4156                                            struct btrfs_ordered_extent,
4157                                            trans_list);
4158                 list_del_init(&ordered->trans_list);
4159                 spin_unlock(&fs_info->trans_lock);
4160
4161                 btrfs_put_ordered_extent(ordered);
4162                 spin_lock(&fs_info->trans_lock);
4163         }
4164         spin_unlock(&fs_info->trans_lock);
4165 }
4166
4167 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4168                                    struct btrfs_root *root)
4169 {
4170         btrfs_destroy_delayed_refs(cur_trans, root);
4171
4172         cur_trans->state = TRANS_STATE_COMMIT_START;
4173         wake_up(&root->fs_info->transaction_blocked_wait);
4174
4175         cur_trans->state = TRANS_STATE_UNBLOCKED;
4176         wake_up(&root->fs_info->transaction_wait);
4177
4178         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4179         btrfs_destroy_delayed_inodes(root);
4180         btrfs_assert_delayed_root_empty(root);
4181
4182         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4183                                      EXTENT_DIRTY);
4184         btrfs_destroy_pinned_extent(root,
4185                                     root->fs_info->pinned_extents);
4186
4187         cur_trans->state =TRANS_STATE_COMPLETED;
4188         wake_up(&cur_trans->commit_wait);
4189
4190         /*
4191         memset(cur_trans, 0, sizeof(*cur_trans));
4192         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4193         */
4194 }
4195
4196 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4197 {
4198         struct btrfs_transaction *t;
4199
4200         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4201
4202         spin_lock(&root->fs_info->trans_lock);
4203         while (!list_empty(&root->fs_info->trans_list)) {
4204                 t = list_first_entry(&root->fs_info->trans_list,
4205                                      struct btrfs_transaction, list);
4206                 if (t->state >= TRANS_STATE_COMMIT_START) {
4207                         atomic_inc(&t->use_count);
4208                         spin_unlock(&root->fs_info->trans_lock);
4209                         btrfs_wait_for_commit(root, t->transid);
4210                         btrfs_put_transaction(t);
4211                         spin_lock(&root->fs_info->trans_lock);
4212                         continue;
4213                 }
4214                 if (t == root->fs_info->running_transaction) {
4215                         t->state = TRANS_STATE_COMMIT_DOING;
4216                         spin_unlock(&root->fs_info->trans_lock);
4217                         /*
4218                          * We wait for 0 num_writers since we don't hold a trans
4219                          * handle open currently for this transaction.
4220                          */
4221                         wait_event(t->writer_wait,
4222                                    atomic_read(&t->num_writers) == 0);
4223                 } else {
4224                         spin_unlock(&root->fs_info->trans_lock);
4225                 }
4226                 btrfs_cleanup_one_transaction(t, root);
4227
4228                 spin_lock(&root->fs_info->trans_lock);
4229                 if (t == root->fs_info->running_transaction)
4230                         root->fs_info->running_transaction = NULL;
4231                 list_del_init(&t->list);
4232                 spin_unlock(&root->fs_info->trans_lock);
4233
4234                 btrfs_put_transaction(t);
4235                 trace_btrfs_transaction_commit(root);
4236                 spin_lock(&root->fs_info->trans_lock);
4237         }
4238         spin_unlock(&root->fs_info->trans_lock);
4239         btrfs_destroy_all_ordered_extents(root->fs_info);
4240         btrfs_destroy_delayed_inodes(root);
4241         btrfs_assert_delayed_root_empty(root);
4242         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4243         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4244         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4245
4246         return 0;
4247 }
4248
4249 static struct extent_io_ops btree_extent_io_ops = {
4250         .readpage_end_io_hook = btree_readpage_end_io_hook,
4251         .readpage_io_failed_hook = btree_io_failed_hook,
4252         .submit_bio_hook = btree_submit_bio_hook,
4253         /* note we're sharing with inode.c for the merge bio hook */
4254         .merge_bio_hook = btrfs_merge_bio_hook,
4255 };