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         int ret;
1634
1635         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1636                 return fs_info->tree_root;
1637         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1638                 return fs_info->extent_root;
1639         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1640                 return fs_info->chunk_root;
1641         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1642                 return fs_info->dev_root;
1643         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1644                 return fs_info->csum_root;
1645         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1646                 return fs_info->quota_root ? fs_info->quota_root :
1647                                              ERR_PTR(-ENOENT);
1648         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1649                 return fs_info->uuid_root ? fs_info->uuid_root :
1650                                             ERR_PTR(-ENOENT);
1651 again:
1652         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1653         if (root) {
1654                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1655                         return ERR_PTR(-ENOENT);
1656                 return root;
1657         }
1658
1659         root = btrfs_read_fs_root(fs_info->tree_root, location);
1660         if (IS_ERR(root))
1661                 return root;
1662
1663         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1664                 ret = -ENOENT;
1665                 goto fail;
1666         }
1667
1668         ret = btrfs_init_fs_root(root);
1669         if (ret)
1670                 goto fail;
1671
1672         ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1673                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1674         if (ret < 0)
1675                 goto fail;
1676         if (ret == 0)
1677                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1678
1679         ret = btrfs_insert_fs_root(fs_info, root);
1680         if (ret) {
1681                 if (ret == -EEXIST) {
1682                         free_fs_root(root);
1683                         goto again;
1684                 }
1685                 goto fail;
1686         }
1687         return root;
1688 fail:
1689         free_fs_root(root);
1690         return ERR_PTR(ret);
1691 }
1692
1693 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1694 {
1695         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1696         int ret = 0;
1697         struct btrfs_device *device;
1698         struct backing_dev_info *bdi;
1699
1700         rcu_read_lock();
1701         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1702                 if (!device->bdev)
1703                         continue;
1704                 bdi = blk_get_backing_dev_info(device->bdev);
1705                 if (bdi_congested(bdi, bdi_bits)) {
1706                         ret = 1;
1707                         break;
1708                 }
1709         }
1710         rcu_read_unlock();
1711         return ret;
1712 }
1713
1714 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1715 {
1716         int err;
1717
1718         bdi->capabilities = BDI_CAP_MAP_COPY;
1719         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1720         if (err)
1721                 return err;
1722
1723         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1724         bdi->congested_fn       = btrfs_congested_fn;
1725         bdi->congested_data     = info;
1726         return 0;
1727 }
1728
1729 /*
1730  * called by the kthread helper functions to finally call the bio end_io
1731  * functions.  This is where read checksum verification actually happens
1732  */
1733 static void end_workqueue_fn(struct btrfs_work *work)
1734 {
1735         struct bio *bio;
1736         struct btrfs_end_io_wq *end_io_wq;
1737         int error;
1738
1739         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1740         bio = end_io_wq->bio;
1741
1742         error = end_io_wq->error;
1743         bio->bi_private = end_io_wq->private;
1744         bio->bi_end_io = end_io_wq->end_io;
1745         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1746         bio_endio_nodec(bio, error);
1747 }
1748
1749 static int cleaner_kthread(void *arg)
1750 {
1751         struct btrfs_root *root = arg;
1752         int again;
1753
1754         do {
1755                 again = 0;
1756
1757                 /* Make the cleaner go to sleep early. */
1758                 if (btrfs_need_cleaner_sleep(root))
1759                         goto sleep;
1760
1761                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1762                         goto sleep;
1763
1764                 /*
1765                  * Avoid the problem that we change the status of the fs
1766                  * during the above check and trylock.
1767                  */
1768                 if (btrfs_need_cleaner_sleep(root)) {
1769                         mutex_unlock(&root->fs_info->cleaner_mutex);
1770                         goto sleep;
1771                 }
1772
1773                 btrfs_run_delayed_iputs(root);
1774                 btrfs_delete_unused_bgs(root->fs_info);
1775                 again = btrfs_clean_one_deleted_snapshot(root);
1776                 mutex_unlock(&root->fs_info->cleaner_mutex);
1777
1778                 /*
1779                  * The defragger has dealt with the R/O remount and umount,
1780                  * needn't do anything special here.
1781                  */
1782                 btrfs_run_defrag_inodes(root->fs_info);
1783 sleep:
1784                 if (!try_to_freeze() && !again) {
1785                         set_current_state(TASK_INTERRUPTIBLE);
1786                         if (!kthread_should_stop())
1787                                 schedule();
1788                         __set_current_state(TASK_RUNNING);
1789                 }
1790         } while (!kthread_should_stop());
1791         return 0;
1792 }
1793
1794 static int transaction_kthread(void *arg)
1795 {
1796         struct btrfs_root *root = arg;
1797         struct btrfs_trans_handle *trans;
1798         struct btrfs_transaction *cur;
1799         u64 transid;
1800         unsigned long now;
1801         unsigned long delay;
1802         bool cannot_commit;
1803
1804         do {
1805                 cannot_commit = false;
1806                 delay = HZ * root->fs_info->commit_interval;
1807                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1808
1809                 spin_lock(&root->fs_info->trans_lock);
1810                 cur = root->fs_info->running_transaction;
1811                 if (!cur) {
1812                         spin_unlock(&root->fs_info->trans_lock);
1813                         goto sleep;
1814                 }
1815
1816                 now = get_seconds();
1817                 if (cur->state < TRANS_STATE_BLOCKED &&
1818                     (now < cur->start_time ||
1819                      now - cur->start_time < root->fs_info->commit_interval)) {
1820                         spin_unlock(&root->fs_info->trans_lock);
1821                         delay = HZ * 5;
1822                         goto sleep;
1823                 }
1824                 transid = cur->transid;
1825                 spin_unlock(&root->fs_info->trans_lock);
1826
1827                 /* If the file system is aborted, this will always fail. */
1828                 trans = btrfs_attach_transaction(root);
1829                 if (IS_ERR(trans)) {
1830                         if (PTR_ERR(trans) != -ENOENT)
1831                                 cannot_commit = true;
1832                         goto sleep;
1833                 }
1834                 if (transid == trans->transid) {
1835                         btrfs_commit_transaction(trans, root);
1836                 } else {
1837                         btrfs_end_transaction(trans, root);
1838                 }
1839 sleep:
1840                 wake_up_process(root->fs_info->cleaner_kthread);
1841                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1842
1843                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1844                                       &root->fs_info->fs_state)))
1845                         btrfs_cleanup_transaction(root);
1846                 if (!try_to_freeze()) {
1847                         set_current_state(TASK_INTERRUPTIBLE);
1848                         if (!kthread_should_stop() &&
1849                             (!btrfs_transaction_blocked(root->fs_info) ||
1850                              cannot_commit))
1851                                 schedule_timeout(delay);
1852                         __set_current_state(TASK_RUNNING);
1853                 }
1854         } while (!kthread_should_stop());
1855         return 0;
1856 }
1857
1858 /*
1859  * this will find the highest generation in the array of
1860  * root backups.  The index of the highest array is returned,
1861  * or -1 if we can't find anything.
1862  *
1863  * We check to make sure the array is valid by comparing the
1864  * generation of the latest  root in the array with the generation
1865  * in the super block.  If they don't match we pitch it.
1866  */
1867 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1868 {
1869         u64 cur;
1870         int newest_index = -1;
1871         struct btrfs_root_backup *root_backup;
1872         int i;
1873
1874         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1875                 root_backup = info->super_copy->super_roots + i;
1876                 cur = btrfs_backup_tree_root_gen(root_backup);
1877                 if (cur == newest_gen)
1878                         newest_index = i;
1879         }
1880
1881         /* check to see if we actually wrapped around */
1882         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1883                 root_backup = info->super_copy->super_roots;
1884                 cur = btrfs_backup_tree_root_gen(root_backup);
1885                 if (cur == newest_gen)
1886                         newest_index = 0;
1887         }
1888         return newest_index;
1889 }
1890
1891
1892 /*
1893  * find the oldest backup so we know where to store new entries
1894  * in the backup array.  This will set the backup_root_index
1895  * field in the fs_info struct
1896  */
1897 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1898                                      u64 newest_gen)
1899 {
1900         int newest_index = -1;
1901
1902         newest_index = find_newest_super_backup(info, newest_gen);
1903         /* if there was garbage in there, just move along */
1904         if (newest_index == -1) {
1905                 info->backup_root_index = 0;
1906         } else {
1907                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1908         }
1909 }
1910
1911 /*
1912  * copy all the root pointers into the super backup array.
1913  * this will bump the backup pointer by one when it is
1914  * done
1915  */
1916 static void backup_super_roots(struct btrfs_fs_info *info)
1917 {
1918         int next_backup;
1919         struct btrfs_root_backup *root_backup;
1920         int last_backup;
1921
1922         next_backup = info->backup_root_index;
1923         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1924                 BTRFS_NUM_BACKUP_ROOTS;
1925
1926         /*
1927          * just overwrite the last backup if we're at the same generation
1928          * this happens only at umount
1929          */
1930         root_backup = info->super_for_commit->super_roots + last_backup;
1931         if (btrfs_backup_tree_root_gen(root_backup) ==
1932             btrfs_header_generation(info->tree_root->node))
1933                 next_backup = last_backup;
1934
1935         root_backup = info->super_for_commit->super_roots + next_backup;
1936
1937         /*
1938          * make sure all of our padding and empty slots get zero filled
1939          * regardless of which ones we use today
1940          */
1941         memset(root_backup, 0, sizeof(*root_backup));
1942
1943         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1944
1945         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1946         btrfs_set_backup_tree_root_gen(root_backup,
1947                                btrfs_header_generation(info->tree_root->node));
1948
1949         btrfs_set_backup_tree_root_level(root_backup,
1950                                btrfs_header_level(info->tree_root->node));
1951
1952         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1953         btrfs_set_backup_chunk_root_gen(root_backup,
1954                                btrfs_header_generation(info->chunk_root->node));
1955         btrfs_set_backup_chunk_root_level(root_backup,
1956                                btrfs_header_level(info->chunk_root->node));
1957
1958         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1959         btrfs_set_backup_extent_root_gen(root_backup,
1960                                btrfs_header_generation(info->extent_root->node));
1961         btrfs_set_backup_extent_root_level(root_backup,
1962                                btrfs_header_level(info->extent_root->node));
1963
1964         /*
1965          * we might commit during log recovery, which happens before we set
1966          * the fs_root.  Make sure it is valid before we fill it in.
1967          */
1968         if (info->fs_root && info->fs_root->node) {
1969                 btrfs_set_backup_fs_root(root_backup,
1970                                          info->fs_root->node->start);
1971                 btrfs_set_backup_fs_root_gen(root_backup,
1972                                btrfs_header_generation(info->fs_root->node));
1973                 btrfs_set_backup_fs_root_level(root_backup,
1974                                btrfs_header_level(info->fs_root->node));
1975         }
1976
1977         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1978         btrfs_set_backup_dev_root_gen(root_backup,
1979                                btrfs_header_generation(info->dev_root->node));
1980         btrfs_set_backup_dev_root_level(root_backup,
1981                                        btrfs_header_level(info->dev_root->node));
1982
1983         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1984         btrfs_set_backup_csum_root_gen(root_backup,
1985                                btrfs_header_generation(info->csum_root->node));
1986         btrfs_set_backup_csum_root_level(root_backup,
1987                                btrfs_header_level(info->csum_root->node));
1988
1989         btrfs_set_backup_total_bytes(root_backup,
1990                              btrfs_super_total_bytes(info->super_copy));
1991         btrfs_set_backup_bytes_used(root_backup,
1992                              btrfs_super_bytes_used(info->super_copy));
1993         btrfs_set_backup_num_devices(root_backup,
1994                              btrfs_super_num_devices(info->super_copy));
1995
1996         /*
1997          * if we don't copy this out to the super_copy, it won't get remembered
1998          * for the next commit
1999          */
2000         memcpy(&info->super_copy->super_roots,
2001                &info->super_for_commit->super_roots,
2002                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2003 }
2004
2005 /*
2006  * this copies info out of the root backup array and back into
2007  * the in-memory super block.  It is meant to help iterate through
2008  * the array, so you send it the number of backups you've already
2009  * tried and the last backup index you used.
2010  *
2011  * this returns -1 when it has tried all the backups
2012  */
2013 static noinline int next_root_backup(struct btrfs_fs_info *info,
2014                                      struct btrfs_super_block *super,
2015                                      int *num_backups_tried, int *backup_index)
2016 {
2017         struct btrfs_root_backup *root_backup;
2018         int newest = *backup_index;
2019
2020         if (*num_backups_tried == 0) {
2021                 u64 gen = btrfs_super_generation(super);
2022
2023                 newest = find_newest_super_backup(info, gen);
2024                 if (newest == -1)
2025                         return -1;
2026
2027                 *backup_index = newest;
2028                 *num_backups_tried = 1;
2029         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2030                 /* we've tried all the backups, all done */
2031                 return -1;
2032         } else {
2033                 /* jump to the next oldest backup */
2034                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2035                         BTRFS_NUM_BACKUP_ROOTS;
2036                 *backup_index = newest;
2037                 *num_backups_tried += 1;
2038         }
2039         root_backup = super->super_roots + newest;
2040
2041         btrfs_set_super_generation(super,
2042                                    btrfs_backup_tree_root_gen(root_backup));
2043         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2044         btrfs_set_super_root_level(super,
2045                                    btrfs_backup_tree_root_level(root_backup));
2046         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2047
2048         /*
2049          * fixme: the total bytes and num_devices need to match or we should
2050          * need a fsck
2051          */
2052         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2053         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2054         return 0;
2055 }
2056
2057 /* helper to cleanup workers */
2058 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2059 {
2060         btrfs_destroy_workqueue(fs_info->fixup_workers);
2061         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2062         btrfs_destroy_workqueue(fs_info->workers);
2063         btrfs_destroy_workqueue(fs_info->endio_workers);
2064         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2065         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2066         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2067         btrfs_destroy_workqueue(fs_info->rmw_workers);
2068         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2069         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2070         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2071         btrfs_destroy_workqueue(fs_info->submit_workers);
2072         btrfs_destroy_workqueue(fs_info->delayed_workers);
2073         btrfs_destroy_workqueue(fs_info->caching_workers);
2074         btrfs_destroy_workqueue(fs_info->readahead_workers);
2075         btrfs_destroy_workqueue(fs_info->flush_workers);
2076         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2077         btrfs_destroy_workqueue(fs_info->extent_workers);
2078 }
2079
2080 static void free_root_extent_buffers(struct btrfs_root *root)
2081 {
2082         if (root) {
2083                 free_extent_buffer(root->node);
2084                 free_extent_buffer(root->commit_root);
2085                 root->node = NULL;
2086                 root->commit_root = NULL;
2087         }
2088 }
2089
2090 /* helper to cleanup tree roots */
2091 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2092 {
2093         free_root_extent_buffers(info->tree_root);
2094
2095         free_root_extent_buffers(info->dev_root);
2096         free_root_extent_buffers(info->extent_root);
2097         free_root_extent_buffers(info->csum_root);
2098         free_root_extent_buffers(info->quota_root);
2099         free_root_extent_buffers(info->uuid_root);
2100         if (chunk_root)
2101                 free_root_extent_buffers(info->chunk_root);
2102 }
2103
2104 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2105 {
2106         int ret;
2107         struct btrfs_root *gang[8];
2108         int i;
2109
2110         while (!list_empty(&fs_info->dead_roots)) {
2111                 gang[0] = list_entry(fs_info->dead_roots.next,
2112                                      struct btrfs_root, root_list);
2113                 list_del(&gang[0]->root_list);
2114
2115                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2116                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2117                 } else {
2118                         free_extent_buffer(gang[0]->node);
2119                         free_extent_buffer(gang[0]->commit_root);
2120                         btrfs_put_fs_root(gang[0]);
2121                 }
2122         }
2123
2124         while (1) {
2125                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2126                                              (void **)gang, 0,
2127                                              ARRAY_SIZE(gang));
2128                 if (!ret)
2129                         break;
2130                 for (i = 0; i < ret; i++)
2131                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2132         }
2133
2134         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2135                 btrfs_free_log_root_tree(NULL, fs_info);
2136                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2137                                             fs_info->pinned_extents);
2138         }
2139 }
2140
2141 int open_ctree(struct super_block *sb,
2142                struct btrfs_fs_devices *fs_devices,
2143                char *options)
2144 {
2145         u32 sectorsize;
2146         u32 nodesize;
2147         u32 stripesize;
2148         u64 generation;
2149         u64 features;
2150         struct btrfs_key location;
2151         struct buffer_head *bh;
2152         struct btrfs_super_block *disk_super;
2153         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2154         struct btrfs_root *tree_root;
2155         struct btrfs_root *extent_root;
2156         struct btrfs_root *csum_root;
2157         struct btrfs_root *chunk_root;
2158         struct btrfs_root *dev_root;
2159         struct btrfs_root *quota_root;
2160         struct btrfs_root *uuid_root;
2161         struct btrfs_root *log_tree_root;
2162         int ret;
2163         int err = -EINVAL;
2164         int num_backups_tried = 0;
2165         int backup_index = 0;
2166         int max_active;
2167         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2168         bool create_uuid_tree;
2169         bool check_uuid_tree;
2170
2171         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2172         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2173         if (!tree_root || !chunk_root) {
2174                 err = -ENOMEM;
2175                 goto fail;
2176         }
2177
2178         ret = init_srcu_struct(&fs_info->subvol_srcu);
2179         if (ret) {
2180                 err = ret;
2181                 goto fail;
2182         }
2183
2184         ret = setup_bdi(fs_info, &fs_info->bdi);
2185         if (ret) {
2186                 err = ret;
2187                 goto fail_srcu;
2188         }
2189
2190         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2191         if (ret) {
2192                 err = ret;
2193                 goto fail_bdi;
2194         }
2195         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2196                                         (1 + ilog2(nr_cpu_ids));
2197
2198         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2199         if (ret) {
2200                 err = ret;
2201                 goto fail_dirty_metadata_bytes;
2202         }
2203
2204         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2205         if (ret) {
2206                 err = ret;
2207                 goto fail_delalloc_bytes;
2208         }
2209
2210         fs_info->btree_inode = new_inode(sb);
2211         if (!fs_info->btree_inode) {
2212                 err = -ENOMEM;
2213                 goto fail_bio_counter;
2214         }
2215
2216         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2217
2218         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2219         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2220         INIT_LIST_HEAD(&fs_info->trans_list);
2221         INIT_LIST_HEAD(&fs_info->dead_roots);
2222         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2223         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2224         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2225         spin_lock_init(&fs_info->delalloc_root_lock);
2226         spin_lock_init(&fs_info->trans_lock);
2227         spin_lock_init(&fs_info->fs_roots_radix_lock);
2228         spin_lock_init(&fs_info->delayed_iput_lock);
2229         spin_lock_init(&fs_info->defrag_inodes_lock);
2230         spin_lock_init(&fs_info->free_chunk_lock);
2231         spin_lock_init(&fs_info->tree_mod_seq_lock);
2232         spin_lock_init(&fs_info->super_lock);
2233         spin_lock_init(&fs_info->qgroup_op_lock);
2234         spin_lock_init(&fs_info->buffer_lock);
2235         spin_lock_init(&fs_info->unused_bgs_lock);
2236         rwlock_init(&fs_info->tree_mod_log_lock);
2237         mutex_init(&fs_info->reloc_mutex);
2238         mutex_init(&fs_info->delalloc_root_mutex);
2239         seqlock_init(&fs_info->profiles_lock);
2240
2241         init_completion(&fs_info->kobj_unregister);
2242         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2243         INIT_LIST_HEAD(&fs_info->space_info);
2244         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2245         INIT_LIST_HEAD(&fs_info->unused_bgs);
2246         btrfs_mapping_init(&fs_info->mapping_tree);
2247         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2248                              BTRFS_BLOCK_RSV_GLOBAL);
2249         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2250                              BTRFS_BLOCK_RSV_DELALLOC);
2251         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2252         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2253         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2254         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2255                              BTRFS_BLOCK_RSV_DELOPS);
2256         atomic_set(&fs_info->nr_async_submits, 0);
2257         atomic_set(&fs_info->async_delalloc_pages, 0);
2258         atomic_set(&fs_info->async_submit_draining, 0);
2259         atomic_set(&fs_info->nr_async_bios, 0);
2260         atomic_set(&fs_info->defrag_running, 0);
2261         atomic_set(&fs_info->qgroup_op_seq, 0);
2262         atomic64_set(&fs_info->tree_mod_seq, 0);
2263         fs_info->sb = sb;
2264         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2265         fs_info->metadata_ratio = 0;
2266         fs_info->defrag_inodes = RB_ROOT;
2267         fs_info->free_chunk_space = 0;
2268         fs_info->tree_mod_log = RB_ROOT;
2269         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2270         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2271         /* readahead state */
2272         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2273         spin_lock_init(&fs_info->reada_lock);
2274
2275         fs_info->thread_pool_size = min_t(unsigned long,
2276                                           num_online_cpus() + 2, 8);
2277
2278         INIT_LIST_HEAD(&fs_info->ordered_roots);
2279         spin_lock_init(&fs_info->ordered_root_lock);
2280         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2281                                         GFP_NOFS);
2282         if (!fs_info->delayed_root) {
2283                 err = -ENOMEM;
2284                 goto fail_iput;
2285         }
2286         btrfs_init_delayed_root(fs_info->delayed_root);
2287
2288         mutex_init(&fs_info->scrub_lock);
2289         atomic_set(&fs_info->scrubs_running, 0);
2290         atomic_set(&fs_info->scrub_pause_req, 0);
2291         atomic_set(&fs_info->scrubs_paused, 0);
2292         atomic_set(&fs_info->scrub_cancel_req, 0);
2293         init_waitqueue_head(&fs_info->replace_wait);
2294         init_waitqueue_head(&fs_info->scrub_pause_wait);
2295         fs_info->scrub_workers_refcnt = 0;
2296 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2297         fs_info->check_integrity_print_mask = 0;
2298 #endif
2299
2300         spin_lock_init(&fs_info->balance_lock);
2301         mutex_init(&fs_info->balance_mutex);
2302         atomic_set(&fs_info->balance_running, 0);
2303         atomic_set(&fs_info->balance_pause_req, 0);
2304         atomic_set(&fs_info->balance_cancel_req, 0);
2305         fs_info->balance_ctl = NULL;
2306         init_waitqueue_head(&fs_info->balance_wait_q);
2307         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2308
2309         sb->s_blocksize = 4096;
2310         sb->s_blocksize_bits = blksize_bits(4096);
2311         sb->s_bdi = &fs_info->bdi;
2312
2313         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2314         set_nlink(fs_info->btree_inode, 1);
2315         /*
2316          * we set the i_size on the btree inode to the max possible int.
2317          * the real end of the address space is determined by all of
2318          * the devices in the system
2319          */
2320         fs_info->btree_inode->i_size = OFFSET_MAX;
2321         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2322         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2323
2324         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2325         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2326                              fs_info->btree_inode->i_mapping);
2327         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2328         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2329
2330         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2331
2332         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2333         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2334                sizeof(struct btrfs_key));
2335         set_bit(BTRFS_INODE_DUMMY,
2336                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2337         btrfs_insert_inode_hash(fs_info->btree_inode);
2338
2339         spin_lock_init(&fs_info->block_group_cache_lock);
2340         fs_info->block_group_cache_tree = RB_ROOT;
2341         fs_info->first_logical_byte = (u64)-1;
2342
2343         extent_io_tree_init(&fs_info->freed_extents[0],
2344                              fs_info->btree_inode->i_mapping);
2345         extent_io_tree_init(&fs_info->freed_extents[1],
2346                              fs_info->btree_inode->i_mapping);
2347         fs_info->pinned_extents = &fs_info->freed_extents[0];
2348         fs_info->do_barriers = 1;
2349
2350
2351         mutex_init(&fs_info->ordered_operations_mutex);
2352         mutex_init(&fs_info->ordered_extent_flush_mutex);
2353         mutex_init(&fs_info->tree_log_mutex);
2354         mutex_init(&fs_info->chunk_mutex);
2355         mutex_init(&fs_info->transaction_kthread_mutex);
2356         mutex_init(&fs_info->cleaner_mutex);
2357         mutex_init(&fs_info->volume_mutex);
2358         init_rwsem(&fs_info->commit_root_sem);
2359         init_rwsem(&fs_info->cleanup_work_sem);
2360         init_rwsem(&fs_info->subvol_sem);
2361         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2362         fs_info->dev_replace.lock_owner = 0;
2363         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2364         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2365         mutex_init(&fs_info->dev_replace.lock_management_lock);
2366         mutex_init(&fs_info->dev_replace.lock);
2367
2368         spin_lock_init(&fs_info->qgroup_lock);
2369         mutex_init(&fs_info->qgroup_ioctl_lock);
2370         fs_info->qgroup_tree = RB_ROOT;
2371         fs_info->qgroup_op_tree = RB_ROOT;
2372         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2373         fs_info->qgroup_seq = 1;
2374         fs_info->quota_enabled = 0;
2375         fs_info->pending_quota_state = 0;
2376         fs_info->qgroup_ulist = NULL;
2377         mutex_init(&fs_info->qgroup_rescan_lock);
2378
2379         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2380         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2381
2382         init_waitqueue_head(&fs_info->transaction_throttle);
2383         init_waitqueue_head(&fs_info->transaction_wait);
2384         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2385         init_waitqueue_head(&fs_info->async_submit_wait);
2386
2387         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2388
2389         ret = btrfs_alloc_stripe_hash_table(fs_info);
2390         if (ret) {
2391                 err = ret;
2392                 goto fail_alloc;
2393         }
2394
2395         __setup_root(4096, 4096, 4096, tree_root,
2396                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2397
2398         invalidate_bdev(fs_devices->latest_bdev);
2399
2400         /*
2401          * Read super block and check the signature bytes only
2402          */
2403         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2404         if (!bh) {
2405                 err = -EINVAL;
2406                 goto fail_alloc;
2407         }
2408
2409         /*
2410          * We want to check superblock checksum, the type is stored inside.
2411          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2412          */
2413         if (btrfs_check_super_csum(bh->b_data)) {
2414                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2415                 err = -EINVAL;
2416                 goto fail_alloc;
2417         }
2418
2419         /*
2420          * super_copy is zeroed at allocation time and we never touch the
2421          * following bytes up to INFO_SIZE, the checksum is calculated from
2422          * the whole block of INFO_SIZE
2423          */
2424         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2425         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2426                sizeof(*fs_info->super_for_commit));
2427         brelse(bh);
2428
2429         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2430
2431         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2432         if (ret) {
2433                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2434                 err = -EINVAL;
2435                 goto fail_alloc;
2436         }
2437
2438         disk_super = fs_info->super_copy;
2439         if (!btrfs_super_root(disk_super))
2440                 goto fail_alloc;
2441
2442         /* check FS state, whether FS is broken. */
2443         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2444                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2445
2446         /*
2447          * run through our array of backup supers and setup
2448          * our ring pointer to the oldest one
2449          */
2450         generation = btrfs_super_generation(disk_super);
2451         find_oldest_super_backup(fs_info, generation);
2452
2453         /*
2454          * In the long term, we'll store the compression type in the super
2455          * block, and it'll be used for per file compression control.
2456          */
2457         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2458
2459         ret = btrfs_parse_options(tree_root, options);
2460         if (ret) {
2461                 err = ret;
2462                 goto fail_alloc;
2463         }
2464
2465         features = btrfs_super_incompat_flags(disk_super) &
2466                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2467         if (features) {
2468                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2469                        "unsupported optional features (%Lx).\n",
2470                        features);
2471                 err = -EINVAL;
2472                 goto fail_alloc;
2473         }
2474
2475         /*
2476          * Leafsize and nodesize were always equal, this is only a sanity check.
2477          */
2478         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2479             btrfs_super_nodesize(disk_super)) {
2480                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2481                        "blocksizes don't match.  node %d leaf %d\n",
2482                        btrfs_super_nodesize(disk_super),
2483                        le32_to_cpu(disk_super->__unused_leafsize));
2484                 err = -EINVAL;
2485                 goto fail_alloc;
2486         }
2487         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2488                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2489                        "blocksize (%d) was too large\n",
2490                        btrfs_super_nodesize(disk_super));
2491                 err = -EINVAL;
2492                 goto fail_alloc;
2493         }
2494
2495         features = btrfs_super_incompat_flags(disk_super);
2496         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2497         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2498                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2499
2500         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2501                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2502
2503         /*
2504          * flag our filesystem as having big metadata blocks if
2505          * they are bigger than the page size
2506          */
2507         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2508                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2509                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2510                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2511         }
2512
2513         nodesize = btrfs_super_nodesize(disk_super);
2514         sectorsize = btrfs_super_sectorsize(disk_super);
2515         stripesize = btrfs_super_stripesize(disk_super);
2516         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2517         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2518
2519         /*
2520          * mixed block groups end up with duplicate but slightly offset
2521          * extent buffers for the same range.  It leads to corruptions
2522          */
2523         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2524             (sectorsize != nodesize)) {
2525                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2526                                 "are not allowed for mixed block groups on %s\n",
2527                                 sb->s_id);
2528                 goto fail_alloc;
2529         }
2530
2531         /*
2532          * Needn't use the lock because there is no other task which will
2533          * update the flag.
2534          */
2535         btrfs_set_super_incompat_flags(disk_super, features);
2536
2537         features = btrfs_super_compat_ro_flags(disk_super) &
2538                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2539         if (!(sb->s_flags & MS_RDONLY) && features) {
2540                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2541                        "unsupported option features (%Lx).\n",
2542                        features);
2543                 err = -EINVAL;
2544                 goto fail_alloc;
2545         }
2546
2547         max_active = fs_info->thread_pool_size;
2548
2549         fs_info->workers =
2550                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2551                                       max_active, 16);
2552
2553         fs_info->delalloc_workers =
2554                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2555
2556         fs_info->flush_workers =
2557                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2558
2559         fs_info->caching_workers =
2560                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2561
2562         /*
2563          * a higher idle thresh on the submit workers makes it much more
2564          * likely that bios will be send down in a sane order to the
2565          * devices
2566          */
2567         fs_info->submit_workers =
2568                 btrfs_alloc_workqueue("submit", flags,
2569                                       min_t(u64, fs_devices->num_devices,
2570                                             max_active), 64);
2571
2572         fs_info->fixup_workers =
2573                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2574
2575         /*
2576          * endios are largely parallel and should have a very
2577          * low idle thresh
2578          */
2579         fs_info->endio_workers =
2580                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2581         fs_info->endio_meta_workers =
2582                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2583         fs_info->endio_meta_write_workers =
2584                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2585         fs_info->endio_raid56_workers =
2586                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2587         fs_info->endio_repair_workers =
2588                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2589         fs_info->rmw_workers =
2590                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2591         fs_info->endio_write_workers =
2592                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2593         fs_info->endio_freespace_worker =
2594                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2595         fs_info->delayed_workers =
2596                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2597         fs_info->readahead_workers =
2598                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2599         fs_info->qgroup_rescan_workers =
2600                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2601         fs_info->extent_workers =
2602                 btrfs_alloc_workqueue("extent-refs", flags,
2603                                       min_t(u64, fs_devices->num_devices,
2604                                             max_active), 8);
2605
2606         if (!(fs_info->workers && fs_info->delalloc_workers &&
2607               fs_info->submit_workers && fs_info->flush_workers &&
2608               fs_info->endio_workers && fs_info->endio_meta_workers &&
2609               fs_info->endio_meta_write_workers &&
2610               fs_info->endio_repair_workers &&
2611               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2612               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2613               fs_info->caching_workers && fs_info->readahead_workers &&
2614               fs_info->fixup_workers && fs_info->delayed_workers &&
2615               fs_info->extent_workers &&
2616               fs_info->qgroup_rescan_workers)) {
2617                 err = -ENOMEM;
2618                 goto fail_sb_buffer;
2619         }
2620
2621         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2622         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2623                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2624
2625         tree_root->nodesize = nodesize;
2626         tree_root->sectorsize = sectorsize;
2627         tree_root->stripesize = stripesize;
2628
2629         sb->s_blocksize = sectorsize;
2630         sb->s_blocksize_bits = blksize_bits(sectorsize);
2631
2632         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2633                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2634                 goto fail_sb_buffer;
2635         }
2636
2637         if (sectorsize != PAGE_SIZE) {
2638                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2639                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2640                 goto fail_sb_buffer;
2641         }
2642
2643         mutex_lock(&fs_info->chunk_mutex);
2644         ret = btrfs_read_sys_array(tree_root);
2645         mutex_unlock(&fs_info->chunk_mutex);
2646         if (ret) {
2647                 printk(KERN_WARNING "BTRFS: failed to read the system "
2648                        "array on %s\n", sb->s_id);
2649                 goto fail_sb_buffer;
2650         }
2651
2652         generation = btrfs_super_chunk_root_generation(disk_super);
2653
2654         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2655                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2656
2657         chunk_root->node = read_tree_block(chunk_root,
2658                                            btrfs_super_chunk_root(disk_super),
2659                                            generation);
2660         if (!chunk_root->node ||
2661             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2662                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2663                        sb->s_id);
2664                 goto fail_tree_roots;
2665         }
2666         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2667         chunk_root->commit_root = btrfs_root_node(chunk_root);
2668
2669         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2670            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2671
2672         ret = btrfs_read_chunk_tree(chunk_root);
2673         if (ret) {
2674                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2675                        sb->s_id);
2676                 goto fail_tree_roots;
2677         }
2678
2679         /*
2680          * keep the device that is marked to be the target device for the
2681          * dev_replace procedure
2682          */
2683         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2684
2685         if (!fs_devices->latest_bdev) {
2686                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2687                        sb->s_id);
2688                 goto fail_tree_roots;
2689         }
2690
2691 retry_root_backup:
2692         generation = btrfs_super_generation(disk_super);
2693
2694         tree_root->node = read_tree_block(tree_root,
2695                                           btrfs_super_root(disk_super),
2696                                           generation);
2697         if (!tree_root->node ||
2698             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2699                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2700                        sb->s_id);
2701
2702                 goto recovery_tree_root;
2703         }
2704
2705         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2706         tree_root->commit_root = btrfs_root_node(tree_root);
2707         btrfs_set_root_refs(&tree_root->root_item, 1);
2708
2709         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2710         location.type = BTRFS_ROOT_ITEM_KEY;
2711         location.offset = 0;
2712
2713         extent_root = btrfs_read_tree_root(tree_root, &location);
2714         if (IS_ERR(extent_root)) {
2715                 ret = PTR_ERR(extent_root);
2716                 goto recovery_tree_root;
2717         }
2718         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2719         fs_info->extent_root = extent_root;
2720
2721         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2722         dev_root = btrfs_read_tree_root(tree_root, &location);
2723         if (IS_ERR(dev_root)) {
2724                 ret = PTR_ERR(dev_root);
2725                 goto recovery_tree_root;
2726         }
2727         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2728         fs_info->dev_root = dev_root;
2729         btrfs_init_devices_late(fs_info);
2730
2731         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2732         csum_root = btrfs_read_tree_root(tree_root, &location);
2733         if (IS_ERR(csum_root)) {
2734                 ret = PTR_ERR(csum_root);
2735                 goto recovery_tree_root;
2736         }
2737         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2738         fs_info->csum_root = csum_root;
2739
2740         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2741         quota_root = btrfs_read_tree_root(tree_root, &location);
2742         if (!IS_ERR(quota_root)) {
2743                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2744                 fs_info->quota_enabled = 1;
2745                 fs_info->pending_quota_state = 1;
2746                 fs_info->quota_root = quota_root;
2747         }
2748
2749         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2750         uuid_root = btrfs_read_tree_root(tree_root, &location);
2751         if (IS_ERR(uuid_root)) {
2752                 ret = PTR_ERR(uuid_root);
2753                 if (ret != -ENOENT)
2754                         goto recovery_tree_root;
2755                 create_uuid_tree = true;
2756                 check_uuid_tree = false;
2757         } else {
2758                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2759                 fs_info->uuid_root = uuid_root;
2760                 create_uuid_tree = false;
2761                 check_uuid_tree =
2762                     generation != btrfs_super_uuid_tree_generation(disk_super);
2763         }
2764
2765         fs_info->generation = generation;
2766         fs_info->last_trans_committed = generation;
2767
2768         ret = btrfs_recover_balance(fs_info);
2769         if (ret) {
2770                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2771                 goto fail_block_groups;
2772         }
2773
2774         ret = btrfs_init_dev_stats(fs_info);
2775         if (ret) {
2776                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2777                        ret);
2778                 goto fail_block_groups;
2779         }
2780
2781         ret = btrfs_init_dev_replace(fs_info);
2782         if (ret) {
2783                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2784                 goto fail_block_groups;
2785         }
2786
2787         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2788
2789         ret = btrfs_sysfs_add_one(fs_info);
2790         if (ret) {
2791                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2792                 goto fail_block_groups;
2793         }
2794
2795         ret = btrfs_init_space_info(fs_info);
2796         if (ret) {
2797                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2798                 goto fail_sysfs;
2799         }
2800
2801         ret = btrfs_read_block_groups(extent_root);
2802         if (ret) {
2803                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2804                 goto fail_sysfs;
2805         }
2806         fs_info->num_tolerated_disk_barrier_failures =
2807                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2808         if (fs_info->fs_devices->missing_devices >
2809              fs_info->num_tolerated_disk_barrier_failures &&
2810             !(sb->s_flags & MS_RDONLY)) {
2811                 printk(KERN_WARNING "BTRFS: "
2812                         "too many missing devices, writeable mount is not allowed\n");
2813                 goto fail_sysfs;
2814         }
2815
2816         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2817                                                "btrfs-cleaner");
2818         if (IS_ERR(fs_info->cleaner_kthread))
2819                 goto fail_sysfs;
2820
2821         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2822                                                    tree_root,
2823                                                    "btrfs-transaction");
2824         if (IS_ERR(fs_info->transaction_kthread))
2825                 goto fail_cleaner;
2826
2827         if (!btrfs_test_opt(tree_root, SSD) &&
2828             !btrfs_test_opt(tree_root, NOSSD) &&
2829             !fs_info->fs_devices->rotating) {
2830                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2831                        "mode\n");
2832                 btrfs_set_opt(fs_info->mount_opt, SSD);
2833         }
2834
2835         /*
2836          * Mount does not set all options immediatelly, we can do it now and do
2837          * not have to wait for transaction commit
2838          */
2839         btrfs_apply_pending_changes(fs_info);
2840
2841 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2842         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2843                 ret = btrfsic_mount(tree_root, fs_devices,
2844                                     btrfs_test_opt(tree_root,
2845                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2846                                     1 : 0,
2847                                     fs_info->check_integrity_print_mask);
2848                 if (ret)
2849                         printk(KERN_WARNING "BTRFS: failed to initialize"
2850                                " integrity check module %s\n", sb->s_id);
2851         }
2852 #endif
2853         ret = btrfs_read_qgroup_config(fs_info);
2854         if (ret)
2855                 goto fail_trans_kthread;
2856
2857         /* do not make disk changes in broken FS */
2858         if (btrfs_super_log_root(disk_super) != 0) {
2859                 u64 bytenr = btrfs_super_log_root(disk_super);
2860
2861                 if (fs_devices->rw_devices == 0) {
2862                         printk(KERN_WARNING "BTRFS: log replay required "
2863                                "on RO media\n");
2864                         err = -EIO;
2865                         goto fail_qgroup;
2866                 }
2867
2868                 log_tree_root = btrfs_alloc_root(fs_info);
2869                 if (!log_tree_root) {
2870                         err = -ENOMEM;
2871                         goto fail_qgroup;
2872                 }
2873
2874                 __setup_root(nodesize, sectorsize, stripesize,
2875                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2876
2877                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2878                                                       generation + 1);
2879                 if (!log_tree_root->node ||
2880                     !extent_buffer_uptodate(log_tree_root->node)) {
2881                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2882                         free_extent_buffer(log_tree_root->node);
2883                         kfree(log_tree_root);
2884                         goto fail_qgroup;
2885                 }
2886                 /* returns with log_tree_root freed on success */
2887                 ret = btrfs_recover_log_trees(log_tree_root);
2888                 if (ret) {
2889                         btrfs_error(tree_root->fs_info, ret,
2890                                     "Failed to recover log tree");
2891                         free_extent_buffer(log_tree_root->node);
2892                         kfree(log_tree_root);
2893                         goto fail_qgroup;
2894                 }
2895
2896                 if (sb->s_flags & MS_RDONLY) {
2897                         ret = btrfs_commit_super(tree_root);
2898                         if (ret)
2899                                 goto fail_qgroup;
2900                 }
2901         }
2902
2903         ret = btrfs_find_orphan_roots(tree_root);
2904         if (ret)
2905                 goto fail_qgroup;
2906
2907         if (!(sb->s_flags & MS_RDONLY)) {
2908                 ret = btrfs_cleanup_fs_roots(fs_info);
2909                 if (ret)
2910                         goto fail_qgroup;
2911
2912                 mutex_lock(&fs_info->cleaner_mutex);
2913                 ret = btrfs_recover_relocation(tree_root);
2914                 mutex_unlock(&fs_info->cleaner_mutex);
2915                 if (ret < 0) {
2916                         printk(KERN_WARNING
2917                                "BTRFS: failed to recover relocation\n");
2918                         err = -EINVAL;
2919                         goto fail_qgroup;
2920                 }
2921         }
2922
2923         location.objectid = BTRFS_FS_TREE_OBJECTID;
2924         location.type = BTRFS_ROOT_ITEM_KEY;
2925         location.offset = 0;
2926
2927         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2928         if (IS_ERR(fs_info->fs_root)) {
2929                 err = PTR_ERR(fs_info->fs_root);
2930                 goto fail_qgroup;
2931         }
2932
2933         if (sb->s_flags & MS_RDONLY)
2934                 return 0;
2935
2936         down_read(&fs_info->cleanup_work_sem);
2937         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2938             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2939                 up_read(&fs_info->cleanup_work_sem);
2940                 close_ctree(tree_root);
2941                 return ret;
2942         }
2943         up_read(&fs_info->cleanup_work_sem);
2944
2945         ret = btrfs_resume_balance_async(fs_info);
2946         if (ret) {
2947                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2948                 close_ctree(tree_root);
2949                 return ret;
2950         }
2951
2952         ret = btrfs_resume_dev_replace_async(fs_info);
2953         if (ret) {
2954                 pr_warn("BTRFS: failed to resume dev_replace\n");
2955                 close_ctree(tree_root);
2956                 return ret;
2957         }
2958
2959         btrfs_qgroup_rescan_resume(fs_info);
2960
2961         if (create_uuid_tree) {
2962                 pr_info("BTRFS: creating UUID tree\n");
2963                 ret = btrfs_create_uuid_tree(fs_info);
2964                 if (ret) {
2965                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2966                                 ret);
2967                         close_ctree(tree_root);
2968                         return ret;
2969                 }
2970         } else if (check_uuid_tree ||
2971                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2972                 pr_info("BTRFS: checking UUID tree\n");
2973                 ret = btrfs_check_uuid_tree(fs_info);
2974                 if (ret) {
2975                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2976                                 ret);
2977                         close_ctree(tree_root);
2978                         return ret;
2979                 }
2980         } else {
2981                 fs_info->update_uuid_tree_gen = 1;
2982         }
2983
2984         fs_info->open = 1;
2985
2986         return 0;
2987
2988 fail_qgroup:
2989         btrfs_free_qgroup_config(fs_info);
2990 fail_trans_kthread:
2991         kthread_stop(fs_info->transaction_kthread);
2992         btrfs_cleanup_transaction(fs_info->tree_root);
2993         btrfs_free_fs_roots(fs_info);
2994 fail_cleaner:
2995         kthread_stop(fs_info->cleaner_kthread);
2996
2997         /*
2998          * make sure we're done with the btree inode before we stop our
2999          * kthreads
3000          */
3001         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3002
3003 fail_sysfs:
3004         btrfs_sysfs_remove_one(fs_info);
3005
3006 fail_block_groups:
3007         btrfs_put_block_group_cache(fs_info);
3008         btrfs_free_block_groups(fs_info);
3009
3010 fail_tree_roots:
3011         free_root_pointers(fs_info, 1);
3012         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3013
3014 fail_sb_buffer:
3015         btrfs_stop_all_workers(fs_info);
3016 fail_alloc:
3017 fail_iput:
3018         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3019
3020         iput(fs_info->btree_inode);
3021 fail_bio_counter:
3022         percpu_counter_destroy(&fs_info->bio_counter);
3023 fail_delalloc_bytes:
3024         percpu_counter_destroy(&fs_info->delalloc_bytes);
3025 fail_dirty_metadata_bytes:
3026         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3027 fail_bdi:
3028         bdi_destroy(&fs_info->bdi);
3029 fail_srcu:
3030         cleanup_srcu_struct(&fs_info->subvol_srcu);
3031 fail:
3032         btrfs_free_stripe_hash_table(fs_info);
3033         btrfs_close_devices(fs_info->fs_devices);
3034         return err;
3035
3036 recovery_tree_root:
3037         if (!btrfs_test_opt(tree_root, RECOVERY))
3038                 goto fail_tree_roots;
3039
3040         free_root_pointers(fs_info, 0);
3041
3042         /* don't use the log in recovery mode, it won't be valid */
3043         btrfs_set_super_log_root(disk_super, 0);
3044
3045         /* we can't trust the free space cache either */
3046         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3047
3048         ret = next_root_backup(fs_info, fs_info->super_copy,
3049                                &num_backups_tried, &backup_index);
3050         if (ret == -1)
3051                 goto fail_block_groups;
3052         goto retry_root_backup;
3053 }
3054
3055 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3056 {
3057         if (uptodate) {
3058                 set_buffer_uptodate(bh);
3059         } else {
3060                 struct btrfs_device *device = (struct btrfs_device *)
3061                         bh->b_private;
3062
3063                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3064                                           "I/O error on %s\n",
3065                                           rcu_str_deref(device->name));
3066                 /* note, we dont' set_buffer_write_io_error because we have
3067                  * our own ways of dealing with the IO errors
3068                  */
3069                 clear_buffer_uptodate(bh);
3070                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3071         }
3072         unlock_buffer(bh);
3073         put_bh(bh);
3074 }
3075
3076 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3077 {
3078         struct buffer_head *bh;
3079         struct buffer_head *latest = NULL;
3080         struct btrfs_super_block *super;
3081         int i;
3082         u64 transid = 0;
3083         u64 bytenr;
3084
3085         /* we would like to check all the supers, but that would make
3086          * a btrfs mount succeed after a mkfs from a different FS.
3087          * So, we need to add a special mount option to scan for
3088          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3089          */
3090         for (i = 0; i < 1; i++) {
3091                 bytenr = btrfs_sb_offset(i);
3092                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3093                                         i_size_read(bdev->bd_inode))
3094                         break;
3095                 bh = __bread(bdev, bytenr / 4096,
3096                                         BTRFS_SUPER_INFO_SIZE);
3097                 if (!bh)
3098                         continue;
3099
3100                 super = (struct btrfs_super_block *)bh->b_data;
3101                 if (btrfs_super_bytenr(super) != bytenr ||
3102                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3103                         brelse(bh);
3104                         continue;
3105                 }
3106
3107                 if (!latest || btrfs_super_generation(super) > transid) {
3108                         brelse(latest);
3109                         latest = bh;
3110                         transid = btrfs_super_generation(super);
3111                 } else {
3112                         brelse(bh);
3113                 }
3114         }
3115         return latest;
3116 }
3117
3118 /*
3119  * this should be called twice, once with wait == 0 and
3120  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3121  * we write are pinned.
3122  *
3123  * They are released when wait == 1 is done.
3124  * max_mirrors must be the same for both runs, and it indicates how
3125  * many supers on this one device should be written.
3126  *
3127  * max_mirrors == 0 means to write them all.
3128  */
3129 static int write_dev_supers(struct btrfs_device *device,
3130                             struct btrfs_super_block *sb,
3131                             int do_barriers, int wait, int max_mirrors)
3132 {
3133         struct buffer_head *bh;
3134         int i;
3135         int ret;
3136         int errors = 0;
3137         u32 crc;
3138         u64 bytenr;
3139
3140         if (max_mirrors == 0)
3141                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3142
3143         for (i = 0; i < max_mirrors; i++) {
3144                 bytenr = btrfs_sb_offset(i);
3145                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3146                     device->commit_total_bytes)
3147                         break;
3148
3149                 if (wait) {
3150                         bh = __find_get_block(device->bdev, bytenr / 4096,
3151                                               BTRFS_SUPER_INFO_SIZE);
3152                         if (!bh) {
3153                                 errors++;
3154                                 continue;
3155                         }
3156                         wait_on_buffer(bh);
3157                         if (!buffer_uptodate(bh))
3158                                 errors++;
3159
3160                         /* drop our reference */
3161                         brelse(bh);
3162
3163                         /* drop the reference from the wait == 0 run */
3164                         brelse(bh);
3165                         continue;
3166                 } else {
3167                         btrfs_set_super_bytenr(sb, bytenr);
3168
3169                         crc = ~(u32)0;
3170                         crc = btrfs_csum_data((char *)sb +
3171                                               BTRFS_CSUM_SIZE, crc,
3172                                               BTRFS_SUPER_INFO_SIZE -
3173                                               BTRFS_CSUM_SIZE);
3174                         btrfs_csum_final(crc, sb->csum);
3175
3176                         /*
3177                          * one reference for us, and we leave it for the
3178                          * caller
3179                          */
3180                         bh = __getblk(device->bdev, bytenr / 4096,
3181                                       BTRFS_SUPER_INFO_SIZE);
3182                         if (!bh) {
3183                                 printk(KERN_ERR "BTRFS: couldn't get super "
3184                                        "buffer head for bytenr %Lu\n", bytenr);
3185                                 errors++;
3186                                 continue;
3187                         }
3188
3189                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3190
3191                         /* one reference for submit_bh */
3192                         get_bh(bh);
3193
3194                         set_buffer_uptodate(bh);
3195                         lock_buffer(bh);
3196                         bh->b_end_io = btrfs_end_buffer_write_sync;
3197                         bh->b_private = device;
3198                 }
3199
3200                 /*
3201                  * we fua the first super.  The others we allow
3202                  * to go down lazy.
3203                  */
3204                 if (i == 0)
3205                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3206                 else
3207                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3208                 if (ret)
3209                         errors++;
3210         }
3211         return errors < i ? 0 : -1;
3212 }
3213
3214 /*
3215  * endio for the write_dev_flush, this will wake anyone waiting
3216  * for the barrier when it is done
3217  */
3218 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3219 {
3220         if (err) {
3221                 if (err == -EOPNOTSUPP)
3222                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3223                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3224         }
3225         if (bio->bi_private)
3226                 complete(bio->bi_private);
3227         bio_put(bio);
3228 }
3229
3230 /*
3231  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3232  * sent down.  With wait == 1, it waits for the previous flush.
3233  *
3234  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3235  * capable
3236  */
3237 static int write_dev_flush(struct btrfs_device *device, int wait)
3238 {
3239         struct bio *bio;
3240         int ret = 0;
3241
3242         if (device->nobarriers)
3243                 return 0;
3244
3245         if (wait) {
3246                 bio = device->flush_bio;
3247                 if (!bio)
3248                         return 0;
3249
3250                 wait_for_completion(&device->flush_wait);
3251
3252                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3253                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3254                                       rcu_str_deref(device->name));
3255                         device->nobarriers = 1;
3256                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3257                         ret = -EIO;
3258                         btrfs_dev_stat_inc_and_print(device,
3259                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3260                 }
3261
3262                 /* drop the reference from the wait == 0 run */
3263                 bio_put(bio);
3264                 device->flush_bio = NULL;
3265
3266                 return ret;
3267         }
3268
3269         /*
3270          * one reference for us, and we leave it for the
3271          * caller
3272          */
3273         device->flush_bio = NULL;
3274         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3275         if (!bio)
3276                 return -ENOMEM;
3277
3278         bio->bi_end_io = btrfs_end_empty_barrier;
3279         bio->bi_bdev = device->bdev;
3280         init_completion(&device->flush_wait);
3281         bio->bi_private = &device->flush_wait;
3282         device->flush_bio = bio;
3283
3284         bio_get(bio);
3285         btrfsic_submit_bio(WRITE_FLUSH, bio);
3286
3287         return 0;
3288 }
3289
3290 /*
3291  * send an empty flush down to each device in parallel,
3292  * then wait for them
3293  */
3294 static int barrier_all_devices(struct btrfs_fs_info *info)
3295 {
3296         struct list_head *head;
3297         struct btrfs_device *dev;
3298         int errors_send = 0;
3299         int errors_wait = 0;
3300         int ret;
3301
3302         /* send down all the barriers */
3303         head = &info->fs_devices->devices;
3304         list_for_each_entry_rcu(dev, head, dev_list) {
3305                 if (dev->missing)
3306                         continue;
3307                 if (!dev->bdev) {
3308                         errors_send++;
3309                         continue;
3310                 }
3311                 if (!dev->in_fs_metadata || !dev->writeable)
3312                         continue;
3313
3314                 ret = write_dev_flush(dev, 0);
3315                 if (ret)
3316                         errors_send++;
3317         }
3318
3319         /* wait for all the barriers */
3320         list_for_each_entry_rcu(dev, head, dev_list) {
3321                 if (dev->missing)
3322                         continue;
3323                 if (!dev->bdev) {
3324                         errors_wait++;
3325                         continue;
3326                 }
3327                 if (!dev->in_fs_metadata || !dev->writeable)
3328                         continue;
3329
3330                 ret = write_dev_flush(dev, 1);
3331                 if (ret)
3332                         errors_wait++;
3333         }
3334         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3335             errors_wait > info->num_tolerated_disk_barrier_failures)
3336                 return -EIO;
3337         return 0;
3338 }
3339
3340 int btrfs_calc_num_tolerated_disk_barrier_failures(
3341         struct btrfs_fs_info *fs_info)
3342 {
3343         struct btrfs_ioctl_space_info space;
3344         struct btrfs_space_info *sinfo;
3345         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3346                        BTRFS_BLOCK_GROUP_SYSTEM,
3347                        BTRFS_BLOCK_GROUP_METADATA,
3348                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3349         int num_types = 4;
3350         int i;
3351         int c;
3352         int num_tolerated_disk_barrier_failures =
3353                 (int)fs_info->fs_devices->num_devices;
3354
3355         for (i = 0; i < num_types; i++) {
3356                 struct btrfs_space_info *tmp;
3357
3358                 sinfo = NULL;
3359                 rcu_read_lock();
3360                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3361                         if (tmp->flags == types[i]) {
3362                                 sinfo = tmp;
3363                                 break;
3364                         }
3365                 }
3366                 rcu_read_unlock();
3367
3368                 if (!sinfo)
3369                         continue;
3370
3371                 down_read(&sinfo->groups_sem);
3372                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3373                         if (!list_empty(&sinfo->block_groups[c])) {
3374                                 u64 flags;
3375
3376                                 btrfs_get_block_group_info(
3377                                         &sinfo->block_groups[c], &space);
3378                                 if (space.total_bytes == 0 ||
3379                                     space.used_bytes == 0)
3380                                         continue;
3381                                 flags = space.flags;
3382                                 /*
3383                                  * return
3384                                  * 0: if dup, single or RAID0 is configured for
3385                                  *    any of metadata, system or data, else
3386                                  * 1: if RAID5 is configured, or if RAID1 or
3387                                  *    RAID10 is configured and only two mirrors
3388                                  *    are used, else
3389                                  * 2: if RAID6 is configured, else
3390                                  * num_mirrors - 1: if RAID1 or RAID10 is
3391                                  *                  configured and more than
3392                                  *                  2 mirrors are used.
3393                                  */
3394                                 if (num_tolerated_disk_barrier_failures > 0 &&
3395                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3396                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3397                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3398                                       == 0)))
3399                                         num_tolerated_disk_barrier_failures = 0;
3400                                 else if (num_tolerated_disk_barrier_failures > 1) {
3401                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3402                                             BTRFS_BLOCK_GROUP_RAID5 |
3403                                             BTRFS_BLOCK_GROUP_RAID10)) {
3404                                                 num_tolerated_disk_barrier_failures = 1;
3405                                         } else if (flags &
3406                                                    BTRFS_BLOCK_GROUP_RAID6) {
3407                                                 num_tolerated_disk_barrier_failures = 2;
3408                                         }
3409                                 }
3410                         }
3411                 }
3412                 up_read(&sinfo->groups_sem);
3413         }
3414
3415         return num_tolerated_disk_barrier_failures;
3416 }
3417
3418 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3419 {
3420         struct list_head *head;
3421         struct btrfs_device *dev;
3422         struct btrfs_super_block *sb;
3423         struct btrfs_dev_item *dev_item;
3424         int ret;
3425         int do_barriers;
3426         int max_errors;
3427         int total_errors = 0;
3428         u64 flags;
3429
3430         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3431         backup_super_roots(root->fs_info);
3432
3433         sb = root->fs_info->super_for_commit;
3434         dev_item = &sb->dev_item;
3435
3436         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3437         head = &root->fs_info->fs_devices->devices;
3438         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3439
3440         if (do_barriers) {
3441                 ret = barrier_all_devices(root->fs_info);
3442                 if (ret) {
3443                         mutex_unlock(
3444                                 &root->fs_info->fs_devices->device_list_mutex);
3445                         btrfs_error(root->fs_info, ret,
3446                                     "errors while submitting device barriers.");
3447                         return ret;
3448                 }
3449         }
3450
3451         list_for_each_entry_rcu(dev, head, dev_list) {
3452                 if (!dev->bdev) {
3453                         total_errors++;
3454                         continue;
3455                 }
3456                 if (!dev->in_fs_metadata || !dev->writeable)
3457                         continue;
3458
3459                 btrfs_set_stack_device_generation(dev_item, 0);
3460                 btrfs_set_stack_device_type(dev_item, dev->type);
3461                 btrfs_set_stack_device_id(dev_item, dev->devid);
3462                 btrfs_set_stack_device_total_bytes(dev_item,
3463                                                    dev->commit_total_bytes);
3464                 btrfs_set_stack_device_bytes_used(dev_item,
3465                                                   dev->commit_bytes_used);
3466                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3467                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3468                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3469                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3470                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3471
3472                 flags = btrfs_super_flags(sb);
3473                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3474
3475                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3476                 if (ret)
3477                         total_errors++;
3478         }
3479         if (total_errors > max_errors) {
3480                 btrfs_err(root->fs_info, "%d errors while writing supers",
3481                        total_errors);
3482                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3483
3484                 /* FUA is masked off if unsupported and can't be the reason */
3485                 btrfs_error(root->fs_info, -EIO,
3486                             "%d errors while writing supers", total_errors);
3487                 return -EIO;
3488         }
3489
3490         total_errors = 0;
3491         list_for_each_entry_rcu(dev, head, dev_list) {
3492                 if (!dev->bdev)
3493                         continue;
3494                 if (!dev->in_fs_metadata || !dev->writeable)
3495                         continue;
3496
3497                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3498                 if (ret)
3499                         total_errors++;
3500         }
3501         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3502         if (total_errors > max_errors) {
3503                 btrfs_error(root->fs_info, -EIO,
3504                             "%d errors while writing supers", total_errors);
3505                 return -EIO;
3506         }
3507         return 0;
3508 }
3509
3510 int write_ctree_super(struct btrfs_trans_handle *trans,
3511                       struct btrfs_root *root, int max_mirrors)
3512 {
3513         return write_all_supers(root, max_mirrors);
3514 }
3515
3516 /* Drop a fs root from the radix tree and free it. */
3517 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3518                                   struct btrfs_root *root)
3519 {
3520         spin_lock(&fs_info->fs_roots_radix_lock);
3521         radix_tree_delete(&fs_info->fs_roots_radix,
3522                           (unsigned long)root->root_key.objectid);
3523         spin_unlock(&fs_info->fs_roots_radix_lock);
3524
3525         if (btrfs_root_refs(&root->root_item) == 0)
3526                 synchronize_srcu(&fs_info->subvol_srcu);
3527
3528         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3529                 btrfs_free_log(NULL, root);
3530
3531         if (root->free_ino_pinned)
3532                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3533         if (root->free_ino_ctl)
3534                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3535         free_fs_root(root);
3536 }
3537
3538 static void free_fs_root(struct btrfs_root *root)
3539 {
3540         iput(root->ino_cache_inode);
3541         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3542         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3543         root->orphan_block_rsv = NULL;
3544         if (root->anon_dev)
3545                 free_anon_bdev(root->anon_dev);
3546         if (root->subv_writers)
3547                 btrfs_free_subvolume_writers(root->subv_writers);
3548         free_extent_buffer(root->node);
3549         free_extent_buffer(root->commit_root);
3550         kfree(root->free_ino_ctl);
3551         kfree(root->free_ino_pinned);
3552         kfree(root->name);
3553         btrfs_put_fs_root(root);
3554 }
3555
3556 void btrfs_free_fs_root(struct btrfs_root *root)
3557 {
3558         free_fs_root(root);
3559 }
3560
3561 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3562 {
3563         u64 root_objectid = 0;
3564         struct btrfs_root *gang[8];
3565         int i = 0;
3566         int err = 0;
3567         unsigned int ret = 0;
3568         int index;
3569
3570         while (1) {
3571                 index = srcu_read_lock(&fs_info->subvol_srcu);
3572                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3573                                              (void **)gang, root_objectid,
3574                                              ARRAY_SIZE(gang));
3575                 if (!ret) {
3576                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3577                         break;
3578                 }
3579                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3580
3581                 for (i = 0; i < ret; i++) {
3582                         /* Avoid to grab roots in dead_roots */
3583                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3584                                 gang[i] = NULL;
3585                                 continue;
3586                         }
3587                         /* grab all the search result for later use */
3588                         gang[i] = btrfs_grab_fs_root(gang[i]);
3589                 }
3590                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3591
3592                 for (i = 0; i < ret; i++) {
3593                         if (!gang[i])
3594                                 continue;
3595                         root_objectid = gang[i]->root_key.objectid;
3596                         err = btrfs_orphan_cleanup(gang[i]);
3597                         if (err)
3598                                 break;
3599                         btrfs_put_fs_root(gang[i]);
3600                 }
3601                 root_objectid++;
3602         }
3603
3604         /* release the uncleaned roots due to error */
3605         for (; i < ret; i++) {
3606                 if (gang[i])
3607                         btrfs_put_fs_root(gang[i]);
3608         }
3609         return err;
3610 }
3611
3612 int btrfs_commit_super(struct btrfs_root *root)
3613 {
3614         struct btrfs_trans_handle *trans;
3615
3616         mutex_lock(&root->fs_info->cleaner_mutex);
3617         btrfs_run_delayed_iputs(root);
3618         mutex_unlock(&root->fs_info->cleaner_mutex);
3619         wake_up_process(root->fs_info->cleaner_kthread);
3620
3621         /* wait until ongoing cleanup work done */
3622         down_write(&root->fs_info->cleanup_work_sem);
3623         up_write(&root->fs_info->cleanup_work_sem);
3624
3625         trans = btrfs_join_transaction(root);
3626         if (IS_ERR(trans))
3627                 return PTR_ERR(trans);
3628         return btrfs_commit_transaction(trans, root);
3629 }
3630
3631 void close_ctree(struct btrfs_root *root)
3632 {
3633         struct btrfs_fs_info *fs_info = root->fs_info;
3634         int ret;
3635
3636         fs_info->closing = 1;
3637         smp_mb();
3638
3639         /* wait for the uuid_scan task to finish */
3640         down(&fs_info->uuid_tree_rescan_sem);
3641         /* avoid complains from lockdep et al., set sem back to initial state */
3642         up(&fs_info->uuid_tree_rescan_sem);
3643
3644         /* pause restriper - we want to resume on mount */
3645         btrfs_pause_balance(fs_info);
3646
3647         btrfs_dev_replace_suspend_for_unmount(fs_info);
3648
3649         btrfs_scrub_cancel(fs_info);
3650
3651         /* wait for any defraggers to finish */
3652         wait_event(fs_info->transaction_wait,
3653                    (atomic_read(&fs_info->defrag_running) == 0));
3654
3655         /* clear out the rbtree of defraggable inodes */
3656         btrfs_cleanup_defrag_inodes(fs_info);
3657
3658         cancel_work_sync(&fs_info->async_reclaim_work);
3659
3660         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3661                 ret = btrfs_commit_super(root);
3662                 if (ret)
3663                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3664         }
3665
3666         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3667                 btrfs_error_commit_super(root);
3668
3669         kthread_stop(fs_info->transaction_kthread);
3670         kthread_stop(fs_info->cleaner_kthread);
3671
3672         fs_info->closing = 2;
3673         smp_mb();
3674
3675         btrfs_free_qgroup_config(root->fs_info);
3676
3677         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3678                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3679                        percpu_counter_sum(&fs_info->delalloc_bytes));
3680         }
3681
3682         btrfs_sysfs_remove_one(fs_info);
3683
3684         btrfs_free_fs_roots(fs_info);
3685
3686         btrfs_put_block_group_cache(fs_info);
3687
3688         btrfs_free_block_groups(fs_info);
3689
3690         /*
3691          * we must make sure there is not any read request to
3692          * submit after we stopping all workers.
3693          */
3694         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3695         btrfs_stop_all_workers(fs_info);
3696
3697         fs_info->open = 0;
3698         free_root_pointers(fs_info, 1);
3699
3700         iput(fs_info->btree_inode);
3701
3702 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3703         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3704                 btrfsic_unmount(root, fs_info->fs_devices);
3705 #endif
3706
3707         btrfs_close_devices(fs_info->fs_devices);
3708         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3709
3710         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3711         percpu_counter_destroy(&fs_info->delalloc_bytes);
3712         percpu_counter_destroy(&fs_info->bio_counter);
3713         bdi_destroy(&fs_info->bdi);
3714         cleanup_srcu_struct(&fs_info->subvol_srcu);
3715
3716         btrfs_free_stripe_hash_table(fs_info);
3717
3718         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3719         root->orphan_block_rsv = NULL;
3720
3721         lock_chunks(root);
3722         while (!list_empty(&fs_info->pinned_chunks)) {
3723                 struct extent_map *em;
3724
3725                 em = list_first_entry(&fs_info->pinned_chunks,
3726                                       struct extent_map, list);
3727                 list_del_init(&em->list);
3728                 free_extent_map(em);
3729         }
3730         unlock_chunks(root);
3731 }
3732
3733 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3734                           int atomic)
3735 {
3736         int ret;
3737         struct inode *btree_inode = buf->pages[0]->mapping->host;
3738
3739         ret = extent_buffer_uptodate(buf);
3740         if (!ret)
3741                 return ret;
3742
3743         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3744                                     parent_transid, atomic);
3745         if (ret == -EAGAIN)
3746                 return ret;
3747         return !ret;
3748 }
3749
3750 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3751 {
3752         return set_extent_buffer_uptodate(buf);
3753 }
3754
3755 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3756 {
3757         struct btrfs_root *root;
3758         u64 transid = btrfs_header_generation(buf);
3759         int was_dirty;
3760
3761 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3762         /*
3763          * This is a fast path so only do this check if we have sanity tests
3764          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3765          * outside of the sanity tests.
3766          */
3767         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3768                 return;
3769 #endif
3770         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3771         btrfs_assert_tree_locked(buf);
3772         if (transid != root->fs_info->generation)
3773                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3774                        "found %llu running %llu\n",
3775                         buf->start, transid, root->fs_info->generation);
3776         was_dirty = set_extent_buffer_dirty(buf);
3777         if (!was_dirty)
3778                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3779                                      buf->len,
3780                                      root->fs_info->dirty_metadata_batch);
3781 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3782         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3783                 btrfs_print_leaf(root, buf);
3784                 ASSERT(0);
3785         }
3786 #endif
3787 }
3788
3789 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3790                                         int flush_delayed)
3791 {
3792         /*
3793          * looks as though older kernels can get into trouble with
3794          * this code, they end up stuck in balance_dirty_pages forever
3795          */
3796         int ret;
3797
3798         if (current->flags & PF_MEMALLOC)
3799                 return;
3800
3801         if (flush_delayed)
3802                 btrfs_balance_delayed_items(root);
3803
3804         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3805                                      BTRFS_DIRTY_METADATA_THRESH);
3806         if (ret > 0) {
3807                 balance_dirty_pages_ratelimited(
3808                                    root->fs_info->btree_inode->i_mapping);
3809         }
3810         return;
3811 }
3812
3813 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3814 {
3815         __btrfs_btree_balance_dirty(root, 1);
3816 }
3817
3818 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3819 {
3820         __btrfs_btree_balance_dirty(root, 0);
3821 }
3822
3823 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3824 {
3825         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3826         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3827 }
3828
3829 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3830                               int read_only)
3831 {
3832         struct btrfs_super_block *sb = fs_info->super_copy;
3833         int ret = 0;
3834
3835         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3836                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3837                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3838                 ret = -EINVAL;
3839         }
3840         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3841                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3842                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3843                 ret = -EINVAL;
3844         }
3845         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3846                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3847                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3848                 ret = -EINVAL;
3849         }
3850
3851         /*
3852          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3853          * items yet, they'll be verified later. Issue just a warning.
3854          */
3855         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3856                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3857                                 btrfs_super_root(sb));
3858         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3859                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3860                                 btrfs_super_chunk_root(sb));
3861         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3862                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3863                                 btrfs_super_log_root(sb));
3864
3865         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3866                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3867                                 fs_info->fsid, sb->dev_item.fsid);
3868                 ret = -EINVAL;
3869         }
3870
3871         /*
3872          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3873          * done later
3874          */
3875         if (btrfs_super_num_devices(sb) > (1UL << 31))
3876                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3877                                 btrfs_super_num_devices(sb));
3878
3879         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3880                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3881                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3882                 ret = -EINVAL;
3883         }
3884
3885         /*
3886          * The generation is a global counter, we'll trust it more than the others
3887          * but it's still possible that it's the one that's wrong.
3888          */
3889         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3890                 printk(KERN_WARNING
3891                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3892                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3893         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3894             && btrfs_super_cache_generation(sb) != (u64)-1)
3895                 printk(KERN_WARNING
3896                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3897                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3898
3899         return ret;
3900 }
3901
3902 static void btrfs_error_commit_super(struct btrfs_root *root)
3903 {
3904         mutex_lock(&root->fs_info->cleaner_mutex);
3905         btrfs_run_delayed_iputs(root);
3906         mutex_unlock(&root->fs_info->cleaner_mutex);
3907
3908         down_write(&root->fs_info->cleanup_work_sem);
3909         up_write(&root->fs_info->cleanup_work_sem);
3910
3911         /* cleanup FS via transaction */
3912         btrfs_cleanup_transaction(root);
3913 }
3914
3915 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3916 {
3917         struct btrfs_ordered_extent *ordered;
3918
3919         spin_lock(&root->ordered_extent_lock);
3920         /*
3921          * This will just short circuit the ordered completion stuff which will
3922          * make sure the ordered extent gets properly cleaned up.
3923          */
3924         list_for_each_entry(ordered, &root->ordered_extents,
3925                             root_extent_list)
3926                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3927         spin_unlock(&root->ordered_extent_lock);
3928 }
3929
3930 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3931 {
3932         struct btrfs_root *root;
3933         struct list_head splice;
3934
3935         INIT_LIST_HEAD(&splice);
3936
3937         spin_lock(&fs_info->ordered_root_lock);
3938         list_splice_init(&fs_info->ordered_roots, &splice);
3939         while (!list_empty(&splice)) {
3940                 root = list_first_entry(&splice, struct btrfs_root,
3941                                         ordered_root);
3942                 list_move_tail(&root->ordered_root,
3943                                &fs_info->ordered_roots);
3944
3945                 spin_unlock(&fs_info->ordered_root_lock);
3946                 btrfs_destroy_ordered_extents(root);
3947
3948                 cond_resched();
3949                 spin_lock(&fs_info->ordered_root_lock);
3950         }
3951         spin_unlock(&fs_info->ordered_root_lock);
3952 }
3953
3954 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3955                                       struct btrfs_root *root)
3956 {
3957         struct rb_node *node;
3958         struct btrfs_delayed_ref_root *delayed_refs;
3959         struct btrfs_delayed_ref_node *ref;
3960         int ret = 0;
3961
3962         delayed_refs = &trans->delayed_refs;
3963
3964         spin_lock(&delayed_refs->lock);
3965         if (atomic_read(&delayed_refs->num_entries) == 0) {
3966                 spin_unlock(&delayed_refs->lock);
3967                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3968                 return ret;
3969         }
3970
3971         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3972                 struct btrfs_delayed_ref_head *head;
3973                 bool pin_bytes = false;
3974
3975                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3976                                 href_node);
3977                 if (!mutex_trylock(&head->mutex)) {
3978                         atomic_inc(&head->node.refs);
3979                         spin_unlock(&delayed_refs->lock);
3980
3981                         mutex_lock(&head->mutex);
3982                         mutex_unlock(&head->mutex);
3983                         btrfs_put_delayed_ref(&head->node);
3984                         spin_lock(&delayed_refs->lock);
3985                         continue;
3986                 }
3987                 spin_lock(&head->lock);
3988                 while ((node = rb_first(&head->ref_root)) != NULL) {
3989                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3990                                        rb_node);
3991                         ref->in_tree = 0;
3992                         rb_erase(&ref->rb_node, &head->ref_root);
3993                         atomic_dec(&delayed_refs->num_entries);
3994                         btrfs_put_delayed_ref(ref);
3995                 }
3996                 if (head->must_insert_reserved)
3997                         pin_bytes = true;
3998                 btrfs_free_delayed_extent_op(head->extent_op);
3999                 delayed_refs->num_heads--;
4000                 if (head->processing == 0)
4001                         delayed_refs->num_heads_ready--;
4002                 atomic_dec(&delayed_refs->num_entries);
4003                 head->node.in_tree = 0;
4004                 rb_erase(&head->href_node, &delayed_refs->href_root);
4005                 spin_unlock(&head->lock);
4006                 spin_unlock(&delayed_refs->lock);
4007                 mutex_unlock(&head->mutex);
4008
4009                 if (pin_bytes)
4010                         btrfs_pin_extent(root, head->node.bytenr,
4011                                          head->node.num_bytes, 1);
4012                 btrfs_put_delayed_ref(&head->node);
4013                 cond_resched();
4014                 spin_lock(&delayed_refs->lock);
4015         }
4016
4017         spin_unlock(&delayed_refs->lock);
4018
4019         return ret;
4020 }
4021
4022 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4023 {
4024         struct btrfs_inode *btrfs_inode;
4025         struct list_head splice;
4026
4027         INIT_LIST_HEAD(&splice);
4028
4029         spin_lock(&root->delalloc_lock);
4030         list_splice_init(&root->delalloc_inodes, &splice);
4031
4032         while (!list_empty(&splice)) {
4033                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4034                                                delalloc_inodes);
4035
4036                 list_del_init(&btrfs_inode->delalloc_inodes);
4037                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4038                           &btrfs_inode->runtime_flags);
4039                 spin_unlock(&root->delalloc_lock);
4040
4041                 btrfs_invalidate_inodes(btrfs_inode->root);
4042
4043                 spin_lock(&root->delalloc_lock);
4044         }
4045
4046         spin_unlock(&root->delalloc_lock);
4047 }
4048
4049 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4050 {
4051         struct btrfs_root *root;
4052         struct list_head splice;
4053
4054         INIT_LIST_HEAD(&splice);
4055
4056         spin_lock(&fs_info->delalloc_root_lock);
4057         list_splice_init(&fs_info->delalloc_roots, &splice);
4058         while (!list_empty(&splice)) {
4059                 root = list_first_entry(&splice, struct btrfs_root,
4060                                          delalloc_root);
4061                 list_del_init(&root->delalloc_root);
4062                 root = btrfs_grab_fs_root(root);
4063                 BUG_ON(!root);
4064                 spin_unlock(&fs_info->delalloc_root_lock);
4065
4066                 btrfs_destroy_delalloc_inodes(root);
4067                 btrfs_put_fs_root(root);
4068
4069                 spin_lock(&fs_info->delalloc_root_lock);
4070         }
4071         spin_unlock(&fs_info->delalloc_root_lock);
4072 }
4073
4074 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4075                                         struct extent_io_tree *dirty_pages,
4076                                         int mark)
4077 {
4078         int ret;
4079         struct extent_buffer *eb;
4080         u64 start = 0;
4081         u64 end;
4082
4083         while (1) {
4084                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4085                                             mark, NULL);
4086                 if (ret)
4087                         break;
4088
4089                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4090                 while (start <= end) {
4091                         eb = btrfs_find_tree_block(root, start);
4092                         start += root->nodesize;
4093                         if (!eb)
4094                                 continue;
4095                         wait_on_extent_buffer_writeback(eb);
4096
4097                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4098                                                &eb->bflags))
4099                                 clear_extent_buffer_dirty(eb);
4100                         free_extent_buffer_stale(eb);
4101                 }
4102         }
4103
4104         return ret;
4105 }
4106
4107 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4108                                        struct extent_io_tree *pinned_extents)
4109 {
4110         struct extent_io_tree *unpin;
4111         u64 start;
4112         u64 end;
4113         int ret;
4114         bool loop = true;
4115
4116         unpin = pinned_extents;
4117 again:
4118         while (1) {
4119                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4120                                             EXTENT_DIRTY, NULL);
4121                 if (ret)
4122                         break;
4123
4124                 /* opt_discard */
4125                 if (btrfs_test_opt(root, DISCARD))
4126                         ret = btrfs_error_discard_extent(root, start,
4127                                                          end + 1 - start,
4128                                                          NULL);
4129
4130                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4131                 btrfs_error_unpin_extent_range(root, start, end);
4132                 cond_resched();
4133         }
4134
4135         if (loop) {
4136                 if (unpin == &root->fs_info->freed_extents[0])
4137                         unpin = &root->fs_info->freed_extents[1];
4138                 else
4139                         unpin = &root->fs_info->freed_extents[0];
4140                 loop = false;
4141                 goto again;
4142         }
4143
4144         return 0;
4145 }
4146
4147 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4148                                        struct btrfs_fs_info *fs_info)
4149 {
4150         struct btrfs_ordered_extent *ordered;
4151
4152         spin_lock(&fs_info->trans_lock);
4153         while (!list_empty(&cur_trans->pending_ordered)) {
4154                 ordered = list_first_entry(&cur_trans->pending_ordered,
4155                                            struct btrfs_ordered_extent,
4156                                            trans_list);
4157                 list_del_init(&ordered->trans_list);
4158                 spin_unlock(&fs_info->trans_lock);
4159
4160                 btrfs_put_ordered_extent(ordered);
4161                 spin_lock(&fs_info->trans_lock);
4162         }
4163         spin_unlock(&fs_info->trans_lock);
4164 }
4165
4166 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4167                                    struct btrfs_root *root)
4168 {
4169         btrfs_destroy_delayed_refs(cur_trans, root);
4170
4171         cur_trans->state = TRANS_STATE_COMMIT_START;
4172         wake_up(&root->fs_info->transaction_blocked_wait);
4173
4174         cur_trans->state = TRANS_STATE_UNBLOCKED;
4175         wake_up(&root->fs_info->transaction_wait);
4176
4177         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4178         btrfs_destroy_delayed_inodes(root);
4179         btrfs_assert_delayed_root_empty(root);
4180
4181         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4182                                      EXTENT_DIRTY);
4183         btrfs_destroy_pinned_extent(root,
4184                                     root->fs_info->pinned_extents);
4185
4186         cur_trans->state =TRANS_STATE_COMPLETED;
4187         wake_up(&cur_trans->commit_wait);
4188
4189         /*
4190         memset(cur_trans, 0, sizeof(*cur_trans));
4191         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4192         */
4193 }
4194
4195 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4196 {
4197         struct btrfs_transaction *t;
4198
4199         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4200
4201         spin_lock(&root->fs_info->trans_lock);
4202         while (!list_empty(&root->fs_info->trans_list)) {
4203                 t = list_first_entry(&root->fs_info->trans_list,
4204                                      struct btrfs_transaction, list);
4205                 if (t->state >= TRANS_STATE_COMMIT_START) {
4206                         atomic_inc(&t->use_count);
4207                         spin_unlock(&root->fs_info->trans_lock);
4208                         btrfs_wait_for_commit(root, t->transid);
4209                         btrfs_put_transaction(t);
4210                         spin_lock(&root->fs_info->trans_lock);
4211                         continue;
4212                 }
4213                 if (t == root->fs_info->running_transaction) {
4214                         t->state = TRANS_STATE_COMMIT_DOING;
4215                         spin_unlock(&root->fs_info->trans_lock);
4216                         /*
4217                          * We wait for 0 num_writers since we don't hold a trans
4218                          * handle open currently for this transaction.
4219                          */
4220                         wait_event(t->writer_wait,
4221                                    atomic_read(&t->num_writers) == 0);
4222                 } else {
4223                         spin_unlock(&root->fs_info->trans_lock);
4224                 }
4225                 btrfs_cleanup_one_transaction(t, root);
4226
4227                 spin_lock(&root->fs_info->trans_lock);
4228                 if (t == root->fs_info->running_transaction)
4229                         root->fs_info->running_transaction = NULL;
4230                 list_del_init(&t->list);
4231                 spin_unlock(&root->fs_info->trans_lock);
4232
4233                 btrfs_put_transaction(t);
4234                 trace_btrfs_transaction_commit(root);
4235                 spin_lock(&root->fs_info->trans_lock);
4236         }
4237         spin_unlock(&root->fs_info->trans_lock);
4238         btrfs_destroy_all_ordered_extents(root->fs_info);
4239         btrfs_destroy_delayed_inodes(root);
4240         btrfs_assert_delayed_root_empty(root);
4241         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4242         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4243         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4244
4245         return 0;
4246 }
4247
4248 static struct extent_io_ops btree_extent_io_ops = {
4249         .readpage_end_io_hook = btree_readpage_end_io_hook,
4250         .readpage_io_failed_hook = btree_io_failed_hook,
4251         .submit_bio_hook = btree_submit_bio_hook,
4252         /* note we're sharing with inode.c for the merge bio hook */
4253         .merge_bio_hook = btrfs_merge_bio_hook,
4254 };