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 const 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_fs_info *fs_info,
 278                            struct extent_buffer *buf,
 279                            int verify)
 280 {
 281         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 282         char *result = NULL;
 283         unsigned long len;
 284         unsigned long cur_len;
 285         unsigned long offset = BTRFS_CSUM_SIZE;
 286         char *kaddr;
 287         unsigned long map_start;
 288         unsigned long map_len;
 289         int err;
 290         u32 crc = ~(u32)0;
 291         unsigned long inline_result;
 292
 293         len = buf->len - offset;
 294         while (len > 0) {
 295                 err = map_private_extent_buffer(buf, offset, 32,
 296                                         &kaddr, &map_start, &map_len);
 297                 if (err)
 298                         return 1;
 299                 cur_len = min(len, map_len - (offset - map_start));
 300                 crc = btrfs_csum_data(kaddr + offset - map_start,
 301                                       crc, cur_len);
 302                 len -= cur_len;
 303                 offset += cur_len;
 304         }
 305         if (csum_size > sizeof(inline_result)) {
 306                 result = kzalloc(csum_size, GFP_NOFS);
 307                 if (!result)
 308                         return 1;
 309         } else {
 310                 result = (char *)&inline_result;
 311         }
 312
 313         btrfs_csum_final(crc, result);
 314
 315         if (verify) {
 316                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 317                         u32 val;
 318                         u32 found = 0;
 319                         memcpy(&found, result, csum_size);
 320
 321                         read_extent_buffer(buf, &val, 0, csum_size);
 322                         printk_ratelimited(KERN_WARNING
 323                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 324                                 "level %d\n",
 325                                 fs_info->sb->s_id, buf->start,
 326                                 val, found, btrfs_header_level(buf));
 327                         if (result != (char *)&inline_result)
 328                                 kfree(result);
 329                         return 1;
 330                 }
 331         } else {
 332                 write_extent_buffer(buf, result, 0, csum_size);
 333         }
 334         if (result != (char *)&inline_result)
 335                 kfree(result);
 336         return 0;
 337 }
 338
 339 /*
 340  * we can't consider a given block up to date unless the transid of the
 341  * block matches the transid in the parent node's pointer.  This is how we
 342  * detect blocks that either didn't get written at all or got written
 343  * in the wrong place.
 344  */
 345 static int verify_parent_transid(struct extent_io_tree *io_tree,
 346                                  struct extent_buffer *eb, u64 parent_transid,
 347                                  int atomic)
 348 {
 349         struct extent_state *cached_state = NULL;
 350         int ret;
 351         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 352
 353         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 354                 return 0;
 355
 356         if (atomic)
 357                 return -EAGAIN;
 358
 359         if (need_lock) {
 360                 btrfs_tree_read_lock(eb);
 361                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 362         }
 363
 364         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 365                          0, &cached_state);
 366         if (extent_buffer_uptodate(eb) &&
 367             btrfs_header_generation(eb) == parent_transid) {
 368                 ret = 0;
 369                 goto out;
 370         }
 371         printk_ratelimited(KERN_ERR
 372             "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 373                         eb->fs_info->sb->s_id, eb->start,
 374                         parent_transid, btrfs_header_generation(eb));
 375         ret = 1;
 376
 377         /*
 378          * Things reading via commit roots that don't have normal protection,
 379          * like send, can have a really old block in cache that may point at a
 380          * block that has been free'd and re-allocated.  So don't clear uptodate
 381          * if we find an eb that is under IO (dirty/writeback) because we could
 382          * end up reading in the stale data and then writing it back out and
 383          * making everybody very sad.
 384          */
 385         if (!extent_buffer_under_io(eb))
 386                 clear_extent_buffer_uptodate(eb);
 387 out:
 388         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 389                              &cached_state, GFP_NOFS);
 390         if (need_lock)
 391                 btrfs_tree_read_unlock_blocking(eb);
 392         return ret;
 393 }
 394
 395 /*
 396  * Return 0 if the superblock checksum type matches the checksum value of that
 397  * algorithm. Pass the raw disk superblock data.
 398  */
 399 static int btrfs_check_super_csum(char *raw_disk_sb)
 400 {
 401         struct btrfs_super_block *disk_sb =
 402                 (struct btrfs_super_block *)raw_disk_sb;
 403         u16 csum_type = btrfs_super_csum_type(disk_sb);
 404         int ret = 0;
 405
 406         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 407                 u32 crc = ~(u32)0;
 408                 const int csum_size = sizeof(crc);
 409                 char result[csum_size];
 410
 411                 /*
 412                  * The super_block structure does not span the whole
 413                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 414                  * is filled with zeros and is included in the checkum.
 415                  */
 416                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 417                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 418                 btrfs_csum_final(crc, result);
 419
 420                 if (memcmp(raw_disk_sb, result, csum_size))
 421                         ret = 1;
 422         }
 423
 424         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 425                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 426                                 csum_type);
 427                 ret = 1;
 428         }
 429
 430         return ret;
 431 }
 432
 433 /*
 434  * helper to read a given tree block, doing retries as required when
 435  * the checksums don't match and we have alternate mirrors to try.
 436  */
 437 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 438                                           struct extent_buffer *eb,
 439                                           u64 start, u64 parent_transid)
 440 {
 441         struct extent_io_tree *io_tree;
 442         int failed = 0;
 443         int ret;
 444         int num_copies = 0;
 445         int mirror_num = 0;
 446         int failed_mirror = 0;
 447
 448         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 449         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 450         while (1) {
 451                 ret = read_extent_buffer_pages(io_tree, eb, start,
 452                                                WAIT_COMPLETE,
 453                                                btree_get_extent, mirror_num);
 454                 if (!ret) {
 455                         if (!verify_parent_transid(io_tree, eb,
 456                                                    parent_transid, 0))
 457                                 break;
 458                         else
 459                                 ret = -EIO;
 460                 }
 461
 462                 /*
 463                  * This buffer's crc is fine, but its contents are corrupted, so
 464                  * there is no reason to read the other copies, they won't be
 465                  * any less wrong.
 466                  */
 467                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 468                         break;
 469
 470                 num_copies = btrfs_num_copies(root->fs_info,
 471                                               eb->start, eb->len);
 472                 if (num_copies == 1)
 473                         break;
 474
 475                 if (!failed_mirror) {
 476                         failed = 1;
 477                         failed_mirror = eb->read_mirror;
 478                 }
 479
 480                 mirror_num++;
 481                 if (mirror_num == failed_mirror)
 482                         mirror_num++;
 483
 484                 if (mirror_num > num_copies)
 485                         break;
 486         }
 487
 488         if (failed && !ret && failed_mirror)
 489                 repair_eb_io_failure(root, eb, failed_mirror);
 490
 491         return ret;
 492 }
 493
 494 /*
 495  * checksum a dirty tree block before IO.  This has extra checks to make sure
 496  * we only fill in the checksum field in the first page of a multi-page block
 497  */
 498
 499 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 500 {
 501         u64 start = page_offset(page);
 502         u64 found_start;
 503         struct extent_buffer *eb;
 504
 505         eb = (struct extent_buffer *)page->private;
 506         if (page != eb->pages[0])
 507                 return 0;
 508         found_start = btrfs_header_bytenr(eb);
 509         if (WARN_ON(found_start != start || !PageUptodate(page)))
 510                 return 0;
 511         csum_tree_block(fs_info, eb, 0);
 512         return 0;
 513 }
 514
 515 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 516                                  struct extent_buffer *eb)
 517 {
 518         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 519         u8 fsid[BTRFS_UUID_SIZE];
 520         int ret = 1;
 521
 522         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 523         while (fs_devices) {
 524                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 525                         ret = 0;
 526                         break;
 527                 }
 528                 fs_devices = fs_devices->seed;
 529         }
 530         return ret;
 531 }
 532
 533 #define CORRUPT(reason, eb, root, slot)                         \
 534         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 535                    "root=%llu, slot=%d", reason,                        \
 536                btrfs_header_bytenr(eb), root->objectid, slot)
 537
 538 static noinline int check_leaf(struct btrfs_root *root,
 539                                struct extent_buffer *leaf)
 540 {
 541         struct btrfs_key key;
 542         struct btrfs_key leaf_key;
 543         u32 nritems = btrfs_header_nritems(leaf);
 544         int slot;
 545
 546         if (nritems == 0)
 547                 return 0;
 548
 549         /* Check the 0 item */
 550         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 551             BTRFS_LEAF_DATA_SIZE(root)) {
 552                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 553                 return -EIO;
 554         }
 555
 556         /*
 557          * Check to make sure each items keys are in the correct order and their
 558          * offsets make sense.  We only have to loop through nritems-1 because
 559          * we check the current slot against the next slot, which verifies the
 560          * next slot's offset+size makes sense and that the current's slot
 561          * offset is correct.
 562          */
 563         for (slot = 0; slot < nritems - 1; slot++) {
 564                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 565                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 566
 567                 /* Make sure the keys are in the right order */
 568                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 569                         CORRUPT("bad key order", leaf, root, slot);
 570                         return -EIO;
 571                 }
 572
 573                 /*
 574                  * Make sure the offset and ends are right, remember that the
 575                  * item data starts at the end of the leaf and grows towards the
 576                  * front.
 577                  */
 578                 if (btrfs_item_offset_nr(leaf, slot) !=
 579                         btrfs_item_end_nr(leaf, slot + 1)) {
 580                         CORRUPT("slot offset bad", leaf, root, slot);
 581                         return -EIO;
 582                 }
 583
 584                 /*
 585                  * Check to make sure that we don't point outside of the leaf,
 586                  * just incase all the items are consistent to eachother, but
 587                  * all point outside of the leaf.
 588                  */
 589                 if (btrfs_item_end_nr(leaf, slot) >
 590                     BTRFS_LEAF_DATA_SIZE(root)) {
 591                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 592                         return -EIO;
 593                 }
 594         }
 595
 596         return 0;
 597 }
 598
 599 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 600                                       u64 phy_offset, struct page *page,
 601                                       u64 start, u64 end, int mirror)
 602 {
 603         u64 found_start;
 604         int found_level;
 605         struct extent_buffer *eb;
 606         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 607         int ret = 0;
 608         int reads_done;
 609
 610         if (!page->private)
 611                 goto out;
 612
 613         eb = (struct extent_buffer *)page->private;
 614
 615         /* the pending IO might have been the only thing that kept this buffer
 616          * in memory.  Make sure we have a ref for all this other checks
 617          */
 618         extent_buffer_get(eb);
 619
 620         reads_done = atomic_dec_and_test(&eb->io_pages);
 621         if (!reads_done)
 622                 goto err;
 623
 624         eb->read_mirror = mirror;
 625         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 626                 ret = -EIO;
 627                 goto err;
 628         }
 629
 630         found_start = btrfs_header_bytenr(eb);
 631         if (found_start != eb->start) {
 632                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
 633                                "%llu %llu\n",
 634                                eb->fs_info->sb->s_id, found_start, eb->start);
 635                 ret = -EIO;
 636                 goto err;
 637         }
 638         if (check_tree_block_fsid(root->fs_info, eb)) {
 639                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
 640                                eb->fs_info->sb->s_id, eb->start);
 641                 ret = -EIO;
 642                 goto err;
 643         }
 644         found_level = btrfs_header_level(eb);
 645         if (found_level >= BTRFS_MAX_LEVEL) {
 646                 btrfs_err(root->fs_info, "bad tree block level %d",
 647                            (int)btrfs_header_level(eb));
 648                 ret = -EIO;
 649                 goto err;
 650         }
 651
 652         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 653                                        eb, found_level);
 654
 655         ret = csum_tree_block(root->fs_info, eb, 1);
 656         if (ret) {
 657                 ret = -EIO;
 658                 goto err;
 659         }
 660
 661         /*
 662          * If this is a leaf block and it is corrupt, set the corrupt bit so
 663          * that we don't try and read the other copies of this block, just
 664          * return -EIO.
 665          */
 666         if (found_level == 0 && check_leaf(root, eb)) {
 667                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 668                 ret = -EIO;
 669         }
 670
 671         if (!ret)
 672                 set_extent_buffer_uptodate(eb);
 673 err:
 674         if (reads_done &&
 675             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 676                 btree_readahead_hook(root, eb, eb->start, ret);
 677
 678         if (ret) {
 679                 /*
 680                  * our io error hook is going to dec the io pages
 681                  * again, we have to make sure it has something
 682                  * to decrement
 683                  */
 684                 atomic_inc(&eb->io_pages);
 685                 clear_extent_buffer_uptodate(eb);
 686         }
 687         free_extent_buffer(eb);
 688 out:
 689         return ret;
 690 }
 691
 692 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 693 {
 694         struct extent_buffer *eb;
 695         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 696
 697         eb = (struct extent_buffer *)page->private;
 698         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 699         eb->read_mirror = failed_mirror;
 700         atomic_dec(&eb->io_pages);
 701         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 702                 btree_readahead_hook(root, eb, eb->start, -EIO);
 703         return -EIO;    /* we fixed nothing */
 704 }
 705
 706 static void end_workqueue_bio(struct bio *bio, int err)
 707 {
 708         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 709         struct btrfs_fs_info *fs_info;
 710         struct btrfs_workqueue *wq;
 711         btrfs_work_func_t func;
 712
 713         fs_info = end_io_wq->info;
 714         end_io_wq->error = err;
 715
 716         if (bio->bi_rw & REQ_WRITE) {
 717                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 718                         wq = fs_info->endio_meta_write_workers;
 719                         func = btrfs_endio_meta_write_helper;
 720                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 721                         wq = fs_info->endio_freespace_worker;
 722                         func = btrfs_freespace_write_helper;
 723                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 724                         wq = fs_info->endio_raid56_workers;
 725                         func = btrfs_endio_raid56_helper;
 726                 } else {
 727                         wq = fs_info->endio_write_workers;
 728                         func = btrfs_endio_write_helper;
 729                 }
 730         } else {
 731                 if (unlikely(end_io_wq->metadata ==
 732                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 733                         wq = fs_info->endio_repair_workers;
 734                         func = btrfs_endio_repair_helper;
 735                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 736                         wq = fs_info->endio_raid56_workers;
 737                         func = btrfs_endio_raid56_helper;
 738                 } else if (end_io_wq->metadata) {
 739                         wq = fs_info->endio_meta_workers;
 740                         func = btrfs_endio_meta_helper;
 741                 } else {
 742                         wq = fs_info->endio_workers;
 743                         func = btrfs_endio_helper;
 744                 }
 745         }
 746
 747         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 748         btrfs_queue_work(wq, &end_io_wq->work);
 749 }
 750
 751 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 752                         enum btrfs_wq_endio_type metadata)
 753 {
 754         struct btrfs_end_io_wq *end_io_wq;
 755
 756         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 757         if (!end_io_wq)
 758                 return -ENOMEM;
 759
 760         end_io_wq->private = bio->bi_private;
 761         end_io_wq->end_io = bio->bi_end_io;
 762         end_io_wq->info = info;
 763         end_io_wq->error = 0;
 764         end_io_wq->bio = bio;
 765         end_io_wq->metadata = metadata;
 766
 767         bio->bi_private = end_io_wq;
 768         bio->bi_end_io = end_workqueue_bio;
 769         return 0;
 770 }
 771
 772 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 773 {
 774         unsigned long limit = min_t(unsigned long,
 775                                     info->thread_pool_size,
 776                                     info->fs_devices->open_devices);
 777         return 256 * limit;
 778 }
 779
 780 static void run_one_async_start(struct btrfs_work *work)
 781 {
 782         struct async_submit_bio *async;
 783         int ret;
 784
 785         async = container_of(work, struct  async_submit_bio, work);
 786         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 787                                       async->mirror_num, async->bio_flags,
 788                                       async->bio_offset);
 789         if (ret)
 790                 async->error = ret;
 791 }
 792
 793 static void run_one_async_done(struct btrfs_work *work)
 794 {
 795         struct btrfs_fs_info *fs_info;
 796         struct async_submit_bio *async;
 797         int limit;
 798
 799         async = container_of(work, struct  async_submit_bio, work);
 800         fs_info = BTRFS_I(async->inode)->root->fs_info;
 801
 802         limit = btrfs_async_submit_limit(fs_info);
 803         limit = limit * 2 / 3;
 804
 805         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 806             waitqueue_active(&fs_info->async_submit_wait))
 807                 wake_up(&fs_info->async_submit_wait);
 808
 809         /* If an error occured we just want to clean up the bio and move on */
 810         if (async->error) {
 811                 bio_endio(async->bio, async->error);
 812                 return;
 813         }
 814
 815         async->submit_bio_done(async->inode, async->rw, async->bio,
 816                                async->mirror_num, async->bio_flags,
 817                                async->bio_offset);
 818 }
 819
 820 static void run_one_async_free(struct btrfs_work *work)
 821 {
 822         struct async_submit_bio *async;
 823
 824         async = container_of(work, struct  async_submit_bio, work);
 825         kfree(async);
 826 }
 827
 828 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 829                         int rw, struct bio *bio, int mirror_num,
 830                         unsigned long bio_flags,
 831                         u64 bio_offset,
 832                         extent_submit_bio_hook_t *submit_bio_start,
 833                         extent_submit_bio_hook_t *submit_bio_done)
 834 {
 835         struct async_submit_bio *async;
 836
 837         async = kmalloc(sizeof(*async), GFP_NOFS);
 838         if (!async)
 839                 return -ENOMEM;
 840
 841         async->inode = inode;
 842         async->rw = rw;
 843         async->bio = bio;
 844         async->mirror_num = mirror_num;
 845         async->submit_bio_start = submit_bio_start;
 846         async->submit_bio_done = submit_bio_done;
 847
 848         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 849                         run_one_async_done, run_one_async_free);
 850
 851         async->bio_flags = bio_flags;
 852         async->bio_offset = bio_offset;
 853
 854         async->error = 0;
 855
 856         atomic_inc(&fs_info->nr_async_submits);
 857
 858         if (rw & REQ_SYNC)
 859                 btrfs_set_work_high_priority(&async->work);
 860
 861         btrfs_queue_work(fs_info->workers, &async->work);
 862
 863         while (atomic_read(&fs_info->async_submit_draining) &&
 864               atomic_read(&fs_info->nr_async_submits)) {
 865                 wait_event(fs_info->async_submit_wait,
 866                            (atomic_read(&fs_info->nr_async_submits) == 0));
 867         }
 868
 869         return 0;
 870 }
 871
 872 static int btree_csum_one_bio(struct bio *bio)
 873 {
 874         struct bio_vec *bvec;
 875         struct btrfs_root *root;
 876         int i, ret = 0;
 877
 878         bio_for_each_segment_all(bvec, bio, i) {
 879                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 880                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 881                 if (ret)
 882                         break;
 883         }
 884
 885         return ret;
 886 }
 887
 888 static int __btree_submit_bio_start(struct inode *inode, int rw,
 889                                     struct bio *bio, int mirror_num,
 890                                     unsigned long bio_flags,
 891                                     u64 bio_offset)
 892 {
 893         /*
 894          * when we're called for a write, we're already in the async
 895          * submission context.  Just jump into btrfs_map_bio
 896          */
 897         return btree_csum_one_bio(bio);
 898 }
 899
 900 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 901                                  int mirror_num, unsigned long bio_flags,
 902                                  u64 bio_offset)
 903 {
 904         int ret;
 905
 906         /*
 907          * when we're called for a write, we're already in the async
 908          * submission context.  Just jump into btrfs_map_bio
 909          */
 910         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 911         if (ret)
 912                 bio_endio(bio, ret);
 913         return ret;
 914 }
 915
 916 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 917 {
 918         if (bio_flags & EXTENT_BIO_TREE_LOG)
 919                 return 0;
 920 #ifdef CONFIG_X86
 921         if (cpu_has_xmm4_2)
 922                 return 0;
 923 #endif
 924         return 1;
 925 }
 926
 927 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 928                                  int mirror_num, unsigned long bio_flags,
 929                                  u64 bio_offset)
 930 {
 931         int async = check_async_write(inode, bio_flags);
 932         int ret;
 933
 934         if (!(rw & REQ_WRITE)) {
 935                 /*
 936                  * called for a read, do the setup so that checksum validation
 937                  * can happen in the async kernel threads
 938                  */
 939                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 940                                           bio, BTRFS_WQ_ENDIO_METADATA);
 941                 if (ret)
 942                         goto out_w_error;
 943                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 944                                     mirror_num, 0);
 945         } else if (!async) {
 946                 ret = btree_csum_one_bio(bio);
 947                 if (ret)
 948                         goto out_w_error;
 949                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 950                                     mirror_num, 0);
 951         } else {
 952                 /*
 953                  * kthread helpers are used to submit writes so that
 954                  * checksumming can happen in parallel across all CPUs
 955                  */
 956                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 957                                           inode, rw, bio, mirror_num, 0,
 958                                           bio_offset,
 959                                           __btree_submit_bio_start,
 960                                           __btree_submit_bio_done);
 961         }
 962
 963         if (ret) {
 964 out_w_error:
 965                 bio_endio(bio, ret);
 966         }
 967         return ret;
 968 }
 969
 970 #ifdef CONFIG_MIGRATION
 971 static int btree_migratepage(struct address_space *mapping,
 972                         struct page *newpage, struct page *page,
 973                         enum migrate_mode mode)
 974 {
 975         /*
 976          * we can't safely write a btree page from here,
 977          * we haven't done the locking hook
 978          */
 979         if (PageDirty(page))
 980                 return -EAGAIN;
 981         /*
 982          * Buffers may be managed in a filesystem specific way.
 983          * We must have no buffers or drop them.
 984          */
 985         if (page_has_private(page) &&
 986             !try_to_release_page(page, GFP_KERNEL))
 987                 return -EAGAIN;
 988         return migrate_page(mapping, newpage, page, mode);
 989 }
 990 #endif
 991
 992
 993 static int btree_writepages(struct address_space *mapping,
 994                             struct writeback_control *wbc)
 995 {
 996         struct btrfs_fs_info *fs_info;
 997         int ret;
 998
 999         if (wbc->sync_mode == WB_SYNC_NONE) {
1000
1001                 if (wbc->for_kupdate)
1002                         return 0;
1003
1004                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1005                 /* this is a bit racy, but that's ok */
1006                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1007                                              BTRFS_DIRTY_METADATA_THRESH);
1008                 if (ret < 0)
1009                         return 0;
1010         }
1011         return btree_write_cache_pages(mapping, wbc);
1012 }
1013
1014 static int btree_readpage(struct file *file, struct page *page)
1015 {
1016         struct extent_io_tree *tree;
1017         tree = &BTRFS_I(page->mapping->host)->io_tree;
1018         return extent_read_full_page(tree, page, btree_get_extent, 0);
1019 }
1020
1021 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1022 {
1023         if (PageWriteback(page) || PageDirty(page))
1024                 return 0;
1025
1026         return try_release_extent_buffer(page);
1027 }
1028
1029 static void btree_invalidatepage(struct page *page, unsigned int offset,
1030                                  unsigned int length)
1031 {
1032         struct extent_io_tree *tree;
1033         tree = &BTRFS_I(page->mapping->host)->io_tree;
1034         extent_invalidatepage(tree, page, offset);
1035         btree_releasepage(page, GFP_NOFS);
1036         if (PagePrivate(page)) {
1037                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1038                            "page private not zero on page %llu",
1039                            (unsigned long long)page_offset(page));
1040                 ClearPagePrivate(page);
1041                 set_page_private(page, 0);
1042                 page_cache_release(page);
1043         }
1044 }
1045
1046 static int btree_set_page_dirty(struct page *page)
1047 {
1048 #ifdef DEBUG
1049         struct extent_buffer *eb;
1050
1051         BUG_ON(!PagePrivate(page));
1052         eb = (struct extent_buffer *)page->private;
1053         BUG_ON(!eb);
1054         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1055         BUG_ON(!atomic_read(&eb->refs));
1056         btrfs_assert_tree_locked(eb);
1057 #endif
1058         return __set_page_dirty_nobuffers(page);
1059 }
1060
1061 static const struct address_space_operations btree_aops = {
1062         .readpage       = btree_readpage,
1063         .writepages     = btree_writepages,
1064         .releasepage    = btree_releasepage,
1065         .invalidatepage = btree_invalidatepage,
1066 #ifdef CONFIG_MIGRATION
1067         .migratepage    = btree_migratepage,
1068 #endif
1069         .set_page_dirty = btree_set_page_dirty,
1070 };
1071
1072 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1073 {
1074         struct extent_buffer *buf = NULL;
1075         struct inode *btree_inode = root->fs_info->btree_inode;
1076
1077         buf = btrfs_find_create_tree_block(root, bytenr);
1078         if (!buf)
1079                 return;
1080         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1081                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1082         free_extent_buffer(buf);
1083 }
1084
1085 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1086                          int mirror_num, struct extent_buffer **eb)
1087 {
1088         struct extent_buffer *buf = NULL;
1089         struct inode *btree_inode = root->fs_info->btree_inode;
1090         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1091         int ret;
1092
1093         buf = btrfs_find_create_tree_block(root, bytenr);
1094         if (!buf)
1095                 return 0;
1096
1097         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1098
1099         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1100                                        btree_get_extent, mirror_num);
1101         if (ret) {
1102                 free_extent_buffer(buf);
1103                 return ret;
1104         }
1105
1106         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1107                 free_extent_buffer(buf);
1108                 return -EIO;
1109         } else if (extent_buffer_uptodate(buf)) {
1110                 *eb = buf;
1111         } else {
1112                 free_extent_buffer(buf);
1113         }
1114         return 0;
1115 }
1116
1117 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1118                                             u64 bytenr)
1119 {
1120         return find_extent_buffer(fs_info, bytenr);
1121 }
1122
1123 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1124                                                  u64 bytenr)
1125 {
1126         if (btrfs_test_is_dummy_root(root))
1127                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1128         return alloc_extent_buffer(root->fs_info, bytenr);
1129 }
1130
1131
1132 int btrfs_write_tree_block(struct extent_buffer *buf)
1133 {
1134         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1135                                         buf->start + buf->len - 1);
1136 }
1137
1138 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1139 {
1140         return filemap_fdatawait_range(buf->pages[0]->mapping,
1141                                        buf->start, buf->start + buf->len - 1);
1142 }
1143
1144 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1145                                       u64 parent_transid)
1146 {
1147         struct extent_buffer *buf = NULL;
1148         int ret;
1149
1150         buf = btrfs_find_create_tree_block(root, bytenr);
1151         if (!buf)
1152                 return NULL;
1153
1154         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1155         if (ret) {
1156                 free_extent_buffer(buf);
1157                 return NULL;
1158         }
1159         return buf;
1160
1161 }
1162
1163 void clean_tree_block(struct btrfs_trans_handle *trans,
1164                       struct btrfs_fs_info *fs_info,
1165                       struct extent_buffer *buf)
1166 {
1167         if (btrfs_header_generation(buf) ==
1168             fs_info->running_transaction->transid) {
1169                 btrfs_assert_tree_locked(buf);
1170
1171                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1172                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1173                                              -buf->len,
1174                                              fs_info->dirty_metadata_batch);
1175                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1176                         btrfs_set_lock_blocking(buf);
1177                         clear_extent_buffer_dirty(buf);
1178                 }
1179         }
1180 }
1181
1182 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1183 {
1184         struct btrfs_subvolume_writers *writers;
1185         int ret;
1186
1187         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1188         if (!writers)
1189                 return ERR_PTR(-ENOMEM);
1190
1191         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1192         if (ret < 0) {
1193                 kfree(writers);
1194                 return ERR_PTR(ret);
1195         }
1196
1197         init_waitqueue_head(&writers->wait);
1198         return writers;
1199 }
1200
1201 static void
1202 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1203 {
1204         percpu_counter_destroy(&writers->counter);
1205         kfree(writers);
1206 }
1207
1208 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1209                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1210                          u64 objectid)
1211 {
1212         root->node = NULL;
1213         root->commit_root = NULL;
1214         root->sectorsize = sectorsize;
1215         root->nodesize = nodesize;
1216         root->stripesize = stripesize;
1217         root->state = 0;
1218         root->orphan_cleanup_state = 0;
1219
1220         root->objectid = objectid;
1221         root->last_trans = 0;
1222         root->highest_objectid = 0;
1223         root->nr_delalloc_inodes = 0;
1224         root->nr_ordered_extents = 0;
1225         root->name = NULL;
1226         root->inode_tree = RB_ROOT;
1227         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1228         root->block_rsv = NULL;
1229         root->orphan_block_rsv = NULL;
1230
1231         INIT_LIST_HEAD(&root->dirty_list);
1232         INIT_LIST_HEAD(&root->root_list);
1233         INIT_LIST_HEAD(&root->delalloc_inodes);
1234         INIT_LIST_HEAD(&root->delalloc_root);
1235         INIT_LIST_HEAD(&root->ordered_extents);
1236         INIT_LIST_HEAD(&root->ordered_root);
1237         INIT_LIST_HEAD(&root->logged_list[0]);
1238         INIT_LIST_HEAD(&root->logged_list[1]);
1239         spin_lock_init(&root->orphan_lock);
1240         spin_lock_init(&root->inode_lock);
1241         spin_lock_init(&root->delalloc_lock);
1242         spin_lock_init(&root->ordered_extent_lock);
1243         spin_lock_init(&root->accounting_lock);
1244         spin_lock_init(&root->log_extents_lock[0]);
1245         spin_lock_init(&root->log_extents_lock[1]);
1246         mutex_init(&root->objectid_mutex);
1247         mutex_init(&root->log_mutex);
1248         mutex_init(&root->ordered_extent_mutex);
1249         mutex_init(&root->delalloc_mutex);
1250         init_waitqueue_head(&root->log_writer_wait);
1251         init_waitqueue_head(&root->log_commit_wait[0]);
1252         init_waitqueue_head(&root->log_commit_wait[1]);
1253         INIT_LIST_HEAD(&root->log_ctxs[0]);
1254         INIT_LIST_HEAD(&root->log_ctxs[1]);
1255         atomic_set(&root->log_commit[0], 0);
1256         atomic_set(&root->log_commit[1], 0);
1257         atomic_set(&root->log_writers, 0);
1258         atomic_set(&root->log_batch, 0);
1259         atomic_set(&root->orphan_inodes, 0);
1260         atomic_set(&root->refs, 1);
1261         atomic_set(&root->will_be_snapshoted, 0);
1262         root->log_transid = 0;
1263         root->log_transid_committed = -1;
1264         root->last_log_commit = 0;
1265         if (fs_info)
1266                 extent_io_tree_init(&root->dirty_log_pages,
1267                                      fs_info->btree_inode->i_mapping);
1268
1269         memset(&root->root_key, 0, sizeof(root->root_key));
1270         memset(&root->root_item, 0, sizeof(root->root_item));
1271         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1272         if (fs_info)
1273                 root->defrag_trans_start = fs_info->generation;
1274         else
1275                 root->defrag_trans_start = 0;
1276         root->root_key.objectid = objectid;
1277         root->anon_dev = 0;
1278
1279         spin_lock_init(&root->root_item_lock);
1280 }
1281
1282 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1283 {
1284         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1285         if (root)
1286                 root->fs_info = fs_info;
1287         return root;
1288 }
1289
1290 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1291 /* Should only be used by the testing infrastructure */
1292 struct btrfs_root *btrfs_alloc_dummy_root(void)
1293 {
1294         struct btrfs_root *root;
1295
1296         root = btrfs_alloc_root(NULL);
1297         if (!root)
1298                 return ERR_PTR(-ENOMEM);
1299         __setup_root(4096, 4096, 4096, root, NULL, 1);
1300         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1301         root->alloc_bytenr = 0;
1302
1303         return root;
1304 }
1305 #endif
1306
1307 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1308                                      struct btrfs_fs_info *fs_info,
1309                                      u64 objectid)
1310 {
1311         struct extent_buffer *leaf;
1312         struct btrfs_root *tree_root = fs_info->tree_root;
1313         struct btrfs_root *root;
1314         struct btrfs_key key;
1315         int ret = 0;
1316         uuid_le uuid;
1317
1318         root = btrfs_alloc_root(fs_info);
1319         if (!root)
1320                 return ERR_PTR(-ENOMEM);
1321
1322         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1323                 tree_root->stripesize, root, fs_info, objectid);
1324         root->root_key.objectid = objectid;
1325         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1326         root->root_key.offset = 0;
1327
1328         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1329         if (IS_ERR(leaf)) {
1330                 ret = PTR_ERR(leaf);
1331                 leaf = NULL;
1332                 goto fail;
1333         }
1334
1335         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1336         btrfs_set_header_bytenr(leaf, leaf->start);
1337         btrfs_set_header_generation(leaf, trans->transid);
1338         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1339         btrfs_set_header_owner(leaf, objectid);
1340         root->node = leaf;
1341
1342         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1343                             BTRFS_FSID_SIZE);
1344         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1345                             btrfs_header_chunk_tree_uuid(leaf),
1346                             BTRFS_UUID_SIZE);
1347         btrfs_mark_buffer_dirty(leaf);
1348
1349         root->commit_root = btrfs_root_node(root);
1350         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1351
1352         root->root_item.flags = 0;
1353         root->root_item.byte_limit = 0;
1354         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1355         btrfs_set_root_generation(&root->root_item, trans->transid);
1356         btrfs_set_root_level(&root->root_item, 0);
1357         btrfs_set_root_refs(&root->root_item, 1);
1358         btrfs_set_root_used(&root->root_item, leaf->len);
1359         btrfs_set_root_last_snapshot(&root->root_item, 0);
1360         btrfs_set_root_dirid(&root->root_item, 0);
1361         uuid_le_gen(&uuid);
1362         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1363         root->root_item.drop_level = 0;
1364
1365         key.objectid = objectid;
1366         key.type = BTRFS_ROOT_ITEM_KEY;
1367         key.offset = 0;
1368         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1369         if (ret)
1370                 goto fail;
1371
1372         btrfs_tree_unlock(leaf);
1373
1374         return root;
1375
1376 fail:
1377         if (leaf) {
1378                 btrfs_tree_unlock(leaf);
1379                 free_extent_buffer(root->commit_root);
1380                 free_extent_buffer(leaf);
1381         }
1382         kfree(root);
1383
1384         return ERR_PTR(ret);
1385 }
1386
1387 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1388                                          struct btrfs_fs_info *fs_info)
1389 {
1390         struct btrfs_root *root;
1391         struct btrfs_root *tree_root = fs_info->tree_root;
1392         struct extent_buffer *leaf;
1393
1394         root = btrfs_alloc_root(fs_info);
1395         if (!root)
1396                 return ERR_PTR(-ENOMEM);
1397
1398         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1399                      tree_root->stripesize, root, fs_info,
1400                      BTRFS_TREE_LOG_OBJECTID);
1401
1402         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1403         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1404         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1405
1406         /*
1407          * DON'T set REF_COWS for log trees
1408          *
1409          * log trees do not get reference counted because they go away
1410          * before a real commit is actually done.  They do store pointers
1411          * to file data extents, and those reference counts still get
1412          * updated (along with back refs to the log tree).
1413          */
1414
1415         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1416                         NULL, 0, 0, 0);
1417         if (IS_ERR(leaf)) {
1418                 kfree(root);
1419                 return ERR_CAST(leaf);
1420         }
1421
1422         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1423         btrfs_set_header_bytenr(leaf, leaf->start);
1424         btrfs_set_header_generation(leaf, trans->transid);
1425         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1426         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1427         root->node = leaf;
1428
1429         write_extent_buffer(root->node, root->fs_info->fsid,
1430                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1431         btrfs_mark_buffer_dirty(root->node);
1432         btrfs_tree_unlock(root->node);
1433         return root;
1434 }
1435
1436 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1437                              struct btrfs_fs_info *fs_info)
1438 {
1439         struct btrfs_root *log_root;
1440
1441         log_root = alloc_log_tree(trans, fs_info);
1442         if (IS_ERR(log_root))
1443                 return PTR_ERR(log_root);
1444         WARN_ON(fs_info->log_root_tree);
1445         fs_info->log_root_tree = log_root;
1446         return 0;
1447 }
1448
1449 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1450                        struct btrfs_root *root)
1451 {
1452         struct btrfs_root *log_root;
1453         struct btrfs_inode_item *inode_item;
1454
1455         log_root = alloc_log_tree(trans, root->fs_info);
1456         if (IS_ERR(log_root))
1457                 return PTR_ERR(log_root);
1458
1459         log_root->last_trans = trans->transid;
1460         log_root->root_key.offset = root->root_key.objectid;
1461
1462         inode_item = &log_root->root_item.inode;
1463         btrfs_set_stack_inode_generation(inode_item, 1);
1464         btrfs_set_stack_inode_size(inode_item, 3);
1465         btrfs_set_stack_inode_nlink(inode_item, 1);
1466         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1467         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1468
1469         btrfs_set_root_node(&log_root->root_item, log_root->node);
1470
1471         WARN_ON(root->log_root);
1472         root->log_root = log_root;
1473         root->log_transid = 0;
1474         root->log_transid_committed = -1;
1475         root->last_log_commit = 0;
1476         return 0;
1477 }
1478
1479 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1480                                                struct btrfs_key *key)
1481 {
1482         struct btrfs_root *root;
1483         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1484         struct btrfs_path *path;
1485         u64 generation;
1486         int ret;
1487
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         root = btrfs_alloc_root(fs_info);
1493         if (!root) {
1494                 ret = -ENOMEM;
1495                 goto alloc_fail;
1496         }
1497
1498         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1499                 tree_root->stripesize, root, fs_info, key->objectid);
1500
1501         ret = btrfs_find_root(tree_root, key, path,
1502                               &root->root_item, &root->root_key);
1503         if (ret) {
1504                 if (ret > 0)
1505                         ret = -ENOENT;
1506                 goto find_fail;
1507         }
1508
1509         generation = btrfs_root_generation(&root->root_item);
1510         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1511                                      generation);
1512         if (!root->node) {
1513                 ret = -ENOMEM;
1514                 goto find_fail;
1515         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1516                 ret = -EIO;
1517                 goto read_fail;
1518         }
1519         root->commit_root = btrfs_root_node(root);
1520 out:
1521         btrfs_free_path(path);
1522         return root;
1523
1524 read_fail:
1525         free_extent_buffer(root->node);
1526 find_fail:
1527         kfree(root);
1528 alloc_fail:
1529         root = ERR_PTR(ret);
1530         goto out;
1531 }
1532
1533 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1534                                       struct btrfs_key *location)
1535 {
1536         struct btrfs_root *root;
1537
1538         root = btrfs_read_tree_root(tree_root, location);
1539         if (IS_ERR(root))
1540                 return root;
1541
1542         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1543                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1544                 btrfs_check_and_init_root_item(&root->root_item);
1545         }
1546
1547         return root;
1548 }
1549
1550 int btrfs_init_fs_root(struct btrfs_root *root)
1551 {
1552         int ret;
1553         struct btrfs_subvolume_writers *writers;
1554
1555         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1556         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1557                                         GFP_NOFS);
1558         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1559                 ret = -ENOMEM;
1560                 goto fail;
1561         }
1562
1563         writers = btrfs_alloc_subvolume_writers();
1564         if (IS_ERR(writers)) {
1565                 ret = PTR_ERR(writers);
1566                 goto fail;
1567         }
1568         root->subv_writers = writers;
1569
1570         btrfs_init_free_ino_ctl(root);
1571         spin_lock_init(&root->ino_cache_lock);
1572         init_waitqueue_head(&root->ino_cache_wait);
1573
1574         ret = get_anon_bdev(&root->anon_dev);
1575         if (ret)
1576                 goto free_writers;
1577
1578         mutex_lock(&root->objectid_mutex);
1579         ret = btrfs_find_highest_objectid(root,
1580                                         &root->highest_objectid);
1581         if (ret) {
1582                 mutex_unlock(&root->objectid_mutex);
1583                 goto free_root_dev;
1584         }
1585
1586         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1587
1588         mutex_unlock(&root->objectid_mutex);
1589
1590         return 0;
1591
1592 free_root_dev:
1593         free_anon_bdev(root->anon_dev);
1594 free_writers:
1595         btrfs_free_subvolume_writers(root->subv_writers);
1596 fail:
1597         kfree(root->free_ino_ctl);
1598         kfree(root->free_ino_pinned);
1599         return ret;
1600 }
1601
1602 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1603                                                u64 root_id)
1604 {
1605         struct btrfs_root *root;
1606
1607         spin_lock(&fs_info->fs_roots_radix_lock);
1608         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1609                                  (unsigned long)root_id);
1610         spin_unlock(&fs_info->fs_roots_radix_lock);
1611         return root;
1612 }
1613
1614 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1615                          struct btrfs_root *root)
1616 {
1617         int ret;
1618
1619         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1620         if (ret)
1621                 return ret;
1622
1623         spin_lock(&fs_info->fs_roots_radix_lock);
1624         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1625                                 (unsigned long)root->root_key.objectid,
1626                                 root);
1627         if (ret == 0)
1628                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1629         spin_unlock(&fs_info->fs_roots_radix_lock);
1630         radix_tree_preload_end();
1631
1632         return ret;
1633 }
1634
1635 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1636                                      struct btrfs_key *location,
1637                                      bool check_ref)
1638 {
1639         struct btrfs_root *root;
1640         struct btrfs_path *path;
1641         struct btrfs_key key;
1642         int ret;
1643
1644         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1645                 return fs_info->tree_root;
1646         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1647                 return fs_info->extent_root;
1648         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1649                 return fs_info->chunk_root;
1650         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1651                 return fs_info->dev_root;
1652         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1653                 return fs_info->csum_root;
1654         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1655                 return fs_info->quota_root ? fs_info->quota_root :
1656                                              ERR_PTR(-ENOENT);
1657         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1658                 return fs_info->uuid_root ? fs_info->uuid_root :
1659                                             ERR_PTR(-ENOENT);
1660 again:
1661         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1662         if (root) {
1663                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1664                         return ERR_PTR(-ENOENT);
1665                 return root;
1666         }
1667
1668         root = btrfs_read_fs_root(fs_info->tree_root, location);
1669         if (IS_ERR(root))
1670                 return root;
1671
1672         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1673                 ret = -ENOENT;
1674                 goto fail;
1675         }
1676
1677         ret = btrfs_init_fs_root(root);
1678         if (ret)
1679                 goto fail;
1680
1681         path = btrfs_alloc_path();
1682         if (!path) {
1683                 ret = -ENOMEM;
1684                 goto fail;
1685         }
1686         key.objectid = BTRFS_ORPHAN_OBJECTID;
1687         key.type = BTRFS_ORPHAN_ITEM_KEY;
1688         key.offset = location->objectid;
1689
1690         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1691         btrfs_free_path(path);
1692         if (ret < 0)
1693                 goto fail;
1694         if (ret == 0)
1695                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1696
1697         ret = btrfs_insert_fs_root(fs_info, root);
1698         if (ret) {
1699                 if (ret == -EEXIST) {
1700                         free_fs_root(root);
1701                         goto again;
1702                 }
1703                 goto fail;
1704         }
1705         return root;
1706 fail:
1707         free_fs_root(root);
1708         return ERR_PTR(ret);
1709 }
1710
1711 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1712 {
1713         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1714         int ret = 0;
1715         struct btrfs_device *device;
1716         struct backing_dev_info *bdi;
1717
1718         rcu_read_lock();
1719         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1720                 if (!device->bdev)
1721                         continue;
1722                 bdi = blk_get_backing_dev_info(device->bdev);
1723                 if (bdi_congested(bdi, bdi_bits)) {
1724                         ret = 1;
1725                         break;
1726                 }
1727         }
1728         rcu_read_unlock();
1729         return ret;
1730 }
1731
1732 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1733 {
1734         int err;
1735
1736         err = bdi_setup_and_register(bdi, "btrfs");
1737         if (err)
1738                 return err;
1739
1740         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1741         bdi->congested_fn       = btrfs_congested_fn;
1742         bdi->congested_data     = info;
1743         return 0;
1744 }
1745
1746 /*
1747  * called by the kthread helper functions to finally call the bio end_io
1748  * functions.  This is where read checksum verification actually happens
1749  */
1750 static void end_workqueue_fn(struct btrfs_work *work)
1751 {
1752         struct bio *bio;
1753         struct btrfs_end_io_wq *end_io_wq;
1754         int error;
1755
1756         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1757         bio = end_io_wq->bio;
1758
1759         error = end_io_wq->error;
1760         bio->bi_private = end_io_wq->private;
1761         bio->bi_end_io = end_io_wq->end_io;
1762         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1763         bio_endio_nodec(bio, error);
1764 }
1765
1766 static int cleaner_kthread(void *arg)
1767 {
1768         struct btrfs_root *root = arg;
1769         int again;
1770
1771         do {
1772                 again = 0;
1773
1774                 /* Make the cleaner go to sleep early. */
1775                 if (btrfs_need_cleaner_sleep(root))
1776                         goto sleep;
1777
1778                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1779                         goto sleep;
1780
1781                 /*
1782                  * Avoid the problem that we change the status of the fs
1783                  * during the above check and trylock.
1784                  */
1785                 if (btrfs_need_cleaner_sleep(root)) {
1786                         mutex_unlock(&root->fs_info->cleaner_mutex);
1787                         goto sleep;
1788                 }
1789
1790                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1791                 btrfs_run_delayed_iputs(root);
1792                 btrfs_delete_unused_bgs(root->fs_info);
1793                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1794
1795                 again = btrfs_clean_one_deleted_snapshot(root);
1796                 mutex_unlock(&root->fs_info->cleaner_mutex);
1797
1798                 /*
1799                  * The defragger has dealt with the R/O remount and umount,
1800                  * needn't do anything special here.
1801                  */
1802                 btrfs_run_defrag_inodes(root->fs_info);
1803 sleep:
1804                 if (!try_to_freeze() && !again) {
1805                         set_current_state(TASK_INTERRUPTIBLE);
1806                         if (!kthread_should_stop())
1807                                 schedule();
1808                         __set_current_state(TASK_RUNNING);
1809                 }
1810         } while (!kthread_should_stop());
1811         return 0;
1812 }
1813
1814 static int transaction_kthread(void *arg)
1815 {
1816         struct btrfs_root *root = arg;
1817         struct btrfs_trans_handle *trans;
1818         struct btrfs_transaction *cur;
1819         u64 transid;
1820         unsigned long now;
1821         unsigned long delay;
1822         bool cannot_commit;
1823
1824         do {
1825                 cannot_commit = false;
1826                 delay = HZ * root->fs_info->commit_interval;
1827                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1828
1829                 spin_lock(&root->fs_info->trans_lock);
1830                 cur = root->fs_info->running_transaction;
1831                 if (!cur) {
1832                         spin_unlock(&root->fs_info->trans_lock);
1833                         goto sleep;
1834                 }
1835
1836                 now = get_seconds();
1837                 if (cur->state < TRANS_STATE_BLOCKED &&
1838                     (now < cur->start_time ||
1839                      now - cur->start_time < root->fs_info->commit_interval)) {
1840                         spin_unlock(&root->fs_info->trans_lock);
1841                         delay = HZ * 5;
1842                         goto sleep;
1843                 }
1844                 transid = cur->transid;
1845                 spin_unlock(&root->fs_info->trans_lock);
1846
1847                 /* If the file system is aborted, this will always fail. */
1848                 trans = btrfs_attach_transaction(root);
1849                 if (IS_ERR(trans)) {
1850                         if (PTR_ERR(trans) != -ENOENT)
1851                                 cannot_commit = true;
1852                         goto sleep;
1853                 }
1854                 if (transid == trans->transid) {
1855                         btrfs_commit_transaction(trans, root);
1856                 } else {
1857                         btrfs_end_transaction(trans, root);
1858                 }
1859 sleep:
1860                 wake_up_process(root->fs_info->cleaner_kthread);
1861                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1862
1863                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1864                                       &root->fs_info->fs_state)))
1865                         btrfs_cleanup_transaction(root);
1866                 if (!try_to_freeze()) {
1867                         set_current_state(TASK_INTERRUPTIBLE);
1868                         if (!kthread_should_stop() &&
1869                             (!btrfs_transaction_blocked(root->fs_info) ||
1870                              cannot_commit))
1871                                 schedule_timeout(delay);
1872                         __set_current_state(TASK_RUNNING);
1873                 }
1874         } while (!kthread_should_stop());
1875         return 0;
1876 }
1877
1878 /*
1879  * this will find the highest generation in the array of
1880  * root backups.  The index of the highest array is returned,
1881  * or -1 if we can't find anything.
1882  *
1883  * We check to make sure the array is valid by comparing the
1884  * generation of the latest  root in the array with the generation
1885  * in the super block.  If they don't match we pitch it.
1886  */
1887 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1888 {
1889         u64 cur;
1890         int newest_index = -1;
1891         struct btrfs_root_backup *root_backup;
1892         int i;
1893
1894         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1895                 root_backup = info->super_copy->super_roots + i;
1896                 cur = btrfs_backup_tree_root_gen(root_backup);
1897                 if (cur == newest_gen)
1898                         newest_index = i;
1899         }
1900
1901         /* check to see if we actually wrapped around */
1902         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1903                 root_backup = info->super_copy->super_roots;
1904                 cur = btrfs_backup_tree_root_gen(root_backup);
1905                 if (cur == newest_gen)
1906                         newest_index = 0;
1907         }
1908         return newest_index;
1909 }
1910
1911
1912 /*
1913  * find the oldest backup so we know where to store new entries
1914  * in the backup array.  This will set the backup_root_index
1915  * field in the fs_info struct
1916  */
1917 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1918                                      u64 newest_gen)
1919 {
1920         int newest_index = -1;
1921
1922         newest_index = find_newest_super_backup(info, newest_gen);
1923         /* if there was garbage in there, just move along */
1924         if (newest_index == -1) {
1925                 info->backup_root_index = 0;
1926         } else {
1927                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1928         }
1929 }
1930
1931 /*
1932  * copy all the root pointers into the super backup array.
1933  * this will bump the backup pointer by one when it is
1934  * done
1935  */
1936 static void backup_super_roots(struct btrfs_fs_info *info)
1937 {
1938         int next_backup;
1939         struct btrfs_root_backup *root_backup;
1940         int last_backup;
1941
1942         next_backup = info->backup_root_index;
1943         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1944                 BTRFS_NUM_BACKUP_ROOTS;
1945
1946         /*
1947          * just overwrite the last backup if we're at the same generation
1948          * this happens only at umount
1949          */
1950         root_backup = info->super_for_commit->super_roots + last_backup;
1951         if (btrfs_backup_tree_root_gen(root_backup) ==
1952             btrfs_header_generation(info->tree_root->node))
1953                 next_backup = last_backup;
1954
1955         root_backup = info->super_for_commit->super_roots + next_backup;
1956
1957         /*
1958          * make sure all of our padding and empty slots get zero filled
1959          * regardless of which ones we use today
1960          */
1961         memset(root_backup, 0, sizeof(*root_backup));
1962
1963         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1964
1965         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1966         btrfs_set_backup_tree_root_gen(root_backup,
1967                                btrfs_header_generation(info->tree_root->node));
1968
1969         btrfs_set_backup_tree_root_level(root_backup,
1970                                btrfs_header_level(info->tree_root->node));
1971
1972         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1973         btrfs_set_backup_chunk_root_gen(root_backup,
1974                                btrfs_header_generation(info->chunk_root->node));
1975         btrfs_set_backup_chunk_root_level(root_backup,
1976                                btrfs_header_level(info->chunk_root->node));
1977
1978         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1979         btrfs_set_backup_extent_root_gen(root_backup,
1980                                btrfs_header_generation(info->extent_root->node));
1981         btrfs_set_backup_extent_root_level(root_backup,
1982                                btrfs_header_level(info->extent_root->node));
1983
1984         /*
1985          * we might commit during log recovery, which happens before we set
1986          * the fs_root.  Make sure it is valid before we fill it in.
1987          */
1988         if (info->fs_root && info->fs_root->node) {
1989                 btrfs_set_backup_fs_root(root_backup,
1990                                          info->fs_root->node->start);
1991                 btrfs_set_backup_fs_root_gen(root_backup,
1992                                btrfs_header_generation(info->fs_root->node));
1993                 btrfs_set_backup_fs_root_level(root_backup,
1994                                btrfs_header_level(info->fs_root->node));
1995         }
1996
1997         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1998         btrfs_set_backup_dev_root_gen(root_backup,
1999                                btrfs_header_generation(info->dev_root->node));
2000         btrfs_set_backup_dev_root_level(root_backup,
2001                                        btrfs_header_level(info->dev_root->node));
2002
2003         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2004         btrfs_set_backup_csum_root_gen(root_backup,
2005                                btrfs_header_generation(info->csum_root->node));
2006         btrfs_set_backup_csum_root_level(root_backup,
2007                                btrfs_header_level(info->csum_root->node));
2008
2009         btrfs_set_backup_total_bytes(root_backup,
2010                              btrfs_super_total_bytes(info->super_copy));
2011         btrfs_set_backup_bytes_used(root_backup,
2012                              btrfs_super_bytes_used(info->super_copy));
2013         btrfs_set_backup_num_devices(root_backup,
2014                              btrfs_super_num_devices(info->super_copy));
2015
2016         /*
2017          * if we don't copy this out to the super_copy, it won't get remembered
2018          * for the next commit
2019          */
2020         memcpy(&info->super_copy->super_roots,
2021                &info->super_for_commit->super_roots,
2022                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2023 }
2024
2025 /*
2026  * this copies info out of the root backup array and back into
2027  * the in-memory super block.  It is meant to help iterate through
2028  * the array, so you send it the number of backups you've already
2029  * tried and the last backup index you used.
2030  *
2031  * this returns -1 when it has tried all the backups
2032  */
2033 static noinline int next_root_backup(struct btrfs_fs_info *info,
2034                                      struct btrfs_super_block *super,
2035                                      int *num_backups_tried, int *backup_index)
2036 {
2037         struct btrfs_root_backup *root_backup;
2038         int newest = *backup_index;
2039
2040         if (*num_backups_tried == 0) {
2041                 u64 gen = btrfs_super_generation(super);
2042
2043                 newest = find_newest_super_backup(info, gen);
2044                 if (newest == -1)
2045                         return -1;
2046
2047                 *backup_index = newest;
2048                 *num_backups_tried = 1;
2049         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2050                 /* we've tried all the backups, all done */
2051                 return -1;
2052         } else {
2053                 /* jump to the next oldest backup */
2054                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2055                         BTRFS_NUM_BACKUP_ROOTS;
2056                 *backup_index = newest;
2057                 *num_backups_tried += 1;
2058         }
2059         root_backup = super->super_roots + newest;
2060
2061         btrfs_set_super_generation(super,
2062                                    btrfs_backup_tree_root_gen(root_backup));
2063         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2064         btrfs_set_super_root_level(super,
2065                                    btrfs_backup_tree_root_level(root_backup));
2066         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2067
2068         /*
2069          * fixme: the total bytes and num_devices need to match or we should
2070          * need a fsck
2071          */
2072         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2073         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2074         return 0;
2075 }
2076
2077 /* helper to cleanup workers */
2078 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2079 {
2080         btrfs_destroy_workqueue(fs_info->fixup_workers);
2081         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2082         btrfs_destroy_workqueue(fs_info->workers);
2083         btrfs_destroy_workqueue(fs_info->endio_workers);
2084         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2085         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2086         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2087         btrfs_destroy_workqueue(fs_info->rmw_workers);
2088         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2089         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2090         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2091         btrfs_destroy_workqueue(fs_info->submit_workers);
2092         btrfs_destroy_workqueue(fs_info->delayed_workers);
2093         btrfs_destroy_workqueue(fs_info->caching_workers);
2094         btrfs_destroy_workqueue(fs_info->readahead_workers);
2095         btrfs_destroy_workqueue(fs_info->flush_workers);
2096         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2097         btrfs_destroy_workqueue(fs_info->extent_workers);
2098 }
2099
2100 static void free_root_extent_buffers(struct btrfs_root *root)
2101 {
2102         if (root) {
2103                 free_extent_buffer(root->node);
2104                 free_extent_buffer(root->commit_root);
2105                 root->node = NULL;
2106                 root->commit_root = NULL;
2107         }
2108 }
2109
2110 /* helper to cleanup tree roots */
2111 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2112 {
2113         free_root_extent_buffers(info->tree_root);
2114
2115         free_root_extent_buffers(info->dev_root);
2116         free_root_extent_buffers(info->extent_root);
2117         free_root_extent_buffers(info->csum_root);
2118         free_root_extent_buffers(info->quota_root);
2119         free_root_extent_buffers(info->uuid_root);
2120         if (chunk_root)
2121                 free_root_extent_buffers(info->chunk_root);
2122 }
2123
2124 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2125 {
2126         int ret;
2127         struct btrfs_root *gang[8];
2128         int i;
2129
2130         while (!list_empty(&fs_info->dead_roots)) {
2131                 gang[0] = list_entry(fs_info->dead_roots.next,
2132                                      struct btrfs_root, root_list);
2133                 list_del(&gang[0]->root_list);
2134
2135                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2136                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2137                 } else {
2138                         free_extent_buffer(gang[0]->node);
2139                         free_extent_buffer(gang[0]->commit_root);
2140                         btrfs_put_fs_root(gang[0]);
2141                 }
2142         }
2143
2144         while (1) {
2145                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2146                                              (void **)gang, 0,
2147                                              ARRAY_SIZE(gang));
2148                 if (!ret)
2149                         break;
2150                 for (i = 0; i < ret; i++)
2151                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2152         }
2153
2154         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2155                 btrfs_free_log_root_tree(NULL, fs_info);
2156                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2157                                             fs_info->pinned_extents);
2158         }
2159 }
2160
2161 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2162 {
2163         mutex_init(&fs_info->scrub_lock);
2164         atomic_set(&fs_info->scrubs_running, 0);
2165         atomic_set(&fs_info->scrub_pause_req, 0);
2166         atomic_set(&fs_info->scrubs_paused, 0);
2167         atomic_set(&fs_info->scrub_cancel_req, 0);
2168         init_waitqueue_head(&fs_info->scrub_pause_wait);
2169         fs_info->scrub_workers_refcnt = 0;
2170 }
2171
2172 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2173 {
2174         spin_lock_init(&fs_info->balance_lock);
2175         mutex_init(&fs_info->balance_mutex);
2176         atomic_set(&fs_info->balance_running, 0);
2177         atomic_set(&fs_info->balance_pause_req, 0);
2178         atomic_set(&fs_info->balance_cancel_req, 0);
2179         fs_info->balance_ctl = NULL;
2180         init_waitqueue_head(&fs_info->balance_wait_q);
2181 }
2182
2183 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2184                                    struct btrfs_root *tree_root)
2185 {
2186         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2187         set_nlink(fs_info->btree_inode, 1);
2188         /*
2189          * we set the i_size on the btree inode to the max possible int.
2190          * the real end of the address space is determined by all of
2191          * the devices in the system
2192          */
2193         fs_info->btree_inode->i_size = OFFSET_MAX;
2194         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2195
2196         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2197         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2198                              fs_info->btree_inode->i_mapping);
2199         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2200         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2201
2202         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2203
2204         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2205         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2206                sizeof(struct btrfs_key));
2207         set_bit(BTRFS_INODE_DUMMY,
2208                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2209         btrfs_insert_inode_hash(fs_info->btree_inode);
2210 }
2211
2212 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2213 {
2214         fs_info->dev_replace.lock_owner = 0;
2215         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2216         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2217         mutex_init(&fs_info->dev_replace.lock_management_lock);
2218         mutex_init(&fs_info->dev_replace.lock);
2219         init_waitqueue_head(&fs_info->replace_wait);
2220 }
2221
2222 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2223 {
2224         spin_lock_init(&fs_info->qgroup_lock);
2225         mutex_init(&fs_info->qgroup_ioctl_lock);
2226         fs_info->qgroup_tree = RB_ROOT;
2227         fs_info->qgroup_op_tree = RB_ROOT;
2228         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2229         fs_info->qgroup_seq = 1;
2230         fs_info->quota_enabled = 0;
2231         fs_info->pending_quota_state = 0;
2232         fs_info->qgroup_ulist = NULL;
2233         mutex_init(&fs_info->qgroup_rescan_lock);
2234 }
2235
2236 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2237                 struct btrfs_fs_devices *fs_devices)
2238 {
2239         int max_active = fs_info->thread_pool_size;
2240         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2241
2242         fs_info->workers =
2243                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2244                                       max_active, 16);
2245
2246         fs_info->delalloc_workers =
2247                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2248
2249         fs_info->flush_workers =
2250                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2251
2252         fs_info->caching_workers =
2253                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2254
2255         /*
2256          * a higher idle thresh on the submit workers makes it much more
2257          * likely that bios will be send down in a sane order to the
2258          * devices
2259          */
2260         fs_info->submit_workers =
2261                 btrfs_alloc_workqueue("submit", flags,
2262                                       min_t(u64, fs_devices->num_devices,
2263                                             max_active), 64);
2264
2265         fs_info->fixup_workers =
2266                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2267
2268         /*
2269          * endios are largely parallel and should have a very
2270          * low idle thresh
2271          */
2272         fs_info->endio_workers =
2273                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2274         fs_info->endio_meta_workers =
2275                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2276         fs_info->endio_meta_write_workers =
2277                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2278         fs_info->endio_raid56_workers =
2279                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2280         fs_info->endio_repair_workers =
2281                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2282         fs_info->rmw_workers =
2283                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2284         fs_info->endio_write_workers =
2285                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2286         fs_info->endio_freespace_worker =
2287                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2288         fs_info->delayed_workers =
2289                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2290         fs_info->readahead_workers =
2291                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2292         fs_info->qgroup_rescan_workers =
2293                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2294         fs_info->extent_workers =
2295                 btrfs_alloc_workqueue("extent-refs", flags,
2296                                       min_t(u64, fs_devices->num_devices,
2297                                             max_active), 8);
2298
2299         if (!(fs_info->workers && fs_info->delalloc_workers &&
2300               fs_info->submit_workers && fs_info->flush_workers &&
2301               fs_info->endio_workers && fs_info->endio_meta_workers &&
2302               fs_info->endio_meta_write_workers &&
2303               fs_info->endio_repair_workers &&
2304               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2305               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2306               fs_info->caching_workers && fs_info->readahead_workers &&
2307               fs_info->fixup_workers && fs_info->delayed_workers &&
2308               fs_info->extent_workers &&
2309               fs_info->qgroup_rescan_workers)) {
2310                 return -ENOMEM;
2311         }
2312
2313         return 0;
2314 }
2315
2316 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2317                             struct btrfs_fs_devices *fs_devices)
2318 {
2319         int ret;
2320         struct btrfs_root *tree_root = fs_info->tree_root;
2321         struct btrfs_root *log_tree_root;
2322         struct btrfs_super_block *disk_super = fs_info->super_copy;
2323         u64 bytenr = btrfs_super_log_root(disk_super);
2324
2325         if (fs_devices->rw_devices == 0) {
2326                 printk(KERN_WARNING "BTRFS: log replay required "
2327                        "on RO media\n");
2328                 return -EIO;
2329         }
2330
2331         log_tree_root = btrfs_alloc_root(fs_info);
2332         if (!log_tree_root)
2333                 return -ENOMEM;
2334
2335         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2336                         tree_root->stripesize, log_tree_root, fs_info,
2337                         BTRFS_TREE_LOG_OBJECTID);
2338
2339         log_tree_root->node = read_tree_block(tree_root, bytenr,
2340                         fs_info->generation + 1);
2341         if (!log_tree_root->node ||
2342             !extent_buffer_uptodate(log_tree_root->node)) {
2343                 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2344                 free_extent_buffer(log_tree_root->node);
2345                 kfree(log_tree_root);
2346                 return -EIO;
2347         }
2348         /* returns with log_tree_root freed on success */
2349         ret = btrfs_recover_log_trees(log_tree_root);
2350         if (ret) {
2351                 btrfs_error(tree_root->fs_info, ret,
2352                             "Failed to recover log tree");
2353                 free_extent_buffer(log_tree_root->node);
2354                 kfree(log_tree_root);
2355                 return ret;
2356         }
2357
2358         if (fs_info->sb->s_flags & MS_RDONLY) {
2359                 ret = btrfs_commit_super(tree_root);
2360                 if (ret)
2361                         return ret;
2362         }
2363
2364         return 0;
2365 }
2366
2367 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2368                             struct btrfs_root *tree_root)
2369 {
2370         struct btrfs_root *root;
2371         struct btrfs_key location;
2372         int ret;
2373
2374         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2375         location.type = BTRFS_ROOT_ITEM_KEY;
2376         location.offset = 0;
2377
2378         root = btrfs_read_tree_root(tree_root, &location);
2379         if (IS_ERR(root))
2380                 return PTR_ERR(root);
2381         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382         fs_info->extent_root = root;
2383
2384         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2385         root = btrfs_read_tree_root(tree_root, &location);
2386         if (IS_ERR(root))
2387                 return PTR_ERR(root);
2388         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389         fs_info->dev_root = root;
2390         btrfs_init_devices_late(fs_info);
2391
2392         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2393         root = btrfs_read_tree_root(tree_root, &location);
2394         if (IS_ERR(root))
2395                 return PTR_ERR(root);
2396         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2397         fs_info->csum_root = root;
2398
2399         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2400         root = btrfs_read_tree_root(tree_root, &location);
2401         if (!IS_ERR(root)) {
2402                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2403                 fs_info->quota_enabled = 1;
2404                 fs_info->pending_quota_state = 1;
2405                 fs_info->quota_root = root;
2406         }
2407
2408         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2409         root = btrfs_read_tree_root(tree_root, &location);
2410         if (IS_ERR(root)) {
2411                 ret = PTR_ERR(root);
2412                 if (ret != -ENOENT)
2413                         return ret;
2414         } else {
2415                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2416                 fs_info->uuid_root = root;
2417         }
2418
2419         return 0;
2420 }
2421
2422 int open_ctree(struct super_block *sb,
2423                struct btrfs_fs_devices *fs_devices,
2424                char *options)
2425 {
2426         u32 sectorsize;
2427         u32 nodesize;
2428         u32 stripesize;
2429         u64 generation;
2430         u64 features;
2431         struct btrfs_key location;
2432         struct buffer_head *bh;
2433         struct btrfs_super_block *disk_super;
2434         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2435         struct btrfs_root *tree_root;
2436         struct btrfs_root *chunk_root;
2437         int ret;
2438         int err = -EINVAL;
2439         int num_backups_tried = 0;
2440         int backup_index = 0;
2441         int max_active;
2442
2443         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2444         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2445         if (!tree_root || !chunk_root) {
2446                 err = -ENOMEM;
2447                 goto fail;
2448         }
2449
2450         ret = init_srcu_struct(&fs_info->subvol_srcu);
2451         if (ret) {
2452                 err = ret;
2453                 goto fail;
2454         }
2455
2456         ret = setup_bdi(fs_info, &fs_info->bdi);
2457         if (ret) {
2458                 err = ret;
2459                 goto fail_srcu;
2460         }
2461
2462         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2463         if (ret) {
2464                 err = ret;
2465                 goto fail_bdi;
2466         }
2467         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2468                                         (1 + ilog2(nr_cpu_ids));
2469
2470         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2471         if (ret) {
2472                 err = ret;
2473                 goto fail_dirty_metadata_bytes;
2474         }
2475
2476         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2477         if (ret) {
2478                 err = ret;
2479                 goto fail_delalloc_bytes;
2480         }
2481
2482         fs_info->btree_inode = new_inode(sb);
2483         if (!fs_info->btree_inode) {
2484                 err = -ENOMEM;
2485                 goto fail_bio_counter;
2486         }
2487
2488         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2489
2490         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2491         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2492         INIT_LIST_HEAD(&fs_info->trans_list);
2493         INIT_LIST_HEAD(&fs_info->dead_roots);
2494         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2495         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2496         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2497         spin_lock_init(&fs_info->delalloc_root_lock);
2498         spin_lock_init(&fs_info->trans_lock);
2499         spin_lock_init(&fs_info->fs_roots_radix_lock);
2500         spin_lock_init(&fs_info->delayed_iput_lock);
2501         spin_lock_init(&fs_info->defrag_inodes_lock);
2502         spin_lock_init(&fs_info->free_chunk_lock);
2503         spin_lock_init(&fs_info->tree_mod_seq_lock);
2504         spin_lock_init(&fs_info->super_lock);
2505         spin_lock_init(&fs_info->qgroup_op_lock);
2506         spin_lock_init(&fs_info->buffer_lock);
2507         spin_lock_init(&fs_info->unused_bgs_lock);
2508         rwlock_init(&fs_info->tree_mod_log_lock);
2509         mutex_init(&fs_info->unused_bg_unpin_mutex);
2510         mutex_init(&fs_info->reloc_mutex);
2511         mutex_init(&fs_info->delalloc_root_mutex);
2512         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2513         seqlock_init(&fs_info->profiles_lock);
2514
2515         init_completion(&fs_info->kobj_unregister);
2516         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2517         INIT_LIST_HEAD(&fs_info->space_info);
2518         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2519         INIT_LIST_HEAD(&fs_info->unused_bgs);
2520         btrfs_mapping_init(&fs_info->mapping_tree);
2521         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2522                              BTRFS_BLOCK_RSV_GLOBAL);
2523         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2524                              BTRFS_BLOCK_RSV_DELALLOC);
2525         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2526         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2527         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2528         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2529                              BTRFS_BLOCK_RSV_DELOPS);
2530         atomic_set(&fs_info->nr_async_submits, 0);
2531         atomic_set(&fs_info->async_delalloc_pages, 0);
2532         atomic_set(&fs_info->async_submit_draining, 0);
2533         atomic_set(&fs_info->nr_async_bios, 0);
2534         atomic_set(&fs_info->defrag_running, 0);
2535         atomic_set(&fs_info->qgroup_op_seq, 0);
2536         atomic64_set(&fs_info->tree_mod_seq, 0);
2537         fs_info->sb = sb;
2538         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2539         fs_info->metadata_ratio = 0;
2540         fs_info->defrag_inodes = RB_ROOT;
2541         fs_info->free_chunk_space = 0;
2542         fs_info->tree_mod_log = RB_ROOT;
2543         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2544         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2545         /* readahead state */
2546         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2547         spin_lock_init(&fs_info->reada_lock);
2548
2549         fs_info->thread_pool_size = min_t(unsigned long,
2550                                           num_online_cpus() + 2, 8);
2551
2552         INIT_LIST_HEAD(&fs_info->ordered_roots);
2553         spin_lock_init(&fs_info->ordered_root_lock);
2554         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2555                                         GFP_NOFS);
2556         if (!fs_info->delayed_root) {
2557                 err = -ENOMEM;
2558                 goto fail_iput;
2559         }
2560         btrfs_init_delayed_root(fs_info->delayed_root);
2561
2562         btrfs_init_scrub(fs_info);
2563 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2564         fs_info->check_integrity_print_mask = 0;
2565 #endif
2566         btrfs_init_balance(fs_info);
2567         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2568
2569         sb->s_blocksize = 4096;
2570         sb->s_blocksize_bits = blksize_bits(4096);
2571         sb->s_bdi = &fs_info->bdi;
2572
2573         btrfs_init_btree_inode(fs_info, tree_root);
2574
2575         spin_lock_init(&fs_info->block_group_cache_lock);
2576         fs_info->block_group_cache_tree = RB_ROOT;
2577         fs_info->first_logical_byte = (u64)-1;
2578
2579         extent_io_tree_init(&fs_info->freed_extents[0],
2580                              fs_info->btree_inode->i_mapping);
2581         extent_io_tree_init(&fs_info->freed_extents[1],
2582                              fs_info->btree_inode->i_mapping);
2583         fs_info->pinned_extents = &fs_info->freed_extents[0];
2584         fs_info->do_barriers = 1;
2585
2586
2587         mutex_init(&fs_info->ordered_operations_mutex);
2588         mutex_init(&fs_info->ordered_extent_flush_mutex);
2589         mutex_init(&fs_info->tree_log_mutex);
2590         mutex_init(&fs_info->chunk_mutex);
2591         mutex_init(&fs_info->transaction_kthread_mutex);
2592         mutex_init(&fs_info->cleaner_mutex);
2593         mutex_init(&fs_info->volume_mutex);
2594         mutex_init(&fs_info->ro_block_group_mutex);
2595         init_rwsem(&fs_info->commit_root_sem);
2596         init_rwsem(&fs_info->cleanup_work_sem);
2597         init_rwsem(&fs_info->subvol_sem);
2598         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2599
2600         btrfs_init_dev_replace_locks(fs_info);
2601         btrfs_init_qgroup(fs_info);
2602
2603         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2604         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2605
2606         init_waitqueue_head(&fs_info->transaction_throttle);
2607         init_waitqueue_head(&fs_info->transaction_wait);
2608         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2609         init_waitqueue_head(&fs_info->async_submit_wait);
2610
2611         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2612
2613         ret = btrfs_alloc_stripe_hash_table(fs_info);
2614         if (ret) {
2615                 err = ret;
2616                 goto fail_alloc;
2617         }
2618
2619         __setup_root(4096, 4096, 4096, tree_root,
2620                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2621
2622         invalidate_bdev(fs_devices->latest_bdev);
2623
2624         /*
2625          * Read super block and check the signature bytes only
2626          */
2627         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2628         if (!bh) {
2629                 err = -EINVAL;
2630                 goto fail_alloc;
2631         }
2632
2633         /*
2634          * We want to check superblock checksum, the type is stored inside.
2635          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2636          */
2637         if (btrfs_check_super_csum(bh->b_data)) {
2638                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2639                 err = -EINVAL;
2640                 brelse(bh);
2641                 goto fail_alloc;
2642         }
2643
2644         /*
2645          * super_copy is zeroed at allocation time and we never touch the
2646          * following bytes up to INFO_SIZE, the checksum is calculated from
2647          * the whole block of INFO_SIZE
2648          */
2649         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2650         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2651                sizeof(*fs_info->super_for_commit));
2652         brelse(bh);
2653
2654         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2655
2656         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2657         if (ret) {
2658                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2659                 err = -EINVAL;
2660                 goto fail_alloc;
2661         }
2662
2663         disk_super = fs_info->super_copy;
2664         if (!btrfs_super_root(disk_super))
2665                 goto fail_alloc;
2666
2667         /* check FS state, whether FS is broken. */
2668         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2669                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2670
2671         /*
2672          * run through our array of backup supers and setup
2673          * our ring pointer to the oldest one
2674          */
2675         generation = btrfs_super_generation(disk_super);
2676         find_oldest_super_backup(fs_info, generation);
2677
2678         /*
2679          * In the long term, we'll store the compression type in the super
2680          * block, and it'll be used for per file compression control.
2681          */
2682         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2683
2684         ret = btrfs_parse_options(tree_root, options);
2685         if (ret) {
2686                 err = ret;
2687                 goto fail_alloc;
2688         }
2689
2690         features = btrfs_super_incompat_flags(disk_super) &
2691                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2692         if (features) {
2693                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2694                        "unsupported optional features (%Lx).\n",
2695                        features);
2696                 err = -EINVAL;
2697                 goto fail_alloc;
2698         }
2699
2700         /*
2701          * Leafsize and nodesize were always equal, this is only a sanity check.
2702          */
2703         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2704             btrfs_super_nodesize(disk_super)) {
2705                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2706                        "blocksizes don't match.  node %d leaf %d\n",
2707                        btrfs_super_nodesize(disk_super),
2708                        le32_to_cpu(disk_super->__unused_leafsize));
2709                 err = -EINVAL;
2710                 goto fail_alloc;
2711         }
2712         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2713                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2714                        "blocksize (%d) was too large\n",
2715                        btrfs_super_nodesize(disk_super));
2716                 err = -EINVAL;
2717                 goto fail_alloc;
2718         }
2719
2720         features = btrfs_super_incompat_flags(disk_super);
2721         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2722         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2723                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2724
2725         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2726                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2727
2728         /*
2729          * flag our filesystem as having big metadata blocks if
2730          * they are bigger than the page size
2731          */
2732         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2733                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2734                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2735                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2736         }
2737
2738         nodesize = btrfs_super_nodesize(disk_super);
2739         sectorsize = btrfs_super_sectorsize(disk_super);
2740         stripesize = btrfs_super_stripesize(disk_super);
2741         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2742         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2743
2744         /*
2745          * mixed block groups end up with duplicate but slightly offset
2746          * extent buffers for the same range.  It leads to corruptions
2747          */
2748         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2749             (sectorsize != nodesize)) {
2750                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2751                                 "are not allowed for mixed block groups on %s\n",
2752                                 sb->s_id);
2753                 goto fail_alloc;
2754         }
2755
2756         /*
2757          * Needn't use the lock because there is no other task which will
2758          * update the flag.
2759          */
2760         btrfs_set_super_incompat_flags(disk_super, features);
2761
2762         features = btrfs_super_compat_ro_flags(disk_super) &
2763                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2764         if (!(sb->s_flags & MS_RDONLY) && features) {
2765                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2766                        "unsupported option features (%Lx).\n",
2767                        features);
2768                 err = -EINVAL;
2769                 goto fail_alloc;
2770         }
2771
2772         max_active = fs_info->thread_pool_size;
2773
2774         ret = btrfs_init_workqueues(fs_info, fs_devices);
2775         if (ret) {
2776                 err = ret;
2777                 goto fail_sb_buffer;
2778         }
2779
2780         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2781         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2782                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2783
2784         tree_root->nodesize = nodesize;
2785         tree_root->sectorsize = sectorsize;
2786         tree_root->stripesize = stripesize;
2787
2788         sb->s_blocksize = sectorsize;
2789         sb->s_blocksize_bits = blksize_bits(sectorsize);
2790
2791         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2792                 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2793                 goto fail_sb_buffer;
2794         }
2795
2796         if (sectorsize != PAGE_SIZE) {
2797                 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2798                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2799                 goto fail_sb_buffer;
2800         }
2801
2802         mutex_lock(&fs_info->chunk_mutex);
2803         ret = btrfs_read_sys_array(tree_root);
2804         mutex_unlock(&fs_info->chunk_mutex);
2805         if (ret) {
2806                 printk(KERN_ERR "BTRFS: failed to read the system "
2807                        "array on %s\n", sb->s_id);
2808                 goto fail_sb_buffer;
2809         }
2810
2811         generation = btrfs_super_chunk_root_generation(disk_super);
2812
2813         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2814                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2815
2816         chunk_root->node = read_tree_block(chunk_root,
2817                                            btrfs_super_chunk_root(disk_super),
2818                                            generation);
2819         if (!chunk_root->node ||
2820             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2821                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2822                        sb->s_id);
2823                 goto fail_tree_roots;
2824         }
2825         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2826         chunk_root->commit_root = btrfs_root_node(chunk_root);
2827
2828         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2829            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2830
2831         ret = btrfs_read_chunk_tree(chunk_root);
2832         if (ret) {
2833                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2834                        sb->s_id);
2835                 goto fail_tree_roots;
2836         }
2837
2838         /*
2839          * keep the device that is marked to be the target device for the
2840          * dev_replace procedure
2841          */
2842         btrfs_close_extra_devices(fs_devices, 0);
2843
2844         if (!fs_devices->latest_bdev) {
2845                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2846                        sb->s_id);
2847                 goto fail_tree_roots;
2848         }
2849
2850 retry_root_backup:
2851         generation = btrfs_super_generation(disk_super);
2852
2853         tree_root->node = read_tree_block(tree_root,
2854                                           btrfs_super_root(disk_super),
2855                                           generation);
2856         if (!tree_root->node ||
2857             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2858                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2859                        sb->s_id);
2860
2861                 goto recovery_tree_root;
2862         }
2863
2864         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2865         tree_root->commit_root = btrfs_root_node(tree_root);
2866         btrfs_set_root_refs(&tree_root->root_item, 1);
2867
2868         mutex_lock(&tree_root->objectid_mutex);
2869         ret = btrfs_find_highest_objectid(tree_root,
2870                                         &tree_root->highest_objectid);
2871         if (ret) {
2872                 mutex_unlock(&tree_root->objectid_mutex);
2873                 goto recovery_tree_root;
2874         }
2875
2876         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2877
2878         mutex_unlock(&tree_root->objectid_mutex);
2879
2880         ret = btrfs_read_roots(fs_info, tree_root);
2881         if (ret)
2882                 goto recovery_tree_root;
2883
2884         fs_info->generation = generation;
2885         fs_info->last_trans_committed = generation;
2886
2887         ret = btrfs_recover_balance(fs_info);
2888         if (ret) {
2889                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2890                 goto fail_block_groups;
2891         }
2892
2893         ret = btrfs_init_dev_stats(fs_info);
2894         if (ret) {
2895                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2896                        ret);
2897                 goto fail_block_groups;
2898         }
2899
2900         ret = btrfs_init_dev_replace(fs_info);
2901         if (ret) {
2902                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2903                 goto fail_block_groups;
2904         }
2905
2906         btrfs_close_extra_devices(fs_devices, 1);
2907
2908         ret = btrfs_sysfs_add_one(fs_info);
2909         if (ret) {
2910                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2911                 goto fail_block_groups;
2912         }
2913
2914         ret = btrfs_init_space_info(fs_info);
2915         if (ret) {
2916                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2917                 goto fail_sysfs;
2918         }
2919
2920         ret = btrfs_read_block_groups(fs_info->extent_root);
2921         if (ret) {
2922                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2923                 goto fail_sysfs;
2924         }
2925         fs_info->num_tolerated_disk_barrier_failures =
2926                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2927         if (fs_info->fs_devices->missing_devices >
2928              fs_info->num_tolerated_disk_barrier_failures &&
2929             !(sb->s_flags & MS_RDONLY)) {
2930                 printk(KERN_WARNING "BTRFS: "
2931                         "too many missing devices, writeable mount is not allowed\n");
2932                 goto fail_sysfs;
2933         }
2934
2935         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2936                                                "btrfs-cleaner");
2937         if (IS_ERR(fs_info->cleaner_kthread))
2938                 goto fail_sysfs;
2939
2940         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2941                                                    tree_root,
2942                                                    "btrfs-transaction");
2943         if (IS_ERR(fs_info->transaction_kthread))
2944                 goto fail_cleaner;
2945
2946         if (!btrfs_test_opt(tree_root, SSD) &&
2947             !btrfs_test_opt(tree_root, NOSSD) &&
2948             !fs_info->fs_devices->rotating) {
2949                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2950                        "mode\n");
2951                 btrfs_set_opt(fs_info->mount_opt, SSD);
2952         }
2953
2954         /*
2955          * Mount does not set all options immediatelly, we can do it now and do
2956          * not have to wait for transaction commit
2957          */
2958         btrfs_apply_pending_changes(fs_info);
2959
2960 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2961         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2962                 ret = btrfsic_mount(tree_root, fs_devices,
2963                                     btrfs_test_opt(tree_root,
2964                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2965                                     1 : 0,
2966                                     fs_info->check_integrity_print_mask);
2967                 if (ret)
2968                         printk(KERN_WARNING "BTRFS: failed to initialize"
2969                                " integrity check module %s\n", sb->s_id);
2970         }
2971 #endif
2972         ret = btrfs_read_qgroup_config(fs_info);
2973         if (ret)
2974                 goto fail_trans_kthread;
2975
2976         /* do not make disk changes in broken FS */
2977         if (btrfs_super_log_root(disk_super) != 0) {
2978                 ret = btrfs_replay_log(fs_info, fs_devices);
2979                 if (ret) {
2980                         err = ret;
2981                         goto fail_qgroup;
2982                 }
2983         }
2984
2985         ret = btrfs_find_orphan_roots(tree_root);
2986         if (ret)
2987                 goto fail_qgroup;
2988
2989         if (!(sb->s_flags & MS_RDONLY)) {
2990                 ret = btrfs_cleanup_fs_roots(fs_info);
2991                 if (ret)
2992                         goto fail_qgroup;
2993
2994                 mutex_lock(&fs_info->cleaner_mutex);
2995                 ret = btrfs_recover_relocation(tree_root);
2996                 mutex_unlock(&fs_info->cleaner_mutex);
2997                 if (ret < 0) {
2998                         printk(KERN_WARNING
2999                                "BTRFS: failed to recover relocation\n");
3000                         err = -EINVAL;
3001                         goto fail_qgroup;
3002                 }
3003         }
3004
3005         location.objectid = BTRFS_FS_TREE_OBJECTID;
3006         location.type = BTRFS_ROOT_ITEM_KEY;
3007         location.offset = 0;
3008
3009         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3010         if (IS_ERR(fs_info->fs_root)) {
3011                 err = PTR_ERR(fs_info->fs_root);
3012                 goto fail_qgroup;
3013         }
3014
3015         if (sb->s_flags & MS_RDONLY)
3016                 return 0;
3017
3018         down_read(&fs_info->cleanup_work_sem);
3019         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3020             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3021                 up_read(&fs_info->cleanup_work_sem);
3022                 close_ctree(tree_root);
3023                 return ret;
3024         }
3025         up_read(&fs_info->cleanup_work_sem);
3026
3027         ret = btrfs_resume_balance_async(fs_info);
3028         if (ret) {
3029                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3030                 close_ctree(tree_root);
3031                 return ret;
3032         }
3033
3034         ret = btrfs_resume_dev_replace_async(fs_info);
3035         if (ret) {
3036                 pr_warn("BTRFS: failed to resume dev_replace\n");
3037                 close_ctree(tree_root);
3038                 return ret;
3039         }
3040
3041         btrfs_qgroup_rescan_resume(fs_info);
3042
3043         if (!fs_info->uuid_root) {
3044                 pr_info("BTRFS: creating UUID tree\n");
3045                 ret = btrfs_create_uuid_tree(fs_info);
3046                 if (ret) {
3047                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3048                                 ret);
3049                         close_ctree(tree_root);
3050                         return ret;
3051                 }
3052         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3053                    fs_info->generation !=
3054                                 btrfs_super_uuid_tree_generation(disk_super)) {
3055                 pr_info("BTRFS: checking UUID tree\n");
3056                 ret = btrfs_check_uuid_tree(fs_info);
3057                 if (ret) {
3058                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3059                                 ret);
3060                         close_ctree(tree_root);
3061                         return ret;
3062                 }
3063         } else {
3064                 fs_info->update_uuid_tree_gen = 1;
3065         }
3066
3067         fs_info->open = 1;
3068
3069         return 0;
3070
3071 fail_qgroup:
3072         btrfs_free_qgroup_config(fs_info);
3073 fail_trans_kthread:
3074         kthread_stop(fs_info->transaction_kthread);
3075         btrfs_cleanup_transaction(fs_info->tree_root);
3076         btrfs_free_fs_roots(fs_info);
3077 fail_cleaner:
3078         kthread_stop(fs_info->cleaner_kthread);
3079
3080         /*
3081          * make sure we're done with the btree inode before we stop our
3082          * kthreads
3083          */
3084         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3085
3086 fail_sysfs:
3087         btrfs_sysfs_remove_one(fs_info);
3088
3089 fail_block_groups:
3090         btrfs_put_block_group_cache(fs_info);
3091         btrfs_free_block_groups(fs_info);
3092
3093 fail_tree_roots:
3094         free_root_pointers(fs_info, 1);
3095         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3096
3097 fail_sb_buffer:
3098         btrfs_stop_all_workers(fs_info);
3099 fail_alloc:
3100 fail_iput:
3101         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3102
3103         iput(fs_info->btree_inode);
3104 fail_bio_counter:
3105         percpu_counter_destroy(&fs_info->bio_counter);
3106 fail_delalloc_bytes:
3107         percpu_counter_destroy(&fs_info->delalloc_bytes);
3108 fail_dirty_metadata_bytes:
3109         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3110 fail_bdi:
3111         bdi_destroy(&fs_info->bdi);
3112 fail_srcu:
3113         cleanup_srcu_struct(&fs_info->subvol_srcu);
3114 fail:
3115         btrfs_free_stripe_hash_table(fs_info);
3116         btrfs_close_devices(fs_info->fs_devices);
3117         return err;
3118
3119 recovery_tree_root:
3120         if (!btrfs_test_opt(tree_root, RECOVERY))
3121                 goto fail_tree_roots;
3122
3123         free_root_pointers(fs_info, 0);
3124
3125         /* don't use the log in recovery mode, it won't be valid */
3126         btrfs_set_super_log_root(disk_super, 0);
3127
3128         /* we can't trust the free space cache either */
3129         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3130
3131         ret = next_root_backup(fs_info, fs_info->super_copy,
3132                                &num_backups_tried, &backup_index);
3133         if (ret == -1)
3134                 goto fail_block_groups;
3135         goto retry_root_backup;
3136 }
3137
3138 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3139 {
3140         if (uptodate) {
3141                 set_buffer_uptodate(bh);
3142         } else {
3143                 struct btrfs_device *device = (struct btrfs_device *)
3144                         bh->b_private;
3145
3146                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3147                                           "I/O error on %s\n",
3148                                           rcu_str_deref(device->name));
3149                 /* note, we dont' set_buffer_write_io_error because we have
3150                  * our own ways of dealing with the IO errors
3151                  */
3152                 clear_buffer_uptodate(bh);
3153                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3154         }
3155         unlock_buffer(bh);
3156         put_bh(bh);
3157 }
3158
3159 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3160 {
3161         struct buffer_head *bh;
3162         struct buffer_head *latest = NULL;
3163         struct btrfs_super_block *super;
3164         int i;
3165         u64 transid = 0;
3166         u64 bytenr;
3167
3168         /* we would like to check all the supers, but that would make
3169          * a btrfs mount succeed after a mkfs from a different FS.
3170          * So, we need to add a special mount option to scan for
3171          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3172          */
3173         for (i = 0; i < 1; i++) {
3174                 bytenr = btrfs_sb_offset(i);
3175                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3176                                         i_size_read(bdev->bd_inode))
3177                         break;
3178                 bh = __bread(bdev, bytenr / 4096,
3179                                         BTRFS_SUPER_INFO_SIZE);
3180                 if (!bh)
3181                         continue;
3182
3183                 super = (struct btrfs_super_block *)bh->b_data;
3184                 if (btrfs_super_bytenr(super) != bytenr ||
3185                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3186                         brelse(bh);
3187                         continue;
3188                 }
3189
3190                 if (!latest || btrfs_super_generation(super) > transid) {
3191                         brelse(latest);
3192                         latest = bh;
3193                         transid = btrfs_super_generation(super);
3194                 } else {
3195                         brelse(bh);
3196                 }
3197         }
3198         return latest;
3199 }
3200
3201 /*
3202  * this should be called twice, once with wait == 0 and
3203  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3204  * we write are pinned.
3205  *
3206  * They are released when wait == 1 is done.
3207  * max_mirrors must be the same for both runs, and it indicates how
3208  * many supers on this one device should be written.
3209  *
3210  * max_mirrors == 0 means to write them all.
3211  */
3212 static int write_dev_supers(struct btrfs_device *device,
3213                             struct btrfs_super_block *sb,
3214                             int do_barriers, int wait, int max_mirrors)
3215 {
3216         struct buffer_head *bh;
3217         int i;
3218         int ret;
3219         int errors = 0;
3220         u32 crc;
3221         u64 bytenr;
3222
3223         if (max_mirrors == 0)
3224                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3225
3226         for (i = 0; i < max_mirrors; i++) {
3227                 bytenr = btrfs_sb_offset(i);
3228                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3229                     device->commit_total_bytes)
3230                         break;
3231
3232                 if (wait) {
3233                         bh = __find_get_block(device->bdev, bytenr / 4096,
3234                                               BTRFS_SUPER_INFO_SIZE);
3235                         if (!bh) {
3236                                 errors++;
3237                                 continue;
3238                         }
3239                         wait_on_buffer(bh);
3240                         if (!buffer_uptodate(bh))
3241                                 errors++;
3242
3243                         /* drop our reference */
3244                         brelse(bh);
3245
3246                         /* drop the reference from the wait == 0 run */
3247                         brelse(bh);
3248                         continue;
3249                 } else {
3250                         btrfs_set_super_bytenr(sb, bytenr);
3251
3252                         crc = ~(u32)0;
3253                         crc = btrfs_csum_data((char *)sb +
3254                                               BTRFS_CSUM_SIZE, crc,
3255                                               BTRFS_SUPER_INFO_SIZE -
3256                                               BTRFS_CSUM_SIZE);
3257                         btrfs_csum_final(crc, sb->csum);
3258
3259                         /*
3260                          * one reference for us, and we leave it for the
3261                          * caller
3262                          */
3263                         bh = __getblk(device->bdev, bytenr / 4096,
3264                                       BTRFS_SUPER_INFO_SIZE);
3265                         if (!bh) {
3266                                 printk(KERN_ERR "BTRFS: couldn't get super "
3267                                        "buffer head for bytenr %Lu\n", bytenr);
3268                                 errors++;
3269                                 continue;
3270                         }
3271
3272                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3273
3274                         /* one reference for submit_bh */
3275                         get_bh(bh);
3276
3277                         set_buffer_uptodate(bh);
3278                         lock_buffer(bh);
3279                         bh->b_end_io = btrfs_end_buffer_write_sync;
3280                         bh->b_private = device;
3281                 }
3282
3283                 /*
3284                  * we fua the first super.  The others we allow
3285                  * to go down lazy.
3286                  */
3287                 if (i == 0)
3288                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3289                 else
3290                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3291                 if (ret)
3292                         errors++;
3293         }
3294         return errors < i ? 0 : -1;
3295 }
3296
3297 /*
3298  * endio for the write_dev_flush, this will wake anyone waiting
3299  * for the barrier when it is done
3300  */
3301 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3302 {
3303         if (err) {
3304                 if (err == -EOPNOTSUPP)
3305                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3306                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3307         }
3308         if (bio->bi_private)
3309                 complete(bio->bi_private);
3310         bio_put(bio);
3311 }
3312
3313 /*
3314  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3315  * sent down.  With wait == 1, it waits for the previous flush.
3316  *
3317  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3318  * capable
3319  */
3320 static int write_dev_flush(struct btrfs_device *device, int wait)
3321 {
3322         struct bio *bio;
3323         int ret = 0;
3324
3325         if (device->nobarriers)
3326                 return 0;
3327
3328         if (wait) {
3329                 bio = device->flush_bio;
3330                 if (!bio)
3331                         return 0;
3332
3333                 wait_for_completion(&device->flush_wait);
3334
3335                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3336                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3337                                       rcu_str_deref(device->name));
3338                         device->nobarriers = 1;
3339                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3340                         ret = -EIO;
3341                         btrfs_dev_stat_inc_and_print(device,
3342                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3343                 }
3344
3345                 /* drop the reference from the wait == 0 run */
3346                 bio_put(bio);
3347                 device->flush_bio = NULL;
3348
3349                 return ret;
3350         }
3351
3352         /*
3353          * one reference for us, and we leave it for the
3354          * caller
3355          */
3356         device->flush_bio = NULL;
3357         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3358         if (!bio)
3359                 return -ENOMEM;
3360
3361         bio->bi_end_io = btrfs_end_empty_barrier;
3362         bio->bi_bdev = device->bdev;
3363         init_completion(&device->flush_wait);
3364         bio->bi_private = &device->flush_wait;
3365         device->flush_bio = bio;
3366
3367         bio_get(bio);
3368         btrfsic_submit_bio(WRITE_FLUSH, bio);
3369
3370         return 0;
3371 }
3372
3373 /*
3374  * send an empty flush down to each device in parallel,
3375  * then wait for them
3376  */
3377 static int barrier_all_devices(struct btrfs_fs_info *info)
3378 {
3379         struct list_head *head;
3380         struct btrfs_device *dev;
3381         int errors_send = 0;
3382         int errors_wait = 0;
3383         int ret;
3384
3385         /* send down all the barriers */
3386         head = &info->fs_devices->devices;
3387         list_for_each_entry_rcu(dev, head, dev_list) {
3388                 if (dev->missing)
3389                         continue;
3390                 if (!dev->bdev) {
3391                         errors_send++;
3392                         continue;
3393                 }
3394                 if (!dev->in_fs_metadata || !dev->writeable)
3395                         continue;
3396
3397                 ret = write_dev_flush(dev, 0);
3398                 if (ret)
3399                         errors_send++;
3400         }
3401
3402         /* wait for all the barriers */
3403         list_for_each_entry_rcu(dev, head, dev_list) {
3404                 if (dev->missing)
3405                         continue;
3406                 if (!dev->bdev) {
3407                         errors_wait++;
3408                         continue;
3409                 }
3410                 if (!dev->in_fs_metadata || !dev->writeable)
3411                         continue;
3412
3413                 ret = write_dev_flush(dev, 1);
3414                 if (ret)
3415                         errors_wait++;
3416         }
3417         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3418             errors_wait > info->num_tolerated_disk_barrier_failures)
3419                 return -EIO;
3420         return 0;
3421 }
3422
3423 int btrfs_calc_num_tolerated_disk_barrier_failures(
3424         struct btrfs_fs_info *fs_info)
3425 {
3426         struct btrfs_ioctl_space_info space;
3427         struct btrfs_space_info *sinfo;
3428         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3429                        BTRFS_BLOCK_GROUP_SYSTEM,
3430                        BTRFS_BLOCK_GROUP_METADATA,
3431                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3432         int num_types = 4;
3433         int i;
3434         int c;
3435         int num_tolerated_disk_barrier_failures =
3436                 (int)fs_info->fs_devices->num_devices;
3437
3438         for (i = 0; i < num_types; i++) {
3439                 struct btrfs_space_info *tmp;
3440
3441                 sinfo = NULL;
3442                 rcu_read_lock();
3443                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3444                         if (tmp->flags == types[i]) {
3445                                 sinfo = tmp;
3446                                 break;
3447                         }
3448                 }
3449                 rcu_read_unlock();
3450
3451                 if (!sinfo)
3452                         continue;
3453
3454                 down_read(&sinfo->groups_sem);
3455                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3456                         if (!list_empty(&sinfo->block_groups[c])) {
3457                                 u64 flags;
3458
3459                                 btrfs_get_block_group_info(
3460                                         &sinfo->block_groups[c], &space);
3461                                 if (space.total_bytes == 0 ||
3462                                     space.used_bytes == 0)
3463                                         continue;
3464                                 flags = space.flags;
3465                                 /*
3466                                  * return
3467                                  * 0: if dup, single or RAID0 is configured for
3468                                  *    any of metadata, system or data, else
3469                                  * 1: if RAID5 is configured, or if RAID1 or
3470                                  *    RAID10 is configured and only two mirrors
3471                                  *    are used, else
3472                                  * 2: if RAID6 is configured, else
3473                                  * num_mirrors - 1: if RAID1 or RAID10 is
3474                                  *                  configured and more than
3475                                  *                  2 mirrors are used.
3476                                  */
3477                                 if (num_tolerated_disk_barrier_failures > 0 &&
3478                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3479                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3480                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3481                                       == 0)))
3482                                         num_tolerated_disk_barrier_failures = 0;
3483                                 else if (num_tolerated_disk_barrier_failures > 1) {
3484                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3485                                             BTRFS_BLOCK_GROUP_RAID5 |
3486                                             BTRFS_BLOCK_GROUP_RAID10)) {
3487                                                 num_tolerated_disk_barrier_failures = 1;
3488                                         } else if (flags &
3489                                                    BTRFS_BLOCK_GROUP_RAID6) {
3490                                                 num_tolerated_disk_barrier_failures = 2;
3491                                         }
3492                                 }
3493                         }
3494                 }
3495                 up_read(&sinfo->groups_sem);
3496         }
3497
3498         return num_tolerated_disk_barrier_failures;
3499 }
3500
3501 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3502 {
3503         struct list_head *head;
3504         struct btrfs_device *dev;
3505         struct btrfs_super_block *sb;
3506         struct btrfs_dev_item *dev_item;
3507         int ret;
3508         int do_barriers;
3509         int max_errors;
3510         int total_errors = 0;
3511         u64 flags;
3512
3513         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3514         backup_super_roots(root->fs_info);
3515
3516         sb = root->fs_info->super_for_commit;
3517         dev_item = &sb->dev_item;
3518
3519         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3520         head = &root->fs_info->fs_devices->devices;
3521         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3522
3523         if (do_barriers) {
3524                 ret = barrier_all_devices(root->fs_info);
3525                 if (ret) {
3526                         mutex_unlock(
3527                                 &root->fs_info->fs_devices->device_list_mutex);
3528                         btrfs_error(root->fs_info, ret,
3529                                     "errors while submitting device barriers.");
3530                         return ret;
3531                 }
3532         }
3533
3534         list_for_each_entry_rcu(dev, head, dev_list) {
3535                 if (!dev->bdev) {
3536                         total_errors++;
3537                         continue;
3538                 }
3539                 if (!dev->in_fs_metadata || !dev->writeable)
3540                         continue;
3541
3542                 btrfs_set_stack_device_generation(dev_item, 0);
3543                 btrfs_set_stack_device_type(dev_item, dev->type);
3544                 btrfs_set_stack_device_id(dev_item, dev->devid);
3545                 btrfs_set_stack_device_total_bytes(dev_item,
3546                                                    dev->commit_total_bytes);
3547                 btrfs_set_stack_device_bytes_used(dev_item,
3548                                                   dev->commit_bytes_used);
3549                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3550                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3551                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3552                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3553                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3554
3555                 flags = btrfs_super_flags(sb);
3556                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3557
3558                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3559                 if (ret)
3560                         total_errors++;
3561         }
3562         if (total_errors > max_errors) {
3563                 btrfs_err(root->fs_info, "%d errors while writing supers",
3564                        total_errors);
3565                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3566
3567                 /* FUA is masked off if unsupported and can't be the reason */
3568                 btrfs_error(root->fs_info, -EIO,
3569                             "%d errors while writing supers", total_errors);
3570                 return -EIO;
3571         }
3572
3573         total_errors = 0;
3574         list_for_each_entry_rcu(dev, head, dev_list) {
3575                 if (!dev->bdev)
3576                         continue;
3577                 if (!dev->in_fs_metadata || !dev->writeable)
3578                         continue;
3579
3580                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3581                 if (ret)
3582                         total_errors++;
3583         }
3584         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3585         if (total_errors > max_errors) {
3586                 btrfs_error(root->fs_info, -EIO,
3587                             "%d errors while writing supers", total_errors);
3588                 return -EIO;
3589         }
3590         return 0;
3591 }
3592
3593 int write_ctree_super(struct btrfs_trans_handle *trans,
3594                       struct btrfs_root *root, int max_mirrors)
3595 {
3596         return write_all_supers(root, max_mirrors);
3597 }
3598
3599 /* Drop a fs root from the radix tree and free it. */
3600 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3601                                   struct btrfs_root *root)
3602 {
3603         spin_lock(&fs_info->fs_roots_radix_lock);
3604         radix_tree_delete(&fs_info->fs_roots_radix,
3605                           (unsigned long)root->root_key.objectid);
3606         spin_unlock(&fs_info->fs_roots_radix_lock);
3607
3608         if (btrfs_root_refs(&root->root_item) == 0)
3609                 synchronize_srcu(&fs_info->subvol_srcu);
3610
3611         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3612                 btrfs_free_log(NULL, root);
3613
3614         if (root->free_ino_pinned)
3615                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3616         if (root->free_ino_ctl)
3617                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3618         free_fs_root(root);
3619 }
3620
3621 static void free_fs_root(struct btrfs_root *root)
3622 {
3623         iput(root->ino_cache_inode);
3624         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3625         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3626         root->orphan_block_rsv = NULL;
3627         if (root->anon_dev)
3628                 free_anon_bdev(root->anon_dev);
3629         if (root->subv_writers)
3630                 btrfs_free_subvolume_writers(root->subv_writers);
3631         free_extent_buffer(root->node);
3632         free_extent_buffer(root->commit_root);
3633         kfree(root->free_ino_ctl);
3634         kfree(root->free_ino_pinned);
3635         kfree(root->name);
3636         btrfs_put_fs_root(root);
3637 }
3638
3639 void btrfs_free_fs_root(struct btrfs_root *root)
3640 {
3641         free_fs_root(root);
3642 }
3643
3644 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3645 {
3646         u64 root_objectid = 0;
3647         struct btrfs_root *gang[8];
3648         int i = 0;
3649         int err = 0;
3650         unsigned int ret = 0;
3651         int index;
3652
3653         while (1) {
3654                 index = srcu_read_lock(&fs_info->subvol_srcu);
3655                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3656                                              (void **)gang, root_objectid,
3657                                              ARRAY_SIZE(gang));
3658                 if (!ret) {
3659                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3660                         break;
3661                 }
3662                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3663
3664                 for (i = 0; i < ret; i++) {
3665                         /* Avoid to grab roots in dead_roots */
3666                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3667                                 gang[i] = NULL;
3668                                 continue;
3669                         }
3670                         /* grab all the search result for later use */
3671                         gang[i] = btrfs_grab_fs_root(gang[i]);
3672                 }
3673                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3674
3675                 for (i = 0; i < ret; i++) {
3676                         if (!gang[i])
3677                                 continue;
3678                         root_objectid = gang[i]->root_key.objectid;
3679                         err = btrfs_orphan_cleanup(gang[i]);
3680                         if (err)
3681                                 break;
3682                         btrfs_put_fs_root(gang[i]);
3683                 }
3684                 root_objectid++;
3685         }
3686
3687         /* release the uncleaned roots due to error */
3688         for (; i < ret; i++) {
3689                 if (gang[i])
3690                         btrfs_put_fs_root(gang[i]);
3691         }
3692         return err;
3693 }
3694
3695 int btrfs_commit_super(struct btrfs_root *root)
3696 {
3697         struct btrfs_trans_handle *trans;
3698
3699         mutex_lock(&root->fs_info->cleaner_mutex);
3700         btrfs_run_delayed_iputs(root);
3701         mutex_unlock(&root->fs_info->cleaner_mutex);
3702         wake_up_process(root->fs_info->cleaner_kthread);
3703
3704         /* wait until ongoing cleanup work done */
3705         down_write(&root->fs_info->cleanup_work_sem);
3706         up_write(&root->fs_info->cleanup_work_sem);
3707
3708         trans = btrfs_join_transaction(root);
3709         if (IS_ERR(trans))
3710                 return PTR_ERR(trans);
3711         return btrfs_commit_transaction(trans, root);
3712 }
3713
3714 void close_ctree(struct btrfs_root *root)
3715 {
3716         struct btrfs_fs_info *fs_info = root->fs_info;
3717         int ret;
3718
3719         fs_info->closing = 1;
3720         smp_mb();
3721
3722         /* wait for the uuid_scan task to finish */
3723         down(&fs_info->uuid_tree_rescan_sem);
3724         /* avoid complains from lockdep et al., set sem back to initial state */
3725         up(&fs_info->uuid_tree_rescan_sem);
3726
3727         /* pause restriper - we want to resume on mount */
3728         btrfs_pause_balance(fs_info);
3729
3730         btrfs_dev_replace_suspend_for_unmount(fs_info);
3731
3732         btrfs_scrub_cancel(fs_info);
3733
3734         /* wait for any defraggers to finish */
3735         wait_event(fs_info->transaction_wait,
3736                    (atomic_read(&fs_info->defrag_running) == 0));
3737
3738         /* clear out the rbtree of defraggable inodes */
3739         btrfs_cleanup_defrag_inodes(fs_info);
3740
3741         cancel_work_sync(&fs_info->async_reclaim_work);
3742
3743         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3744                 ret = btrfs_commit_super(root);
3745                 if (ret)
3746                         btrfs_err(fs_info, "commit super ret %d", ret);
3747         }
3748
3749         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3750                 btrfs_error_commit_super(root);
3751
3752         kthread_stop(fs_info->transaction_kthread);
3753         kthread_stop(fs_info->cleaner_kthread);
3754
3755         fs_info->closing = 2;
3756         smp_mb();
3757
3758         btrfs_free_qgroup_config(fs_info);
3759
3760         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3761                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3762                        percpu_counter_sum(&fs_info->delalloc_bytes));
3763         }
3764
3765         btrfs_sysfs_remove_one(fs_info);
3766
3767         btrfs_free_fs_roots(fs_info);
3768
3769         btrfs_put_block_group_cache(fs_info);
3770
3771         btrfs_free_block_groups(fs_info);
3772
3773         /*
3774          * we must make sure there is not any read request to
3775          * submit after we stopping all workers.
3776          */
3777         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3778         btrfs_stop_all_workers(fs_info);
3779
3780         fs_info->open = 0;
3781         free_root_pointers(fs_info, 1);
3782
3783         iput(fs_info->btree_inode);
3784
3785 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3786         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3787                 btrfsic_unmount(root, fs_info->fs_devices);
3788 #endif
3789
3790         btrfs_close_devices(fs_info->fs_devices);
3791         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3792
3793         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3794         percpu_counter_destroy(&fs_info->delalloc_bytes);
3795         percpu_counter_destroy(&fs_info->bio_counter);
3796         bdi_destroy(&fs_info->bdi);
3797         cleanup_srcu_struct(&fs_info->subvol_srcu);
3798
3799         btrfs_free_stripe_hash_table(fs_info);
3800
3801         __btrfs_free_block_rsv(root->orphan_block_rsv);
3802         root->orphan_block_rsv = NULL;
3803
3804         lock_chunks(root);
3805         while (!list_empty(&fs_info->pinned_chunks)) {
3806                 struct extent_map *em;
3807
3808                 em = list_first_entry(&fs_info->pinned_chunks,
3809                                       struct extent_map, list);
3810                 list_del_init(&em->list);
3811                 free_extent_map(em);
3812         }
3813         unlock_chunks(root);
3814 }
3815
3816 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3817                           int atomic)
3818 {
3819         int ret;
3820         struct inode *btree_inode = buf->pages[0]->mapping->host;
3821
3822         ret = extent_buffer_uptodate(buf);
3823         if (!ret)
3824                 return ret;
3825
3826         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3827                                     parent_transid, atomic);
3828         if (ret == -EAGAIN)
3829                 return ret;
3830         return !ret;
3831 }
3832
3833 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3834 {
3835         return set_extent_buffer_uptodate(buf);
3836 }
3837
3838 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3839 {
3840         struct btrfs_root *root;
3841         u64 transid = btrfs_header_generation(buf);
3842         int was_dirty;
3843
3844 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3845         /*
3846          * This is a fast path so only do this check if we have sanity tests
3847          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3848          * outside of the sanity tests.
3849          */
3850         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3851                 return;
3852 #endif
3853         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3854         btrfs_assert_tree_locked(buf);
3855         if (transid != root->fs_info->generation)
3856                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3857                        "found %llu running %llu\n",
3858                         buf->start, transid, root->fs_info->generation);
3859         was_dirty = set_extent_buffer_dirty(buf);
3860         if (!was_dirty)
3861                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3862                                      buf->len,
3863                                      root->fs_info->dirty_metadata_batch);
3864 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3865         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3866                 btrfs_print_leaf(root, buf);
3867                 ASSERT(0);
3868         }
3869 #endif
3870 }
3871
3872 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3873                                         int flush_delayed)
3874 {
3875         /*
3876          * looks as though older kernels can get into trouble with
3877          * this code, they end up stuck in balance_dirty_pages forever
3878          */
3879         int ret;
3880
3881         if (current->flags & PF_MEMALLOC)
3882                 return;
3883
3884         if (flush_delayed)
3885                 btrfs_balance_delayed_items(root);
3886
3887         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3888                                      BTRFS_DIRTY_METADATA_THRESH);
3889         if (ret > 0) {
3890                 balance_dirty_pages_ratelimited(
3891                                    root->fs_info->btree_inode->i_mapping);
3892         }
3893         return;
3894 }
3895
3896 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3897 {
3898         __btrfs_btree_balance_dirty(root, 1);
3899 }
3900
3901 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3902 {
3903         __btrfs_btree_balance_dirty(root, 0);
3904 }
3905
3906 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3907 {
3908         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3909         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3910 }
3911
3912 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3913                               int read_only)
3914 {
3915         struct btrfs_super_block *sb = fs_info->super_copy;
3916         int ret = 0;
3917
3918         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3919                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3920                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3921                 ret = -EINVAL;
3922         }
3923         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3924                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3925                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3926                 ret = -EINVAL;
3927         }
3928         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3929                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3930                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3931                 ret = -EINVAL;
3932         }
3933
3934         /*
3935          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3936          * items yet, they'll be verified later. Issue just a warning.
3937          */
3938         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3939                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3940                                 btrfs_super_root(sb));
3941         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3942                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3943                                 btrfs_super_chunk_root(sb));
3944         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3945                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3946                                 btrfs_super_log_root(sb));
3947
3948         /*
3949          * Check the lower bound, the alignment and other constraints are
3950          * checked later.
3951          */
3952         if (btrfs_super_nodesize(sb) < 4096) {
3953                 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3954                                 btrfs_super_nodesize(sb));
3955                 ret = -EINVAL;
3956         }
3957         if (btrfs_super_sectorsize(sb) < 4096) {
3958                 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3959                                 btrfs_super_sectorsize(sb));
3960                 ret = -EINVAL;
3961         }
3962
3963         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3964                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3965                                 fs_info->fsid, sb->dev_item.fsid);
3966                 ret = -EINVAL;
3967         }
3968
3969         /*
3970          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3971          * done later
3972          */
3973         if (btrfs_super_num_devices(sb) > (1UL << 31))
3974                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3975                                 btrfs_super_num_devices(sb));
3976         if (btrfs_super_num_devices(sb) == 0) {
3977                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3978                 ret = -EINVAL;
3979         }
3980
3981         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3982                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3983                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3984                 ret = -EINVAL;
3985         }
3986
3987         /*
3988          * Obvious sys_chunk_array corruptions, it must hold at least one key
3989          * and one chunk
3990          */
3991         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3992                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3993                                 btrfs_super_sys_array_size(sb),
3994                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3995                 ret = -EINVAL;
3996         }
3997         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3998                         + sizeof(struct btrfs_chunk)) {
3999                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4000                                 btrfs_super_sys_array_size(sb),
4001                                 sizeof(struct btrfs_disk_key)
4002                                 + sizeof(struct btrfs_chunk));
4003                 ret = -EINVAL;
4004         }
4005
4006         /*
4007          * The generation is a global counter, we'll trust it more than the others
4008          * but it's still possible that it's the one that's wrong.
4009          */
4010         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4011                 printk(KERN_WARNING
4012                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4013                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4014         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4015             && btrfs_super_cache_generation(sb) != (u64)-1)
4016                 printk(KERN_WARNING
4017                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4018                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4019
4020         return ret;
4021 }
4022
4023 static void btrfs_error_commit_super(struct btrfs_root *root)
4024 {
4025         mutex_lock(&root->fs_info->cleaner_mutex);
4026         btrfs_run_delayed_iputs(root);
4027         mutex_unlock(&root->fs_info->cleaner_mutex);
4028
4029         down_write(&root->fs_info->cleanup_work_sem);
4030         up_write(&root->fs_info->cleanup_work_sem);
4031
4032         /* cleanup FS via transaction */
4033         btrfs_cleanup_transaction(root);
4034 }
4035
4036 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4037 {
4038         struct btrfs_ordered_extent *ordered;
4039
4040         spin_lock(&root->ordered_extent_lock);
4041         /*
4042          * This will just short circuit the ordered completion stuff which will
4043          * make sure the ordered extent gets properly cleaned up.
4044          */
4045         list_for_each_entry(ordered, &root->ordered_extents,
4046                             root_extent_list)
4047                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4048         spin_unlock(&root->ordered_extent_lock);
4049 }
4050
4051 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4052 {
4053         struct btrfs_root *root;
4054         struct list_head splice;
4055
4056         INIT_LIST_HEAD(&splice);
4057
4058         spin_lock(&fs_info->ordered_root_lock);
4059         list_splice_init(&fs_info->ordered_roots, &splice);
4060         while (!list_empty(&splice)) {
4061                 root = list_first_entry(&splice, struct btrfs_root,
4062                                         ordered_root);
4063                 list_move_tail(&root->ordered_root,
4064                                &fs_info->ordered_roots);
4065
4066                 spin_unlock(&fs_info->ordered_root_lock);
4067                 btrfs_destroy_ordered_extents(root);
4068
4069                 cond_resched();
4070                 spin_lock(&fs_info->ordered_root_lock);
4071         }
4072         spin_unlock(&fs_info->ordered_root_lock);
4073 }
4074
4075 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4076                                       struct btrfs_root *root)
4077 {
4078         struct rb_node *node;
4079         struct btrfs_delayed_ref_root *delayed_refs;
4080         struct btrfs_delayed_ref_node *ref;
4081         int ret = 0;
4082
4083         delayed_refs = &trans->delayed_refs;
4084
4085         spin_lock(&delayed_refs->lock);
4086         if (atomic_read(&delayed_refs->num_entries) == 0) {
4087                 spin_unlock(&delayed_refs->lock);
4088                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4089                 return ret;
4090         }
4091
4092         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4093                 struct btrfs_delayed_ref_head *head;
4094                 bool pin_bytes = false;
4095
4096                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4097                                 href_node);
4098                 if (!mutex_trylock(&head->mutex)) {
4099                         atomic_inc(&head->node.refs);
4100                         spin_unlock(&delayed_refs->lock);
4101
4102                         mutex_lock(&head->mutex);
4103                         mutex_unlock(&head->mutex);
4104                         btrfs_put_delayed_ref(&head->node);
4105                         spin_lock(&delayed_refs->lock);
4106                         continue;
4107                 }
4108                 spin_lock(&head->lock);
4109                 while ((node = rb_first(&head->ref_root)) != NULL) {
4110                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
4111                                        rb_node);
4112                         ref->in_tree = 0;
4113                         rb_erase(&ref->rb_node, &head->ref_root);
4114                         atomic_dec(&delayed_refs->num_entries);
4115                         btrfs_put_delayed_ref(ref);
4116                 }
4117                 if (head->must_insert_reserved)
4118                         pin_bytes = true;
4119                 btrfs_free_delayed_extent_op(head->extent_op);
4120                 delayed_refs->num_heads--;
4121                 if (head->processing == 0)
4122                         delayed_refs->num_heads_ready--;
4123                 atomic_dec(&delayed_refs->num_entries);
4124                 head->node.in_tree = 0;
4125                 rb_erase(&head->href_node, &delayed_refs->href_root);
4126                 spin_unlock(&head->lock);
4127                 spin_unlock(&delayed_refs->lock);
4128                 mutex_unlock(&head->mutex);
4129
4130                 if (pin_bytes)
4131                         btrfs_pin_extent(root, head->node.bytenr,
4132                                          head->node.num_bytes, 1);
4133                 btrfs_put_delayed_ref(&head->node);
4134                 cond_resched();
4135                 spin_lock(&delayed_refs->lock);
4136         }
4137
4138         spin_unlock(&delayed_refs->lock);
4139
4140         return ret;
4141 }
4142
4143 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4144 {
4145         struct btrfs_inode *btrfs_inode;
4146         struct list_head splice;
4147
4148         INIT_LIST_HEAD(&splice);
4149
4150         spin_lock(&root->delalloc_lock);
4151         list_splice_init(&root->delalloc_inodes, &splice);
4152
4153         while (!list_empty(&splice)) {
4154                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4155                                                delalloc_inodes);
4156
4157                 list_del_init(&btrfs_inode->delalloc_inodes);
4158                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4159                           &btrfs_inode->runtime_flags);
4160                 spin_unlock(&root->delalloc_lock);
4161
4162                 btrfs_invalidate_inodes(btrfs_inode->root);
4163
4164                 spin_lock(&root->delalloc_lock);
4165         }
4166
4167         spin_unlock(&root->delalloc_lock);
4168 }
4169
4170 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4171 {
4172         struct btrfs_root *root;
4173         struct list_head splice;
4174
4175         INIT_LIST_HEAD(&splice);
4176
4177         spin_lock(&fs_info->delalloc_root_lock);
4178         list_splice_init(&fs_info->delalloc_roots, &splice);
4179         while (!list_empty(&splice)) {
4180                 root = list_first_entry(&splice, struct btrfs_root,
4181                                          delalloc_root);
4182                 list_del_init(&root->delalloc_root);
4183                 root = btrfs_grab_fs_root(root);
4184                 BUG_ON(!root);
4185                 spin_unlock(&fs_info->delalloc_root_lock);
4186
4187                 btrfs_destroy_delalloc_inodes(root);
4188                 btrfs_put_fs_root(root);
4189
4190                 spin_lock(&fs_info->delalloc_root_lock);
4191         }
4192         spin_unlock(&fs_info->delalloc_root_lock);
4193 }
4194
4195 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4196                                         struct extent_io_tree *dirty_pages,
4197                                         int mark)
4198 {
4199         int ret;
4200         struct extent_buffer *eb;
4201         u64 start = 0;
4202         u64 end;
4203
4204         while (1) {
4205                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4206                                             mark, NULL);
4207                 if (ret)
4208                         break;
4209
4210                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4211                 while (start <= end) {
4212                         eb = btrfs_find_tree_block(root->fs_info, start);
4213                         start += root->nodesize;
4214                         if (!eb)
4215                                 continue;
4216                         wait_on_extent_buffer_writeback(eb);
4217
4218                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4219                                                &eb->bflags))
4220                                 clear_extent_buffer_dirty(eb);
4221                         free_extent_buffer_stale(eb);
4222                 }
4223         }
4224
4225         return ret;
4226 }
4227
4228 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4229                                        struct extent_io_tree *pinned_extents)
4230 {
4231         struct extent_io_tree *unpin;
4232         u64 start;
4233         u64 end;
4234         int ret;
4235         bool loop = true;
4236
4237         unpin = pinned_extents;
4238 again:
4239         while (1) {
4240                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4241                                             EXTENT_DIRTY, NULL);
4242                 if (ret)
4243                         break;
4244
4245                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4246                 btrfs_error_unpin_extent_range(root, start, end);
4247                 cond_resched();
4248         }
4249
4250         if (loop) {
4251                 if (unpin == &root->fs_info->freed_extents[0])
4252                         unpin = &root->fs_info->freed_extents[1];
4253                 else
4254                         unpin = &root->fs_info->freed_extents[0];
4255                 loop = false;
4256                 goto again;
4257         }
4258
4259         return 0;
4260 }
4261
4262 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4263                                        struct btrfs_fs_info *fs_info)
4264 {
4265         struct btrfs_ordered_extent *ordered;
4266
4267         spin_lock(&fs_info->trans_lock);
4268         while (!list_empty(&cur_trans->pending_ordered)) {
4269                 ordered = list_first_entry(&cur_trans->pending_ordered,
4270                                            struct btrfs_ordered_extent,
4271                                            trans_list);
4272                 list_del_init(&ordered->trans_list);
4273                 spin_unlock(&fs_info->trans_lock);
4274
4275                 btrfs_put_ordered_extent(ordered);
4276                 spin_lock(&fs_info->trans_lock);
4277         }
4278         spin_unlock(&fs_info->trans_lock);
4279 }
4280
4281 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4282                                    struct btrfs_root *root)
4283 {
4284         btrfs_destroy_delayed_refs(cur_trans, root);
4285
4286         cur_trans->state = TRANS_STATE_COMMIT_START;
4287         wake_up(&root->fs_info->transaction_blocked_wait);
4288
4289         cur_trans->state = TRANS_STATE_UNBLOCKED;
4290         wake_up(&root->fs_info->transaction_wait);
4291
4292         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4293         btrfs_destroy_delayed_inodes(root);
4294         btrfs_assert_delayed_root_empty(root);
4295
4296         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4297                                      EXTENT_DIRTY);
4298         btrfs_destroy_pinned_extent(root,
4299                                     root->fs_info->pinned_extents);
4300
4301         cur_trans->state =TRANS_STATE_COMPLETED;
4302         wake_up(&cur_trans->commit_wait);
4303
4304         /*
4305         memset(cur_trans, 0, sizeof(*cur_trans));
4306         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4307         */
4308 }
4309
4310 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4311 {
4312         struct btrfs_transaction *t;
4313
4314         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4315
4316         spin_lock(&root->fs_info->trans_lock);
4317         while (!list_empty(&root->fs_info->trans_list)) {
4318                 t = list_first_entry(&root->fs_info->trans_list,
4319                                      struct btrfs_transaction, list);
4320                 if (t->state >= TRANS_STATE_COMMIT_START) {
4321                         atomic_inc(&t->use_count);
4322                         spin_unlock(&root->fs_info->trans_lock);
4323                         btrfs_wait_for_commit(root, t->transid);
4324                         btrfs_put_transaction(t);
4325                         spin_lock(&root->fs_info->trans_lock);
4326                         continue;
4327                 }
4328                 if (t == root->fs_info->running_transaction) {
4329                         t->state = TRANS_STATE_COMMIT_DOING;
4330                         spin_unlock(&root->fs_info->trans_lock);
4331                         /*
4332                          * We wait for 0 num_writers since we don't hold a trans
4333                          * handle open currently for this transaction.
4334                          */
4335                         wait_event(t->writer_wait,
4336                                    atomic_read(&t->num_writers) == 0);
4337                 } else {
4338                         spin_unlock(&root->fs_info->trans_lock);
4339                 }
4340                 btrfs_cleanup_one_transaction(t, root);
4341
4342                 spin_lock(&root->fs_info->trans_lock);
4343                 if (t == root->fs_info->running_transaction)
4344                         root->fs_info->running_transaction = NULL;
4345                 list_del_init(&t->list);
4346                 spin_unlock(&root->fs_info->trans_lock);
4347
4348                 btrfs_put_transaction(t);
4349                 trace_btrfs_transaction_commit(root);
4350                 spin_lock(&root->fs_info->trans_lock);
4351         }
4352         spin_unlock(&root->fs_info->trans_lock);
4353         btrfs_destroy_all_ordered_extents(root->fs_info);
4354         btrfs_destroy_delayed_inodes(root);
4355         btrfs_assert_delayed_root_empty(root);
4356         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4357         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4358         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4359
4360         return 0;
4361 }
4362
4363 static const struct extent_io_ops btree_extent_io_ops = {
4364         .readpage_end_io_hook = btree_readpage_end_io_hook,
4365         .readpage_io_failed_hook = btree_io_failed_hook,
4366         .submit_bio_hook = btree_submit_bio_hook,
4367         /* note we're sharing with inode.c for the merge bio hook */
4368         .merge_bio_hook = btrfs_merge_bio_hook,
4369 };