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)
 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 = bio->bi_error;
 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                 async->bio->bi_error = async->error;
 812                 bio_endio(async->bio);
 813                 return;
 814         }
 815
 816         async->submit_bio_done(async->inode, async->rw, async->bio,
 817                                async->mirror_num, async->bio_flags,
 818                                async->bio_offset);
 819 }
 820
 821 static void run_one_async_free(struct btrfs_work *work)
 822 {
 823         struct async_submit_bio *async;
 824
 825         async = container_of(work, struct  async_submit_bio, work);
 826         kfree(async);
 827 }
 828
 829 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 830                         int rw, struct bio *bio, int mirror_num,
 831                         unsigned long bio_flags,
 832                         u64 bio_offset,
 833                         extent_submit_bio_hook_t *submit_bio_start,
 834                         extent_submit_bio_hook_t *submit_bio_done)
 835 {
 836         struct async_submit_bio *async;
 837
 838         async = kmalloc(sizeof(*async), GFP_NOFS);
 839         if (!async)
 840                 return -ENOMEM;
 841
 842         async->inode = inode;
 843         async->rw = rw;
 844         async->bio = bio;
 845         async->mirror_num = mirror_num;
 846         async->submit_bio_start = submit_bio_start;
 847         async->submit_bio_done = submit_bio_done;
 848
 849         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 850                         run_one_async_done, run_one_async_free);
 851
 852         async->bio_flags = bio_flags;
 853         async->bio_offset = bio_offset;
 854
 855         async->error = 0;
 856
 857         atomic_inc(&fs_info->nr_async_submits);
 858
 859         if (rw & REQ_SYNC)
 860                 btrfs_set_work_high_priority(&async->work);
 861
 862         btrfs_queue_work(fs_info->workers, &async->work);
 863
 864         while (atomic_read(&fs_info->async_submit_draining) &&
 865               atomic_read(&fs_info->nr_async_submits)) {
 866                 wait_event(fs_info->async_submit_wait,
 867                            (atomic_read(&fs_info->nr_async_submits) == 0));
 868         }
 869
 870         return 0;
 871 }
 872
 873 static int btree_csum_one_bio(struct bio *bio)
 874 {
 875         struct bio_vec *bvec;
 876         struct btrfs_root *root;
 877         int i, ret = 0;
 878
 879         bio_for_each_segment_all(bvec, bio, i) {
 880                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 881                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 882                 if (ret)
 883                         break;
 884         }
 885
 886         return ret;
 887 }
 888
 889 static int __btree_submit_bio_start(struct inode *inode, int rw,
 890                                     struct bio *bio, int mirror_num,
 891                                     unsigned long bio_flags,
 892                                     u64 bio_offset)
 893 {
 894         /*
 895          * when we're called for a write, we're already in the async
 896          * submission context.  Just jump into btrfs_map_bio
 897          */
 898         return btree_csum_one_bio(bio);
 899 }
 900
 901 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 902                                  int mirror_num, unsigned long bio_flags,
 903                                  u64 bio_offset)
 904 {
 905         int ret;
 906
 907         /*
 908          * when we're called for a write, we're already in the async
 909          * submission context.  Just jump into btrfs_map_bio
 910          */
 911         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 912         if (ret) {
 913                 bio->bi_error = ret;
 914                 bio_endio(bio);
 915         }
 916         return ret;
 917 }
 918
 919 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 920 {
 921         if (bio_flags & EXTENT_BIO_TREE_LOG)
 922                 return 0;
 923 #ifdef CONFIG_X86
 924         if (cpu_has_xmm4_2)
 925                 return 0;
 926 #endif
 927         return 1;
 928 }
 929
 930 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 931                                  int mirror_num, unsigned long bio_flags,
 932                                  u64 bio_offset)
 933 {
 934         int async = check_async_write(inode, bio_flags);
 935         int ret;
 936
 937         if (!(rw & REQ_WRITE)) {
 938                 /*
 939                  * called for a read, do the setup so that checksum validation
 940                  * can happen in the async kernel threads
 941                  */
 942                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 943                                           bio, BTRFS_WQ_ENDIO_METADATA);
 944                 if (ret)
 945                         goto out_w_error;
 946                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 947                                     mirror_num, 0);
 948         } else if (!async) {
 949                 ret = btree_csum_one_bio(bio);
 950                 if (ret)
 951                         goto out_w_error;
 952                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 953                                     mirror_num, 0);
 954         } else {
 955                 /*
 956                  * kthread helpers are used to submit writes so that
 957                  * checksumming can happen in parallel across all CPUs
 958                  */
 959                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 960                                           inode, rw, bio, mirror_num, 0,
 961                                           bio_offset,
 962                                           __btree_submit_bio_start,
 963                                           __btree_submit_bio_done);
 964         }
 965
 966         if (ret)
 967                 goto out_w_error;
 968         return 0;
 969
 970 out_w_error:
 971         bio->bi_error = ret;
 972         bio_endio(bio);
 973         return ret;
 974 }
 975
 976 #ifdef CONFIG_MIGRATION
 977 static int btree_migratepage(struct address_space *mapping,
 978                         struct page *newpage, struct page *page,
 979                         enum migrate_mode mode)
 980 {
 981         /*
 982          * we can't safely write a btree page from here,
 983          * we haven't done the locking hook
 984          */
 985         if (PageDirty(page))
 986                 return -EAGAIN;
 987         /*
 988          * Buffers may be managed in a filesystem specific way.
 989          * We must have no buffers or drop them.
 990          */
 991         if (page_has_private(page) &&
 992             !try_to_release_page(page, GFP_KERNEL))
 993                 return -EAGAIN;
 994         return migrate_page(mapping, newpage, page, mode);
 995 }
 996 #endif
 997
 998
 999 static int btree_writepages(struct address_space *mapping,
1000                             struct writeback_control *wbc)
1001 {
1002         struct btrfs_fs_info *fs_info;
1003         int ret;
1004
1005         if (wbc->sync_mode == WB_SYNC_NONE) {
1006
1007                 if (wbc->for_kupdate)
1008                         return 0;
1009
1010                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1011                 /* this is a bit racy, but that's ok */
1012                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1013                                              BTRFS_DIRTY_METADATA_THRESH);
1014                 if (ret < 0)
1015                         return 0;
1016         }
1017         return btree_write_cache_pages(mapping, wbc);
1018 }
1019
1020 static int btree_readpage(struct file *file, struct page *page)
1021 {
1022         struct extent_io_tree *tree;
1023         tree = &BTRFS_I(page->mapping->host)->io_tree;
1024         return extent_read_full_page(tree, page, btree_get_extent, 0);
1025 }
1026
1027 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1028 {
1029         if (PageWriteback(page) || PageDirty(page))
1030                 return 0;
1031
1032         return try_release_extent_buffer(page);
1033 }
1034
1035 static void btree_invalidatepage(struct page *page, unsigned int offset,
1036                                  unsigned int length)
1037 {
1038         struct extent_io_tree *tree;
1039         tree = &BTRFS_I(page->mapping->host)->io_tree;
1040         extent_invalidatepage(tree, page, offset);
1041         btree_releasepage(page, GFP_NOFS);
1042         if (PagePrivate(page)) {
1043                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1044                            "page private not zero on page %llu",
1045                            (unsigned long long)page_offset(page));
1046                 ClearPagePrivate(page);
1047                 set_page_private(page, 0);
1048                 page_cache_release(page);
1049         }
1050 }
1051
1052 static int btree_set_page_dirty(struct page *page)
1053 {
1054 #ifdef DEBUG
1055         struct extent_buffer *eb;
1056
1057         BUG_ON(!PagePrivate(page));
1058         eb = (struct extent_buffer *)page->private;
1059         BUG_ON(!eb);
1060         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1061         BUG_ON(!atomic_read(&eb->refs));
1062         btrfs_assert_tree_locked(eb);
1063 #endif
1064         return __set_page_dirty_nobuffers(page);
1065 }
1066
1067 static const struct address_space_operations btree_aops = {
1068         .readpage       = btree_readpage,
1069         .writepages     = btree_writepages,
1070         .releasepage    = btree_releasepage,
1071         .invalidatepage = btree_invalidatepage,
1072 #ifdef CONFIG_MIGRATION
1073         .migratepage    = btree_migratepage,
1074 #endif
1075         .set_page_dirty = btree_set_page_dirty,
1076 };
1077
1078 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1079 {
1080         struct extent_buffer *buf = NULL;
1081         struct inode *btree_inode = root->fs_info->btree_inode;
1082
1083         buf = btrfs_find_create_tree_block(root, bytenr);
1084         if (!buf)
1085                 return;
1086         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1087                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1088         free_extent_buffer(buf);
1089 }
1090
1091 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1092                          int mirror_num, struct extent_buffer **eb)
1093 {
1094         struct extent_buffer *buf = NULL;
1095         struct inode *btree_inode = root->fs_info->btree_inode;
1096         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1097         int ret;
1098
1099         buf = btrfs_find_create_tree_block(root, bytenr);
1100         if (!buf)
1101                 return 0;
1102
1103         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1104
1105         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1106                                        btree_get_extent, mirror_num);
1107         if (ret) {
1108                 free_extent_buffer(buf);
1109                 return ret;
1110         }
1111
1112         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1113                 free_extent_buffer(buf);
1114                 return -EIO;
1115         } else if (extent_buffer_uptodate(buf)) {
1116                 *eb = buf;
1117         } else {
1118                 free_extent_buffer(buf);
1119         }
1120         return 0;
1121 }
1122
1123 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1124                                             u64 bytenr)
1125 {
1126         return find_extent_buffer(fs_info, bytenr);
1127 }
1128
1129 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1130                                                  u64 bytenr)
1131 {
1132         if (btrfs_test_is_dummy_root(root))
1133                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1134         return alloc_extent_buffer(root->fs_info, bytenr);
1135 }
1136
1137
1138 int btrfs_write_tree_block(struct extent_buffer *buf)
1139 {
1140         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1141                                         buf->start + buf->len - 1);
1142 }
1143
1144 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1145 {
1146         return filemap_fdatawait_range(buf->pages[0]->mapping,
1147                                        buf->start, buf->start + buf->len - 1);
1148 }
1149
1150 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1151                                       u64 parent_transid)
1152 {
1153         struct extent_buffer *buf = NULL;
1154         int ret;
1155
1156         buf = btrfs_find_create_tree_block(root, bytenr);
1157         if (!buf)
1158                 return ERR_PTR(-ENOMEM);
1159
1160         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1161         if (ret) {
1162                 free_extent_buffer(buf);
1163                 return ERR_PTR(ret);
1164         }
1165         return buf;
1166
1167 }
1168
1169 void clean_tree_block(struct btrfs_trans_handle *trans,
1170                       struct btrfs_fs_info *fs_info,
1171                       struct extent_buffer *buf)
1172 {
1173         if (btrfs_header_generation(buf) ==
1174             fs_info->running_transaction->transid) {
1175                 btrfs_assert_tree_locked(buf);
1176
1177                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1178                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1179                                              -buf->len,
1180                                              fs_info->dirty_metadata_batch);
1181                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182                         btrfs_set_lock_blocking(buf);
1183                         clear_extent_buffer_dirty(buf);
1184                 }
1185         }
1186 }
1187
1188 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1189 {
1190         struct btrfs_subvolume_writers *writers;
1191         int ret;
1192
1193         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1194         if (!writers)
1195                 return ERR_PTR(-ENOMEM);
1196
1197         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1198         if (ret < 0) {
1199                 kfree(writers);
1200                 return ERR_PTR(ret);
1201         }
1202
1203         init_waitqueue_head(&writers->wait);
1204         return writers;
1205 }
1206
1207 static void
1208 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1209 {
1210         percpu_counter_destroy(&writers->counter);
1211         kfree(writers);
1212 }
1213
1214 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1215                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1216                          u64 objectid)
1217 {
1218         root->node = NULL;
1219         root->commit_root = NULL;
1220         root->sectorsize = sectorsize;
1221         root->nodesize = nodesize;
1222         root->stripesize = stripesize;
1223         root->state = 0;
1224         root->orphan_cleanup_state = 0;
1225
1226         root->objectid = objectid;
1227         root->last_trans = 0;
1228         root->highest_objectid = 0;
1229         root->nr_delalloc_inodes = 0;
1230         root->nr_ordered_extents = 0;
1231         root->name = NULL;
1232         root->inode_tree = RB_ROOT;
1233         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1234         root->block_rsv = NULL;
1235         root->orphan_block_rsv = NULL;
1236
1237         INIT_LIST_HEAD(&root->dirty_list);
1238         INIT_LIST_HEAD(&root->root_list);
1239         INIT_LIST_HEAD(&root->delalloc_inodes);
1240         INIT_LIST_HEAD(&root->delalloc_root);
1241         INIT_LIST_HEAD(&root->ordered_extents);
1242         INIT_LIST_HEAD(&root->ordered_root);
1243         INIT_LIST_HEAD(&root->logged_list[0]);
1244         INIT_LIST_HEAD(&root->logged_list[1]);
1245         spin_lock_init(&root->orphan_lock);
1246         spin_lock_init(&root->inode_lock);
1247         spin_lock_init(&root->delalloc_lock);
1248         spin_lock_init(&root->ordered_extent_lock);
1249         spin_lock_init(&root->accounting_lock);
1250         spin_lock_init(&root->log_extents_lock[0]);
1251         spin_lock_init(&root->log_extents_lock[1]);
1252         mutex_init(&root->objectid_mutex);
1253         mutex_init(&root->log_mutex);
1254         mutex_init(&root->ordered_extent_mutex);
1255         mutex_init(&root->delalloc_mutex);
1256         init_waitqueue_head(&root->log_writer_wait);
1257         init_waitqueue_head(&root->log_commit_wait[0]);
1258         init_waitqueue_head(&root->log_commit_wait[1]);
1259         INIT_LIST_HEAD(&root->log_ctxs[0]);
1260         INIT_LIST_HEAD(&root->log_ctxs[1]);
1261         atomic_set(&root->log_commit[0], 0);
1262         atomic_set(&root->log_commit[1], 0);
1263         atomic_set(&root->log_writers, 0);
1264         atomic_set(&root->log_batch, 0);
1265         atomic_set(&root->orphan_inodes, 0);
1266         atomic_set(&root->refs, 1);
1267         atomic_set(&root->will_be_snapshoted, 0);
1268         root->log_transid = 0;
1269         root->log_transid_committed = -1;
1270         root->last_log_commit = 0;
1271         if (fs_info)
1272                 extent_io_tree_init(&root->dirty_log_pages,
1273                                      fs_info->btree_inode->i_mapping);
1274
1275         memset(&root->root_key, 0, sizeof(root->root_key));
1276         memset(&root->root_item, 0, sizeof(root->root_item));
1277         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1278         if (fs_info)
1279                 root->defrag_trans_start = fs_info->generation;
1280         else
1281                 root->defrag_trans_start = 0;
1282         root->root_key.objectid = objectid;
1283         root->anon_dev = 0;
1284
1285         spin_lock_init(&root->root_item_lock);
1286 }
1287
1288 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1289 {
1290         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1291         if (root)
1292                 root->fs_info = fs_info;
1293         return root;
1294 }
1295
1296 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1297 /* Should only be used by the testing infrastructure */
1298 struct btrfs_root *btrfs_alloc_dummy_root(void)
1299 {
1300         struct btrfs_root *root;
1301
1302         root = btrfs_alloc_root(NULL);
1303         if (!root)
1304                 return ERR_PTR(-ENOMEM);
1305         __setup_root(4096, 4096, 4096, root, NULL, 1);
1306         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1307         root->alloc_bytenr = 0;
1308
1309         return root;
1310 }
1311 #endif
1312
1313 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1314                                      struct btrfs_fs_info *fs_info,
1315                                      u64 objectid)
1316 {
1317         struct extent_buffer *leaf;
1318         struct btrfs_root *tree_root = fs_info->tree_root;
1319         struct btrfs_root *root;
1320         struct btrfs_key key;
1321         int ret = 0;
1322         uuid_le uuid;
1323
1324         root = btrfs_alloc_root(fs_info);
1325         if (!root)
1326                 return ERR_PTR(-ENOMEM);
1327
1328         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1329                 tree_root->stripesize, root, fs_info, objectid);
1330         root->root_key.objectid = objectid;
1331         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1332         root->root_key.offset = 0;
1333
1334         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1335         if (IS_ERR(leaf)) {
1336                 ret = PTR_ERR(leaf);
1337                 leaf = NULL;
1338                 goto fail;
1339         }
1340
1341         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1342         btrfs_set_header_bytenr(leaf, leaf->start);
1343         btrfs_set_header_generation(leaf, trans->transid);
1344         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1345         btrfs_set_header_owner(leaf, objectid);
1346         root->node = leaf;
1347
1348         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1349                             BTRFS_FSID_SIZE);
1350         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1351                             btrfs_header_chunk_tree_uuid(leaf),
1352                             BTRFS_UUID_SIZE);
1353         btrfs_mark_buffer_dirty(leaf);
1354
1355         root->commit_root = btrfs_root_node(root);
1356         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1357
1358         root->root_item.flags = 0;
1359         root->root_item.byte_limit = 0;
1360         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1361         btrfs_set_root_generation(&root->root_item, trans->transid);
1362         btrfs_set_root_level(&root->root_item, 0);
1363         btrfs_set_root_refs(&root->root_item, 1);
1364         btrfs_set_root_used(&root->root_item, leaf->len);
1365         btrfs_set_root_last_snapshot(&root->root_item, 0);
1366         btrfs_set_root_dirid(&root->root_item, 0);
1367         uuid_le_gen(&uuid);
1368         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1369         root->root_item.drop_level = 0;
1370
1371         key.objectid = objectid;
1372         key.type = BTRFS_ROOT_ITEM_KEY;
1373         key.offset = 0;
1374         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1375         if (ret)
1376                 goto fail;
1377
1378         btrfs_tree_unlock(leaf);
1379
1380         return root;
1381
1382 fail:
1383         if (leaf) {
1384                 btrfs_tree_unlock(leaf);
1385                 free_extent_buffer(root->commit_root);
1386                 free_extent_buffer(leaf);
1387         }
1388         kfree(root);
1389
1390         return ERR_PTR(ret);
1391 }
1392
1393 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1394                                          struct btrfs_fs_info *fs_info)
1395 {
1396         struct btrfs_root *root;
1397         struct btrfs_root *tree_root = fs_info->tree_root;
1398         struct extent_buffer *leaf;
1399
1400         root = btrfs_alloc_root(fs_info);
1401         if (!root)
1402                 return ERR_PTR(-ENOMEM);
1403
1404         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1405                      tree_root->stripesize, root, fs_info,
1406                      BTRFS_TREE_LOG_OBJECTID);
1407
1408         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1409         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1410         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1411
1412         /*
1413          * DON'T set REF_COWS for log trees
1414          *
1415          * log trees do not get reference counted because they go away
1416          * before a real commit is actually done.  They do store pointers
1417          * to file data extents, and those reference counts still get
1418          * updated (along with back refs to the log tree).
1419          */
1420
1421         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1422                         NULL, 0, 0, 0);
1423         if (IS_ERR(leaf)) {
1424                 kfree(root);
1425                 return ERR_CAST(leaf);
1426         }
1427
1428         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1429         btrfs_set_header_bytenr(leaf, leaf->start);
1430         btrfs_set_header_generation(leaf, trans->transid);
1431         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1432         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1433         root->node = leaf;
1434
1435         write_extent_buffer(root->node, root->fs_info->fsid,
1436                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1437         btrfs_mark_buffer_dirty(root->node);
1438         btrfs_tree_unlock(root->node);
1439         return root;
1440 }
1441
1442 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1443                              struct btrfs_fs_info *fs_info)
1444 {
1445         struct btrfs_root *log_root;
1446
1447         log_root = alloc_log_tree(trans, fs_info);
1448         if (IS_ERR(log_root))
1449                 return PTR_ERR(log_root);
1450         WARN_ON(fs_info->log_root_tree);
1451         fs_info->log_root_tree = log_root;
1452         return 0;
1453 }
1454
1455 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1456                        struct btrfs_root *root)
1457 {
1458         struct btrfs_root *log_root;
1459         struct btrfs_inode_item *inode_item;
1460
1461         log_root = alloc_log_tree(trans, root->fs_info);
1462         if (IS_ERR(log_root))
1463                 return PTR_ERR(log_root);
1464
1465         log_root->last_trans = trans->transid;
1466         log_root->root_key.offset = root->root_key.objectid;
1467
1468         inode_item = &log_root->root_item.inode;
1469         btrfs_set_stack_inode_generation(inode_item, 1);
1470         btrfs_set_stack_inode_size(inode_item, 3);
1471         btrfs_set_stack_inode_nlink(inode_item, 1);
1472         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1473         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1474
1475         btrfs_set_root_node(&log_root->root_item, log_root->node);
1476
1477         WARN_ON(root->log_root);
1478         root->log_root = log_root;
1479         root->log_transid = 0;
1480         root->log_transid_committed = -1;
1481         root->last_log_commit = 0;
1482         return 0;
1483 }
1484
1485 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1486                                                struct btrfs_key *key)
1487 {
1488         struct btrfs_root *root;
1489         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1490         struct btrfs_path *path;
1491         u64 generation;
1492         int ret;
1493
1494         path = btrfs_alloc_path();
1495         if (!path)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         root = btrfs_alloc_root(fs_info);
1499         if (!root) {
1500                 ret = -ENOMEM;
1501                 goto alloc_fail;
1502         }
1503
1504         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1505                 tree_root->stripesize, root, fs_info, key->objectid);
1506
1507         ret = btrfs_find_root(tree_root, key, path,
1508                               &root->root_item, &root->root_key);
1509         if (ret) {
1510                 if (ret > 0)
1511                         ret = -ENOENT;
1512                 goto find_fail;
1513         }
1514
1515         generation = btrfs_root_generation(&root->root_item);
1516         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1517                                      generation);
1518         if (IS_ERR(root->node)) {
1519                 ret = PTR_ERR(root->node);
1520                 goto find_fail;
1521         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1522                 ret = -EIO;
1523                 free_extent_buffer(root->node);
1524                 goto find_fail;
1525         }
1526         root->commit_root = btrfs_root_node(root);
1527 out:
1528         btrfs_free_path(path);
1529         return root;
1530
1531 find_fail:
1532         kfree(root);
1533 alloc_fail:
1534         root = ERR_PTR(ret);
1535         goto out;
1536 }
1537
1538 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1539                                       struct btrfs_key *location)
1540 {
1541         struct btrfs_root *root;
1542
1543         root = btrfs_read_tree_root(tree_root, location);
1544         if (IS_ERR(root))
1545                 return root;
1546
1547         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1548                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1549                 btrfs_check_and_init_root_item(&root->root_item);
1550         }
1551
1552         return root;
1553 }
1554
1555 int btrfs_init_fs_root(struct btrfs_root *root)
1556 {
1557         int ret;
1558         struct btrfs_subvolume_writers *writers;
1559
1560         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1561         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1562                                         GFP_NOFS);
1563         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1564                 ret = -ENOMEM;
1565                 goto fail;
1566         }
1567
1568         writers = btrfs_alloc_subvolume_writers();
1569         if (IS_ERR(writers)) {
1570                 ret = PTR_ERR(writers);
1571                 goto fail;
1572         }
1573         root->subv_writers = writers;
1574
1575         btrfs_init_free_ino_ctl(root);
1576         spin_lock_init(&root->ino_cache_lock);
1577         init_waitqueue_head(&root->ino_cache_wait);
1578
1579         ret = get_anon_bdev(&root->anon_dev);
1580         if (ret)
1581                 goto free_writers;
1582         return 0;
1583
1584 free_writers:
1585         btrfs_free_subvolume_writers(root->subv_writers);
1586 fail:
1587         kfree(root->free_ino_ctl);
1588         kfree(root->free_ino_pinned);
1589         return ret;
1590 }
1591
1592 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1593                                                u64 root_id)
1594 {
1595         struct btrfs_root *root;
1596
1597         spin_lock(&fs_info->fs_roots_radix_lock);
1598         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1599                                  (unsigned long)root_id);
1600         spin_unlock(&fs_info->fs_roots_radix_lock);
1601         return root;
1602 }
1603
1604 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1605                          struct btrfs_root *root)
1606 {
1607         int ret;
1608
1609         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1610         if (ret)
1611                 return ret;
1612
1613         spin_lock(&fs_info->fs_roots_radix_lock);
1614         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1615                                 (unsigned long)root->root_key.objectid,
1616                                 root);
1617         if (ret == 0)
1618                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1619         spin_unlock(&fs_info->fs_roots_radix_lock);
1620         radix_tree_preload_end();
1621
1622         return ret;
1623 }
1624
1625 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1626                                      struct btrfs_key *location,
1627                                      bool check_ref)
1628 {
1629         struct btrfs_root *root;
1630         struct btrfs_path *path;
1631         struct btrfs_key key;
1632         int ret;
1633
1634         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1635                 return fs_info->tree_root;
1636         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1637                 return fs_info->extent_root;
1638         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1639                 return fs_info->chunk_root;
1640         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1641                 return fs_info->dev_root;
1642         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1643                 return fs_info->csum_root;
1644         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1645                 return fs_info->quota_root ? fs_info->quota_root :
1646                                              ERR_PTR(-ENOENT);
1647         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1648                 return fs_info->uuid_root ? fs_info->uuid_root :
1649                                             ERR_PTR(-ENOENT);
1650 again:
1651         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1652         if (root) {
1653                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1654                         return ERR_PTR(-ENOENT);
1655                 return root;
1656         }
1657
1658         root = btrfs_read_fs_root(fs_info->tree_root, location);
1659         if (IS_ERR(root))
1660                 return root;
1661
1662         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1663                 ret = -ENOENT;
1664                 goto fail;
1665         }
1666
1667         ret = btrfs_init_fs_root(root);
1668         if (ret)
1669                 goto fail;
1670
1671         path = btrfs_alloc_path();
1672         if (!path) {
1673                 ret = -ENOMEM;
1674                 goto fail;
1675         }
1676         key.objectid = BTRFS_ORPHAN_OBJECTID;
1677         key.type = BTRFS_ORPHAN_ITEM_KEY;
1678         key.offset = location->objectid;
1679
1680         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1681         btrfs_free_path(path);
1682         if (ret < 0)
1683                 goto fail;
1684         if (ret == 0)
1685                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1686
1687         ret = btrfs_insert_fs_root(fs_info, root);
1688         if (ret) {
1689                 if (ret == -EEXIST) {
1690                         free_fs_root(root);
1691                         goto again;
1692                 }
1693                 goto fail;
1694         }
1695         return root;
1696 fail:
1697         free_fs_root(root);
1698         return ERR_PTR(ret);
1699 }
1700
1701 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1702 {
1703         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1704         int ret = 0;
1705         struct btrfs_device *device;
1706         struct backing_dev_info *bdi;
1707
1708         rcu_read_lock();
1709         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1710                 if (!device->bdev)
1711                         continue;
1712                 bdi = blk_get_backing_dev_info(device->bdev);
1713                 if (bdi_congested(bdi, bdi_bits)) {
1714                         ret = 1;
1715                         break;
1716                 }
1717         }
1718         rcu_read_unlock();
1719         return ret;
1720 }
1721
1722 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1723 {
1724         int err;
1725
1726         err = bdi_setup_and_register(bdi, "btrfs");
1727         if (err)
1728                 return err;
1729
1730         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1731         bdi->congested_fn       = btrfs_congested_fn;
1732         bdi->congested_data     = info;
1733         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1734         return 0;
1735 }
1736
1737 /*
1738  * called by the kthread helper functions to finally call the bio end_io
1739  * functions.  This is where read checksum verification actually happens
1740  */
1741 static void end_workqueue_fn(struct btrfs_work *work)
1742 {
1743         struct bio *bio;
1744         struct btrfs_end_io_wq *end_io_wq;
1745
1746         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1747         bio = end_io_wq->bio;
1748
1749         bio->bi_error = end_io_wq->error;
1750         bio->bi_private = end_io_wq->private;
1751         bio->bi_end_io = end_io_wq->end_io;
1752         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1753         bio_endio(bio);
1754 }
1755
1756 static int cleaner_kthread(void *arg)
1757 {
1758         struct btrfs_root *root = arg;
1759         int again;
1760         struct btrfs_trans_handle *trans;
1761
1762         do {
1763                 again = 0;
1764
1765                 /* Make the cleaner go to sleep early. */
1766                 if (btrfs_need_cleaner_sleep(root))
1767                         goto sleep;
1768
1769                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1770                         goto sleep;
1771
1772                 /*
1773                  * Avoid the problem that we change the status of the fs
1774                  * during the above check and trylock.
1775                  */
1776                 if (btrfs_need_cleaner_sleep(root)) {
1777                         mutex_unlock(&root->fs_info->cleaner_mutex);
1778                         goto sleep;
1779                 }
1780
1781                 btrfs_run_delayed_iputs(root);
1782                 again = btrfs_clean_one_deleted_snapshot(root);
1783                 mutex_unlock(&root->fs_info->cleaner_mutex);
1784
1785                 /*
1786                  * The defragger has dealt with the R/O remount and umount,
1787                  * needn't do anything special here.
1788                  */
1789                 btrfs_run_defrag_inodes(root->fs_info);
1790
1791                 /*
1792                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1793                  * with relocation (btrfs_relocate_chunk) and relocation
1794                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1795                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1796                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1797                  * unused block groups.
1798                  */
1799                 btrfs_delete_unused_bgs(root->fs_info);
1800 sleep:
1801                 if (!try_to_freeze() && !again) {
1802                         set_current_state(TASK_INTERRUPTIBLE);
1803                         if (!kthread_should_stop())
1804                                 schedule();
1805                         __set_current_state(TASK_RUNNING);
1806                 }
1807         } while (!kthread_should_stop());
1808
1809         /*
1810          * Transaction kthread is stopped before us and wakes us up.
1811          * However we might have started a new transaction and COWed some
1812          * tree blocks when deleting unused block groups for example. So
1813          * make sure we commit the transaction we started to have a clean
1814          * shutdown when evicting the btree inode - if it has dirty pages
1815          * when we do the final iput() on it, eviction will trigger a
1816          * writeback for it which will fail with null pointer dereferences
1817          * since work queues and other resources were already released and
1818          * destroyed by the time the iput/eviction/writeback is made.
1819          */
1820         trans = btrfs_attach_transaction(root);
1821         if (IS_ERR(trans)) {
1822                 if (PTR_ERR(trans) != -ENOENT)
1823                         btrfs_err(root->fs_info,
1824                                   "cleaner transaction attach returned %ld",
1825                                   PTR_ERR(trans));
1826         } else {
1827                 int ret;
1828
1829                 ret = btrfs_commit_transaction(trans, root);
1830                 if (ret)
1831                         btrfs_err(root->fs_info,
1832                                   "cleaner open transaction commit returned %d",
1833                                   ret);
1834         }
1835
1836         return 0;
1837 }
1838
1839 static int transaction_kthread(void *arg)
1840 {
1841         struct btrfs_root *root = arg;
1842         struct btrfs_trans_handle *trans;
1843         struct btrfs_transaction *cur;
1844         u64 transid;
1845         unsigned long now;
1846         unsigned long delay;
1847         bool cannot_commit;
1848
1849         do {
1850                 cannot_commit = false;
1851                 delay = HZ * root->fs_info->commit_interval;
1852                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1853
1854                 spin_lock(&root->fs_info->trans_lock);
1855                 cur = root->fs_info->running_transaction;
1856                 if (!cur) {
1857                         spin_unlock(&root->fs_info->trans_lock);
1858                         goto sleep;
1859                 }
1860
1861                 now = get_seconds();
1862                 if (cur->state < TRANS_STATE_BLOCKED &&
1863                     (now < cur->start_time ||
1864                      now - cur->start_time < root->fs_info->commit_interval)) {
1865                         spin_unlock(&root->fs_info->trans_lock);
1866                         delay = HZ * 5;
1867                         goto sleep;
1868                 }
1869                 transid = cur->transid;
1870                 spin_unlock(&root->fs_info->trans_lock);
1871
1872                 /* If the file system is aborted, this will always fail. */
1873                 trans = btrfs_attach_transaction(root);
1874                 if (IS_ERR(trans)) {
1875                         if (PTR_ERR(trans) != -ENOENT)
1876                                 cannot_commit = true;
1877                         goto sleep;
1878                 }
1879                 if (transid == trans->transid) {
1880                         btrfs_commit_transaction(trans, root);
1881                 } else {
1882                         btrfs_end_transaction(trans, root);
1883                 }
1884 sleep:
1885                 wake_up_process(root->fs_info->cleaner_kthread);
1886                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1887
1888                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1889                                       &root->fs_info->fs_state)))
1890                         btrfs_cleanup_transaction(root);
1891                 if (!try_to_freeze()) {
1892                         set_current_state(TASK_INTERRUPTIBLE);
1893                         if (!kthread_should_stop() &&
1894                             (!btrfs_transaction_blocked(root->fs_info) ||
1895                              cannot_commit))
1896                                 schedule_timeout(delay);
1897                         __set_current_state(TASK_RUNNING);
1898                 }
1899         } while (!kthread_should_stop());
1900         return 0;
1901 }
1902
1903 /*
1904  * this will find the highest generation in the array of
1905  * root backups.  The index of the highest array is returned,
1906  * or -1 if we can't find anything.
1907  *
1908  * We check to make sure the array is valid by comparing the
1909  * generation of the latest  root in the array with the generation
1910  * in the super block.  If they don't match we pitch it.
1911  */
1912 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1913 {
1914         u64 cur;
1915         int newest_index = -1;
1916         struct btrfs_root_backup *root_backup;
1917         int i;
1918
1919         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1920                 root_backup = info->super_copy->super_roots + i;
1921                 cur = btrfs_backup_tree_root_gen(root_backup);
1922                 if (cur == newest_gen)
1923                         newest_index = i;
1924         }
1925
1926         /* check to see if we actually wrapped around */
1927         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1928                 root_backup = info->super_copy->super_roots;
1929                 cur = btrfs_backup_tree_root_gen(root_backup);
1930                 if (cur == newest_gen)
1931                         newest_index = 0;
1932         }
1933         return newest_index;
1934 }
1935
1936
1937 /*
1938  * find the oldest backup so we know where to store new entries
1939  * in the backup array.  This will set the backup_root_index
1940  * field in the fs_info struct
1941  */
1942 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1943                                      u64 newest_gen)
1944 {
1945         int newest_index = -1;
1946
1947         newest_index = find_newest_super_backup(info, newest_gen);
1948         /* if there was garbage in there, just move along */
1949         if (newest_index == -1) {
1950                 info->backup_root_index = 0;
1951         } else {
1952                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1953         }
1954 }
1955
1956 /*
1957  * copy all the root pointers into the super backup array.
1958  * this will bump the backup pointer by one when it is
1959  * done
1960  */
1961 static void backup_super_roots(struct btrfs_fs_info *info)
1962 {
1963         int next_backup;
1964         struct btrfs_root_backup *root_backup;
1965         int last_backup;
1966
1967         next_backup = info->backup_root_index;
1968         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1969                 BTRFS_NUM_BACKUP_ROOTS;
1970
1971         /*
1972          * just overwrite the last backup if we're at the same generation
1973          * this happens only at umount
1974          */
1975         root_backup = info->super_for_commit->super_roots + last_backup;
1976         if (btrfs_backup_tree_root_gen(root_backup) ==
1977             btrfs_header_generation(info->tree_root->node))
1978                 next_backup = last_backup;
1979
1980         root_backup = info->super_for_commit->super_roots + next_backup;
1981
1982         /*
1983          * make sure all of our padding and empty slots get zero filled
1984          * regardless of which ones we use today
1985          */
1986         memset(root_backup, 0, sizeof(*root_backup));
1987
1988         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1989
1990         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1991         btrfs_set_backup_tree_root_gen(root_backup,
1992                                btrfs_header_generation(info->tree_root->node));
1993
1994         btrfs_set_backup_tree_root_level(root_backup,
1995                                btrfs_header_level(info->tree_root->node));
1996
1997         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1998         btrfs_set_backup_chunk_root_gen(root_backup,
1999                                btrfs_header_generation(info->chunk_root->node));
2000         btrfs_set_backup_chunk_root_level(root_backup,
2001                                btrfs_header_level(info->chunk_root->node));
2002
2003         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2004         btrfs_set_backup_extent_root_gen(root_backup,
2005                                btrfs_header_generation(info->extent_root->node));
2006         btrfs_set_backup_extent_root_level(root_backup,
2007                                btrfs_header_level(info->extent_root->node));
2008
2009         /*
2010          * we might commit during log recovery, which happens before we set
2011          * the fs_root.  Make sure it is valid before we fill it in.
2012          */
2013         if (info->fs_root && info->fs_root->node) {
2014                 btrfs_set_backup_fs_root(root_backup,
2015                                          info->fs_root->node->start);
2016                 btrfs_set_backup_fs_root_gen(root_backup,
2017                                btrfs_header_generation(info->fs_root->node));
2018                 btrfs_set_backup_fs_root_level(root_backup,
2019                                btrfs_header_level(info->fs_root->node));
2020         }
2021
2022         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2023         btrfs_set_backup_dev_root_gen(root_backup,
2024                                btrfs_header_generation(info->dev_root->node));
2025         btrfs_set_backup_dev_root_level(root_backup,
2026                                        btrfs_header_level(info->dev_root->node));
2027
2028         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2029         btrfs_set_backup_csum_root_gen(root_backup,
2030                                btrfs_header_generation(info->csum_root->node));
2031         btrfs_set_backup_csum_root_level(root_backup,
2032                                btrfs_header_level(info->csum_root->node));
2033
2034         btrfs_set_backup_total_bytes(root_backup,
2035                              btrfs_super_total_bytes(info->super_copy));
2036         btrfs_set_backup_bytes_used(root_backup,
2037                              btrfs_super_bytes_used(info->super_copy));
2038         btrfs_set_backup_num_devices(root_backup,
2039                              btrfs_super_num_devices(info->super_copy));
2040
2041         /*
2042          * if we don't copy this out to the super_copy, it won't get remembered
2043          * for the next commit
2044          */
2045         memcpy(&info->super_copy->super_roots,
2046                &info->super_for_commit->super_roots,
2047                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2048 }
2049
2050 /*
2051  * this copies info out of the root backup array and back into
2052  * the in-memory super block.  It is meant to help iterate through
2053  * the array, so you send it the number of backups you've already
2054  * tried and the last backup index you used.
2055  *
2056  * this returns -1 when it has tried all the backups
2057  */
2058 static noinline int next_root_backup(struct btrfs_fs_info *info,
2059                                      struct btrfs_super_block *super,
2060                                      int *num_backups_tried, int *backup_index)
2061 {
2062         struct btrfs_root_backup *root_backup;
2063         int newest = *backup_index;
2064
2065         if (*num_backups_tried == 0) {
2066                 u64 gen = btrfs_super_generation(super);
2067
2068                 newest = find_newest_super_backup(info, gen);
2069                 if (newest == -1)
2070                         return -1;
2071
2072                 *backup_index = newest;
2073                 *num_backups_tried = 1;
2074         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2075                 /* we've tried all the backups, all done */
2076                 return -1;
2077         } else {
2078                 /* jump to the next oldest backup */
2079                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2080                         BTRFS_NUM_BACKUP_ROOTS;
2081                 *backup_index = newest;
2082                 *num_backups_tried += 1;
2083         }
2084         root_backup = super->super_roots + newest;
2085
2086         btrfs_set_super_generation(super,
2087                                    btrfs_backup_tree_root_gen(root_backup));
2088         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2089         btrfs_set_super_root_level(super,
2090                                    btrfs_backup_tree_root_level(root_backup));
2091         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2092
2093         /*
2094          * fixme: the total bytes and num_devices need to match or we should
2095          * need a fsck
2096          */
2097         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2098         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2099         return 0;
2100 }
2101
2102 /* helper to cleanup workers */
2103 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2104 {
2105         btrfs_destroy_workqueue(fs_info->fixup_workers);
2106         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2107         btrfs_destroy_workqueue(fs_info->workers);
2108         btrfs_destroy_workqueue(fs_info->endio_workers);
2109         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2110         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2111         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2112         btrfs_destroy_workqueue(fs_info->rmw_workers);
2113         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2114         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2115         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2116         btrfs_destroy_workqueue(fs_info->submit_workers);
2117         btrfs_destroy_workqueue(fs_info->delayed_workers);
2118         btrfs_destroy_workqueue(fs_info->caching_workers);
2119         btrfs_destroy_workqueue(fs_info->readahead_workers);
2120         btrfs_destroy_workqueue(fs_info->flush_workers);
2121         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2122         btrfs_destroy_workqueue(fs_info->extent_workers);
2123 }
2124
2125 static void free_root_extent_buffers(struct btrfs_root *root)
2126 {
2127         if (root) {
2128                 free_extent_buffer(root->node);
2129                 free_extent_buffer(root->commit_root);
2130                 root->node = NULL;
2131                 root->commit_root = NULL;
2132         }
2133 }
2134
2135 /* helper to cleanup tree roots */
2136 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2137 {
2138         free_root_extent_buffers(info->tree_root);
2139
2140         free_root_extent_buffers(info->dev_root);
2141         free_root_extent_buffers(info->extent_root);
2142         free_root_extent_buffers(info->csum_root);
2143         free_root_extent_buffers(info->quota_root);
2144         free_root_extent_buffers(info->uuid_root);
2145         if (chunk_root)
2146                 free_root_extent_buffers(info->chunk_root);
2147 }
2148
2149 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2150 {
2151         int ret;
2152         struct btrfs_root *gang[8];
2153         int i;
2154
2155         while (!list_empty(&fs_info->dead_roots)) {
2156                 gang[0] = list_entry(fs_info->dead_roots.next,
2157                                      struct btrfs_root, root_list);
2158                 list_del(&gang[0]->root_list);
2159
2160                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2161                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2162                 } else {
2163                         free_extent_buffer(gang[0]->node);
2164                         free_extent_buffer(gang[0]->commit_root);
2165                         btrfs_put_fs_root(gang[0]);
2166                 }
2167         }
2168
2169         while (1) {
2170                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2171                                              (void **)gang, 0,
2172                                              ARRAY_SIZE(gang));
2173                 if (!ret)
2174                         break;
2175                 for (i = 0; i < ret; i++)
2176                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2177         }
2178
2179         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2180                 btrfs_free_log_root_tree(NULL, fs_info);
2181                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2182                                             fs_info->pinned_extents);
2183         }
2184 }
2185
2186 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2187 {
2188         mutex_init(&fs_info->scrub_lock);
2189         atomic_set(&fs_info->scrubs_running, 0);
2190         atomic_set(&fs_info->scrub_pause_req, 0);
2191         atomic_set(&fs_info->scrubs_paused, 0);
2192         atomic_set(&fs_info->scrub_cancel_req, 0);
2193         init_waitqueue_head(&fs_info->scrub_pause_wait);
2194         fs_info->scrub_workers_refcnt = 0;
2195 }
2196
2197 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2198 {
2199         spin_lock_init(&fs_info->balance_lock);
2200         mutex_init(&fs_info->balance_mutex);
2201         atomic_set(&fs_info->balance_running, 0);
2202         atomic_set(&fs_info->balance_pause_req, 0);
2203         atomic_set(&fs_info->balance_cancel_req, 0);
2204         fs_info->balance_ctl = NULL;
2205         init_waitqueue_head(&fs_info->balance_wait_q);
2206 }
2207
2208 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2209                                    struct btrfs_root *tree_root)
2210 {
2211         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2212         set_nlink(fs_info->btree_inode, 1);
2213         /*
2214          * we set the i_size on the btree inode to the max possible int.
2215          * the real end of the address space is determined by all of
2216          * the devices in the system
2217          */
2218         fs_info->btree_inode->i_size = OFFSET_MAX;
2219         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2220
2221         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2222         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2223                              fs_info->btree_inode->i_mapping);
2224         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2225         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2226
2227         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2228
2229         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2230         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2231                sizeof(struct btrfs_key));
2232         set_bit(BTRFS_INODE_DUMMY,
2233                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2234         btrfs_insert_inode_hash(fs_info->btree_inode);
2235 }
2236
2237 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2238 {
2239         fs_info->dev_replace.lock_owner = 0;
2240         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2241         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2242         mutex_init(&fs_info->dev_replace.lock_management_lock);
2243         mutex_init(&fs_info->dev_replace.lock);
2244         init_waitqueue_head(&fs_info->replace_wait);
2245 }
2246
2247 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2248 {
2249         spin_lock_init(&fs_info->qgroup_lock);
2250         mutex_init(&fs_info->qgroup_ioctl_lock);
2251         fs_info->qgroup_tree = RB_ROOT;
2252         fs_info->qgroup_op_tree = RB_ROOT;
2253         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2254         fs_info->qgroup_seq = 1;
2255         fs_info->quota_enabled = 0;
2256         fs_info->pending_quota_state = 0;
2257         fs_info->qgroup_ulist = NULL;
2258         mutex_init(&fs_info->qgroup_rescan_lock);
2259 }
2260
2261 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2262                 struct btrfs_fs_devices *fs_devices)
2263 {
2264         int max_active = fs_info->thread_pool_size;
2265         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2266
2267         fs_info->workers =
2268                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2269                                       max_active, 16);
2270
2271         fs_info->delalloc_workers =
2272                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2273
2274         fs_info->flush_workers =
2275                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2276
2277         fs_info->caching_workers =
2278                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2279
2280         /*
2281          * a higher idle thresh on the submit workers makes it much more
2282          * likely that bios will be send down in a sane order to the
2283          * devices
2284          */
2285         fs_info->submit_workers =
2286                 btrfs_alloc_workqueue("submit", flags,
2287                                       min_t(u64, fs_devices->num_devices,
2288                                             max_active), 64);
2289
2290         fs_info->fixup_workers =
2291                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2292
2293         /*
2294          * endios are largely parallel and should have a very
2295          * low idle thresh
2296          */
2297         fs_info->endio_workers =
2298                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2299         fs_info->endio_meta_workers =
2300                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2301         fs_info->endio_meta_write_workers =
2302                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2303         fs_info->endio_raid56_workers =
2304                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2305         fs_info->endio_repair_workers =
2306                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2307         fs_info->rmw_workers =
2308                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2309         fs_info->endio_write_workers =
2310                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2311         fs_info->endio_freespace_worker =
2312                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2313         fs_info->delayed_workers =
2314                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2315         fs_info->readahead_workers =
2316                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2317         fs_info->qgroup_rescan_workers =
2318                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2319         fs_info->extent_workers =
2320                 btrfs_alloc_workqueue("extent-refs", flags,
2321                                       min_t(u64, fs_devices->num_devices,
2322                                             max_active), 8);
2323
2324         if (!(fs_info->workers && fs_info->delalloc_workers &&
2325               fs_info->submit_workers && fs_info->flush_workers &&
2326               fs_info->endio_workers && fs_info->endio_meta_workers &&
2327               fs_info->endio_meta_write_workers &&
2328               fs_info->endio_repair_workers &&
2329               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2330               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2331               fs_info->caching_workers && fs_info->readahead_workers &&
2332               fs_info->fixup_workers && fs_info->delayed_workers &&
2333               fs_info->extent_workers &&
2334               fs_info->qgroup_rescan_workers)) {
2335                 return -ENOMEM;
2336         }
2337
2338         return 0;
2339 }
2340
2341 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2342                             struct btrfs_fs_devices *fs_devices)
2343 {
2344         int ret;
2345         struct btrfs_root *tree_root = fs_info->tree_root;
2346         struct btrfs_root *log_tree_root;
2347         struct btrfs_super_block *disk_super = fs_info->super_copy;
2348         u64 bytenr = btrfs_super_log_root(disk_super);
2349
2350         if (fs_devices->rw_devices == 0) {
2351                 printk(KERN_WARNING "BTRFS: log replay required "
2352                        "on RO media\n");
2353                 return -EIO;
2354         }
2355
2356         log_tree_root = btrfs_alloc_root(fs_info);
2357         if (!log_tree_root)
2358                 return -ENOMEM;
2359
2360         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2361                         tree_root->stripesize, log_tree_root, fs_info,
2362                         BTRFS_TREE_LOG_OBJECTID);
2363
2364         log_tree_root->node = read_tree_block(tree_root, bytenr,
2365                         fs_info->generation + 1);
2366         if (IS_ERR(log_tree_root->node)) {
2367                 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2368                 ret = PTR_ERR(log_tree_root->node);
2369                 kfree(log_tree_root);
2370                 return ret;
2371         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2372                 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2373                 free_extent_buffer(log_tree_root->node);
2374                 kfree(log_tree_root);
2375                 return -EIO;
2376         }
2377         /* returns with log_tree_root freed on success */
2378         ret = btrfs_recover_log_trees(log_tree_root);
2379         if (ret) {
2380                 btrfs_error(tree_root->fs_info, ret,
2381                             "Failed to recover log tree");
2382                 free_extent_buffer(log_tree_root->node);
2383                 kfree(log_tree_root);
2384                 return ret;
2385         }
2386
2387         if (fs_info->sb->s_flags & MS_RDONLY) {
2388                 ret = btrfs_commit_super(tree_root);
2389                 if (ret)
2390                         return ret;
2391         }
2392
2393         return 0;
2394 }
2395
2396 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2397                             struct btrfs_root *tree_root)
2398 {
2399         struct btrfs_root *root;
2400         struct btrfs_key location;
2401         int ret;
2402
2403         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2404         location.type = BTRFS_ROOT_ITEM_KEY;
2405         location.offset = 0;
2406
2407         root = btrfs_read_tree_root(tree_root, &location);
2408         if (IS_ERR(root))
2409                 return PTR_ERR(root);
2410         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2411         fs_info->extent_root = root;
2412
2413         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2414         root = btrfs_read_tree_root(tree_root, &location);
2415         if (IS_ERR(root))
2416                 return PTR_ERR(root);
2417         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2418         fs_info->dev_root = root;
2419         btrfs_init_devices_late(fs_info);
2420
2421         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2422         root = btrfs_read_tree_root(tree_root, &location);
2423         if (IS_ERR(root))
2424                 return PTR_ERR(root);
2425         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2426         fs_info->csum_root = root;
2427
2428         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2429         root = btrfs_read_tree_root(tree_root, &location);
2430         if (!IS_ERR(root)) {
2431                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2432                 fs_info->quota_enabled = 1;
2433                 fs_info->pending_quota_state = 1;
2434                 fs_info->quota_root = root;
2435         }
2436
2437         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2438         root = btrfs_read_tree_root(tree_root, &location);
2439         if (IS_ERR(root)) {
2440                 ret = PTR_ERR(root);
2441                 if (ret != -ENOENT)
2442                         return ret;
2443         } else {
2444                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2445                 fs_info->uuid_root = root;
2446         }
2447
2448         return 0;
2449 }
2450
2451 int open_ctree(struct super_block *sb,
2452                struct btrfs_fs_devices *fs_devices,
2453                char *options)
2454 {
2455         u32 sectorsize;
2456         u32 nodesize;
2457         u32 stripesize;
2458         u64 generation;
2459         u64 features;
2460         struct btrfs_key location;
2461         struct buffer_head *bh;
2462         struct btrfs_super_block *disk_super;
2463         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2464         struct btrfs_root *tree_root;
2465         struct btrfs_root *chunk_root;
2466         int ret;
2467         int err = -EINVAL;
2468         int num_backups_tried = 0;
2469         int backup_index = 0;
2470         int max_active;
2471
2472         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2473         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2474         if (!tree_root || !chunk_root) {
2475                 err = -ENOMEM;
2476                 goto fail;
2477         }
2478
2479         ret = init_srcu_struct(&fs_info->subvol_srcu);
2480         if (ret) {
2481                 err = ret;
2482                 goto fail;
2483         }
2484
2485         ret = setup_bdi(fs_info, &fs_info->bdi);
2486         if (ret) {
2487                 err = ret;
2488                 goto fail_srcu;
2489         }
2490
2491         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2492         if (ret) {
2493                 err = ret;
2494                 goto fail_bdi;
2495         }
2496         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2497                                         (1 + ilog2(nr_cpu_ids));
2498
2499         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2500         if (ret) {
2501                 err = ret;
2502                 goto fail_dirty_metadata_bytes;
2503         }
2504
2505         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2506         if (ret) {
2507                 err = ret;
2508                 goto fail_delalloc_bytes;
2509         }
2510
2511         fs_info->btree_inode = new_inode(sb);
2512         if (!fs_info->btree_inode) {
2513                 err = -ENOMEM;
2514                 goto fail_bio_counter;
2515         }
2516
2517         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2518
2519         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2520         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2521         INIT_LIST_HEAD(&fs_info->trans_list);
2522         INIT_LIST_HEAD(&fs_info->dead_roots);
2523         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2524         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2525         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2526         spin_lock_init(&fs_info->delalloc_root_lock);
2527         spin_lock_init(&fs_info->trans_lock);
2528         spin_lock_init(&fs_info->fs_roots_radix_lock);
2529         spin_lock_init(&fs_info->delayed_iput_lock);
2530         spin_lock_init(&fs_info->defrag_inodes_lock);
2531         spin_lock_init(&fs_info->free_chunk_lock);
2532         spin_lock_init(&fs_info->tree_mod_seq_lock);
2533         spin_lock_init(&fs_info->super_lock);
2534         spin_lock_init(&fs_info->qgroup_op_lock);
2535         spin_lock_init(&fs_info->buffer_lock);
2536         spin_lock_init(&fs_info->unused_bgs_lock);
2537         rwlock_init(&fs_info->tree_mod_log_lock);
2538         mutex_init(&fs_info->unused_bg_unpin_mutex);
2539         mutex_init(&fs_info->delete_unused_bgs_mutex);
2540         mutex_init(&fs_info->reloc_mutex);
2541         mutex_init(&fs_info->delalloc_root_mutex);
2542         seqlock_init(&fs_info->profiles_lock);
2543         init_rwsem(&fs_info->delayed_iput_sem);
2544
2545         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2546         INIT_LIST_HEAD(&fs_info->space_info);
2547         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2548         INIT_LIST_HEAD(&fs_info->unused_bgs);
2549         btrfs_mapping_init(&fs_info->mapping_tree);
2550         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2551                              BTRFS_BLOCK_RSV_GLOBAL);
2552         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2553                              BTRFS_BLOCK_RSV_DELALLOC);
2554         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2555         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2556         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2557         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2558                              BTRFS_BLOCK_RSV_DELOPS);
2559         atomic_set(&fs_info->nr_async_submits, 0);
2560         atomic_set(&fs_info->async_delalloc_pages, 0);
2561         atomic_set(&fs_info->async_submit_draining, 0);
2562         atomic_set(&fs_info->nr_async_bios, 0);
2563         atomic_set(&fs_info->defrag_running, 0);
2564         atomic_set(&fs_info->qgroup_op_seq, 0);
2565         atomic64_set(&fs_info->tree_mod_seq, 0);
2566         fs_info->sb = sb;
2567         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2568         fs_info->metadata_ratio = 0;
2569         fs_info->defrag_inodes = RB_ROOT;
2570         fs_info->free_chunk_space = 0;
2571         fs_info->tree_mod_log = RB_ROOT;
2572         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2573         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2574         /* readahead state */
2575         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2576         spin_lock_init(&fs_info->reada_lock);
2577
2578         fs_info->thread_pool_size = min_t(unsigned long,
2579                                           num_online_cpus() + 2, 8);
2580
2581         INIT_LIST_HEAD(&fs_info->ordered_roots);
2582         spin_lock_init(&fs_info->ordered_root_lock);
2583         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2584                                         GFP_NOFS);
2585         if (!fs_info->delayed_root) {
2586                 err = -ENOMEM;
2587                 goto fail_iput;
2588         }
2589         btrfs_init_delayed_root(fs_info->delayed_root);
2590
2591         btrfs_init_scrub(fs_info);
2592 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2593         fs_info->check_integrity_print_mask = 0;
2594 #endif
2595         btrfs_init_balance(fs_info);
2596         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2597
2598         sb->s_blocksize = 4096;
2599         sb->s_blocksize_bits = blksize_bits(4096);
2600         sb->s_bdi = &fs_info->bdi;
2601
2602         btrfs_init_btree_inode(fs_info, tree_root);
2603
2604         spin_lock_init(&fs_info->block_group_cache_lock);
2605         fs_info->block_group_cache_tree = RB_ROOT;
2606         fs_info->first_logical_byte = (u64)-1;
2607
2608         extent_io_tree_init(&fs_info->freed_extents[0],
2609                              fs_info->btree_inode->i_mapping);
2610         extent_io_tree_init(&fs_info->freed_extents[1],
2611                              fs_info->btree_inode->i_mapping);
2612         fs_info->pinned_extents = &fs_info->freed_extents[0];
2613         fs_info->do_barriers = 1;
2614
2615
2616         mutex_init(&fs_info->ordered_operations_mutex);
2617         mutex_init(&fs_info->tree_log_mutex);
2618         mutex_init(&fs_info->chunk_mutex);
2619         mutex_init(&fs_info->transaction_kthread_mutex);
2620         mutex_init(&fs_info->cleaner_mutex);
2621         mutex_init(&fs_info->volume_mutex);
2622         mutex_init(&fs_info->ro_block_group_mutex);
2623         init_rwsem(&fs_info->commit_root_sem);
2624         init_rwsem(&fs_info->cleanup_work_sem);
2625         init_rwsem(&fs_info->subvol_sem);
2626         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2627
2628         btrfs_init_dev_replace_locks(fs_info);
2629         btrfs_init_qgroup(fs_info);
2630
2631         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2632         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2633
2634         init_waitqueue_head(&fs_info->transaction_throttle);
2635         init_waitqueue_head(&fs_info->transaction_wait);
2636         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2637         init_waitqueue_head(&fs_info->async_submit_wait);
2638
2639         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2640
2641         ret = btrfs_alloc_stripe_hash_table(fs_info);
2642         if (ret) {
2643                 err = ret;
2644                 goto fail_alloc;
2645         }
2646
2647         __setup_root(4096, 4096, 4096, tree_root,
2648                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2649
2650         invalidate_bdev(fs_devices->latest_bdev);
2651
2652         /*
2653          * Read super block and check the signature bytes only
2654          */
2655         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2656         if (!bh) {
2657                 err = -EINVAL;
2658                 goto fail_alloc;
2659         }
2660
2661         /*
2662          * We want to check superblock checksum, the type is stored inside.
2663          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2664          */
2665         if (btrfs_check_super_csum(bh->b_data)) {
2666                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2667                 err = -EINVAL;
2668                 goto fail_alloc;
2669         }
2670
2671         /*
2672          * super_copy is zeroed at allocation time and we never touch the
2673          * following bytes up to INFO_SIZE, the checksum is calculated from
2674          * the whole block of INFO_SIZE
2675          */
2676         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2677         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2678                sizeof(*fs_info->super_for_commit));
2679         brelse(bh);
2680
2681         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2682
2683         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2684         if (ret) {
2685                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2686                 err = -EINVAL;
2687                 goto fail_alloc;
2688         }
2689
2690         disk_super = fs_info->super_copy;
2691         if (!btrfs_super_root(disk_super))
2692                 goto fail_alloc;
2693
2694         /* check FS state, whether FS is broken. */
2695         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2696                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2697
2698         /*
2699          * run through our array of backup supers and setup
2700          * our ring pointer to the oldest one
2701          */
2702         generation = btrfs_super_generation(disk_super);
2703         find_oldest_super_backup(fs_info, generation);
2704
2705         /*
2706          * In the long term, we'll store the compression type in the super
2707          * block, and it'll be used for per file compression control.
2708          */
2709         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2710
2711         ret = btrfs_parse_options(tree_root, options);
2712         if (ret) {
2713                 err = ret;
2714                 goto fail_alloc;
2715         }
2716
2717         features = btrfs_super_incompat_flags(disk_super) &
2718                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2719         if (features) {
2720                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2721                        "unsupported optional features (%Lx).\n",
2722                        features);
2723                 err = -EINVAL;
2724                 goto fail_alloc;
2725         }
2726
2727         /*
2728          * Leafsize and nodesize were always equal, this is only a sanity check.
2729          */
2730         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2731             btrfs_super_nodesize(disk_super)) {
2732                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2733                        "blocksizes don't match.  node %d leaf %d\n",
2734                        btrfs_super_nodesize(disk_super),
2735                        le32_to_cpu(disk_super->__unused_leafsize));
2736                 err = -EINVAL;
2737                 goto fail_alloc;
2738         }
2739         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2740                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2741                        "blocksize (%d) was too large\n",
2742                        btrfs_super_nodesize(disk_super));
2743                 err = -EINVAL;
2744                 goto fail_alloc;
2745         }
2746
2747         features = btrfs_super_incompat_flags(disk_super);
2748         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2749         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2750                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2751
2752         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2753                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2754
2755         /*
2756          * flag our filesystem as having big metadata blocks if
2757          * they are bigger than the page size
2758          */
2759         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2760                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2761                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2762                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2763         }
2764
2765         nodesize = btrfs_super_nodesize(disk_super);
2766         sectorsize = btrfs_super_sectorsize(disk_super);
2767         stripesize = btrfs_super_stripesize(disk_super);
2768         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2769         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2770
2771         /*
2772          * mixed block groups end up with duplicate but slightly offset
2773          * extent buffers for the same range.  It leads to corruptions
2774          */
2775         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2776             (sectorsize != nodesize)) {
2777                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2778                                 "are not allowed for mixed block groups on %s\n",
2779                                 sb->s_id);
2780                 goto fail_alloc;
2781         }
2782
2783         /*
2784          * Needn't use the lock because there is no other task which will
2785          * update the flag.
2786          */
2787         btrfs_set_super_incompat_flags(disk_super, features);
2788
2789         features = btrfs_super_compat_ro_flags(disk_super) &
2790                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2791         if (!(sb->s_flags & MS_RDONLY) && features) {
2792                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2793                        "unsupported option features (%Lx).\n",
2794                        features);
2795                 err = -EINVAL;
2796                 goto fail_alloc;
2797         }
2798
2799         max_active = fs_info->thread_pool_size;
2800
2801         ret = btrfs_init_workqueues(fs_info, fs_devices);
2802         if (ret) {
2803                 err = ret;
2804                 goto fail_sb_buffer;
2805         }
2806
2807         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2808         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2809                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2810
2811         tree_root->nodesize = nodesize;
2812         tree_root->sectorsize = sectorsize;
2813         tree_root->stripesize = stripesize;
2814
2815         sb->s_blocksize = sectorsize;
2816         sb->s_blocksize_bits = blksize_bits(sectorsize);
2817
2818         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2819                 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2820                 goto fail_sb_buffer;
2821         }
2822
2823         if (sectorsize != PAGE_SIZE) {
2824                 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2825                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2826                 goto fail_sb_buffer;
2827         }
2828
2829         mutex_lock(&fs_info->chunk_mutex);
2830         ret = btrfs_read_sys_array(tree_root);
2831         mutex_unlock(&fs_info->chunk_mutex);
2832         if (ret) {
2833                 printk(KERN_ERR "BTRFS: failed to read the system "
2834                        "array on %s\n", sb->s_id);
2835                 goto fail_sb_buffer;
2836         }
2837
2838         generation = btrfs_super_chunk_root_generation(disk_super);
2839
2840         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2841                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2842
2843         chunk_root->node = read_tree_block(chunk_root,
2844                                            btrfs_super_chunk_root(disk_super),
2845                                            generation);
2846         if (IS_ERR(chunk_root->node) ||
2847             !extent_buffer_uptodate(chunk_root->node)) {
2848                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2849                        sb->s_id);
2850                 chunk_root->node = NULL;
2851                 goto fail_tree_roots;
2852         }
2853         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2854         chunk_root->commit_root = btrfs_root_node(chunk_root);
2855
2856         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2857            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2858
2859         ret = btrfs_read_chunk_tree(chunk_root);
2860         if (ret) {
2861                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2862                        sb->s_id);
2863                 goto fail_tree_roots;
2864         }
2865
2866         /*
2867          * keep the device that is marked to be the target device for the
2868          * dev_replace procedure
2869          */
2870         btrfs_close_extra_devices(fs_devices, 0);
2871
2872         if (!fs_devices->latest_bdev) {
2873                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2874                        sb->s_id);
2875                 goto fail_tree_roots;
2876         }
2877
2878 retry_root_backup:
2879         generation = btrfs_super_generation(disk_super);
2880
2881         tree_root->node = read_tree_block(tree_root,
2882                                           btrfs_super_root(disk_super),
2883                                           generation);
2884         if (IS_ERR(tree_root->node) ||
2885             !extent_buffer_uptodate(tree_root->node)) {
2886                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2887                        sb->s_id);
2888                 tree_root->node = NULL;
2889                 goto recovery_tree_root;
2890         }
2891
2892         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2893         tree_root->commit_root = btrfs_root_node(tree_root);
2894         btrfs_set_root_refs(&tree_root->root_item, 1);
2895
2896         ret = btrfs_read_roots(fs_info, tree_root);
2897         if (ret)
2898                 goto recovery_tree_root;
2899
2900         fs_info->generation = generation;
2901         fs_info->last_trans_committed = generation;
2902
2903         ret = btrfs_recover_balance(fs_info);
2904         if (ret) {
2905                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2906                 goto fail_block_groups;
2907         }
2908
2909         ret = btrfs_init_dev_stats(fs_info);
2910         if (ret) {
2911                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2912                        ret);
2913                 goto fail_block_groups;
2914         }
2915
2916         ret = btrfs_init_dev_replace(fs_info);
2917         if (ret) {
2918                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2919                 goto fail_block_groups;
2920         }
2921
2922         btrfs_close_extra_devices(fs_devices, 1);
2923
2924         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2925         if (ret) {
2926                 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2927                 goto fail_block_groups;
2928         }
2929
2930         ret = btrfs_sysfs_add_device(fs_devices);
2931         if (ret) {
2932                 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2933                 goto fail_fsdev_sysfs;
2934         }
2935
2936         ret = btrfs_sysfs_add_one(fs_info);
2937         if (ret) {
2938                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2939                 goto fail_fsdev_sysfs;
2940         }
2941
2942         ret = btrfs_init_space_info(fs_info);
2943         if (ret) {
2944                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2945                 goto fail_sysfs;
2946         }
2947
2948         ret = btrfs_read_block_groups(fs_info->extent_root);
2949         if (ret) {
2950                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2951                 goto fail_sysfs;
2952         }
2953         fs_info->num_tolerated_disk_barrier_failures =
2954                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2955         if (fs_info->fs_devices->missing_devices >
2956              fs_info->num_tolerated_disk_barrier_failures &&
2957             !(sb->s_flags & MS_RDONLY)) {
2958                 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2959                         fs_info->fs_devices->missing_devices,
2960                         fs_info->num_tolerated_disk_barrier_failures);
2961                 goto fail_sysfs;
2962         }
2963
2964         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2965                                                "btrfs-cleaner");
2966         if (IS_ERR(fs_info->cleaner_kthread))
2967                 goto fail_sysfs;
2968
2969         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2970                                                    tree_root,
2971                                                    "btrfs-transaction");
2972         if (IS_ERR(fs_info->transaction_kthread))
2973                 goto fail_cleaner;
2974
2975         if (!btrfs_test_opt(tree_root, SSD) &&
2976             !btrfs_test_opt(tree_root, NOSSD) &&
2977             !fs_info->fs_devices->rotating) {
2978                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2979                        "mode\n");
2980                 btrfs_set_opt(fs_info->mount_opt, SSD);
2981         }
2982
2983         /*
2984          * Mount does not set all options immediatelly, we can do it now and do
2985          * not have to wait for transaction commit
2986          */
2987         btrfs_apply_pending_changes(fs_info);
2988
2989 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2990         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2991                 ret = btrfsic_mount(tree_root, fs_devices,
2992                                     btrfs_test_opt(tree_root,
2993                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2994                                     1 : 0,
2995                                     fs_info->check_integrity_print_mask);
2996                 if (ret)
2997                         printk(KERN_WARNING "BTRFS: failed to initialize"
2998                                " integrity check module %s\n", sb->s_id);
2999         }
3000 #endif
3001         ret = btrfs_read_qgroup_config(fs_info);
3002         if (ret)
3003                 goto fail_trans_kthread;
3004
3005         /* do not make disk changes in broken FS */
3006         if (btrfs_super_log_root(disk_super) != 0) {
3007                 ret = btrfs_replay_log(fs_info, fs_devices);
3008                 if (ret) {
3009                         err = ret;
3010                         goto fail_qgroup;
3011                 }
3012         }
3013
3014         ret = btrfs_find_orphan_roots(tree_root);
3015         if (ret)
3016                 goto fail_qgroup;
3017
3018         if (!(sb->s_flags & MS_RDONLY)) {
3019                 ret = btrfs_cleanup_fs_roots(fs_info);
3020                 if (ret)
3021                         goto fail_qgroup;
3022
3023                 mutex_lock(&fs_info->cleaner_mutex);
3024                 ret = btrfs_recover_relocation(tree_root);
3025                 mutex_unlock(&fs_info->cleaner_mutex);
3026                 if (ret < 0) {
3027                         printk(KERN_WARNING
3028                                "BTRFS: failed to recover relocation\n");
3029                         err = -EINVAL;
3030                         goto fail_qgroup;
3031                 }
3032         }
3033
3034         location.objectid = BTRFS_FS_TREE_OBJECTID;
3035         location.type = BTRFS_ROOT_ITEM_KEY;
3036         location.offset = 0;
3037
3038         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3039         if (IS_ERR(fs_info->fs_root)) {
3040                 err = PTR_ERR(fs_info->fs_root);
3041                 goto fail_qgroup;
3042         }
3043
3044         if (sb->s_flags & MS_RDONLY)
3045                 return 0;
3046
3047         down_read(&fs_info->cleanup_work_sem);
3048         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3049             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3050                 up_read(&fs_info->cleanup_work_sem);
3051                 close_ctree(tree_root);
3052                 return ret;
3053         }
3054         up_read(&fs_info->cleanup_work_sem);
3055
3056         ret = btrfs_resume_balance_async(fs_info);
3057         if (ret) {
3058                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3059                 close_ctree(tree_root);
3060                 return ret;
3061         }
3062
3063         ret = btrfs_resume_dev_replace_async(fs_info);
3064         if (ret) {
3065                 pr_warn("BTRFS: failed to resume dev_replace\n");
3066                 close_ctree(tree_root);
3067                 return ret;
3068         }
3069
3070         btrfs_qgroup_rescan_resume(fs_info);
3071
3072         if (!fs_info->uuid_root) {
3073                 pr_info("BTRFS: creating UUID tree\n");
3074                 ret = btrfs_create_uuid_tree(fs_info);
3075                 if (ret) {
3076                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3077                                 ret);
3078                         close_ctree(tree_root);
3079                         return ret;
3080                 }
3081         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3082                    fs_info->generation !=
3083                                 btrfs_super_uuid_tree_generation(disk_super)) {
3084                 pr_info("BTRFS: checking UUID tree\n");
3085                 ret = btrfs_check_uuid_tree(fs_info);
3086                 if (ret) {
3087                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3088                                 ret);
3089                         close_ctree(tree_root);
3090                         return ret;
3091                 }
3092         } else {
3093                 fs_info->update_uuid_tree_gen = 1;
3094         }
3095
3096         fs_info->open = 1;
3097
3098         return 0;
3099
3100 fail_qgroup:
3101         btrfs_free_qgroup_config(fs_info);
3102 fail_trans_kthread:
3103         kthread_stop(fs_info->transaction_kthread);
3104         btrfs_cleanup_transaction(fs_info->tree_root);
3105         btrfs_free_fs_roots(fs_info);
3106 fail_cleaner:
3107         kthread_stop(fs_info->cleaner_kthread);
3108
3109         /*
3110          * make sure we're done with the btree inode before we stop our
3111          * kthreads
3112          */
3113         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3114
3115 fail_sysfs:
3116         btrfs_sysfs_remove_one(fs_info);
3117
3118 fail_fsdev_sysfs:
3119         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3120
3121 fail_block_groups:
3122         btrfs_put_block_group_cache(fs_info);
3123         btrfs_free_block_groups(fs_info);
3124
3125 fail_tree_roots:
3126         free_root_pointers(fs_info, 1);
3127         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3128
3129 fail_sb_buffer:
3130         btrfs_stop_all_workers(fs_info);
3131 fail_alloc:
3132 fail_iput:
3133         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3134
3135         iput(fs_info->btree_inode);
3136 fail_bio_counter:
3137         percpu_counter_destroy(&fs_info->bio_counter);
3138 fail_delalloc_bytes:
3139         percpu_counter_destroy(&fs_info->delalloc_bytes);
3140 fail_dirty_metadata_bytes:
3141         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3142 fail_bdi:
3143         bdi_destroy(&fs_info->bdi);
3144 fail_srcu:
3145         cleanup_srcu_struct(&fs_info->subvol_srcu);
3146 fail:
3147         btrfs_free_stripe_hash_table(fs_info);
3148         btrfs_close_devices(fs_info->fs_devices);
3149         return err;
3150
3151 recovery_tree_root:
3152         if (!btrfs_test_opt(tree_root, RECOVERY))
3153                 goto fail_tree_roots;
3154
3155         free_root_pointers(fs_info, 0);
3156
3157         /* don't use the log in recovery mode, it won't be valid */
3158         btrfs_set_super_log_root(disk_super, 0);
3159
3160         /* we can't trust the free space cache either */
3161         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3162
3163         ret = next_root_backup(fs_info, fs_info->super_copy,
3164                                &num_backups_tried, &backup_index);
3165         if (ret == -1)
3166                 goto fail_block_groups;
3167         goto retry_root_backup;
3168 }
3169
3170 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3171 {
3172         if (uptodate) {
3173                 set_buffer_uptodate(bh);
3174         } else {
3175                 struct btrfs_device *device = (struct btrfs_device *)
3176                         bh->b_private;
3177
3178                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3179                                           "I/O error on %s\n",
3180                                           rcu_str_deref(device->name));
3181                 /* note, we dont' set_buffer_write_io_error because we have
3182                  * our own ways of dealing with the IO errors
3183                  */
3184                 clear_buffer_uptodate(bh);
3185                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3186         }
3187         unlock_buffer(bh);
3188         put_bh(bh);
3189 }
3190
3191 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3192 {
3193         struct buffer_head *bh;
3194         struct buffer_head *latest = NULL;
3195         struct btrfs_super_block *super;
3196         int i;
3197         u64 transid = 0;
3198         u64 bytenr;
3199
3200         /* we would like to check all the supers, but that would make
3201          * a btrfs mount succeed after a mkfs from a different FS.
3202          * So, we need to add a special mount option to scan for
3203          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3204          */
3205         for (i = 0; i < 1; i++) {
3206                 bytenr = btrfs_sb_offset(i);
3207                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3208                                         i_size_read(bdev->bd_inode))
3209                         break;
3210                 bh = __bread(bdev, bytenr / 4096,
3211                                         BTRFS_SUPER_INFO_SIZE);
3212                 if (!bh)
3213                         continue;
3214
3215                 super = (struct btrfs_super_block *)bh->b_data;
3216                 if (btrfs_super_bytenr(super) != bytenr ||
3217                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3218                         brelse(bh);
3219                         continue;
3220                 }
3221
3222                 if (!latest || btrfs_super_generation(super) > transid) {
3223                         brelse(latest);
3224                         latest = bh;
3225                         transid = btrfs_super_generation(super);
3226                 } else {
3227                         brelse(bh);
3228                 }
3229         }
3230         return latest;
3231 }
3232
3233 /*
3234  * this should be called twice, once with wait == 0 and
3235  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3236  * we write are pinned.
3237  *
3238  * They are released when wait == 1 is done.
3239  * max_mirrors must be the same for both runs, and it indicates how
3240  * many supers on this one device should be written.
3241  *
3242  * max_mirrors == 0 means to write them all.
3243  */
3244 static int write_dev_supers(struct btrfs_device *device,
3245                             struct btrfs_super_block *sb,
3246                             int do_barriers, int wait, int max_mirrors)
3247 {
3248         struct buffer_head *bh;
3249         int i;
3250         int ret;
3251         int errors = 0;
3252         u32 crc;
3253         u64 bytenr;
3254
3255         if (max_mirrors == 0)
3256                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3257
3258         for (i = 0; i < max_mirrors; i++) {
3259                 bytenr = btrfs_sb_offset(i);
3260                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3261                     device->commit_total_bytes)
3262                         break;
3263
3264                 if (wait) {
3265                         bh = __find_get_block(device->bdev, bytenr / 4096,
3266                                               BTRFS_SUPER_INFO_SIZE);
3267                         if (!bh) {
3268                                 errors++;
3269                                 continue;
3270                         }
3271                         wait_on_buffer(bh);
3272                         if (!buffer_uptodate(bh))
3273                                 errors++;
3274
3275                         /* drop our reference */
3276                         brelse(bh);
3277
3278                         /* drop the reference from the wait == 0 run */
3279                         brelse(bh);
3280                         continue;
3281                 } else {
3282                         btrfs_set_super_bytenr(sb, bytenr);
3283
3284                         crc = ~(u32)0;
3285                         crc = btrfs_csum_data((char *)sb +
3286                                               BTRFS_CSUM_SIZE, crc,
3287                                               BTRFS_SUPER_INFO_SIZE -
3288                                               BTRFS_CSUM_SIZE);
3289                         btrfs_csum_final(crc, sb->csum);
3290
3291                         /*
3292                          * one reference for us, and we leave it for the
3293                          * caller
3294                          */
3295                         bh = __getblk(device->bdev, bytenr / 4096,
3296                                       BTRFS_SUPER_INFO_SIZE);
3297                         if (!bh) {
3298                                 printk(KERN_ERR "BTRFS: couldn't get super "
3299                                        "buffer head for bytenr %Lu\n", bytenr);
3300                                 errors++;
3301                                 continue;
3302                         }
3303
3304                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3305
3306                         /* one reference for submit_bh */
3307                         get_bh(bh);
3308
3309                         set_buffer_uptodate(bh);
3310                         lock_buffer(bh);
3311                         bh->b_end_io = btrfs_end_buffer_write_sync;
3312                         bh->b_private = device;
3313                 }
3314
3315                 /*
3316                  * we fua the first super.  The others we allow
3317                  * to go down lazy.
3318                  */
3319                 if (i == 0)
3320                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3321                 else
3322                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3323                 if (ret)
3324                         errors++;
3325         }
3326         return errors < i ? 0 : -1;
3327 }
3328
3329 /*
3330  * endio for the write_dev_flush, this will wake anyone waiting
3331  * for the barrier when it is done
3332  */
3333 static void btrfs_end_empty_barrier(struct bio *bio)
3334 {
3335         if (bio->bi_private)
3336                 complete(bio->bi_private);
3337         bio_put(bio);
3338 }
3339
3340 /*
3341  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3342  * sent down.  With wait == 1, it waits for the previous flush.
3343  *
3344  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3345  * capable
3346  */
3347 static int write_dev_flush(struct btrfs_device *device, int wait)
3348 {
3349         struct bio *bio;
3350         int ret = 0;
3351
3352         if (device->nobarriers)
3353                 return 0;
3354
3355         if (wait) {
3356                 bio = device->flush_bio;
3357                 if (!bio)
3358                         return 0;
3359
3360                 wait_for_completion(&device->flush_wait);
3361
3362                 if (bio->bi_error) {
3363                         ret = bio->bi_error;
3364                         btrfs_dev_stat_inc_and_print(device,
3365                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3366                 }
3367
3368                 /* drop the reference from the wait == 0 run */
3369                 bio_put(bio);
3370                 device->flush_bio = NULL;
3371
3372                 return ret;
3373         }
3374
3375         /*
3376          * one reference for us, and we leave it for the
3377          * caller
3378          */
3379         device->flush_bio = NULL;
3380         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3381         if (!bio)
3382                 return -ENOMEM;
3383
3384         bio->bi_end_io = btrfs_end_empty_barrier;
3385         bio->bi_bdev = device->bdev;
3386         init_completion(&device->flush_wait);
3387         bio->bi_private = &device->flush_wait;
3388         device->flush_bio = bio;
3389
3390         bio_get(bio);
3391         btrfsic_submit_bio(WRITE_FLUSH, bio);
3392
3393         return 0;
3394 }
3395
3396 /*
3397  * send an empty flush down to each device in parallel,
3398  * then wait for them
3399  */
3400 static int barrier_all_devices(struct btrfs_fs_info *info)
3401 {
3402         struct list_head *head;
3403         struct btrfs_device *dev;
3404         int errors_send = 0;
3405         int errors_wait = 0;
3406         int ret;
3407
3408         /* send down all the barriers */
3409         head = &info->fs_devices->devices;
3410         list_for_each_entry_rcu(dev, head, dev_list) {
3411                 if (dev->missing)
3412                         continue;
3413                 if (!dev->bdev) {
3414                         errors_send++;
3415                         continue;
3416                 }
3417                 if (!dev->in_fs_metadata || !dev->writeable)
3418                         continue;
3419
3420                 ret = write_dev_flush(dev, 0);
3421                 if (ret)
3422                         errors_send++;
3423         }
3424
3425         /* wait for all the barriers */
3426         list_for_each_entry_rcu(dev, head, dev_list) {
3427                 if (dev->missing)
3428                         continue;
3429                 if (!dev->bdev) {
3430                         errors_wait++;
3431                         continue;
3432                 }
3433                 if (!dev->in_fs_metadata || !dev->writeable)
3434                         continue;
3435
3436                 ret = write_dev_flush(dev, 1);
3437                 if (ret)
3438                         errors_wait++;
3439         }
3440         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3441             errors_wait > info->num_tolerated_disk_barrier_failures)
3442                 return -EIO;
3443         return 0;
3444 }
3445
3446 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3447 {
3448         if ((flags & (BTRFS_BLOCK_GROUP_DUP |
3449                       BTRFS_BLOCK_GROUP_RAID0 |
3450                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)) ||
3451             ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0))
3452                 return 0;
3453
3454         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3455                      BTRFS_BLOCK_GROUP_RAID5 |
3456                      BTRFS_BLOCK_GROUP_RAID10))
3457                 return 1;
3458
3459         if (flags & BTRFS_BLOCK_GROUP_RAID6)
3460                 return 2;
3461
3462         pr_warn("BTRFS: unknown raid type: %llu\n", flags);
3463         return 0;
3464 }
3465
3466 int btrfs_calc_num_tolerated_disk_barrier_failures(
3467         struct btrfs_fs_info *fs_info)
3468 {
3469         struct btrfs_ioctl_space_info space;
3470         struct btrfs_space_info *sinfo;
3471         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3472                        BTRFS_BLOCK_GROUP_SYSTEM,
3473                        BTRFS_BLOCK_GROUP_METADATA,
3474                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3475         int i;
3476         int c;
3477         int num_tolerated_disk_barrier_failures =
3478                 (int)fs_info->fs_devices->num_devices;
3479
3480         for (i = 0; i < ARRAY_SIZE(types); i++) {
3481                 struct btrfs_space_info *tmp;
3482
3483                 sinfo = NULL;
3484                 rcu_read_lock();
3485                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3486                         if (tmp->flags == types[i]) {
3487                                 sinfo = tmp;
3488                                 break;
3489                         }
3490                 }
3491                 rcu_read_unlock();
3492
3493                 if (!sinfo)
3494                         continue;
3495
3496                 down_read(&sinfo->groups_sem);
3497                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3498                         u64 flags;
3499
3500                         if (list_empty(&sinfo->block_groups[c]))
3501                                 continue;
3502
3503                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3504                                                    &space);
3505                         if (space.total_bytes == 0 || space.used_bytes == 0)
3506                                 continue;
3507                         flags = space.flags;
3508
3509                         num_tolerated_disk_barrier_failures = min(
3510                                 num_tolerated_disk_barrier_failures,
3511                                 btrfs_get_num_tolerated_disk_barrier_failures(
3512                                         flags));
3513                 }
3514                 up_read(&sinfo->groups_sem);
3515         }
3516
3517         return num_tolerated_disk_barrier_failures;
3518 }
3519
3520 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3521 {
3522         struct list_head *head;
3523         struct btrfs_device *dev;
3524         struct btrfs_super_block *sb;
3525         struct btrfs_dev_item *dev_item;
3526         int ret;
3527         int do_barriers;
3528         int max_errors;
3529         int total_errors = 0;
3530         u64 flags;
3531
3532         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3533         backup_super_roots(root->fs_info);
3534
3535         sb = root->fs_info->super_for_commit;
3536         dev_item = &sb->dev_item;
3537
3538         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3539         head = &root->fs_info->fs_devices->devices;
3540         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3541
3542         if (do_barriers) {
3543                 ret = barrier_all_devices(root->fs_info);
3544                 if (ret) {
3545                         mutex_unlock(
3546                                 &root->fs_info->fs_devices->device_list_mutex);
3547                         btrfs_error(root->fs_info, ret,
3548                                     "errors while submitting device barriers.");
3549                         return ret;
3550                 }
3551         }
3552
3553         list_for_each_entry_rcu(dev, head, dev_list) {
3554                 if (!dev->bdev) {
3555                         total_errors++;
3556                         continue;
3557                 }
3558                 if (!dev->in_fs_metadata || !dev->writeable)
3559                         continue;
3560
3561                 btrfs_set_stack_device_generation(dev_item, 0);
3562                 btrfs_set_stack_device_type(dev_item, dev->type);
3563                 btrfs_set_stack_device_id(dev_item, dev->devid);
3564                 btrfs_set_stack_device_total_bytes(dev_item,
3565                                                    dev->commit_total_bytes);
3566                 btrfs_set_stack_device_bytes_used(dev_item,
3567                                                   dev->commit_bytes_used);
3568                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3569                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3570                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3571                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3572                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3573
3574                 flags = btrfs_super_flags(sb);
3575                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3576
3577                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3578                 if (ret)
3579                         total_errors++;
3580         }
3581         if (total_errors > max_errors) {
3582                 btrfs_err(root->fs_info, "%d errors while writing supers",
3583                        total_errors);
3584                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3585
3586                 /* FUA is masked off if unsupported and can't be the reason */
3587                 btrfs_error(root->fs_info, -EIO,
3588                             "%d errors while writing supers", total_errors);
3589                 return -EIO;
3590         }
3591
3592         total_errors = 0;
3593         list_for_each_entry_rcu(dev, head, dev_list) {
3594                 if (!dev->bdev)
3595                         continue;
3596                 if (!dev->in_fs_metadata || !dev->writeable)
3597                         continue;
3598
3599                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3600                 if (ret)
3601                         total_errors++;
3602         }
3603         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3604         if (total_errors > max_errors) {
3605                 btrfs_error(root->fs_info, -EIO,
3606                             "%d errors while writing supers", total_errors);
3607                 return -EIO;
3608         }
3609         return 0;
3610 }
3611
3612 int write_ctree_super(struct btrfs_trans_handle *trans,
3613                       struct btrfs_root *root, int max_mirrors)
3614 {
3615         return write_all_supers(root, max_mirrors);
3616 }
3617
3618 /* Drop a fs root from the radix tree and free it. */
3619 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3620                                   struct btrfs_root *root)
3621 {
3622         spin_lock(&fs_info->fs_roots_radix_lock);
3623         radix_tree_delete(&fs_info->fs_roots_radix,
3624                           (unsigned long)root->root_key.objectid);
3625         spin_unlock(&fs_info->fs_roots_radix_lock);
3626
3627         if (btrfs_root_refs(&root->root_item) == 0)
3628                 synchronize_srcu(&fs_info->subvol_srcu);
3629
3630         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3631                 btrfs_free_log(NULL, root);
3632
3633         if (root->free_ino_pinned)
3634                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3635         if (root->free_ino_ctl)
3636                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3637         free_fs_root(root);
3638 }
3639
3640 static void free_fs_root(struct btrfs_root *root)
3641 {
3642         iput(root->ino_cache_inode);
3643         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3644         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3645         root->orphan_block_rsv = NULL;
3646         if (root->anon_dev)
3647                 free_anon_bdev(root->anon_dev);
3648         if (root->subv_writers)
3649                 btrfs_free_subvolume_writers(root->subv_writers);
3650         free_extent_buffer(root->node);
3651         free_extent_buffer(root->commit_root);
3652         kfree(root->free_ino_ctl);
3653         kfree(root->free_ino_pinned);
3654         kfree(root->name);
3655         btrfs_put_fs_root(root);
3656 }
3657
3658 void btrfs_free_fs_root(struct btrfs_root *root)
3659 {
3660         free_fs_root(root);
3661 }
3662
3663 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3664 {
3665         u64 root_objectid = 0;
3666         struct btrfs_root *gang[8];
3667         int i = 0;
3668         int err = 0;
3669         unsigned int ret = 0;
3670         int index;
3671
3672         while (1) {
3673                 index = srcu_read_lock(&fs_info->subvol_srcu);
3674                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3675                                              (void **)gang, root_objectid,
3676                                              ARRAY_SIZE(gang));
3677                 if (!ret) {
3678                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3679                         break;
3680                 }
3681                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3682
3683                 for (i = 0; i < ret; i++) {
3684                         /* Avoid to grab roots in dead_roots */
3685                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3686                                 gang[i] = NULL;
3687                                 continue;
3688                         }
3689                         /* grab all the search result for later use */
3690                         gang[i] = btrfs_grab_fs_root(gang[i]);
3691                 }
3692                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3693
3694                 for (i = 0; i < ret; i++) {
3695                         if (!gang[i])
3696                                 continue;
3697                         root_objectid = gang[i]->root_key.objectid;
3698                         err = btrfs_orphan_cleanup(gang[i]);
3699                         if (err)
3700                                 break;
3701                         btrfs_put_fs_root(gang[i]);
3702                 }
3703                 root_objectid++;
3704         }
3705
3706         /* release the uncleaned roots due to error */
3707         for (; i < ret; i++) {
3708                 if (gang[i])
3709                         btrfs_put_fs_root(gang[i]);
3710         }
3711         return err;
3712 }
3713
3714 int btrfs_commit_super(struct btrfs_root *root)
3715 {
3716         struct btrfs_trans_handle *trans;
3717
3718         mutex_lock(&root->fs_info->cleaner_mutex);
3719         btrfs_run_delayed_iputs(root);
3720         mutex_unlock(&root->fs_info->cleaner_mutex);
3721         wake_up_process(root->fs_info->cleaner_kthread);
3722
3723         /* wait until ongoing cleanup work done */
3724         down_write(&root->fs_info->cleanup_work_sem);
3725         up_write(&root->fs_info->cleanup_work_sem);
3726
3727         trans = btrfs_join_transaction(root);
3728         if (IS_ERR(trans))
3729                 return PTR_ERR(trans);
3730         return btrfs_commit_transaction(trans, root);
3731 }
3732
3733 void close_ctree(struct btrfs_root *root)
3734 {
3735         struct btrfs_fs_info *fs_info = root->fs_info;
3736         int ret;
3737
3738         fs_info->closing = 1;
3739         smp_mb();
3740
3741         /* wait for the uuid_scan task to finish */
3742         down(&fs_info->uuid_tree_rescan_sem);
3743         /* avoid complains from lockdep et al., set sem back to initial state */
3744         up(&fs_info->uuid_tree_rescan_sem);
3745
3746         /* pause restriper - we want to resume on mount */
3747         btrfs_pause_balance(fs_info);
3748
3749         btrfs_dev_replace_suspend_for_unmount(fs_info);
3750
3751         btrfs_scrub_cancel(fs_info);
3752
3753         /* wait for any defraggers to finish */
3754         wait_event(fs_info->transaction_wait,
3755                    (atomic_read(&fs_info->defrag_running) == 0));
3756
3757         /* clear out the rbtree of defraggable inodes */
3758         btrfs_cleanup_defrag_inodes(fs_info);
3759
3760         cancel_work_sync(&fs_info->async_reclaim_work);
3761
3762         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3763                 /*
3764                  * If the cleaner thread is stopped and there are
3765                  * block groups queued for removal, the deletion will be
3766                  * skipped when we quit the cleaner thread.
3767                  */
3768                 btrfs_delete_unused_bgs(root->fs_info);
3769
3770                 ret = btrfs_commit_super(root);
3771                 if (ret)
3772                         btrfs_err(fs_info, "commit super ret %d", ret);
3773         }
3774
3775         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3776                 btrfs_error_commit_super(root);
3777
3778         kthread_stop(fs_info->transaction_kthread);
3779         kthread_stop(fs_info->cleaner_kthread);
3780
3781         fs_info->closing = 2;
3782         smp_mb();
3783
3784         btrfs_free_qgroup_config(fs_info);
3785
3786         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3787                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3788                        percpu_counter_sum(&fs_info->delalloc_bytes));
3789         }
3790
3791         btrfs_sysfs_remove_one(fs_info);
3792         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3793
3794         btrfs_free_fs_roots(fs_info);
3795
3796         btrfs_put_block_group_cache(fs_info);
3797
3798         btrfs_free_block_groups(fs_info);
3799
3800         /*
3801          * we must make sure there is not any read request to
3802          * submit after we stopping all workers.
3803          */
3804         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3805         btrfs_stop_all_workers(fs_info);
3806
3807         fs_info->open = 0;
3808         free_root_pointers(fs_info, 1);
3809
3810         iput(fs_info->btree_inode);
3811
3812 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3813         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3814                 btrfsic_unmount(root, fs_info->fs_devices);
3815 #endif
3816
3817         btrfs_close_devices(fs_info->fs_devices);
3818         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3819
3820         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3821         percpu_counter_destroy(&fs_info->delalloc_bytes);
3822         percpu_counter_destroy(&fs_info->bio_counter);
3823         bdi_destroy(&fs_info->bdi);
3824         cleanup_srcu_struct(&fs_info->subvol_srcu);
3825
3826         btrfs_free_stripe_hash_table(fs_info);
3827
3828         __btrfs_free_block_rsv(root->orphan_block_rsv);
3829         root->orphan_block_rsv = NULL;
3830
3831         lock_chunks(root);
3832         while (!list_empty(&fs_info->pinned_chunks)) {
3833                 struct extent_map *em;
3834
3835                 em = list_first_entry(&fs_info->pinned_chunks,
3836                                       struct extent_map, list);
3837                 list_del_init(&em->list);
3838                 free_extent_map(em);
3839         }
3840         unlock_chunks(root);
3841 }
3842
3843 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3844                           int atomic)
3845 {
3846         int ret;
3847         struct inode *btree_inode = buf->pages[0]->mapping->host;
3848
3849         ret = extent_buffer_uptodate(buf);
3850         if (!ret)
3851                 return ret;
3852
3853         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3854                                     parent_transid, atomic);
3855         if (ret == -EAGAIN)
3856                 return ret;
3857         return !ret;
3858 }
3859
3860 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3861 {
3862         return set_extent_buffer_uptodate(buf);
3863 }
3864
3865 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3866 {
3867         struct btrfs_root *root;
3868         u64 transid = btrfs_header_generation(buf);
3869         int was_dirty;
3870
3871 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3872         /*
3873          * This is a fast path so only do this check if we have sanity tests
3874          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3875          * outside of the sanity tests.
3876          */
3877         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3878                 return;
3879 #endif
3880         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3881         btrfs_assert_tree_locked(buf);
3882         if (transid != root->fs_info->generation)
3883                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3884                        "found %llu running %llu\n",
3885                         buf->start, transid, root->fs_info->generation);
3886         was_dirty = set_extent_buffer_dirty(buf);
3887         if (!was_dirty)
3888                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3889                                      buf->len,
3890                                      root->fs_info->dirty_metadata_batch);
3891 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3892         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3893                 btrfs_print_leaf(root, buf);
3894                 ASSERT(0);
3895         }
3896 #endif
3897 }
3898
3899 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3900                                         int flush_delayed)
3901 {
3902         /*
3903          * looks as though older kernels can get into trouble with
3904          * this code, they end up stuck in balance_dirty_pages forever
3905          */
3906         int ret;
3907
3908         if (current->flags & PF_MEMALLOC)
3909                 return;
3910
3911         if (flush_delayed)
3912                 btrfs_balance_delayed_items(root);
3913
3914         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3915                                      BTRFS_DIRTY_METADATA_THRESH);
3916         if (ret > 0) {
3917                 balance_dirty_pages_ratelimited(
3918                                    root->fs_info->btree_inode->i_mapping);
3919         }
3920         return;
3921 }
3922
3923 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3924 {
3925         __btrfs_btree_balance_dirty(root, 1);
3926 }
3927
3928 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3929 {
3930         __btrfs_btree_balance_dirty(root, 0);
3931 }
3932
3933 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3934 {
3935         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3936         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3937 }
3938
3939 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3940                               int read_only)
3941 {
3942         struct btrfs_super_block *sb = fs_info->super_copy;
3943         int ret = 0;
3944
3945         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3946                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3947                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3948                 ret = -EINVAL;
3949         }
3950         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3951                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3952                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3953                 ret = -EINVAL;
3954         }
3955         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3956                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3957                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3958                 ret = -EINVAL;
3959         }
3960
3961         /*
3962          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3963          * items yet, they'll be verified later. Issue just a warning.
3964          */
3965         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3966                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3967                                 btrfs_super_root(sb));
3968         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3969                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3970                                 btrfs_super_chunk_root(sb));
3971         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3972                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3973                                 btrfs_super_log_root(sb));
3974
3975         /*
3976          * Check the lower bound, the alignment and other constraints are
3977          * checked later.
3978          */
3979         if (btrfs_super_nodesize(sb) < 4096) {
3980                 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3981                                 btrfs_super_nodesize(sb));
3982                 ret = -EINVAL;
3983         }
3984         if (btrfs_super_sectorsize(sb) < 4096) {
3985                 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3986                                 btrfs_super_sectorsize(sb));
3987                 ret = -EINVAL;
3988         }
3989
3990         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3991                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3992                                 fs_info->fsid, sb->dev_item.fsid);
3993                 ret = -EINVAL;
3994         }
3995
3996         /*
3997          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3998          * done later
3999          */
4000         if (btrfs_super_num_devices(sb) > (1UL << 31))
4001                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4002                                 btrfs_super_num_devices(sb));
4003         if (btrfs_super_num_devices(sb) == 0) {
4004                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4005                 ret = -EINVAL;
4006         }
4007
4008         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4009                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4010                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4011                 ret = -EINVAL;
4012         }
4013
4014         /*
4015          * Obvious sys_chunk_array corruptions, it must hold at least one key
4016          * and one chunk
4017          */
4018         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4019                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4020                                 btrfs_super_sys_array_size(sb),
4021                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4022                 ret = -EINVAL;
4023         }
4024         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4025                         + sizeof(struct btrfs_chunk)) {
4026                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4027                                 btrfs_super_sys_array_size(sb),
4028                                 sizeof(struct btrfs_disk_key)
4029                                 + sizeof(struct btrfs_chunk));
4030                 ret = -EINVAL;
4031         }
4032
4033         /*
4034          * The generation is a global counter, we'll trust it more than the others
4035          * but it's still possible that it's the one that's wrong.
4036          */
4037         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4038                 printk(KERN_WARNING
4039                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4040                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4041         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4042             && btrfs_super_cache_generation(sb) != (u64)-1)
4043                 printk(KERN_WARNING
4044                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4045                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4046
4047         return ret;
4048 }
4049
4050 static void btrfs_error_commit_super(struct btrfs_root *root)
4051 {
4052         mutex_lock(&root->fs_info->cleaner_mutex);
4053         btrfs_run_delayed_iputs(root);
4054         mutex_unlock(&root->fs_info->cleaner_mutex);
4055
4056         down_write(&root->fs_info->cleanup_work_sem);
4057         up_write(&root->fs_info->cleanup_work_sem);
4058
4059         /* cleanup FS via transaction */
4060         btrfs_cleanup_transaction(root);
4061 }
4062
4063 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4064 {
4065         struct btrfs_ordered_extent *ordered;
4066
4067         spin_lock(&root->ordered_extent_lock);
4068         /*
4069          * This will just short circuit the ordered completion stuff which will
4070          * make sure the ordered extent gets properly cleaned up.
4071          */
4072         list_for_each_entry(ordered, &root->ordered_extents,
4073                             root_extent_list)
4074                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4075         spin_unlock(&root->ordered_extent_lock);
4076 }
4077
4078 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4079 {
4080         struct btrfs_root *root;
4081         struct list_head splice;
4082
4083         INIT_LIST_HEAD(&splice);
4084
4085         spin_lock(&fs_info->ordered_root_lock);
4086         list_splice_init(&fs_info->ordered_roots, &splice);
4087         while (!list_empty(&splice)) {
4088                 root = list_first_entry(&splice, struct btrfs_root,
4089                                         ordered_root);
4090                 list_move_tail(&root->ordered_root,
4091                                &fs_info->ordered_roots);
4092
4093                 spin_unlock(&fs_info->ordered_root_lock);
4094                 btrfs_destroy_ordered_extents(root);
4095
4096                 cond_resched();
4097                 spin_lock(&fs_info->ordered_root_lock);
4098         }
4099         spin_unlock(&fs_info->ordered_root_lock);
4100 }
4101
4102 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4103                                       struct btrfs_root *root)
4104 {
4105         struct rb_node *node;
4106         struct btrfs_delayed_ref_root *delayed_refs;
4107         struct btrfs_delayed_ref_node *ref;
4108         int ret = 0;
4109
4110         delayed_refs = &trans->delayed_refs;
4111
4112         spin_lock(&delayed_refs->lock);
4113         if (atomic_read(&delayed_refs->num_entries) == 0) {
4114                 spin_unlock(&delayed_refs->lock);
4115                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4116                 return ret;
4117         }
4118
4119         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4120                 struct btrfs_delayed_ref_head *head;
4121                 struct btrfs_delayed_ref_node *tmp;
4122                 bool pin_bytes = false;
4123
4124                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4125                                 href_node);
4126                 if (!mutex_trylock(&head->mutex)) {
4127                         atomic_inc(&head->node.refs);
4128                         spin_unlock(&delayed_refs->lock);
4129
4130                         mutex_lock(&head->mutex);
4131                         mutex_unlock(&head->mutex);
4132                         btrfs_put_delayed_ref(&head->node);
4133                         spin_lock(&delayed_refs->lock);
4134                         continue;
4135                 }
4136                 spin_lock(&head->lock);
4137                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4138                                                  list) {
4139                         ref->in_tree = 0;
4140                         list_del(&ref->list);
4141                         atomic_dec(&delayed_refs->num_entries);
4142                         btrfs_put_delayed_ref(ref);
4143                 }
4144                 if (head->must_insert_reserved)
4145                         pin_bytes = true;
4146                 btrfs_free_delayed_extent_op(head->extent_op);
4147                 delayed_refs->num_heads--;
4148                 if (head->processing == 0)
4149                         delayed_refs->num_heads_ready--;
4150                 atomic_dec(&delayed_refs->num_entries);
4151                 head->node.in_tree = 0;
4152                 rb_erase(&head->href_node, &delayed_refs->href_root);
4153                 spin_unlock(&head->lock);
4154                 spin_unlock(&delayed_refs->lock);
4155                 mutex_unlock(&head->mutex);
4156
4157                 if (pin_bytes)
4158                         btrfs_pin_extent(root, head->node.bytenr,
4159                                          head->node.num_bytes, 1);
4160                 btrfs_put_delayed_ref(&head->node);
4161                 cond_resched();
4162                 spin_lock(&delayed_refs->lock);
4163         }
4164
4165         spin_unlock(&delayed_refs->lock);
4166
4167         return ret;
4168 }
4169
4170 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4171 {
4172         struct btrfs_inode *btrfs_inode;
4173         struct list_head splice;
4174
4175         INIT_LIST_HEAD(&splice);
4176
4177         spin_lock(&root->delalloc_lock);
4178         list_splice_init(&root->delalloc_inodes, &splice);
4179
4180         while (!list_empty(&splice)) {
4181                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4182                                                delalloc_inodes);
4183
4184                 list_del_init(&btrfs_inode->delalloc_inodes);
4185                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4186                           &btrfs_inode->runtime_flags);
4187                 spin_unlock(&root->delalloc_lock);
4188
4189                 btrfs_invalidate_inodes(btrfs_inode->root);
4190
4191                 spin_lock(&root->delalloc_lock);
4192         }
4193
4194         spin_unlock(&root->delalloc_lock);
4195 }
4196
4197 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4198 {
4199         struct btrfs_root *root;
4200         struct list_head splice;
4201
4202         INIT_LIST_HEAD(&splice);
4203
4204         spin_lock(&fs_info->delalloc_root_lock);
4205         list_splice_init(&fs_info->delalloc_roots, &splice);
4206         while (!list_empty(&splice)) {
4207                 root = list_first_entry(&splice, struct btrfs_root,
4208                                          delalloc_root);
4209                 list_del_init(&root->delalloc_root);
4210                 root = btrfs_grab_fs_root(root);
4211                 BUG_ON(!root);
4212                 spin_unlock(&fs_info->delalloc_root_lock);
4213
4214                 btrfs_destroy_delalloc_inodes(root);
4215                 btrfs_put_fs_root(root);
4216
4217                 spin_lock(&fs_info->delalloc_root_lock);
4218         }
4219         spin_unlock(&fs_info->delalloc_root_lock);
4220 }
4221
4222 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4223                                         struct extent_io_tree *dirty_pages,
4224                                         int mark)
4225 {
4226         int ret;
4227         struct extent_buffer *eb;
4228         u64 start = 0;
4229         u64 end;
4230
4231         while (1) {
4232                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4233                                             mark, NULL);
4234                 if (ret)
4235                         break;
4236
4237                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4238                 while (start <= end) {
4239                         eb = btrfs_find_tree_block(root->fs_info, start);
4240                         start += root->nodesize;
4241                         if (!eb)
4242                                 continue;
4243                         wait_on_extent_buffer_writeback(eb);
4244
4245                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4246                                                &eb->bflags))
4247                                 clear_extent_buffer_dirty(eb);
4248                         free_extent_buffer_stale(eb);
4249                 }
4250         }
4251
4252         return ret;
4253 }
4254
4255 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4256                                        struct extent_io_tree *pinned_extents)
4257 {
4258         struct extent_io_tree *unpin;
4259         u64 start;
4260         u64 end;
4261         int ret;
4262         bool loop = true;
4263
4264         unpin = pinned_extents;
4265 again:
4266         while (1) {
4267                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4268                                             EXTENT_DIRTY, NULL);
4269                 if (ret)
4270                         break;
4271
4272                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4273                 btrfs_error_unpin_extent_range(root, start, end);
4274                 cond_resched();
4275         }
4276
4277         if (loop) {
4278                 if (unpin == &root->fs_info->freed_extents[0])
4279                         unpin = &root->fs_info->freed_extents[1];
4280                 else
4281                         unpin = &root->fs_info->freed_extents[0];
4282                 loop = false;
4283                 goto again;
4284         }
4285
4286         return 0;
4287 }
4288
4289 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4290                                        struct btrfs_fs_info *fs_info)
4291 {
4292         struct btrfs_ordered_extent *ordered;
4293
4294         spin_lock(&fs_info->trans_lock);
4295         while (!list_empty(&cur_trans->pending_ordered)) {
4296                 ordered = list_first_entry(&cur_trans->pending_ordered,
4297                                            struct btrfs_ordered_extent,
4298                                            trans_list);
4299                 list_del_init(&ordered->trans_list);
4300                 spin_unlock(&fs_info->trans_lock);
4301
4302                 btrfs_put_ordered_extent(ordered);
4303                 spin_lock(&fs_info->trans_lock);
4304         }
4305         spin_unlock(&fs_info->trans_lock);
4306 }
4307
4308 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4309                                    struct btrfs_root *root)
4310 {
4311         btrfs_destroy_delayed_refs(cur_trans, root);
4312
4313         cur_trans->state = TRANS_STATE_COMMIT_START;
4314         wake_up(&root->fs_info->transaction_blocked_wait);
4315
4316         cur_trans->state = TRANS_STATE_UNBLOCKED;
4317         wake_up(&root->fs_info->transaction_wait);
4318
4319         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4320         btrfs_destroy_delayed_inodes(root);
4321         btrfs_assert_delayed_root_empty(root);
4322
4323         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4324                                      EXTENT_DIRTY);
4325         btrfs_destroy_pinned_extent(root,
4326                                     root->fs_info->pinned_extents);
4327
4328         cur_trans->state =TRANS_STATE_COMPLETED;
4329         wake_up(&cur_trans->commit_wait);
4330
4331         /*
4332         memset(cur_trans, 0, sizeof(*cur_trans));
4333         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4334         */
4335 }
4336
4337 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4338 {
4339         struct btrfs_transaction *t;
4340
4341         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4342
4343         spin_lock(&root->fs_info->trans_lock);
4344         while (!list_empty(&root->fs_info->trans_list)) {
4345                 t = list_first_entry(&root->fs_info->trans_list,
4346                                      struct btrfs_transaction, list);
4347                 if (t->state >= TRANS_STATE_COMMIT_START) {
4348                         atomic_inc(&t->use_count);
4349                         spin_unlock(&root->fs_info->trans_lock);
4350                         btrfs_wait_for_commit(root, t->transid);
4351                         btrfs_put_transaction(t);
4352                         spin_lock(&root->fs_info->trans_lock);
4353                         continue;
4354                 }
4355                 if (t == root->fs_info->running_transaction) {
4356                         t->state = TRANS_STATE_COMMIT_DOING;
4357                         spin_unlock(&root->fs_info->trans_lock);
4358                         /*
4359                          * We wait for 0 num_writers since we don't hold a trans
4360                          * handle open currently for this transaction.
4361                          */
4362                         wait_event(t->writer_wait,
4363                                    atomic_read(&t->num_writers) == 0);
4364                 } else {
4365                         spin_unlock(&root->fs_info->trans_lock);
4366                 }
4367                 btrfs_cleanup_one_transaction(t, root);
4368
4369                 spin_lock(&root->fs_info->trans_lock);
4370                 if (t == root->fs_info->running_transaction)
4371                         root->fs_info->running_transaction = NULL;
4372                 list_del_init(&t->list);
4373                 spin_unlock(&root->fs_info->trans_lock);
4374
4375                 btrfs_put_transaction(t);
4376                 trace_btrfs_transaction_commit(root);
4377                 spin_lock(&root->fs_info->trans_lock);
4378         }
4379         spin_unlock(&root->fs_info->trans_lock);
4380         btrfs_destroy_all_ordered_extents(root->fs_info);
4381         btrfs_destroy_delayed_inodes(root);
4382         btrfs_assert_delayed_root_empty(root);
4383         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4384         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4385         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4386
4387         return 0;
4388 }
4389
4390 static const struct extent_io_ops btree_extent_io_ops = {
4391         .readpage_end_io_hook = btree_readpage_end_io_hook,
4392         .readpage_io_failed_hook = btree_io_failed_hook,
4393         .submit_bio_hook = btree_submit_bio_hook,
4394         /* note we're sharing with inode.c for the merge bio hook */
4395         .merge_bio_hook = btrfs_merge_bio_hook,
4396 };