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/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include <linux/ratelimit.h>
  33 #include <asm/unaligned.h>
  34 #include "compat.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46
  47 static struct extent_io_ops btree_extent_io_ops;
  48 static void end_workqueue_fn(struct btrfs_work *work);
  49 static void free_fs_root(struct btrfs_root *root);
  50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  51                                     int read_only);
  52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  55                                       struct btrfs_root *root);
  56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  64
  65 /*
  66  * end_io_wq structs are used to do processing in task context when an IO is
  67  * complete.  This is used during reads to verify checksums, and it is used
  68  * by writes to insert metadata for new file extents after IO is complete.
  69  */
  70 struct end_io_wq {
  71         struct bio *bio;
  72         bio_end_io_t *end_io;
  73         void *private;
  74         struct btrfs_fs_info *info;
  75         int error;
  76         int metadata;
  77         struct list_head list;
  78         struct btrfs_work work;
  79 };
  80
  81 /*
  82  * async submit bios are used to offload expensive checksumming
  83  * onto the worker threads.  They checksum file and metadata bios
  84  * just before they are sent down the IO stack.
  85  */
  86 struct async_submit_bio {
  87         struct inode *inode;
  88         struct bio *bio;
  89         struct list_head list;
  90         extent_submit_bio_hook_t *submit_bio_start;
  91         extent_submit_bio_hook_t *submit_bio_done;
  92         int rw;
  93         int mirror_num;
  94         unsigned long bio_flags;
  95         /*
  96          * bio_offset is optional, can be used if the pages in the bio
  97          * can't tell us where in the file the bio should go
  98          */
  99         u64 bio_offset;
 100         struct btrfs_work work;
 101 };
 102
 103 /*
 104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 106  * the level the eb occupies in the tree.
 107  *
 108  * Different roots are used for different purposes and may nest inside each
 109  * other and they require separate keysets.  As lockdep keys should be
 110  * static, assign keysets according to the purpose of the root as indicated
 111  * by btrfs_root->objectid.  This ensures that all special purpose roots
 112  * have separate keysets.
 113  *
 114  * Lock-nesting across peer nodes is always done with the immediate parent
 115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 116  * subclass to avoid triggering lockdep warning in such cases.
 117  *
 118  * The key is set by the readpage_end_io_hook after the buffer has passed
 119  * csum validation but before the pages are unlocked.  It is also set by
 120  * btrfs_init_new_buffer on freshly allocated blocks.
 121  *
 122  * We also add a check to make sure the highest level of the tree is the
 123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 124  * needs update as well.
 125  */
 126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 127 # if BTRFS_MAX_LEVEL != 8
 128 #  error
 129 # endif
 130
 131 static struct btrfs_lockdep_keyset {
 132         u64                     id;             /* root objectid */
 133         const char              *name_stem;     /* lock name stem */
 134         char                    names[BTRFS_MAX_LEVEL + 1][20];
 135         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 136 } btrfs_lockdep_keysets[] = {
 137         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 138         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 139         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 140         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 141         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 142         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 143         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
 144         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 145         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 146         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 147         { .id = 0,                              .name_stem = "tree"     },
 148 };
 149
 150 void __init btrfs_init_lockdep(void)
 151 {
 152         int i, j;
 153
 154         /* initialize lockdep class names */
 155         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 156                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 157
 158                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 159                         snprintf(ks->names[j], sizeof(ks->names[j]),
 160                                  "btrfs-%s-%02d", ks->name_stem, j);
 161         }
 162 }
 163
 164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 165                                     int level)
 166 {
 167         struct btrfs_lockdep_keyset *ks;
 168
 169         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 170
 171         /* find the matching keyset, id 0 is the default entry */
 172         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 173                 if (ks->id == objectid)
 174                         break;
 175
 176         lockdep_set_class_and_name(&eb->lock,
 177                                    &ks->keys[level], ks->names[level]);
 178 }
 179
 180 #endif
 181
 182 /*
 183  * extents on the btree inode are pretty simple, there's one extent
 184  * that covers the entire device
 185  */
 186 static struct extent_map *btree_get_extent(struct inode *inode,
 187                 struct page *page, size_t pg_offset, u64 start, u64 len,
 188                 int create)
 189 {
 190         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 191         struct extent_map *em;
 192         int ret;
 193
 194         read_lock(&em_tree->lock);
 195         em = lookup_extent_mapping(em_tree, start, len);
 196         if (em) {
 197                 em->bdev =
 198                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 199                 read_unlock(&em_tree->lock);
 200                 goto out;
 201         }
 202         read_unlock(&em_tree->lock);
 203
 204         em = alloc_extent_map();
 205         if (!em) {
 206                 em = ERR_PTR(-ENOMEM);
 207                 goto out;
 208         }
 209         em->start = 0;
 210         em->len = (u64)-1;
 211         em->block_len = (u64)-1;
 212         em->block_start = 0;
 213         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 214
 215         write_lock(&em_tree->lock);
 216         ret = add_extent_mapping(em_tree, em);
 217         if (ret == -EEXIST) {
 218                 u64 failed_start = em->start;
 219                 u64 failed_len = em->len;
 220
 221                 free_extent_map(em);
 222                 em = lookup_extent_mapping(em_tree, start, len);
 223                 if (em) {
 224                         ret = 0;
 225                 } else {
 226                         em = lookup_extent_mapping(em_tree, failed_start,
 227                                                    failed_len);
 228                         ret = -EIO;
 229                 }
 230         } else if (ret) {
 231                 free_extent_map(em);
 232                 em = NULL;
 233         }
 234         write_unlock(&em_tree->lock);
 235
 236         if (ret)
 237                 em = ERR_PTR(ret);
 238 out:
 239         return em;
 240 }
 241
 242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 243 {
 244         return crc32c(seed, data, len);
 245 }
 246
 247 void btrfs_csum_final(u32 crc, char *result)
 248 {
 249         put_unaligned_le32(~crc, result);
 250 }
 251
 252 /*
 253  * compute the csum for a btree block, and either verify it or write it
 254  * into the csum field of the block.
 255  */
 256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 257                            int verify)
 258 {
 259         u16 csum_size =
 260                 btrfs_super_csum_size(&root->fs_info->super_copy);
 261         char *result = NULL;
 262         unsigned long len;
 263         unsigned long cur_len;
 264         unsigned long offset = BTRFS_CSUM_SIZE;
 265         char *kaddr;
 266         unsigned long map_start;
 267         unsigned long map_len;
 268         int err;
 269         u32 crc = ~(u32)0;
 270         unsigned long inline_result;
 271
 272         len = buf->len - offset;
 273         while (len > 0) {
 274                 err = map_private_extent_buffer(buf, offset, 32,
 275                                         &kaddr, &map_start, &map_len);
 276                 if (err)
 277                         return 1;
 278                 cur_len = min(len, map_len - (offset - map_start));
 279                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 280                                       crc, cur_len);
 281                 len -= cur_len;
 282                 offset += cur_len;
 283         }
 284         if (csum_size > sizeof(inline_result)) {
 285                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 286                 if (!result)
 287                         return 1;
 288         } else {
 289                 result = (char *)&inline_result;
 290         }
 291
 292         btrfs_csum_final(crc, result);
 293
 294         if (verify) {
 295                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 296                         u32 val;
 297                         u32 found = 0;
 298                         memcpy(&found, result, csum_size);
 299
 300                         read_extent_buffer(buf, &val, 0, csum_size);
 301                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
 302                                        "failed on %llu wanted %X found %X "
 303                                        "level %d\n",
 304                                        root->fs_info->sb->s_id,
 305                                        (unsigned long long)buf->start, val, found,
 306                                        btrfs_header_level(buf));
 307                         if (result != (char *)&inline_result)
 308                                 kfree(result);
 309                         return 1;
 310                 }
 311         } else {
 312                 write_extent_buffer(buf, result, 0, csum_size);
 313         }
 314         if (result != (char *)&inline_result)
 315                 kfree(result);
 316         return 0;
 317 }
 318
 319 /*
 320  * we can't consider a given block up to date unless the transid of the
 321  * block matches the transid in the parent node's pointer.  This is how we
 322  * detect blocks that either didn't get written at all or got written
 323  * in the wrong place.
 324  */
 325 static int verify_parent_transid(struct extent_io_tree *io_tree,
 326                                  struct extent_buffer *eb, u64 parent_transid)
 327 {
 328         struct extent_state *cached_state = NULL;
 329         int ret;
 330
 331         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 332                 return 0;
 333
 334         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 335                          0, &cached_state, GFP_NOFS);
 336         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 337             btrfs_header_generation(eb) == parent_transid) {
 338                 ret = 0;
 339                 goto out;
 340         }
 341         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 342                        "found %llu\n",
 343                        (unsigned long long)eb->start,
 344                        (unsigned long long)parent_transid,
 345                        (unsigned long long)btrfs_header_generation(eb));
 346         ret = 1;
 347         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 348 out:
 349         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 350                              &cached_state, GFP_NOFS);
 351         return ret;
 352 }
 353
 354 /*
 355  * helper to read a given tree block, doing retries as required when
 356  * the checksums don't match and we have alternate mirrors to try.
 357  */
 358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 359                                           struct extent_buffer *eb,
 360                                           u64 start, u64 parent_transid)
 361 {
 362         struct extent_io_tree *io_tree;
 363         int ret;
 364         int num_copies = 0;
 365         int mirror_num = 0;
 366
 367         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 368         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 369         while (1) {
 370                 ret = read_extent_buffer_pages(io_tree, eb, start,
 371                                                WAIT_COMPLETE,
 372                                                btree_get_extent, mirror_num);
 373                 if (!ret &&
 374                     !verify_parent_transid(io_tree, eb, parent_transid))
 375                         return ret;
 376
 377                 /*
 378                  * This buffer's crc is fine, but its contents are corrupted, so
 379                  * there is no reason to read the other copies, they won't be
 380                  * any less wrong.
 381                  */
 382                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 383                         return ret;
 384
 385                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 386                                               eb->start, eb->len);
 387                 if (num_copies == 1)
 388                         return ret;
 389
 390                 mirror_num++;
 391                 if (mirror_num > num_copies)
 392                         return ret;
 393         }
 394         return -EIO;
 395 }
 396
 397 /*
 398  * checksum a dirty tree block before IO.  This has extra checks to make sure
 399  * we only fill in the checksum field in the first page of a multi-page block
 400  */
 401
 402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 403 {
 404         struct extent_io_tree *tree;
 405         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 406         u64 found_start;
 407         unsigned long len;
 408         struct extent_buffer *eb;
 409         int ret;
 410
 411         tree = &BTRFS_I(page->mapping->host)->io_tree;
 412
 413         if (page->private == EXTENT_PAGE_PRIVATE) {
 414                 WARN_ON(1);
 415                 goto out;
 416         }
 417         if (!page->private) {
 418                 WARN_ON(1);
 419                 goto out;
 420         }
 421         len = page->private >> 2;
 422         WARN_ON(len == 0);
 423
 424         eb = alloc_extent_buffer(tree, start, len, page);
 425         if (eb == NULL) {
 426                 WARN_ON(1);
 427                 goto out;
 428         }
 429         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 430                                              btrfs_header_generation(eb));
 431         BUG_ON(ret);
 432         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 433
 434         found_start = btrfs_header_bytenr(eb);
 435         if (found_start != start) {
 436                 WARN_ON(1);
 437                 goto err;
 438         }
 439         if (eb->first_page != page) {
 440                 WARN_ON(1);
 441                 goto err;
 442         }
 443         if (!PageUptodate(page)) {
 444                 WARN_ON(1);
 445                 goto err;
 446         }
 447         csum_tree_block(root, eb, 0);
 448 err:
 449         free_extent_buffer(eb);
 450 out:
 451         return 0;
 452 }
 453
 454 static int check_tree_block_fsid(struct btrfs_root *root,
 455                                  struct extent_buffer *eb)
 456 {
 457         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 458         u8 fsid[BTRFS_UUID_SIZE];
 459         int ret = 1;
 460
 461         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 462                            BTRFS_FSID_SIZE);
 463         while (fs_devices) {
 464                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 465                         ret = 0;
 466                         break;
 467                 }
 468                 fs_devices = fs_devices->seed;
 469         }
 470         return ret;
 471 }
 472
 473 #define CORRUPT(reason, eb, root, slot)                         \
 474         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 475                "root=%llu, slot=%d\n", reason,                  \
 476                (unsigned long long)btrfs_header_bytenr(eb),     \
 477                (unsigned long long)root->objectid, slot)
 478
 479 static noinline int check_leaf(struct btrfs_root *root,
 480                                struct extent_buffer *leaf)
 481 {
 482         struct btrfs_key key;
 483         struct btrfs_key leaf_key;
 484         u32 nritems = btrfs_header_nritems(leaf);
 485         int slot;
 486
 487         if (nritems == 0)
 488                 return 0;
 489
 490         /* Check the 0 item */
 491         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 492             BTRFS_LEAF_DATA_SIZE(root)) {
 493                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 494                 return -EIO;
 495         }
 496
 497         /*
 498          * Check to make sure each items keys are in the correct order and their
 499          * offsets make sense.  We only have to loop through nritems-1 because
 500          * we check the current slot against the next slot, which verifies the
 501          * next slot's offset+size makes sense and that the current's slot
 502          * offset is correct.
 503          */
 504         for (slot = 0; slot < nritems - 1; slot++) {
 505                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 506                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 507
 508                 /* Make sure the keys are in the right order */
 509                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 510                         CORRUPT("bad key order", leaf, root, slot);
 511                         return -EIO;
 512                 }
 513
 514                 /*
 515                  * Make sure the offset and ends are right, remember that the
 516                  * item data starts at the end of the leaf and grows towards the
 517                  * front.
 518                  */
 519                 if (btrfs_item_offset_nr(leaf, slot) !=
 520                         btrfs_item_end_nr(leaf, slot + 1)) {
 521                         CORRUPT("slot offset bad", leaf, root, slot);
 522                         return -EIO;
 523                 }
 524
 525                 /*
 526                  * Check to make sure that we don't point outside of the leaf,
 527                  * just incase all the items are consistent to eachother, but
 528                  * all point outside of the leaf.
 529                  */
 530                 if (btrfs_item_end_nr(leaf, slot) >
 531                     BTRFS_LEAF_DATA_SIZE(root)) {
 532                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 533                         return -EIO;
 534                 }
 535         }
 536
 537         return 0;
 538 }
 539
 540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 541                                struct extent_state *state)
 542 {
 543         struct extent_io_tree *tree;
 544         u64 found_start;
 545         int found_level;
 546         unsigned long len;
 547         struct extent_buffer *eb;
 548         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 549         int ret = 0;
 550
 551         tree = &BTRFS_I(page->mapping->host)->io_tree;
 552         if (page->private == EXTENT_PAGE_PRIVATE)
 553                 goto out;
 554         if (!page->private)
 555                 goto out;
 556
 557         len = page->private >> 2;
 558         WARN_ON(len == 0);
 559
 560         eb = alloc_extent_buffer(tree, start, len, page);
 561         if (eb == NULL) {
 562                 ret = -EIO;
 563                 goto out;
 564         }
 565
 566         found_start = btrfs_header_bytenr(eb);
 567         if (found_start != start) {
 568                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
 569                                "%llu %llu\n",
 570                                (unsigned long long)found_start,
 571                                (unsigned long long)eb->start);
 572                 ret = -EIO;
 573                 goto err;
 574         }
 575         if (eb->first_page != page) {
 576                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 577                        eb->first_page->index, page->index);
 578                 WARN_ON(1);
 579                 ret = -EIO;
 580                 goto err;
 581         }
 582         if (check_tree_block_fsid(root, eb)) {
 583                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
 584                                (unsigned long long)eb->start);
 585                 ret = -EIO;
 586                 goto err;
 587         }
 588         found_level = btrfs_header_level(eb);
 589
 590         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 591                                        eb, found_level);
 592
 593         ret = csum_tree_block(root, eb, 1);
 594         if (ret) {
 595                 ret = -EIO;
 596                 goto err;
 597         }
 598
 599         /*
 600          * If this is a leaf block and it is corrupt, set the corrupt bit so
 601          * that we don't try and read the other copies of this block, just
 602          * return -EIO.
 603          */
 604         if (found_level == 0 && check_leaf(root, eb)) {
 605                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 606                 ret = -EIO;
 607         }
 608
 609         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 610         end = eb->start + end - 1;
 611 err:
 612         free_extent_buffer(eb);
 613 out:
 614         return ret;
 615 }
 616
 617 static void end_workqueue_bio(struct bio *bio, int err)
 618 {
 619         struct end_io_wq *end_io_wq = bio->bi_private;
 620         struct btrfs_fs_info *fs_info;
 621
 622         fs_info = end_io_wq->info;
 623         end_io_wq->error = err;
 624         end_io_wq->work.func = end_workqueue_fn;
 625         end_io_wq->work.flags = 0;
 626
 627         if (bio->bi_rw & REQ_WRITE) {
 628                 if (end_io_wq->metadata == 1)
 629                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 630                                            &end_io_wq->work);
 631                 else if (end_io_wq->metadata == 2)
 632                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 633                                            &end_io_wq->work);
 634                 else
 635                         btrfs_queue_worker(&fs_info->endio_write_workers,
 636                                            &end_io_wq->work);
 637         } else {
 638                 if (end_io_wq->metadata)
 639                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 640                                            &end_io_wq->work);
 641                 else
 642                         btrfs_queue_worker(&fs_info->endio_workers,
 643                                            &end_io_wq->work);
 644         }
 645 }
 646
 647 /*
 648  * For the metadata arg you want
 649  *
 650  * 0 - if data
 651  * 1 - if normal metadta
 652  * 2 - if writing to the free space cache area
 653  */
 654 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 655                         int metadata)
 656 {
 657         struct end_io_wq *end_io_wq;
 658         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 659         if (!end_io_wq)
 660                 return -ENOMEM;
 661
 662         end_io_wq->private = bio->bi_private;
 663         end_io_wq->end_io = bio->bi_end_io;
 664         end_io_wq->info = info;
 665         end_io_wq->error = 0;
 666         end_io_wq->bio = bio;
 667         end_io_wq->metadata = metadata;
 668
 669         bio->bi_private = end_io_wq;
 670         bio->bi_end_io = end_workqueue_bio;
 671         return 0;
 672 }
 673
 674 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 675 {
 676         unsigned long limit = min_t(unsigned long,
 677                                     info->workers.max_workers,
 678                                     info->fs_devices->open_devices);
 679         return 256 * limit;
 680 }
 681
 682 static void run_one_async_start(struct btrfs_work *work)
 683 {
 684         struct async_submit_bio *async;
 685
 686         async = container_of(work, struct  async_submit_bio, work);
 687         async->submit_bio_start(async->inode, async->rw, async->bio,
 688                                async->mirror_num, async->bio_flags,
 689                                async->bio_offset);
 690 }
 691
 692 static void run_one_async_done(struct btrfs_work *work)
 693 {
 694         struct btrfs_fs_info *fs_info;
 695         struct async_submit_bio *async;
 696         int limit;
 697
 698         async = container_of(work, struct  async_submit_bio, work);
 699         fs_info = BTRFS_I(async->inode)->root->fs_info;
 700
 701         limit = btrfs_async_submit_limit(fs_info);
 702         limit = limit * 2 / 3;
 703
 704         atomic_dec(&fs_info->nr_async_submits);
 705
 706         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 707             waitqueue_active(&fs_info->async_submit_wait))
 708                 wake_up(&fs_info->async_submit_wait);
 709
 710         async->submit_bio_done(async->inode, async->rw, async->bio,
 711                                async->mirror_num, async->bio_flags,
 712                                async->bio_offset);
 713 }
 714
 715 static void run_one_async_free(struct btrfs_work *work)
 716 {
 717         struct async_submit_bio *async;
 718
 719         async = container_of(work, struct  async_submit_bio, work);
 720         kfree(async);
 721 }
 722
 723 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 724                         int rw, struct bio *bio, int mirror_num,
 725                         unsigned long bio_flags,
 726                         u64 bio_offset,
 727                         extent_submit_bio_hook_t *submit_bio_start,
 728                         extent_submit_bio_hook_t *submit_bio_done)
 729 {
 730         struct async_submit_bio *async;
 731
 732         async = kmalloc(sizeof(*async), GFP_NOFS);
 733         if (!async)
 734                 return -ENOMEM;
 735
 736         async->inode = inode;
 737         async->rw = rw;
 738         async->bio = bio;
 739         async->mirror_num = mirror_num;
 740         async->submit_bio_start = submit_bio_start;
 741         async->submit_bio_done = submit_bio_done;
 742
 743         async->work.func = run_one_async_start;
 744         async->work.ordered_func = run_one_async_done;
 745         async->work.ordered_free = run_one_async_free;
 746
 747         async->work.flags = 0;
 748         async->bio_flags = bio_flags;
 749         async->bio_offset = bio_offset;
 750
 751         atomic_inc(&fs_info->nr_async_submits);
 752
 753         if (rw & REQ_SYNC)
 754                 btrfs_set_work_high_prio(&async->work);
 755
 756         btrfs_queue_worker(&fs_info->workers, &async->work);
 757
 758         while (atomic_read(&fs_info->async_submit_draining) &&
 759               atomic_read(&fs_info->nr_async_submits)) {
 760                 wait_event(fs_info->async_submit_wait,
 761                            (atomic_read(&fs_info->nr_async_submits) == 0));
 762         }
 763
 764         return 0;
 765 }
 766
 767 static int btree_csum_one_bio(struct bio *bio)
 768 {
 769         struct bio_vec *bvec = bio->bi_io_vec;
 770         int bio_index = 0;
 771         struct btrfs_root *root;
 772
 773         WARN_ON(bio->bi_vcnt <= 0);
 774         while (bio_index < bio->bi_vcnt) {
 775                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 776                 csum_dirty_buffer(root, bvec->bv_page);
 777                 bio_index++;
 778                 bvec++;
 779         }
 780         return 0;
 781 }
 782
 783 static int __btree_submit_bio_start(struct inode *inode, int rw,
 784                                     struct bio *bio, int mirror_num,
 785                                     unsigned long bio_flags,
 786                                     u64 bio_offset)
 787 {
 788         /*
 789          * when we're called for a write, we're already in the async
 790          * submission context.  Just jump into btrfs_map_bio
 791          */
 792         btree_csum_one_bio(bio);
 793         return 0;
 794 }
 795
 796 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 797                                  int mirror_num, unsigned long bio_flags,
 798                                  u64 bio_offset)
 799 {
 800         /*
 801          * when we're called for a write, we're already in the async
 802          * submission context.  Just jump into btrfs_map_bio
 803          */
 804         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 805 }
 806
 807 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 808                                  int mirror_num, unsigned long bio_flags,
 809                                  u64 bio_offset)
 810 {
 811         int ret;
 812
 813         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 814                                           bio, 1);
 815         BUG_ON(ret);
 816
 817         if (!(rw & REQ_WRITE)) {
 818                 /*
 819                  * called for a read, do the setup so that checksum validation
 820                  * can happen in the async kernel threads
 821                  */
 822                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 823                                      mirror_num, 0);
 824         }
 825
 826         /*
 827          * kthread helpers are used to submit writes so that checksumming
 828          * can happen in parallel across all CPUs
 829          */
 830         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 831                                    inode, rw, bio, mirror_num, 0,
 832                                    bio_offset,
 833                                    __btree_submit_bio_start,
 834                                    __btree_submit_bio_done);
 835 }
 836
 837 #ifdef CONFIG_MIGRATION
 838 static int btree_migratepage(struct address_space *mapping,
 839                         struct page *newpage, struct page *page)
 840 {
 841         /*
 842          * we can't safely write a btree page from here,
 843          * we haven't done the locking hook
 844          */
 845         if (PageDirty(page))
 846                 return -EAGAIN;
 847         /*
 848          * Buffers may be managed in a filesystem specific way.
 849          * We must have no buffers or drop them.
 850          */
 851         if (page_has_private(page) &&
 852             !try_to_release_page(page, GFP_KERNEL))
 853                 return -EAGAIN;
 854         return migrate_page(mapping, newpage, page);
 855 }
 856 #endif
 857
 858 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 859 {
 860         struct extent_io_tree *tree;
 861         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 862         struct extent_buffer *eb;
 863         int was_dirty;
 864
 865         tree = &BTRFS_I(page->mapping->host)->io_tree;
 866         if (!(current->flags & PF_MEMALLOC)) {
 867                 return extent_write_full_page(tree, page,
 868                                               btree_get_extent, wbc);
 869         }
 870
 871         redirty_page_for_writepage(wbc, page);
 872         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 873         WARN_ON(!eb);
 874
 875         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 876         if (!was_dirty) {
 877                 spin_lock(&root->fs_info->delalloc_lock);
 878                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 879                 spin_unlock(&root->fs_info->delalloc_lock);
 880         }
 881         free_extent_buffer(eb);
 882
 883         unlock_page(page);
 884         return 0;
 885 }
 886
 887 static int btree_writepages(struct address_space *mapping,
 888                             struct writeback_control *wbc)
 889 {
 890         struct extent_io_tree *tree;
 891         tree = &BTRFS_I(mapping->host)->io_tree;
 892         if (wbc->sync_mode == WB_SYNC_NONE) {
 893                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 894                 u64 num_dirty;
 895                 unsigned long thresh = 32 * 1024 * 1024;
 896
 897                 if (wbc->for_kupdate)
 898                         return 0;
 899
 900                 /* this is a bit racy, but that's ok */
 901                 num_dirty = root->fs_info->dirty_metadata_bytes;
 902                 if (num_dirty < thresh)
 903                         return 0;
 904         }
 905         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 906 }
 907
 908 static int btree_readpage(struct file *file, struct page *page)
 909 {
 910         struct extent_io_tree *tree;
 911         tree = &BTRFS_I(page->mapping->host)->io_tree;
 912         return extent_read_full_page(tree, page, btree_get_extent);
 913 }
 914
 915 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 916 {
 917         struct extent_io_tree *tree;
 918         struct extent_map_tree *map;
 919         int ret;
 920
 921         if (PageWriteback(page) || PageDirty(page))
 922                 return 0;
 923
 924         tree = &BTRFS_I(page->mapping->host)->io_tree;
 925         map = &BTRFS_I(page->mapping->host)->extent_tree;
 926
 927         ret = try_release_extent_state(map, tree, page, gfp_flags);
 928         if (!ret)
 929                 return 0;
 930
 931         ret = try_release_extent_buffer(tree, page);
 932         if (ret == 1) {
 933                 ClearPagePrivate(page);
 934                 set_page_private(page, 0);
 935                 page_cache_release(page);
 936         }
 937
 938         return ret;
 939 }
 940
 941 static void btree_invalidatepage(struct page *page, unsigned long offset)
 942 {
 943         struct extent_io_tree *tree;
 944         tree = &BTRFS_I(page->mapping->host)->io_tree;
 945         extent_invalidatepage(tree, page, offset);
 946         btree_releasepage(page, GFP_NOFS);
 947         if (PagePrivate(page)) {
 948                 printk(KERN_WARNING "btrfs warning page private not zero "
 949                        "on page %llu\n", (unsigned long long)page_offset(page));
 950                 ClearPagePrivate(page);
 951                 set_page_private(page, 0);
 952                 page_cache_release(page);
 953         }
 954 }
 955
 956 static const struct address_space_operations btree_aops = {
 957         .readpage       = btree_readpage,
 958         .writepage      = btree_writepage,
 959         .writepages     = btree_writepages,
 960         .releasepage    = btree_releasepage,
 961         .invalidatepage = btree_invalidatepage,
 962 #ifdef CONFIG_MIGRATION
 963         .migratepage    = btree_migratepage,
 964 #endif
 965 };
 966
 967 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 968                          u64 parent_transid)
 969 {
 970         struct extent_buffer *buf = NULL;
 971         struct inode *btree_inode = root->fs_info->btree_inode;
 972         int ret = 0;
 973
 974         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 975         if (!buf)
 976                 return 0;
 977         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 978                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
 979         free_extent_buffer(buf);
 980         return ret;
 981 }
 982
 983 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 984                                             u64 bytenr, u32 blocksize)
 985 {
 986         struct inode *btree_inode = root->fs_info->btree_inode;
 987         struct extent_buffer *eb;
 988         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 989                                 bytenr, blocksize);
 990         return eb;
 991 }
 992
 993 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 994                                                  u64 bytenr, u32 blocksize)
 995 {
 996         struct inode *btree_inode = root->fs_info->btree_inode;
 997         struct extent_buffer *eb;
 998
 999         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1000                                  bytenr, blocksize, NULL);
1001         return eb;
1002 }
1003
1004
1005 int btrfs_write_tree_block(struct extent_buffer *buf)
1006 {
1007         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1008                                         buf->start + buf->len - 1);
1009 }
1010
1011 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1012 {
1013         return filemap_fdatawait_range(buf->first_page->mapping,
1014                                        buf->start, buf->start + buf->len - 1);
1015 }
1016
1017 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1018                                       u32 blocksize, u64 parent_transid)
1019 {
1020         struct extent_buffer *buf = NULL;
1021         int ret;
1022
1023         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1024         if (!buf)
1025                 return NULL;
1026
1027         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1028
1029         if (ret == 0)
1030                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1031         return buf;
1032
1033 }
1034
1035 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1036                      struct extent_buffer *buf)
1037 {
1038         struct inode *btree_inode = root->fs_info->btree_inode;
1039         if (btrfs_header_generation(buf) ==
1040             root->fs_info->running_transaction->transid) {
1041                 btrfs_assert_tree_locked(buf);
1042
1043                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1044                         spin_lock(&root->fs_info->delalloc_lock);
1045                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1046                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1047                         else
1048                                 WARN_ON(1);
1049                         spin_unlock(&root->fs_info->delalloc_lock);
1050                 }
1051
1052                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1053                 btrfs_set_lock_blocking(buf);
1054                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1055                                           buf);
1056         }
1057         return 0;
1058 }
1059
1060 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1061                         u32 stripesize, struct btrfs_root *root,
1062                         struct btrfs_fs_info *fs_info,
1063                         u64 objectid)
1064 {
1065         root->node = NULL;
1066         root->commit_root = NULL;
1067         root->sectorsize = sectorsize;
1068         root->nodesize = nodesize;
1069         root->leafsize = leafsize;
1070         root->stripesize = stripesize;
1071         root->ref_cows = 0;
1072         root->track_dirty = 0;
1073         root->in_radix = 0;
1074         root->orphan_item_inserted = 0;
1075         root->orphan_cleanup_state = 0;
1076
1077         root->fs_info = fs_info;
1078         root->objectid = objectid;
1079         root->last_trans = 0;
1080         root->highest_objectid = 0;
1081         root->name = NULL;
1082         root->inode_tree = RB_ROOT;
1083         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1084         root->block_rsv = NULL;
1085         root->orphan_block_rsv = NULL;
1086
1087         INIT_LIST_HEAD(&root->dirty_list);
1088         INIT_LIST_HEAD(&root->orphan_list);
1089         INIT_LIST_HEAD(&root->root_list);
1090         spin_lock_init(&root->orphan_lock);
1091         spin_lock_init(&root->inode_lock);
1092         spin_lock_init(&root->accounting_lock);
1093         mutex_init(&root->objectid_mutex);
1094         mutex_init(&root->log_mutex);
1095         init_waitqueue_head(&root->log_writer_wait);
1096         init_waitqueue_head(&root->log_commit_wait[0]);
1097         init_waitqueue_head(&root->log_commit_wait[1]);
1098         atomic_set(&root->log_commit[0], 0);
1099         atomic_set(&root->log_commit[1], 0);
1100         atomic_set(&root->log_writers, 0);
1101         root->log_batch = 0;
1102         root->log_transid = 0;
1103         root->last_log_commit = 0;
1104         extent_io_tree_init(&root->dirty_log_pages,
1105                              fs_info->btree_inode->i_mapping);
1106
1107         memset(&root->root_key, 0, sizeof(root->root_key));
1108         memset(&root->root_item, 0, sizeof(root->root_item));
1109         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1110         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1111         root->defrag_trans_start = fs_info->generation;
1112         init_completion(&root->kobj_unregister);
1113         root->defrag_running = 0;
1114         root->root_key.objectid = objectid;
1115         root->anon_dev = 0;
1116         return 0;
1117 }
1118
1119 static int find_and_setup_root(struct btrfs_root *tree_root,
1120                                struct btrfs_fs_info *fs_info,
1121                                u64 objectid,
1122                                struct btrfs_root *root)
1123 {
1124         int ret;
1125         u32 blocksize;
1126         u64 generation;
1127
1128         __setup_root(tree_root->nodesize, tree_root->leafsize,
1129                      tree_root->sectorsize, tree_root->stripesize,
1130                      root, fs_info, objectid);
1131         ret = btrfs_find_last_root(tree_root, objectid,
1132                                    &root->root_item, &root->root_key);
1133         if (ret > 0)
1134                 return -ENOENT;
1135         BUG_ON(ret);
1136
1137         generation = btrfs_root_generation(&root->root_item);
1138         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1139         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1140                                      blocksize, generation);
1141         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1142                 free_extent_buffer(root->node);
1143                 return -EIO;
1144         }
1145         root->commit_root = btrfs_root_node(root);
1146         return 0;
1147 }
1148
1149 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1150                                          struct btrfs_fs_info *fs_info)
1151 {
1152         struct btrfs_root *root;
1153         struct btrfs_root *tree_root = fs_info->tree_root;
1154         struct extent_buffer *leaf;
1155
1156         root = kzalloc(sizeof(*root), GFP_NOFS);
1157         if (!root)
1158                 return ERR_PTR(-ENOMEM);
1159
1160         __setup_root(tree_root->nodesize, tree_root->leafsize,
1161                      tree_root->sectorsize, tree_root->stripesize,
1162                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1163
1164         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1165         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1166         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1167         /*
1168          * log trees do not get reference counted because they go away
1169          * before a real commit is actually done.  They do store pointers
1170          * to file data extents, and those reference counts still get
1171          * updated (along with back refs to the log tree).
1172          */
1173         root->ref_cows = 0;
1174
1175         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1176                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1177         if (IS_ERR(leaf)) {
1178                 kfree(root);
1179                 return ERR_CAST(leaf);
1180         }
1181
1182         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1183         btrfs_set_header_bytenr(leaf, leaf->start);
1184         btrfs_set_header_generation(leaf, trans->transid);
1185         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1186         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1187         root->node = leaf;
1188
1189         write_extent_buffer(root->node, root->fs_info->fsid,
1190                             (unsigned long)btrfs_header_fsid(root->node),
1191                             BTRFS_FSID_SIZE);
1192         btrfs_mark_buffer_dirty(root->node);
1193         btrfs_tree_unlock(root->node);
1194         return root;
1195 }
1196
1197 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1198                              struct btrfs_fs_info *fs_info)
1199 {
1200         struct btrfs_root *log_root;
1201
1202         log_root = alloc_log_tree(trans, fs_info);
1203         if (IS_ERR(log_root))
1204                 return PTR_ERR(log_root);
1205         WARN_ON(fs_info->log_root_tree);
1206         fs_info->log_root_tree = log_root;
1207         return 0;
1208 }
1209
1210 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1211                        struct btrfs_root *root)
1212 {
1213         struct btrfs_root *log_root;
1214         struct btrfs_inode_item *inode_item;
1215
1216         log_root = alloc_log_tree(trans, root->fs_info);
1217         if (IS_ERR(log_root))
1218                 return PTR_ERR(log_root);
1219
1220         log_root->last_trans = trans->transid;
1221         log_root->root_key.offset = root->root_key.objectid;
1222
1223         inode_item = &log_root->root_item.inode;
1224         inode_item->generation = cpu_to_le64(1);
1225         inode_item->size = cpu_to_le64(3);
1226         inode_item->nlink = cpu_to_le32(1);
1227         inode_item->nbytes = cpu_to_le64(root->leafsize);
1228         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1229
1230         btrfs_set_root_node(&log_root->root_item, log_root->node);
1231
1232         WARN_ON(root->log_root);
1233         root->log_root = log_root;
1234         root->log_transid = 0;
1235         root->last_log_commit = 0;
1236         return 0;
1237 }
1238
1239 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1240                                                struct btrfs_key *location)
1241 {
1242         struct btrfs_root *root;
1243         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1244         struct btrfs_path *path;
1245         struct extent_buffer *l;
1246         u64 generation;
1247         u32 blocksize;
1248         int ret = 0;
1249
1250         root = kzalloc(sizeof(*root), GFP_NOFS);
1251         if (!root)
1252                 return ERR_PTR(-ENOMEM);
1253         if (location->offset == (u64)-1) {
1254                 ret = find_and_setup_root(tree_root, fs_info,
1255                                           location->objectid, root);
1256                 if (ret) {
1257                         kfree(root);
1258                         return ERR_PTR(ret);
1259                 }
1260                 goto out;
1261         }
1262
1263         __setup_root(tree_root->nodesize, tree_root->leafsize,
1264                      tree_root->sectorsize, tree_root->stripesize,
1265                      root, fs_info, location->objectid);
1266
1267         path = btrfs_alloc_path();
1268         if (!path) {
1269                 kfree(root);
1270                 return ERR_PTR(-ENOMEM);
1271         }
1272         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1273         if (ret == 0) {
1274                 l = path->nodes[0];
1275                 read_extent_buffer(l, &root->root_item,
1276                                 btrfs_item_ptr_offset(l, path->slots[0]),
1277                                 sizeof(root->root_item));
1278                 memcpy(&root->root_key, location, sizeof(*location));
1279         }
1280         btrfs_free_path(path);
1281         if (ret) {
1282                 kfree(root);
1283                 if (ret > 0)
1284                         ret = -ENOENT;
1285                 return ERR_PTR(ret);
1286         }
1287
1288         generation = btrfs_root_generation(&root->root_item);
1289         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1290         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1291                                      blocksize, generation);
1292         root->commit_root = btrfs_root_node(root);
1293         BUG_ON(!root->node);
1294 out:
1295         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1296                 root->ref_cows = 1;
1297                 btrfs_check_and_init_root_item(&root->root_item);
1298         }
1299
1300         return root;
1301 }
1302
1303 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1304                                               struct btrfs_key *location)
1305 {
1306         struct btrfs_root *root;
1307         int ret;
1308
1309         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1310                 return fs_info->tree_root;
1311         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1312                 return fs_info->extent_root;
1313         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1314                 return fs_info->chunk_root;
1315         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1316                 return fs_info->dev_root;
1317         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1318                 return fs_info->csum_root;
1319 again:
1320         spin_lock(&fs_info->fs_roots_radix_lock);
1321         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1322                                  (unsigned long)location->objectid);
1323         spin_unlock(&fs_info->fs_roots_radix_lock);
1324         if (root)
1325                 return root;
1326
1327         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1328         if (IS_ERR(root))
1329                 return root;
1330
1331         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1332         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1333                                         GFP_NOFS);
1334         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1335                 ret = -ENOMEM;
1336                 goto fail;
1337         }
1338
1339         btrfs_init_free_ino_ctl(root);
1340         mutex_init(&root->fs_commit_mutex);
1341         spin_lock_init(&root->cache_lock);
1342         init_waitqueue_head(&root->cache_wait);
1343
1344         ret = get_anon_bdev(&root->anon_dev);
1345         if (ret)
1346                 goto fail;
1347
1348         if (btrfs_root_refs(&root->root_item) == 0) {
1349                 ret = -ENOENT;
1350                 goto fail;
1351         }
1352
1353         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1354         if (ret < 0)
1355                 goto fail;
1356         if (ret == 0)
1357                 root->orphan_item_inserted = 1;
1358
1359         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1360         if (ret)
1361                 goto fail;
1362
1363         spin_lock(&fs_info->fs_roots_radix_lock);
1364         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1365                                 (unsigned long)root->root_key.objectid,
1366                                 root);
1367         if (ret == 0)
1368                 root->in_radix = 1;
1369
1370         spin_unlock(&fs_info->fs_roots_radix_lock);
1371         radix_tree_preload_end();
1372         if (ret) {
1373                 if (ret == -EEXIST) {
1374                         free_fs_root(root);
1375                         goto again;
1376                 }
1377                 goto fail;
1378         }
1379
1380         ret = btrfs_find_dead_roots(fs_info->tree_root,
1381                                     root->root_key.objectid);
1382         WARN_ON(ret);
1383         return root;
1384 fail:
1385         free_fs_root(root);
1386         return ERR_PTR(ret);
1387 }
1388
1389 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1390 {
1391         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1392         int ret = 0;
1393         struct btrfs_device *device;
1394         struct backing_dev_info *bdi;
1395
1396         rcu_read_lock();
1397         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1398                 if (!device->bdev)
1399                         continue;
1400                 bdi = blk_get_backing_dev_info(device->bdev);
1401                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1402                         ret = 1;
1403                         break;
1404                 }
1405         }
1406         rcu_read_unlock();
1407         return ret;
1408 }
1409
1410 /*
1411  * If this fails, caller must call bdi_destroy() to get rid of the
1412  * bdi again.
1413  */
1414 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1415 {
1416         int err;
1417
1418         bdi->capabilities = BDI_CAP_MAP_COPY;
1419         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1420         if (err)
1421                 return err;
1422
1423         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1424         bdi->congested_fn       = btrfs_congested_fn;
1425         bdi->congested_data     = info;
1426         return 0;
1427 }
1428
1429 static int bio_ready_for_csum(struct bio *bio)
1430 {
1431         u64 length = 0;
1432         u64 buf_len = 0;
1433         u64 start = 0;
1434         struct page *page;
1435         struct extent_io_tree *io_tree = NULL;
1436         struct bio_vec *bvec;
1437         int i;
1438         int ret;
1439
1440         bio_for_each_segment(bvec, bio, i) {
1441                 page = bvec->bv_page;
1442                 if (page->private == EXTENT_PAGE_PRIVATE) {
1443                         length += bvec->bv_len;
1444                         continue;
1445                 }
1446                 if (!page->private) {
1447                         length += bvec->bv_len;
1448                         continue;
1449                 }
1450                 length = bvec->bv_len;
1451                 buf_len = page->private >> 2;
1452                 start = page_offset(page) + bvec->bv_offset;
1453                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1454         }
1455         /* are we fully contained in this bio? */
1456         if (buf_len <= length)
1457                 return 1;
1458
1459         ret = extent_range_uptodate(io_tree, start + length,
1460                                     start + buf_len - 1);
1461         return ret;
1462 }
1463
1464 /*
1465  * called by the kthread helper functions to finally call the bio end_io
1466  * functions.  This is where read checksum verification actually happens
1467  */
1468 static void end_workqueue_fn(struct btrfs_work *work)
1469 {
1470         struct bio *bio;
1471         struct end_io_wq *end_io_wq;
1472         struct btrfs_fs_info *fs_info;
1473         int error;
1474
1475         end_io_wq = container_of(work, struct end_io_wq, work);
1476         bio = end_io_wq->bio;
1477         fs_info = end_io_wq->info;
1478
1479         /* metadata bio reads are special because the whole tree block must
1480          * be checksummed at once.  This makes sure the entire block is in
1481          * ram and up to date before trying to verify things.  For
1482          * blocksize <= pagesize, it is basically a noop
1483          */
1484         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1485             !bio_ready_for_csum(bio)) {
1486                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1487                                    &end_io_wq->work);
1488                 return;
1489         }
1490         error = end_io_wq->error;
1491         bio->bi_private = end_io_wq->private;
1492         bio->bi_end_io = end_io_wq->end_io;
1493         kfree(end_io_wq);
1494         bio_endio(bio, error);
1495 }
1496
1497 static int cleaner_kthread(void *arg)
1498 {
1499         struct btrfs_root *root = arg;
1500
1501         do {
1502                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1503
1504                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1505                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1506                         btrfs_run_delayed_iputs(root);
1507                         btrfs_clean_old_snapshots(root);
1508                         mutex_unlock(&root->fs_info->cleaner_mutex);
1509                         btrfs_run_defrag_inodes(root->fs_info);
1510                 }
1511
1512                 if (freezing(current)) {
1513                         refrigerator();
1514                 } else {
1515                         set_current_state(TASK_INTERRUPTIBLE);
1516                         if (!kthread_should_stop())
1517                                 schedule();
1518                         __set_current_state(TASK_RUNNING);
1519                 }
1520         } while (!kthread_should_stop());
1521         return 0;
1522 }
1523
1524 static int transaction_kthread(void *arg)
1525 {
1526         struct btrfs_root *root = arg;
1527         struct btrfs_trans_handle *trans;
1528         struct btrfs_transaction *cur;
1529         u64 transid;
1530         unsigned long now;
1531         unsigned long delay;
1532         int ret;
1533
1534         do {
1535                 delay = HZ * 30;
1536                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1538
1539                 spin_lock(&root->fs_info->trans_lock);
1540                 cur = root->fs_info->running_transaction;
1541                 if (!cur) {
1542                         spin_unlock(&root->fs_info->trans_lock);
1543                         goto sleep;
1544                 }
1545
1546                 now = get_seconds();
1547                 if (!cur->blocked &&
1548                     (now < cur->start_time || now - cur->start_time < 30)) {
1549                         spin_unlock(&root->fs_info->trans_lock);
1550                         delay = HZ * 5;
1551                         goto sleep;
1552                 }
1553                 transid = cur->transid;
1554                 spin_unlock(&root->fs_info->trans_lock);
1555
1556                 trans = btrfs_join_transaction(root);
1557                 BUG_ON(IS_ERR(trans));
1558                 if (transid == trans->transid) {
1559                         ret = btrfs_commit_transaction(trans, root);
1560                         BUG_ON(ret);
1561                 } else {
1562                         btrfs_end_transaction(trans, root);
1563                 }
1564 sleep:
1565                 wake_up_process(root->fs_info->cleaner_kthread);
1566                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1567
1568                 if (freezing(current)) {
1569                         refrigerator();
1570                 } else {
1571                         set_current_state(TASK_INTERRUPTIBLE);
1572                         if (!kthread_should_stop() &&
1573                             !btrfs_transaction_blocked(root->fs_info))
1574                                 schedule_timeout(delay);
1575                         __set_current_state(TASK_RUNNING);
1576                 }
1577         } while (!kthread_should_stop());
1578         return 0;
1579 }
1580
1581 struct btrfs_root *open_ctree(struct super_block *sb,
1582                               struct btrfs_fs_devices *fs_devices,
1583                               char *options)
1584 {
1585         u32 sectorsize;
1586         u32 nodesize;
1587         u32 leafsize;
1588         u32 blocksize;
1589         u32 stripesize;
1590         u64 generation;
1591         u64 features;
1592         struct btrfs_key location;
1593         struct buffer_head *bh;
1594         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1595                                                  GFP_NOFS);
1596         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1597                                                  GFP_NOFS);
1598         struct btrfs_root *tree_root = btrfs_sb(sb);
1599         struct btrfs_fs_info *fs_info = NULL;
1600         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1601                                                 GFP_NOFS);
1602         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1603                                               GFP_NOFS);
1604         struct btrfs_root *log_tree_root;
1605
1606         int ret;
1607         int err = -EINVAL;
1608
1609         struct btrfs_super_block *disk_super;
1610
1611         if (!extent_root || !tree_root || !tree_root->fs_info ||
1612             !chunk_root || !dev_root || !csum_root) {
1613                 err = -ENOMEM;
1614                 goto fail;
1615         }
1616         fs_info = tree_root->fs_info;
1617
1618         ret = init_srcu_struct(&fs_info->subvol_srcu);
1619         if (ret) {
1620                 err = ret;
1621                 goto fail;
1622         }
1623
1624         ret = setup_bdi(fs_info, &fs_info->bdi);
1625         if (ret) {
1626                 err = ret;
1627                 goto fail_srcu;
1628         }
1629
1630         fs_info->btree_inode = new_inode(sb);
1631         if (!fs_info->btree_inode) {
1632                 err = -ENOMEM;
1633                 goto fail_bdi;
1634         }
1635
1636         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1637
1638         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1639         INIT_LIST_HEAD(&fs_info->trans_list);
1640         INIT_LIST_HEAD(&fs_info->dead_roots);
1641         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1642         INIT_LIST_HEAD(&fs_info->hashers);
1643         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1644         INIT_LIST_HEAD(&fs_info->ordered_operations);
1645         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1646         spin_lock_init(&fs_info->delalloc_lock);
1647         spin_lock_init(&fs_info->trans_lock);
1648         spin_lock_init(&fs_info->ref_cache_lock);
1649         spin_lock_init(&fs_info->fs_roots_radix_lock);
1650         spin_lock_init(&fs_info->delayed_iput_lock);
1651         spin_lock_init(&fs_info->defrag_inodes_lock);
1652         mutex_init(&fs_info->reloc_mutex);
1653
1654         init_completion(&fs_info->kobj_unregister);
1655         fs_info->tree_root = tree_root;
1656         fs_info->extent_root = extent_root;
1657         fs_info->csum_root = csum_root;
1658         fs_info->chunk_root = chunk_root;
1659         fs_info->dev_root = dev_root;
1660         fs_info->fs_devices = fs_devices;
1661         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1662         INIT_LIST_HEAD(&fs_info->space_info);
1663         btrfs_mapping_init(&fs_info->mapping_tree);
1664         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1665         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1666         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1667         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1668         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1669         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1670         mutex_init(&fs_info->durable_block_rsv_mutex);
1671         atomic_set(&fs_info->nr_async_submits, 0);
1672         atomic_set(&fs_info->async_delalloc_pages, 0);
1673         atomic_set(&fs_info->async_submit_draining, 0);
1674         atomic_set(&fs_info->nr_async_bios, 0);
1675         atomic_set(&fs_info->defrag_running, 0);
1676         fs_info->sb = sb;
1677         fs_info->max_inline = 8192 * 1024;
1678         fs_info->metadata_ratio = 0;
1679         fs_info->defrag_inodes = RB_ROOT;
1680         fs_info->trans_no_join = 0;
1681
1682         fs_info->thread_pool_size = min_t(unsigned long,
1683                                           num_online_cpus() + 2, 8);
1684
1685         INIT_LIST_HEAD(&fs_info->ordered_extents);
1686         spin_lock_init(&fs_info->ordered_extent_lock);
1687         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1688                                         GFP_NOFS);
1689         if (!fs_info->delayed_root) {
1690                 err = -ENOMEM;
1691                 goto fail_iput;
1692         }
1693         btrfs_init_delayed_root(fs_info->delayed_root);
1694
1695         mutex_init(&fs_info->scrub_lock);
1696         atomic_set(&fs_info->scrubs_running, 0);
1697         atomic_set(&fs_info->scrub_pause_req, 0);
1698         atomic_set(&fs_info->scrubs_paused, 0);
1699         atomic_set(&fs_info->scrub_cancel_req, 0);
1700         init_waitqueue_head(&fs_info->scrub_pause_wait);
1701         init_rwsem(&fs_info->scrub_super_lock);
1702         fs_info->scrub_workers_refcnt = 0;
1703
1704         sb->s_blocksize = 4096;
1705         sb->s_blocksize_bits = blksize_bits(4096);
1706         sb->s_bdi = &fs_info->bdi;
1707
1708         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1709         fs_info->btree_inode->i_nlink = 1;
1710         /*
1711          * we set the i_size on the btree inode to the max possible int.
1712          * the real end of the address space is determined by all of
1713          * the devices in the system
1714          */
1715         fs_info->btree_inode->i_size = OFFSET_MAX;
1716         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1717         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1718
1719         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1720         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1721                              fs_info->btree_inode->i_mapping);
1722         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1723
1724         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1725
1726         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1727         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1728                sizeof(struct btrfs_key));
1729         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1730         insert_inode_hash(fs_info->btree_inode);
1731
1732         spin_lock_init(&fs_info->block_group_cache_lock);
1733         fs_info->block_group_cache_tree = RB_ROOT;
1734
1735         extent_io_tree_init(&fs_info->freed_extents[0],
1736                              fs_info->btree_inode->i_mapping);
1737         extent_io_tree_init(&fs_info->freed_extents[1],
1738                              fs_info->btree_inode->i_mapping);
1739         fs_info->pinned_extents = &fs_info->freed_extents[0];
1740         fs_info->do_barriers = 1;
1741
1742
1743         mutex_init(&fs_info->ordered_operations_mutex);
1744         mutex_init(&fs_info->tree_log_mutex);
1745         mutex_init(&fs_info->chunk_mutex);
1746         mutex_init(&fs_info->transaction_kthread_mutex);
1747         mutex_init(&fs_info->cleaner_mutex);
1748         mutex_init(&fs_info->volume_mutex);
1749         init_rwsem(&fs_info->extent_commit_sem);
1750         init_rwsem(&fs_info->cleanup_work_sem);
1751         init_rwsem(&fs_info->subvol_sem);
1752
1753         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1754         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1755
1756         init_waitqueue_head(&fs_info->transaction_throttle);
1757         init_waitqueue_head(&fs_info->transaction_wait);
1758         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1759         init_waitqueue_head(&fs_info->async_submit_wait);
1760
1761         __setup_root(4096, 4096, 4096, 4096, tree_root,
1762                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1763
1764         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1765         if (!bh) {
1766                 err = -EINVAL;
1767                 goto fail_alloc;
1768         }
1769
1770         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1771         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1772                sizeof(fs_info->super_for_commit));
1773         brelse(bh);
1774
1775         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1776
1777         disk_super = &fs_info->super_copy;
1778         if (!btrfs_super_root(disk_super))
1779                 goto fail_alloc;
1780
1781         /* check FS state, whether FS is broken. */
1782         fs_info->fs_state |= btrfs_super_flags(disk_super);
1783
1784         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1785
1786         /*
1787          * In the long term, we'll store the compression type in the super
1788          * block, and it'll be used for per file compression control.
1789          */
1790         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1791
1792         ret = btrfs_parse_options(tree_root, options);
1793         if (ret) {
1794                 err = ret;
1795                 goto fail_alloc;
1796         }
1797
1798         features = btrfs_super_incompat_flags(disk_super) &
1799                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1800         if (features) {
1801                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1802                        "unsupported optional features (%Lx).\n",
1803                        (unsigned long long)features);
1804                 err = -EINVAL;
1805                 goto fail_alloc;
1806         }
1807
1808         features = btrfs_super_incompat_flags(disk_super);
1809         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1810         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1811                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1812         btrfs_set_super_incompat_flags(disk_super, features);
1813
1814         features = btrfs_super_compat_ro_flags(disk_super) &
1815                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1816         if (!(sb->s_flags & MS_RDONLY) && features) {
1817                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1818                        "unsupported option features (%Lx).\n",
1819                        (unsigned long long)features);
1820                 err = -EINVAL;
1821                 goto fail_alloc;
1822         }
1823
1824         btrfs_init_workers(&fs_info->generic_worker,
1825                            "genwork", 1, NULL);
1826
1827         btrfs_init_workers(&fs_info->workers, "worker",
1828                            fs_info->thread_pool_size,
1829                            &fs_info->generic_worker);
1830
1831         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1832                            fs_info->thread_pool_size,
1833                            &fs_info->generic_worker);
1834
1835         btrfs_init_workers(&fs_info->submit_workers, "submit",
1836                            min_t(u64, fs_devices->num_devices,
1837                            fs_info->thread_pool_size),
1838                            &fs_info->generic_worker);
1839
1840         btrfs_init_workers(&fs_info->caching_workers, "cache",
1841                            2, &fs_info->generic_worker);
1842
1843         /* a higher idle thresh on the submit workers makes it much more
1844          * likely that bios will be send down in a sane order to the
1845          * devices
1846          */
1847         fs_info->submit_workers.idle_thresh = 64;
1848
1849         fs_info->workers.idle_thresh = 16;
1850         fs_info->workers.ordered = 1;
1851
1852         fs_info->delalloc_workers.idle_thresh = 2;
1853         fs_info->delalloc_workers.ordered = 1;
1854
1855         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1856                            &fs_info->generic_worker);
1857         btrfs_init_workers(&fs_info->endio_workers, "endio",
1858                            fs_info->thread_pool_size,
1859                            &fs_info->generic_worker);
1860         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1861                            fs_info->thread_pool_size,
1862                            &fs_info->generic_worker);
1863         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1864                            "endio-meta-write", fs_info->thread_pool_size,
1865                            &fs_info->generic_worker);
1866         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1867                            fs_info->thread_pool_size,
1868                            &fs_info->generic_worker);
1869         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1870                            1, &fs_info->generic_worker);
1871         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1872                            fs_info->thread_pool_size,
1873                            &fs_info->generic_worker);
1874
1875         /*
1876          * endios are largely parallel and should have a very
1877          * low idle thresh
1878          */
1879         fs_info->endio_workers.idle_thresh = 4;
1880         fs_info->endio_meta_workers.idle_thresh = 4;
1881
1882         fs_info->endio_write_workers.idle_thresh = 2;
1883         fs_info->endio_meta_write_workers.idle_thresh = 2;
1884
1885         btrfs_start_workers(&fs_info->workers, 1);
1886         btrfs_start_workers(&fs_info->generic_worker, 1);
1887         btrfs_start_workers(&fs_info->submit_workers, 1);
1888         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1889         btrfs_start_workers(&fs_info->fixup_workers, 1);
1890         btrfs_start_workers(&fs_info->endio_workers, 1);
1891         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1892         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1893         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1894         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1895         btrfs_start_workers(&fs_info->delayed_workers, 1);
1896         btrfs_start_workers(&fs_info->caching_workers, 1);
1897
1898         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1899         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1900                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1901
1902         nodesize = btrfs_super_nodesize(disk_super);
1903         leafsize = btrfs_super_leafsize(disk_super);
1904         sectorsize = btrfs_super_sectorsize(disk_super);
1905         stripesize = btrfs_super_stripesize(disk_super);
1906         tree_root->nodesize = nodesize;
1907         tree_root->leafsize = leafsize;
1908         tree_root->sectorsize = sectorsize;
1909         tree_root->stripesize = stripesize;
1910
1911         sb->s_blocksize = sectorsize;
1912         sb->s_blocksize_bits = blksize_bits(sectorsize);
1913
1914         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1915                     sizeof(disk_super->magic))) {
1916                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1917                 goto fail_sb_buffer;
1918         }
1919
1920         mutex_lock(&fs_info->chunk_mutex);
1921         ret = btrfs_read_sys_array(tree_root);
1922         mutex_unlock(&fs_info->chunk_mutex);
1923         if (ret) {
1924                 printk(KERN_WARNING "btrfs: failed to read the system "
1925                        "array on %s\n", sb->s_id);
1926                 goto fail_sb_buffer;
1927         }
1928
1929         blocksize = btrfs_level_size(tree_root,
1930                                      btrfs_super_chunk_root_level(disk_super));
1931         generation = btrfs_super_chunk_root_generation(disk_super);
1932
1933         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1934                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1935
1936         chunk_root->node = read_tree_block(chunk_root,
1937                                            btrfs_super_chunk_root(disk_super),
1938                                            blocksize, generation);
1939         BUG_ON(!chunk_root->node);
1940         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1941                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1942                        sb->s_id);
1943                 goto fail_chunk_root;
1944         }
1945         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1946         chunk_root->commit_root = btrfs_root_node(chunk_root);
1947
1948         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1949            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1950            BTRFS_UUID_SIZE);
1951
1952         mutex_lock(&fs_info->chunk_mutex);
1953         ret = btrfs_read_chunk_tree(chunk_root);
1954         mutex_unlock(&fs_info->chunk_mutex);
1955         if (ret) {
1956                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1957                        sb->s_id);
1958                 goto fail_chunk_root;
1959         }
1960
1961         btrfs_close_extra_devices(fs_devices);
1962
1963         blocksize = btrfs_level_size(tree_root,
1964                                      btrfs_super_root_level(disk_super));
1965         generation = btrfs_super_generation(disk_super);
1966
1967         tree_root->node = read_tree_block(tree_root,
1968                                           btrfs_super_root(disk_super),
1969                                           blocksize, generation);
1970         if (!tree_root->node)
1971                 goto fail_chunk_root;
1972         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1973                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1974                        sb->s_id);
1975                 goto fail_tree_root;
1976         }
1977         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1978         tree_root->commit_root = btrfs_root_node(tree_root);
1979
1980         ret = find_and_setup_root(tree_root, fs_info,
1981                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1982         if (ret)
1983                 goto fail_tree_root;
1984         extent_root->track_dirty = 1;
1985
1986         ret = find_and_setup_root(tree_root, fs_info,
1987                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1988         if (ret)
1989                 goto fail_extent_root;
1990         dev_root->track_dirty = 1;
1991
1992         ret = find_and_setup_root(tree_root, fs_info,
1993                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1994         if (ret)
1995                 goto fail_dev_root;
1996
1997         csum_root->track_dirty = 1;
1998
1999         fs_info->generation = generation;
2000         fs_info->last_trans_committed = generation;
2001         fs_info->data_alloc_profile = (u64)-1;
2002         fs_info->metadata_alloc_profile = (u64)-1;
2003         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2004
2005         ret = btrfs_init_space_info(fs_info);
2006         if (ret) {
2007                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2008                 goto fail_block_groups;
2009         }
2010
2011         ret = btrfs_read_block_groups(extent_root);
2012         if (ret) {
2013                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2014                 goto fail_block_groups;
2015         }
2016
2017         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2018                                                "btrfs-cleaner");
2019         if (IS_ERR(fs_info->cleaner_kthread))
2020                 goto fail_block_groups;
2021
2022         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2023                                                    tree_root,
2024                                                    "btrfs-transaction");
2025         if (IS_ERR(fs_info->transaction_kthread))
2026                 goto fail_cleaner;
2027
2028         if (!btrfs_test_opt(tree_root, SSD) &&
2029             !btrfs_test_opt(tree_root, NOSSD) &&
2030             !fs_info->fs_devices->rotating) {
2031                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2032                        "mode\n");
2033                 btrfs_set_opt(fs_info->mount_opt, SSD);
2034         }
2035
2036         /* do not make disk changes in broken FS */
2037         if (btrfs_super_log_root(disk_super) != 0 &&
2038             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2039                 u64 bytenr = btrfs_super_log_root(disk_super);
2040
2041                 if (fs_devices->rw_devices == 0) {
2042                         printk(KERN_WARNING "Btrfs log replay required "
2043                                "on RO media\n");
2044                         err = -EIO;
2045                         goto fail_trans_kthread;
2046                 }
2047                 blocksize =
2048                      btrfs_level_size(tree_root,
2049                                       btrfs_super_log_root_level(disk_super));
2050
2051                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2052                 if (!log_tree_root) {
2053                         err = -ENOMEM;
2054                         goto fail_trans_kthread;
2055                 }
2056
2057                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2058                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2059
2060                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2061                                                       blocksize,
2062                                                       generation + 1);
2063                 ret = btrfs_recover_log_trees(log_tree_root);
2064                 BUG_ON(ret);
2065
2066                 if (sb->s_flags & MS_RDONLY) {
2067                         ret =  btrfs_commit_super(tree_root);
2068                         BUG_ON(ret);
2069                 }
2070         }
2071
2072         ret = btrfs_find_orphan_roots(tree_root);
2073         BUG_ON(ret);
2074
2075         if (!(sb->s_flags & MS_RDONLY)) {
2076                 ret = btrfs_cleanup_fs_roots(fs_info);
2077                 BUG_ON(ret);
2078
2079                 ret = btrfs_recover_relocation(tree_root);
2080                 if (ret < 0) {
2081                         printk(KERN_WARNING
2082                                "btrfs: failed to recover relocation\n");
2083                         err = -EINVAL;
2084                         goto fail_trans_kthread;
2085                 }
2086         }
2087
2088         location.objectid = BTRFS_FS_TREE_OBJECTID;
2089         location.type = BTRFS_ROOT_ITEM_KEY;
2090         location.offset = (u64)-1;
2091
2092         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2093         if (!fs_info->fs_root)
2094                 goto fail_trans_kthread;
2095         if (IS_ERR(fs_info->fs_root)) {
2096                 err = PTR_ERR(fs_info->fs_root);
2097                 goto fail_trans_kthread;
2098         }
2099
2100         if (!(sb->s_flags & MS_RDONLY)) {
2101                 down_read(&fs_info->cleanup_work_sem);
2102                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2103                 if (!err)
2104                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2105                 up_read(&fs_info->cleanup_work_sem);
2106                 if (err) {
2107                         close_ctree(tree_root);
2108                         return ERR_PTR(err);
2109                 }
2110         }
2111
2112         return tree_root;
2113
2114 fail_trans_kthread:
2115         kthread_stop(fs_info->transaction_kthread);
2116 fail_cleaner:
2117         kthread_stop(fs_info->cleaner_kthread);
2118
2119         /*
2120          * make sure we're done with the btree inode before we stop our
2121          * kthreads
2122          */
2123         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2124         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2125
2126 fail_block_groups:
2127         btrfs_free_block_groups(fs_info);
2128         free_extent_buffer(csum_root->node);
2129         free_extent_buffer(csum_root->commit_root);
2130 fail_dev_root:
2131         free_extent_buffer(dev_root->node);
2132         free_extent_buffer(dev_root->commit_root);
2133 fail_extent_root:
2134         free_extent_buffer(extent_root->node);
2135         free_extent_buffer(extent_root->commit_root);
2136 fail_tree_root:
2137         free_extent_buffer(tree_root->node);
2138         free_extent_buffer(tree_root->commit_root);
2139 fail_chunk_root:
2140         free_extent_buffer(chunk_root->node);
2141         free_extent_buffer(chunk_root->commit_root);
2142 fail_sb_buffer:
2143         btrfs_stop_workers(&fs_info->generic_worker);
2144         btrfs_stop_workers(&fs_info->fixup_workers);
2145         btrfs_stop_workers(&fs_info->delalloc_workers);
2146         btrfs_stop_workers(&fs_info->workers);
2147         btrfs_stop_workers(&fs_info->endio_workers);
2148         btrfs_stop_workers(&fs_info->endio_meta_workers);
2149         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2150         btrfs_stop_workers(&fs_info->endio_write_workers);
2151         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2152         btrfs_stop_workers(&fs_info->submit_workers);
2153         btrfs_stop_workers(&fs_info->delayed_workers);
2154         btrfs_stop_workers(&fs_info->caching_workers);
2155 fail_alloc:
2156         kfree(fs_info->delayed_root);
2157 fail_iput:
2158         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2159         iput(fs_info->btree_inode);
2160
2161         btrfs_close_devices(fs_info->fs_devices);
2162         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2163 fail_bdi:
2164         bdi_destroy(&fs_info->bdi);
2165 fail_srcu:
2166         cleanup_srcu_struct(&fs_info->subvol_srcu);
2167 fail:
2168         kfree(extent_root);
2169         kfree(tree_root);
2170         kfree(fs_info);
2171         kfree(chunk_root);
2172         kfree(dev_root);
2173         kfree(csum_root);
2174         return ERR_PTR(err);
2175 }
2176
2177 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2178 {
2179         char b[BDEVNAME_SIZE];
2180
2181         if (uptodate) {
2182                 set_buffer_uptodate(bh);
2183         } else {
2184                 printk_ratelimited(KERN_WARNING "lost page write due to "
2185                                         "I/O error on %s\n",
2186                                        bdevname(bh->b_bdev, b));
2187                 /* note, we dont' set_buffer_write_io_error because we have
2188                  * our own ways of dealing with the IO errors
2189                  */
2190                 clear_buffer_uptodate(bh);
2191         }
2192         unlock_buffer(bh);
2193         put_bh(bh);
2194 }
2195
2196 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2197 {
2198         struct buffer_head *bh;
2199         struct buffer_head *latest = NULL;
2200         struct btrfs_super_block *super;
2201         int i;
2202         u64 transid = 0;
2203         u64 bytenr;
2204
2205         /* we would like to check all the supers, but that would make
2206          * a btrfs mount succeed after a mkfs from a different FS.
2207          * So, we need to add a special mount option to scan for
2208          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2209          */
2210         for (i = 0; i < 1; i++) {
2211                 bytenr = btrfs_sb_offset(i);
2212                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2213                         break;
2214                 bh = __bread(bdev, bytenr / 4096, 4096);
2215                 if (!bh)
2216                         continue;
2217
2218                 super = (struct btrfs_super_block *)bh->b_data;
2219                 if (btrfs_super_bytenr(super) != bytenr ||
2220                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2221                             sizeof(super->magic))) {
2222                         brelse(bh);
2223                         continue;
2224                 }
2225
2226                 if (!latest || btrfs_super_generation(super) > transid) {
2227                         brelse(latest);
2228                         latest = bh;
2229                         transid = btrfs_super_generation(super);
2230                 } else {
2231                         brelse(bh);
2232                 }
2233         }
2234         return latest;
2235 }
2236
2237 /*
2238  * this should be called twice, once with wait == 0 and
2239  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2240  * we write are pinned.
2241  *
2242  * They are released when wait == 1 is done.
2243  * max_mirrors must be the same for both runs, and it indicates how
2244  * many supers on this one device should be written.
2245  *
2246  * max_mirrors == 0 means to write them all.
2247  */
2248 static int write_dev_supers(struct btrfs_device *device,
2249                             struct btrfs_super_block *sb,
2250                             int do_barriers, int wait, int max_mirrors)
2251 {
2252         struct buffer_head *bh;
2253         int i;
2254         int ret;
2255         int errors = 0;
2256         u32 crc;
2257         u64 bytenr;
2258         int last_barrier = 0;
2259
2260         if (max_mirrors == 0)
2261                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2262
2263         /* make sure only the last submit_bh does a barrier */
2264         if (do_barriers) {
2265                 for (i = 0; i < max_mirrors; i++) {
2266                         bytenr = btrfs_sb_offset(i);
2267                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2268                             device->total_bytes)
2269                                 break;
2270                         last_barrier = i;
2271                 }
2272         }
2273
2274         for (i = 0; i < max_mirrors; i++) {
2275                 bytenr = btrfs_sb_offset(i);
2276                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2277                         break;
2278
2279                 if (wait) {
2280                         bh = __find_get_block(device->bdev, bytenr / 4096,
2281                                               BTRFS_SUPER_INFO_SIZE);
2282                         BUG_ON(!bh);
2283                         wait_on_buffer(bh);
2284                         if (!buffer_uptodate(bh))
2285                                 errors++;
2286
2287                         /* drop our reference */
2288                         brelse(bh);
2289
2290                         /* drop the reference from the wait == 0 run */
2291                         brelse(bh);
2292                         continue;
2293                 } else {
2294                         btrfs_set_super_bytenr(sb, bytenr);
2295
2296                         crc = ~(u32)0;
2297                         crc = btrfs_csum_data(NULL, (char *)sb +
2298                                               BTRFS_CSUM_SIZE, crc,
2299                                               BTRFS_SUPER_INFO_SIZE -
2300                                               BTRFS_CSUM_SIZE);
2301                         btrfs_csum_final(crc, sb->csum);
2302
2303                         /*
2304                          * one reference for us, and we leave it for the
2305                          * caller
2306                          */
2307                         bh = __getblk(device->bdev, bytenr / 4096,
2308                                       BTRFS_SUPER_INFO_SIZE);
2309                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2310
2311                         /* one reference for submit_bh */
2312                         get_bh(bh);
2313
2314                         set_buffer_uptodate(bh);
2315                         lock_buffer(bh);
2316                         bh->b_end_io = btrfs_end_buffer_write_sync;
2317                 }
2318
2319                 if (i == last_barrier && do_barriers)
2320                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2321                 else
2322                         ret = submit_bh(WRITE_SYNC, bh);
2323
2324                 if (ret)
2325                         errors++;
2326         }
2327         return errors < i ? 0 : -1;
2328 }
2329
2330 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2331 {
2332         struct list_head *head;
2333         struct btrfs_device *dev;
2334         struct btrfs_super_block *sb;
2335         struct btrfs_dev_item *dev_item;
2336         int ret;
2337         int do_barriers;
2338         int max_errors;
2339         int total_errors = 0;
2340         u64 flags;
2341
2342         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2343         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2344
2345         sb = &root->fs_info->super_for_commit;
2346         dev_item = &sb->dev_item;
2347
2348         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2349         head = &root->fs_info->fs_devices->devices;
2350         list_for_each_entry_rcu(dev, head, dev_list) {
2351                 if (!dev->bdev) {
2352                         total_errors++;
2353                         continue;
2354                 }
2355                 if (!dev->in_fs_metadata || !dev->writeable)
2356                         continue;
2357
2358                 btrfs_set_stack_device_generation(dev_item, 0);
2359                 btrfs_set_stack_device_type(dev_item, dev->type);
2360                 btrfs_set_stack_device_id(dev_item, dev->devid);
2361                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2362                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2363                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2364                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2365                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2366                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2367                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2368
2369                 flags = btrfs_super_flags(sb);
2370                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2371
2372                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2373                 if (ret)
2374                         total_errors++;
2375         }
2376         if (total_errors > max_errors) {
2377                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2378                        total_errors);
2379                 BUG();
2380         }
2381
2382         total_errors = 0;
2383         list_for_each_entry_rcu(dev, head, dev_list) {
2384                 if (!dev->bdev)
2385                         continue;
2386                 if (!dev->in_fs_metadata || !dev->writeable)
2387                         continue;
2388
2389                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2390                 if (ret)
2391                         total_errors++;
2392         }
2393         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2394         if (total_errors > max_errors) {
2395                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2396                        total_errors);
2397                 BUG();
2398         }
2399         return 0;
2400 }
2401
2402 int write_ctree_super(struct btrfs_trans_handle *trans,
2403                       struct btrfs_root *root, int max_mirrors)
2404 {
2405         int ret;
2406
2407         ret = write_all_supers(root, max_mirrors);
2408         return ret;
2409 }
2410
2411 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2412 {
2413         spin_lock(&fs_info->fs_roots_radix_lock);
2414         radix_tree_delete(&fs_info->fs_roots_radix,
2415                           (unsigned long)root->root_key.objectid);
2416         spin_unlock(&fs_info->fs_roots_radix_lock);
2417
2418         if (btrfs_root_refs(&root->root_item) == 0)
2419                 synchronize_srcu(&fs_info->subvol_srcu);
2420
2421         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2422         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2423         free_fs_root(root);
2424         return 0;
2425 }
2426
2427 static void free_fs_root(struct btrfs_root *root)
2428 {
2429         iput(root->cache_inode);
2430         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2431         if (root->anon_dev)
2432                 free_anon_bdev(root->anon_dev);
2433         free_extent_buffer(root->node);
2434         free_extent_buffer(root->commit_root);
2435         kfree(root->free_ino_ctl);
2436         kfree(root->free_ino_pinned);
2437         kfree(root->name);
2438         kfree(root);
2439 }
2440
2441 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2442 {
2443         int ret;
2444         struct btrfs_root *gang[8];
2445         int i;
2446
2447         while (!list_empty(&fs_info->dead_roots)) {
2448                 gang[0] = list_entry(fs_info->dead_roots.next,
2449                                      struct btrfs_root, root_list);
2450                 list_del(&gang[0]->root_list);
2451
2452                 if (gang[0]->in_radix) {
2453                         btrfs_free_fs_root(fs_info, gang[0]);
2454                 } else {
2455                         free_extent_buffer(gang[0]->node);
2456                         free_extent_buffer(gang[0]->commit_root);
2457                         kfree(gang[0]);
2458                 }
2459         }
2460
2461         while (1) {
2462                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2463                                              (void **)gang, 0,
2464                                              ARRAY_SIZE(gang));
2465                 if (!ret)
2466                         break;
2467                 for (i = 0; i < ret; i++)
2468                         btrfs_free_fs_root(fs_info, gang[i]);
2469         }
2470         return 0;
2471 }
2472
2473 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2474 {
2475         u64 root_objectid = 0;
2476         struct btrfs_root *gang[8];
2477         int i;
2478         int ret;
2479
2480         while (1) {
2481                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2482                                              (void **)gang, root_objectid,
2483                                              ARRAY_SIZE(gang));
2484                 if (!ret)
2485                         break;
2486
2487                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2488                 for (i = 0; i < ret; i++) {
2489                         int err;
2490
2491                         root_objectid = gang[i]->root_key.objectid;
2492                         err = btrfs_orphan_cleanup(gang[i]);
2493                         if (err)
2494                                 return err;
2495                 }
2496                 root_objectid++;
2497         }
2498         return 0;
2499 }
2500
2501 int btrfs_commit_super(struct btrfs_root *root)
2502 {
2503         struct btrfs_trans_handle *trans;
2504         int ret;
2505
2506         mutex_lock(&root->fs_info->cleaner_mutex);
2507         btrfs_run_delayed_iputs(root);
2508         btrfs_clean_old_snapshots(root);
2509         mutex_unlock(&root->fs_info->cleaner_mutex);
2510
2511         /* wait until ongoing cleanup work done */
2512         down_write(&root->fs_info->cleanup_work_sem);
2513         up_write(&root->fs_info->cleanup_work_sem);
2514
2515         trans = btrfs_join_transaction(root);
2516         if (IS_ERR(trans))
2517                 return PTR_ERR(trans);
2518         ret = btrfs_commit_transaction(trans, root);
2519         BUG_ON(ret);
2520         /* run commit again to drop the original snapshot */
2521         trans = btrfs_join_transaction(root);
2522         if (IS_ERR(trans))
2523                 return PTR_ERR(trans);
2524         btrfs_commit_transaction(trans, root);
2525         ret = btrfs_write_and_wait_transaction(NULL, root);
2526         BUG_ON(ret);
2527
2528         ret = write_ctree_super(NULL, root, 0);
2529         return ret;
2530 }
2531
2532 int close_ctree(struct btrfs_root *root)
2533 {
2534         struct btrfs_fs_info *fs_info = root->fs_info;
2535         int ret;
2536
2537         fs_info->closing = 1;
2538         smp_mb();
2539
2540         btrfs_scrub_cancel(root);
2541
2542         /* wait for any defraggers to finish */
2543         wait_event(fs_info->transaction_wait,
2544                    (atomic_read(&fs_info->defrag_running) == 0));
2545
2546         /* clear out the rbtree of defraggable inodes */
2547         btrfs_run_defrag_inodes(root->fs_info);
2548
2549         btrfs_put_block_group_cache(fs_info);
2550
2551         /*
2552          * Here come 2 situations when btrfs is broken to flip readonly:
2553          *
2554          * 1. when btrfs flips readonly somewhere else before
2555          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2556          * and btrfs will skip to write sb directly to keep
2557          * ERROR state on disk.
2558          *
2559          * 2. when btrfs flips readonly just in btrfs_commit_super,
2560          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2561          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2562          * btrfs will cleanup all FS resources first and write sb then.
2563          */
2564         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2565                 ret = btrfs_commit_super(root);
2566                 if (ret)
2567                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2568         }
2569
2570         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2571                 ret = btrfs_error_commit_super(root);
2572                 if (ret)
2573                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2574         }
2575
2576         kthread_stop(root->fs_info->transaction_kthread);
2577         kthread_stop(root->fs_info->cleaner_kthread);
2578
2579         fs_info->closing = 2;
2580         smp_mb();
2581
2582         if (fs_info->delalloc_bytes) {
2583                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2584                        (unsigned long long)fs_info->delalloc_bytes);
2585         }
2586         if (fs_info->total_ref_cache_size) {
2587                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2588                        (unsigned long long)fs_info->total_ref_cache_size);
2589         }
2590
2591         free_extent_buffer(fs_info->extent_root->node);
2592         free_extent_buffer(fs_info->extent_root->commit_root);
2593         free_extent_buffer(fs_info->tree_root->node);
2594         free_extent_buffer(fs_info->tree_root->commit_root);
2595         free_extent_buffer(root->fs_info->chunk_root->node);
2596         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2597         free_extent_buffer(root->fs_info->dev_root->node);
2598         free_extent_buffer(root->fs_info->dev_root->commit_root);
2599         free_extent_buffer(root->fs_info->csum_root->node);
2600         free_extent_buffer(root->fs_info->csum_root->commit_root);
2601
2602         btrfs_free_block_groups(root->fs_info);
2603
2604         del_fs_roots(fs_info);
2605
2606         iput(fs_info->btree_inode);
2607         kfree(fs_info->delayed_root);
2608
2609         btrfs_stop_workers(&fs_info->generic_worker);
2610         btrfs_stop_workers(&fs_info->fixup_workers);
2611         btrfs_stop_workers(&fs_info->delalloc_workers);
2612         btrfs_stop_workers(&fs_info->workers);
2613         btrfs_stop_workers(&fs_info->endio_workers);
2614         btrfs_stop_workers(&fs_info->endio_meta_workers);
2615         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2616         btrfs_stop_workers(&fs_info->endio_write_workers);
2617         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2618         btrfs_stop_workers(&fs_info->submit_workers);
2619         btrfs_stop_workers(&fs_info->delayed_workers);
2620         btrfs_stop_workers(&fs_info->caching_workers);
2621
2622         btrfs_close_devices(fs_info->fs_devices);
2623         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2624
2625         bdi_destroy(&fs_info->bdi);
2626         cleanup_srcu_struct(&fs_info->subvol_srcu);
2627
2628         kfree(fs_info->extent_root);
2629         kfree(fs_info->tree_root);
2630         kfree(fs_info->chunk_root);
2631         kfree(fs_info->dev_root);
2632         kfree(fs_info->csum_root);
2633         kfree(fs_info);
2634
2635         return 0;
2636 }
2637
2638 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2639 {
2640         int ret;
2641         struct inode *btree_inode = buf->first_page->mapping->host;
2642
2643         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2644                                      NULL);
2645         if (!ret)
2646                 return ret;
2647
2648         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2649                                     parent_transid);
2650         return !ret;
2651 }
2652
2653 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2654 {
2655         struct inode *btree_inode = buf->first_page->mapping->host;
2656         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2657                                           buf);
2658 }
2659
2660 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2661 {
2662         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2663         u64 transid = btrfs_header_generation(buf);
2664         struct inode *btree_inode = root->fs_info->btree_inode;
2665         int was_dirty;
2666
2667         btrfs_assert_tree_locked(buf);
2668         if (transid != root->fs_info->generation) {
2669                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2670                        "found %llu running %llu\n",
2671                         (unsigned long long)buf->start,
2672                         (unsigned long long)transid,
2673                         (unsigned long long)root->fs_info->generation);
2674                 WARN_ON(1);
2675         }
2676         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2677                                             buf);
2678         if (!was_dirty) {
2679                 spin_lock(&root->fs_info->delalloc_lock);
2680                 root->fs_info->dirty_metadata_bytes += buf->len;
2681                 spin_unlock(&root->fs_info->delalloc_lock);
2682         }
2683 }
2684
2685 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2686 {
2687         /*
2688          * looks as though older kernels can get into trouble with
2689          * this code, they end up stuck in balance_dirty_pages forever
2690          */
2691         u64 num_dirty;
2692         unsigned long thresh = 32 * 1024 * 1024;
2693
2694         if (current->flags & PF_MEMALLOC)
2695                 return;
2696
2697         btrfs_balance_delayed_items(root);
2698
2699         num_dirty = root->fs_info->dirty_metadata_bytes;
2700
2701         if (num_dirty > thresh) {
2702                 balance_dirty_pages_ratelimited_nr(
2703                                    root->fs_info->btree_inode->i_mapping, 1);
2704         }
2705         return;
2706 }
2707
2708 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2709 {
2710         /*
2711          * looks as though older kernels can get into trouble with
2712          * this code, they end up stuck in balance_dirty_pages forever
2713          */
2714         u64 num_dirty;
2715         unsigned long thresh = 32 * 1024 * 1024;
2716
2717         if (current->flags & PF_MEMALLOC)
2718                 return;
2719
2720         num_dirty = root->fs_info->dirty_metadata_bytes;
2721
2722         if (num_dirty > thresh) {
2723                 balance_dirty_pages_ratelimited_nr(
2724                                    root->fs_info->btree_inode->i_mapping, 1);
2725         }
2726         return;
2727 }
2728
2729 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2730 {
2731         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2732         int ret;
2733         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2734         if (ret == 0)
2735                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2736         return ret;
2737 }
2738
2739 int btree_lock_page_hook(struct page *page)
2740 {
2741         struct inode *inode = page->mapping->host;
2742         struct btrfs_root *root = BTRFS_I(inode)->root;
2743         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2744         struct extent_buffer *eb;
2745         unsigned long len;
2746         u64 bytenr = page_offset(page);
2747
2748         if (page->private == EXTENT_PAGE_PRIVATE)
2749                 goto out;
2750
2751         len = page->private >> 2;
2752         eb = find_extent_buffer(io_tree, bytenr, len);
2753         if (!eb)
2754                 goto out;
2755
2756         btrfs_tree_lock(eb);
2757         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2758
2759         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2760                 spin_lock(&root->fs_info->delalloc_lock);
2761                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2762                         root->fs_info->dirty_metadata_bytes -= eb->len;
2763                 else
2764                         WARN_ON(1);
2765                 spin_unlock(&root->fs_info->delalloc_lock);
2766         }
2767
2768         btrfs_tree_unlock(eb);
2769         free_extent_buffer(eb);
2770 out:
2771         lock_page(page);
2772         return 0;
2773 }
2774
2775 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2776                               int read_only)
2777 {
2778         if (read_only)
2779                 return;
2780
2781         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2782                 printk(KERN_WARNING "warning: mount fs with errors, "
2783                        "running btrfsck is recommended\n");
2784 }
2785
2786 int btrfs_error_commit_super(struct btrfs_root *root)
2787 {
2788         int ret;
2789
2790         mutex_lock(&root->fs_info->cleaner_mutex);
2791         btrfs_run_delayed_iputs(root);
2792         mutex_unlock(&root->fs_info->cleaner_mutex);
2793
2794         down_write(&root->fs_info->cleanup_work_sem);
2795         up_write(&root->fs_info->cleanup_work_sem);
2796
2797         /* cleanup FS via transaction */
2798         btrfs_cleanup_transaction(root);
2799
2800         ret = write_ctree_super(NULL, root, 0);
2801
2802         return ret;
2803 }
2804
2805 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2806 {
2807         struct btrfs_inode *btrfs_inode;
2808         struct list_head splice;
2809
2810         INIT_LIST_HEAD(&splice);
2811
2812         mutex_lock(&root->fs_info->ordered_operations_mutex);
2813         spin_lock(&root->fs_info->ordered_extent_lock);
2814
2815         list_splice_init(&root->fs_info->ordered_operations, &splice);
2816         while (!list_empty(&splice)) {
2817                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2818                                          ordered_operations);
2819
2820                 list_del_init(&btrfs_inode->ordered_operations);
2821
2822                 btrfs_invalidate_inodes(btrfs_inode->root);
2823         }
2824
2825         spin_unlock(&root->fs_info->ordered_extent_lock);
2826         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2827
2828         return 0;
2829 }
2830
2831 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2832 {
2833         struct list_head splice;
2834         struct btrfs_ordered_extent *ordered;
2835         struct inode *inode;
2836
2837         INIT_LIST_HEAD(&splice);
2838
2839         spin_lock(&root->fs_info->ordered_extent_lock);
2840
2841         list_splice_init(&root->fs_info->ordered_extents, &splice);
2842         while (!list_empty(&splice)) {
2843                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2844                                      root_extent_list);
2845
2846                 list_del_init(&ordered->root_extent_list);
2847                 atomic_inc(&ordered->refs);
2848
2849                 /* the inode may be getting freed (in sys_unlink path). */
2850                 inode = igrab(ordered->inode);
2851
2852                 spin_unlock(&root->fs_info->ordered_extent_lock);
2853                 if (inode)
2854                         iput(inode);
2855
2856                 atomic_set(&ordered->refs, 1);
2857                 btrfs_put_ordered_extent(ordered);
2858
2859                 spin_lock(&root->fs_info->ordered_extent_lock);
2860         }
2861
2862         spin_unlock(&root->fs_info->ordered_extent_lock);
2863
2864         return 0;
2865 }
2866
2867 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2868                                       struct btrfs_root *root)
2869 {
2870         struct rb_node *node;
2871         struct btrfs_delayed_ref_root *delayed_refs;
2872         struct btrfs_delayed_ref_node *ref;
2873         int ret = 0;
2874
2875         delayed_refs = &trans->delayed_refs;
2876
2877         spin_lock(&delayed_refs->lock);
2878         if (delayed_refs->num_entries == 0) {
2879                 spin_unlock(&delayed_refs->lock);
2880                 printk(KERN_INFO "delayed_refs has NO entry\n");
2881                 return ret;
2882         }
2883
2884         node = rb_first(&delayed_refs->root);
2885         while (node) {
2886                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2887                 node = rb_next(node);
2888
2889                 ref->in_tree = 0;
2890                 rb_erase(&ref->rb_node, &delayed_refs->root);
2891                 delayed_refs->num_entries--;
2892
2893                 atomic_set(&ref->refs, 1);
2894                 if (btrfs_delayed_ref_is_head(ref)) {
2895                         struct btrfs_delayed_ref_head *head;
2896
2897                         head = btrfs_delayed_node_to_head(ref);
2898                         mutex_lock(&head->mutex);
2899                         kfree(head->extent_op);
2900                         delayed_refs->num_heads--;
2901                         if (list_empty(&head->cluster))
2902                                 delayed_refs->num_heads_ready--;
2903                         list_del_init(&head->cluster);
2904                         mutex_unlock(&head->mutex);
2905                 }
2906
2907                 spin_unlock(&delayed_refs->lock);
2908                 btrfs_put_delayed_ref(ref);
2909
2910                 cond_resched();
2911                 spin_lock(&delayed_refs->lock);
2912         }
2913
2914         spin_unlock(&delayed_refs->lock);
2915
2916         return ret;
2917 }
2918
2919 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2920 {
2921         struct btrfs_pending_snapshot *snapshot;
2922         struct list_head splice;
2923
2924         INIT_LIST_HEAD(&splice);
2925
2926         list_splice_init(&t->pending_snapshots, &splice);
2927
2928         while (!list_empty(&splice)) {
2929                 snapshot = list_entry(splice.next,
2930                                       struct btrfs_pending_snapshot,
2931                                       list);
2932
2933                 list_del_init(&snapshot->list);
2934
2935                 kfree(snapshot);
2936         }
2937
2938         return 0;
2939 }
2940
2941 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2942 {
2943         struct btrfs_inode *btrfs_inode;
2944         struct list_head splice;
2945
2946         INIT_LIST_HEAD(&splice);
2947
2948         spin_lock(&root->fs_info->delalloc_lock);
2949         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2950
2951         while (!list_empty(&splice)) {
2952                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2953                                     delalloc_inodes);
2954
2955                 list_del_init(&btrfs_inode->delalloc_inodes);
2956
2957                 btrfs_invalidate_inodes(btrfs_inode->root);
2958         }
2959
2960         spin_unlock(&root->fs_info->delalloc_lock);
2961
2962         return 0;
2963 }
2964
2965 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2966                                         struct extent_io_tree *dirty_pages,
2967                                         int mark)
2968 {
2969         int ret;
2970         struct page *page;
2971         struct inode *btree_inode = root->fs_info->btree_inode;
2972         struct extent_buffer *eb;
2973         u64 start = 0;
2974         u64 end;
2975         u64 offset;
2976         unsigned long index;
2977
2978         while (1) {
2979                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2980                                             mark);
2981                 if (ret)
2982                         break;
2983
2984                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2985                 while (start <= end) {
2986                         index = start >> PAGE_CACHE_SHIFT;
2987                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2988                         page = find_get_page(btree_inode->i_mapping, index);
2989                         if (!page)
2990                                 continue;
2991                         offset = page_offset(page);
2992
2993                         spin_lock(&dirty_pages->buffer_lock);
2994                         eb = radix_tree_lookup(
2995                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2996                                                offset >> PAGE_CACHE_SHIFT);
2997                         spin_unlock(&dirty_pages->buffer_lock);
2998                         if (eb) {
2999                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3000                                                          &eb->bflags);
3001                                 atomic_set(&eb->refs, 1);
3002                         }
3003                         if (PageWriteback(page))
3004                                 end_page_writeback(page);
3005
3006                         lock_page(page);
3007                         if (PageDirty(page)) {
3008                                 clear_page_dirty_for_io(page);
3009                                 spin_lock_irq(&page->mapping->tree_lock);
3010                                 radix_tree_tag_clear(&page->mapping->page_tree,
3011                                                         page_index(page),
3012                                                         PAGECACHE_TAG_DIRTY);
3013                                 spin_unlock_irq(&page->mapping->tree_lock);
3014                         }
3015
3016                         page->mapping->a_ops->invalidatepage(page, 0);
3017                         unlock_page(page);
3018                 }
3019         }
3020
3021         return ret;
3022 }
3023
3024 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3025                                        struct extent_io_tree *pinned_extents)
3026 {
3027         struct extent_io_tree *unpin;
3028         u64 start;
3029         u64 end;
3030         int ret;
3031
3032         unpin = pinned_extents;
3033         while (1) {
3034                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3035                                             EXTENT_DIRTY);
3036                 if (ret)
3037                         break;
3038
3039                 /* opt_discard */
3040                 if (btrfs_test_opt(root, DISCARD))
3041                         ret = btrfs_error_discard_extent(root, start,
3042                                                          end + 1 - start,
3043                                                          NULL);
3044
3045                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3046                 btrfs_error_unpin_extent_range(root, start, end);
3047                 cond_resched();
3048         }
3049
3050         return 0;
3051 }
3052
3053 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3054 {
3055         struct btrfs_transaction *t;
3056         LIST_HEAD(list);
3057
3058         WARN_ON(1);
3059
3060         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3061
3062         spin_lock(&root->fs_info->trans_lock);
3063         list_splice_init(&root->fs_info->trans_list, &list);
3064         root->fs_info->trans_no_join = 1;
3065         spin_unlock(&root->fs_info->trans_lock);
3066
3067         while (!list_empty(&list)) {
3068                 t = list_entry(list.next, struct btrfs_transaction, list);
3069                 if (!t)
3070                         break;
3071
3072                 btrfs_destroy_ordered_operations(root);
3073
3074                 btrfs_destroy_ordered_extents(root);
3075
3076                 btrfs_destroy_delayed_refs(t, root);
3077
3078                 btrfs_block_rsv_release(root,
3079                                         &root->fs_info->trans_block_rsv,
3080                                         t->dirty_pages.dirty_bytes);
3081
3082                 /* FIXME: cleanup wait for commit */
3083                 t->in_commit = 1;
3084                 t->blocked = 1;
3085                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3086                         wake_up(&root->fs_info->transaction_blocked_wait);
3087
3088                 t->blocked = 0;
3089                 if (waitqueue_active(&root->fs_info->transaction_wait))
3090                         wake_up(&root->fs_info->transaction_wait);
3091
3092                 t->commit_done = 1;
3093                 if (waitqueue_active(&t->commit_wait))
3094                         wake_up(&t->commit_wait);
3095
3096                 btrfs_destroy_pending_snapshots(t);
3097
3098                 btrfs_destroy_delalloc_inodes(root);
3099
3100                 spin_lock(&root->fs_info->trans_lock);
3101                 root->fs_info->running_transaction = NULL;
3102                 spin_unlock(&root->fs_info->trans_lock);
3103
3104                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3105                                              EXTENT_DIRTY);
3106
3107                 btrfs_destroy_pinned_extent(root,
3108                                             root->fs_info->pinned_extents);
3109
3110                 atomic_set(&t->use_count, 0);
3111                 list_del_init(&t->list);
3112                 memset(t, 0, sizeof(*t));
3113                 kmem_cache_free(btrfs_transaction_cachep, t);
3114         }
3115
3116         spin_lock(&root->fs_info->trans_lock);
3117         root->fs_info->trans_no_join = 0;
3118         spin_unlock(&root->fs_info->trans_lock);
3119         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3120
3121         return 0;
3122 }
3123
3124 static struct extent_io_ops btree_extent_io_ops = {
3125         .write_cache_pages_lock_hook = btree_lock_page_hook,
3126         .readpage_end_io_hook = btree_readpage_end_io_hook,
3127         .submit_bio_hook = btree_submit_bio_hook,
3128         /* note we're sharing with inode.c for the merge bio hook */
3129         .merge_bio_hook = btrfs_merge_bio_hook,
3130 };