fs/btrfs/extent-tree.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 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include "hash.h"
  22 #include "crc32c.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27 #include "volumes.h"
  28 #include "locking.h"
  29
  30 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  32 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  33
  34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  35
  36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  37                                  btrfs_root *extent_root);
  38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  39                                btrfs_root *extent_root);
  40 static struct btrfs_block_group_cache *
  41 __btrfs_find_block_group(struct btrfs_root *root,
  42                          struct btrfs_block_group_cache *hint,
  43                          u64 search_start, int data, int owner);
  44
  45 void maybe_lock_mutex(struct btrfs_root *root)
  46 {
  47         if (root != root->fs_info->extent_root &&
  48             root != root->fs_info->chunk_root &&
  49             root != root->fs_info->dev_root) {
  50                 mutex_lock(&root->fs_info->alloc_mutex);
  51         }
  52 }
  53
  54 void maybe_unlock_mutex(struct btrfs_root *root)
  55 {
  56         if (root != root->fs_info->extent_root &&
  57             root != root->fs_info->chunk_root &&
  58             root != root->fs_info->dev_root) {
  59                 mutex_unlock(&root->fs_info->alloc_mutex);
  60         }
  61 }
  62
  63 static int cache_block_group(struct btrfs_root *root,
  64                              struct btrfs_block_group_cache *block_group)
  65 {
  66         struct btrfs_path *path;
  67         int ret;
  68         struct btrfs_key key;
  69         struct extent_buffer *leaf;
  70         struct extent_io_tree *free_space_cache;
  71         int slot;
  72         u64 last = 0;
  73         u64 hole_size;
  74         u64 first_free;
  75         int found = 0;
  76
  77         if (!block_group)
  78                 return 0;
  79
  80         root = root->fs_info->extent_root;
  81         free_space_cache = &root->fs_info->free_space_cache;
  82
  83         if (block_group->cached)
  84                 return 0;
  85
  86         path = btrfs_alloc_path();
  87         if (!path)
  88                 return -ENOMEM;
  89
  90         path->reada = 2;
  91         /*
  92          * we get into deadlocks with paths held by callers of this function.
  93          * since the alloc_mutex is protecting things right now, just
  94          * skip the locking here
  95          */
  96         path->skip_locking = 1;
  97         first_free = block_group->key.objectid;
  98         key.objectid = block_group->key.objectid;
  99         key.offset = 0;
 100         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 101         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 102         if (ret < 0)
 103                 return ret;
 104         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 105         if (ret < 0)
 106                 return ret;
 107         if (ret == 0) {
 108                 leaf = path->nodes[0];
 109                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 110                 if (key.objectid + key.offset > first_free)
 111                         first_free = key.objectid + key.offset;
 112         }
 113         while(1) {
 114                 leaf = path->nodes[0];
 115                 slot = path->slots[0];
 116                 if (slot >= btrfs_header_nritems(leaf)) {
 117                         ret = btrfs_next_leaf(root, path);
 118                         if (ret < 0)
 119                                 goto err;
 120                         if (ret == 0) {
 121                                 continue;
 122                         } else {
 123                                 break;
 124                         }
 125                 }
 126                 btrfs_item_key_to_cpu(leaf, &key, slot);
 127                 if (key.objectid < block_group->key.objectid) {
 128                         goto next;
 129                 }
 130                 if (key.objectid >= block_group->key.objectid +
 131                     block_group->key.offset) {
 132                         break;
 133                 }
 134
 135                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 136                         if (!found) {
 137                                 last = first_free;
 138                                 found = 1;
 139                         }
 140                         if (key.objectid > last) {
 141                                 hole_size = key.objectid - last;
 142                                 set_extent_dirty(free_space_cache, last,
 143                                                  last + hole_size - 1,
 144                                                  GFP_NOFS);
 145                         }
 146                         last = key.objectid + key.offset;
 147                 }
 148 next:
 149                 path->slots[0]++;
 150         }
 151
 152         if (!found)
 153                 last = first_free;
 154         if (block_group->key.objectid +
 155             block_group->key.offset > last) {
 156                 hole_size = block_group->key.objectid +
 157                         block_group->key.offset - last;
 158                 set_extent_dirty(free_space_cache, last,
 159                                  last + hole_size - 1, GFP_NOFS);
 160         }
 161         block_group->cached = 1;
 162 err:
 163         btrfs_free_path(path);
 164         return 0;
 165 }
 166
 167 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 168                                                        btrfs_fs_info *info,
 169                                                          u64 bytenr)
 170 {
 171         struct extent_io_tree *block_group_cache;
 172         struct btrfs_block_group_cache *block_group = NULL;
 173         u64 ptr;
 174         u64 start;
 175         u64 end;
 176         int ret;
 177
 178         bytenr = max_t(u64, bytenr,
 179                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 180         block_group_cache = &info->block_group_cache;
 181         ret = find_first_extent_bit(block_group_cache,
 182                                     bytenr, &start, &end,
 183                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 184                                     BLOCK_GROUP_SYSTEM);
 185         if (ret) {
 186                 return NULL;
 187         }
 188         ret = get_state_private(block_group_cache, start, &ptr);
 189         if (ret)
 190                 return NULL;
 191
 192         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 193         return block_group;
 194 }
 195
 196 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 197                                                          btrfs_fs_info *info,
 198                                                          u64 bytenr)
 199 {
 200         struct extent_io_tree *block_group_cache;
 201         struct btrfs_block_group_cache *block_group = NULL;
 202         u64 ptr;
 203         u64 start;
 204         u64 end;
 205         int ret;
 206
 207         bytenr = max_t(u64, bytenr,
 208                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 209         block_group_cache = &info->block_group_cache;
 210         ret = find_first_extent_bit(block_group_cache,
 211                                     bytenr, &start, &end,
 212                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 213                                     BLOCK_GROUP_SYSTEM);
 214         if (ret) {
 215                 return NULL;
 216         }
 217         ret = get_state_private(block_group_cache, start, &ptr);
 218         if (ret)
 219                 return NULL;
 220
 221         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 222         if (block_group->key.objectid <= bytenr && bytenr <
 223             block_group->key.objectid + block_group->key.offset)
 224                 return block_group;
 225         return NULL;
 226 }
 227
 228 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 229 {
 230         return (cache->flags & bits) == bits;
 231 }
 232
 233 static int noinline find_search_start(struct btrfs_root *root,
 234                               struct btrfs_block_group_cache **cache_ret,
 235                               u64 *start_ret, u64 num, int data)
 236 {
 237         int ret;
 238         struct btrfs_block_group_cache *cache = *cache_ret;
 239         struct extent_io_tree *free_space_cache;
 240         struct extent_state *state;
 241         u64 last;
 242         u64 start = 0;
 243         u64 cache_miss = 0;
 244         u64 total_fs_bytes;
 245         u64 search_start = *start_ret;
 246         int wrapped = 0;
 247
 248         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 249         free_space_cache = &root->fs_info->free_space_cache;
 250
 251         if (!cache)
 252                 goto out;
 253
 254 again:
 255         ret = cache_block_group(root, cache);
 256         if (ret) {
 257                 goto out;
 258         }
 259
 260         last = max(search_start, cache->key.objectid);
 261         if (!block_group_bits(cache, data) || cache->ro)
 262                 goto new_group;
 263
 264         spin_lock_irq(&free_space_cache->lock);
 265         state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 266         while(1) {
 267                 if (!state) {
 268                         if (!cache_miss)
 269                                 cache_miss = last;
 270                         spin_unlock_irq(&free_space_cache->lock);
 271                         goto new_group;
 272                 }
 273
 274                 start = max(last, state->start);
 275                 last = state->end + 1;
 276                 if (last - start < num) {
 277                         do {
 278                                 state = extent_state_next(state);
 279                         } while(state && !(state->state & EXTENT_DIRTY));
 280                         continue;
 281                 }
 282                 spin_unlock_irq(&free_space_cache->lock);
 283                 if (cache->ro) {
 284                         goto new_group;
 285                 }
 286                 if (start + num > cache->key.objectid + cache->key.offset)
 287                         goto new_group;
 288                 if (!block_group_bits(cache, data)) {
 289                         printk("block group bits don't match %Lu %d\n", cache->flags, data);
 290                 }
 291                 *start_ret = start;
 292                 return 0;
 293         }
 294 out:
 295         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 296         if (!cache) {
 297                 printk("Unable to find block group for %Lu\n", search_start);
 298                 WARN_ON(1);
 299         }
 300         return -ENOSPC;
 301
 302 new_group:
 303         last = cache->key.objectid + cache->key.offset;
 304 wrapped:
 305         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 306         if (!cache || cache->key.objectid >= total_fs_bytes) {
 307 no_cache:
 308                 if (!wrapped) {
 309                         wrapped = 1;
 310                         last = search_start;
 311                         goto wrapped;
 312                 }
 313                 goto out;
 314         }
 315         if (cache_miss && !cache->cached) {
 316                 cache_block_group(root, cache);
 317                 last = cache_miss;
 318                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 319         }
 320         cache_miss = 0;
 321         cache = __btrfs_find_block_group(root, cache, last, data, 0);
 322         if (!cache)
 323                 goto no_cache;
 324         *cache_ret = cache;
 325         goto again;
 326 }
 327
 328 static u64 div_factor(u64 num, int factor)
 329 {
 330         if (factor == 10)
 331                 return num;
 332         num *= factor;
 333         do_div(num, 10);
 334         return num;
 335 }
 336
 337 static int block_group_state_bits(u64 flags)
 338 {
 339         int bits = 0;
 340         if (flags & BTRFS_BLOCK_GROUP_DATA)
 341                 bits |= BLOCK_GROUP_DATA;
 342         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 343                 bits |= BLOCK_GROUP_METADATA;
 344         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 345                 bits |= BLOCK_GROUP_SYSTEM;
 346         return bits;
 347 }
 348
 349 static struct btrfs_block_group_cache *
 350 __btrfs_find_block_group(struct btrfs_root *root,
 351                          struct btrfs_block_group_cache *hint,
 352                          u64 search_start, int data, int owner)
 353 {
 354         struct btrfs_block_group_cache *cache;
 355         struct extent_io_tree *block_group_cache;
 356         struct btrfs_block_group_cache *found_group = NULL;
 357         struct btrfs_fs_info *info = root->fs_info;
 358         u64 used;
 359         u64 last = 0;
 360         u64 start;
 361         u64 end;
 362         u64 free_check;
 363         u64 ptr;
 364         int bit;
 365         int ret;
 366         int full_search = 0;
 367         int factor = 10;
 368         int wrapped = 0;
 369
 370         block_group_cache = &info->block_group_cache;
 371
 372         if (data & BTRFS_BLOCK_GROUP_METADATA)
 373                 factor = 9;
 374
 375         bit = block_group_state_bits(data);
 376
 377         if (search_start) {
 378                 struct btrfs_block_group_cache *shint;
 379                 shint = btrfs_lookup_first_block_group(info, search_start);
 380                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 381                         used = btrfs_block_group_used(&shint->item);
 382                         if (used + shint->pinned <
 383                             div_factor(shint->key.offset, factor)) {
 384                                 return shint;
 385                         }
 386                 }
 387         }
 388         if (hint && !hint->ro && block_group_bits(hint, data)) {
 389                 used = btrfs_block_group_used(&hint->item);
 390                 if (used + hint->pinned <
 391                     div_factor(hint->key.offset, factor)) {
 392                         return hint;
 393                 }
 394                 last = hint->key.objectid + hint->key.offset;
 395         } else {
 396                 if (hint)
 397                         last = max(hint->key.objectid, search_start);
 398                 else
 399                         last = search_start;
 400         }
 401 again:
 402         while(1) {
 403                 ret = find_first_extent_bit(block_group_cache, last,
 404                                             &start, &end, bit);
 405                 if (ret)
 406                         break;
 407
 408                 ret = get_state_private(block_group_cache, start, &ptr);
 409                 if (ret) {
 410                         last = end + 1;
 411                         continue;
 412                 }
 413
 414                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 415                 last = cache->key.objectid + cache->key.offset;
 416                 used = btrfs_block_group_used(&cache->item);
 417
 418                 if (!cache->ro && block_group_bits(cache, data)) {
 419                         free_check = div_factor(cache->key.offset, factor);
 420                         if (used + cache->pinned < free_check) {
 421                                 found_group = cache;
 422                                 goto found;
 423                         }
 424                 }
 425                 cond_resched();
 426         }
 427         if (!wrapped) {
 428                 last = search_start;
 429                 wrapped = 1;
 430                 goto again;
 431         }
 432         if (!full_search && factor < 10) {
 433                 last = search_start;
 434                 full_search = 1;
 435                 factor = 10;
 436                 goto again;
 437         }
 438 found:
 439         return found_group;
 440 }
 441
 442 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 443                                                  struct btrfs_block_group_cache
 444                                                  *hint, u64 search_start,
 445                                                  int data, int owner)
 446 {
 447
 448         struct btrfs_block_group_cache *ret;
 449         mutex_lock(&root->fs_info->alloc_mutex);
 450         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 451         mutex_unlock(&root->fs_info->alloc_mutex);
 452         return ret;
 453 }
 454 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 455                            u64 owner, u64 owner_offset)
 456 {
 457         u32 high_crc = ~(u32)0;
 458         u32 low_crc = ~(u32)0;
 459         __le64 lenum;
 460         lenum = cpu_to_le64(root_objectid);
 461         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 462         lenum = cpu_to_le64(ref_generation);
 463         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 464         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 465                 lenum = cpu_to_le64(owner);
 466                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 467                 lenum = cpu_to_le64(owner_offset);
 468                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 469         }
 470         return ((u64)high_crc << 32) | (u64)low_crc;
 471 }
 472
 473 static int match_extent_ref(struct extent_buffer *leaf,
 474                             struct btrfs_extent_ref *disk_ref,
 475                             struct btrfs_extent_ref *cpu_ref)
 476 {
 477         int ret;
 478         int len;
 479
 480         if (cpu_ref->objectid)
 481                 len = sizeof(*cpu_ref);
 482         else
 483                 len = 2 * sizeof(u64);
 484         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 485                                    len);
 486         return ret == 0;
 487 }
 488
 489 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 490                                           struct btrfs_root *root,
 491                                           struct btrfs_path *path, u64 bytenr,
 492                                           u64 root_objectid,
 493                                           u64 ref_generation, u64 owner,
 494                                           u64 owner_offset, int del)
 495 {
 496         u64 hash;
 497         struct btrfs_key key;
 498         struct btrfs_key found_key;
 499         struct btrfs_extent_ref ref;
 500         struct extent_buffer *leaf;
 501         struct btrfs_extent_ref *disk_ref;
 502         int ret;
 503         int ret2;
 504
 505         btrfs_set_stack_ref_root(&ref, root_objectid);
 506         btrfs_set_stack_ref_generation(&ref, ref_generation);
 507         btrfs_set_stack_ref_objectid(&ref, owner);
 508         btrfs_set_stack_ref_offset(&ref, owner_offset);
 509
 510         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 511                                owner_offset);
 512         key.offset = hash;
 513         key.objectid = bytenr;
 514         key.type = BTRFS_EXTENT_REF_KEY;
 515
 516         while (1) {
 517                 ret = btrfs_search_slot(trans, root, &key, path,
 518                                         del ? -1 : 0, del);
 519                 if (ret < 0)
 520                         goto out;
 521                 leaf = path->nodes[0];
 522                 if (ret != 0) {
 523                         u32 nritems = btrfs_header_nritems(leaf);
 524                         if (path->slots[0] >= nritems) {
 525                                 ret2 = btrfs_next_leaf(root, path);
 526                                 if (ret2)
 527                                         goto out;
 528                                 leaf = path->nodes[0];
 529                         }
 530                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 531                         if (found_key.objectid != bytenr ||
 532                             found_key.type != BTRFS_EXTENT_REF_KEY)
 533                                 goto out;
 534                         key.offset = found_key.offset;
 535                         if (del) {
 536                                 btrfs_release_path(root, path);
 537                                 continue;
 538                         }
 539                 }
 540                 disk_ref = btrfs_item_ptr(path->nodes[0],
 541                                           path->slots[0],
 542                                           struct btrfs_extent_ref);
 543                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 544                         ret = 0;
 545                         goto out;
 546                 }
 547                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 548                 key.offset = found_key.offset + 1;
 549                 btrfs_release_path(root, path);
 550         }
 551 out:
 552         return ret;
 553 }
 554
 555 /*
 556  * Back reference rules.  Back refs have three main goals:
 557  *
 558  * 1) differentiate between all holders of references to an extent so that
 559  *    when a reference is dropped we can make sure it was a valid reference
 560  *    before freeing the extent.
 561  *
 562  * 2) Provide enough information to quickly find the holders of an extent
 563  *    if we notice a given block is corrupted or bad.
 564  *
 565  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 566  *    maintenance.  This is actually the same as #2, but with a slightly
 567  *    different use case.
 568  *
 569  * File extents can be referenced by:
 570  *
 571  * - multiple snapshots, subvolumes, or different generations in one subvol
 572  * - different files inside a single subvolume (in theory, not implemented yet)
 573  * - different offsets inside a file (bookend extents in file.c)
 574  *
 575  * The extent ref structure has fields for:
 576  *
 577  * - Objectid of the subvolume root
 578  * - Generation number of the tree holding the reference
 579  * - objectid of the file holding the reference
 580  * - offset in the file corresponding to the key holding the reference
 581  *
 582  * When a file extent is allocated the fields are filled in:
 583  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 584  *
 585  * When a leaf is cow'd new references are added for every file extent found
 586  * in the leaf.  It looks the same as the create case, but trans->transid
 587  * will be different when the block is cow'd.
 588  *
 589  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 590  *
 591  * When a file extent is removed either during snapshot deletion or file
 592  * truncation, the corresponding back reference is found
 593  * by searching for:
 594  *
 595  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 596  *      inode objectid, offset in file)
 597  *
 598  * Btree extents can be referenced by:
 599  *
 600  * - Different subvolumes
 601  * - Different generations of the same subvolume
 602  *
 603  * Storing sufficient information for a full reverse mapping of a btree
 604  * block would require storing the lowest key of the block in the backref,
 605  * and it would require updating that lowest key either before write out or
 606  * every time it changed.  Instead, the objectid of the lowest key is stored
 607  * along with the level of the tree block.  This provides a hint
 608  * about where in the btree the block can be found.  Searches through the
 609  * btree only need to look for a pointer to that block, so they stop one
 610  * level higher than the level recorded in the backref.
 611  *
 612  * Some btrees do not do reference counting on their extents.  These
 613  * include the extent tree and the tree of tree roots.  Backrefs for these
 614  * trees always have a generation of zero.
 615  *
 616  * When a tree block is created, back references are inserted:
 617  *
 618  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 619  *
 620  * When a tree block is cow'd in a reference counted root,
 621  * new back references are added for all the blocks it points to.
 622  * These are of the form (trans->transid will have increased since creation):
 623  *
 624  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 625  *
 626  * Because the lowest_key_objectid and the level are just hints
 627  * they are not used when backrefs are deleted.  When a backref is deleted:
 628  *
 629  * if backref was for a tree root:
 630  *     root_objectid = root->root_key.objectid
 631  * else
 632  *     root_objectid = btrfs_header_owner(parent)
 633  *
 634  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 635  *
 636  * Back Reference Key hashing:
 637  *
 638  * Back references have four fields, each 64 bits long.  Unfortunately,
 639  * This is hashed into a single 64 bit number and placed into the key offset.
 640  * The key objectid corresponds to the first byte in the extent, and the
 641  * key type is set to BTRFS_EXTENT_REF_KEY
 642  */
 643 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 644                                  struct btrfs_root *root,
 645                                  struct btrfs_path *path, u64 bytenr,
 646                                  u64 root_objectid, u64 ref_generation,
 647                                  u64 owner, u64 owner_offset)
 648 {
 649         u64 hash;
 650         struct btrfs_key key;
 651         struct btrfs_extent_ref ref;
 652         struct btrfs_extent_ref *disk_ref;
 653         int ret;
 654
 655         btrfs_set_stack_ref_root(&ref, root_objectid);
 656         btrfs_set_stack_ref_generation(&ref, ref_generation);
 657         btrfs_set_stack_ref_objectid(&ref, owner);
 658         btrfs_set_stack_ref_offset(&ref, owner_offset);
 659
 660         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 661                                owner_offset);
 662         key.offset = hash;
 663         key.objectid = bytenr;
 664         key.type = BTRFS_EXTENT_REF_KEY;
 665
 666         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 667         while (ret == -EEXIST) {
 668                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 669                                           struct btrfs_extent_ref);
 670                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 671                         goto out;
 672                 key.offset++;
 673                 btrfs_release_path(root, path);
 674                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 675                                               sizeof(ref));
 676         }
 677         if (ret)
 678                 goto out;
 679         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 680                                   struct btrfs_extent_ref);
 681         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 682                             sizeof(ref));
 683         btrfs_mark_buffer_dirty(path->nodes[0]);
 684 out:
 685         btrfs_release_path(root, path);
 686         return ret;
 687 }
 688
 689 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 690                                 struct btrfs_root *root,
 691                                 u64 bytenr, u64 num_bytes,
 692                                 u64 root_objectid, u64 ref_generation,
 693                                 u64 owner, u64 owner_offset)
 694 {
 695         struct btrfs_path *path;
 696         int ret;
 697         struct btrfs_key key;
 698         struct extent_buffer *l;
 699         struct btrfs_extent_item *item;
 700         u32 refs;
 701
 702         WARN_ON(num_bytes < root->sectorsize);
 703         path = btrfs_alloc_path();
 704         if (!path)
 705                 return -ENOMEM;
 706
 707         path->reada = 1;
 708         key.objectid = bytenr;
 709         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 710         key.offset = num_bytes;
 711         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 712                                 0, 1);
 713         if (ret < 0)
 714                 return ret;
 715         if (ret != 0) {
 716                 BUG();
 717         }
 718         BUG_ON(ret != 0);
 719         l = path->nodes[0];
 720         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 721         refs = btrfs_extent_refs(l, item);
 722         btrfs_set_extent_refs(l, item, refs + 1);
 723         btrfs_mark_buffer_dirty(path->nodes[0]);
 724
 725         btrfs_release_path(root->fs_info->extent_root, path);
 726
 727         path->reada = 1;
 728         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 729                                           path, bytenr, root_objectid,
 730                                           ref_generation, owner, owner_offset);
 731         BUG_ON(ret);
 732         finish_current_insert(trans, root->fs_info->extent_root);
 733         del_pending_extents(trans, root->fs_info->extent_root);
 734
 735         btrfs_free_path(path);
 736         return 0;
 737 }
 738
 739 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 740                                 struct btrfs_root *root,
 741                                 u64 bytenr, u64 num_bytes,
 742                                 u64 root_objectid, u64 ref_generation,
 743                                 u64 owner, u64 owner_offset)
 744 {
 745         int ret;
 746
 747         mutex_lock(&root->fs_info->alloc_mutex);
 748         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
 749                                      root_objectid, ref_generation,
 750                                      owner, owner_offset);
 751         mutex_unlock(&root->fs_info->alloc_mutex);
 752         return ret;
 753 }
 754
 755 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 756                          struct btrfs_root *root)
 757 {
 758         finish_current_insert(trans, root->fs_info->extent_root);
 759         del_pending_extents(trans, root->fs_info->extent_root);
 760         return 0;
 761 }
 762
 763 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 764                              struct btrfs_root *root, u64 bytenr,
 765                              u64 num_bytes, u32 *refs)
 766 {
 767         struct btrfs_path *path;
 768         int ret;
 769         struct btrfs_key key;
 770         struct extent_buffer *l;
 771         struct btrfs_extent_item *item;
 772
 773         WARN_ON(num_bytes < root->sectorsize);
 774         path = btrfs_alloc_path();
 775         path->reada = 1;
 776         key.objectid = bytenr;
 777         key.offset = num_bytes;
 778         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 779         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 780                                 0, 0);
 781         if (ret < 0)
 782                 goto out;
 783         if (ret != 0) {
 784                 btrfs_print_leaf(root, path->nodes[0]);
 785                 printk("failed to find block number %Lu\n", bytenr);
 786                 BUG();
 787         }
 788         l = path->nodes[0];
 789         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 790         *refs = btrfs_extent_refs(l, item);
 791 out:
 792         btrfs_free_path(path);
 793         return 0;
 794 }
 795
 796 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 797                                   struct btrfs_path *count_path,
 798                                   u64 expected_owner,
 799                                   u64 first_extent)
 800 {
 801         struct btrfs_root *extent_root = root->fs_info->extent_root;
 802         struct btrfs_path *path;
 803         u64 bytenr;
 804         u64 found_objectid;
 805         u64 found_owner;
 806         u64 root_objectid = root->root_key.objectid;
 807         u32 total_count = 0;
 808         u32 extent_refs;
 809         u32 cur_count;
 810         u32 nritems;
 811         int ret;
 812         struct btrfs_key key;
 813         struct btrfs_key found_key;
 814         struct extent_buffer *l;
 815         struct btrfs_extent_item *item;
 816         struct btrfs_extent_ref *ref_item;
 817         int level = -1;
 818
 819         /* FIXME, needs locking */
 820         BUG();
 821
 822         mutex_lock(&root->fs_info->alloc_mutex);
 823         path = btrfs_alloc_path();
 824 again:
 825         if (level == -1)
 826                 bytenr = first_extent;
 827         else
 828                 bytenr = count_path->nodes[level]->start;
 829
 830         cur_count = 0;
 831         key.objectid = bytenr;
 832         key.offset = 0;
 833
 834         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 835         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 836         if (ret < 0)
 837                 goto out;
 838         BUG_ON(ret == 0);
 839
 840         l = path->nodes[0];
 841         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 842
 843         if (found_key.objectid != bytenr ||
 844             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 845                 goto out;
 846         }
 847
 848         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 849         extent_refs = btrfs_extent_refs(l, item);
 850         while (1) {
 851                 l = path->nodes[0];
 852                 nritems = btrfs_header_nritems(l);
 853                 if (path->slots[0] >= nritems) {
 854                         ret = btrfs_next_leaf(extent_root, path);
 855                         if (ret == 0)
 856                                 continue;
 857                         break;
 858                 }
 859                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 860                 if (found_key.objectid != bytenr)
 861                         break;
 862
 863                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 864                         path->slots[0]++;
 865                         continue;
 866                 }
 867
 868                 cur_count++;
 869                 ref_item = btrfs_item_ptr(l, path->slots[0],
 870                                           struct btrfs_extent_ref);
 871                 found_objectid = btrfs_ref_root(l, ref_item);
 872
 873                 if (found_objectid != root_objectid) {
 874                         total_count = 2;
 875                         goto out;
 876                 }
 877                 if (level == -1) {
 878                         found_owner = btrfs_ref_objectid(l, ref_item);
 879                         if (found_owner != expected_owner) {
 880                                 total_count = 2;
 881                                 goto out;
 882                         }
 883                         /*
 884                          * nasty.  we don't count a reference held by
 885                          * the running transaction.  This allows nodatacow
 886                          * to avoid cow most of the time
 887                          */
 888                         if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
 889                             btrfs_ref_generation(l, ref_item) ==
 890                             root->fs_info->generation) {
 891                                 extent_refs--;
 892                         }
 893                 }
 894                 total_count = 1;
 895                 path->slots[0]++;
 896         }
 897         /*
 898          * if there is more than one reference against a data extent,
 899          * we have to assume the other ref is another snapshot
 900          */
 901         if (level == -1 && extent_refs > 1) {
 902                 total_count = 2;
 903                 goto out;
 904         }
 905         if (cur_count == 0) {
 906                 total_count = 0;
 907                 goto out;
 908         }
 909         if (level >= 0 && root->node == count_path->nodes[level])
 910                 goto out;
 911         level++;
 912         btrfs_release_path(root, path);
 913         goto again;
 914
 915 out:
 916         btrfs_free_path(path);
 917         mutex_unlock(&root->fs_info->alloc_mutex);
 918         return total_count;
 919 }
 920
 921 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 922                   struct extent_buffer *buf)
 923 {
 924         u64 bytenr;
 925         u32 nritems;
 926         struct btrfs_key key;
 927         struct btrfs_file_extent_item *fi;
 928         int i;
 929         int level;
 930         int ret;
 931         int faili;
 932
 933         if (!root->ref_cows)
 934                 return 0;
 935
 936         mutex_lock(&root->fs_info->alloc_mutex);
 937         level = btrfs_header_level(buf);
 938         nritems = btrfs_header_nritems(buf);
 939         for (i = 0; i < nritems; i++) {
 940                 if (level == 0) {
 941                         u64 disk_bytenr;
 942                         btrfs_item_key_to_cpu(buf, &key, i);
 943                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 944                                 continue;
 945                         fi = btrfs_item_ptr(buf, i,
 946                                             struct btrfs_file_extent_item);
 947                         if (btrfs_file_extent_type(buf, fi) ==
 948                             BTRFS_FILE_EXTENT_INLINE)
 949                                 continue;
 950                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 951                         if (disk_bytenr == 0)
 952                                 continue;
 953                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
 954                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 955                                     root->root_key.objectid, trans->transid,
 956                                     key.objectid, key.offset);
 957                         if (ret) {
 958                                 faili = i;
 959                                 goto fail;
 960                         }
 961                 } else {
 962                         bytenr = btrfs_node_blockptr(buf, i);
 963                         btrfs_node_key_to_cpu(buf, &key, i);
 964                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
 965                                            btrfs_level_size(root, level - 1),
 966                                            root->root_key.objectid,
 967                                            trans->transid,
 968                                            level - 1, key.objectid);
 969                         if (ret) {
 970                                 faili = i;
 971                                 goto fail;
 972                         }
 973                 }
 974         }
 975         mutex_unlock(&root->fs_info->alloc_mutex);
 976         return 0;
 977 fail:
 978         WARN_ON(1);
 979 #if 0
 980         for (i =0; i < faili; i++) {
 981                 if (level == 0) {
 982                         u64 disk_bytenr;
 983                         btrfs_item_key_to_cpu(buf, &key, i);
 984                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 985                                 continue;
 986                         fi = btrfs_item_ptr(buf, i,
 987                                             struct btrfs_file_extent_item);
 988                         if (btrfs_file_extent_type(buf, fi) ==
 989                             BTRFS_FILE_EXTENT_INLINE)
 990                                 continue;
 991                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 992                         if (disk_bytenr == 0)
 993                                 continue;
 994                         err = btrfs_free_extent(trans, root, disk_bytenr,
 995                                     btrfs_file_extent_disk_num_bytes(buf,
 996                                                                       fi), 0);
 997                         BUG_ON(err);
 998                 } else {
 999                         bytenr = btrfs_node_blockptr(buf, i);
1000                         err = btrfs_free_extent(trans, root, bytenr,
1001                                         btrfs_level_size(root, level - 1), 0);
1002                         BUG_ON(err);
1003                 }
1004         }
1005 #endif
1006         mutex_unlock(&root->fs_info->alloc_mutex);
1007         return ret;
1008 }
1009
1010 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1011                                  struct btrfs_root *root,
1012                                  struct btrfs_path *path,
1013                                  struct btrfs_block_group_cache *cache)
1014 {
1015         int ret;
1016         int pending_ret;
1017         struct btrfs_root *extent_root = root->fs_info->extent_root;
1018         unsigned long bi;
1019         struct extent_buffer *leaf;
1020
1021         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1022         if (ret < 0)
1023                 goto fail;
1024         BUG_ON(ret);
1025
1026         leaf = path->nodes[0];
1027         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1028         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1029         btrfs_mark_buffer_dirty(leaf);
1030         btrfs_release_path(extent_root, path);
1031 fail:
1032         finish_current_insert(trans, extent_root);
1033         pending_ret = del_pending_extents(trans, extent_root);
1034         if (ret)
1035                 return ret;
1036         if (pending_ret)
1037                 return pending_ret;
1038         return 0;
1039
1040 }
1041
1042 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1043                                    struct btrfs_root *root)
1044 {
1045         struct extent_io_tree *block_group_cache;
1046         struct btrfs_block_group_cache *cache;
1047         int ret;
1048         int err = 0;
1049         int werr = 0;
1050         struct btrfs_path *path;
1051         u64 last = 0;
1052         u64 start;
1053         u64 end;
1054         u64 ptr;
1055
1056         block_group_cache = &root->fs_info->block_group_cache;
1057         path = btrfs_alloc_path();
1058         if (!path)
1059                 return -ENOMEM;
1060
1061         mutex_lock(&root->fs_info->alloc_mutex);
1062         while(1) {
1063                 ret = find_first_extent_bit(block_group_cache, last,
1064                                             &start, &end, BLOCK_GROUP_DIRTY);
1065                 if (ret)
1066                         break;
1067
1068                 last = end + 1;
1069                 ret = get_state_private(block_group_cache, start, &ptr);
1070                 if (ret)
1071                         break;
1072                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1073                 err = write_one_cache_group(trans, root,
1074                                             path, cache);
1075                 /*
1076                  * if we fail to write the cache group, we want
1077                  * to keep it marked dirty in hopes that a later
1078                  * write will work
1079                  */
1080                 if (err) {
1081                         werr = err;
1082                         continue;
1083                 }
1084                 clear_extent_bits(block_group_cache, start, end,
1085                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1086         }
1087         btrfs_free_path(path);
1088         mutex_unlock(&root->fs_info->alloc_mutex);
1089         return werr;
1090 }
1091
1092 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1093                                                   u64 flags)
1094 {
1095         struct list_head *head = &info->space_info;
1096         struct list_head *cur;
1097         struct btrfs_space_info *found;
1098         list_for_each(cur, head) {
1099                 found = list_entry(cur, struct btrfs_space_info, list);
1100                 if (found->flags == flags)
1101                         return found;
1102         }
1103         return NULL;
1104
1105 }
1106
1107 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1108                              u64 total_bytes, u64 bytes_used,
1109                              struct btrfs_space_info **space_info)
1110 {
1111         struct btrfs_space_info *found;
1112
1113         found = __find_space_info(info, flags);
1114         if (found) {
1115                 found->total_bytes += total_bytes;
1116                 found->bytes_used += bytes_used;
1117                 found->full = 0;
1118                 WARN_ON(found->total_bytes < found->bytes_used);
1119                 *space_info = found;
1120                 return 0;
1121         }
1122         found = kmalloc(sizeof(*found), GFP_NOFS);
1123         if (!found)
1124                 return -ENOMEM;
1125
1126         list_add(&found->list, &info->space_info);
1127         found->flags = flags;
1128         found->total_bytes = total_bytes;
1129         found->bytes_used = bytes_used;
1130         found->bytes_pinned = 0;
1131         found->full = 0;
1132         found->force_alloc = 0;
1133         *space_info = found;
1134         return 0;
1135 }
1136
1137 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1138 {
1139         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1140                                    BTRFS_BLOCK_GROUP_RAID1 |
1141                                    BTRFS_BLOCK_GROUP_RAID10 |
1142                                    BTRFS_BLOCK_GROUP_DUP);
1143         if (extra_flags) {
1144                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1145                         fs_info->avail_data_alloc_bits |= extra_flags;
1146                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1147                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1148                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1149                         fs_info->avail_system_alloc_bits |= extra_flags;
1150         }
1151 }
1152
1153 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1154 {
1155         u64 num_devices = root->fs_info->fs_devices->num_devices;
1156
1157         if (num_devices == 1)
1158                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1159         if (num_devices < 4)
1160                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1161
1162         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1163             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1164                       BTRFS_BLOCK_GROUP_RAID10))) {
1165                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1166         }
1167
1168         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1169             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1170                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1171         }
1172
1173         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1174             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1175              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1176              (flags & BTRFS_BLOCK_GROUP_DUP)))
1177                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1178         return flags;
1179 }
1180
1181 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1182                           struct btrfs_root *extent_root, u64 alloc_bytes,
1183                           u64 flags, int force)
1184 {
1185         struct btrfs_space_info *space_info;
1186         u64 thresh;
1187         u64 start;
1188         u64 num_bytes;
1189         int ret;
1190
1191         flags = reduce_alloc_profile(extent_root, flags);
1192
1193         space_info = __find_space_info(extent_root->fs_info, flags);
1194         if (!space_info) {
1195                 ret = update_space_info(extent_root->fs_info, flags,
1196                                         0, 0, &space_info);
1197                 BUG_ON(ret);
1198         }
1199         BUG_ON(!space_info);
1200
1201         if (space_info->force_alloc) {
1202                 force = 1;
1203                 space_info->force_alloc = 0;
1204         }
1205         if (space_info->full)
1206                 goto out;
1207
1208         thresh = div_factor(space_info->total_bytes, 6);
1209         if (!force &&
1210            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1211             thresh)
1212                 goto out;
1213
1214         mutex_lock(&extent_root->fs_info->chunk_mutex);
1215         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1216         if (ret == -ENOSPC) {
1217 printk("space info full %Lu\n", flags);
1218                 space_info->full = 1;
1219                 goto out_unlock;
1220         }
1221         BUG_ON(ret);
1222
1223         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1224                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1225         BUG_ON(ret);
1226 out_unlock:
1227         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1228 out:
1229         return 0;
1230 }
1231
1232 static int update_block_group(struct btrfs_trans_handle *trans,
1233                               struct btrfs_root *root,
1234                               u64 bytenr, u64 num_bytes, int alloc,
1235                               int mark_free)
1236 {
1237         struct btrfs_block_group_cache *cache;
1238         struct btrfs_fs_info *info = root->fs_info;
1239         u64 total = num_bytes;
1240         u64 old_val;
1241         u64 byte_in_group;
1242         u64 start;
1243         u64 end;
1244
1245         while(total) {
1246                 cache = btrfs_lookup_block_group(info, bytenr);
1247                 if (!cache) {
1248                         return -1;
1249                 }
1250                 byte_in_group = bytenr - cache->key.objectid;
1251                 WARN_ON(byte_in_group > cache->key.offset);
1252                 start = cache->key.objectid;
1253                 end = start + cache->key.offset - 1;
1254                 set_extent_bits(&info->block_group_cache, start, end,
1255                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1256
1257                 old_val = btrfs_block_group_used(&cache->item);
1258                 num_bytes = min(total, cache->key.offset - byte_in_group);
1259                 if (alloc) {
1260                         old_val += num_bytes;
1261                         cache->space_info->bytes_used += num_bytes;
1262                 } else {
1263                         old_val -= num_bytes;
1264                         cache->space_info->bytes_used -= num_bytes;
1265                         if (mark_free) {
1266                                 set_extent_dirty(&info->free_space_cache,
1267                                                  bytenr, bytenr + num_bytes - 1,
1268                                                  GFP_NOFS);
1269                         }
1270                 }
1271                 btrfs_set_block_group_used(&cache->item, old_val);
1272                 total -= num_bytes;
1273                 bytenr += num_bytes;
1274         }
1275         return 0;
1276 }
1277
1278 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1279 {
1280         u64 start;
1281         u64 end;
1282         int ret;
1283         ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1284                                     search_start, &start, &end,
1285                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1286                                     BLOCK_GROUP_SYSTEM);
1287         if (ret)
1288                 return 0;
1289         return start;
1290 }
1291
1292
1293 static int update_pinned_extents(struct btrfs_root *root,
1294                                 u64 bytenr, u64 num, int pin)
1295 {
1296         u64 len;
1297         struct btrfs_block_group_cache *cache;
1298         struct btrfs_fs_info *fs_info = root->fs_info;
1299
1300         if (pin) {
1301                 set_extent_dirty(&fs_info->pinned_extents,
1302                                 bytenr, bytenr + num - 1, GFP_NOFS);
1303         } else {
1304                 clear_extent_dirty(&fs_info->pinned_extents,
1305                                 bytenr, bytenr + num - 1, GFP_NOFS);
1306         }
1307         while (num > 0) {
1308                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1309                 if (!cache) {
1310                         u64 first = first_logical_byte(root, bytenr);
1311                         WARN_ON(first < bytenr);
1312                         len = min(first - bytenr, num);
1313                 } else {
1314                         len = min(num, cache->key.offset -
1315                                   (bytenr - cache->key.objectid));
1316                 }
1317                 if (pin) {
1318                         if (cache) {
1319                                 cache->pinned += len;
1320                                 cache->space_info->bytes_pinned += len;
1321                         }
1322                         fs_info->total_pinned += len;
1323                 } else {
1324                         if (cache) {
1325                                 cache->pinned -= len;
1326                                 cache->space_info->bytes_pinned -= len;
1327                         }
1328                         fs_info->total_pinned -= len;
1329                 }
1330                 bytenr += len;
1331                 num -= len;
1332         }
1333         return 0;
1334 }
1335
1336 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1337 {
1338         u64 last = 0;
1339         u64 start;
1340         u64 end;
1341         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1342         int ret;
1343
1344         while(1) {
1345                 ret = find_first_extent_bit(pinned_extents, last,
1346                                             &start, &end, EXTENT_DIRTY);
1347                 if (ret)
1348                         break;
1349                 set_extent_dirty(copy, start, end, GFP_NOFS);
1350                 last = end + 1;
1351         }
1352         return 0;
1353 }
1354
1355 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1356                                struct btrfs_root *root,
1357                                struct extent_io_tree *unpin)
1358 {
1359         u64 start;
1360         u64 end;
1361         int ret;
1362         struct extent_io_tree *free_space_cache;
1363         free_space_cache = &root->fs_info->free_space_cache;
1364
1365         mutex_lock(&root->fs_info->alloc_mutex);
1366         while(1) {
1367                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1368                                             EXTENT_DIRTY);
1369                 if (ret)
1370                         break;
1371                 update_pinned_extents(root, start, end + 1 - start, 0);
1372                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1373                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1374         }
1375         mutex_unlock(&root->fs_info->alloc_mutex);
1376         return 0;
1377 }
1378
1379 static int finish_current_insert(struct btrfs_trans_handle *trans,
1380                                  struct btrfs_root *extent_root)
1381 {
1382         u64 start;
1383         u64 end;
1384         struct btrfs_fs_info *info = extent_root->fs_info;
1385         struct extent_buffer *eb;
1386         struct btrfs_path *path;
1387         struct btrfs_key ins;
1388         struct btrfs_disk_key first;
1389         struct btrfs_extent_item extent_item;
1390         int ret;
1391         int level;
1392         int err = 0;
1393
1394         btrfs_set_stack_extent_refs(&extent_item, 1);
1395         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1396         path = btrfs_alloc_path();
1397
1398         while(1) {
1399                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1400                                             &end, EXTENT_LOCKED);
1401                 if (ret)
1402                         break;
1403
1404                 ins.objectid = start;
1405                 ins.offset = end + 1 - start;
1406                 err = btrfs_insert_item(trans, extent_root, &ins,
1407                                         &extent_item, sizeof(extent_item));
1408                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1409                                   GFP_NOFS);
1410                 eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1411                                      trans->transid);
1412                 btrfs_tree_lock(eb);
1413                 level = btrfs_header_level(eb);
1414                 if (level == 0) {
1415                         btrfs_item_key(eb, &first, 0);
1416                 } else {
1417                         btrfs_node_key(eb, &first, 0);
1418                 }
1419                 btrfs_tree_unlock(eb);
1420                 free_extent_buffer(eb);
1421                 /*
1422                  * the first key is just a hint, so the race we've created
1423                  * against reading it is fine
1424                  */
1425                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1426                                           start, extent_root->root_key.objectid,
1427                                           0, level,
1428                                           btrfs_disk_key_objectid(&first));
1429                 BUG_ON(err);
1430         }
1431         btrfs_free_path(path);
1432         return 0;
1433 }
1434
1435 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1436                           int pending)
1437 {
1438         int err = 0;
1439
1440         if (!pending) {
1441                 struct extent_buffer *buf;
1442                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1443                 if (buf) {
1444                         if (!btrfs_try_tree_lock(buf) &&
1445                             btrfs_buffer_uptodate(buf, 0)) {
1446                                 u64 transid =
1447                                     root->fs_info->running_transaction->transid;
1448                                 u64 header_transid =
1449                                         btrfs_header_generation(buf);
1450                                 if (header_transid == transid &&
1451                                     !btrfs_header_flag(buf,
1452                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1453                                         clean_tree_block(NULL, root, buf);
1454                                         btrfs_tree_unlock(buf);
1455                                         free_extent_buffer(buf);
1456                                         return 1;
1457                                 }
1458                                 btrfs_tree_unlock(buf);
1459                         }
1460                         free_extent_buffer(buf);
1461                 }
1462                 update_pinned_extents(root, bytenr, num_bytes, 1);
1463         } else {
1464                 set_extent_bits(&root->fs_info->pending_del,
1465                                 bytenr, bytenr + num_bytes - 1,
1466                                 EXTENT_LOCKED, GFP_NOFS);
1467         }
1468         BUG_ON(err < 0);
1469         return 0;
1470 }
1471
1472 /*
1473  * remove an extent from the root, returns 0 on success
1474  */
1475 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1476                          *root, u64 bytenr, u64 num_bytes,
1477                          u64 root_objectid, u64 ref_generation,
1478                          u64 owner_objectid, u64 owner_offset, int pin,
1479                          int mark_free)
1480 {
1481         struct btrfs_path *path;
1482         struct btrfs_key key;
1483         struct btrfs_fs_info *info = root->fs_info;
1484         struct btrfs_root *extent_root = info->extent_root;
1485         struct extent_buffer *leaf;
1486         int ret;
1487         int extent_slot = 0;
1488         int found_extent = 0;
1489         int num_to_del = 1;
1490         struct btrfs_extent_item *ei;
1491         u32 refs;
1492
1493         key.objectid = bytenr;
1494         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1495         key.offset = num_bytes;
1496         path = btrfs_alloc_path();
1497         if (!path)
1498                 return -ENOMEM;
1499
1500         path->reada = 1;
1501         ret = lookup_extent_backref(trans, extent_root, path,
1502                                     bytenr, root_objectid,
1503                                     ref_generation,
1504                                     owner_objectid, owner_offset, 1);
1505         if (ret == 0) {
1506                 struct btrfs_key found_key;
1507                 extent_slot = path->slots[0];
1508                 while(extent_slot > 0) {
1509                         extent_slot--;
1510                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1511                                               extent_slot);
1512                         if (found_key.objectid != bytenr)
1513                                 break;
1514                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1515                             found_key.offset == num_bytes) {
1516                                 found_extent = 1;
1517                                 break;
1518                         }
1519                         if (path->slots[0] - extent_slot > 5)
1520                                 break;
1521                 }
1522                 if (!found_extent)
1523                         ret = btrfs_del_item(trans, extent_root, path);
1524         } else {
1525                 btrfs_print_leaf(extent_root, path->nodes[0]);
1526                 WARN_ON(1);
1527                 printk("Unable to find ref byte nr %Lu root %Lu "
1528                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1529                        root_objectid, ref_generation, owner_objectid,
1530                        owner_offset);
1531         }
1532         if (!found_extent) {
1533                 btrfs_release_path(extent_root, path);
1534                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1535                 if (ret < 0)
1536                         return ret;
1537                 BUG_ON(ret);
1538                 extent_slot = path->slots[0];
1539         }
1540
1541         leaf = path->nodes[0];
1542         ei = btrfs_item_ptr(leaf, extent_slot,
1543                             struct btrfs_extent_item);
1544         refs = btrfs_extent_refs(leaf, ei);
1545         BUG_ON(refs == 0);
1546         refs -= 1;
1547         btrfs_set_extent_refs(leaf, ei, refs);
1548
1549         btrfs_mark_buffer_dirty(leaf);
1550
1551         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1552                 /* if the back ref and the extent are next to each other
1553                  * they get deleted below in one shot
1554                  */
1555                 path->slots[0] = extent_slot;
1556                 num_to_del = 2;
1557         } else if (found_extent) {
1558                 /* otherwise delete the extent back ref */
1559                 ret = btrfs_del_item(trans, extent_root, path);
1560                 BUG_ON(ret);
1561                 /* if refs are 0, we need to setup the path for deletion */
1562                 if (refs == 0) {
1563                         btrfs_release_path(extent_root, path);
1564                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1565                                                 -1, 1);
1566                         if (ret < 0)
1567                                 return ret;
1568                         BUG_ON(ret);
1569                 }
1570         }
1571
1572         if (refs == 0) {
1573                 u64 super_used;
1574                 u64 root_used;
1575
1576                 if (pin) {
1577                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1578                         if (ret > 0)
1579                                 mark_free = 1;
1580                         BUG_ON(ret < 0);
1581                 }
1582
1583                 /* block accounting for super block */
1584                 spin_lock_irq(&info->delalloc_lock);
1585                 super_used = btrfs_super_bytes_used(&info->super_copy);
1586                 btrfs_set_super_bytes_used(&info->super_copy,
1587                                            super_used - num_bytes);
1588                 spin_unlock_irq(&info->delalloc_lock);
1589
1590                 /* block accounting for root item */
1591                 root_used = btrfs_root_used(&root->root_item);
1592                 btrfs_set_root_used(&root->root_item,
1593                                            root_used - num_bytes);
1594                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1595                                       num_to_del);
1596                 if (ret) {
1597                         return ret;
1598                 }
1599                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1600                                          mark_free);
1601                 BUG_ON(ret);
1602         }
1603         btrfs_free_path(path);
1604         finish_current_insert(trans, extent_root);
1605         return ret;
1606 }
1607
1608 /*
1609  * find all the blocks marked as pending in the radix tree and remove
1610  * them from the extent map
1611  */
1612 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1613                                btrfs_root *extent_root)
1614 {
1615         int ret;
1616         int err = 0;
1617         u64 start;
1618         u64 end;
1619         struct extent_io_tree *pending_del;
1620         struct extent_io_tree *pinned_extents;
1621
1622         pending_del = &extent_root->fs_info->pending_del;
1623         pinned_extents = &extent_root->fs_info->pinned_extents;
1624
1625         while(1) {
1626                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1627                                             EXTENT_LOCKED);
1628                 if (ret)
1629                         break;
1630                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1631                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1632                                   GFP_NOFS);
1633                 ret = __free_extent(trans, extent_root,
1634                                      start, end + 1 - start,
1635                                      extent_root->root_key.objectid,
1636                                      0, 0, 0, 0, 0);
1637                 if (ret)
1638                         err = ret;
1639         }
1640         return err;
1641 }
1642
1643 /*
1644  * remove an extent from the root, returns 0 on success
1645  */
1646 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1647                                struct btrfs_root *root, u64 bytenr,
1648                                u64 num_bytes, u64 root_objectid,
1649                                u64 ref_generation, u64 owner_objectid,
1650                                u64 owner_offset, int pin)
1651 {
1652         struct btrfs_root *extent_root = root->fs_info->extent_root;
1653         int pending_ret;
1654         int ret;
1655
1656         WARN_ON(num_bytes < root->sectorsize);
1657         if (!root->ref_cows)
1658                 ref_generation = 0;
1659
1660         if (root == extent_root) {
1661                 pin_down_bytes(root, bytenr, num_bytes, 1);
1662                 return 0;
1663         }
1664         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1665                             ref_generation, owner_objectid, owner_offset,
1666                             pin, pin == 0);
1667         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1668         return ret ? ret : pending_ret;
1669 }
1670
1671 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1672                       struct btrfs_root *root, u64 bytenr,
1673                       u64 num_bytes, u64 root_objectid,
1674                       u64 ref_generation, u64 owner_objectid,
1675                       u64 owner_offset, int pin)
1676 {
1677         int ret;
1678
1679         maybe_lock_mutex(root);
1680         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1681                                   root_objectid, ref_generation,
1682                                   owner_objectid, owner_offset, pin);
1683         maybe_unlock_mutex(root);
1684         return ret;
1685 }
1686
1687 static u64 stripe_align(struct btrfs_root *root, u64 val)
1688 {
1689         u64 mask = ((u64)root->stripesize - 1);
1690         u64 ret = (val + mask) & ~mask;
1691         return ret;
1692 }
1693
1694 /*
1695  * walks the btree of allocated extents and find a hole of a given size.
1696  * The key ins is changed to record the hole:
1697  * ins->objectid == block start
1698  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1699  * ins->offset == number of blocks
1700  * Any available blocks before search_start are skipped.
1701  */
1702 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1703                                      struct btrfs_root *orig_root,
1704                                      u64 num_bytes, u64 empty_size,
1705                                      u64 search_start, u64 search_end,
1706                                      u64 hint_byte, struct btrfs_key *ins,
1707                                      u64 exclude_start, u64 exclude_nr,
1708                                      int data)
1709 {
1710         int ret;
1711         u64 orig_search_start;
1712         struct btrfs_root * root = orig_root->fs_info->extent_root;
1713         struct btrfs_fs_info *info = root->fs_info;
1714         u64 total_needed = num_bytes;
1715         u64 *last_ptr = NULL;
1716         struct btrfs_block_group_cache *block_group;
1717         int full_scan = 0;
1718         int wrapped = 0;
1719         int chunk_alloc_done = 0;
1720         int empty_cluster = 2 * 1024 * 1024;
1721         int allowed_chunk_alloc = 0;
1722
1723         WARN_ON(num_bytes < root->sectorsize);
1724         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1725
1726         if (orig_root->ref_cows || empty_size)
1727                 allowed_chunk_alloc = 1;
1728
1729         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1730                 last_ptr = &root->fs_info->last_alloc;
1731                 empty_cluster = 256 * 1024;
1732         }
1733
1734         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1735                 last_ptr = &root->fs_info->last_data_alloc;
1736         }
1737
1738         if (last_ptr) {
1739                 if (*last_ptr)
1740                         hint_byte = *last_ptr;
1741                 else {
1742                         empty_size += empty_cluster;
1743                 }
1744         }
1745
1746         search_start = max(search_start, first_logical_byte(root, 0));
1747         orig_search_start = search_start;
1748
1749         if (search_end == (u64)-1)
1750                 search_end = btrfs_super_total_bytes(&info->super_copy);
1751
1752         if (hint_byte) {
1753                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1754                 if (!block_group)
1755                         hint_byte = search_start;
1756                 block_group = __btrfs_find_block_group(root, block_group,
1757                                                      hint_byte, data, 1);
1758                 if (last_ptr && *last_ptr == 0 && block_group)
1759                         hint_byte = block_group->key.objectid;
1760         } else {
1761                 block_group = __btrfs_find_block_group(root,
1762                                                      trans->block_group,
1763                                                      search_start, data, 1);
1764         }
1765         search_start = max(search_start, hint_byte);
1766
1767         total_needed += empty_size;
1768
1769 check_failed:
1770         if (!block_group) {
1771                 block_group = btrfs_lookup_first_block_group(info,
1772                                                              search_start);
1773                 if (!block_group)
1774                         block_group = btrfs_lookup_first_block_group(info,
1775                                                        orig_search_start);
1776         }
1777         if (full_scan && !chunk_alloc_done) {
1778                 if (allowed_chunk_alloc) {
1779                         do_chunk_alloc(trans, root,
1780                                      num_bytes + 2 * 1024 * 1024, data, 1);
1781                         allowed_chunk_alloc = 0;
1782                 } else if (block_group && block_group_bits(block_group, data)) {
1783                         block_group->space_info->force_alloc = 1;
1784                 }
1785                 chunk_alloc_done = 1;
1786         }
1787         ret = find_search_start(root, &block_group, &search_start,
1788                                 total_needed, data);
1789         if (ret == -ENOSPC && last_ptr && *last_ptr) {
1790                 *last_ptr = 0;
1791                 block_group = btrfs_lookup_first_block_group(info,
1792                                                              orig_search_start);
1793                 search_start = orig_search_start;
1794                 ret = find_search_start(root, &block_group, &search_start,
1795                                         total_needed, data);
1796         }
1797         if (ret == -ENOSPC)
1798                 goto enospc;
1799         if (ret)
1800                 goto error;
1801
1802         if (last_ptr && *last_ptr && search_start != *last_ptr) {
1803                 *last_ptr = 0;
1804                 if (!empty_size) {
1805                         empty_size += empty_cluster;
1806                         total_needed += empty_size;
1807                 }
1808                 block_group = btrfs_lookup_first_block_group(info,
1809                                                        orig_search_start);
1810                 search_start = orig_search_start;
1811                 ret = find_search_start(root, &block_group,
1812                                         &search_start, total_needed, data);
1813                 if (ret == -ENOSPC)
1814                         goto enospc;
1815                 if (ret)
1816                         goto error;
1817         }
1818
1819         search_start = stripe_align(root, search_start);
1820         ins->objectid = search_start;
1821         ins->offset = num_bytes;
1822
1823         if (ins->objectid + num_bytes >= search_end)
1824                 goto enospc;
1825
1826         if (ins->objectid + num_bytes >
1827             block_group->key.objectid + block_group->key.offset) {
1828                 search_start = block_group->key.objectid +
1829                         block_group->key.offset;
1830                 goto new_group;
1831         }
1832
1833         if (test_range_bit(&info->extent_ins, ins->objectid,
1834                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1835                 search_start = ins->objectid + num_bytes;
1836                 goto new_group;
1837         }
1838
1839         if (test_range_bit(&info->pinned_extents, ins->objectid,
1840                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1841                 search_start = ins->objectid + num_bytes;
1842                 goto new_group;
1843         }
1844
1845         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1846             ins->objectid < exclude_start + exclude_nr)) {
1847                 search_start = exclude_start + exclude_nr;
1848                 goto new_group;
1849         }
1850
1851         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1852                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1853                 if (block_group)
1854                         trans->block_group = block_group;
1855         }
1856         ins->offset = num_bytes;
1857         if (last_ptr) {
1858                 *last_ptr = ins->objectid + ins->offset;
1859                 if (*last_ptr ==
1860                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1861                         *last_ptr = 0;
1862                 }
1863         }
1864         return 0;
1865
1866 new_group:
1867         if (search_start + num_bytes >= search_end) {
1868 enospc:
1869                 search_start = orig_search_start;
1870                 if (full_scan) {
1871                         ret = -ENOSPC;
1872                         goto error;
1873                 }
1874                 if (wrapped) {
1875                         if (!full_scan)
1876                                 total_needed -= empty_size;
1877                         full_scan = 1;
1878                 } else
1879                         wrapped = 1;
1880         }
1881         block_group = btrfs_lookup_first_block_group(info, search_start);
1882         cond_resched();
1883         block_group = __btrfs_find_block_group(root, block_group,
1884                                              search_start, data, 0);
1885         goto check_failed;
1886
1887 error:
1888         return ret;
1889 }
1890
1891 /*
1892  * finds a free extent and does all the dirty work required for allocation
1893  * returns the key for the extent through ins, and a tree buffer for
1894  * the first block of the extent through buf.
1895  *
1896  * returns 0 if everything worked, non-zero otherwise.
1897  */
1898 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1899                        struct btrfs_root *root,
1900                        u64 num_bytes, u64 min_alloc_size,
1901                        u64 root_objectid, u64 ref_generation,
1902                        u64 owner, u64 owner_offset,
1903                        u64 empty_size, u64 hint_byte,
1904                        u64 search_end, struct btrfs_key *ins, u64 data)
1905 {
1906         int ret;
1907         int pending_ret;
1908         u64 super_used;
1909         u64 root_used;
1910         u64 search_start = 0;
1911         u64 alloc_profile;
1912         u32 sizes[2];
1913         struct btrfs_fs_info *info = root->fs_info;
1914         struct btrfs_root *extent_root = info->extent_root;
1915         struct btrfs_extent_item *extent_item;
1916         struct btrfs_extent_ref *ref;
1917         struct btrfs_path *path;
1918         struct btrfs_key keys[2];
1919
1920         maybe_lock_mutex(root);
1921
1922         if (data) {
1923                 alloc_profile = info->avail_data_alloc_bits &
1924                                 info->data_alloc_profile;
1925                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1926         } else if (root == root->fs_info->chunk_root) {
1927                 alloc_profile = info->avail_system_alloc_bits &
1928                                 info->system_alloc_profile;
1929                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1930         } else {
1931                 alloc_profile = info->avail_metadata_alloc_bits &
1932                                 info->metadata_alloc_profile;
1933                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1934         }
1935 again:
1936         data = reduce_alloc_profile(root, data);
1937         /*
1938          * the only place that sets empty_size is btrfs_realloc_node, which
1939          * is not called recursively on allocations
1940          */
1941         if (empty_size || root->ref_cows) {
1942                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1943                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1944                                      2 * 1024 * 1024,
1945                                      BTRFS_BLOCK_GROUP_METADATA |
1946                                      (info->metadata_alloc_profile &
1947                                       info->avail_metadata_alloc_bits), 0);
1948                         BUG_ON(ret);
1949                 }
1950                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1951                                      num_bytes + 2 * 1024 * 1024, data, 0);
1952                 BUG_ON(ret);
1953         }
1954
1955         WARN_ON(num_bytes < root->sectorsize);
1956         ret = find_free_extent(trans, root, num_bytes, empty_size,
1957                                search_start, search_end, hint_byte, ins,
1958                                trans->alloc_exclude_start,
1959                                trans->alloc_exclude_nr, data);
1960
1961         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1962                 num_bytes = num_bytes >> 1;
1963                 num_bytes = max(num_bytes, min_alloc_size);
1964                 do_chunk_alloc(trans, root->fs_info->extent_root,
1965                                num_bytes, data, 1);
1966                 goto again;
1967         }
1968         if (ret) {
1969                 printk("allocation failed flags %Lu\n", data);
1970         }
1971         if (ret) {
1972                 BUG();
1973                 goto out;
1974         }
1975
1976         /* block accounting for super block */
1977         spin_lock_irq(&info->delalloc_lock);
1978         super_used = btrfs_super_bytes_used(&info->super_copy);
1979         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1980         spin_unlock_irq(&info->delalloc_lock);
1981
1982         /* block accounting for root item */
1983         root_used = btrfs_root_used(&root->root_item);
1984         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1985
1986         clear_extent_dirty(&root->fs_info->free_space_cache,
1987                            ins->objectid, ins->objectid + ins->offset - 1,
1988                            GFP_NOFS);
1989
1990         if (root == extent_root) {
1991                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1992                                 ins->objectid + ins->offset - 1,
1993                                 EXTENT_LOCKED, GFP_NOFS);
1994                 goto update_block;
1995         }
1996
1997         WARN_ON(trans->alloc_exclude_nr);
1998         trans->alloc_exclude_start = ins->objectid;
1999         trans->alloc_exclude_nr = ins->offset;
2000
2001         memcpy(&keys[0], ins, sizeof(*ins));
2002         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2003                                          owner, owner_offset);
2004         keys[1].objectid = ins->objectid;
2005         keys[1].type = BTRFS_EXTENT_REF_KEY;
2006         sizes[0] = sizeof(*extent_item);
2007         sizes[1] = sizeof(*ref);
2008
2009         path = btrfs_alloc_path();
2010         BUG_ON(!path);
2011
2012         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2013                                        sizes, 2);
2014
2015         BUG_ON(ret);
2016         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2017                                      struct btrfs_extent_item);
2018         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2019         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2020                              struct btrfs_extent_ref);
2021
2022         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2023         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2024         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2025         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2026
2027         btrfs_mark_buffer_dirty(path->nodes[0]);
2028
2029         trans->alloc_exclude_start = 0;
2030         trans->alloc_exclude_nr = 0;
2031         btrfs_free_path(path);
2032         finish_current_insert(trans, extent_root);
2033         pending_ret = del_pending_extents(trans, extent_root);
2034
2035         if (ret)
2036                 goto out;
2037         if (pending_ret) {
2038                 ret = pending_ret;
2039                 goto out;
2040         }
2041
2042 update_block:
2043         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2044         if (ret) {
2045                 printk("update block group failed for %Lu %Lu\n",
2046                        ins->objectid, ins->offset);
2047                 BUG();
2048         }
2049 out:
2050         maybe_unlock_mutex(root);
2051         return ret;
2052 }
2053 /*
2054  * helper function to allocate a block for a given tree
2055  * returns the tree buffer or NULL.
2056  */
2057 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2058                                              struct btrfs_root *root,
2059                                              u32 blocksize,
2060                                              u64 root_objectid,
2061                                              u64 ref_generation,
2062                                              u64 first_objectid,
2063                                              int level,
2064                                              u64 hint,
2065                                              u64 empty_size)
2066 {
2067         struct btrfs_key ins;
2068         int ret;
2069         struct extent_buffer *buf;
2070
2071         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2072                                  root_objectid, ref_generation,
2073                                  level, first_objectid, empty_size, hint,
2074                                  (u64)-1, &ins, 0);
2075         if (ret) {
2076                 BUG_ON(ret > 0);
2077                 return ERR_PTR(ret);
2078         }
2079         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2080         if (!buf) {
2081                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2082                                   root->root_key.objectid, ref_generation,
2083                                   0, 0, 0);
2084                 return ERR_PTR(-ENOMEM);
2085         }
2086         btrfs_set_header_generation(buf, trans->transid);
2087         btrfs_tree_lock(buf);
2088         clean_tree_block(trans, root, buf);
2089         btrfs_set_buffer_uptodate(buf);
2090
2091         if (PageDirty(buf->first_page)) {
2092                 printk("page %lu dirty\n", buf->first_page->index);
2093                 WARN_ON(1);
2094         }
2095
2096         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2097                          buf->start + buf->len - 1, GFP_NOFS);
2098         trans->blocks_used++;
2099         return buf;
2100 }
2101
2102 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2103                                   struct btrfs_root *root,
2104                                   struct extent_buffer *leaf)
2105 {
2106         u64 leaf_owner;
2107         u64 leaf_generation;
2108         struct btrfs_key key;
2109         struct btrfs_file_extent_item *fi;
2110         int i;
2111         int nritems;
2112         int ret;
2113
2114         BUG_ON(!btrfs_is_leaf(leaf));
2115         nritems = btrfs_header_nritems(leaf);
2116         leaf_owner = btrfs_header_owner(leaf);
2117         leaf_generation = btrfs_header_generation(leaf);
2118
2119         for (i = 0; i < nritems; i++) {
2120                 u64 disk_bytenr;
2121
2122                 btrfs_item_key_to_cpu(leaf, &key, i);
2123                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2124                         continue;
2125                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2126                 if (btrfs_file_extent_type(leaf, fi) ==
2127                     BTRFS_FILE_EXTENT_INLINE)
2128                         continue;
2129                 /*
2130                  * FIXME make sure to insert a trans record that
2131                  * repeats the snapshot del on crash
2132                  */
2133                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2134                 if (disk_bytenr == 0)
2135                         continue;
2136                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2137                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2138                                 leaf_owner, leaf_generation,
2139                                 key.objectid, key.offset, 0);
2140                 BUG_ON(ret);
2141         }
2142         return 0;
2143 }
2144
2145 static void noinline reada_walk_down(struct btrfs_root *root,
2146                                      struct extent_buffer *node,
2147                                      int slot)
2148 {
2149         u64 bytenr;
2150         u64 last = 0;
2151         u32 nritems;
2152         u32 refs;
2153         u32 blocksize;
2154         int ret;
2155         int i;
2156         int level;
2157         int skipped = 0;
2158
2159         nritems = btrfs_header_nritems(node);
2160         level = btrfs_header_level(node);
2161         if (level)
2162                 return;
2163
2164         for (i = slot; i < nritems && skipped < 32; i++) {
2165                 bytenr = btrfs_node_blockptr(node, i);
2166                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2167                              (last > bytenr && last - bytenr > 32 * 1024))) {
2168                         skipped++;
2169                         continue;
2170                 }
2171                 blocksize = btrfs_level_size(root, level - 1);
2172                 if (i != slot) {
2173                         ret = lookup_extent_ref(NULL, root, bytenr,
2174                                                 blocksize, &refs);
2175                         BUG_ON(ret);
2176                         if (refs != 1) {
2177                                 skipped++;
2178                                 continue;
2179                         }
2180                 }
2181                 ret = readahead_tree_block(root, bytenr, blocksize,
2182                                            btrfs_node_ptr_generation(node, i));
2183                 last = bytenr + blocksize;
2184                 cond_resched();
2185                 if (ret)
2186                         break;
2187         }
2188 }
2189
2190 /*
2191  * we want to avoid as much random IO as we can with the alloc mutex
2192  * held, so drop the lock and do the lookup, then do it again with the
2193  * lock held.
2194  */
2195 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2196                               u32 *refs)
2197 {
2198         mutex_unlock(&root->fs_info->alloc_mutex);
2199         lookup_extent_ref(NULL, root, start, len, refs);
2200         cond_resched();
2201         mutex_lock(&root->fs_info->alloc_mutex);
2202         return lookup_extent_ref(NULL, root, start, len, refs);
2203 }
2204
2205 /*
2206  * helper function for drop_snapshot, this walks down the tree dropping ref
2207  * counts as it goes.
2208  */
2209 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2210                                    struct btrfs_root *root,
2211                                    struct btrfs_path *path, int *level)
2212 {
2213         u64 root_owner;
2214         u64 root_gen;
2215         u64 bytenr;
2216         u64 ptr_gen;
2217         struct extent_buffer *next;
2218         struct extent_buffer *cur;
2219         struct extent_buffer *parent;
2220         u32 blocksize;
2221         int ret;
2222         u32 refs;
2223
2224         mutex_lock(&root->fs_info->alloc_mutex);
2225
2226         WARN_ON(*level < 0);
2227         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2228         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2229                                 path->nodes[*level]->len, &refs);
2230         BUG_ON(ret);
2231         if (refs > 1)
2232                 goto out;
2233
2234         /*
2235          * walk down to the last node level and free all the leaves
2236          */
2237         while(*level >= 0) {
2238                 WARN_ON(*level < 0);
2239                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2240                 cur = path->nodes[*level];
2241
2242                 if (btrfs_header_level(cur) != *level)
2243                         WARN_ON(1);
2244
2245                 if (path->slots[*level] >=
2246                     btrfs_header_nritems(cur))
2247                         break;
2248                 if (*level == 0) {
2249                         ret = drop_leaf_ref(trans, root, cur);
2250                         BUG_ON(ret);
2251                         break;
2252                 }
2253                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2254                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2255                 blocksize = btrfs_level_size(root, *level - 1);
2256
2257                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2258                 BUG_ON(ret);
2259                 if (refs != 1) {
2260                         parent = path->nodes[*level];
2261                         root_owner = btrfs_header_owner(parent);
2262                         root_gen = btrfs_header_generation(parent);
2263                         path->slots[*level]++;
2264                         ret = __btrfs_free_extent(trans, root, bytenr,
2265                                                 blocksize, root_owner,
2266                                                 root_gen, 0, 0, 1);
2267                         BUG_ON(ret);
2268                         continue;
2269                 }
2270                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2271                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2272                         free_extent_buffer(next);
2273                         mutex_unlock(&root->fs_info->alloc_mutex);
2274
2275                         if (path->slots[*level] == 0)
2276                                 reada_walk_down(root, cur, path->slots[*level]);
2277
2278                         next = read_tree_block(root, bytenr, blocksize,
2279                                                ptr_gen);
2280                         cond_resched();
2281                         mutex_lock(&root->fs_info->alloc_mutex);
2282
2283                         /* we've dropped the lock, double check */
2284                         ret = drop_snap_lookup_refcount(root, bytenr,
2285                                                 blocksize, &refs);
2286                         BUG_ON(ret);
2287                         if (refs != 1) {
2288                                 parent = path->nodes[*level];
2289                                 root_owner = btrfs_header_owner(parent);
2290                                 root_gen = btrfs_header_generation(parent);
2291
2292                                 path->slots[*level]++;
2293                                 free_extent_buffer(next);
2294                                 ret = __btrfs_free_extent(trans, root, bytenr,
2295                                                         blocksize,
2296                                                         root_owner,
2297                                                         root_gen, 0, 0, 1);
2298                                 BUG_ON(ret);
2299                                 continue;
2300                         }
2301                 }
2302                 WARN_ON(*level <= 0);
2303                 if (path->nodes[*level-1])
2304                         free_extent_buffer(path->nodes[*level-1]);
2305                 path->nodes[*level-1] = next;
2306                 *level = btrfs_header_level(next);
2307                 path->slots[*level] = 0;
2308         }
2309 out:
2310         WARN_ON(*level < 0);
2311         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2312
2313         if (path->nodes[*level] == root->node) {
2314                 root_owner = root->root_key.objectid;
2315                 parent = path->nodes[*level];
2316         } else {
2317                 parent = path->nodes[*level + 1];
2318                 root_owner = btrfs_header_owner(parent);
2319         }
2320
2321         root_gen = btrfs_header_generation(parent);
2322         ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2323                                 path->nodes[*level]->len,
2324                                 root_owner, root_gen, 0, 0, 1);
2325         free_extent_buffer(path->nodes[*level]);
2326         path->nodes[*level] = NULL;
2327         *level += 1;
2328         BUG_ON(ret);
2329         mutex_unlock(&root->fs_info->alloc_mutex);
2330         cond_resched();
2331         return 0;
2332 }
2333
2334 /*
2335  * helper for dropping snapshots.  This walks back up the tree in the path
2336  * to find the first node higher up where we haven't yet gone through
2337  * all the slots
2338  */
2339 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2340                                  struct btrfs_root *root,
2341                                  struct btrfs_path *path, int *level)
2342 {
2343         u64 root_owner;
2344         u64 root_gen;
2345         struct btrfs_root_item *root_item = &root->root_item;
2346         int i;
2347         int slot;
2348         int ret;
2349
2350         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2351                 slot = path->slots[i];
2352                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2353                         struct extent_buffer *node;
2354                         struct btrfs_disk_key disk_key;
2355                         node = path->nodes[i];
2356                         path->slots[i]++;
2357                         *level = i;
2358                         WARN_ON(*level == 0);
2359                         btrfs_node_key(node, &disk_key, path->slots[i]);
2360                         memcpy(&root_item->drop_progress,
2361                                &disk_key, sizeof(disk_key));
2362                         root_item->drop_level = i;
2363                         return 0;
2364                 } else {
2365                         if (path->nodes[*level] == root->node) {
2366                                 root_owner = root->root_key.objectid;
2367                                 root_gen =
2368                                    btrfs_header_generation(path->nodes[*level]);
2369                         } else {
2370                                 struct extent_buffer *node;
2371                                 node = path->nodes[*level + 1];
2372                                 root_owner = btrfs_header_owner(node);
2373                                 root_gen = btrfs_header_generation(node);
2374                         }
2375                         ret = btrfs_free_extent(trans, root,
2376                                                 path->nodes[*level]->start,
2377                                                 path->nodes[*level]->len,
2378                                                 root_owner, root_gen, 0, 0, 1);
2379                         BUG_ON(ret);
2380                         free_extent_buffer(path->nodes[*level]);
2381                         path->nodes[*level] = NULL;
2382                         *level = i + 1;
2383                 }
2384         }
2385         return 1;
2386 }
2387
2388 /*
2389  * drop the reference count on the tree rooted at 'snap'.  This traverses
2390  * the tree freeing any blocks that have a ref count of zero after being
2391  * decremented.
2392  */
2393 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2394                         *root)
2395 {
2396         int ret = 0;
2397         int wret;
2398         int level;
2399         struct btrfs_path *path;
2400         int i;
2401         int orig_level;
2402         struct btrfs_root_item *root_item = &root->root_item;
2403
2404         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2405         path = btrfs_alloc_path();
2406         BUG_ON(!path);
2407
2408         level = btrfs_header_level(root->node);
2409         orig_level = level;
2410         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2411                 path->nodes[level] = root->node;
2412                 extent_buffer_get(root->node);
2413                 path->slots[level] = 0;
2414         } else {
2415                 struct btrfs_key key;
2416                 struct btrfs_disk_key found_key;
2417                 struct extent_buffer *node;
2418
2419                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2420                 level = root_item->drop_level;
2421                 path->lowest_level = level;
2422                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2423                 if (wret < 0) {
2424                         ret = wret;
2425                         goto out;
2426                 }
2427                 node = path->nodes[level];
2428                 btrfs_node_key(node, &found_key, path->slots[level]);
2429                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2430                                sizeof(found_key)));
2431                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2432                         if (path->nodes[i] && path->locks[i]) {
2433                                 path->locks[i] = 0;
2434                                 btrfs_tree_unlock(path->nodes[i]);
2435                         }
2436                 }
2437         }
2438         while(1) {
2439                 wret = walk_down_tree(trans, root, path, &level);
2440                 if (wret > 0)
2441                         break;
2442                 if (wret < 0)
2443                         ret = wret;
2444
2445                 wret = walk_up_tree(trans, root, path, &level);
2446                 if (wret > 0)
2447                         break;
2448                 if (wret < 0)
2449                         ret = wret;
2450                 if (trans->transaction->in_commit) {
2451                         ret = -EAGAIN;
2452                         break;
2453                 }
2454         }
2455         for (i = 0; i <= orig_level; i++) {
2456                 if (path->nodes[i]) {
2457                         free_extent_buffer(path->nodes[i]);
2458                         path->nodes[i] = NULL;
2459                 }
2460         }
2461 out:
2462         btrfs_free_path(path);
2463         return ret;
2464 }
2465
2466 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2467 {
2468         u64 start;
2469         u64 end;
2470         u64 ptr;
2471         int ret;
2472
2473         mutex_lock(&info->alloc_mutex);
2474         while(1) {
2475                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2476                                             &start, &end, (unsigned int)-1);
2477                 if (ret)
2478                         break;
2479                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2480                 if (!ret)
2481                         kfree((void *)(unsigned long)ptr);
2482                 clear_extent_bits(&info->block_group_cache, start,
2483                                   end, (unsigned int)-1, GFP_NOFS);
2484         }
2485         while(1) {
2486                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2487                                             &start, &end, EXTENT_DIRTY);
2488                 if (ret)
2489                         break;
2490                 clear_extent_dirty(&info->free_space_cache, start,
2491                                    end, GFP_NOFS);
2492         }
2493         mutex_unlock(&info->alloc_mutex);
2494         return 0;
2495 }
2496
2497 static unsigned long calc_ra(unsigned long start, unsigned long last,
2498                              unsigned long nr)
2499 {
2500         return min(last, start + nr - 1);
2501 }
2502
2503 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2504                                          u64 len)
2505 {
2506         u64 page_start;
2507         u64 page_end;
2508         unsigned long last_index;
2509         unsigned long i;
2510         struct page *page;
2511         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2512         struct file_ra_state *ra;
2513         unsigned long total_read = 0;
2514         unsigned long ra_pages;
2515         struct btrfs_trans_handle *trans;
2516
2517         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2518
2519         mutex_lock(&inode->i_mutex);
2520         i = start >> PAGE_CACHE_SHIFT;
2521         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2522
2523         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2524
2525         file_ra_state_init(ra, inode->i_mapping);
2526
2527         for (; i <= last_index; i++) {
2528                 if (total_read % ra_pages == 0) {
2529                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2530                                        calc_ra(i, last_index, ra_pages));
2531                 }
2532                 total_read++;
2533                 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2534                         goto truncate_racing;
2535
2536                 page = grab_cache_page(inode->i_mapping, i);
2537                 if (!page) {
2538                         goto out_unlock;
2539                 }
2540                 if (!PageUptodate(page)) {
2541                         btrfs_readpage(NULL, page);
2542                         lock_page(page);
2543                         if (!PageUptodate(page)) {
2544                                 unlock_page(page);
2545                                 page_cache_release(page);
2546                                 goto out_unlock;
2547                         }
2548                 }
2549 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2550                 ClearPageDirty(page);
2551 #else
2552                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2553 #endif
2554                 wait_on_page_writeback(page);
2555                 set_page_extent_mapped(page);
2556                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2557                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2558
2559                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2560
2561                 set_extent_delalloc(io_tree, page_start,
2562                                     page_end, GFP_NOFS);
2563                 set_page_dirty(page);
2564
2565                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2566                 unlock_page(page);
2567                 page_cache_release(page);
2568         }
2569         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2570                                            total_read);
2571
2572 out_unlock:
2573         kfree(ra);
2574         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2575         if (trans) {
2576                 btrfs_add_ordered_inode(inode);
2577                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2578                 mark_inode_dirty(inode);
2579         }
2580         mutex_unlock(&inode->i_mutex);
2581         return 0;
2582
2583 truncate_racing:
2584         vmtruncate(inode, inode->i_size);
2585         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2586                                            total_read);
2587         goto out_unlock;
2588 }
2589
2590 /*
2591  * The back references tell us which tree holds a ref on a block,
2592  * but it is possible for the tree root field in the reference to
2593  * reflect the original root before a snapshot was made.  In this
2594  * case we should search through all the children of a given root
2595  * to find potential holders of references on a block.
2596  *
2597  * Instead, we do something a little less fancy and just search
2598  * all the roots for a given key/block combination.
2599  */
2600 static int find_root_for_ref(struct btrfs_root *root,
2601                              struct btrfs_path *path,
2602                              struct btrfs_key *key0,
2603                              int level,
2604                              int file_key,
2605                              struct btrfs_root **found_root,
2606                              u64 bytenr)
2607 {
2608         struct btrfs_key root_location;
2609         struct btrfs_root *cur_root = *found_root;
2610         struct btrfs_file_extent_item *file_extent;
2611         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2612         u64 found_bytenr;
2613         int ret;
2614         int i;
2615
2616         root_location.offset = (u64)-1;
2617         root_location.type = BTRFS_ROOT_ITEM_KEY;
2618         path->lowest_level = level;
2619         path->reada = 0;
2620         while(1) {
2621                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2622                 found_bytenr = 0;
2623                 if (ret == 0 && file_key) {
2624                         struct extent_buffer *leaf = path->nodes[0];
2625                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
2626                                              struct btrfs_file_extent_item);
2627                         if (btrfs_file_extent_type(leaf, file_extent) ==
2628                             BTRFS_FILE_EXTENT_REG) {
2629                                 found_bytenr =
2630                                         btrfs_file_extent_disk_bytenr(leaf,
2631                                                                file_extent);
2632                        }
2633                 } else if (!file_key) {
2634                         if (path->nodes[level])
2635                                 found_bytenr = path->nodes[level]->start;
2636                 }
2637
2638                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2639                         if (!path->nodes[i])
2640                                 break;
2641                         free_extent_buffer(path->nodes[i]);
2642                         path->nodes[i] = NULL;
2643                 }
2644                 btrfs_release_path(cur_root, path);
2645
2646                 if (found_bytenr == bytenr) {
2647                         *found_root = cur_root;
2648                         ret = 0;
2649                         goto out;
2650                 }
2651                 ret = btrfs_search_root(root->fs_info->tree_root,
2652                                         root_search_start, &root_search_start);
2653                 if (ret)
2654                         break;
2655
2656                 root_location.objectid = root_search_start;
2657                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2658                                                       &root_location);
2659                 if (!cur_root) {
2660                         ret = 1;
2661                         break;
2662                 }
2663         }
2664 out:
2665         path->lowest_level = 0;
2666         return ret;
2667 }
2668
2669 /*
2670  * note, this releases the path
2671  */
2672 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2673                                   struct btrfs_path *path,
2674                                   struct btrfs_key *extent_key,
2675                                   u64 *last_file_objectid,
2676                                   u64 *last_file_offset,
2677                                   u64 *last_file_root,
2678                                   u64 last_extent)
2679 {
2680         struct inode *inode;
2681         struct btrfs_root *found_root;
2682         struct btrfs_key root_location;
2683         struct btrfs_key found_key;
2684         struct btrfs_extent_ref *ref;
2685         u64 ref_root;
2686         u64 ref_gen;
2687         u64 ref_objectid;
2688         u64 ref_offset;
2689         int ret;
2690         int level;
2691
2692         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2693                              struct btrfs_extent_ref);
2694         ref_root = btrfs_ref_root(path->nodes[0], ref);
2695         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2696         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2697         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2698         btrfs_release_path(extent_root, path);
2699
2700         root_location.objectid = ref_root;
2701         if (ref_gen == 0)
2702                 root_location.offset = 0;
2703         else
2704                 root_location.offset = (u64)-1;
2705         root_location.type = BTRFS_ROOT_ITEM_KEY;
2706
2707         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2708                                                 &root_location);
2709         BUG_ON(!found_root);
2710
2711         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2712                 found_key.objectid = ref_objectid;
2713                 found_key.type = BTRFS_EXTENT_DATA_KEY;
2714                 found_key.offset = ref_offset;
2715                 level = 0;
2716
2717                 if (last_extent == extent_key->objectid &&
2718                     *last_file_objectid == ref_objectid &&
2719                     *last_file_offset == ref_offset &&
2720                     *last_file_root == ref_root)
2721                         goto out;
2722
2723                 ret = find_root_for_ref(extent_root, path, &found_key,
2724                                         level, 1, &found_root,
2725                                         extent_key->objectid);
2726
2727                 if (ret)
2728                         goto out;
2729
2730                 if (last_extent == extent_key->objectid &&
2731                     *last_file_objectid == ref_objectid &&
2732                     *last_file_offset == ref_offset &&
2733                     *last_file_root == ref_root)
2734                         goto out;
2735
2736                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2737                                           ref_objectid, found_root);
2738                 if (inode->i_state & I_NEW) {
2739                         /* the inode and parent dir are two different roots */
2740                         BTRFS_I(inode)->root = found_root;
2741                         BTRFS_I(inode)->location.objectid = ref_objectid;
2742                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2743                         BTRFS_I(inode)->location.offset = 0;
2744                         btrfs_read_locked_inode(inode);
2745                         unlock_new_inode(inode);
2746
2747                 }
2748                 /* this can happen if the reference is not against
2749                  * the latest version of the tree root
2750                  */
2751                 if (is_bad_inode(inode)) {
2752                         goto out;
2753                 }
2754                 *last_file_objectid = inode->i_ino;
2755                 *last_file_root = found_root->root_key.objectid;
2756                 *last_file_offset = ref_offset;
2757
2758                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2759                 iput(inode);
2760         } else {
2761                 struct btrfs_trans_handle *trans;
2762                 struct extent_buffer *eb;
2763                 int i;
2764
2765                 eb = read_tree_block(found_root, extent_key->objectid,
2766                                      extent_key->offset, 0);
2767                 btrfs_tree_lock(eb);
2768                 level = btrfs_header_level(eb);
2769
2770                 if (level == 0)
2771                         btrfs_item_key_to_cpu(eb, &found_key, 0);
2772                 else
2773                         btrfs_node_key_to_cpu(eb, &found_key, 0);
2774
2775                 btrfs_tree_unlock(eb);
2776                 free_extent_buffer(eb);
2777
2778                 ret = find_root_for_ref(extent_root, path, &found_key,
2779                                         level, 0, &found_root,
2780                                         extent_key->objectid);
2781
2782                 if (ret)
2783                         goto out;
2784
2785                 trans = btrfs_start_transaction(found_root, 1);
2786
2787                 path->lowest_level = level;
2788                 path->reada = 2;
2789                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2790                                         0, 1);
2791                 path->lowest_level = 0;
2792                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2793                         if (!path->nodes[i])
2794                                 break;
2795                         free_extent_buffer(path->nodes[i]);
2796                         path->nodes[i] = NULL;
2797                 }
2798                 btrfs_release_path(found_root, path);
2799                 if (found_root == found_root->fs_info->extent_root)
2800                         btrfs_extent_post_op(trans, found_root);
2801                 btrfs_end_transaction(trans, found_root);
2802         }
2803
2804 out:
2805         return 0;
2806 }
2807
2808 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2809                                     struct btrfs_path *path,
2810                                     struct btrfs_key *extent_key)
2811 {
2812         int ret;
2813         struct btrfs_trans_handle *trans;
2814
2815         trans = btrfs_start_transaction(extent_root, 1);
2816         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2817         if (ret > 0) {
2818                 ret = -EIO;
2819                 goto out;
2820         }
2821         if (ret < 0)
2822                 goto out;
2823         ret = btrfs_del_item(trans, extent_root, path);
2824 out:
2825         btrfs_end_transaction(trans, extent_root);
2826         return ret;
2827 }
2828
2829 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2830                                         struct btrfs_path *path,
2831                                         struct btrfs_key *extent_key)
2832 {
2833         struct btrfs_key key;
2834         struct btrfs_key found_key;
2835         struct extent_buffer *leaf;
2836         u64 last_file_objectid = 0;
2837         u64 last_file_root = 0;
2838         u64 last_file_offset = (u64)-1;
2839         u64 last_extent = 0;
2840         u32 nritems;
2841         u32 item_size;
2842         int ret = 0;
2843
2844         if (extent_key->objectid == 0) {
2845                 ret = del_extent_zero(extent_root, path, extent_key);
2846                 goto out;
2847         }
2848         key.objectid = extent_key->objectid;
2849         key.type = BTRFS_EXTENT_REF_KEY;
2850         key.offset = 0;
2851
2852         while(1) {
2853                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2854
2855                 if (ret < 0)
2856                         goto out;
2857
2858                 ret = 0;
2859                 leaf = path->nodes[0];
2860                 nritems = btrfs_header_nritems(leaf);
2861                 if (path->slots[0] == nritems) {
2862                         ret = btrfs_next_leaf(extent_root, path);
2863                         if (ret > 0) {
2864                                 ret = 0;
2865                                 goto out;
2866                         }
2867                         if (ret < 0)
2868                                 goto out;
2869                         leaf = path->nodes[0];
2870                 }
2871
2872                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2873                 if (found_key.objectid != extent_key->objectid) {
2874                         break;
2875                 }
2876
2877                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2878                         break;
2879                 }
2880
2881                 key.offset = found_key.offset + 1;
2882                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2883
2884                 ret = relocate_one_reference(extent_root, path, extent_key,
2885                                              &last_file_objectid,
2886                                              &last_file_offset,
2887                                              &last_file_root, last_extent);
2888                 if (ret)
2889                         goto out;
2890                 last_extent = extent_key->objectid;
2891         }
2892         ret = 0;
2893 out:
2894         btrfs_release_path(extent_root, path);
2895         return ret;
2896 }
2897
2898 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2899 {
2900         u64 num_devices;
2901         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2902                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2903
2904         num_devices = root->fs_info->fs_devices->num_devices;
2905         if (num_devices == 1) {
2906                 stripped |= BTRFS_BLOCK_GROUP_DUP;
2907                 stripped = flags & ~stripped;
2908
2909                 /* turn raid0 into single device chunks */
2910                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2911                         return stripped;
2912
2913                 /* turn mirroring into duplication */
2914                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2915                              BTRFS_BLOCK_GROUP_RAID10))
2916                         return stripped | BTRFS_BLOCK_GROUP_DUP;
2917                 return flags;
2918         } else {
2919                 /* they already had raid on here, just return */
2920                 if (flags & stripped)
2921                         return flags;
2922
2923                 stripped |= BTRFS_BLOCK_GROUP_DUP;
2924                 stripped = flags & ~stripped;
2925
2926                 /* switch duplicated blocks with raid1 */
2927                 if (flags & BTRFS_BLOCK_GROUP_DUP)
2928                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
2929
2930                 /* turn single device chunks into raid0 */
2931                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2932         }
2933         return flags;
2934 }
2935
2936 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2937                      struct btrfs_block_group_cache *shrink_block_group,
2938                      int force)
2939 {
2940         struct btrfs_trans_handle *trans;
2941         u64 new_alloc_flags;
2942         u64 calc;
2943
2944         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2945
2946                 trans = btrfs_start_transaction(root, 1);
2947                 new_alloc_flags = update_block_group_flags(root,
2948                                                    shrink_block_group->flags);
2949                 if (new_alloc_flags != shrink_block_group->flags) {
2950                         calc =
2951                              btrfs_block_group_used(&shrink_block_group->item);
2952                 } else {
2953                         calc = shrink_block_group->key.offset;
2954                 }
2955                 do_chunk_alloc(trans, root->fs_info->extent_root,
2956                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
2957                 btrfs_end_transaction(trans, root);
2958         }
2959         return 0;
2960 }
2961
2962 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2963 {
2964         struct btrfs_trans_handle *trans;
2965         struct btrfs_root *tree_root = root->fs_info->tree_root;
2966         struct btrfs_path *path;
2967         u64 cur_byte;
2968         u64 total_found;
2969         u64 shrink_last_byte;
2970         struct btrfs_block_group_cache *shrink_block_group;
2971         struct btrfs_fs_info *info = root->fs_info;
2972         struct btrfs_key key;
2973         struct btrfs_key found_key;
2974         struct extent_buffer *leaf;
2975         u32 nritems;
2976         int ret;
2977         int progress;
2978
2979         mutex_lock(&root->fs_info->alloc_mutex);
2980         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2981                                                       shrink_start);
2982         BUG_ON(!shrink_block_group);
2983
2984         shrink_last_byte = shrink_block_group->key.objectid +
2985                 shrink_block_group->key.offset;
2986
2987         shrink_block_group->space_info->total_bytes -=
2988                 shrink_block_group->key.offset;
2989         path = btrfs_alloc_path();
2990         root = root->fs_info->extent_root;
2991         path->reada = 2;
2992
2993         printk("btrfs relocating block group %llu flags %llu\n",
2994                (unsigned long long)shrink_start,
2995                (unsigned long long)shrink_block_group->flags);
2996
2997         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
2998
2999 again:
3000
3001         shrink_block_group->ro = 1;
3002
3003         total_found = 0;
3004         progress = 0;
3005         key.objectid = shrink_start;
3006         key.offset = 0;
3007         key.type = 0;
3008         cur_byte = key.objectid;
3009
3010         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3011         if (ret < 0)
3012                 goto out;
3013
3014         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3015         if (ret < 0)
3016                 goto out;
3017
3018         if (ret == 0) {
3019                 leaf = path->nodes[0];
3020                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3021                 if (found_key.objectid + found_key.offset > shrink_start &&
3022                     found_key.objectid < shrink_last_byte) {
3023                         cur_byte = found_key.objectid;
3024                         key.objectid = cur_byte;
3025                 }
3026         }
3027         btrfs_release_path(root, path);
3028
3029         while(1) {
3030                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3031                 if (ret < 0)
3032                         goto out;
3033
3034                 leaf = path->nodes[0];
3035                 nritems = btrfs_header_nritems(leaf);
3036 next:
3037                 if (path->slots[0] >= nritems) {
3038                         ret = btrfs_next_leaf(root, path);
3039                         if (ret < 0)
3040                                 goto out;
3041                         if (ret == 1) {
3042                                 ret = 0;
3043                                 break;
3044                         }
3045                         leaf = path->nodes[0];
3046                         nritems = btrfs_header_nritems(leaf);
3047                 }
3048
3049                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3050
3051                 if (found_key.objectid >= shrink_last_byte)
3052                         break;
3053
3054                 if (progress && need_resched()) {
3055                         memcpy(&key, &found_key, sizeof(key));
3056                         cond_resched();
3057                         btrfs_release_path(root, path);
3058                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3059                         progress = 0;
3060                         goto next;
3061                 }
3062                 progress = 1;
3063
3064                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3065                     found_key.objectid + found_key.offset <= cur_byte) {
3066                         memcpy(&key, &found_key, sizeof(key));
3067                         key.offset++;
3068                         path->slots[0]++;
3069                         goto next;
3070                 }
3071
3072                 total_found++;
3073                 cur_byte = found_key.objectid + found_key.offset;
3074                 key.objectid = cur_byte;
3075                 btrfs_release_path(root, path);
3076                 ret = relocate_one_extent(root, path, &found_key);
3077                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3078         }
3079
3080         btrfs_release_path(root, path);
3081
3082         if (total_found > 0) {
3083                 printk("btrfs relocate found %llu last extent was %llu\n",
3084                        (unsigned long long)total_found,
3085                        (unsigned long long)found_key.objectid);
3086                 trans = btrfs_start_transaction(tree_root, 1);
3087                 btrfs_commit_transaction(trans, tree_root);
3088
3089                 btrfs_clean_old_snapshots(tree_root);
3090
3091                 trans = btrfs_start_transaction(tree_root, 1);
3092                 btrfs_commit_transaction(trans, tree_root);
3093                 goto again;
3094         }
3095
3096         /*
3097          * we've freed all the extents, now remove the block
3098          * group item from the tree
3099          */
3100         trans = btrfs_start_transaction(root, 1);
3101         memcpy(&key, &shrink_block_group->key, sizeof(key));
3102
3103         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3104         if (ret > 0)
3105                 ret = -EIO;
3106         if (ret < 0)
3107                 goto out;
3108
3109         clear_extent_bits(&info->block_group_cache, key.objectid,
3110                           key.objectid + key.offset - 1,
3111                           (unsigned int)-1, GFP_NOFS);
3112
3113
3114         clear_extent_bits(&info->free_space_cache,
3115                            key.objectid, key.objectid + key.offset - 1,
3116                            (unsigned int)-1, GFP_NOFS);
3117
3118         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3119         kfree(shrink_block_group);
3120
3121         btrfs_del_item(trans, root, path);
3122         btrfs_commit_transaction(trans, root);
3123
3124         /* the code to unpin extents might set a few bits in the free
3125          * space cache for this range again
3126          */
3127         clear_extent_bits(&info->free_space_cache,
3128                            key.objectid, key.objectid + key.offset - 1,
3129                            (unsigned int)-1, GFP_NOFS);
3130 out:
3131         btrfs_free_path(path);
3132         mutex_unlock(&root->fs_info->alloc_mutex);
3133         return ret;
3134 }
3135
3136 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3137                            struct btrfs_key *key)
3138 {
3139         int ret = 0;
3140         struct btrfs_key found_key;
3141         struct extent_buffer *leaf;
3142         int slot;
3143
3144         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3145         if (ret < 0)
3146                 goto out;
3147
3148         while(1) {
3149                 slot = path->slots[0];
3150                 leaf = path->nodes[0];
3151                 if (slot >= btrfs_header_nritems(leaf)) {
3152                         ret = btrfs_next_leaf(root, path);
3153                         if (ret == 0)
3154                                 continue;
3155                         if (ret < 0)
3156                                 goto out;
3157                         break;
3158                 }
3159                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3160
3161                 if (found_key.objectid >= key->objectid &&
3162                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3163                         ret = 0;
3164                         goto out;
3165                 }
3166                 path->slots[0]++;
3167         }
3168         ret = -ENOENT;
3169 out:
3170         return ret;
3171 }
3172
3173 int btrfs_read_block_groups(struct btrfs_root *root)
3174 {
3175         struct btrfs_path *path;
3176         int ret;
3177         int bit;
3178         struct btrfs_block_group_cache *cache;
3179         struct btrfs_fs_info *info = root->fs_info;
3180         struct btrfs_space_info *space_info;
3181         struct extent_io_tree *block_group_cache;
3182         struct btrfs_key key;
3183         struct btrfs_key found_key;
3184         struct extent_buffer *leaf;
3185
3186         block_group_cache = &info->block_group_cache;
3187         root = info->extent_root;
3188         key.objectid = 0;
3189         key.offset = 0;
3190         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3191         path = btrfs_alloc_path();
3192         if (!path)
3193                 return -ENOMEM;
3194
3195         mutex_lock(&root->fs_info->alloc_mutex);
3196         while(1) {
3197                 ret = find_first_block_group(root, path, &key);
3198                 if (ret > 0) {
3199                         ret = 0;
3200                         goto error;
3201                 }
3202                 if (ret != 0)
3203                         goto error;
3204
3205                 leaf = path->nodes[0];
3206                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3207                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3208                 if (!cache) {
3209                         ret = -ENOMEM;
3210                         break;
3211                 }
3212
3213                 read_extent_buffer(leaf, &cache->item,
3214                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3215                                    sizeof(cache->item));
3216                 memcpy(&cache->key, &found_key, sizeof(found_key));
3217
3218                 key.objectid = found_key.objectid + found_key.offset;
3219                 btrfs_release_path(root, path);
3220                 cache->flags = btrfs_block_group_flags(&cache->item);
3221                 bit = 0;
3222                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3223                         bit = BLOCK_GROUP_DATA;
3224                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3225                         bit = BLOCK_GROUP_SYSTEM;
3226                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3227                         bit = BLOCK_GROUP_METADATA;
3228                 }
3229                 set_avail_alloc_bits(info, cache->flags);
3230
3231                 ret = update_space_info(info, cache->flags, found_key.offset,
3232                                         btrfs_block_group_used(&cache->item),
3233                                         &space_info);
3234                 BUG_ON(ret);
3235                 cache->space_info = space_info;
3236
3237                 /* use EXTENT_LOCKED to prevent merging */
3238                 set_extent_bits(block_group_cache, found_key.objectid,
3239                                 found_key.objectid + found_key.offset - 1,
3240                                 bit | EXTENT_LOCKED, GFP_NOFS);
3241                 set_state_private(block_group_cache, found_key.objectid,
3242                                   (unsigned long)cache);
3243
3244                 if (key.objectid >=
3245                     btrfs_super_total_bytes(&info->super_copy))
3246                         break;
3247         }
3248         ret = 0;
3249 error:
3250         btrfs_free_path(path);
3251         mutex_unlock(&root->fs_info->alloc_mutex);
3252         return ret;
3253 }
3254
3255 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3256                            struct btrfs_root *root, u64 bytes_used,
3257                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3258                            u64 size)
3259 {
3260         int ret;
3261         int bit = 0;
3262         struct btrfs_root *extent_root;
3263         struct btrfs_block_group_cache *cache;
3264         struct extent_io_tree *block_group_cache;
3265
3266         extent_root = root->fs_info->extent_root;
3267         block_group_cache = &root->fs_info->block_group_cache;
3268
3269         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3270         BUG_ON(!cache);
3271         cache->key.objectid = chunk_offset;
3272         cache->key.offset = size;
3273         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3274
3275         btrfs_set_block_group_used(&cache->item, bytes_used);
3276         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3277         cache->flags = type;
3278         btrfs_set_block_group_flags(&cache->item, type);
3279
3280         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3281                                 &cache->space_info);
3282         BUG_ON(ret);
3283
3284         bit = block_group_state_bits(type);
3285         set_extent_bits(block_group_cache, chunk_offset,
3286                         chunk_offset + size - 1,
3287                         bit | EXTENT_LOCKED, GFP_NOFS);
3288
3289         set_state_private(block_group_cache, chunk_offset,
3290                           (unsigned long)cache);
3291         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3292                                 sizeof(cache->item));
3293         BUG_ON(ret);
3294
3295         finish_current_insert(trans, extent_root);
3296         ret = del_pending_extents(trans, extent_root);
3297         BUG_ON(ret);
3298         set_avail_alloc_bits(extent_root->fs_info, type);
3299
3300         return 0;
3301 }