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
  19 #include <linux/sched.h>
  20 #include <linux/crc32c.h>
  21 #include "hash.h"
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "print-tree.h"
  25 #include "transaction.h"
  26
  27 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
  28 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  29 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  30
  31 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  32                                  btrfs_root *extent_root);
  33 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  34                                btrfs_root *extent_root);
  35
  36 static int cache_block_group(struct btrfs_root *root,
  37                              struct btrfs_block_group_cache *block_group)
  38 {
  39         struct btrfs_path *path;
  40         int ret;
  41         struct btrfs_key key;
  42         struct extent_buffer *leaf;
  43         struct extent_map_tree *free_space_cache;
  44         int slot;
  45         u64 last = 0;
  46         u64 hole_size;
  47         u64 first_free;
  48         int found = 0;
  49
  50         if (!block_group)
  51                 return 0;
  52
  53         root = root->fs_info->extent_root;
  54         free_space_cache = &root->fs_info->free_space_cache;
  55
  56         if (block_group->cached)
  57                 return 0;
  58
  59         path = btrfs_alloc_path();
  60         if (!path)
  61                 return -ENOMEM;
  62
  63         path->reada = 2;
  64         first_free = block_group->key.objectid;
  65         key.objectid = block_group->key.objectid;
  66         key.offset = 0;
  67
  68         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  69         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  70
  71         if (ret < 0)
  72                 return ret;
  73
  74         if (ret && path->slots[0] > 0)
  75                 path->slots[0]--;
  76
  77         while(1) {
  78                 leaf = path->nodes[0];
  79                 slot = path->slots[0];
  80                 if (slot >= btrfs_header_nritems(leaf)) {
  81                         ret = btrfs_next_leaf(root, path);
  82                         if (ret < 0)
  83                                 goto err;
  84                         if (ret == 0) {
  85                                 continue;
  86                         } else {
  87                                 break;
  88                         }
  89                 }
  90
  91                 btrfs_item_key_to_cpu(leaf, &key, slot);
  92                 if (key.objectid < block_group->key.objectid) {
  93                         if (btrfs_key_type(&key) != BTRFS_EXTENT_REF_KEY &&
  94                             key.objectid + key.offset > first_free)
  95                                 first_free = key.objectid + key.offset;
  96                         goto next;
  97                 }
  98
  99                 if (key.objectid >= block_group->key.objectid +
 100                     block_group->key.offset) {
 101                         break;
 102                 }
 103
 104                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 105                         if (!found) {
 106                                 last = first_free;
 107                                 found = 1;
 108                         }
 109                         if (key.objectid > last) {
 110                                 hole_size = key.objectid - last;
 111                                 set_extent_dirty(free_space_cache, last,
 112                                                  last + hole_size - 1,
 113                                                  GFP_NOFS);
 114                         }
 115                         last = key.objectid + key.offset;
 116                 }
 117 next:
 118                 path->slots[0]++;
 119         }
 120
 121         if (!found)
 122                 last = first_free;
 123         if (block_group->key.objectid +
 124             block_group->key.offset > last) {
 125                 hole_size = block_group->key.objectid +
 126                         block_group->key.offset - last;
 127                 set_extent_dirty(free_space_cache, last,
 128                                  last + hole_size - 1, GFP_NOFS);
 129         }
 130         block_group->cached = 1;
 131 err:
 132         btrfs_free_path(path);
 133         return 0;
 134 }
 135
 136 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 137                                                          btrfs_fs_info *info,
 138                                                          u64 bytenr)
 139 {
 140         struct extent_map_tree *block_group_cache;
 141         struct btrfs_block_group_cache *block_group = NULL;
 142         u64 ptr;
 143         u64 start;
 144         u64 end;
 145         int ret;
 146
 147         block_group_cache = &info->block_group_cache;
 148         ret = find_first_extent_bit(block_group_cache,
 149                                     bytenr, &start, &end,
 150                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
 151         if (ret) {
 152                 return NULL;
 153         }
 154         ret = get_state_private(block_group_cache, start, &ptr);
 155         if (ret)
 156                 return NULL;
 157
 158         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 159         if (block_group->key.objectid <= bytenr && bytenr <
 160             block_group->key.objectid + block_group->key.offset)
 161                 return block_group;
 162         return NULL;
 163 }
 164 static u64 find_search_start(struct btrfs_root *root,
 165                              struct btrfs_block_group_cache **cache_ret,
 166                              u64 search_start, int num,
 167                              int data, int full_scan)
 168 {
 169         int ret;
 170         struct btrfs_block_group_cache *cache = *cache_ret;
 171         u64 last;
 172         u64 start = 0;
 173         u64 end = 0;
 174         u64 cache_miss = 0;
 175         int wrapped = 0;
 176
 177         if (!cache) {
 178                 goto out;
 179         }
 180 again:
 181         ret = cache_block_group(root, cache);
 182         if (ret)
 183                 goto out;
 184
 185         last = max(search_start, cache->key.objectid);
 186
 187         while(1) {
 188                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 189                                             last, &start, &end, EXTENT_DIRTY);
 190                 if (ret) {
 191                         if (!cache_miss)
 192                                 cache_miss = last;
 193                         goto new_group;
 194                 }
 195
 196                 start = max(last, start);
 197                 last = end + 1;
 198                 if (last - start < num) {
 199                         if (last == cache->key.objectid + cache->key.offset)
 200                                 cache_miss = start;
 201                         continue;
 202                 }
 203                 if (data != BTRFS_BLOCK_GROUP_MIXED &&
 204                     start + num > cache->key.objectid + cache->key.offset)
 205                         goto new_group;
 206                 return start;
 207         }
 208 out:
 209         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 210         if (!cache) {
 211                 printk("Unable to find block group for %Lu\n",
 212                        search_start);
 213                 WARN_ON(1);
 214                 return search_start;
 215         }
 216         return search_start;
 217
 218 new_group:
 219         last = cache->key.objectid + cache->key.offset;
 220 wrapped:
 221         cache = btrfs_lookup_block_group(root->fs_info, last);
 222         if (!cache) {
 223 no_cache:
 224                 if (!wrapped) {
 225                         wrapped = 1;
 226                         last = search_start;
 227                         data = BTRFS_BLOCK_GROUP_MIXED;
 228                         goto wrapped;
 229                 }
 230                 goto out;
 231         }
 232         if (cache_miss && !cache->cached) {
 233                 cache_block_group(root, cache);
 234                 last = cache_miss;
 235                 cache = btrfs_lookup_block_group(root->fs_info, last);
 236         }
 237         cache = btrfs_find_block_group(root, cache, last, data, 0);
 238         if (!cache)
 239                 goto no_cache;
 240         *cache_ret = cache;
 241         cache_miss = 0;
 242         goto again;
 243 }
 244
 245 static u64 div_factor(u64 num, int factor)
 246 {
 247         if (factor == 10)
 248                 return num;
 249         num *= factor;
 250         do_div(num, 10);
 251         return num;
 252 }
 253
 254 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 255                                                  struct btrfs_block_group_cache
 256                                                  *hint, u64 search_start,
 257                                                  int data, int owner)
 258 {
 259         struct btrfs_block_group_cache *cache;
 260         struct extent_map_tree *block_group_cache;
 261         struct btrfs_block_group_cache *found_group = NULL;
 262         struct btrfs_fs_info *info = root->fs_info;
 263         u64 used;
 264         u64 last = 0;
 265         u64 hint_last;
 266         u64 start;
 267         u64 end;
 268         u64 free_check;
 269         u64 ptr;
 270         int bit;
 271         int ret;
 272         int full_search = 0;
 273         int factor = 8;
 274         int data_swap = 0;
 275
 276         block_group_cache = &info->block_group_cache;
 277
 278         if (!owner)
 279                 factor = 8;
 280
 281         if (data == BTRFS_BLOCK_GROUP_MIXED) {
 282                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 283                 factor = 10;
 284         } else if (data)
 285                 bit = BLOCK_GROUP_DATA;
 286         else
 287                 bit = BLOCK_GROUP_METADATA;
 288
 289         if (search_start) {
 290                 struct btrfs_block_group_cache *shint;
 291                 shint = btrfs_lookup_block_group(info, search_start);
 292                 if (shint && (shint->data == data ||
 293                               shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 294                         used = btrfs_block_group_used(&shint->item);
 295                         if (used + shint->pinned <
 296                             div_factor(shint->key.offset, factor)) {
 297                                 return shint;
 298                         }
 299                 }
 300         }
 301         if (hint && (hint->data == data ||
 302                      hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 303                 used = btrfs_block_group_used(&hint->item);
 304                 if (used + hint->pinned <
 305                     div_factor(hint->key.offset, factor)) {
 306                         return hint;
 307                 }
 308                 last = hint->key.objectid + hint->key.offset;
 309                 hint_last = last;
 310         } else {
 311                 if (hint)
 312                         hint_last = max(hint->key.objectid, search_start);
 313                 else
 314                         hint_last = search_start;
 315
 316                 last = hint_last;
 317         }
 318 again:
 319         while(1) {
 320                 ret = find_first_extent_bit(block_group_cache, last,
 321                                             &start, &end, bit);
 322                 if (ret)
 323                         break;
 324
 325                 ret = get_state_private(block_group_cache, start, &ptr);
 326                 if (ret)
 327                         break;
 328
 329                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 330                 last = cache->key.objectid + cache->key.offset;
 331                 used = btrfs_block_group_used(&cache->item);
 332
 333                 if (full_search)
 334                         free_check = cache->key.offset;
 335                 else
 336                         free_check = div_factor(cache->key.offset, factor);
 337                 if (used + cache->pinned < free_check) {
 338                         found_group = cache;
 339                         goto found;
 340                 }
 341                 cond_resched();
 342         }
 343         if (!full_search) {
 344                 last = search_start;
 345                 full_search = 1;
 346                 goto again;
 347         }
 348         if (!data_swap) {
 349                 data_swap = 1;
 350                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 351                 last = search_start;
 352                 goto again;
 353         }
 354 found:
 355         return found_group;
 356 }
 357
 358 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 359                            u64 owner, u64 owner_offset)
 360 {
 361         u32 high_crc = ~(u32)0;
 362         u32 low_crc = ~(u32)0;
 363         __le64 lenum;
 364
 365         lenum = cpu_to_le64(root_objectid);
 366         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 367         lenum = cpu_to_le64(ref_generation);
 368         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 369
 370 #if 0
 371         lenum = cpu_to_le64(owner);
 372         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 373         lenum = cpu_to_le64(owner_offset);
 374         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 375 #endif
 376         return ((u64)high_crc << 32) | (u64)low_crc;
 377 }
 378
 379 static int match_extent_ref(struct extent_buffer *leaf,
 380                             struct btrfs_extent_ref *disk_ref,
 381                             struct btrfs_extent_ref *cpu_ref)
 382 {
 383         int ret;
 384         int len;
 385
 386         if (cpu_ref->objectid)
 387                 len = sizeof(*cpu_ref);
 388         else
 389                 len = 2 * sizeof(u64);
 390         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 391                                    len);
 392         return ret == 0;
 393 }
 394
 395 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
 396                                  struct btrfs_root *root,
 397                                  struct btrfs_path *path, u64 bytenr,
 398                                  u64 root_objectid, u64 ref_generation,
 399                                  u64 owner, u64 owner_offset, int del)
 400 {
 401         u64 hash;
 402         struct btrfs_key key;
 403         struct btrfs_key found_key;
 404         struct btrfs_extent_ref ref;
 405         struct extent_buffer *leaf;
 406         struct btrfs_extent_ref *disk_ref;
 407         int ret;
 408         int ret2;
 409
 410         btrfs_set_stack_ref_root(&ref, root_objectid);
 411         btrfs_set_stack_ref_generation(&ref, ref_generation);
 412         btrfs_set_stack_ref_objectid(&ref, owner);
 413         btrfs_set_stack_ref_offset(&ref, owner_offset);
 414
 415         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 416                                owner_offset);
 417         key.offset = hash;
 418         key.objectid = bytenr;
 419         key.type = BTRFS_EXTENT_REF_KEY;
 420
 421         while (1) {
 422                 ret = btrfs_search_slot(trans, root, &key, path,
 423                                         del ? -1 : 0, del);
 424                 if (ret < 0)
 425                         goto out;
 426                 leaf = path->nodes[0];
 427                 if (ret != 0) {
 428                         u32 nritems = btrfs_header_nritems(leaf);
 429                         if (path->slots[0] >= nritems) {
 430                                 ret2 = btrfs_next_leaf(root, path);
 431                                 if (ret2)
 432                                         goto out;
 433                                 leaf = path->nodes[0];
 434                         }
 435                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 436                         if (found_key.objectid != bytenr ||
 437                             found_key.type != BTRFS_EXTENT_REF_KEY)
 438                                 goto out;
 439                         key.offset = found_key.offset;
 440                         if (del) {
 441                                 btrfs_release_path(root, path);
 442                                 continue;
 443                         }
 444                 }
 445                 disk_ref = btrfs_item_ptr(path->nodes[0],
 446                                           path->slots[0],
 447                                           struct btrfs_extent_ref);
 448                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 449                         ret = 0;
 450                         goto out;
 451                 }
 452                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 453                 key.offset = found_key.offset + 1;
 454                 btrfs_release_path(root, path);
 455         }
 456 out:
 457         return ret;
 458 }
 459
 460 /*
 461  * Back reference rules.  Back refs have three main goals:
 462  *
 463  * 1) differentiate between all holders of references to an extent so that
 464  *    when a reference is dropped we can make sure it was a valid reference
 465  *    before freeing the extent.
 466  *
 467  * 2) Provide enough information to quickly find the holders of an extent
 468  *    if we notice a given block is corrupted or bad.
 469  *
 470  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 471  *    maintenance.  This is actually the same as #2, but with a slightly
 472  *    different use case.
 473  *
 474  * File extents can be referenced by:
 475  *
 476  * - multiple snapshots, subvolumes, or different generations in one subvol
 477  * - different files inside a single subvolume (in theory, not implemented yet)
 478  * - different offsets inside a file (bookend extents in file.c)
 479  *
 480  * The extent ref structure has fields for:
 481  *
 482  * - Objectid of the subvolume root
 483  * - Generation number of the tree holding the reference
 484  * - objectid of the file holding the reference
 485  * - offset in the file corresponding to the key holding the reference
 486  *
 487  * When a file extent is allocated the fields are filled in:
 488  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 489  *
 490  * When a leaf is cow'd new references are added for every file extent found
 491  * in the leaf.  It looks the same as the create case, but trans->transid
 492  * will be different when the block is cow'd.
 493  *
 494  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 495  *
 496  * When a file extent is removed either during snapshot deletion or file
 497  * truncation, the corresponding back reference is found
 498  * by searching for:
 499  *
 500  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 501  *      inode objectid, offset in file)
 502  *
 503  * Btree extents can be referenced by:
 504  *
 505  * - Different subvolumes
 506  * - Different generations of the same subvolume
 507  *
 508  * Storing sufficient information for a full reverse mapping of a btree
 509  * block would require storing the lowest key of the block in the backref,
 510  * and it would require updating that lowest key either before write out or
 511  * every time it changed.  Instead, the objectid of the lowest key is stored
 512  * along with the level of the tree block.  This provides a hint
 513  * about where in the btree the block can be found.  Searches through the
 514  * btree only need to look for a pointer to that block, so they stop one
 515  * level higher than the level recorded in the backref.
 516  *
 517  * Some btrees do not do reference counting on their extents.  These
 518  * include the extent tree and the tree of tree roots.  Backrefs for these
 519  * trees always have a generation of zero.
 520  *
 521  * When a tree block is created, back references are inserted:
 522  *
 523  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 524  *
 525  * When a tree block is cow'd in a reference counted root,
 526  * new back references are added for all the blocks it points to.
 527  * These are of the form (trans->transid will have increased since creation):
 528  *
 529  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 530  *
 531  * Because the lowest_key_objectid and the level are just hints
 532  * they are not used when backrefs are deleted.  When a backref is deleted:
 533  *
 534  * if backref was for a tree root:
 535  *     root_objectid = root->root_key.objectid
 536  * else
 537  *     root_objectid = btrfs_header_owner(parent)
 538  *
 539  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 540  *
 541  * Back Reference Key hashing:
 542  *
 543  * Back references have four fields, each 64 bits long.  Unfortunately,
 544  * This is hashed into a single 64 bit number and placed into the key offset.
 545  * The key objectid corresponds to the first byte in the extent, and the
 546  * key type is set to BTRFS_EXTENT_REF_KEY
 547  */
 548 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 549                                  struct btrfs_root *root,
 550                                  struct btrfs_path *path, u64 bytenr,
 551                                  u64 root_objectid, u64 ref_generation,
 552                                  u64 owner, u64 owner_offset)
 553 {
 554         u64 hash;
 555         struct btrfs_key key;
 556         struct btrfs_extent_ref ref;
 557         struct btrfs_extent_ref *disk_ref;
 558         int ret;
 559
 560         btrfs_set_stack_ref_root(&ref, root_objectid);
 561         btrfs_set_stack_ref_generation(&ref, ref_generation);
 562         btrfs_set_stack_ref_objectid(&ref, owner);
 563         btrfs_set_stack_ref_offset(&ref, owner_offset);
 564
 565         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 566                                owner_offset);
 567         key.offset = hash;
 568         key.objectid = bytenr;
 569         key.type = BTRFS_EXTENT_REF_KEY;
 570
 571         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 572         while (ret == -EEXIST) {
 573                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 574                                           struct btrfs_extent_ref);
 575                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 576                         goto out;
 577                 key.offset++;
 578                 btrfs_release_path(root, path);
 579                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 580                                               sizeof(ref));
 581         }
 582         if (ret)
 583                 goto out;
 584         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 585                                   struct btrfs_extent_ref);
 586         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 587                             sizeof(ref));
 588         btrfs_mark_buffer_dirty(path->nodes[0]);
 589 out:
 590         btrfs_release_path(root, path);
 591         return ret;
 592 }
 593
 594 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 595                                 struct btrfs_root *root,
 596                                 u64 bytenr, u64 num_bytes,
 597                                 u64 root_objectid, u64 ref_generation,
 598                                 u64 owner, u64 owner_offset)
 599 {
 600         struct btrfs_path *path;
 601         int ret;
 602         struct btrfs_key key;
 603         struct extent_buffer *l;
 604         struct btrfs_extent_item *item;
 605         u32 refs;
 606
 607         WARN_ON(num_bytes < root->sectorsize);
 608         path = btrfs_alloc_path();
 609         if (!path)
 610                 return -ENOMEM;
 611
 612         key.objectid = bytenr;
 613         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 614         key.offset = num_bytes;
 615         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 616                                 0, 1);
 617         if (ret < 0)
 618                 return ret;
 619         if (ret != 0) {
 620                 BUG();
 621         }
 622         BUG_ON(ret != 0);
 623         l = path->nodes[0];
 624         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 625         refs = btrfs_extent_refs(l, item);
 626         btrfs_set_extent_refs(l, item, refs + 1);
 627         btrfs_mark_buffer_dirty(path->nodes[0]);
 628
 629         btrfs_release_path(root->fs_info->extent_root, path);
 630
 631         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 632                                           path, bytenr, root_objectid,
 633                                           ref_generation, owner, owner_offset);
 634         BUG_ON(ret);
 635         finish_current_insert(trans, root->fs_info->extent_root);
 636         del_pending_extents(trans, root->fs_info->extent_root);
 637
 638         btrfs_free_path(path);
 639         return 0;
 640 }
 641
 642 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 643                          struct btrfs_root *root)
 644 {
 645         finish_current_insert(trans, root->fs_info->extent_root);
 646         del_pending_extents(trans, root->fs_info->extent_root);
 647         return 0;
 648 }
 649
 650 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 651                              struct btrfs_root *root, u64 bytenr,
 652                              u64 num_bytes, u32 *refs)
 653 {
 654         struct btrfs_path *path;
 655         int ret;
 656         struct btrfs_key key;
 657         struct extent_buffer *l;
 658         struct btrfs_extent_item *item;
 659
 660         WARN_ON(num_bytes < root->sectorsize);
 661         path = btrfs_alloc_path();
 662         key.objectid = bytenr;
 663         key.offset = num_bytes;
 664         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 665         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 666                                 0, 0);
 667         if (ret < 0)
 668                 goto out;
 669         if (ret != 0) {
 670                 btrfs_print_leaf(root, path->nodes[0]);
 671                 printk("failed to find block number %Lu\n", bytenr);
 672                 BUG();
 673         }
 674         l = path->nodes[0];
 675         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 676         *refs = btrfs_extent_refs(l, item);
 677 out:
 678         btrfs_free_path(path);
 679         return 0;
 680 }
 681
 682 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 683                                   struct btrfs_path *count_path,
 684                                   u64 first_extent)
 685 {
 686         struct btrfs_root *extent_root = root->fs_info->extent_root;
 687         struct btrfs_path *path;
 688         u64 bytenr;
 689         u64 found_objectid;
 690         u64 root_objectid = 0;
 691         u32 total_count = 0;
 692         u32 cur_count;
 693         u32 refs;
 694         u32 nritems;
 695         int ret;
 696         struct btrfs_key key;
 697         struct btrfs_key found_key;
 698         struct extent_buffer *l;
 699         struct btrfs_extent_item *item;
 700         struct btrfs_extent_ref *ref_item;
 701         int level = -1;
 702
 703         path = btrfs_alloc_path();
 704 again:
 705         if (level == -1)
 706                 bytenr = first_extent;
 707         else
 708                 bytenr = count_path->nodes[level]->start;
 709
 710         cur_count = 0;
 711         key.objectid = bytenr;
 712         key.offset = 0;
 713
 714         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 715         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 716         if (ret < 0)
 717                 goto out;
 718         BUG_ON(ret == 0);
 719
 720         l = path->nodes[0];
 721         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 722
 723         if (found_key.objectid != bytenr ||
 724             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 725                 goto out;
 726         }
 727
 728         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 729         refs = btrfs_extent_refs(l, item);
 730         while (1) {
 731                 nritems = btrfs_header_nritems(l);
 732                 if (path->slots[0] >= nritems) {
 733                         ret = btrfs_next_leaf(extent_root, path);
 734                         if (ret == 0)
 735                                 continue;
 736                         break;
 737                 }
 738                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 739                 if (found_key.objectid != bytenr)
 740                         break;
 741                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 742                         path->slots[0]++;
 743                         continue;
 744                 }
 745
 746                 cur_count++;
 747                 ref_item = btrfs_item_ptr(l, path->slots[0],
 748                                           struct btrfs_extent_ref);
 749                 found_objectid = btrfs_ref_root(l, ref_item);
 750
 751                 if (found_objectid != root_objectid)
 752                         total_count++;
 753
 754                 if (total_count > 1)
 755                         goto out;
 756
 757                 if (root_objectid == 0)
 758                         root_objectid = found_objectid;
 759
 760                 path->slots[0]++;
 761         }
 762         if (cur_count == 0) {
 763                 total_count = 0;
 764                 goto out;
 765         }
 766         if (total_count > 1)
 767                 goto out;
 768         if (level >= 0 && root->node == count_path->nodes[level])
 769                 goto out;
 770         level++;
 771         btrfs_release_path(root, path);
 772         goto again;
 773
 774 out:
 775         btrfs_free_path(path);
 776         return total_count;
 777
 778 }
 779
 780 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 781                        struct btrfs_root *root, u64 owner_objectid)
 782 {
 783         u64 generation;
 784         u64 key_objectid;
 785         u64 level;
 786         u32 nritems;
 787         struct btrfs_disk_key disk_key;
 788
 789         level = btrfs_header_level(root->node);
 790         generation = trans->transid;
 791         nritems = btrfs_header_nritems(root->node);
 792         if (nritems > 0) {
 793                 if (level == 0)
 794                         btrfs_item_key(root->node, &disk_key, 0);
 795                 else
 796                         btrfs_node_key(root->node, &disk_key, 0);
 797                 key_objectid = btrfs_disk_key_objectid(&disk_key);
 798         } else {
 799                 key_objectid = 0;
 800         }
 801         return btrfs_inc_extent_ref(trans, root, root->node->start,
 802                                     root->node->len, owner_objectid,
 803                                     generation, level, key_objectid);
 804 }
 805
 806 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 807                   struct extent_buffer *buf)
 808 {
 809         u64 bytenr;
 810         u32 nritems;
 811         struct btrfs_key key;
 812         struct btrfs_file_extent_item *fi;
 813         int i;
 814         int level;
 815         int ret;
 816         int faili;
 817
 818         if (!root->ref_cows)
 819                 return 0;
 820
 821         level = btrfs_header_level(buf);
 822         nritems = btrfs_header_nritems(buf);
 823         for (i = 0; i < nritems; i++) {
 824                 if (level == 0) {
 825                         u64 disk_bytenr;
 826                         btrfs_item_key_to_cpu(buf, &key, i);
 827                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 828                                 continue;
 829                         fi = btrfs_item_ptr(buf, i,
 830                                             struct btrfs_file_extent_item);
 831                         if (btrfs_file_extent_type(buf, fi) ==
 832                             BTRFS_FILE_EXTENT_INLINE)
 833                                 continue;
 834                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 835                         if (disk_bytenr == 0)
 836                                 continue;
 837                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
 838                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 839                                     root->root_key.objectid, trans->transid,
 840                                     key.objectid, key.offset);
 841                         if (ret) {
 842                                 faili = i;
 843                                 goto fail;
 844                         }
 845                 } else {
 846                         bytenr = btrfs_node_blockptr(buf, i);
 847                         btrfs_node_key_to_cpu(buf, &key, i);
 848                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
 849                                            btrfs_level_size(root, level - 1),
 850                                            root->root_key.objectid,
 851                                            trans->transid,
 852                                            level - 1, key.objectid);
 853                         if (ret) {
 854                                 faili = i;
 855                                 goto fail;
 856                         }
 857                 }
 858         }
 859         return 0;
 860 fail:
 861         WARN_ON(1);
 862 #if 0
 863         for (i =0; i < faili; i++) {
 864                 if (level == 0) {
 865                         u64 disk_bytenr;
 866                         btrfs_item_key_to_cpu(buf, &key, i);
 867                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 868                                 continue;
 869                         fi = btrfs_item_ptr(buf, i,
 870                                             struct btrfs_file_extent_item);
 871                         if (btrfs_file_extent_type(buf, fi) ==
 872                             BTRFS_FILE_EXTENT_INLINE)
 873                                 continue;
 874                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 875                         if (disk_bytenr == 0)
 876                                 continue;
 877                         err = btrfs_free_extent(trans, root, disk_bytenr,
 878                                     btrfs_file_extent_disk_num_bytes(buf,
 879                                                                       fi), 0);
 880                         BUG_ON(err);
 881                 } else {
 882                         bytenr = btrfs_node_blockptr(buf, i);
 883                         err = btrfs_free_extent(trans, root, bytenr,
 884                                         btrfs_level_size(root, level - 1), 0);
 885                         BUG_ON(err);
 886                 }
 887         }
 888 #endif
 889         return ret;
 890 }
 891
 892 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 893                                  struct btrfs_root *root,
 894                                  struct btrfs_path *path,
 895                                  struct btrfs_block_group_cache *cache)
 896 {
 897         int ret;
 898         int pending_ret;
 899         struct btrfs_root *extent_root = root->fs_info->extent_root;
 900         unsigned long bi;
 901         struct extent_buffer *leaf;
 902
 903         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 904         if (ret < 0)
 905                 goto fail;
 906         BUG_ON(ret);
 907
 908         leaf = path->nodes[0];
 909         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 910         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 911         btrfs_mark_buffer_dirty(leaf);
 912         btrfs_release_path(extent_root, path);
 913 fail:
 914         finish_current_insert(trans, extent_root);
 915         pending_ret = del_pending_extents(trans, extent_root);
 916         if (ret)
 917                 return ret;
 918         if (pending_ret)
 919                 return pending_ret;
 920         return 0;
 921
 922 }
 923
 924 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 925                                    struct btrfs_root *root)
 926 {
 927         struct extent_map_tree *block_group_cache;
 928         struct btrfs_block_group_cache *cache;
 929         int ret;
 930         int err = 0;
 931         int werr = 0;
 932         struct btrfs_path *path;
 933         u64 last = 0;
 934         u64 start;
 935         u64 end;
 936         u64 ptr;
 937
 938         block_group_cache = &root->fs_info->block_group_cache;
 939         path = btrfs_alloc_path();
 940         if (!path)
 941                 return -ENOMEM;
 942
 943         while(1) {
 944                 ret = find_first_extent_bit(block_group_cache, last,
 945                                             &start, &end, BLOCK_GROUP_DIRTY);
 946                 if (ret)
 947                         break;
 948
 949                 last = end + 1;
 950                 ret = get_state_private(block_group_cache, start, &ptr);
 951                 if (ret)
 952                         break;
 953
 954                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 955                 err = write_one_cache_group(trans, root,
 956                                             path, cache);
 957                 /*
 958                  * if we fail to write the cache group, we want
 959                  * to keep it marked dirty in hopes that a later
 960                  * write will work
 961                  */
 962                 if (err) {
 963                         werr = err;
 964                         continue;
 965                 }
 966                 clear_extent_bits(block_group_cache, start, end,
 967                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 968         }
 969         btrfs_free_path(path);
 970         return werr;
 971 }
 972
 973 static int update_block_group(struct btrfs_trans_handle *trans,
 974                               struct btrfs_root *root,
 975                               u64 bytenr, u64 num_bytes, int alloc,
 976                               int mark_free, int data)
 977 {
 978         struct btrfs_block_group_cache *cache;
 979         struct btrfs_fs_info *info = root->fs_info;
 980         u64 total = num_bytes;
 981         u64 old_val;
 982         u64 byte_in_group;
 983         u64 start;
 984         u64 end;
 985
 986         while(total) {
 987                 cache = btrfs_lookup_block_group(info, bytenr);
 988                 if (!cache) {
 989                         return -1;
 990                 }
 991                 byte_in_group = bytenr - cache->key.objectid;
 992                 WARN_ON(byte_in_group > cache->key.offset);
 993                 start = cache->key.objectid;
 994                 end = start + cache->key.offset - 1;
 995                 set_extent_bits(&info->block_group_cache, start, end,
 996                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
 997
 998                 old_val = btrfs_block_group_used(&cache->item);
 999                 num_bytes = min(total, cache->key.offset - byte_in_group);
1000                 if (alloc) {
1001                         if (cache->data != data &&
1002                             old_val < (cache->key.offset >> 1)) {
1003                                 int bit_to_clear;
1004                                 int bit_to_set;
1005                                 cache->data = data;
1006                                 if (data) {
1007                                         bit_to_clear = BLOCK_GROUP_METADATA;
1008                                         bit_to_set = BLOCK_GROUP_DATA;
1009                                         cache->item.flags &=
1010                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1011                                         cache->item.flags |=
1012                                                 BTRFS_BLOCK_GROUP_DATA;
1013                                 } else {
1014                                         bit_to_clear = BLOCK_GROUP_DATA;
1015                                         bit_to_set = BLOCK_GROUP_METADATA;
1016                                         cache->item.flags &=
1017                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1018                                         cache->item.flags &=
1019                                                 ~BTRFS_BLOCK_GROUP_DATA;
1020                                 }
1021                                 clear_extent_bits(&info->block_group_cache,
1022                                                   start, end, bit_to_clear,
1023                                                   GFP_NOFS);
1024                                 set_extent_bits(&info->block_group_cache,
1025                                                 start, end, bit_to_set,
1026                                                 GFP_NOFS);
1027                         } else if (cache->data != data &&
1028                                    cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1029                                 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1030                                 set_extent_bits(&info->block_group_cache,
1031                                                 start, end,
1032                                                 BLOCK_GROUP_DATA |
1033                                                 BLOCK_GROUP_METADATA,
1034                                                 GFP_NOFS);
1035                         }
1036                         old_val += num_bytes;
1037                 } else {
1038                         old_val -= num_bytes;
1039                         if (mark_free) {
1040                                 set_extent_dirty(&info->free_space_cache,
1041                                                  bytenr, bytenr + num_bytes - 1,
1042                                                  GFP_NOFS);
1043                         }
1044                 }
1045                 btrfs_set_block_group_used(&cache->item, old_val);
1046                 total -= num_bytes;
1047                 bytenr += num_bytes;
1048         }
1049         return 0;
1050 }
1051 static int update_pinned_extents(struct btrfs_root *root,
1052                                 u64 bytenr, u64 num, int pin)
1053 {
1054         u64 len;
1055         struct btrfs_block_group_cache *cache;
1056         struct btrfs_fs_info *fs_info = root->fs_info;
1057
1058         if (pin) {
1059                 set_extent_dirty(&fs_info->pinned_extents,
1060                                 bytenr, bytenr + num - 1, GFP_NOFS);
1061         } else {
1062                 clear_extent_dirty(&fs_info->pinned_extents,
1063                                 bytenr, bytenr + num - 1, GFP_NOFS);
1064         }
1065         while (num > 0) {
1066                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1067                 WARN_ON(!cache);
1068                 len = min(num, cache->key.offset -
1069                           (bytenr - cache->key.objectid));
1070                 if (pin) {
1071                         cache->pinned += len;
1072                         fs_info->total_pinned += len;
1073                 } else {
1074                         cache->pinned -= len;
1075                         fs_info->total_pinned -= len;
1076                 }
1077                 bytenr += len;
1078                 num -= len;
1079         }
1080         return 0;
1081 }
1082
1083 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
1084 {
1085         u64 last = 0;
1086         u64 start;
1087         u64 end;
1088         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
1089         int ret;
1090
1091         while(1) {
1092                 ret = find_first_extent_bit(pinned_extents, last,
1093                                             &start, &end, EXTENT_DIRTY);
1094                 if (ret)
1095                         break;
1096                 set_extent_dirty(copy, start, end, GFP_NOFS);
1097                 last = end + 1;
1098         }
1099         return 0;
1100 }
1101
1102 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1103                                struct btrfs_root *root,
1104                                struct extent_map_tree *unpin)
1105 {
1106         u64 start;
1107         u64 end;
1108         int ret;
1109         struct extent_map_tree *free_space_cache;
1110         free_space_cache = &root->fs_info->free_space_cache;
1111
1112         while(1) {
1113                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1114                                             EXTENT_DIRTY);
1115                 if (ret)
1116                         break;
1117                 update_pinned_extents(root, start, end + 1 - start, 0);
1118                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1119                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1120         }
1121         return 0;
1122 }
1123
1124 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
1125                                  btrfs_root *extent_root)
1126 {
1127         u64 start;
1128         u64 end;
1129         struct btrfs_fs_info *info = extent_root->fs_info;
1130         struct extent_buffer *eb;
1131         struct btrfs_path *path;
1132         struct btrfs_key ins;
1133         struct btrfs_disk_key first;
1134         struct btrfs_extent_item extent_item;
1135         int ret;
1136         int level;
1137         int err = 0;
1138
1139         btrfs_set_stack_extent_refs(&extent_item, 1);
1140         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1141         path = btrfs_alloc_path();
1142
1143         while(1) {
1144                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1145                                             &end, EXTENT_LOCKED);
1146                 if (ret)
1147                         break;
1148
1149                 ins.objectid = start;
1150                 ins.offset = end + 1 - start;
1151                 err = btrfs_insert_item(trans, extent_root, &ins,
1152                                         &extent_item, sizeof(extent_item));
1153                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1154                                   GFP_NOFS);
1155                 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1156                 level = btrfs_header_level(eb);
1157                 if (level == 0) {
1158                         btrfs_item_key(eb, &first, 0);
1159                 } else {
1160                         btrfs_node_key(eb, &first, 0);
1161                 }
1162                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1163                                           start, extent_root->root_key.objectid,
1164                                           0, level,
1165                                           btrfs_disk_key_objectid(&first));
1166                 BUG_ON(err);
1167                 free_extent_buffer(eb);
1168         }
1169         btrfs_free_path(path);
1170         return 0;
1171 }
1172
1173 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1174                           int pending)
1175 {
1176         int err = 0;
1177         struct extent_buffer *buf;
1178
1179         if (!pending) {
1180                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1181                 if (buf) {
1182                         if (btrfs_buffer_uptodate(buf)) {
1183                                 u64 transid =
1184                                     root->fs_info->running_transaction->transid;
1185                                 if (btrfs_header_generation(buf) == transid) {
1186                                         free_extent_buffer(buf);
1187                                         return 1;
1188                                 }
1189                         }
1190                         free_extent_buffer(buf);
1191                 }
1192                 update_pinned_extents(root, bytenr, num_bytes, 1);
1193         } else {
1194                 set_extent_bits(&root->fs_info->pending_del,
1195                                 bytenr, bytenr + num_bytes - 1,
1196                                 EXTENT_LOCKED, GFP_NOFS);
1197         }
1198         BUG_ON(err < 0);
1199         return 0;
1200 }
1201
1202 /*
1203  * remove an extent from the root, returns 0 on success
1204  */
1205 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1206                          *root, u64 bytenr, u64 num_bytes,
1207                          u64 root_objectid, u64 ref_generation,
1208                          u64 owner_objectid, u64 owner_offset, int pin,
1209                          int mark_free)
1210 {
1211         struct btrfs_path *path;
1212         struct btrfs_key key;
1213         struct btrfs_fs_info *info = root->fs_info;
1214         struct btrfs_root *extent_root = info->extent_root;
1215         struct extent_buffer *leaf;
1216         int ret;
1217         struct btrfs_extent_item *ei;
1218         u32 refs;
1219
1220         key.objectid = bytenr;
1221         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1222         key.offset = num_bytes;
1223
1224         path = btrfs_alloc_path();
1225         if (!path)
1226                 return -ENOMEM;
1227
1228         ret = lookup_extent_backref(trans, extent_root, path,
1229                                     bytenr, root_objectid,
1230                                     ref_generation,
1231                                     owner_objectid, owner_offset, 1);
1232         if (ret == 0) {
1233                 ret = btrfs_del_item(trans, extent_root, path);
1234         } else {
1235                 btrfs_print_leaf(extent_root, path->nodes[0]);
1236                 WARN_ON(1);
1237                 printk("Unable to find ref byte nr %Lu root %Lu "
1238                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1239                        root_objectid, ref_generation, owner_objectid,
1240                        owner_offset);
1241         }
1242         btrfs_release_path(extent_root, path);
1243         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1244         if (ret < 0)
1245                 return ret;
1246         BUG_ON(ret);
1247
1248         leaf = path->nodes[0];
1249         ei = btrfs_item_ptr(leaf, path->slots[0],
1250                             struct btrfs_extent_item);
1251         refs = btrfs_extent_refs(leaf, ei);
1252         BUG_ON(refs == 0);
1253         refs -= 1;
1254         btrfs_set_extent_refs(leaf, ei, refs);
1255         btrfs_mark_buffer_dirty(leaf);
1256
1257         if (refs == 0) {
1258                 u64 super_used;
1259                 u64 root_used;
1260
1261                 if (pin) {
1262                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1263                         if (ret > 0)
1264                                 mark_free = 1;
1265                         BUG_ON(ret < 0);
1266                 }
1267
1268                 /* block accounting for super block */
1269                 super_used = btrfs_super_bytes_used(&info->super_copy);
1270                 btrfs_set_super_bytes_used(&info->super_copy,
1271                                            super_used - num_bytes);
1272
1273                 /* block accounting for root item */
1274                 root_used = btrfs_root_used(&root->root_item);
1275                 btrfs_set_root_used(&root->root_item,
1276                                            root_used - num_bytes);
1277
1278                 ret = btrfs_del_item(trans, extent_root, path);
1279                 if (ret) {
1280                         return ret;
1281                 }
1282                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1283                                          mark_free, 0);
1284                 BUG_ON(ret);
1285         }
1286         btrfs_free_path(path);
1287         finish_current_insert(trans, extent_root);
1288         return ret;
1289 }
1290
1291 /*
1292  * find all the blocks marked as pending in the radix tree and remove
1293  * them from the extent map
1294  */
1295 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1296                                btrfs_root *extent_root)
1297 {
1298         int ret;
1299         int err = 0;
1300         u64 start;
1301         u64 end;
1302         struct extent_map_tree *pending_del;
1303         struct extent_map_tree *pinned_extents;
1304
1305         pending_del = &extent_root->fs_info->pending_del;
1306         pinned_extents = &extent_root->fs_info->pinned_extents;
1307
1308         while(1) {
1309                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1310                                             EXTENT_LOCKED);
1311                 if (ret)
1312                         break;
1313                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1314                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1315                                   GFP_NOFS);
1316                 ret = __free_extent(trans, extent_root,
1317                                      start, end + 1 - start,
1318                                      extent_root->root_key.objectid,
1319                                      0, 0, 0, 0, 0);
1320                 if (ret)
1321                         err = ret;
1322         }
1323         return err;
1324 }
1325
1326 /*
1327  * remove an extent from the root, returns 0 on success
1328  */
1329 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1330                       *root, u64 bytenr, u64 num_bytes,
1331                       u64 root_objectid, u64 ref_generation,
1332                       u64 owner_objectid, u64 owner_offset, int pin)
1333 {
1334         struct btrfs_root *extent_root = root->fs_info->extent_root;
1335         int pending_ret;
1336         int ret;
1337
1338         WARN_ON(num_bytes < root->sectorsize);
1339         if (!root->ref_cows)
1340                 ref_generation = 0;
1341
1342         if (root == extent_root) {
1343                 pin_down_bytes(root, bytenr, num_bytes, 1);
1344                 return 0;
1345         }
1346         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1347                             ref_generation, owner_objectid, owner_offset,
1348                             pin, pin == 0);
1349         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1350         return ret ? ret : pending_ret;
1351 }
1352
1353 static u64 stripe_align(struct btrfs_root *root, u64 val)
1354 {
1355         u64 mask = ((u64)root->stripesize - 1);
1356         u64 ret = (val + mask) & ~mask;
1357         return ret;
1358 }
1359
1360 /*
1361  * walks the btree of allocated extents and find a hole of a given size.
1362  * The key ins is changed to record the hole:
1363  * ins->objectid == block start
1364  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1365  * ins->offset == number of blocks
1366  * Any available blocks before search_start are skipped.
1367  */
1368 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1369                             *orig_root, u64 num_bytes, u64 empty_size,
1370                             u64 search_start, u64 search_end, u64 hint_byte,
1371                             struct btrfs_key *ins, u64 exclude_start,
1372                             u64 exclude_nr, int data)
1373 {
1374         struct btrfs_path *path;
1375         struct btrfs_key key;
1376         u64 hole_size = 0;
1377         u64 aligned;
1378         int ret;
1379         int slot = 0;
1380         u64 last_byte = 0;
1381         u64 orig_search_start = search_start;
1382         int start_found;
1383         struct extent_buffer *l;
1384         struct btrfs_root * root = orig_root->fs_info->extent_root;
1385         struct btrfs_fs_info *info = root->fs_info;
1386         u64 total_needed = num_bytes;
1387         int level;
1388         struct btrfs_block_group_cache *block_group;
1389         int full_scan = 0;
1390         int wrapped = 0;
1391         u64 cached_start;
1392
1393         WARN_ON(num_bytes < root->sectorsize);
1394         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1395
1396         level = btrfs_header_level(root->node);
1397
1398         if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1399                 data = BTRFS_BLOCK_GROUP_MIXED;
1400         }
1401
1402         if (search_end == (u64)-1)
1403                 search_end = btrfs_super_total_bytes(&info->super_copy);
1404         if (hint_byte) {
1405                 block_group = btrfs_lookup_block_group(info, hint_byte);
1406                 if (!block_group)
1407                         hint_byte = search_start;
1408                 block_group = btrfs_find_block_group(root, block_group,
1409                                                      hint_byte, data, 1);
1410         } else {
1411                 block_group = btrfs_find_block_group(root,
1412                                                      trans->block_group,
1413                                                      search_start, data, 1);
1414         }
1415
1416         total_needed += empty_size;
1417         path = btrfs_alloc_path();
1418 check_failed:
1419         if (!block_group) {
1420                 block_group = btrfs_lookup_block_group(info, search_start);
1421                 if (!block_group)
1422                         block_group = btrfs_lookup_block_group(info,
1423                                                        orig_search_start);
1424         }
1425         search_start = find_search_start(root, &block_group, search_start,
1426                                          total_needed, data, full_scan);
1427         search_start = stripe_align(root, search_start);
1428         cached_start = search_start;
1429         btrfs_init_path(path);
1430         ins->objectid = search_start;
1431         ins->offset = 0;
1432         start_found = 0;
1433         path->reada = 2;
1434
1435         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1436         if (ret < 0)
1437                 goto error;
1438
1439         if (path->slots[0] > 0) {
1440                 path->slots[0]--;
1441         }
1442
1443         l = path->nodes[0];
1444         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1445
1446         /*
1447          * walk backwards to find the first extent item key
1448          */
1449         while(btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1450                 if (path->slots[0] == 0) {
1451                         ret = btrfs_prev_leaf(root, path);
1452                         if (ret != 0) {
1453                                 ret = btrfs_search_slot(trans, root, ins,
1454                                                         path, 0, 0);
1455                                 if (ret < 0)
1456                                         goto error;
1457                                 if (path->slots[0] > 0)
1458                                         path->slots[0]--;
1459                                 break;
1460                         }
1461                 } else {
1462                         path->slots[0]--;
1463                 }
1464                 l = path->nodes[0];
1465                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1466         }
1467         while (1) {
1468                 l = path->nodes[0];
1469                 slot = path->slots[0];
1470                 if (slot >= btrfs_header_nritems(l)) {
1471                         ret = btrfs_next_leaf(root, path);
1472                         if (ret == 0)
1473                                 continue;
1474                         if (ret < 0)
1475                                 goto error;
1476
1477                         search_start = max(search_start,
1478                                            block_group->key.objectid);
1479                         if (!start_found) {
1480                                 aligned = stripe_align(root, search_start);
1481                                 ins->objectid = aligned;
1482                                 if (aligned >= search_end) {
1483                                         ret = -ENOSPC;
1484                                         goto error;
1485                                 }
1486                                 ins->offset = search_end - aligned;
1487                                 start_found = 1;
1488                                 goto check_pending;
1489                         }
1490                         ins->objectid = stripe_align(root,
1491                                                      last_byte > search_start ?
1492                                                      last_byte : search_start);
1493                         if (search_end <= ins->objectid) {
1494                                 ret = -ENOSPC;
1495                                 goto error;
1496                         }
1497                         ins->offset = search_end - ins->objectid;
1498                         BUG_ON(ins->objectid >= search_end);
1499                         goto check_pending;
1500                 }
1501                 btrfs_item_key_to_cpu(l, &key, slot);
1502
1503                 if (key.objectid >= search_start && key.objectid > last_byte &&
1504                     start_found) {
1505                         if (last_byte < search_start)
1506                                 last_byte = search_start;
1507                         aligned = stripe_align(root, last_byte);
1508                         hole_size = key.objectid - aligned;
1509                         if (key.objectid > aligned && hole_size >= num_bytes) {
1510                                 ins->objectid = aligned;
1511                                 ins->offset = hole_size;
1512                                 goto check_pending;
1513                         }
1514                 }
1515                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1516                         if (!start_found && btrfs_key_type(&key) ==
1517                             BTRFS_BLOCK_GROUP_ITEM_KEY) {
1518                                 last_byte = key.objectid;
1519                                 start_found = 1;
1520                         }
1521                         goto next;
1522                 }
1523
1524
1525                 start_found = 1;
1526                 last_byte = key.objectid + key.offset;
1527
1528                 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1529                     last_byte >= block_group->key.objectid +
1530                     block_group->key.offset) {
1531                         btrfs_release_path(root, path);
1532                         search_start = block_group->key.objectid +
1533                                 block_group->key.offset;
1534                         goto new_group;
1535                 }
1536 next:
1537                 path->slots[0]++;
1538                 cond_resched();
1539         }
1540 check_pending:
1541         /* we have to make sure we didn't find an extent that has already
1542          * been allocated by the map tree or the original allocation
1543          */
1544         btrfs_release_path(root, path);
1545         BUG_ON(ins->objectid < search_start);
1546
1547         if (ins->objectid + num_bytes >= search_end)
1548                 goto enospc;
1549         if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1550             ins->objectid + num_bytes > block_group->
1551             key.objectid + block_group->key.offset) {
1552                 search_start = block_group->key.objectid +
1553                         block_group->key.offset;
1554                 goto new_group;
1555         }
1556         if (test_range_bit(&info->extent_ins, ins->objectid,
1557                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1558                 search_start = ins->objectid + num_bytes;
1559                 goto new_group;
1560         }
1561         if (test_range_bit(&info->pinned_extents, ins->objectid,
1562                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1563                 search_start = ins->objectid + num_bytes;
1564                 goto new_group;
1565         }
1566         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1567             ins->objectid < exclude_start + exclude_nr)) {
1568                 search_start = exclude_start + exclude_nr;
1569                 goto new_group;
1570         }
1571         if (!data) {
1572                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1573                 if (block_group)
1574                         trans->block_group = block_group;
1575         }
1576         ins->offset = num_bytes;
1577         btrfs_free_path(path);
1578         return 0;
1579
1580 new_group:
1581         if (search_start + num_bytes >= search_end) {
1582 enospc:
1583                 search_start = orig_search_start;
1584                 if (full_scan) {
1585                         ret = -ENOSPC;
1586                         goto error;
1587                 }
1588                 if (wrapped) {
1589                         if (!full_scan)
1590                                 total_needed -= empty_size;
1591                         full_scan = 1;
1592                         data = BTRFS_BLOCK_GROUP_MIXED;
1593                 } else
1594                         wrapped = 1;
1595         }
1596         block_group = btrfs_lookup_block_group(info, search_start);
1597         cond_resched();
1598         block_group = btrfs_find_block_group(root, block_group,
1599                                              search_start, data, 0);
1600         goto check_failed;
1601
1602 error:
1603         btrfs_release_path(root, path);
1604         btrfs_free_path(path);
1605         return ret;
1606 }
1607 /*
1608  * finds a free extent and does all the dirty work required for allocation
1609  * returns the key for the extent through ins, and a tree buffer for
1610  * the first block of the extent through buf.
1611  *
1612  * returns 0 if everything worked, non-zero otherwise.
1613  */
1614 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1615                        struct btrfs_root *root,
1616                        u64 num_bytes, u64 root_objectid, u64 ref_generation,
1617                        u64 owner, u64 owner_offset,
1618                        u64 empty_size, u64 hint_byte,
1619                        u64 search_end, struct btrfs_key *ins, int data)
1620 {
1621         int ret;
1622         int pending_ret;
1623         u64 super_used, root_used;
1624         u64 search_start = 0;
1625         struct btrfs_fs_info *info = root->fs_info;
1626         struct btrfs_root *extent_root = info->extent_root;
1627         struct btrfs_extent_item extent_item;
1628         struct btrfs_path *path;
1629
1630         btrfs_set_stack_extent_refs(&extent_item, 1);
1631
1632         WARN_ON(num_bytes < root->sectorsize);
1633         ret = find_free_extent(trans, root, num_bytes, empty_size,
1634                                search_start, search_end, hint_byte, ins,
1635                                trans->alloc_exclude_start,
1636                                trans->alloc_exclude_nr, data);
1637         BUG_ON(ret);
1638         if (ret)
1639                 return ret;
1640
1641         /* block accounting for super block */
1642         super_used = btrfs_super_bytes_used(&info->super_copy);
1643         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1644
1645         /* block accounting for root item */
1646         root_used = btrfs_root_used(&root->root_item);
1647         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1648
1649         clear_extent_dirty(&root->fs_info->free_space_cache,
1650                            ins->objectid, ins->objectid + ins->offset - 1,
1651                            GFP_NOFS);
1652
1653         if (root == extent_root) {
1654                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1655                                 ins->objectid + ins->offset - 1,
1656                                 EXTENT_LOCKED, GFP_NOFS);
1657                 WARN_ON(data == 1);
1658                 goto update_block;
1659         }
1660
1661         WARN_ON(trans->alloc_exclude_nr);
1662         trans->alloc_exclude_start = ins->objectid;
1663         trans->alloc_exclude_nr = ins->offset;
1664         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1665                                 sizeof(extent_item));
1666
1667         trans->alloc_exclude_start = 0;
1668         trans->alloc_exclude_nr = 0;
1669         BUG_ON(ret);
1670
1671         path = btrfs_alloc_path();
1672         BUG_ON(!path);
1673         ret = btrfs_insert_extent_backref(trans, extent_root, path,
1674                                           ins->objectid, root_objectid,
1675                                           ref_generation, owner, owner_offset);
1676
1677         BUG_ON(ret);
1678         btrfs_free_path(path);
1679         finish_current_insert(trans, extent_root);
1680         pending_ret = del_pending_extents(trans, extent_root);
1681
1682         if (ret) {
1683                 return ret;
1684         }
1685         if (pending_ret) {
1686                 return pending_ret;
1687         }
1688
1689 update_block:
1690         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1691                                  data);
1692         BUG_ON(ret);
1693         return 0;
1694 }
1695
1696 /*
1697  * helper function to allocate a block for a given tree
1698  * returns the tree buffer or NULL.
1699  */
1700 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1701                                              struct btrfs_root *root,
1702                                              u32 blocksize,
1703                                              u64 root_objectid, u64 hint,
1704                                              u64 empty_size)
1705 {
1706         u64 ref_generation;
1707
1708         if (root->ref_cows)
1709                 ref_generation = trans->transid;
1710         else
1711                 ref_generation = 0;
1712
1713
1714         return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1715                                         ref_generation, 0, 0, hint, empty_size);
1716 }
1717
1718 /*
1719  * helper function to allocate a block for a given tree
1720  * returns the tree buffer or NULL.
1721  */
1722 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1723                                              struct btrfs_root *root,
1724                                              u32 blocksize,
1725                                              u64 root_objectid,
1726                                              u64 ref_generation,
1727                                              u64 first_objectid,
1728                                              int level,
1729                                              u64 hint,
1730                                              u64 empty_size)
1731 {
1732         struct btrfs_key ins;
1733         int ret;
1734         struct extent_buffer *buf;
1735
1736         ret = btrfs_alloc_extent(trans, root, blocksize,
1737                                  root_objectid, ref_generation,
1738                                  level, first_objectid, empty_size, hint,
1739                                  (u64)-1, &ins, 0);
1740         if (ret) {
1741                 BUG_ON(ret > 0);
1742                 return ERR_PTR(ret);
1743         }
1744         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1745         if (!buf) {
1746                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1747                                   root->root_key.objectid, ref_generation,
1748                                   0, 0, 0);
1749                 return ERR_PTR(-ENOMEM);
1750         }
1751         btrfs_set_buffer_uptodate(buf);
1752         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1753                          buf->start + buf->len - 1, GFP_NOFS);
1754         set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1755                         buf->start, buf->start + buf->len - 1,
1756                         EXTENT_CSUM, GFP_NOFS);
1757         buf->flags |= EXTENT_CSUM;
1758         btrfs_set_buffer_defrag(buf);
1759         trans->blocks_used++;
1760         return buf;
1761 }
1762
1763 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1764                          struct btrfs_root *root, struct extent_buffer *leaf)
1765 {
1766         u64 leaf_owner;
1767         u64 leaf_generation;
1768         struct btrfs_key key;
1769         struct btrfs_file_extent_item *fi;
1770         int i;
1771         int nritems;
1772         int ret;
1773
1774         BUG_ON(!btrfs_is_leaf(leaf));
1775         nritems = btrfs_header_nritems(leaf);
1776         leaf_owner = btrfs_header_owner(leaf);
1777         leaf_generation = btrfs_header_generation(leaf);
1778
1779         for (i = 0; i < nritems; i++) {
1780                 u64 disk_bytenr;
1781
1782                 btrfs_item_key_to_cpu(leaf, &key, i);
1783                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1784                         continue;
1785                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1786                 if (btrfs_file_extent_type(leaf, fi) ==
1787                     BTRFS_FILE_EXTENT_INLINE)
1788                         continue;
1789                 /*
1790                  * FIXME make sure to insert a trans record that
1791                  * repeats the snapshot del on crash
1792                  */
1793                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1794                 if (disk_bytenr == 0)
1795                         continue;
1796                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1797                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
1798                                 leaf_owner, leaf_generation,
1799                                 key.objectid, key.offset, 0);
1800                 BUG_ON(ret);
1801         }
1802         return 0;
1803 }
1804
1805 static void reada_walk_down(struct btrfs_root *root,
1806                             struct extent_buffer *node)
1807 {
1808         int i;
1809         u32 nritems;
1810         u64 bytenr;
1811         int ret;
1812         u32 refs;
1813         int level;
1814         u32 blocksize;
1815
1816         nritems = btrfs_header_nritems(node);
1817         level = btrfs_header_level(node);
1818         for (i = 0; i < nritems; i++) {
1819                 bytenr = btrfs_node_blockptr(node, i);
1820                 blocksize = btrfs_level_size(root, level - 1);
1821                 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1822                 BUG_ON(ret);
1823                 if (refs != 1)
1824                         continue;
1825                 mutex_unlock(&root->fs_info->fs_mutex);
1826                 ret = readahead_tree_block(root, bytenr, blocksize);
1827                 cond_resched();
1828                 mutex_lock(&root->fs_info->fs_mutex);
1829                 if (ret)
1830                         break;
1831         }
1832 }
1833
1834 /*
1835  * helper function for drop_snapshot, this walks down the tree dropping ref
1836  * counts as it goes.
1837  */
1838 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1839                           *root, struct btrfs_path *path, int *level)
1840 {
1841         u64 root_owner;
1842         u64 root_gen;
1843         u64 bytenr;
1844         struct extent_buffer *next;
1845         struct extent_buffer *cur;
1846         struct extent_buffer *parent;
1847         u32 blocksize;
1848         int ret;
1849         u32 refs;
1850
1851         WARN_ON(*level < 0);
1852         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1853         ret = lookup_extent_ref(trans, root,
1854                                 path->nodes[*level]->start,
1855                                 path->nodes[*level]->len, &refs);
1856         BUG_ON(ret);
1857         if (refs > 1)
1858                 goto out;
1859
1860         /*
1861          * walk down to the last node level and free all the leaves
1862          */
1863         while(*level >= 0) {
1864                 WARN_ON(*level < 0);
1865                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1866                 cur = path->nodes[*level];
1867
1868                 if (*level > 0 && path->slots[*level] == 0)
1869                         reada_walk_down(root, cur);
1870
1871                 if (btrfs_header_level(cur) != *level)
1872                         WARN_ON(1);
1873
1874                 if (path->slots[*level] >=
1875                     btrfs_header_nritems(cur))
1876                         break;
1877                 if (*level == 0) {
1878                         ret = drop_leaf_ref(trans, root, cur);
1879                         BUG_ON(ret);
1880                         break;
1881                 }
1882                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1883                 blocksize = btrfs_level_size(root, *level - 1);
1884                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1885                 BUG_ON(ret);
1886                 if (refs != 1) {
1887                         parent = path->nodes[*level];
1888                         root_owner = btrfs_header_owner(parent);
1889                         root_gen = btrfs_header_generation(parent);
1890                         path->slots[*level]++;
1891                         ret = btrfs_free_extent(trans, root, bytenr,
1892                                                 blocksize, root_owner,
1893                                                 root_gen, 0, 0, 1);
1894                         BUG_ON(ret);
1895                         continue;
1896                 }
1897                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1898                 if (!next || !btrfs_buffer_uptodate(next)) {
1899                         free_extent_buffer(next);
1900                         mutex_unlock(&root->fs_info->fs_mutex);
1901                         next = read_tree_block(root, bytenr, blocksize);
1902                         mutex_lock(&root->fs_info->fs_mutex);
1903
1904                         /* we dropped the lock, check one more time */
1905                         ret = lookup_extent_ref(trans, root, bytenr,
1906                                                 blocksize, &refs);
1907                         BUG_ON(ret);
1908                         if (refs != 1) {
1909                                 parent = path->nodes[*level];
1910                                 root_owner = btrfs_header_owner(parent);
1911                                 root_gen = btrfs_header_generation(parent);
1912
1913                                 path->slots[*level]++;
1914                                 free_extent_buffer(next);
1915                                 ret = btrfs_free_extent(trans, root, bytenr,
1916                                                         blocksize,
1917                                                         root_owner,
1918                                                         root_gen, 0, 0, 1);
1919                                 BUG_ON(ret);
1920                                 continue;
1921                         }
1922                 }
1923                 WARN_ON(*level <= 0);
1924                 if (path->nodes[*level-1])
1925                         free_extent_buffer(path->nodes[*level-1]);
1926                 path->nodes[*level-1] = next;
1927                 *level = btrfs_header_level(next);
1928                 path->slots[*level] = 0;
1929         }
1930 out:
1931         WARN_ON(*level < 0);
1932         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1933
1934         if (path->nodes[*level] == root->node) {
1935                 root_owner = root->root_key.objectid;
1936                 parent = path->nodes[*level];
1937         } else {
1938                 parent = path->nodes[*level + 1];
1939                 root_owner = btrfs_header_owner(parent);
1940         }
1941
1942         root_gen = btrfs_header_generation(parent);
1943         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1944                                 path->nodes[*level]->len,
1945                                 root_owner, root_gen, 0, 0, 1);
1946         free_extent_buffer(path->nodes[*level]);
1947         path->nodes[*level] = NULL;
1948         *level += 1;
1949         BUG_ON(ret);
1950         return 0;
1951 }
1952
1953 /*
1954  * helper for dropping snapshots.  This walks back up the tree in the path
1955  * to find the first node higher up where we haven't yet gone through
1956  * all the slots
1957  */
1958 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1959                         *root, struct btrfs_path *path, int *level)
1960 {
1961         u64 root_owner;
1962         u64 root_gen;
1963         struct btrfs_root_item *root_item = &root->root_item;
1964         int i;
1965         int slot;
1966         int ret;
1967
1968         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1969                 slot = path->slots[i];
1970                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1971                         struct extent_buffer *node;
1972                         struct btrfs_disk_key disk_key;
1973                         node = path->nodes[i];
1974                         path->slots[i]++;
1975                         *level = i;
1976                         WARN_ON(*level == 0);
1977                         btrfs_node_key(node, &disk_key, path->slots[i]);
1978                         memcpy(&root_item->drop_progress,
1979                                &disk_key, sizeof(disk_key));
1980                         root_item->drop_level = i;
1981                         return 0;
1982                 } else {
1983                         if (path->nodes[*level] == root->node) {
1984                                 root_owner = root->root_key.objectid;
1985                                 root_gen =
1986                                    btrfs_header_generation(path->nodes[*level]);
1987                         } else {
1988                                 struct extent_buffer *node;
1989                                 node = path->nodes[*level + 1];
1990                                 root_owner = btrfs_header_owner(node);
1991                                 root_gen = btrfs_header_generation(node);
1992                         }
1993                         ret = btrfs_free_extent(trans, root,
1994                                                 path->nodes[*level]->start,
1995                                                 path->nodes[*level]->len,
1996                                                 root_owner, root_gen, 0, 0, 1);
1997                         BUG_ON(ret);
1998                         free_extent_buffer(path->nodes[*level]);
1999                         path->nodes[*level] = NULL;
2000                         *level = i + 1;
2001                 }
2002         }
2003         return 1;
2004 }
2005
2006 /*
2007  * drop the reference count on the tree rooted at 'snap'.  This traverses
2008  * the tree freeing any blocks that have a ref count of zero after being
2009  * decremented.
2010  */
2011 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2012                         *root)
2013 {
2014         int ret = 0;
2015         int wret;
2016         int level;
2017         struct btrfs_path *path;
2018         int i;
2019         int orig_level;
2020         struct btrfs_root_item *root_item = &root->root_item;
2021
2022         path = btrfs_alloc_path();
2023         BUG_ON(!path);
2024
2025         level = btrfs_header_level(root->node);
2026         orig_level = level;
2027         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2028                 path->nodes[level] = root->node;
2029                 extent_buffer_get(root->node);
2030                 path->slots[level] = 0;
2031         } else {
2032                 struct btrfs_key key;
2033                 struct btrfs_disk_key found_key;
2034                 struct extent_buffer *node;
2035
2036                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2037                 level = root_item->drop_level;
2038                 path->lowest_level = level;
2039                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2040                 if (wret < 0) {
2041                         ret = wret;
2042                         goto out;
2043                 }
2044                 node = path->nodes[level];
2045                 btrfs_node_key(node, &found_key, path->slots[level]);
2046                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2047                                sizeof(found_key)));
2048         }
2049         while(1) {
2050                 wret = walk_down_tree(trans, root, path, &level);
2051                 if (wret > 0)
2052                         break;
2053                 if (wret < 0)
2054                         ret = wret;
2055
2056                 wret = walk_up_tree(trans, root, path, &level);
2057                 if (wret > 0)
2058                         break;
2059                 if (wret < 0)
2060                         ret = wret;
2061                 ret = -EAGAIN;
2062                 break;
2063         }
2064         for (i = 0; i <= orig_level; i++) {
2065                 if (path->nodes[i]) {
2066                         free_extent_buffer(path->nodes[i]);
2067                         path->nodes[i] = NULL;
2068                 }
2069         }
2070 out:
2071         btrfs_free_path(path);
2072         return ret;
2073 }
2074
2075 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2076 {
2077         u64 start;
2078         u64 end;
2079         u64 ptr;
2080         int ret;
2081         while(1) {
2082                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2083                                             &start, &end, (unsigned int)-1);
2084                 if (ret)
2085                         break;
2086                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2087                 if (!ret)
2088                         kfree((void *)(unsigned long)ptr);
2089                 clear_extent_bits(&info->block_group_cache, start,
2090                                   end, (unsigned int)-1, GFP_NOFS);
2091         }
2092         while(1) {
2093                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2094                                             &start, &end, EXTENT_DIRTY);
2095                 if (ret)
2096                         break;
2097                 clear_extent_dirty(&info->free_space_cache, start,
2098                                    end, GFP_NOFS);
2099         }
2100         return 0;
2101 }
2102
2103 int btrfs_read_block_groups(struct btrfs_root *root)
2104 {
2105         struct btrfs_path *path;
2106         int ret;
2107         int err = 0;
2108         int bit;
2109         struct btrfs_block_group_cache *cache;
2110         struct btrfs_fs_info *info = root->fs_info;
2111         struct extent_map_tree *block_group_cache;
2112         struct btrfs_key key;
2113         struct btrfs_key found_key;
2114         struct extent_buffer *leaf;
2115
2116         block_group_cache = &info->block_group_cache;
2117
2118         root = info->extent_root;
2119         key.objectid = 0;
2120         key.offset = BTRFS_BLOCK_GROUP_SIZE;
2121         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2122
2123         path = btrfs_alloc_path();
2124         if (!path)
2125                 return -ENOMEM;
2126
2127         while(1) {
2128                 ret = btrfs_search_slot(NULL, info->extent_root,
2129                                         &key, path, 0, 0);
2130                 if (ret != 0) {
2131                         err = ret;
2132                         break;
2133                 }
2134                 leaf = path->nodes[0];
2135                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2136                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2137                 if (!cache) {
2138                         err = -1;
2139                         break;
2140                 }
2141
2142                 read_extent_buffer(leaf, &cache->item,
2143                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
2144                                    sizeof(cache->item));
2145                 memcpy(&cache->key, &found_key, sizeof(found_key));
2146                 cache->cached = 0;
2147                 cache->pinned = 0;
2148                 key.objectid = found_key.objectid + found_key.offset;
2149                 btrfs_release_path(root, path);
2150
2151                 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2152                         bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2153                         cache->data = BTRFS_BLOCK_GROUP_MIXED;
2154                 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2155                         bit = BLOCK_GROUP_DATA;
2156                         cache->data = BTRFS_BLOCK_GROUP_DATA;
2157                 } else {
2158                         bit = BLOCK_GROUP_METADATA;
2159                         cache->data = 0;
2160                 }
2161
2162                 /* use EXTENT_LOCKED to prevent merging */
2163                 set_extent_bits(block_group_cache, found_key.objectid,
2164                                 found_key.objectid + found_key.offset - 1,
2165                                 bit | EXTENT_LOCKED, GFP_NOFS);
2166                 set_state_private(block_group_cache, found_key.objectid,
2167                                   (unsigned long)cache);
2168
2169                 if (key.objectid >=
2170                     btrfs_super_total_bytes(&info->super_copy))
2171                         break;
2172         }
2173
2174         btrfs_free_path(path);
2175         return 0;
2176 }