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 <linux/pagemap.h>
  22 #include "hash.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27
  28 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
  29 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  30 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  31
  32 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  33                                  btrfs_root *extent_root);
  34 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  35                                btrfs_root *extent_root);
  36 static int find_previous_extent(struct btrfs_root *root,
  37                                 struct btrfs_path *path)
  38 {
  39         struct btrfs_key found_key;
  40         struct extent_buffer *leaf;
  41         int ret;
  42
  43         while(1) {
  44                 if (path->slots[0] == 0) {
  45                         ret = btrfs_prev_leaf(root, path);
  46                         if (ret != 0)
  47                                 return ret;
  48                 } else {
  49                         path->slots[0]--;
  50                 }
  51                 leaf = path->nodes[0];
  52                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  53                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
  54                         return 0;
  55         }
  56         return 1;
  57 }
  58
  59 static int cache_block_group(struct btrfs_root *root,
  60                              struct btrfs_block_group_cache *block_group)
  61 {
  62         struct btrfs_path *path;
  63         int ret;
  64         struct btrfs_key key;
  65         struct extent_buffer *leaf;
  66         struct extent_map_tree *free_space_cache;
  67         int slot;
  68         u64 last = 0;
  69         u64 hole_size;
  70         u64 first_free;
  71         int found = 0;
  72
  73         if (!block_group)
  74                 return 0;
  75
  76         root = root->fs_info->extent_root;
  77         free_space_cache = &root->fs_info->free_space_cache;
  78
  79         if (block_group->cached)
  80                 return 0;
  81
  82         path = btrfs_alloc_path();
  83         if (!path)
  84                 return -ENOMEM;
  85
  86         path->reada = 2;
  87         first_free = block_group->key.objectid;
  88         key.objectid = block_group->key.objectid;
  89         key.offset = 0;
  90         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  91         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  92         if (ret < 0)
  93                 return ret;
  94         ret = find_previous_extent(root, path);
  95         if (ret < 0)
  96                 return ret;
  97         if (ret == 0) {
  98                 leaf = path->nodes[0];
  99                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 100                 if (key.objectid + key.offset > first_free)
 101                         first_free = key.objectid + key.offset;
 102         }
 103         while(1) {
 104                 leaf = path->nodes[0];
 105                 slot = path->slots[0];
 106                 if (slot >= btrfs_header_nritems(leaf)) {
 107                         ret = btrfs_next_leaf(root, path);
 108                         if (ret < 0)
 109                                 goto err;
 110                         if (ret == 0) {
 111                                 continue;
 112                         } else {
 113                                 break;
 114                         }
 115                 }
 116                 btrfs_item_key_to_cpu(leaf, &key, slot);
 117                 if (key.objectid < block_group->key.objectid) {
 118                         goto next;
 119                 }
 120                 if (key.objectid >= block_group->key.objectid +
 121                     block_group->key.offset) {
 122                         break;
 123                 }
 124
 125                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 126                         if (!found) {
 127                                 last = first_free;
 128                                 found = 1;
 129                         }
 130                         if (key.objectid > last) {
 131                                 hole_size = key.objectid - last;
 132                                 set_extent_dirty(free_space_cache, last,
 133                                                  last + hole_size - 1,
 134                                                  GFP_NOFS);
 135                         }
 136                         last = key.objectid + key.offset;
 137                 }
 138 next:
 139                 path->slots[0]++;
 140         }
 141
 142         if (!found)
 143                 last = first_free;
 144         if (block_group->key.objectid +
 145             block_group->key.offset > last) {
 146                 hole_size = block_group->key.objectid +
 147                         block_group->key.offset - last;
 148                 set_extent_dirty(free_space_cache, last,
 149                                  last + hole_size - 1, GFP_NOFS);
 150         }
 151         block_group->cached = 1;
 152 err:
 153         btrfs_free_path(path);
 154         return 0;
 155 }
 156
 157 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 158                                                          btrfs_fs_info *info,
 159                                                          u64 bytenr)
 160 {
 161         struct extent_map_tree *block_group_cache;
 162         struct btrfs_block_group_cache *block_group = NULL;
 163         u64 ptr;
 164         u64 start;
 165         u64 end;
 166         int ret;
 167
 168         block_group_cache = &info->block_group_cache;
 169         ret = find_first_extent_bit(block_group_cache,
 170                                     bytenr, &start, &end,
 171                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
 172         if (ret) {
 173                 return NULL;
 174         }
 175         ret = get_state_private(block_group_cache, start, &ptr);
 176         if (ret)
 177                 return NULL;
 178
 179         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 180         if (block_group->key.objectid <= bytenr && bytenr <
 181             block_group->key.objectid + block_group->key.offset)
 182                 return block_group;
 183         return NULL;
 184 }
 185 static u64 noinline find_search_start(struct btrfs_root *root,
 186                               struct btrfs_block_group_cache **cache_ret,
 187                               u64 search_start, int num, int data)
 188 {
 189         int ret;
 190         struct btrfs_block_group_cache *cache = *cache_ret;
 191         u64 last;
 192         u64 start = 0;
 193         u64 end = 0;
 194         u64 cache_miss = 0;
 195         int wrapped = 0;
 196
 197         if (!cache) {
 198                 goto out;
 199         }
 200 again:
 201         ret = cache_block_group(root, cache);
 202         if (ret)
 203                 goto out;
 204
 205         last = max(search_start, cache->key.objectid);
 206
 207         while(1) {
 208                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 209                                             last, &start, &end, EXTENT_DIRTY);
 210                 if (ret) {
 211                         if (!cache_miss)
 212                                 cache_miss = last;
 213                         goto new_group;
 214                 }
 215
 216                 start = max(last, start);
 217                 last = end + 1;
 218                 if (last - start < num) {
 219                         if (last == cache->key.objectid + cache->key.offset)
 220                                 cache_miss = start;
 221                         continue;
 222                 }
 223                 if (data != BTRFS_BLOCK_GROUP_MIXED &&
 224                     start + num > cache->key.objectid + cache->key.offset)
 225                         goto new_group;
 226                 return start;
 227         }
 228 out:
 229         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 230         if (!cache) {
 231                 printk("Unable to find block group for %Lu\n",
 232                        search_start);
 233                 WARN_ON(1);
 234                 return search_start;
 235         }
 236         return search_start;
 237
 238 new_group:
 239         last = cache->key.objectid + cache->key.offset;
 240 wrapped:
 241         cache = btrfs_lookup_block_group(root->fs_info, last);
 242         if (!cache) {
 243 no_cache:
 244                 if (!wrapped) {
 245                         wrapped = 1;
 246                         last = search_start;
 247                         data = BTRFS_BLOCK_GROUP_MIXED;
 248                         goto wrapped;
 249                 }
 250                 goto out;
 251         }
 252         if (cache_miss && !cache->cached) {
 253                 cache_block_group(root, cache);
 254                 last = cache_miss;
 255                 cache = btrfs_lookup_block_group(root->fs_info, last);
 256         }
 257         cache = btrfs_find_block_group(root, cache, last, data, 0);
 258         if (!cache)
 259                 goto no_cache;
 260         *cache_ret = cache;
 261         cache_miss = 0;
 262         goto again;
 263 }
 264
 265 static u64 div_factor(u64 num, int factor)
 266 {
 267         if (factor == 10)
 268                 return num;
 269         num *= factor;
 270         do_div(num, 10);
 271         return num;
 272 }
 273
 274 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 275                                                  struct btrfs_block_group_cache
 276                                                  *hint, u64 search_start,
 277                                                  int data, int owner)
 278 {
 279         struct btrfs_block_group_cache *cache;
 280         struct extent_map_tree *block_group_cache;
 281         struct btrfs_block_group_cache *found_group = NULL;
 282         struct btrfs_fs_info *info = root->fs_info;
 283         u64 used;
 284         u64 last = 0;
 285         u64 hint_last;
 286         u64 start;
 287         u64 end;
 288         u64 free_check;
 289         u64 ptr;
 290         int bit;
 291         int ret;
 292         int full_search = 0;
 293         int factor = 8;
 294         int data_swap = 0;
 295
 296         block_group_cache = &info->block_group_cache;
 297
 298         if (!owner)
 299                 factor = 8;
 300
 301         if (data == BTRFS_BLOCK_GROUP_MIXED) {
 302                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 303                 factor = 10;
 304         } else if (data)
 305                 bit = BLOCK_GROUP_DATA;
 306         else
 307                 bit = BLOCK_GROUP_METADATA;
 308
 309         if (search_start) {
 310                 struct btrfs_block_group_cache *shint;
 311                 shint = btrfs_lookup_block_group(info, search_start);
 312                 if (shint && (shint->data == data ||
 313                               shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 314                         used = btrfs_block_group_used(&shint->item);
 315                         if (used + shint->pinned <
 316                             div_factor(shint->key.offset, factor)) {
 317                                 return shint;
 318                         }
 319                 }
 320         }
 321         if (hint && (hint->data == data ||
 322                      hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 323                 used = btrfs_block_group_used(&hint->item);
 324                 if (used + hint->pinned <
 325                     div_factor(hint->key.offset, factor)) {
 326                         return hint;
 327                 }
 328                 last = hint->key.objectid + hint->key.offset;
 329                 hint_last = last;
 330         } else {
 331                 if (hint)
 332                         hint_last = max(hint->key.objectid, search_start);
 333                 else
 334                         hint_last = search_start;
 335
 336                 last = hint_last;
 337         }
 338 again:
 339         while(1) {
 340                 ret = find_first_extent_bit(block_group_cache, last,
 341                                             &start, &end, bit);
 342                 if (ret)
 343                         break;
 344
 345                 ret = get_state_private(block_group_cache, start, &ptr);
 346                 if (ret)
 347                         break;
 348
 349                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 350                 last = cache->key.objectid + cache->key.offset;
 351                 used = btrfs_block_group_used(&cache->item);
 352
 353                 if (full_search)
 354                         free_check = cache->key.offset;
 355                 else
 356                         free_check = div_factor(cache->key.offset, factor);
 357                 if (used + cache->pinned < free_check) {
 358                         found_group = cache;
 359                         goto found;
 360                 }
 361                 cond_resched();
 362         }
 363         if (!full_search) {
 364                 last = search_start;
 365                 full_search = 1;
 366                 goto again;
 367         }
 368         if (!data_swap) {
 369                 data_swap = 1;
 370                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 371                 last = search_start;
 372                 goto again;
 373         }
 374 found:
 375         return found_group;
 376 }
 377
 378 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 379                            u64 owner, u64 owner_offset)
 380 {
 381         u32 high_crc = ~(u32)0;
 382         u32 low_crc = ~(u32)0;
 383         __le64 lenum;
 384
 385         lenum = cpu_to_le64(root_objectid);
 386         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 387         lenum = cpu_to_le64(ref_generation);
 388         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 389
 390 #if 0
 391         lenum = cpu_to_le64(owner);
 392         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 393         lenum = cpu_to_le64(owner_offset);
 394         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 395 #endif
 396         return ((u64)high_crc << 32) | (u64)low_crc;
 397 }
 398
 399 static int match_extent_ref(struct extent_buffer *leaf,
 400                             struct btrfs_extent_ref *disk_ref,
 401                             struct btrfs_extent_ref *cpu_ref)
 402 {
 403         int ret;
 404         int len;
 405
 406         if (cpu_ref->objectid)
 407                 len = sizeof(*cpu_ref);
 408         else
 409                 len = 2 * sizeof(u64);
 410         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 411                                    len);
 412         return ret == 0;
 413 }
 414
 415 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 416                                           struct btrfs_root *root,
 417                                           struct btrfs_path *path, u64 bytenr,
 418                                           u64 root_objectid,
 419                                           u64 ref_generation, u64 owner,
 420                                           u64 owner_offset, int del)
 421 {
 422         u64 hash;
 423         struct btrfs_key key;
 424         struct btrfs_key found_key;
 425         struct btrfs_extent_ref ref;
 426         struct extent_buffer *leaf;
 427         struct btrfs_extent_ref *disk_ref;
 428         int ret;
 429         int ret2;
 430
 431         btrfs_set_stack_ref_root(&ref, root_objectid);
 432         btrfs_set_stack_ref_generation(&ref, ref_generation);
 433         btrfs_set_stack_ref_objectid(&ref, owner);
 434         btrfs_set_stack_ref_offset(&ref, owner_offset);
 435
 436         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 437                                owner_offset);
 438         key.offset = hash;
 439         key.objectid = bytenr;
 440         key.type = BTRFS_EXTENT_REF_KEY;
 441
 442         while (1) {
 443                 ret = btrfs_search_slot(trans, root, &key, path,
 444                                         del ? -1 : 0, del);
 445                 if (ret < 0)
 446                         goto out;
 447                 leaf = path->nodes[0];
 448                 if (ret != 0) {
 449                         u32 nritems = btrfs_header_nritems(leaf);
 450                         if (path->slots[0] >= nritems) {
 451                                 ret2 = btrfs_next_leaf(root, path);
 452                                 if (ret2)
 453                                         goto out;
 454                                 leaf = path->nodes[0];
 455                         }
 456                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 457                         if (found_key.objectid != bytenr ||
 458                             found_key.type != BTRFS_EXTENT_REF_KEY)
 459                                 goto out;
 460                         key.offset = found_key.offset;
 461                         if (del) {
 462                                 btrfs_release_path(root, path);
 463                                 continue;
 464                         }
 465                 }
 466                 disk_ref = btrfs_item_ptr(path->nodes[0],
 467                                           path->slots[0],
 468                                           struct btrfs_extent_ref);
 469                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 470                         ret = 0;
 471                         goto out;
 472                 }
 473                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 474                 key.offset = found_key.offset + 1;
 475                 btrfs_release_path(root, path);
 476         }
 477 out:
 478         return ret;
 479 }
 480
 481 /*
 482  * Back reference rules.  Back refs have three main goals:
 483  *
 484  * 1) differentiate between all holders of references to an extent so that
 485  *    when a reference is dropped we can make sure it was a valid reference
 486  *    before freeing the extent.
 487  *
 488  * 2) Provide enough information to quickly find the holders of an extent
 489  *    if we notice a given block is corrupted or bad.
 490  *
 491  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 492  *    maintenance.  This is actually the same as #2, but with a slightly
 493  *    different use case.
 494  *
 495  * File extents can be referenced by:
 496  *
 497  * - multiple snapshots, subvolumes, or different generations in one subvol
 498  * - different files inside a single subvolume (in theory, not implemented yet)
 499  * - different offsets inside a file (bookend extents in file.c)
 500  *
 501  * The extent ref structure has fields for:
 502  *
 503  * - Objectid of the subvolume root
 504  * - Generation number of the tree holding the reference
 505  * - objectid of the file holding the reference
 506  * - offset in the file corresponding to the key holding the reference
 507  *
 508  * When a file extent is allocated the fields are filled in:
 509  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 510  *
 511  * When a leaf is cow'd new references are added for every file extent found
 512  * in the leaf.  It looks the same as the create case, but trans->transid
 513  * will be different when the block is cow'd.
 514  *
 515  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 516  *
 517  * When a file extent is removed either during snapshot deletion or file
 518  * truncation, the corresponding back reference is found
 519  * by searching for:
 520  *
 521  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 522  *      inode objectid, offset in file)
 523  *
 524  * Btree extents can be referenced by:
 525  *
 526  * - Different subvolumes
 527  * - Different generations of the same subvolume
 528  *
 529  * Storing sufficient information for a full reverse mapping of a btree
 530  * block would require storing the lowest key of the block in the backref,
 531  * and it would require updating that lowest key either before write out or
 532  * every time it changed.  Instead, the objectid of the lowest key is stored
 533  * along with the level of the tree block.  This provides a hint
 534  * about where in the btree the block can be found.  Searches through the
 535  * btree only need to look for a pointer to that block, so they stop one
 536  * level higher than the level recorded in the backref.
 537  *
 538  * Some btrees do not do reference counting on their extents.  These
 539  * include the extent tree and the tree of tree roots.  Backrefs for these
 540  * trees always have a generation of zero.
 541  *
 542  * When a tree block is created, back references are inserted:
 543  *
 544  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 545  *
 546  * When a tree block is cow'd in a reference counted root,
 547  * new back references are added for all the blocks it points to.
 548  * These are of the form (trans->transid will have increased since creation):
 549  *
 550  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 551  *
 552  * Because the lowest_key_objectid and the level are just hints
 553  * they are not used when backrefs are deleted.  When a backref is deleted:
 554  *
 555  * if backref was for a tree root:
 556  *     root_objectid = root->root_key.objectid
 557  * else
 558  *     root_objectid = btrfs_header_owner(parent)
 559  *
 560  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 561  *
 562  * Back Reference Key hashing:
 563  *
 564  * Back references have four fields, each 64 bits long.  Unfortunately,
 565  * This is hashed into a single 64 bit number and placed into the key offset.
 566  * The key objectid corresponds to the first byte in the extent, and the
 567  * key type is set to BTRFS_EXTENT_REF_KEY
 568  */
 569 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 570                                  struct btrfs_root *root,
 571                                  struct btrfs_path *path, u64 bytenr,
 572                                  u64 root_objectid, u64 ref_generation,
 573                                  u64 owner, u64 owner_offset)
 574 {
 575         u64 hash;
 576         struct btrfs_key key;
 577         struct btrfs_extent_ref ref;
 578         struct btrfs_extent_ref *disk_ref;
 579         int ret;
 580
 581         btrfs_set_stack_ref_root(&ref, root_objectid);
 582         btrfs_set_stack_ref_generation(&ref, ref_generation);
 583         btrfs_set_stack_ref_objectid(&ref, owner);
 584         btrfs_set_stack_ref_offset(&ref, owner_offset);
 585
 586         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 587                                owner_offset);
 588         key.offset = hash;
 589         key.objectid = bytenr;
 590         key.type = BTRFS_EXTENT_REF_KEY;
 591
 592         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 593         while (ret == -EEXIST) {
 594                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 595                                           struct btrfs_extent_ref);
 596                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 597                         goto out;
 598                 key.offset++;
 599                 btrfs_release_path(root, path);
 600                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 601                                               sizeof(ref));
 602         }
 603         if (ret)
 604                 goto out;
 605         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 606                                   struct btrfs_extent_ref);
 607         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 608                             sizeof(ref));
 609         btrfs_mark_buffer_dirty(path->nodes[0]);
 610 out:
 611         btrfs_release_path(root, path);
 612         return ret;
 613 }
 614
 615 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 616                                 struct btrfs_root *root,
 617                                 u64 bytenr, u64 num_bytes,
 618                                 u64 root_objectid, u64 ref_generation,
 619                                 u64 owner, u64 owner_offset)
 620 {
 621         struct btrfs_path *path;
 622         int ret;
 623         struct btrfs_key key;
 624         struct extent_buffer *l;
 625         struct btrfs_extent_item *item;
 626         u32 refs;
 627
 628         WARN_ON(num_bytes < root->sectorsize);
 629         path = btrfs_alloc_path();
 630         if (!path)
 631                 return -ENOMEM;
 632
 633         key.objectid = bytenr;
 634         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 635         key.offset = num_bytes;
 636         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 637                                 0, 1);
 638         if (ret < 0)
 639                 return ret;
 640         if (ret != 0) {
 641                 BUG();
 642         }
 643         BUG_ON(ret != 0);
 644         l = path->nodes[0];
 645         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 646         refs = btrfs_extent_refs(l, item);
 647         btrfs_set_extent_refs(l, item, refs + 1);
 648         btrfs_mark_buffer_dirty(path->nodes[0]);
 649
 650         btrfs_release_path(root->fs_info->extent_root, path);
 651
 652         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 653                                           path, bytenr, root_objectid,
 654                                           ref_generation, owner, owner_offset);
 655         BUG_ON(ret);
 656         finish_current_insert(trans, root->fs_info->extent_root);
 657         del_pending_extents(trans, root->fs_info->extent_root);
 658
 659         btrfs_free_path(path);
 660         return 0;
 661 }
 662
 663 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 664                          struct btrfs_root *root)
 665 {
 666         finish_current_insert(trans, root->fs_info->extent_root);
 667         del_pending_extents(trans, root->fs_info->extent_root);
 668         return 0;
 669 }
 670
 671 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 672                              struct btrfs_root *root, u64 bytenr,
 673                              u64 num_bytes, u32 *refs)
 674 {
 675         struct btrfs_path *path;
 676         int ret;
 677         struct btrfs_key key;
 678         struct extent_buffer *l;
 679         struct btrfs_extent_item *item;
 680
 681         WARN_ON(num_bytes < root->sectorsize);
 682         path = btrfs_alloc_path();
 683         key.objectid = bytenr;
 684         key.offset = num_bytes;
 685         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 686         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 687                                 0, 0);
 688         if (ret < 0)
 689                 goto out;
 690         if (ret != 0) {
 691                 btrfs_print_leaf(root, path->nodes[0]);
 692                 printk("failed to find block number %Lu\n", bytenr);
 693                 BUG();
 694         }
 695         l = path->nodes[0];
 696         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 697         *refs = btrfs_extent_refs(l, item);
 698 out:
 699         btrfs_free_path(path);
 700         return 0;
 701 }
 702
 703 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 704                                   struct btrfs_path *count_path,
 705                                   u64 first_extent)
 706 {
 707         struct btrfs_root *extent_root = root->fs_info->extent_root;
 708         struct btrfs_path *path;
 709         u64 bytenr;
 710         u64 found_objectid;
 711         u64 root_objectid = root->root_key.objectid;
 712         u32 total_count = 0;
 713         u32 cur_count;
 714         u32 nritems;
 715         int ret;
 716         struct btrfs_key key;
 717         struct btrfs_key found_key;
 718         struct extent_buffer *l;
 719         struct btrfs_extent_item *item;
 720         struct btrfs_extent_ref *ref_item;
 721         int level = -1;
 722
 723         path = btrfs_alloc_path();
 724 again:
 725         if (level == -1)
 726                 bytenr = first_extent;
 727         else
 728                 bytenr = count_path->nodes[level]->start;
 729
 730         cur_count = 0;
 731         key.objectid = bytenr;
 732         key.offset = 0;
 733
 734         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 735         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 736         if (ret < 0)
 737                 goto out;
 738         BUG_ON(ret == 0);
 739
 740         l = path->nodes[0];
 741         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 742
 743         if (found_key.objectid != bytenr ||
 744             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 745                 goto out;
 746         }
 747
 748         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 749         while (1) {
 750                 l = path->nodes[0];
 751                 nritems = btrfs_header_nritems(l);
 752                 if (path->slots[0] >= nritems) {
 753                         ret = btrfs_next_leaf(extent_root, path);
 754                         if (ret == 0)
 755                                 continue;
 756                         break;
 757                 }
 758                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 759                 if (found_key.objectid != bytenr)
 760                         break;
 761
 762                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 763                         path->slots[0]++;
 764                         continue;
 765                 }
 766
 767                 cur_count++;
 768                 ref_item = btrfs_item_ptr(l, path->slots[0],
 769                                           struct btrfs_extent_ref);
 770                 found_objectid = btrfs_ref_root(l, ref_item);
 771
 772                 if (found_objectid != root_objectid) {
 773                         total_count = 2;
 774                         goto out;
 775                 }
 776                 total_count = 1;
 777                 path->slots[0]++;
 778         }
 779         if (cur_count == 0) {
 780                 total_count = 0;
 781                 goto out;
 782         }
 783         if (level >= 0 && root->node == count_path->nodes[level])
 784                 goto out;
 785         level++;
 786         btrfs_release_path(root, path);
 787         goto again;
 788
 789 out:
 790         btrfs_free_path(path);
 791         return total_count;
 792 }
 793 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 794                        struct btrfs_root *root, u64 owner_objectid)
 795 {
 796         u64 generation;
 797         u64 key_objectid;
 798         u64 level;
 799         u32 nritems;
 800         struct btrfs_disk_key disk_key;
 801
 802         level = btrfs_header_level(root->node);
 803         generation = trans->transid;
 804         nritems = btrfs_header_nritems(root->node);
 805         if (nritems > 0) {
 806                 if (level == 0)
 807                         btrfs_item_key(root->node, &disk_key, 0);
 808                 else
 809                         btrfs_node_key(root->node, &disk_key, 0);
 810                 key_objectid = btrfs_disk_key_objectid(&disk_key);
 811         } else {
 812                 key_objectid = 0;
 813         }
 814         return btrfs_inc_extent_ref(trans, root, root->node->start,
 815                                     root->node->len, owner_objectid,
 816                                     generation, level, key_objectid);
 817 }
 818
 819 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 820                   struct extent_buffer *buf)
 821 {
 822         u64 bytenr;
 823         u32 nritems;
 824         struct btrfs_key key;
 825         struct btrfs_file_extent_item *fi;
 826         int i;
 827         int level;
 828         int ret;
 829         int faili;
 830
 831         if (!root->ref_cows)
 832                 return 0;
 833
 834         level = btrfs_header_level(buf);
 835         nritems = btrfs_header_nritems(buf);
 836         for (i = 0; i < nritems; i++) {
 837                 if (level == 0) {
 838                         u64 disk_bytenr;
 839                         btrfs_item_key_to_cpu(buf, &key, i);
 840                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 841                                 continue;
 842                         fi = btrfs_item_ptr(buf, i,
 843                                             struct btrfs_file_extent_item);
 844                         if (btrfs_file_extent_type(buf, fi) ==
 845                             BTRFS_FILE_EXTENT_INLINE)
 846                                 continue;
 847                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 848                         if (disk_bytenr == 0)
 849                                 continue;
 850                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
 851                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 852                                     root->root_key.objectid, trans->transid,
 853                                     key.objectid, key.offset);
 854                         if (ret) {
 855                                 faili = i;
 856                                 goto fail;
 857                         }
 858                 } else {
 859                         bytenr = btrfs_node_blockptr(buf, i);
 860                         btrfs_node_key_to_cpu(buf, &key, i);
 861                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
 862                                            btrfs_level_size(root, level - 1),
 863                                            root->root_key.objectid,
 864                                            trans->transid,
 865                                            level - 1, key.objectid);
 866                         if (ret) {
 867                                 faili = i;
 868                                 goto fail;
 869                         }
 870                 }
 871         }
 872         return 0;
 873 fail:
 874         WARN_ON(1);
 875 #if 0
 876         for (i =0; i < faili; i++) {
 877                 if (level == 0) {
 878                         u64 disk_bytenr;
 879                         btrfs_item_key_to_cpu(buf, &key, i);
 880                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 881                                 continue;
 882                         fi = btrfs_item_ptr(buf, i,
 883                                             struct btrfs_file_extent_item);
 884                         if (btrfs_file_extent_type(buf, fi) ==
 885                             BTRFS_FILE_EXTENT_INLINE)
 886                                 continue;
 887                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 888                         if (disk_bytenr == 0)
 889                                 continue;
 890                         err = btrfs_free_extent(trans, root, disk_bytenr,
 891                                     btrfs_file_extent_disk_num_bytes(buf,
 892                                                                       fi), 0);
 893                         BUG_ON(err);
 894                 } else {
 895                         bytenr = btrfs_node_blockptr(buf, i);
 896                         err = btrfs_free_extent(trans, root, bytenr,
 897                                         btrfs_level_size(root, level - 1), 0);
 898                         BUG_ON(err);
 899                 }
 900         }
 901 #endif
 902         return ret;
 903 }
 904
 905 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 906                                  struct btrfs_root *root,
 907                                  struct btrfs_path *path,
 908                                  struct btrfs_block_group_cache *cache)
 909 {
 910         int ret;
 911         int pending_ret;
 912         struct btrfs_root *extent_root = root->fs_info->extent_root;
 913         unsigned long bi;
 914         struct extent_buffer *leaf;
 915
 916         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 917         if (ret < 0)
 918                 goto fail;
 919         BUG_ON(ret);
 920
 921         leaf = path->nodes[0];
 922         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 923         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 924         btrfs_mark_buffer_dirty(leaf);
 925         btrfs_release_path(extent_root, path);
 926 fail:
 927         finish_current_insert(trans, extent_root);
 928         pending_ret = del_pending_extents(trans, extent_root);
 929         if (ret)
 930                 return ret;
 931         if (pending_ret)
 932                 return pending_ret;
 933         return 0;
 934
 935 }
 936
 937 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 938                                    struct btrfs_root *root)
 939 {
 940         struct extent_map_tree *block_group_cache;
 941         struct btrfs_block_group_cache *cache;
 942         int ret;
 943         int err = 0;
 944         int werr = 0;
 945         struct btrfs_path *path;
 946         u64 last = 0;
 947         u64 start;
 948         u64 end;
 949         u64 ptr;
 950
 951         block_group_cache = &root->fs_info->block_group_cache;
 952         path = btrfs_alloc_path();
 953         if (!path)
 954                 return -ENOMEM;
 955
 956         while(1) {
 957                 ret = find_first_extent_bit(block_group_cache, last,
 958                                             &start, &end, BLOCK_GROUP_DIRTY);
 959                 if (ret)
 960                         break;
 961
 962                 last = end + 1;
 963                 ret = get_state_private(block_group_cache, start, &ptr);
 964                 if (ret)
 965                         break;
 966
 967                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 968                 err = write_one_cache_group(trans, root,
 969                                             path, cache);
 970                 /*
 971                  * if we fail to write the cache group, we want
 972                  * to keep it marked dirty in hopes that a later
 973                  * write will work
 974                  */
 975                 if (err) {
 976                         werr = err;
 977                         continue;
 978                 }
 979                 clear_extent_bits(block_group_cache, start, end,
 980                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 981         }
 982         btrfs_free_path(path);
 983         return werr;
 984 }
 985
 986 static int update_block_group(struct btrfs_trans_handle *trans,
 987                               struct btrfs_root *root,
 988                               u64 bytenr, u64 num_bytes, int alloc,
 989                               int mark_free, int data)
 990 {
 991         struct btrfs_block_group_cache *cache;
 992         struct btrfs_fs_info *info = root->fs_info;
 993         u64 total = num_bytes;
 994         u64 old_val;
 995         u64 byte_in_group;
 996         u64 start;
 997         u64 end;
 998
 999         while(total) {
1000                 cache = btrfs_lookup_block_group(info, bytenr);
1001                 if (!cache) {
1002                         return -1;
1003                 }
1004                 byte_in_group = bytenr - cache->key.objectid;
1005                 WARN_ON(byte_in_group > cache->key.offset);
1006                 start = cache->key.objectid;
1007                 end = start + cache->key.offset - 1;
1008                 set_extent_bits(&info->block_group_cache, start, end,
1009                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1010
1011                 old_val = btrfs_block_group_used(&cache->item);
1012                 num_bytes = min(total, cache->key.offset - byte_in_group);
1013                 if (alloc) {
1014                         if (cache->data != data &&
1015                             old_val < (cache->key.offset >> 1)) {
1016                                 int bit_to_clear;
1017                                 int bit_to_set;
1018                                 cache->data = data;
1019                                 if (data) {
1020                                         bit_to_clear = BLOCK_GROUP_METADATA;
1021                                         bit_to_set = BLOCK_GROUP_DATA;
1022                                         cache->item.flags &=
1023                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1024                                         cache->item.flags |=
1025                                                 BTRFS_BLOCK_GROUP_DATA;
1026                                 } else {
1027                                         bit_to_clear = BLOCK_GROUP_DATA;
1028                                         bit_to_set = BLOCK_GROUP_METADATA;
1029                                         cache->item.flags &=
1030                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1031                                         cache->item.flags &=
1032                                                 ~BTRFS_BLOCK_GROUP_DATA;
1033                                 }
1034                                 clear_extent_bits(&info->block_group_cache,
1035                                                   start, end, bit_to_clear,
1036                                                   GFP_NOFS);
1037                                 set_extent_bits(&info->block_group_cache,
1038                                                 start, end, bit_to_set,
1039                                                 GFP_NOFS);
1040                         } else if (cache->data != data &&
1041                                    cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1042                                 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1043                                 set_extent_bits(&info->block_group_cache,
1044                                                 start, end,
1045                                                 BLOCK_GROUP_DATA |
1046                                                 BLOCK_GROUP_METADATA,
1047                                                 GFP_NOFS);
1048                         }
1049                         old_val += num_bytes;
1050                 } else {
1051                         old_val -= num_bytes;
1052                         if (mark_free) {
1053                                 set_extent_dirty(&info->free_space_cache,
1054                                                  bytenr, bytenr + num_bytes - 1,
1055                                                  GFP_NOFS);
1056                         }
1057                 }
1058                 btrfs_set_block_group_used(&cache->item, old_val);
1059                 total -= num_bytes;
1060                 bytenr += num_bytes;
1061         }
1062         return 0;
1063 }
1064 static int update_pinned_extents(struct btrfs_root *root,
1065                                 u64 bytenr, u64 num, int pin)
1066 {
1067         u64 len;
1068         struct btrfs_block_group_cache *cache;
1069         struct btrfs_fs_info *fs_info = root->fs_info;
1070
1071         if (pin) {
1072                 set_extent_dirty(&fs_info->pinned_extents,
1073                                 bytenr, bytenr + num - 1, GFP_NOFS);
1074         } else {
1075                 clear_extent_dirty(&fs_info->pinned_extents,
1076                                 bytenr, bytenr + num - 1, GFP_NOFS);
1077         }
1078         while (num > 0) {
1079                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1080                 WARN_ON(!cache);
1081                 len = min(num, cache->key.offset -
1082                           (bytenr - cache->key.objectid));
1083                 if (pin) {
1084                         cache->pinned += len;
1085                         fs_info->total_pinned += len;
1086                 } else {
1087                         cache->pinned -= len;
1088                         fs_info->total_pinned -= len;
1089                 }
1090                 bytenr += len;
1091                 num -= len;
1092         }
1093         return 0;
1094 }
1095
1096 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
1097 {
1098         u64 last = 0;
1099         u64 start;
1100         u64 end;
1101         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
1102         int ret;
1103
1104         while(1) {
1105                 ret = find_first_extent_bit(pinned_extents, last,
1106                                             &start, &end, EXTENT_DIRTY);
1107                 if (ret)
1108                         break;
1109                 set_extent_dirty(copy, start, end, GFP_NOFS);
1110                 last = end + 1;
1111         }
1112         return 0;
1113 }
1114
1115 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1116                                struct btrfs_root *root,
1117                                struct extent_map_tree *unpin)
1118 {
1119         u64 start;
1120         u64 end;
1121         int ret;
1122         struct extent_map_tree *free_space_cache;
1123         free_space_cache = &root->fs_info->free_space_cache;
1124
1125         while(1) {
1126                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1127                                             EXTENT_DIRTY);
1128                 if (ret)
1129                         break;
1130                 update_pinned_extents(root, start, end + 1 - start, 0);
1131                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1132                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1133         }
1134         return 0;
1135 }
1136
1137 static int finish_current_insert(struct btrfs_trans_handle *trans,
1138                                  struct btrfs_root *extent_root)
1139 {
1140         u64 start;
1141         u64 end;
1142         struct btrfs_fs_info *info = extent_root->fs_info;
1143         struct extent_buffer *eb;
1144         struct btrfs_path *path;
1145         struct btrfs_key ins;
1146         struct btrfs_disk_key first;
1147         struct btrfs_extent_item extent_item;
1148         int ret;
1149         int level;
1150         int err = 0;
1151
1152         btrfs_set_stack_extent_refs(&extent_item, 1);
1153         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1154         path = btrfs_alloc_path();
1155
1156         while(1) {
1157                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1158                                             &end, EXTENT_LOCKED);
1159                 if (ret)
1160                         break;
1161
1162                 ins.objectid = start;
1163                 ins.offset = end + 1 - start;
1164                 err = btrfs_insert_item(trans, extent_root, &ins,
1165                                         &extent_item, sizeof(extent_item));
1166                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1167                                   GFP_NOFS);
1168                 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1169                 level = btrfs_header_level(eb);
1170                 if (level == 0) {
1171                         btrfs_item_key(eb, &first, 0);
1172                 } else {
1173                         btrfs_node_key(eb, &first, 0);
1174                 }
1175                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1176                                           start, extent_root->root_key.objectid,
1177                                           0, level,
1178                                           btrfs_disk_key_objectid(&first));
1179                 BUG_ON(err);
1180                 free_extent_buffer(eb);
1181         }
1182         btrfs_free_path(path);
1183         return 0;
1184 }
1185
1186 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1187                           int pending)
1188 {
1189         int err = 0;
1190         struct extent_buffer *buf;
1191
1192         if (!pending) {
1193                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1194                 if (buf) {
1195                         if (btrfs_buffer_uptodate(buf)) {
1196                                 u64 transid =
1197                                     root->fs_info->running_transaction->transid;
1198                                 u64 header_transid =
1199                                         btrfs_header_generation(buf);
1200                                 if (header_transid == transid) {
1201                                         free_extent_buffer(buf);
1202                                         return 1;
1203                                 }
1204                         }
1205                         free_extent_buffer(buf);
1206                 }
1207                 update_pinned_extents(root, bytenr, num_bytes, 1);
1208         } else {
1209                 set_extent_bits(&root->fs_info->pending_del,
1210                                 bytenr, bytenr + num_bytes - 1,
1211                                 EXTENT_LOCKED, GFP_NOFS);
1212         }
1213         BUG_ON(err < 0);
1214         return 0;
1215 }
1216
1217 /*
1218  * remove an extent from the root, returns 0 on success
1219  */
1220 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1221                          *root, u64 bytenr, u64 num_bytes,
1222                          u64 root_objectid, u64 ref_generation,
1223                          u64 owner_objectid, u64 owner_offset, int pin,
1224                          int mark_free)
1225 {
1226         struct btrfs_path *path;
1227         struct btrfs_key key;
1228         struct btrfs_fs_info *info = root->fs_info;
1229         struct btrfs_root *extent_root = info->extent_root;
1230         struct extent_buffer *leaf;
1231         int ret;
1232         struct btrfs_extent_item *ei;
1233         u32 refs;
1234
1235         key.objectid = bytenr;
1236         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1237         key.offset = num_bytes;
1238
1239         path = btrfs_alloc_path();
1240         if (!path)
1241                 return -ENOMEM;
1242
1243         ret = lookup_extent_backref(trans, extent_root, path,
1244                                     bytenr, root_objectid,
1245                                     ref_generation,
1246                                     owner_objectid, owner_offset, 1);
1247         if (ret == 0) {
1248                 ret = btrfs_del_item(trans, extent_root, path);
1249         } else {
1250                 btrfs_print_leaf(extent_root, path->nodes[0]);
1251                 WARN_ON(1);
1252                 printk("Unable to find ref byte nr %Lu root %Lu "
1253                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1254                        root_objectid, ref_generation, owner_objectid,
1255                        owner_offset);
1256         }
1257         btrfs_release_path(extent_root, path);
1258         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1259         if (ret < 0)
1260                 return ret;
1261         BUG_ON(ret);
1262
1263         leaf = path->nodes[0];
1264         ei = btrfs_item_ptr(leaf, path->slots[0],
1265                             struct btrfs_extent_item);
1266         refs = btrfs_extent_refs(leaf, ei);
1267         BUG_ON(refs == 0);
1268         refs -= 1;
1269         btrfs_set_extent_refs(leaf, ei, refs);
1270         btrfs_mark_buffer_dirty(leaf);
1271
1272         if (refs == 0) {
1273                 u64 super_used;
1274                 u64 root_used;
1275
1276                 if (pin) {
1277                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1278                         if (ret > 0)
1279                                 mark_free = 1;
1280                         BUG_ON(ret < 0);
1281                 }
1282
1283                 /* block accounting for super block */
1284                 super_used = btrfs_super_bytes_used(&info->super_copy);
1285                 btrfs_set_super_bytes_used(&info->super_copy,
1286                                            super_used - num_bytes);
1287
1288                 /* block accounting for root item */
1289                 root_used = btrfs_root_used(&root->root_item);
1290                 btrfs_set_root_used(&root->root_item,
1291                                            root_used - num_bytes);
1292
1293                 ret = btrfs_del_item(trans, extent_root, path);
1294                 if (ret) {
1295                         return ret;
1296                 }
1297                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1298                                          mark_free, 0);
1299                 BUG_ON(ret);
1300         }
1301         btrfs_free_path(path);
1302         finish_current_insert(trans, extent_root);
1303         return ret;
1304 }
1305
1306 /*
1307  * find all the blocks marked as pending in the radix tree and remove
1308  * them from the extent map
1309  */
1310 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1311                                btrfs_root *extent_root)
1312 {
1313         int ret;
1314         int err = 0;
1315         u64 start;
1316         u64 end;
1317         struct extent_map_tree *pending_del;
1318         struct extent_map_tree *pinned_extents;
1319
1320         pending_del = &extent_root->fs_info->pending_del;
1321         pinned_extents = &extent_root->fs_info->pinned_extents;
1322
1323         while(1) {
1324                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1325                                             EXTENT_LOCKED);
1326                 if (ret)
1327                         break;
1328                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1329                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1330                                   GFP_NOFS);
1331                 ret = __free_extent(trans, extent_root,
1332                                      start, end + 1 - start,
1333                                      extent_root->root_key.objectid,
1334                                      0, 0, 0, 0, 0);
1335                 if (ret)
1336                         err = ret;
1337         }
1338         return err;
1339 }
1340
1341 /*
1342  * remove an extent from the root, returns 0 on success
1343  */
1344 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1345                       *root, u64 bytenr, u64 num_bytes,
1346                       u64 root_objectid, u64 ref_generation,
1347                       u64 owner_objectid, u64 owner_offset, int pin)
1348 {
1349         struct btrfs_root *extent_root = root->fs_info->extent_root;
1350         int pending_ret;
1351         int ret;
1352
1353         WARN_ON(num_bytes < root->sectorsize);
1354         if (!root->ref_cows)
1355                 ref_generation = 0;
1356
1357         if (root == extent_root) {
1358                 pin_down_bytes(root, bytenr, num_bytes, 1);
1359                 return 0;
1360         }
1361         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1362                             ref_generation, owner_objectid, owner_offset,
1363                             pin, pin == 0);
1364         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1365         return ret ? ret : pending_ret;
1366 }
1367
1368 static u64 stripe_align(struct btrfs_root *root, u64 val)
1369 {
1370         u64 mask = ((u64)root->stripesize - 1);
1371         u64 ret = (val + mask) & ~mask;
1372         return ret;
1373 }
1374
1375 /*
1376  * walks the btree of allocated extents and find a hole of a given size.
1377  * The key ins is changed to record the hole:
1378  * ins->objectid == block start
1379  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1380  * ins->offset == number of blocks
1381  * Any available blocks before search_start are skipped.
1382  */
1383 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1384                                      struct btrfs_root *orig_root,
1385                                      u64 num_bytes, u64 empty_size,
1386                                      u64 search_start, u64 search_end,
1387                                      u64 hint_byte, struct btrfs_key *ins,
1388                                      u64 exclude_start, u64 exclude_nr,
1389                                      int data)
1390 {
1391         struct btrfs_path *path;
1392         struct btrfs_key key;
1393         u64 hole_size = 0;
1394         u64 aligned;
1395         int ret;
1396         int slot = 0;
1397         u64 last_byte = 0;
1398         u64 orig_search_start = search_start;
1399         int start_found;
1400         struct extent_buffer *l;
1401         struct btrfs_root * root = orig_root->fs_info->extent_root;
1402         struct btrfs_fs_info *info = root->fs_info;
1403         u64 total_needed = num_bytes;
1404         int level;
1405         struct btrfs_block_group_cache *block_group;
1406         int full_scan = 0;
1407         int wrapped = 0;
1408         u64 cached_start;
1409
1410         WARN_ON(num_bytes < root->sectorsize);
1411         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1412
1413         level = btrfs_header_level(root->node);
1414
1415         if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1416                 data = BTRFS_BLOCK_GROUP_MIXED;
1417         }
1418
1419         if (search_end == (u64)-1)
1420                 search_end = btrfs_super_total_bytes(&info->super_copy);
1421         if (hint_byte) {
1422                 block_group = btrfs_lookup_block_group(info, hint_byte);
1423                 if (!block_group)
1424                         hint_byte = search_start;
1425                 block_group = btrfs_find_block_group(root, block_group,
1426                                                      hint_byte, data, 1);
1427         } else {
1428                 block_group = btrfs_find_block_group(root,
1429                                                      trans->block_group,
1430                                                      search_start, data, 1);
1431         }
1432
1433         total_needed += empty_size;
1434         path = btrfs_alloc_path();
1435 check_failed:
1436         if (!block_group) {
1437                 block_group = btrfs_lookup_block_group(info, search_start);
1438                 if (!block_group)
1439                         block_group = btrfs_lookup_block_group(info,
1440                                                        orig_search_start);
1441         }
1442         search_start = find_search_start(root, &block_group, search_start,
1443                                          total_needed, data);
1444         search_start = stripe_align(root, search_start);
1445         cached_start = search_start;
1446         btrfs_init_path(path);
1447         ins->objectid = search_start;
1448         ins->offset = 0;
1449         start_found = 0;
1450         path->reada = 2;
1451
1452         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1453         if (ret < 0)
1454                 goto error;
1455         ret = find_previous_extent(root, path);
1456         if (ret < 0)
1457                 goto error;
1458         l = path->nodes[0];
1459         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1460         while (1) {
1461                 l = path->nodes[0];
1462                 slot = path->slots[0];
1463                 if (slot >= btrfs_header_nritems(l)) {
1464                         ret = btrfs_next_leaf(root, path);
1465                         if (ret == 0)
1466                                 continue;
1467                         if (ret < 0)
1468                                 goto error;
1469
1470                         search_start = max(search_start,
1471                                            block_group->key.objectid);
1472                         if (!start_found) {
1473                                 aligned = stripe_align(root, search_start);
1474                                 ins->objectid = aligned;
1475                                 if (aligned >= search_end) {
1476                                         ret = -ENOSPC;
1477                                         goto error;
1478                                 }
1479                                 ins->offset = search_end - aligned;
1480                                 start_found = 1;
1481                                 goto check_pending;
1482                         }
1483                         ins->objectid = stripe_align(root,
1484                                                      last_byte > search_start ?
1485                                                      last_byte : search_start);
1486                         if (search_end <= ins->objectid) {
1487                                 ret = -ENOSPC;
1488                                 goto error;
1489                         }
1490                         ins->offset = search_end - ins->objectid;
1491                         BUG_ON(ins->objectid >= search_end);
1492                         goto check_pending;
1493                 }
1494                 btrfs_item_key_to_cpu(l, &key, slot);
1495
1496                 if (key.objectid >= search_start && key.objectid > last_byte &&
1497                     start_found) {
1498                         if (last_byte < search_start)
1499                                 last_byte = search_start;
1500                         aligned = stripe_align(root, last_byte);
1501                         hole_size = key.objectid - aligned;
1502                         if (key.objectid > aligned && hole_size >= num_bytes) {
1503                                 ins->objectid = aligned;
1504                                 ins->offset = hole_size;
1505                                 goto check_pending;
1506                         }
1507                 }
1508                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1509                         if (!start_found && btrfs_key_type(&key) ==
1510                             BTRFS_BLOCK_GROUP_ITEM_KEY) {
1511                                 last_byte = key.objectid;
1512                                 start_found = 1;
1513                         }
1514                         goto next;
1515                 }
1516
1517
1518                 start_found = 1;
1519                 last_byte = key.objectid + key.offset;
1520
1521                 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1522                     last_byte >= block_group->key.objectid +
1523                     block_group->key.offset) {
1524                         btrfs_release_path(root, path);
1525                         search_start = block_group->key.objectid +
1526                                 block_group->key.offset;
1527                         goto new_group;
1528                 }
1529 next:
1530                 path->slots[0]++;
1531                 cond_resched();
1532         }
1533 check_pending:
1534         /* we have to make sure we didn't find an extent that has already
1535          * been allocated by the map tree or the original allocation
1536          */
1537         btrfs_release_path(root, path);
1538         BUG_ON(ins->objectid < search_start);
1539
1540         if (ins->objectid + num_bytes >= search_end)
1541                 goto enospc;
1542         if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1543             ins->objectid + num_bytes > block_group->
1544             key.objectid + block_group->key.offset) {
1545                 search_start = block_group->key.objectid +
1546                         block_group->key.offset;
1547                 goto new_group;
1548         }
1549         if (test_range_bit(&info->extent_ins, ins->objectid,
1550                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1551                 search_start = ins->objectid + num_bytes;
1552                 goto new_group;
1553         }
1554         if (test_range_bit(&info->pinned_extents, ins->objectid,
1555                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1556                 search_start = ins->objectid + num_bytes;
1557                 goto new_group;
1558         }
1559         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1560             ins->objectid < exclude_start + exclude_nr)) {
1561                 search_start = exclude_start + exclude_nr;
1562                 goto new_group;
1563         }
1564         if (!data) {
1565                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1566                 if (block_group)
1567                         trans->block_group = block_group;
1568         }
1569         ins->offset = num_bytes;
1570         btrfs_free_path(path);
1571         return 0;
1572
1573 new_group:
1574         if (search_start + num_bytes >= search_end) {
1575 enospc:
1576                 search_start = orig_search_start;
1577                 if (full_scan) {
1578                         ret = -ENOSPC;
1579                         goto error;
1580                 }
1581                 if (wrapped) {
1582                         if (!full_scan)
1583                                 total_needed -= empty_size;
1584                         full_scan = 1;
1585                         data = BTRFS_BLOCK_GROUP_MIXED;
1586                 } else
1587                         wrapped = 1;
1588         }
1589         block_group = btrfs_lookup_block_group(info, search_start);
1590         cond_resched();
1591         block_group = btrfs_find_block_group(root, block_group,
1592                                              search_start, data, 0);
1593         goto check_failed;
1594
1595 error:
1596         btrfs_release_path(root, path);
1597         btrfs_free_path(path);
1598         return ret;
1599 }
1600 /*
1601  * finds a free extent and does all the dirty work required for allocation
1602  * returns the key for the extent through ins, and a tree buffer for
1603  * the first block of the extent through buf.
1604  *
1605  * returns 0 if everything worked, non-zero otherwise.
1606  */
1607 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1608                        struct btrfs_root *root,
1609                        u64 num_bytes, u64 root_objectid, u64 ref_generation,
1610                        u64 owner, u64 owner_offset,
1611                        u64 empty_size, u64 hint_byte,
1612                        u64 search_end, struct btrfs_key *ins, int data)
1613 {
1614         int ret;
1615         int pending_ret;
1616         u64 super_used, root_used;
1617         u64 search_start = 0;
1618         u64 new_hint;
1619         struct btrfs_fs_info *info = root->fs_info;
1620         struct btrfs_root *extent_root = info->extent_root;
1621         struct btrfs_extent_item extent_item;
1622         struct btrfs_path *path;
1623
1624         btrfs_set_stack_extent_refs(&extent_item, 1);
1625
1626         new_hint = max(hint_byte, root->fs_info->alloc_start);
1627         if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1628                 hint_byte = new_hint;
1629
1630         WARN_ON(num_bytes < root->sectorsize);
1631         ret = find_free_extent(trans, root, num_bytes, empty_size,
1632                                search_start, search_end, hint_byte, ins,
1633                                trans->alloc_exclude_start,
1634                                trans->alloc_exclude_nr, data);
1635         BUG_ON(ret);
1636         if (ret)
1637                 return ret;
1638
1639         /* block accounting for super block */
1640         super_used = btrfs_super_bytes_used(&info->super_copy);
1641         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1642
1643         /* block accounting for root item */
1644         root_used = btrfs_root_used(&root->root_item);
1645         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1646
1647         clear_extent_dirty(&root->fs_info->free_space_cache,
1648                            ins->objectid, ins->objectid + ins->offset - 1,
1649                            GFP_NOFS);
1650
1651         if (root == extent_root) {
1652                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1653                                 ins->objectid + ins->offset - 1,
1654                                 EXTENT_LOCKED, GFP_NOFS);
1655                 WARN_ON(data == 1);
1656                 goto update_block;
1657         }
1658
1659         WARN_ON(trans->alloc_exclude_nr);
1660         trans->alloc_exclude_start = ins->objectid;
1661         trans->alloc_exclude_nr = ins->offset;
1662         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1663                                 sizeof(extent_item));
1664
1665         trans->alloc_exclude_start = 0;
1666         trans->alloc_exclude_nr = 0;
1667         BUG_ON(ret);
1668
1669         path = btrfs_alloc_path();
1670         BUG_ON(!path);
1671         ret = btrfs_insert_extent_backref(trans, extent_root, path,
1672                                           ins->objectid, root_objectid,
1673                                           ref_generation, owner, owner_offset);
1674
1675         BUG_ON(ret);
1676         btrfs_free_path(path);
1677         finish_current_insert(trans, extent_root);
1678         pending_ret = del_pending_extents(trans, extent_root);
1679
1680         if (ret) {
1681                 return ret;
1682         }
1683         if (pending_ret) {
1684                 return pending_ret;
1685         }
1686
1687 update_block:
1688         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1689                                  data);
1690         BUG_ON(ret);
1691         return 0;
1692 }
1693
1694 /*
1695  * helper function to allocate a block for a given tree
1696  * returns the tree buffer or NULL.
1697  */
1698 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1699                                              struct btrfs_root *root,
1700                                              u32 blocksize,
1701                                              u64 root_objectid, u64 hint,
1702                                              u64 empty_size)
1703 {
1704         u64 ref_generation;
1705
1706         if (root->ref_cows)
1707                 ref_generation = trans->transid;
1708         else
1709                 ref_generation = 0;
1710
1711
1712         return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1713                                         ref_generation, 0, 0, hint, empty_size);
1714 }
1715
1716 /*
1717  * helper function to allocate a block for a given tree
1718  * returns the tree buffer or NULL.
1719  */
1720 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1721                                              struct btrfs_root *root,
1722                                              u32 blocksize,
1723                                              u64 root_objectid,
1724                                              u64 ref_generation,
1725                                              u64 first_objectid,
1726                                              int level,
1727                                              u64 hint,
1728                                              u64 empty_size)
1729 {
1730         struct btrfs_key ins;
1731         int ret;
1732         struct extent_buffer *buf;
1733
1734         ret = btrfs_alloc_extent(trans, root, blocksize,
1735                                  root_objectid, ref_generation,
1736                                  level, first_objectid, empty_size, hint,
1737                                  (u64)-1, &ins, 0);
1738         if (ret) {
1739                 BUG_ON(ret > 0);
1740                 return ERR_PTR(ret);
1741         }
1742         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1743         if (!buf) {
1744                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1745                                   root->root_key.objectid, ref_generation,
1746                                   0, 0, 0);
1747                 return ERR_PTR(-ENOMEM);
1748         }
1749         btrfs_set_buffer_uptodate(buf);
1750         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1751                          buf->start + buf->len - 1, GFP_NOFS);
1752         set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1753                         buf->start, buf->start + buf->len - 1,
1754                         EXTENT_CSUM, GFP_NOFS);
1755         buf->flags |= EXTENT_CSUM;
1756         btrfs_set_buffer_defrag(buf);
1757         trans->blocks_used++;
1758         return buf;
1759 }
1760
1761 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1762                                   struct btrfs_root *root,
1763                                   struct extent_buffer *leaf)
1764 {
1765         u64 leaf_owner;
1766         u64 leaf_generation;
1767         struct btrfs_key key;
1768         struct btrfs_file_extent_item *fi;
1769         int i;
1770         int nritems;
1771         int ret;
1772
1773         BUG_ON(!btrfs_is_leaf(leaf));
1774         nritems = btrfs_header_nritems(leaf);
1775         leaf_owner = btrfs_header_owner(leaf);
1776         leaf_generation = btrfs_header_generation(leaf);
1777
1778         for (i = 0; i < nritems; i++) {
1779                 u64 disk_bytenr;
1780
1781                 btrfs_item_key_to_cpu(leaf, &key, i);
1782                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1783                         continue;
1784                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1785                 if (btrfs_file_extent_type(leaf, fi) ==
1786                     BTRFS_FILE_EXTENT_INLINE)
1787                         continue;
1788                 /*
1789                  * FIXME make sure to insert a trans record that
1790                  * repeats the snapshot del on crash
1791                  */
1792                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1793                 if (disk_bytenr == 0)
1794                         continue;
1795                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1796                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
1797                                 leaf_owner, leaf_generation,
1798                                 key.objectid, key.offset, 0);
1799                 BUG_ON(ret);
1800         }
1801         return 0;
1802 }
1803
1804 static void noinline reada_walk_down(struct btrfs_root *root,
1805                                      struct extent_buffer *node)
1806 {
1807         int i;
1808         u32 nritems;
1809         u64 bytenr;
1810         int ret;
1811         u32 refs;
1812         int level;
1813         u32 blocksize;
1814
1815         nritems = btrfs_header_nritems(node);
1816         level = btrfs_header_level(node);
1817         for (i = 0; i < nritems; i++) {
1818                 bytenr = btrfs_node_blockptr(node, i);
1819                 blocksize = btrfs_level_size(root, level - 1);
1820                 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1821                 BUG_ON(ret);
1822                 if (refs != 1)
1823                         continue;
1824                 mutex_unlock(&root->fs_info->fs_mutex);
1825                 ret = readahead_tree_block(root, bytenr, blocksize);
1826                 cond_resched();
1827                 mutex_lock(&root->fs_info->fs_mutex);
1828                 if (ret)
1829                         break;
1830         }
1831 }
1832
1833 /*
1834  * helper function for drop_snapshot, this walks down the tree dropping ref
1835  * counts as it goes.
1836  */
1837 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1838                                    struct btrfs_root *root,
1839                                    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 noinline walk_up_tree(struct btrfs_trans_handle *trans,
1959                                  struct btrfs_root *root,
1960                                  struct btrfs_path *path, int *level)
1961 {
1962         u64 root_owner;
1963         u64 root_gen;
1964         struct btrfs_root_item *root_item = &root->root_item;
1965         int i;
1966         int slot;
1967         int ret;
1968
1969         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1970                 slot = path->slots[i];
1971                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1972                         struct extent_buffer *node;
1973                         struct btrfs_disk_key disk_key;
1974                         node = path->nodes[i];
1975                         path->slots[i]++;
1976                         *level = i;
1977                         WARN_ON(*level == 0);
1978                         btrfs_node_key(node, &disk_key, path->slots[i]);
1979                         memcpy(&root_item->drop_progress,
1980                                &disk_key, sizeof(disk_key));
1981                         root_item->drop_level = i;
1982                         return 0;
1983                 } else {
1984                         if (path->nodes[*level] == root->node) {
1985                                 root_owner = root->root_key.objectid;
1986                                 root_gen =
1987                                    btrfs_header_generation(path->nodes[*level]);
1988                         } else {
1989                                 struct extent_buffer *node;
1990                                 node = path->nodes[*level + 1];
1991                                 root_owner = btrfs_header_owner(node);
1992                                 root_gen = btrfs_header_generation(node);
1993                         }
1994                         ret = btrfs_free_extent(trans, root,
1995                                                 path->nodes[*level]->start,
1996                                                 path->nodes[*level]->len,
1997                                                 root_owner, root_gen, 0, 0, 1);
1998                         BUG_ON(ret);
1999                         free_extent_buffer(path->nodes[*level]);
2000                         path->nodes[*level] = NULL;
2001                         *level = i + 1;
2002                 }
2003         }
2004         return 1;
2005 }
2006
2007 /*
2008  * drop the reference count on the tree rooted at 'snap'.  This traverses
2009  * the tree freeing any blocks that have a ref count of zero after being
2010  * decremented.
2011  */
2012 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2013                         *root)
2014 {
2015         int ret = 0;
2016         int wret;
2017         int level;
2018         struct btrfs_path *path;
2019         int i;
2020         int orig_level;
2021         struct btrfs_root_item *root_item = &root->root_item;
2022
2023         path = btrfs_alloc_path();
2024         BUG_ON(!path);
2025
2026         level = btrfs_header_level(root->node);
2027         orig_level = level;
2028         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2029                 path->nodes[level] = root->node;
2030                 extent_buffer_get(root->node);
2031                 path->slots[level] = 0;
2032         } else {
2033                 struct btrfs_key key;
2034                 struct btrfs_disk_key found_key;
2035                 struct extent_buffer *node;
2036
2037                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2038                 level = root_item->drop_level;
2039                 path->lowest_level = level;
2040                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2041                 if (wret < 0) {
2042                         ret = wret;
2043                         goto out;
2044                 }
2045                 node = path->nodes[level];
2046                 btrfs_node_key(node, &found_key, path->slots[level]);
2047                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2048                                sizeof(found_key)));
2049         }
2050         while(1) {
2051                 wret = walk_down_tree(trans, root, path, &level);
2052                 if (wret > 0)
2053                         break;
2054                 if (wret < 0)
2055                         ret = wret;
2056
2057                 wret = walk_up_tree(trans, root, path, &level);
2058                 if (wret > 0)
2059                         break;
2060                 if (wret < 0)
2061                         ret = wret;
2062                 ret = -EAGAIN;
2063                 break;
2064         }
2065         for (i = 0; i <= orig_level; i++) {
2066                 if (path->nodes[i]) {
2067                         free_extent_buffer(path->nodes[i]);
2068                         path->nodes[i] = NULL;
2069                 }
2070         }
2071 out:
2072         btrfs_free_path(path);
2073         return ret;
2074 }
2075
2076 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2077 {
2078         u64 start;
2079         u64 end;
2080         u64 ptr;
2081         int ret;
2082         while(1) {
2083                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2084                                             &start, &end, (unsigned int)-1);
2085                 if (ret)
2086                         break;
2087                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2088                 if (!ret)
2089                         kfree((void *)(unsigned long)ptr);
2090                 clear_extent_bits(&info->block_group_cache, start,
2091                                   end, (unsigned int)-1, GFP_NOFS);
2092         }
2093         while(1) {
2094                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2095                                             &start, &end, EXTENT_DIRTY);
2096                 if (ret)
2097                         break;
2098                 clear_extent_dirty(&info->free_space_cache, start,
2099                                    end, GFP_NOFS);
2100         }
2101         return 0;
2102 }
2103
2104 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2105                                          u64 len)
2106 {
2107         u64 page_start;
2108         u64 page_end;
2109         u64 delalloc_start;
2110         u64 existing_delalloc;
2111         unsigned long last_index;
2112         unsigned long i;
2113         struct page *page;
2114         struct btrfs_root *root = BTRFS_I(inode)->root;
2115         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2116         struct file_ra_state *ra;
2117
2118         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2119
2120         mutex_lock(&inode->i_mutex);
2121         i = start >> PAGE_CACHE_SHIFT;
2122         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2123
2124         file_ra_state_init(ra, inode->i_mapping);
2125         btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2126         kfree(ra);
2127
2128         for (; i <= last_index; i++) {
2129                 page = grab_cache_page(inode->i_mapping, i);
2130                 if (!page)
2131                         goto out_unlock;
2132                 if (!PageUptodate(page)) {
2133                         btrfs_readpage(NULL, page);
2134                         lock_page(page);
2135                         if (!PageUptodate(page)) {
2136                                 unlock_page(page);
2137                                 page_cache_release(page);
2138                                 goto out_unlock;
2139                         }
2140                 }
2141                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2142                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2143
2144                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2145
2146                 delalloc_start = page_start;
2147                 existing_delalloc =
2148                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2149                                          &delalloc_start, page_end,
2150                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2151
2152                 set_extent_delalloc(em_tree, page_start,
2153                                     page_end, GFP_NOFS);
2154
2155                 spin_lock(&root->fs_info->delalloc_lock);
2156                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2157                                                  existing_delalloc;
2158                 spin_unlock(&root->fs_info->delalloc_lock);
2159
2160                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2161                 set_page_dirty(page);
2162                 unlock_page(page);
2163                 page_cache_release(page);
2164         }
2165
2166 out_unlock:
2167         mutex_unlock(&inode->i_mutex);
2168         return 0;
2169 }
2170
2171 /*
2172  * note, this releases the path
2173  */
2174 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2175                                   struct btrfs_path *path,
2176                                   struct btrfs_key *extent_key)
2177 {
2178         struct inode *inode;
2179         struct btrfs_root *found_root;
2180         struct btrfs_key *root_location;
2181         struct btrfs_extent_ref *ref;
2182         u64 ref_root;
2183         u64 ref_gen;
2184         u64 ref_objectid;
2185         u64 ref_offset;
2186         int ret;
2187
2188         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2189                              struct btrfs_extent_ref);
2190         ref_root = btrfs_ref_root(path->nodes[0], ref);
2191         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2192         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2193         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2194         btrfs_release_path(extent_root, path);
2195
2196         root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2197         root_location->objectid = ref_root;
2198         if (ref_gen == 0)
2199                 root_location->offset = 0;
2200         else
2201                 root_location->offset = (u64)-1;
2202         root_location->type = BTRFS_ROOT_ITEM_KEY;
2203
2204         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2205                                                 root_location);
2206         BUG_ON(!found_root);
2207         kfree(root_location);
2208
2209         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2210                 mutex_unlock(&extent_root->fs_info->fs_mutex);
2211                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2212                                           ref_objectid, found_root);
2213                 if (inode->i_state & I_NEW) {
2214                         /* the inode and parent dir are two different roots */
2215                         BTRFS_I(inode)->root = found_root;
2216                         BTRFS_I(inode)->location.objectid = ref_objectid;
2217                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2218                         BTRFS_I(inode)->location.offset = 0;
2219                         btrfs_read_locked_inode(inode);
2220                         unlock_new_inode(inode);
2221
2222                 }
2223                 /* this can happen if the reference is not against
2224                  * the latest version of the tree root
2225                  */
2226                 if (is_bad_inode(inode)) {
2227                         mutex_lock(&extent_root->fs_info->fs_mutex);
2228                         goto out;
2229                 }
2230                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2231                 /* FIXME, data=ordered will help get rid of this */
2232                 filemap_fdatawrite(inode->i_mapping);
2233                 iput(inode);
2234                 mutex_lock(&extent_root->fs_info->fs_mutex);
2235         } else {
2236                 struct btrfs_trans_handle *trans;
2237                 struct btrfs_key found_key;
2238                 struct extent_buffer *eb;
2239                 int level;
2240                 int i;
2241
2242                 trans = btrfs_start_transaction(found_root, 1);
2243                 eb = read_tree_block(found_root, extent_key->objectid,
2244                                      extent_key->offset);
2245                 level = btrfs_header_level(eb);
2246
2247                 if (level == 0)
2248                         btrfs_item_key_to_cpu(eb, &found_key, 0);
2249                 else
2250                         btrfs_node_key_to_cpu(eb, &found_key, 0);
2251
2252                 free_extent_buffer(eb);
2253
2254                 path->lowest_level = level;
2255                 path->reada = 2;
2256                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2257                                         0, 1);
2258                 path->lowest_level = 0;
2259                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2260                         if (!path->nodes[i])
2261                                 break;
2262                         free_extent_buffer(path->nodes[i]);
2263                         path->nodes[i] = NULL;
2264                 }
2265                 btrfs_release_path(found_root, path);
2266                 btrfs_end_transaction(trans, found_root);
2267         }
2268
2269 out:
2270         return 0;
2271 }
2272
2273 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2274                                         struct btrfs_path *path,
2275                                         struct btrfs_key *extent_key)
2276 {
2277         struct btrfs_key key;
2278         struct btrfs_key found_key;
2279         struct extent_buffer *leaf;
2280         u32 nritems;
2281         u32 item_size;
2282         int ret = 0;
2283
2284         key.objectid = extent_key->objectid;
2285         key.type = BTRFS_EXTENT_REF_KEY;
2286         key.offset = 0;
2287
2288         while(1) {
2289                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2290
2291                 BUG_ON(ret == 0);
2292
2293                 if (ret < 0)
2294                         goto out;
2295
2296                 ret = 0;
2297                 leaf = path->nodes[0];
2298                 nritems = btrfs_header_nritems(leaf);
2299                 if (path->slots[0] == nritems)
2300                         goto out;
2301
2302                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2303                 if (found_key.objectid != extent_key->objectid)
2304                         break;
2305
2306                 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2307                         break;
2308
2309                 key.offset = found_key.offset + 1;
2310                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2311
2312                 ret = relocate_one_reference(extent_root, path, extent_key);
2313                 if (ret)
2314                         goto out;
2315         }
2316         ret = 0;
2317 out:
2318         btrfs_release_path(extent_root, path);
2319         return ret;
2320 }
2321
2322 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2323 {
2324         struct btrfs_trans_handle *trans;
2325         struct btrfs_root *tree_root = root->fs_info->tree_root;
2326         struct btrfs_path *path;
2327         u64 cur_byte;
2328         u64 total_found;
2329         struct btrfs_fs_info *info = root->fs_info;
2330         struct extent_map_tree *block_group_cache;
2331         struct btrfs_key key;
2332         struct btrfs_key found_key;
2333         struct extent_buffer *leaf;
2334         u32 nritems;
2335         int ret;
2336         int progress = 0;
2337
2338         btrfs_set_super_total_bytes(&info->super_copy, new_size);
2339         block_group_cache = &info->block_group_cache;
2340         path = btrfs_alloc_path();
2341         root = root->fs_info->extent_root;
2342         path->reada = 2;
2343
2344 again:
2345         total_found = 0;
2346         key.objectid = new_size;
2347         key.offset = 0;
2348         key.type = 0;
2349         cur_byte = key.objectid;
2350
2351         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2352         if (ret < 0)
2353                 goto out;
2354
2355         ret = find_previous_extent(root, path);
2356         if (ret < 0)
2357                 goto out;
2358         if (ret == 0) {
2359                 leaf = path->nodes[0];
2360                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2361                 if (found_key.objectid + found_key.offset > new_size) {
2362                         cur_byte = found_key.objectid;
2363                         key.objectid = cur_byte;
2364                 }
2365         }
2366         btrfs_release_path(root, path);
2367
2368         while(1) {
2369                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2370                 if (ret < 0)
2371                         goto out;
2372
2373                 leaf = path->nodes[0];
2374                 nritems = btrfs_header_nritems(leaf);
2375 next:
2376                 if (path->slots[0] >= nritems) {
2377                         ret = btrfs_next_leaf(root, path);
2378                         if (ret < 0)
2379                                 goto out;
2380                         if (ret == 1) {
2381                                 ret = 0;
2382                                 break;
2383                         }
2384                         leaf = path->nodes[0];
2385                         nritems = btrfs_header_nritems(leaf);
2386                 }
2387
2388                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2389
2390                 if (progress && need_resched()) {
2391                         memcpy(&key, &found_key, sizeof(key));
2392                         mutex_unlock(&root->fs_info->fs_mutex);
2393                         cond_resched();
2394                         mutex_lock(&root->fs_info->fs_mutex);
2395                         btrfs_release_path(root, path);
2396                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
2397                         progress = 0;
2398                         goto next;
2399                 }
2400                 progress = 1;
2401
2402                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2403                     found_key.objectid + found_key.offset <= cur_byte) {
2404                         path->slots[0]++;
2405                         goto next;
2406                 }
2407
2408                 total_found++;
2409                 cur_byte = found_key.objectid + found_key.offset;
2410                 key.objectid = cur_byte;
2411                 btrfs_release_path(root, path);
2412                 ret = relocate_one_extent(root, path, &found_key);
2413         }
2414
2415         btrfs_release_path(root, path);
2416
2417         if (total_found > 0) {
2418                 trans = btrfs_start_transaction(tree_root, 1);
2419                 btrfs_commit_transaction(trans, tree_root);
2420
2421                 mutex_unlock(&root->fs_info->fs_mutex);
2422                 btrfs_clean_old_snapshots(tree_root);
2423                 mutex_lock(&root->fs_info->fs_mutex);
2424
2425                 trans = btrfs_start_transaction(tree_root, 1);
2426                 btrfs_commit_transaction(trans, tree_root);
2427                 goto again;
2428         }
2429
2430         trans = btrfs_start_transaction(root, 1);
2431         key.objectid = new_size;
2432         key.offset = 0;
2433         key.type = 0;
2434         while(1) {
2435                 u64 ptr;
2436
2437                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2438                 if (ret < 0)
2439                         goto out;
2440
2441                 leaf = path->nodes[0];
2442                 nritems = btrfs_header_nritems(leaf);
2443 bg_next:
2444                 if (path->slots[0] >= nritems) {
2445                         ret = btrfs_next_leaf(root, path);
2446                         if (ret < 0)
2447                                 break;
2448                         if (ret == 1) {
2449                                 ret = 0;
2450                                 break;
2451                         }
2452                         leaf = path->nodes[0];
2453                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2454
2455                         /*
2456                          * btrfs_next_leaf doesn't cow buffers, we have to
2457                          * do the search again
2458                          */
2459                         memcpy(&key, &found_key, sizeof(key));
2460                         btrfs_release_path(root, path);
2461                         goto resched_check;
2462                 }
2463
2464                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2465                 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2466                         printk("shrinker found key %Lu %u %Lu\n",
2467                                 found_key.objectid, found_key.type,
2468                                 found_key.offset);
2469                         path->slots[0]++;
2470                         goto bg_next;
2471                 }
2472                 ret = get_state_private(&info->block_group_cache,
2473                                         found_key.objectid, &ptr);
2474                 if (!ret)
2475                         kfree((void *)(unsigned long)ptr);
2476
2477                 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2478                                   found_key.objectid + found_key.offset - 1,
2479                                   (unsigned int)-1, GFP_NOFS);
2480
2481                 key.objectid = found_key.objectid + 1;
2482                 btrfs_del_item(trans, root, path);
2483                 btrfs_release_path(root, path);
2484 resched_check:
2485                 if (need_resched()) {
2486                         mutex_unlock(&root->fs_info->fs_mutex);
2487                         cond_resched();
2488                         mutex_lock(&root->fs_info->fs_mutex);
2489                 }
2490         }
2491         clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2492                            GFP_NOFS);
2493         btrfs_commit_transaction(trans, root);
2494 out:
2495         btrfs_free_path(path);
2496         return ret;
2497 }
2498
2499 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2500                            struct btrfs_root *root, u64 new_size)
2501 {
2502         struct btrfs_path *path;
2503         u64 nr = 0;
2504         u64 cur_byte;
2505         u64 old_size;
2506         unsigned long rem;
2507         struct btrfs_block_group_cache *cache;
2508         struct btrfs_block_group_item *item;
2509         struct btrfs_fs_info *info = root->fs_info;
2510         struct extent_map_tree *block_group_cache;
2511         struct btrfs_key key;
2512         struct extent_buffer *leaf;
2513         int ret;
2514         int bit;
2515
2516         old_size = btrfs_super_total_bytes(&info->super_copy);
2517         block_group_cache = &info->block_group_cache;
2518
2519         root = info->extent_root;
2520
2521         cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2522
2523         cur_byte = cache->key.objectid + cache->key.offset;
2524         if (cur_byte >= new_size)
2525                 goto set_size;
2526
2527         key.offset = BTRFS_BLOCK_GROUP_SIZE;
2528         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2529
2530         path = btrfs_alloc_path();
2531         if (!path)
2532                 return -ENOMEM;
2533
2534         while(cur_byte < new_size) {
2535                 key.objectid = cur_byte;
2536                 ret = btrfs_insert_empty_item(trans, root, path, &key,
2537                                         sizeof(struct btrfs_block_group_item));
2538                 BUG_ON(ret);
2539                 leaf = path->nodes[0];
2540                 item = btrfs_item_ptr(leaf, path->slots[0],
2541                                       struct btrfs_block_group_item);
2542
2543                 btrfs_set_disk_block_group_used(leaf, item, 0);
2544                 div_long_long_rem(nr, 3, &rem);
2545                 if (rem) {
2546                         btrfs_set_disk_block_group_flags(leaf, item,
2547                                                  BTRFS_BLOCK_GROUP_DATA);
2548                 } else {
2549                         btrfs_set_disk_block_group_flags(leaf, item, 0);
2550                 }
2551                 nr++;
2552
2553                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2554                 BUG_ON(!cache);
2555
2556                 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2557                                    sizeof(cache->item));
2558
2559                 memcpy(&cache->key, &key, sizeof(key));
2560                 cache->cached = 0;
2561                 cache->pinned = 0;
2562                 cur_byte = key.objectid + key.offset;
2563                 btrfs_release_path(root, path);
2564
2565                 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2566                         bit = BLOCK_GROUP_DATA;
2567                         cache->data = BTRFS_BLOCK_GROUP_DATA;
2568                 } else {
2569                         bit = BLOCK_GROUP_METADATA;
2570                         cache->data = 0;
2571                 }
2572
2573                 /* use EXTENT_LOCKED to prevent merging */
2574                 set_extent_bits(block_group_cache, key.objectid,
2575                                 key.objectid + key.offset - 1,
2576                                 bit | EXTENT_LOCKED, GFP_NOFS);
2577                 set_state_private(block_group_cache, key.objectid,
2578                                   (unsigned long)cache);
2579         }
2580         btrfs_free_path(path);
2581 set_size:
2582         btrfs_set_super_total_bytes(&info->super_copy, new_size);
2583         return 0;
2584 }
2585
2586 int btrfs_read_block_groups(struct btrfs_root *root)
2587 {
2588         struct btrfs_path *path;
2589         int ret;
2590         int err = 0;
2591         int bit;
2592         struct btrfs_block_group_cache *cache;
2593         struct btrfs_fs_info *info = root->fs_info;
2594         struct extent_map_tree *block_group_cache;
2595         struct btrfs_key key;
2596         struct btrfs_key found_key;
2597         struct extent_buffer *leaf;
2598
2599         block_group_cache = &info->block_group_cache;
2600
2601         root = info->extent_root;
2602         key.objectid = 0;
2603         key.offset = BTRFS_BLOCK_GROUP_SIZE;
2604         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2605
2606         path = btrfs_alloc_path();
2607         if (!path)
2608                 return -ENOMEM;
2609
2610         while(1) {
2611                 ret = btrfs_search_slot(NULL, info->extent_root,
2612                                         &key, path, 0, 0);
2613                 if (ret != 0) {
2614                         err = ret;
2615                         break;
2616                 }
2617                 leaf = path->nodes[0];
2618                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2619                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2620                 if (!cache) {
2621                         err = -1;
2622                         break;
2623                 }
2624
2625                 read_extent_buffer(leaf, &cache->item,
2626                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
2627                                    sizeof(cache->item));
2628                 memcpy(&cache->key, &found_key, sizeof(found_key));
2629                 cache->cached = 0;
2630                 cache->pinned = 0;
2631                 key.objectid = found_key.objectid + found_key.offset;
2632                 btrfs_release_path(root, path);
2633
2634                 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2635                         bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2636                         cache->data = BTRFS_BLOCK_GROUP_MIXED;
2637                 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2638                         bit = BLOCK_GROUP_DATA;
2639                         cache->data = BTRFS_BLOCK_GROUP_DATA;
2640                 } else {
2641                         bit = BLOCK_GROUP_METADATA;
2642                         cache->data = 0;
2643                 }
2644
2645                 /* use EXTENT_LOCKED to prevent merging */
2646                 set_extent_bits(block_group_cache, found_key.objectid,
2647                                 found_key.objectid + found_key.offset - 1,
2648                                 bit | EXTENT_LOCKED, GFP_NOFS);
2649                 set_state_private(block_group_cache, found_key.objectid,
2650                                   (unsigned long)cache);
2651
2652                 if (key.objectid >=
2653                     btrfs_super_total_bytes(&info->super_copy))
2654                         break;
2655         }
2656
2657         btrfs_free_path(path);
2658         return 0;
2659 }