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 <stdio.h>
  20 #include <stdlib.h>
  21 #include "kerncompat.h"
  22 #include "radix-tree.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27 #include "crc32c.h"
  28 #include "volumes.h"
  29
  30 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  32 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  33
  34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  35
  36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  37                                  btrfs_root *extent_root);
  38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  39                                btrfs_root *extent_root);
  40 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
  41                            struct btrfs_root *root, u64 bytes_used,
  42                            u64 type, u64 chunk_tree, u64 chunk_objectid,
  43                            u64 size);
  44
  45 static int cache_block_group(struct btrfs_root *root,
  46                              struct btrfs_block_group_cache *block_group)
  47 {
  48         struct btrfs_path *path;
  49         int ret;
  50         struct btrfs_key key;
  51         struct extent_buffer *leaf;
  52         struct extent_io_tree *free_space_cache;
  53         int slot;
  54         u64 last = 0;
  55         u64 hole_size;
  56         u64 first_free;
  57         int found = 0;
  58
  59         if (!block_group)
  60                 return 0;
  61
  62         root = root->fs_info->extent_root;
  63         free_space_cache = &root->fs_info->free_space_cache;
  64
  65         if (block_group->cached)
  66                 return 0;
  67
  68         path = btrfs_alloc_path();
  69         if (!path)
  70                 return -ENOMEM;
  71
  72         path->reada = 2;
  73         first_free = block_group->key.objectid;
  74         key.objectid = block_group->key.objectid;
  75         key.offset = 0;
  76         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  77         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  78         if (ret < 0)
  79                 return ret;
  80         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
  81         if (ret < 0)
  82                 return ret;
  83         if (ret == 0) {
  84                 leaf = path->nodes[0];
  85                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  86                 if (key.objectid + key.offset > first_free)
  87                         first_free = key.objectid + key.offset;
  88         }
  89         while(1) {
  90                 leaf = path->nodes[0];
  91                 slot = path->slots[0];
  92                 if (slot >= btrfs_header_nritems(leaf)) {
  93                         ret = btrfs_next_leaf(root, path);
  94                         if (ret < 0)
  95                                 goto err;
  96                         if (ret == 0) {
  97                                 continue;
  98                         } else {
  99                                 break;
 100                         }
 101                 }
 102                 btrfs_item_key_to_cpu(leaf, &key, slot);
 103                 if (key.objectid < block_group->key.objectid) {
 104                         goto next;
 105                 }
 106                 if (key.objectid >= block_group->key.objectid +
 107                     block_group->key.offset) {
 108                         break;
 109                 }
 110
 111                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 112                         if (!found) {
 113                                 last = first_free;
 114                                 found = 1;
 115                         }
 116                         if (key.objectid > last) {
 117                                 hole_size = key.objectid - last;
 118                                 set_extent_dirty(free_space_cache, last,
 119                                                  last + hole_size - 1,
 120                                                  GFP_NOFS);
 121                         }
 122                         last = key.objectid + key.offset;
 123                 }
 124 next:
 125                 path->slots[0]++;
 126         }
 127
 128         if (!found)
 129                 last = first_free;
 130         if (block_group->key.objectid +
 131             block_group->key.offset > last) {
 132                 hole_size = block_group->key.objectid +
 133                         block_group->key.offset - last;
 134                 set_extent_dirty(free_space_cache, last,
 135                                  last + hole_size - 1, GFP_NOFS);
 136         }
 137         block_group->cached = 1;
 138 err:
 139         btrfs_free_path(path);
 140         return 0;
 141 }
 142
 143 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 144                                                          btrfs_fs_info *info,
 145                                                          u64 bytenr)
 146 {
 147         struct extent_io_tree *block_group_cache;
 148         struct btrfs_block_group_cache *block_group = NULL;
 149         u64 ptr;
 150         u64 start;
 151         u64 end;
 152         int ret;
 153
 154         block_group_cache = &info->block_group_cache;
 155         ret = find_first_extent_bit(block_group_cache,
 156                                     bytenr, &start, &end,
 157                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 158                                     BLOCK_GROUP_SYSTEM);
 159         if (ret) {
 160                 return NULL;
 161         }
 162         ret = get_state_private(block_group_cache, start, &ptr);
 163         if (ret)
 164                 return NULL;
 165
 166         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 167         if (block_group->key.objectid <= bytenr && bytenr <
 168             block_group->key.objectid + block_group->key.offset)
 169                 return block_group;
 170         return NULL;
 171 }
 172
 173 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 174 {
 175         return (cache->flags & bits) == bits;
 176 }
 177
 178 static int noinline find_search_start(struct btrfs_root *root,
 179                               struct btrfs_block_group_cache **cache_ret,
 180                               u64 *start_ret, int num, int data)
 181 {
 182         int ret;
 183         struct btrfs_block_group_cache *cache = *cache_ret;
 184         u64 last;
 185         u64 start = 0;
 186         u64 end = 0;
 187         u64 cache_miss = 0;
 188         u64 search_start = *start_ret;
 189         int wrapped = 0;
 190
 191         if (!cache) {
 192                 goto out;
 193         }
 194 again:
 195         ret = cache_block_group(root, cache);
 196         if (ret)
 197                 goto out;
 198
 199         last = max(search_start, cache->key.objectid);
 200         if (!block_group_bits(cache, data)) {
 201                 goto new_group;
 202         }
 203
 204         while(1) {
 205                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 206                                             last, &start, &end, EXTENT_DIRTY);
 207                 if (ret) {
 208                         if (!cache_miss)
 209                                 cache_miss = last;
 210                         goto new_group;
 211                 }
 212
 213                 start = max(last, start);
 214                 last = end + 1;
 215                 if (last - start < num) {
 216                         if (last == cache->key.objectid + cache->key.offset)
 217                                 cache_miss = start;
 218                         continue;
 219                 }
 220                 if (start + num > cache->key.objectid + cache->key.offset)
 221                         goto new_group;
 222                 *start_ret = start;
 223                 return 0;
 224         }
 225 out:
 226         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 227         if (!cache) {
 228                 printk("Unable to find block group for %llu\n",
 229                         (unsigned long long)search_start);
 230                 WARN_ON(1);
 231         }
 232         return -ENOSPC;
 233
 234 new_group:
 235         last = cache->key.objectid + cache->key.offset;
 236 wrapped:
 237         cache = btrfs_lookup_block_group(root->fs_info, last);
 238         if (!cache) {
 239 no_cache:
 240                 if (!wrapped) {
 241                         wrapped = 1;
 242                         last = search_start;
 243                         goto wrapped;
 244                 }
 245                 goto out;
 246         }
 247         if (cache_miss && !cache->cached) {
 248                 cache_block_group(root, cache);
 249                 last = cache_miss;
 250                 cache = btrfs_lookup_block_group(root->fs_info, last);
 251         }
 252         cache = btrfs_find_block_group(root, cache, last, data, 0);
 253         if (!cache)
 254                 goto no_cache;
 255         *cache_ret = cache;
 256         cache_miss = 0;
 257         goto again;
 258 }
 259
 260 static u64 div_factor(u64 num, int factor)
 261 {
 262         if (factor == 10)
 263                 return num;
 264         num *= factor;
 265         num /= 10;
 266         return num;
 267 }
 268
 269 static int block_group_state_bits(u64 flags)
 270 {
 271         int bits = 0;
 272         if (flags & BTRFS_BLOCK_GROUP_DATA)
 273                 bits |= BLOCK_GROUP_DATA;
 274         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 275                 bits |= BLOCK_GROUP_METADATA;
 276         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 277                 bits |= BLOCK_GROUP_SYSTEM;
 278         return bits;
 279 }
 280
 281 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 282                                                  struct btrfs_block_group_cache
 283                                                  *hint, u64 search_start,
 284                                                  int data, int owner)
 285 {
 286         struct btrfs_block_group_cache *cache;
 287         struct extent_io_tree *block_group_cache;
 288         struct btrfs_block_group_cache *found_group = NULL;
 289         struct btrfs_fs_info *info = root->fs_info;
 290         u64 used;
 291         u64 last = 0;
 292         u64 hint_last;
 293         u64 start;
 294         u64 end;
 295         u64 free_check;
 296         u64 ptr;
 297         int bit;
 298         int ret;
 299         int full_search = 0;
 300         int factor = 8;
 301
 302         block_group_cache = &info->block_group_cache;
 303
 304         if (!owner)
 305                 factor = 8;
 306
 307         bit = block_group_state_bits(data);
 308
 309         if (search_start) {
 310                 struct btrfs_block_group_cache *shint;
 311                 shint = btrfs_lookup_block_group(info, search_start);
 312                 if (shint && block_group_bits(shint, data)) {
 313                         used = btrfs_block_group_used(&shint->item);
 314                         if (used + shint->pinned <
 315                             div_factor(shint->key.offset, factor)) {
 316                                 return shint;
 317                         }
 318                 }
 319         }
 320         if (hint && block_group_bits(hint, data)) {
 321                 used = btrfs_block_group_used(&hint->item);
 322                 if (used + hint->pinned <
 323                     div_factor(hint->key.offset, factor)) {
 324                         return hint;
 325                 }
 326                 last = hint->key.objectid + hint->key.offset;
 327                 hint_last = last;
 328         } else {
 329                 if (hint)
 330                         hint_last = max(hint->key.objectid, search_start);
 331                 else
 332                         hint_last = search_start;
 333
 334                 last = hint_last;
 335         }
 336 again:
 337         while(1) {
 338                 ret = find_first_extent_bit(block_group_cache, last,
 339                                             &start, &end, bit);
 340                 if (ret)
 341                         break;
 342
 343                 ret = get_state_private(block_group_cache, start, &ptr);
 344                 if (ret)
 345                         break;
 346
 347                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 348                 last = cache->key.objectid + cache->key.offset;
 349                 used = btrfs_block_group_used(&cache->item);
 350
 351                 if (block_group_bits(cache, data)) {
 352                         if (full_search)
 353                                 free_check = cache->key.offset;
 354                         else
 355                                 free_check = div_factor(cache->key.offset,
 356                                                         factor);
 357
 358                         if (used + cache->pinned < free_check) {
 359                                 found_group = cache;
 360                                 goto found;
 361                         }
 362                 }
 363                 cond_resched();
 364         }
 365         if (!full_search) {
 366                 last = search_start;
 367                 full_search = 1;
 368                 goto again;
 369         }
 370 found:
 371         return found_group;
 372 }
 373
 374 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 375                            u64 owner, u64 owner_offset)
 376 {
 377         u32 high_crc = ~(u32)0;
 378         u32 low_crc = ~(u32)0;
 379         __le64 lenum;
 380
 381         lenum = cpu_to_le64(root_objectid);
 382         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 383         lenum = cpu_to_le64(ref_generation);
 384         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 385         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 386                 lenum = cpu_to_le64(owner);
 387                 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 388                 lenum = cpu_to_le64(owner_offset);
 389                 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 390         }
 391         return ((u64)high_crc << 32) | (u64)low_crc;
 392 }
 393
 394 static int match_extent_ref(struct extent_buffer *leaf,
 395                             struct btrfs_extent_ref *disk_ref,
 396                             struct btrfs_extent_ref *cpu_ref)
 397 {
 398         int ret;
 399         int len;
 400
 401         if (cpu_ref->objectid)
 402                 len = sizeof(*cpu_ref);
 403         else
 404                 len = 2 * sizeof(u64);
 405         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 406                                    len);
 407         return ret == 0;
 408 }
 409
 410 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 411                                           struct btrfs_root *root,
 412                                           struct btrfs_path *path, u64 bytenr,
 413                                           u64 root_objectid,
 414                                           u64 ref_generation, u64 owner,
 415                                           u64 owner_offset, int del)
 416 {
 417         u64 hash;
 418         struct btrfs_key key;
 419         struct btrfs_key found_key;
 420         struct btrfs_extent_ref ref;
 421         struct extent_buffer *leaf;
 422         struct btrfs_extent_ref *disk_ref;
 423         int ret;
 424         int ret2;
 425
 426         btrfs_set_stack_ref_root(&ref, root_objectid);
 427         btrfs_set_stack_ref_generation(&ref, ref_generation);
 428         btrfs_set_stack_ref_objectid(&ref, owner);
 429         btrfs_set_stack_ref_offset(&ref, owner_offset);
 430
 431         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 432                                owner_offset);
 433         key.offset = hash;
 434         key.objectid = bytenr;
 435         key.type = BTRFS_EXTENT_REF_KEY;
 436
 437         while (1) {
 438                 ret = btrfs_search_slot(trans, root, &key, path,
 439                                         del ? -1 : 0, del);
 440                 if (ret < 0)
 441                         goto out;
 442                 leaf = path->nodes[0];
 443                 if (ret != 0) {
 444                         u32 nritems = btrfs_header_nritems(leaf);
 445                         if (path->slots[0] >= nritems) {
 446                                 ret2 = btrfs_next_leaf(root, path);
 447                                 if (ret2)
 448                                         goto out;
 449                                 leaf = path->nodes[0];
 450                         }
 451                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 452                         if (found_key.objectid != bytenr ||
 453                             found_key.type != BTRFS_EXTENT_REF_KEY)
 454                                 goto out;
 455                         key.offset = found_key.offset;
 456                         if (del) {
 457                                 btrfs_release_path(root, path);
 458                                 continue;
 459                         }
 460                 }
 461                 disk_ref = btrfs_item_ptr(path->nodes[0],
 462                                           path->slots[0],
 463                                           struct btrfs_extent_ref);
 464                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 465                         ret = 0;
 466                         goto out;
 467                 }
 468                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 469                 key.offset = found_key.offset + 1;
 470                 btrfs_release_path(root, path);
 471         }
 472 out:
 473         return ret;
 474 }
 475
 476 /*
 477  * Back reference rules.  Back refs have three main goals:
 478  *
 479  * 1) differentiate between all holders of references to an extent so that
 480  *    when a reference is dropped we can make sure it was a valid reference
 481  *    before freeing the extent.
 482  *
 483  * 2) Provide enough information to quickly find the holders of an extent
 484  *    if we notice a given block is corrupted or bad.
 485  *
 486  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 487  *    maintenance.  This is actually the same as #2, but with a slightly
 488  *    different use case.
 489  *
 490  * File extents can be referenced by:
 491  *
 492  * - multiple snapshots, subvolumes, or different generations in one subvol
 493  * - different files inside a single subvolume (in theory, not implemented yet)
 494  * - different offsets inside a file (bookend extents in file.c)
 495  *
 496  * The extent ref structure has fields for:
 497  *
 498  * - Objectid of the subvolume root
 499  * - Generation number of the tree holding the reference
 500  * - objectid of the file holding the reference
 501  * - offset in the file corresponding to the key holding the reference
 502  *
 503  * When a file extent is allocated the fields are filled in:
 504  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 505  *
 506  * When a leaf is cow'd new references are added for every file extent found
 507  * in the leaf.  It looks the same as the create case, but trans->transid
 508  * will be different when the block is cow'd.
 509  *
 510  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 511  *
 512  * When a file extent is removed either during snapshot deletion or file
 513  * truncation, the corresponding back reference is found
 514  * by searching for:
 515  *
 516  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 517  *      inode objectid, offset in file)
 518  *
 519  * Btree extents can be referenced by:
 520  *
 521  * - Different subvolumes
 522  * - Different generations of the same subvolume
 523  *
 524  * Storing sufficient information for a full reverse mapping of a btree
 525  * block would require storing the lowest key of the block in the backref,
 526  * and it would require updating that lowest key either before write out or
 527  * every time it changed.  Instead, the objectid of the lowest key is stored
 528  * along with the level of the tree block.  This provides a hint
 529  * about where in the btree the block can be found.  Searches through the
 530  * btree only need to look for a pointer to that block, so they stop one
 531  * level higher than the level recorded in the backref.
 532  *
 533  * Some btrees do not do reference counting on their extents.  These
 534  * include the extent tree and the tree of tree roots.  Backrefs for these
 535  * trees always have a generation of zero.
 536  *
 537  * When a tree block is created, back references are inserted:
 538  *
 539  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 540  *
 541  * When a tree block is cow'd in a reference counted root,
 542  * new back references are added for all the blocks it points to.
 543  * These are of the form (trans->transid will have increased since creation):
 544  *
 545  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 546  *
 547  * Because the lowest_key_objectid and the level are just hints
 548  * they are not used when backrefs are deleted.  When a backref is deleted:
 549  *
 550  * if backref was for a tree root:
 551  *     root_objectid = root->root_key.objectid
 552  * else
 553  *     root_objectid = btrfs_header_owner(parent)
 554  *
 555  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 556  *
 557  * Back Reference Key hashing:
 558  *
 559  * Back references have four fields, each 64 bits long.  Unfortunately,
 560  * This is hashed into a single 64 bit number and placed into the key offset.
 561  * The key objectid corresponds to the first byte in the extent, and the
 562  * key type is set to BTRFS_EXTENT_REF_KEY
 563  */
 564 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 565                                  struct btrfs_root *root,
 566                                  struct btrfs_path *path, u64 bytenr,
 567                                  u64 root_objectid, u64 ref_generation,
 568                                  u64 owner, u64 owner_offset)
 569 {
 570         u64 hash;
 571         struct btrfs_key key;
 572         struct btrfs_extent_ref ref;
 573         struct btrfs_extent_ref *disk_ref;
 574         int ret;
 575
 576         btrfs_set_stack_ref_root(&ref, root_objectid);
 577         btrfs_set_stack_ref_generation(&ref, ref_generation);
 578         btrfs_set_stack_ref_objectid(&ref, owner);
 579         btrfs_set_stack_ref_offset(&ref, owner_offset);
 580
 581         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 582                                owner_offset);
 583         key.offset = hash;
 584         key.objectid = bytenr;
 585         key.type = BTRFS_EXTENT_REF_KEY;
 586
 587         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 588         while (ret == -EEXIST) {
 589                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 590                                           struct btrfs_extent_ref);
 591                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 592                         goto out;
 593                 key.offset++;
 594                 btrfs_release_path(root, path);
 595                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 596                                               sizeof(ref));
 597         }
 598         if (ret)
 599                 goto out;
 600         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 601                                   struct btrfs_extent_ref);
 602         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 603                             sizeof(ref));
 604         btrfs_mark_buffer_dirty(path->nodes[0]);
 605 out:
 606         btrfs_release_path(root, path);
 607         return ret;
 608 }
 609
 610 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 611                                 struct btrfs_root *root,
 612                                 u64 bytenr, u64 num_bytes,
 613                                 u64 root_objectid, u64 ref_generation,
 614                                 u64 owner, u64 owner_offset)
 615 {
 616         struct btrfs_path *path;
 617         int ret;
 618         struct btrfs_key key;
 619         struct extent_buffer *l;
 620         struct btrfs_extent_item *item;
 621         u32 refs;
 622
 623         WARN_ON(num_bytes < root->sectorsize);
 624         path = btrfs_alloc_path();
 625         if (!path)
 626                 return -ENOMEM;
 627
 628         key.objectid = bytenr;
 629         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 630         key.offset = num_bytes;
 631         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 632                                 0, 1);
 633         if (ret < 0)
 634                 return ret;
 635         if (ret != 0) {
 636                 BUG();
 637         }
 638         BUG_ON(ret != 0);
 639         l = path->nodes[0];
 640         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 641         refs = btrfs_extent_refs(l, item);
 642         btrfs_set_extent_refs(l, item, refs + 1);
 643         btrfs_mark_buffer_dirty(path->nodes[0]);
 644
 645         btrfs_release_path(root->fs_info->extent_root, path);
 646
 647         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 648                                           path, bytenr, root_objectid,
 649                                           ref_generation, owner, owner_offset);
 650         BUG_ON(ret);
 651         finish_current_insert(trans, root->fs_info->extent_root);
 652         del_pending_extents(trans, root->fs_info->extent_root);
 653
 654         btrfs_free_path(path);
 655         return 0;
 656 }
 657
 658 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 659                          struct btrfs_root *root)
 660 {
 661         finish_current_insert(trans, root->fs_info->extent_root);
 662         del_pending_extents(trans, root->fs_info->extent_root);
 663         return 0;
 664 }
 665
 666 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 667                              struct btrfs_root *root, u64 bytenr,
 668                              u64 num_bytes, u32 *refs)
 669 {
 670         struct btrfs_path *path;
 671         int ret;
 672         struct btrfs_key key;
 673         struct extent_buffer *l;
 674         struct btrfs_extent_item *item;
 675
 676         WARN_ON(num_bytes < root->sectorsize);
 677         path = btrfs_alloc_path();
 678         key.objectid = bytenr;
 679         key.offset = num_bytes;
 680         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 681         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 682                                 0, 0);
 683         if (ret < 0)
 684                 goto out;
 685         if (ret != 0) {
 686                 btrfs_print_leaf(root, path->nodes[0]);
 687                 printk("failed to find block number %llu\n",
 688                         (unsigned long long)bytenr);
 689                 BUG();
 690         }
 691         l = path->nodes[0];
 692         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 693         *refs = btrfs_extent_refs(l, item);
 694 out:
 695         btrfs_free_path(path);
 696         return 0;
 697 }
 698
 699 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 700                                   struct btrfs_path *count_path,
 701                                   u64 first_extent)
 702 {
 703         struct btrfs_root *extent_root = root->fs_info->extent_root;
 704         struct btrfs_path *path;
 705         u64 bytenr;
 706         u64 found_objectid;
 707         u64 root_objectid = root->root_key.objectid;
 708         u32 total_count = 0;
 709         u32 cur_count;
 710         u32 refs;
 711         u32 nritems;
 712         int ret;
 713         struct btrfs_key key;
 714         struct btrfs_key found_key;
 715         struct extent_buffer *l;
 716         struct btrfs_extent_item *item;
 717         struct btrfs_extent_ref *ref_item;
 718         int level = -1;
 719
 720         path = btrfs_alloc_path();
 721 again:
 722         if (level == -1)
 723                 bytenr = first_extent;
 724         else
 725                 bytenr = count_path->nodes[level]->start;
 726
 727         cur_count = 0;
 728         key.objectid = bytenr;
 729         key.offset = 0;
 730
 731         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 732         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 733         if (ret < 0)
 734                 goto out;
 735         BUG_ON(ret == 0);
 736
 737         l = path->nodes[0];
 738         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 739
 740         if (found_key.objectid != bytenr ||
 741             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 742                 goto out;
 743         }
 744
 745         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 746         refs = btrfs_extent_refs(l, item);
 747         while (1) {
 748                 nritems = btrfs_header_nritems(l);
 749                 if (path->slots[0] >= nritems) {
 750                         ret = btrfs_next_leaf(extent_root, path);
 751                         if (ret == 0)
 752                                 continue;
 753                         break;
 754                 }
 755                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 756                 if (found_key.objectid != bytenr)
 757                         break;
 758                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 759                         path->slots[0]++;
 760                         continue;
 761                 }
 762
 763                 cur_count++;
 764                 ref_item = btrfs_item_ptr(l, path->slots[0],
 765                                           struct btrfs_extent_ref);
 766                 found_objectid = btrfs_ref_root(l, ref_item);
 767
 768                 if (found_objectid != root_objectid) {
 769                         total_count = 2;
 770                         goto out;
 771                 }
 772                 total_count = 1;
 773                 path->slots[0]++;
 774         }
 775         if (cur_count == 0) {
 776                 total_count = 0;
 777                 goto out;
 778         }
 779         if (level >= 0 && root->node == count_path->nodes[level])
 780                 goto out;
 781         level++;
 782         btrfs_release_path(root, path);
 783         goto again;
 784
 785 out:
 786         btrfs_free_path(path);
 787         return total_count;
 788 }
 789 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 790                        struct btrfs_root *root, u64 owner_objectid)
 791 {
 792         u64 generation;
 793         u64 key_objectid;
 794         u64 level;
 795         u32 nritems;
 796         struct btrfs_disk_key disk_key;
 797
 798         level = btrfs_header_level(root->node);
 799         generation = trans->transid;
 800         nritems = btrfs_header_nritems(root->node);
 801         if (nritems > 0) {
 802                 if (level == 0)
 803                         btrfs_item_key(root->node, &disk_key, 0);
 804                 else
 805                         btrfs_node_key(root->node, &disk_key, 0);
 806                 key_objectid = btrfs_disk_key_objectid(&disk_key);
 807         } else {
 808                 key_objectid = 0;
 809         }
 810         return btrfs_inc_extent_ref(trans, root, root->node->start,
 811                                     root->node->len, owner_objectid,
 812                                     generation, level, key_objectid);
 813 }
 814
 815 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 816                   struct extent_buffer *buf)
 817 {
 818         u64 bytenr;
 819         u32 nritems;
 820         struct btrfs_key key;
 821         struct btrfs_file_extent_item *fi;
 822         int i;
 823         int level;
 824         int ret;
 825         int faili;
 826
 827         if (!root->ref_cows)
 828                 return 0;
 829
 830         level = btrfs_header_level(buf);
 831         nritems = btrfs_header_nritems(buf);
 832         for (i = 0; i < nritems; i++) {
 833                 if (level == 0) {
 834                         u64 disk_bytenr;
 835                         btrfs_item_key_to_cpu(buf, &key, i);
 836                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 837                                 continue;
 838                         fi = btrfs_item_ptr(buf, i,
 839                                             struct btrfs_file_extent_item);
 840                         if (btrfs_file_extent_type(buf, fi) ==
 841                             BTRFS_FILE_EXTENT_INLINE)
 842                                 continue;
 843                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 844                         if (disk_bytenr == 0)
 845                                 continue;
 846                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
 847                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 848                                     root->root_key.objectid, trans->transid,
 849                                     key.objectid, key.offset);
 850                         if (ret) {
 851                                 faili = i;
 852                                 goto fail;
 853                         }
 854                 } else {
 855                         bytenr = btrfs_node_blockptr(buf, i);
 856                         btrfs_node_key_to_cpu(buf, &key, i);
 857                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
 858                                            btrfs_level_size(root, level - 1),
 859                                            root->root_key.objectid,
 860                                            trans->transid,
 861                                            level - 1, key.objectid);
 862                         if (ret) {
 863                                 faili = i;
 864                                 goto fail;
 865                         }
 866                 }
 867         }
 868         return 0;
 869 fail:
 870         WARN_ON(1);
 871 #if 0
 872         for (i =0; i < faili; i++) {
 873                 if (level == 0) {
 874                         u64 disk_bytenr;
 875                         btrfs_item_key_to_cpu(buf, &key, i);
 876                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 877                                 continue;
 878                         fi = btrfs_item_ptr(buf, i,
 879                                             struct btrfs_file_extent_item);
 880                         if (btrfs_file_extent_type(buf, fi) ==
 881                             BTRFS_FILE_EXTENT_INLINE)
 882                                 continue;
 883                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 884                         if (disk_bytenr == 0)
 885                                 continue;
 886                         err = btrfs_free_extent(trans, root, disk_bytenr,
 887                                     btrfs_file_extent_disk_num_bytes(buf,
 888                                                                       fi), 0);
 889                         BUG_ON(err);
 890                 } else {
 891                         bytenr = btrfs_node_blockptr(buf, i);
 892                         err = btrfs_free_extent(trans, root, bytenr,
 893                                         btrfs_level_size(root, level - 1), 0);
 894                         BUG_ON(err);
 895                 }
 896         }
 897 #endif
 898         return ret;
 899 }
 900
 901 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 902                                  struct btrfs_root *root,
 903                                  struct btrfs_path *path,
 904                                  struct btrfs_block_group_cache *cache)
 905 {
 906         int ret;
 907         int pending_ret;
 908         struct btrfs_root *extent_root = root->fs_info->extent_root;
 909         unsigned long bi;
 910         struct extent_buffer *leaf;
 911
 912         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 913         if (ret < 0)
 914                 goto fail;
 915         BUG_ON(ret);
 916
 917         leaf = path->nodes[0];
 918         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 919         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 920         btrfs_mark_buffer_dirty(leaf);
 921         btrfs_release_path(extent_root, path);
 922 fail:
 923         finish_current_insert(trans, extent_root);
 924         pending_ret = del_pending_extents(trans, extent_root);
 925         if (ret)
 926                 return ret;
 927         if (pending_ret)
 928                 return pending_ret;
 929         return 0;
 930
 931 }
 932
 933 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 934                                    struct btrfs_root *root)
 935 {
 936         struct extent_io_tree *block_group_cache;
 937         struct btrfs_block_group_cache *cache;
 938         int ret;
 939         int err = 0;
 940         int werr = 0;
 941         struct btrfs_path *path;
 942         u64 last = 0;
 943         u64 start;
 944         u64 end;
 945         u64 ptr;
 946
 947         block_group_cache = &root->fs_info->block_group_cache;
 948         path = btrfs_alloc_path();
 949         if (!path)
 950                 return -ENOMEM;
 951
 952         while(1) {
 953                 ret = find_first_extent_bit(block_group_cache, last,
 954                                             &start, &end, BLOCK_GROUP_DIRTY);
 955                 if (ret)
 956                         break;
 957
 958                 last = end + 1;
 959                 ret = get_state_private(block_group_cache, start, &ptr);
 960                 if (ret)
 961                         break;
 962
 963                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 964                 err = write_one_cache_group(trans, root,
 965                                             path, cache);
 966                 /*
 967                  * if we fail to write the cache group, we want
 968                  * to keep it marked dirty in hopes that a later
 969                  * write will work
 970                  */
 971                 if (err) {
 972                         werr = err;
 973                         continue;
 974                 }
 975                 clear_extent_bits(block_group_cache, start, end,
 976                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 977         }
 978         btrfs_free_path(path);
 979         return werr;
 980 }
 981
 982 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 983                                                   u64 flags)
 984 {
 985         struct list_head *head = &info->space_info;
 986         struct list_head *cur;
 987         struct btrfs_space_info *found;
 988         list_for_each(cur, head) {
 989                 found = list_entry(cur, struct btrfs_space_info, list);
 990                 if (found->flags == flags)
 991                         return found;
 992         }
 993         return NULL;
 994
 995 }
 996
 997 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 998                              u64 total_bytes, u64 bytes_used,
 999                              struct btrfs_space_info **space_info)
1000 {
1001         struct btrfs_space_info *found;
1002
1003         found = __find_space_info(info, flags);
1004         if (found) {
1005                 found->total_bytes += total_bytes;
1006                 found->bytes_used += bytes_used;
1007                 WARN_ON(found->total_bytes < found->bytes_used);
1008                 *space_info = found;
1009                 return 0;
1010         }
1011         found = kmalloc(sizeof(*found), GFP_NOFS);
1012         if (!found)
1013                 return -ENOMEM;
1014
1015         list_add(&found->list, &info->space_info);
1016         found->flags = flags;
1017         found->total_bytes = total_bytes;
1018         found->bytes_used = bytes_used;
1019         found->bytes_pinned = 0;
1020         found->full = 0;
1021         *space_info = found;
1022         return 0;
1023 }
1024
1025
1026 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1027 {
1028         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1029                                    BTRFS_BLOCK_GROUP_RAID1 |
1030                                    BTRFS_BLOCK_GROUP_DUP);
1031         if (extra_flags) {
1032                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1033                         fs_info->avail_data_alloc_bits |= extra_flags;
1034                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1035                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1036                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1037                         fs_info->avail_system_alloc_bits |= extra_flags;
1038         }
1039 }
1040
1041 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1042                           struct btrfs_root *extent_root, u64 alloc_bytes,
1043                           u64 flags)
1044 {
1045         struct btrfs_space_info *space_info;
1046         u64 thresh;
1047         u64 start;
1048         u64 num_bytes;
1049         int ret;
1050
1051         space_info = __find_space_info(extent_root->fs_info, flags);
1052         if (!space_info) {
1053                 ret = update_space_info(extent_root->fs_info, flags,
1054                                         0, 0, &space_info);
1055                 BUG_ON(ret);
1056         }
1057         BUG_ON(!space_info);
1058
1059         if (space_info->full)
1060                 return 0;
1061
1062         thresh = div_factor(space_info->total_bytes, 7);
1063         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1064             thresh)
1065                 return 0;
1066
1067         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1068         if (ret == -ENOSPC) {
1069 printk("space info full %llu\n", (unsigned long long)flags);
1070                 space_info->full = 1;
1071                 return 0;
1072         }
1073
1074         BUG_ON(ret);
1075
1076         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1077                      extent_root->fs_info->chunk_root->root_key.objectid,
1078                      start, num_bytes);
1079         BUG_ON(ret);
1080         return 0;
1081 }
1082
1083 static int update_block_group(struct btrfs_trans_handle *trans,
1084                               struct btrfs_root *root,
1085                               u64 bytenr, u64 num_bytes, int alloc,
1086                               int mark_free)
1087 {
1088         struct btrfs_block_group_cache *cache;
1089         struct btrfs_fs_info *info = root->fs_info;
1090         u64 total = num_bytes;
1091         u64 old_val;
1092         u64 byte_in_group;
1093         u64 start;
1094         u64 end;
1095
1096         while(total) {
1097                 cache = btrfs_lookup_block_group(info, bytenr);
1098                 if (!cache) {
1099                         return -1;
1100                 }
1101                 byte_in_group = bytenr - cache->key.objectid;
1102                 WARN_ON(byte_in_group > cache->key.offset);
1103                 start = cache->key.objectid;
1104                 end = start + cache->key.offset - 1;
1105                 set_extent_bits(&info->block_group_cache, start, end,
1106                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1107
1108                 old_val = btrfs_block_group_used(&cache->item);
1109                 num_bytes = min(total, cache->key.offset - byte_in_group);
1110                 if (alloc) {
1111                         old_val += num_bytes;
1112                         cache->space_info->bytes_used += num_bytes;
1113                 } else {
1114                         old_val -= num_bytes;
1115                         cache->space_info->bytes_used -= num_bytes;
1116                         if (mark_free) {
1117                                 set_extent_dirty(&info->free_space_cache,
1118                                                  bytenr, bytenr + num_bytes - 1,
1119                                                  GFP_NOFS);
1120                         }
1121                 }
1122                 btrfs_set_block_group_used(&cache->item, old_val);
1123                 total -= num_bytes;
1124                 bytenr += num_bytes;
1125         }
1126         return 0;
1127 }
1128
1129 static int update_pinned_extents(struct btrfs_root *root,
1130                                 u64 bytenr, u64 num, int pin)
1131 {
1132         u64 len;
1133         struct btrfs_block_group_cache *cache;
1134         struct btrfs_fs_info *fs_info = root->fs_info;
1135
1136         if (pin) {
1137                 set_extent_dirty(&fs_info->pinned_extents,
1138                                 bytenr, bytenr + num - 1, GFP_NOFS);
1139         } else {
1140                 clear_extent_dirty(&fs_info->pinned_extents,
1141                                 bytenr, bytenr + num - 1, GFP_NOFS);
1142         }
1143         while (num > 0) {
1144                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1145                 WARN_ON(!cache);
1146                 len = min(num, cache->key.offset -
1147                           (bytenr - cache->key.objectid));
1148                 if (pin) {
1149                         cache->pinned += len;
1150                         cache->space_info->bytes_pinned += len;
1151                         fs_info->total_pinned += len;
1152                 } else {
1153                         cache->pinned -= len;
1154                         cache->space_info->bytes_pinned -= len;
1155                         fs_info->total_pinned -= len;
1156                 }
1157                 bytenr += len;
1158                 num -= len;
1159         }
1160         return 0;
1161 }
1162
1163 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1164 {
1165         u64 last = 0;
1166         u64 start;
1167         u64 end;
1168         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1169         int ret;
1170
1171         while(1) {
1172                 ret = find_first_extent_bit(pinned_extents, last,
1173                                             &start, &end, EXTENT_DIRTY);
1174                 if (ret)
1175                         break;
1176                 set_extent_dirty(copy, start, end, GFP_NOFS);
1177                 last = end + 1;
1178         }
1179         return 0;
1180 }
1181
1182 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1183                                struct btrfs_root *root,
1184                                struct extent_io_tree *unpin)
1185 {
1186         u64 start;
1187         u64 end;
1188         int ret;
1189         struct extent_io_tree *free_space_cache;
1190         free_space_cache = &root->fs_info->free_space_cache;
1191
1192         while(1) {
1193                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1194                                             EXTENT_DIRTY);
1195                 if (ret)
1196                         break;
1197                 update_pinned_extents(root, start, end + 1 - start, 0);
1198                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1199                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1200         }
1201         return 0;
1202 }
1203
1204 static int finish_current_insert(struct btrfs_trans_handle *trans,
1205                                  struct btrfs_root *extent_root)
1206 {
1207         u64 start;
1208         u64 end;
1209         struct btrfs_fs_info *info = extent_root->fs_info;
1210         struct extent_buffer *eb;
1211         struct btrfs_path *path;
1212         struct btrfs_key ins;
1213         struct btrfs_disk_key first;
1214         struct btrfs_extent_item extent_item;
1215         int ret;
1216         int level;
1217         int err = 0;
1218
1219         btrfs_set_stack_extent_refs(&extent_item, 1);
1220         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1221         path = btrfs_alloc_path();
1222
1223         while(1) {
1224                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1225                                             &end, EXTENT_LOCKED);
1226                 if (ret)
1227                         break;
1228
1229                 ins.objectid = start;
1230                 ins.offset = end + 1 - start;
1231                 err = btrfs_insert_item(trans, extent_root, &ins,
1232                                         &extent_item, sizeof(extent_item));
1233                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1234                                   GFP_NOFS);
1235                 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1236                 level = btrfs_header_level(eb);
1237                 if (level == 0) {
1238                         btrfs_item_key(eb, &first, 0);
1239                 } else {
1240                         btrfs_node_key(eb, &first, 0);
1241                 }
1242                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1243                                           start, extent_root->root_key.objectid,
1244                                           0, level,
1245                                           btrfs_disk_key_objectid(&first));
1246                 BUG_ON(err);
1247                 free_extent_buffer(eb);
1248         }
1249         btrfs_free_path(path);
1250         return 0;
1251 }
1252
1253 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1254                           int pending)
1255 {
1256         int err = 0;
1257         struct extent_buffer *buf;
1258
1259         if (!pending) {
1260                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1261                 if (buf) {
1262                         if (btrfs_buffer_uptodate(buf)) {
1263                                 u64 transid =
1264                                     root->fs_info->running_transaction->transid;
1265                                 if (btrfs_header_generation(buf) == transid) {
1266                                         free_extent_buffer(buf);
1267                                         return 1;
1268                                 }
1269                         }
1270                         free_extent_buffer(buf);
1271                 }
1272                 update_pinned_extents(root, bytenr, num_bytes, 1);
1273         } else {
1274                 set_extent_bits(&root->fs_info->pending_del,
1275                                 bytenr, bytenr + num_bytes - 1,
1276                                 EXTENT_LOCKED, GFP_NOFS);
1277         }
1278         BUG_ON(err < 0);
1279         return 0;
1280 }
1281
1282 /*
1283  * remove an extent from the root, returns 0 on success
1284  */
1285 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1286                          *root, u64 bytenr, u64 num_bytes,
1287                          u64 root_objectid, u64 ref_generation,
1288                          u64 owner_objectid, u64 owner_offset, int pin,
1289                          int mark_free)
1290 {
1291         struct btrfs_path *path;
1292         struct btrfs_key key;
1293         struct btrfs_fs_info *info = root->fs_info;
1294         struct btrfs_extent_ops *ops = info->extent_ops;
1295         struct btrfs_root *extent_root = info->extent_root;
1296         struct extent_buffer *leaf;
1297         int ret;
1298         int extent_slot = 0;
1299         int found_extent = 0;
1300         int num_to_del = 1;
1301         struct btrfs_extent_item *ei;
1302         u32 refs;
1303
1304         key.objectid = bytenr;
1305         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1306         key.offset = num_bytes;
1307
1308         path = btrfs_alloc_path();
1309         if (!path)
1310                 return -ENOMEM;
1311
1312         ret = lookup_extent_backref(trans, extent_root, path,
1313                                     bytenr, root_objectid,
1314                                     ref_generation,
1315                                     owner_objectid, owner_offset, 1);
1316         if (ret == 0) {
1317                 struct btrfs_key found_key;
1318                 extent_slot = path->slots[0];
1319                 while(extent_slot > 0) {
1320                         extent_slot--;
1321                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1322                                               extent_slot);
1323                         if (found_key.objectid != bytenr)
1324                                 break;
1325                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1326                             found_key.offset == num_bytes) {
1327                                 found_extent = 1;
1328                                 break;
1329                         }
1330                         if (path->slots[0] - extent_slot > 5)
1331                                 break;
1332                 }
1333                 if (!found_extent)
1334                         ret = btrfs_del_item(trans, extent_root, path);
1335         } else {
1336                 btrfs_print_leaf(extent_root, path->nodes[0]);
1337                 WARN_ON(1);
1338                 printk("Unable to find ref byte nr %llu root %llu "
1339                        " gen %llu owner %llu offset %llu\n",
1340                        (unsigned long long)bytenr,
1341                        (unsigned long long)root_objectid,
1342                        (unsigned long long)ref_generation,
1343                        (unsigned long long)owner_objectid,
1344                        (unsigned long long)owner_offset);
1345         }
1346         if (!found_extent) {
1347                 btrfs_release_path(extent_root, path);
1348                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1349                 if (ret < 0)
1350                         return ret;
1351                 BUG_ON(ret);
1352                 extent_slot = path->slots[0];
1353         }
1354
1355         leaf = path->nodes[0];
1356         ei = btrfs_item_ptr(leaf, extent_slot,
1357                             struct btrfs_extent_item);
1358         refs = btrfs_extent_refs(leaf, ei);
1359         BUG_ON(refs == 0);
1360         refs -= 1;
1361         btrfs_set_extent_refs(leaf, ei, refs);
1362
1363         btrfs_mark_buffer_dirty(leaf);
1364
1365         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1366                 /* if the back ref and the extent are next to each other
1367                  * they get deleted below in one shot
1368                  */
1369                 path->slots[0] = extent_slot;
1370                 num_to_del = 2;
1371         } else if (found_extent) {
1372                 /* otherwise delete the extent back ref */
1373                 ret = btrfs_del_item(trans, extent_root, path);
1374                 BUG_ON(ret);
1375                 /* if refs are 0, we need to setup the path for deletion */
1376                 if (refs == 0) {
1377                         btrfs_release_path(extent_root, path);
1378                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1379                                                 -1, 1);
1380                         if (ret < 0)
1381                                 return ret;
1382                         BUG_ON(ret);
1383                 }
1384         }
1385
1386         if (refs == 0) {
1387                 u64 super_used;
1388                 u64 root_used;
1389
1390                 if (pin) {
1391                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1392                         if (ret > 0)
1393                                 mark_free = 1;
1394                         BUG_ON(ret < 0);
1395                 }
1396
1397                 /* block accounting for super block */
1398                 super_used = btrfs_super_bytes_used(&info->super_copy);
1399                 btrfs_set_super_bytes_used(&info->super_copy,
1400                                            super_used - num_bytes);
1401
1402                 /* block accounting for root item */
1403                 root_used = btrfs_root_used(&root->root_item);
1404                 btrfs_set_root_used(&root->root_item,
1405                                            root_used - num_bytes);
1406                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1407                                       num_to_del);
1408                 if (ret)
1409                         return ret;
1410
1411                 if (ops && ops->free_extent)
1412                         ops->free_extent(root, bytenr, num_bytes);
1413
1414                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1415                                          mark_free);
1416                 BUG_ON(ret);
1417         }
1418         btrfs_free_path(path);
1419         finish_current_insert(trans, extent_root);
1420         return ret;
1421 }
1422
1423 /*
1424  * find all the blocks marked as pending in the radix tree and remove
1425  * them from the extent map
1426  */
1427 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1428                                btrfs_root *extent_root)
1429 {
1430         int ret;
1431         int err = 0;
1432         u64 start;
1433         u64 end;
1434         struct extent_io_tree *pending_del;
1435         struct extent_io_tree *pinned_extents;
1436
1437         pending_del = &extent_root->fs_info->pending_del;
1438         pinned_extents = &extent_root->fs_info->pinned_extents;
1439
1440         while(1) {
1441                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1442                                             EXTENT_LOCKED);
1443                 if (ret)
1444                         break;
1445                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1446                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1447                                   GFP_NOFS);
1448                 ret = __free_extent(trans, extent_root,
1449                                      start, end + 1 - start,
1450                                      extent_root->root_key.objectid,
1451                                      0, 0, 0, 0, 0);
1452                 if (ret)
1453                         err = ret;
1454         }
1455         return err;
1456 }
1457
1458 /*
1459  * remove an extent from the root, returns 0 on success
1460  */
1461 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1462                       *root, u64 bytenr, u64 num_bytes,
1463                       u64 root_objectid, u64 ref_generation,
1464                       u64 owner_objectid, u64 owner_offset, int pin)
1465 {
1466         struct btrfs_root *extent_root = root->fs_info->extent_root;
1467         int pending_ret;
1468         int ret;
1469
1470         WARN_ON(num_bytes < root->sectorsize);
1471         if (!root->ref_cows)
1472                 ref_generation = 0;
1473
1474         if (root == extent_root) {
1475                 pin_down_bytes(root, bytenr, num_bytes, 1);
1476                 return 0;
1477         }
1478         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1479                             ref_generation, owner_objectid, owner_offset,
1480                             pin, pin == 0);
1481         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1482         return ret ? ret : pending_ret;
1483 }
1484
1485 static u64 stripe_align(struct btrfs_root *root, u64 val)
1486 {
1487         u64 mask = ((u64)root->stripesize - 1);
1488         u64 ret = (val + mask) & ~mask;
1489         return ret;
1490 }
1491
1492 /*
1493  * walks the btree of allocated extents and find a hole of a given size.
1494  * The key ins is changed to record the hole:
1495  * ins->objectid == block start
1496  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1497  * ins->offset == number of blocks
1498  * Any available blocks before search_start are skipped.
1499  */
1500 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1501                                      struct btrfs_root *orig_root,
1502                                      u64 num_bytes, u64 empty_size,
1503                                      u64 search_start, u64 search_end,
1504                                      u64 hint_byte, struct btrfs_key *ins,
1505                                      u64 exclude_start, u64 exclude_nr,
1506                                      int data)
1507 {
1508         int ret;
1509         u64 orig_search_start = search_start;
1510         struct btrfs_root * root = orig_root->fs_info->extent_root;
1511         struct btrfs_fs_info *info = root->fs_info;
1512         u64 total_needed = num_bytes;
1513         struct btrfs_block_group_cache *block_group;
1514         int full_scan = 0;
1515         int wrapped = 0;
1516
1517         WARN_ON(num_bytes < root->sectorsize);
1518         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1519
1520         if (search_end == (u64)-1)
1521                 search_end = btrfs_super_total_bytes(&info->super_copy);
1522
1523         if (hint_byte) {
1524                 block_group = btrfs_lookup_block_group(info, hint_byte);
1525                 if (!block_group)
1526                         hint_byte = search_start;
1527                 block_group = btrfs_find_block_group(root, block_group,
1528                                                      hint_byte, data, 1);
1529         } else {
1530                 block_group = btrfs_find_block_group(root,
1531                                                      trans->block_group,
1532                                                      search_start, data, 1);
1533         }
1534
1535         total_needed += empty_size;
1536
1537 check_failed:
1538         if (!block_group) {
1539                 block_group = btrfs_lookup_block_group(info, search_start);
1540                 if (!block_group)
1541                         block_group = btrfs_lookup_block_group(info,
1542                                                        orig_search_start);
1543         }
1544         ret = find_search_start(root, &block_group, &search_start,
1545                                 total_needed, data);
1546         if (ret)
1547                 goto error;
1548
1549         search_start = stripe_align(root, search_start);
1550         ins->objectid = search_start;
1551         ins->offset = num_bytes;
1552
1553         if (ins->objectid + num_bytes >= search_end)
1554                 goto enospc;
1555
1556         if (ins->objectid + num_bytes >
1557             block_group->key.objectid + block_group->key.offset) {
1558                 search_start = block_group->key.objectid +
1559                         block_group->key.offset;
1560                 goto new_group;
1561         }
1562
1563         if (test_range_bit(&info->extent_ins, ins->objectid,
1564                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1565                 search_start = ins->objectid + num_bytes;
1566                 goto new_group;
1567         }
1568
1569         if (test_range_bit(&info->pinned_extents, ins->objectid,
1570                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1571                 search_start = ins->objectid + num_bytes;
1572                 goto new_group;
1573         }
1574
1575         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1576             ins->objectid < exclude_start + exclude_nr)) {
1577                 search_start = exclude_start + exclude_nr;
1578                 goto new_group;
1579         }
1580
1581         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1582                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1583                 if (block_group)
1584                         trans->block_group = block_group;
1585         }
1586         ins->offset = num_bytes;
1587         return 0;
1588
1589 new_group:
1590         if (search_start + num_bytes >= search_end) {
1591 enospc:
1592                 search_start = orig_search_start;
1593                 if (full_scan) {
1594                         ret = -ENOSPC;
1595                         goto error;
1596                 }
1597                 if (wrapped) {
1598                         if (!full_scan)
1599                                 total_needed -= empty_size;
1600                         full_scan = 1;
1601                 } else
1602                         wrapped = 1;
1603         }
1604         block_group = btrfs_lookup_block_group(info, search_start);
1605         cond_resched();
1606         block_group = btrfs_find_block_group(root, block_group,
1607                                              search_start, data, 0);
1608         goto check_failed;
1609
1610 error:
1611         return ret;
1612 }
1613 /*
1614  * finds a free extent and does all the dirty work required for allocation
1615  * returns the key for the extent through ins, and a tree buffer for
1616  * the first block of the extent through buf.
1617  *
1618  * returns 0 if everything worked, non-zero otherwise.
1619  */
1620 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1621                        struct btrfs_root *root,
1622                        u64 num_bytes, u64 root_objectid, u64 ref_generation,
1623                        u64 owner, u64 owner_offset,
1624                        u64 empty_size, u64 hint_byte,
1625                        u64 search_end, struct btrfs_key *ins, int data)
1626 {
1627         int ret;
1628         int pending_ret;
1629         u64 super_used, root_used;
1630         u64 search_start = 0;
1631         u64 alloc_profile;
1632         struct btrfs_fs_info *info = root->fs_info;
1633         struct btrfs_extent_ops *ops = info->extent_ops;
1634         u32 sizes[2];
1635         struct btrfs_root *extent_root = info->extent_root;
1636         struct btrfs_path *path;
1637         struct btrfs_extent_item *extent_item;
1638         struct btrfs_extent_ref *ref;
1639         struct btrfs_key keys[2];
1640
1641         if (data) {
1642                 alloc_profile = info->avail_data_alloc_bits &
1643                                 info->data_alloc_profile;
1644                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1645         } else if (root == info->chunk_root || info->force_system_allocs) {
1646                 alloc_profile = info->avail_system_alloc_bits &
1647                                 info->system_alloc_profile;
1648                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1649         } else {
1650                 alloc_profile = info->avail_metadata_alloc_bits &
1651                                 info->metadata_alloc_profile;
1652                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1653         }
1654
1655         if (root->ref_cows) {
1656                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1657                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1658                                              num_bytes,
1659                                              BTRFS_BLOCK_GROUP_METADATA);
1660                         BUG_ON(ret);
1661                 }
1662                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1663                                      num_bytes + 2 * 1024 * 1024, data);
1664                 BUG_ON(ret);
1665         }
1666
1667         WARN_ON(num_bytes < root->sectorsize);
1668         if (ops && ops->alloc_extent) {
1669                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
1670         } else {
1671                 ret = find_free_extent(trans, root, num_bytes, empty_size,
1672                                         search_start, search_end, hint_byte,
1673                                         ins, trans->alloc_exclude_start,
1674                                         trans->alloc_exclude_nr, data);
1675         }
1676         BUG_ON(ret);
1677         if (ret)
1678                 return ret;
1679
1680         /* block accounting for super block */
1681         super_used = btrfs_super_bytes_used(&info->super_copy);
1682         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1683
1684         /* block accounting for root item */
1685         root_used = btrfs_root_used(&root->root_item);
1686         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1687
1688         clear_extent_dirty(&root->fs_info->free_space_cache,
1689                            ins->objectid, ins->objectid + ins->offset - 1,
1690                            GFP_NOFS);
1691
1692         if (root == extent_root) {
1693                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1694                                 ins->objectid + ins->offset - 1,
1695                                 EXTENT_LOCKED, GFP_NOFS);
1696                 goto update_block;
1697         }
1698
1699         WARN_ON(trans->alloc_exclude_nr);
1700         trans->alloc_exclude_start = ins->objectid;
1701         trans->alloc_exclude_nr = ins->offset;
1702
1703         memcpy(&keys[0], ins, sizeof(*ins));
1704         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1705                                          owner, owner_offset);
1706         keys[1].objectid = ins->objectid;
1707         keys[1].type = BTRFS_EXTENT_REF_KEY;
1708         sizes[0] = sizeof(*extent_item);
1709         sizes[1] = sizeof(*ref);
1710
1711         path = btrfs_alloc_path();
1712         BUG_ON(!path);
1713
1714         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1715                                        sizes, 2);
1716
1717         BUG_ON(ret);
1718         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1719                                      struct btrfs_extent_item);
1720         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1721         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1722                              struct btrfs_extent_ref);
1723
1724         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1725         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1726         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1727         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1728
1729         btrfs_mark_buffer_dirty(path->nodes[0]);
1730
1731         trans->alloc_exclude_start = 0;
1732         trans->alloc_exclude_nr = 0;
1733         btrfs_free_path(path);
1734         finish_current_insert(trans, extent_root);
1735         pending_ret = del_pending_extents(trans, extent_root);
1736
1737         if (ret) {
1738                 return ret;
1739         }
1740         if (pending_ret) {
1741                 return pending_ret;
1742         }
1743
1744 update_block:
1745         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1746         if (ret) {
1747                 printk("update block group failed for %llu %llu\n",
1748                        (unsigned long long)ins->objectid,
1749                        (unsigned long long)ins->offset);
1750                 BUG();
1751         }
1752         return 0;
1753 }
1754
1755 /*
1756  * helper function to allocate a block for a given tree
1757  * returns the tree buffer or NULL.
1758  */
1759 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1760                                              struct btrfs_root *root,
1761                                              u32 blocksize,
1762                                              u64 root_objectid, u64 hint,
1763                                              u64 empty_size)
1764 {
1765         u64 ref_generation;
1766
1767         if (root->ref_cows)
1768                 ref_generation = trans->transid;
1769         else
1770                 ref_generation = 0;
1771
1772
1773         return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1774                                         ref_generation, 0, 0, hint, empty_size);
1775 }
1776
1777 /*
1778  * helper function to allocate a block for a given tree
1779  * returns the tree buffer or NULL.
1780  */
1781 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1782                                              struct btrfs_root *root,
1783                                              u32 blocksize,
1784                                              u64 root_objectid,
1785                                              u64 ref_generation,
1786                                              u64 first_objectid,
1787                                              int level,
1788                                              u64 hint,
1789                                              u64 empty_size)
1790 {
1791         struct btrfs_key ins;
1792         int ret;
1793         struct extent_buffer *buf;
1794
1795         ret = btrfs_alloc_extent(trans, root, blocksize,
1796                                  root_objectid, ref_generation,
1797                                  level, first_objectid, empty_size, hint,
1798                                  (u64)-1, &ins, 0);
1799         if (ret) {
1800                 BUG_ON(ret > 0);
1801                 return ERR_PTR(ret);
1802         }
1803         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1804         if (!buf) {
1805                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1806                                   root->root_key.objectid, ref_generation,
1807                                   0, 0, 0);
1808                 BUG_ON(1);
1809                 return ERR_PTR(-ENOMEM);
1810         }
1811         btrfs_set_buffer_uptodate(buf);
1812         trans->blocks_used++;
1813         return buf;
1814 }
1815
1816 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1817                                   struct btrfs_root *root,
1818                                   struct extent_buffer *leaf)
1819 {
1820         u64 leaf_owner;
1821         u64 leaf_generation;
1822         struct btrfs_key key;
1823         struct btrfs_file_extent_item *fi;
1824         int i;
1825         int nritems;
1826         int ret;
1827
1828         BUG_ON(!btrfs_is_leaf(leaf));
1829         nritems = btrfs_header_nritems(leaf);
1830         leaf_owner = btrfs_header_owner(leaf);
1831         leaf_generation = btrfs_header_generation(leaf);
1832
1833         for (i = 0; i < nritems; i++) {
1834                 u64 disk_bytenr;
1835
1836                 btrfs_item_key_to_cpu(leaf, &key, i);
1837                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1838                         continue;
1839                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1840                 if (btrfs_file_extent_type(leaf, fi) ==
1841                     BTRFS_FILE_EXTENT_INLINE)
1842                         continue;
1843                 /*
1844                  * FIXME make sure to insert a trans record that
1845                  * repeats the snapshot del on crash
1846                  */
1847                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1848                 if (disk_bytenr == 0)
1849                         continue;
1850                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1851                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
1852                                 leaf_owner, leaf_generation,
1853                                 key.objectid, key.offset, 0);
1854                 BUG_ON(ret);
1855         }
1856         return 0;
1857 }
1858
1859 static void noinline reada_walk_down(struct btrfs_root *root,
1860                                      struct extent_buffer *node,
1861                                      int slot)
1862 {
1863         u64 bytenr;
1864         u64 last = 0;
1865         u32 nritems;
1866         u32 refs;
1867         u32 blocksize;
1868         int ret;
1869         int i;
1870         int level;
1871         int skipped = 0;
1872
1873         nritems = btrfs_header_nritems(node);
1874         level = btrfs_header_level(node);
1875         if (level)
1876                 return;
1877
1878         for (i = slot; i < nritems && skipped < 32; i++) {
1879                 bytenr = btrfs_node_blockptr(node, i);
1880                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1881                              (last > bytenr && last - bytenr > 32 * 1024))) {
1882                         skipped++;
1883                         continue;
1884                 }
1885                 blocksize = btrfs_level_size(root, level - 1);
1886                 if (i != slot) {
1887                         ret = lookup_extent_ref(NULL, root, bytenr,
1888                                                 blocksize, &refs);
1889                         BUG_ON(ret);
1890                         if (refs != 1) {
1891                                 skipped++;
1892                                 continue;
1893                         }
1894                 }
1895                 mutex_unlock(&root->fs_info->fs_mutex);
1896                 ret = readahead_tree_block(root, bytenr, blocksize);
1897                 last = bytenr + blocksize;
1898                 cond_resched();
1899                 mutex_lock(&root->fs_info->fs_mutex);
1900                 if (ret)
1901                         break;
1902         }
1903 }
1904
1905 /*
1906  * helper function for drop_snapshot, this walks down the tree dropping ref
1907  * counts as it goes.
1908  */
1909 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1910                                    struct btrfs_root *root,
1911                                    struct btrfs_path *path, int *level)
1912 {
1913         u64 root_owner;
1914         u64 root_gen;
1915         u64 bytenr;
1916         struct extent_buffer *next;
1917         struct extent_buffer *cur;
1918         struct extent_buffer *parent;
1919         u32 blocksize;
1920         int ret;
1921         u32 refs;
1922
1923         WARN_ON(*level < 0);
1924         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1925         ret = lookup_extent_ref(trans, root,
1926                                 path->nodes[*level]->start,
1927                                 path->nodes[*level]->len, &refs);
1928         BUG_ON(ret);
1929         if (refs > 1)
1930                 goto out;
1931
1932         /*
1933          * walk down to the last node level and free all the leaves
1934          */
1935         while(*level >= 0) {
1936                 WARN_ON(*level < 0);
1937                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1938                 cur = path->nodes[*level];
1939
1940                 if (btrfs_header_level(cur) != *level)
1941                         WARN_ON(1);
1942
1943                 if (path->slots[*level] >=
1944                     btrfs_header_nritems(cur))
1945                         break;
1946                 if (*level == 0) {
1947                         ret = drop_leaf_ref(trans, root, cur);
1948                         BUG_ON(ret);
1949                         break;
1950                 }
1951                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1952                 blocksize = btrfs_level_size(root, *level - 1);
1953                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1954                 BUG_ON(ret);
1955                 if (refs != 1) {
1956                         parent = path->nodes[*level];
1957                         root_owner = btrfs_header_owner(parent);
1958                         root_gen = btrfs_header_generation(parent);
1959                         path->slots[*level]++;
1960                         ret = btrfs_free_extent(trans, root, bytenr,
1961                                                 blocksize, root_owner,
1962                                                 root_gen, 0, 0, 1);
1963                         BUG_ON(ret);
1964                         continue;
1965                 }
1966                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1967                 if (!next || !btrfs_buffer_uptodate(next)) {
1968                         free_extent_buffer(next);
1969                         reada_walk_down(root, cur, path->slots[*level]);
1970                         mutex_unlock(&root->fs_info->fs_mutex);
1971                         next = read_tree_block(root, bytenr, blocksize);
1972                         mutex_lock(&root->fs_info->fs_mutex);
1973
1974                         /* we dropped the lock, check one more time */
1975                         ret = lookup_extent_ref(trans, root, bytenr,
1976                                                 blocksize, &refs);
1977                         BUG_ON(ret);
1978                         if (refs != 1) {
1979                                 parent = path->nodes[*level];
1980                                 root_owner = btrfs_header_owner(parent);
1981                                 root_gen = btrfs_header_generation(parent);
1982
1983                                 path->slots[*level]++;
1984                                 free_extent_buffer(next);
1985                                 ret = btrfs_free_extent(trans, root, bytenr,
1986                                                         blocksize,
1987                                                         root_owner,
1988                                                         root_gen, 0, 0, 1);
1989                                 BUG_ON(ret);
1990                                 continue;
1991                         }
1992                 }
1993                 WARN_ON(*level <= 0);
1994                 if (path->nodes[*level-1])
1995                         free_extent_buffer(path->nodes[*level-1]);
1996                 path->nodes[*level-1] = next;
1997                 *level = btrfs_header_level(next);
1998                 path->slots[*level] = 0;
1999         }
2000 out:
2001         WARN_ON(*level < 0);
2002         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2003
2004         if (path->nodes[*level] == root->node) {
2005                 root_owner = root->root_key.objectid;
2006                 parent = path->nodes[*level];
2007         } else {
2008                 parent = path->nodes[*level + 1];
2009                 root_owner = btrfs_header_owner(parent);
2010         }
2011
2012         root_gen = btrfs_header_generation(parent);
2013         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2014                                 path->nodes[*level]->len,
2015                                 root_owner, root_gen, 0, 0, 1);
2016         free_extent_buffer(path->nodes[*level]);
2017         path->nodes[*level] = NULL;
2018         *level += 1;
2019         BUG_ON(ret);
2020         return 0;
2021 }
2022
2023 /*
2024  * helper for dropping snapshots.  This walks back up the tree in the path
2025  * to find the first node higher up where we haven't yet gone through
2026  * all the slots
2027  */
2028 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2029                                  struct btrfs_root *root,
2030                                  struct btrfs_path *path, int *level)
2031 {
2032         u64 root_owner;
2033         u64 root_gen;
2034         struct btrfs_root_item *root_item = &root->root_item;
2035         int i;
2036         int slot;
2037         int ret;
2038
2039         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2040                 slot = path->slots[i];
2041                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2042                         struct extent_buffer *node;
2043                         struct btrfs_disk_key disk_key;
2044                         node = path->nodes[i];
2045                         path->slots[i]++;
2046                         *level = i;
2047                         WARN_ON(*level == 0);
2048                         btrfs_node_key(node, &disk_key, path->slots[i]);
2049                         memcpy(&root_item->drop_progress,
2050                                &disk_key, sizeof(disk_key));
2051                         root_item->drop_level = i;
2052                         return 0;
2053                 } else {
2054                         if (path->nodes[*level] == root->node) {
2055                                 root_owner = root->root_key.objectid;
2056                                 root_gen =
2057                                    btrfs_header_generation(path->nodes[*level]);
2058                         } else {
2059                                 struct extent_buffer *node;
2060                                 node = path->nodes[*level + 1];
2061                                 root_owner = btrfs_header_owner(node);
2062                                 root_gen = btrfs_header_generation(node);
2063                         }
2064                         ret = btrfs_free_extent(trans, root,
2065                                                 path->nodes[*level]->start,
2066                                                 path->nodes[*level]->len,
2067                                                 root_owner, root_gen, 0, 0, 1);
2068                         BUG_ON(ret);
2069                         free_extent_buffer(path->nodes[*level]);
2070                         path->nodes[*level] = NULL;
2071                         *level = i + 1;
2072                 }
2073         }
2074         return 1;
2075 }
2076
2077 /*
2078  * drop the reference count on the tree rooted at 'snap'.  This traverses
2079  * the tree freeing any blocks that have a ref count of zero after being
2080  * decremented.
2081  */
2082 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2083                         *root)
2084 {
2085         int ret = 0;
2086         int wret;
2087         int level;
2088         struct btrfs_path *path;
2089         int i;
2090         int orig_level;
2091         struct btrfs_root_item *root_item = &root->root_item;
2092
2093         path = btrfs_alloc_path();
2094         BUG_ON(!path);
2095
2096         level = btrfs_header_level(root->node);
2097         orig_level = level;
2098         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2099                 path->nodes[level] = root->node;
2100                 extent_buffer_get(root->node);
2101                 path->slots[level] = 0;
2102         } else {
2103                 struct btrfs_key key;
2104                 struct btrfs_disk_key found_key;
2105                 struct extent_buffer *node;
2106
2107                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2108                 level = root_item->drop_level;
2109                 path->lowest_level = level;
2110                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2111                 if (wret < 0) {
2112                         ret = wret;
2113                         goto out;
2114                 }
2115                 node = path->nodes[level];
2116                 btrfs_node_key(node, &found_key, path->slots[level]);
2117                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2118                                sizeof(found_key)));
2119         }
2120         while(1) {
2121                 wret = walk_down_tree(trans, root, path, &level);
2122                 if (wret < 0)
2123                         ret = wret;
2124                 if (wret != 0)
2125                         break;
2126
2127                 wret = walk_up_tree(trans, root, path, &level);
2128                 if (wret < 0)
2129                         ret = wret;
2130                 if (wret != 0)
2131                         break;
2132                 /*
2133                 ret = -EAGAIN;
2134                 break;
2135                 */
2136         }
2137         for (i = 0; i <= orig_level; i++) {
2138                 if (path->nodes[i]) {
2139                         free_extent_buffer(path->nodes[i]);
2140                         path->nodes[i] = NULL;
2141                 }
2142         }
2143 out:
2144         btrfs_free_path(path);
2145         return ret;
2146 }
2147
2148 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2149 {
2150         u64 start;
2151         u64 end;
2152         u64 ptr;
2153         int ret;
2154         while(1) {
2155                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2156                                             &start, &end, (unsigned int)-1);
2157                 if (ret)
2158                         break;
2159                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2160                 if (!ret)
2161                         kfree((void *)(unsigned long)ptr);
2162                 clear_extent_bits(&info->block_group_cache, start,
2163                                   end, (unsigned int)-1, GFP_NOFS);
2164         }
2165         while(1) {
2166                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2167                                             &start, &end, EXTENT_DIRTY);
2168                 if (ret)
2169                         break;
2170                 clear_extent_dirty(&info->free_space_cache, start,
2171                                    end, GFP_NOFS);
2172         }
2173         return 0;
2174 }
2175
2176 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2177                            struct btrfs_key *key)
2178 {
2179         int ret;
2180         struct btrfs_key found_key;
2181         struct extent_buffer *leaf;
2182         int slot;
2183
2184         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2185         if (ret < 0)
2186                 return ret;
2187         while(1) {
2188                 slot = path->slots[0];
2189                 leaf = path->nodes[0];
2190                 if (slot >= btrfs_header_nritems(leaf)) {
2191                         ret = btrfs_next_leaf(root, path);
2192                         if (ret == 0)
2193                                 continue;
2194                         if (ret < 0)
2195                                 goto error;
2196                         break;
2197                 }
2198                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2199
2200                 if (found_key.objectid >= key->objectid &&
2201                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2202                         return 0;
2203                 path->slots[0]++;
2204         }
2205         ret = -ENOENT;
2206 error:
2207         return ret;
2208 }
2209
2210 int btrfs_read_block_groups(struct btrfs_root *root)
2211 {
2212         struct btrfs_path *path;
2213         int ret;
2214         int bit;
2215         struct btrfs_block_group_cache *cache;
2216         struct btrfs_fs_info *info = root->fs_info;
2217         struct btrfs_space_info *space_info;
2218         struct extent_io_tree *block_group_cache;
2219         struct btrfs_key key;
2220         struct btrfs_key found_key;
2221         struct extent_buffer *leaf;
2222
2223         block_group_cache = &info->block_group_cache;
2224
2225         root = info->extent_root;
2226         key.objectid = 0;
2227         key.offset = 0;
2228         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2229         path = btrfs_alloc_path();
2230         if (!path)
2231                 return -ENOMEM;
2232
2233         while(1) {
2234                 ret = find_first_block_group(root, path, &key);
2235                 if (ret > 0) {
2236                         ret = 0;
2237                         goto error;
2238                 }
2239                 if (ret != 0) {
2240                         goto error;
2241                 }
2242                 leaf = path->nodes[0];
2243                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2244                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2245                 if (!cache) {
2246                         ret = -ENOMEM;
2247                         break;
2248                 }
2249
2250                 read_extent_buffer(leaf, &cache->item,
2251                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
2252                                    sizeof(cache->item));
2253                 memcpy(&cache->key, &found_key, sizeof(found_key));
2254                 cache->cached = 0;
2255                 cache->pinned = 0;
2256                 key.objectid = found_key.objectid + found_key.offset;
2257                 btrfs_release_path(root, path);
2258                 cache->flags = btrfs_block_group_flags(&cache->item);
2259                 bit = 0;
2260                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2261                         bit = BLOCK_GROUP_DATA;
2262                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2263                         bit = BLOCK_GROUP_SYSTEM;
2264                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2265                         bit = BLOCK_GROUP_METADATA;
2266                 }
2267                 set_avail_alloc_bits(info, cache->flags);
2268
2269                 ret = update_space_info(info, cache->flags, found_key.offset,
2270                                         btrfs_block_group_used(&cache->item),
2271                                         &space_info);
2272                 BUG_ON(ret);
2273                 cache->space_info = space_info;
2274
2275                 /* use EXTENT_LOCKED to prevent merging */
2276                 set_extent_bits(block_group_cache, found_key.objectid,
2277                                 found_key.objectid + found_key.offset - 1,
2278                                 bit | EXTENT_LOCKED, GFP_NOFS);
2279                 set_state_private(block_group_cache, found_key.objectid,
2280                                   (unsigned long)cache);
2281
2282                 if (key.objectid >=
2283                     btrfs_super_total_bytes(&info->super_copy))
2284                         break;
2285         }
2286         ret = 0;
2287 error:
2288         btrfs_free_path(path);
2289         return ret;
2290 }
2291
2292 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2293                            struct btrfs_root *root, u64 bytes_used,
2294                            u64 type, u64 chunk_tree, u64 chunk_objectid,
2295                            u64 size)
2296 {
2297         int ret;
2298         int bit = 0;
2299         struct btrfs_root *extent_root;
2300         struct btrfs_block_group_cache *cache;
2301         struct extent_io_tree *block_group_cache;
2302
2303         extent_root = root->fs_info->extent_root;
2304         block_group_cache = &root->fs_info->block_group_cache;
2305
2306         cache = kzalloc(sizeof(*cache), GFP_NOFS);
2307         BUG_ON(!cache);
2308         cache->key.objectid = chunk_objectid;
2309         cache->key.offset = size;
2310         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2311         btrfs_set_block_group_used(&cache->item, bytes_used);
2312         btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
2313         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2314         cache->flags = type;
2315         btrfs_set_block_group_flags(&cache->item, type);
2316
2317         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2318                                 &cache->space_info);
2319         BUG_ON(ret);
2320
2321         bit = block_group_state_bits(type);
2322         set_extent_bits(block_group_cache, chunk_objectid,
2323                         chunk_objectid + size - 1,
2324                         bit | EXTENT_LOCKED, GFP_NOFS);
2325         set_state_private(block_group_cache, chunk_objectid,
2326                           (unsigned long)cache);
2327
2328         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2329                                 sizeof(cache->item));
2330         BUG_ON(ret);
2331
2332         finish_current_insert(trans, extent_root);
2333         ret = del_pending_extents(trans, extent_root);
2334         BUG_ON(ret);
2335         set_avail_alloc_bits(extent_root->fs_info, type);
2336         return 0;
2337 }
2338
2339 u64 btrfs_hash_extent_ref(u64 root_objectid, u64 ref_generation,
2340                           u64 owner, u64 owner_offset)
2341 {
2342         return hash_extent_ref(root_objectid, ref_generation,
2343                                owner, owner_offset);
2344 }
2345
2346 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
2347                              struct btrfs_root *root,
2348                              u64 bytenr, u64 num_bytes, int alloc,
2349                              int mark_free)
2350 {
2351         return update_block_group(trans, root, bytenr, num_bytes,
2352                                   alloc, mark_free);
2353 }