fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include <linux/crc32c.h>
  21 #include <linux/pagemap.h>
  22 #include "hash.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27
  28 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
  29 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  30 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  31
  32 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  33                                  btrfs_root *extent_root);
  34 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  35                                btrfs_root *extent_root);
  36 static int find_previous_extent(struct btrfs_root *root,
  37                                 struct btrfs_path *path)
  38 {
  39         struct btrfs_key found_key;
  40         struct extent_buffer *leaf;
  41         int ret;
  42
  43         while(1) {
  44                 if (path->slots[0] == 0) {
  45                         ret = btrfs_prev_leaf(root, path);
  46                         if (ret != 0)
  47                                 return ret;
  48                 } else {
  49                         path->slots[0]--;
  50                 }
  51                 leaf = path->nodes[0];
  52                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  53                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
  54                         return 0;
  55         }
  56         return 1;
  57 }
  58
  59 static int cache_block_group(struct btrfs_root *root,
  60                              struct btrfs_block_group_cache *block_group)
  61 {
  62         struct btrfs_path *path;
  63         int ret;
  64         struct btrfs_key key;
  65         struct extent_buffer *leaf;
  66         struct extent_io_tree *free_space_cache;
  67         int slot;
  68         u64 last = 0;
  69         u64 hole_size;
  70         u64 first_free;
  71         int found = 0;
  72
  73         if (!block_group)
  74                 return 0;
  75
  76         root = root->fs_info->extent_root;
  77         free_space_cache = &root->fs_info->free_space_cache;
  78
  79         if (block_group->cached)
  80                 return 0;
  81
  82         path = btrfs_alloc_path();
  83         if (!path)
  84                 return -ENOMEM;
  85
  86         path->reada = 2;
  87         first_free = block_group->key.objectid;
  88         key.objectid = block_group->key.objectid;
  89         key.offset = 0;
  90         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  91         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  92         if (ret < 0)
  93                 return ret;
  94         ret = find_previous_extent(root, path);
  95         if (ret < 0)
  96                 return ret;
  97         if (ret == 0) {
  98                 leaf = path->nodes[0];
  99                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 100                 if (key.objectid + key.offset > first_free)
 101                         first_free = key.objectid + key.offset;
 102         }
 103         while(1) {
 104                 leaf = path->nodes[0];
 105                 slot = path->slots[0];
 106                 if (slot >= btrfs_header_nritems(leaf)) {
 107                         ret = btrfs_next_leaf(root, path);
 108                         if (ret < 0)
 109                                 goto err;
 110                         if (ret == 0) {
 111                                 continue;
 112                         } else {
 113                                 break;
 114                         }
 115                 }
 116                 btrfs_item_key_to_cpu(leaf, &key, slot);
 117                 if (key.objectid < block_group->key.objectid) {
 118                         goto next;
 119                 }
 120                 if (key.objectid >= block_group->key.objectid +
 121                     block_group->key.offset) {
 122                         break;
 123                 }
 124
 125                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 126                         if (!found) {
 127                                 last = first_free;
 128                                 found = 1;
 129                         }
 130                         if (key.objectid > last) {
 131                                 hole_size = key.objectid - last;
 132                                 set_extent_dirty(free_space_cache, last,
 133                                                  last + hole_size - 1,
 134                                                  GFP_NOFS);
 135                         }
 136                         last = key.objectid + key.offset;
 137                 }
 138 next:
 139                 path->slots[0]++;
 140         }
 141
 142         if (!found)
 143                 last = first_free;
 144         if (block_group->key.objectid +
 145             block_group->key.offset > last) {
 146                 hole_size = block_group->key.objectid +
 147                         block_group->key.offset - last;
 148                 set_extent_dirty(free_space_cache, last,
 149                                  last + hole_size - 1, GFP_NOFS);
 150         }
 151         block_group->cached = 1;
 152 err:
 153         btrfs_free_path(path);
 154         return 0;
 155 }
 156
 157 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 158                                                          btrfs_fs_info *info,
 159                                                          u64 bytenr)
 160 {
 161         struct extent_io_tree *block_group_cache;
 162         struct btrfs_block_group_cache *block_group = NULL;
 163         u64 ptr;
 164         u64 start;
 165         u64 end;
 166         int ret;
 167
 168         block_group_cache = &info->block_group_cache;
 169         ret = find_first_extent_bit(block_group_cache,
 170                                     bytenr, &start, &end,
 171                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
 172         if (ret) {
 173                 return NULL;
 174         }
 175         ret = get_state_private(block_group_cache, start, &ptr);
 176         if (ret)
 177                 return NULL;
 178
 179         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 180         if (block_group->key.objectid <= bytenr && bytenr <
 181             block_group->key.objectid + block_group->key.offset)
 182                 return block_group;
 183         return NULL;
 184 }
 185 static u64 noinline find_search_start(struct btrfs_root *root,
 186                               struct btrfs_block_group_cache **cache_ret,
 187                               u64 search_start, int num, int data)
 188 {
 189         int ret;
 190         struct btrfs_block_group_cache *cache = *cache_ret;
 191         u64 last;
 192         u64 start = 0;
 193         u64 end = 0;
 194         u64 cache_miss = 0;
 195         u64 total_fs_bytes;
 196         int wrapped = 0;
 197
 198         if (!cache) {
 199                 goto out;
 200         }
 201         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 202 again:
 203         ret = cache_block_group(root, cache);
 204         if (ret)
 205                 goto out;
 206
 207         last = max(search_start, cache->key.objectid);
 208
 209         while(1) {
 210                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 211                                             last, &start, &end, EXTENT_DIRTY);
 212                 if (ret) {
 213                         if (!cache_miss)
 214                                 cache_miss = last;
 215                         goto new_group;
 216                 }
 217
 218                 start = max(last, start);
 219                 last = end + 1;
 220                 if (last - start < num) {
 221                         if (last == cache->key.objectid + cache->key.offset)
 222                                 cache_miss = start;
 223                         continue;
 224                 }
 225                 if (data != BTRFS_BLOCK_GROUP_MIXED &&
 226                     start + num > cache->key.objectid + cache->key.offset)
 227                         goto new_group;
 228                 if (start + num  > total_fs_bytes)
 229                         goto new_group;
 230                 return start;
 231         }
 232 out:
 233         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 234         if (!cache) {
 235                 printk("Unable to find block group for %Lu\n",
 236                        search_start);
 237                 WARN_ON(1);
 238                 return search_start;
 239         }
 240         return search_start;
 241
 242 new_group:
 243         last = cache->key.objectid + cache->key.offset;
 244 wrapped:
 245         cache = btrfs_lookup_block_group(root->fs_info, last);
 246         if (!cache || cache->key.objectid >= total_fs_bytes) {
 247 no_cache:
 248                 if (!wrapped) {
 249                         wrapped = 1;
 250                         last = search_start;
 251                         data = BTRFS_BLOCK_GROUP_MIXED;
 252                         goto wrapped;
 253                 }
 254                 goto out;
 255         }
 256         if (cache_miss && !cache->cached) {
 257                 cache_block_group(root, cache);
 258                 last = cache_miss;
 259                 cache = btrfs_lookup_block_group(root->fs_info, last);
 260         }
 261         cache = btrfs_find_block_group(root, cache, last, data, 0);
 262         if (!cache)
 263                 goto no_cache;
 264         *cache_ret = cache;
 265         cache_miss = 0;
 266         goto again;
 267 }
 268
 269 static u64 div_factor(u64 num, int factor)
 270 {
 271         if (factor == 10)
 272                 return num;
 273         num *= factor;
 274         do_div(num, 10);
 275         return num;
 276 }
 277
 278 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 279                                                  struct btrfs_block_group_cache
 280                                                  *hint, u64 search_start,
 281                                                  int data, int owner)
 282 {
 283         struct btrfs_block_group_cache *cache;
 284         struct extent_io_tree *block_group_cache;
 285         struct btrfs_block_group_cache *found_group = NULL;
 286         struct btrfs_fs_info *info = root->fs_info;
 287         u64 used;
 288         u64 last = 0;
 289         u64 hint_last;
 290         u64 start;
 291         u64 end;
 292         u64 free_check;
 293         u64 ptr;
 294         u64 total_fs_bytes;
 295         int bit;
 296         int ret;
 297         int full_search = 0;
 298         int factor = 8;
 299         int data_swap = 0;
 300
 301         block_group_cache = &info->block_group_cache;
 302         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 303
 304         if (!owner)
 305                 factor = 8;
 306
 307         if (data == BTRFS_BLOCK_GROUP_MIXED) {
 308                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 309                 factor = 10;
 310         } else if (data)
 311                 bit = BLOCK_GROUP_DATA;
 312         else
 313                 bit = BLOCK_GROUP_METADATA;
 314
 315         if (search_start && search_start < total_fs_bytes) {
 316                 struct btrfs_block_group_cache *shint;
 317                 shint = btrfs_lookup_block_group(info, search_start);
 318                 if (shint && (shint->data == data ||
 319                               shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 320                         used = btrfs_block_group_used(&shint->item);
 321                         if (used + shint->pinned <
 322                             div_factor(shint->key.offset, factor)) {
 323                                 return shint;
 324                         }
 325                 }
 326         }
 327         if (hint && hint->key.objectid < total_fs_bytes &&
 328             (hint->data == data || hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 329                 used = btrfs_block_group_used(&hint->item);
 330                 if (used + hint->pinned <
 331                     div_factor(hint->key.offset, factor)) {
 332                         return hint;
 333                 }
 334                 last = hint->key.objectid + hint->key.offset;
 335                 hint_last = last;
 336         } else {
 337                 if (hint)
 338                         hint_last = max(hint->key.objectid, search_start);
 339                 else
 340                         hint_last = search_start;
 341
 342                 if (hint_last >= total_fs_bytes)
 343                         hint_last = search_start;
 344                 last = hint_last;
 345         }
 346 again:
 347         while(1) {
 348                 ret = find_first_extent_bit(block_group_cache, last,
 349                                             &start, &end, bit);
 350                 if (ret)
 351                         break;
 352
 353                 ret = get_state_private(block_group_cache, start, &ptr);
 354                 if (ret)
 355                         break;
 356
 357                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 358                 last = cache->key.objectid + cache->key.offset;
 359                 used = btrfs_block_group_used(&cache->item);
 360
 361                 if (cache->key.objectid > total_fs_bytes)
 362                         break;
 363
 364                 if (full_search)
 365                         free_check = cache->key.offset;
 366                 else
 367                         free_check = div_factor(cache->key.offset, factor);
 368                 if (used + cache->pinned < free_check) {
 369                         found_group = cache;
 370                         goto found;
 371                 }
 372                 cond_resched();
 373         }
 374         if (!full_search) {
 375                 last = search_start;
 376                 full_search = 1;
 377                 goto again;
 378         }
 379         if (!data_swap) {
 380                 data_swap = 1;
 381                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 382                 last = search_start;
 383                 goto again;
 384         }
 385 found:
 386         return found_group;
 387 }
 388
 389 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 390                            u64 owner, u64 owner_offset)
 391 {
 392         u32 high_crc = ~(u32)0;
 393         u32 low_crc = ~(u32)0;
 394         __le64 lenum;
 395
 396         lenum = cpu_to_le64(root_objectid);
 397         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 398         lenum = cpu_to_le64(ref_generation);
 399         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 400         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 401                 lenum = cpu_to_le64(owner);
 402                 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 403                 lenum = cpu_to_le64(owner_offset);
 404                 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 405         }
 406         return ((u64)high_crc << 32) | (u64)low_crc;
 407 }
 408
 409 static int match_extent_ref(struct extent_buffer *leaf,
 410                             struct btrfs_extent_ref *disk_ref,
 411                             struct btrfs_extent_ref *cpu_ref)
 412 {
 413         int ret;
 414         int len;
 415
 416         if (cpu_ref->objectid)
 417                 len = sizeof(*cpu_ref);
 418         else
 419                 len = 2 * sizeof(u64);
 420         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 421                                    len);
 422         return ret == 0;
 423 }
 424
 425 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 426                                           struct btrfs_root *root,
 427                                           struct btrfs_path *path, u64 bytenr,
 428                                           u64 root_objectid,
 429                                           u64 ref_generation, u64 owner,
 430                                           u64 owner_offset, int del)
 431 {
 432         u64 hash;
 433         struct btrfs_key key;
 434         struct btrfs_key found_key;
 435         struct btrfs_extent_ref ref;
 436         struct extent_buffer *leaf;
 437         struct btrfs_extent_ref *disk_ref;
 438         int ret;
 439         int ret2;
 440
 441         btrfs_set_stack_ref_root(&ref, root_objectid);
 442         btrfs_set_stack_ref_generation(&ref, ref_generation);
 443         btrfs_set_stack_ref_objectid(&ref, owner);
 444         btrfs_set_stack_ref_offset(&ref, owner_offset);
 445
 446         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 447                                owner_offset);
 448         key.offset = hash;
 449         key.objectid = bytenr;
 450         key.type = BTRFS_EXTENT_REF_KEY;
 451
 452         while (1) {
 453                 ret = btrfs_search_slot(trans, root, &key, path,
 454                                         del ? -1 : 0, del);
 455                 if (ret < 0)
 456                         goto out;
 457                 leaf = path->nodes[0];
 458                 if (ret != 0) {
 459                         u32 nritems = btrfs_header_nritems(leaf);
 460                         if (path->slots[0] >= nritems) {
 461                                 ret2 = btrfs_next_leaf(root, path);
 462                                 if (ret2)
 463                                         goto out;
 464                                 leaf = path->nodes[0];
 465                         }
 466                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 467                         if (found_key.objectid != bytenr ||
 468                             found_key.type != BTRFS_EXTENT_REF_KEY)
 469                                 goto out;
 470                         key.offset = found_key.offset;
 471                         if (del) {
 472                                 btrfs_release_path(root, path);
 473                                 continue;
 474                         }
 475                 }
 476                 disk_ref = btrfs_item_ptr(path->nodes[0],
 477                                           path->slots[0],
 478                                           struct btrfs_extent_ref);
 479                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 480                         ret = 0;
 481                         goto out;
 482                 }
 483                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 484                 key.offset = found_key.offset + 1;
 485                 btrfs_release_path(root, path);
 486         }
 487 out:
 488         return ret;
 489 }
 490
 491 /*
 492  * Back reference rules.  Back refs have three main goals:
 493  *
 494  * 1) differentiate between all holders of references to an extent so that
 495  *    when a reference is dropped we can make sure it was a valid reference
 496  *    before freeing the extent.
 497  *
 498  * 2) Provide enough information to quickly find the holders of an extent
 499  *    if we notice a given block is corrupted or bad.
 500  *
 501  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 502  *    maintenance.  This is actually the same as #2, but with a slightly
 503  *    different use case.
 504  *
 505  * File extents can be referenced by:
 506  *
 507  * - multiple snapshots, subvolumes, or different generations in one subvol
 508  * - different files inside a single subvolume (in theory, not implemented yet)
 509  * - different offsets inside a file (bookend extents in file.c)
 510  *
 511  * The extent ref structure has fields for:
 512  *
 513  * - Objectid of the subvolume root
 514  * - Generation number of the tree holding the reference
 515  * - objectid of the file holding the reference
 516  * - offset in the file corresponding to the key holding the reference
 517  *
 518  * When a file extent is allocated the fields are filled in:
 519  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 520  *
 521  * When a leaf is cow'd new references are added for every file extent found
 522  * in the leaf.  It looks the same as the create case, but trans->transid
 523  * will be different when the block is cow'd.
 524  *
 525  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 526  *
 527  * When a file extent is removed either during snapshot deletion or file
 528  * truncation, the corresponding back reference is found
 529  * by searching for:
 530  *
 531  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 532  *      inode objectid, offset in file)
 533  *
 534  * Btree extents can be referenced by:
 535  *
 536  * - Different subvolumes
 537  * - Different generations of the same subvolume
 538  *
 539  * Storing sufficient information for a full reverse mapping of a btree
 540  * block would require storing the lowest key of the block in the backref,
 541  * and it would require updating that lowest key either before write out or
 542  * every time it changed.  Instead, the objectid of the lowest key is stored
 543  * along with the level of the tree block.  This provides a hint
 544  * about where in the btree the block can be found.  Searches through the
 545  * btree only need to look for a pointer to that block, so they stop one
 546  * level higher than the level recorded in the backref.
 547  *
 548  * Some btrees do not do reference counting on their extents.  These
 549  * include the extent tree and the tree of tree roots.  Backrefs for these
 550  * trees always have a generation of zero.
 551  *
 552  * When a tree block is created, back references are inserted:
 553  *
 554  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 555  *
 556  * When a tree block is cow'd in a reference counted root,
 557  * new back references are added for all the blocks it points to.
 558  * These are of the form (trans->transid will have increased since creation):
 559  *
 560  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 561  *
 562  * Because the lowest_key_objectid and the level are just hints
 563  * they are not used when backrefs are deleted.  When a backref is deleted:
 564  *
 565  * if backref was for a tree root:
 566  *     root_objectid = root->root_key.objectid
 567  * else
 568  *     root_objectid = btrfs_header_owner(parent)
 569  *
 570  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 571  *
 572  * Back Reference Key hashing:
 573  *
 574  * Back references have four fields, each 64 bits long.  Unfortunately,
 575  * This is hashed into a single 64 bit number and placed into the key offset.
 576  * The key objectid corresponds to the first byte in the extent, and the
 577  * key type is set to BTRFS_EXTENT_REF_KEY
 578  */
 579 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 580                                  struct btrfs_root *root,
 581                                  struct btrfs_path *path, u64 bytenr,
 582                                  u64 root_objectid, u64 ref_generation,
 583                                  u64 owner, u64 owner_offset)
 584 {
 585         u64 hash;
 586         struct btrfs_key key;
 587         struct btrfs_extent_ref ref;
 588         struct btrfs_extent_ref *disk_ref;
 589         int ret;
 590
 591         btrfs_set_stack_ref_root(&ref, root_objectid);
 592         btrfs_set_stack_ref_generation(&ref, ref_generation);
 593         btrfs_set_stack_ref_objectid(&ref, owner);
 594         btrfs_set_stack_ref_offset(&ref, owner_offset);
 595
 596         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 597                                owner_offset);
 598         key.offset = hash;
 599         key.objectid = bytenr;
 600         key.type = BTRFS_EXTENT_REF_KEY;
 601
 602         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 603         while (ret == -EEXIST) {
 604                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 605                                           struct btrfs_extent_ref);
 606                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 607                         goto out;
 608                 key.offset++;
 609                 btrfs_release_path(root, path);
 610                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 611                                               sizeof(ref));
 612         }
 613         if (ret)
 614                 goto out;
 615         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 616                                   struct btrfs_extent_ref);
 617         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 618                             sizeof(ref));
 619         btrfs_mark_buffer_dirty(path->nodes[0]);
 620 out:
 621         btrfs_release_path(root, path);
 622         return ret;
 623 }
 624
 625 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 626                                 struct btrfs_root *root,
 627                                 u64 bytenr, u64 num_bytes,
 628                                 u64 root_objectid, u64 ref_generation,
 629                                 u64 owner, u64 owner_offset)
 630 {
 631         struct btrfs_path *path;
 632         int ret;
 633         struct btrfs_key key;
 634         struct extent_buffer *l;
 635         struct btrfs_extent_item *item;
 636         u32 refs;
 637
 638         WARN_ON(num_bytes < root->sectorsize);
 639         path = btrfs_alloc_path();
 640         if (!path)
 641                 return -ENOMEM;
 642
 643         path->reada = 0;
 644         key.objectid = bytenr;
 645         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 646         key.offset = num_bytes;
 647         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 648                                 0, 1);
 649         if (ret < 0)
 650                 return ret;
 651         if (ret != 0) {
 652                 BUG();
 653         }
 654         BUG_ON(ret != 0);
 655         l = path->nodes[0];
 656         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 657         refs = btrfs_extent_refs(l, item);
 658         btrfs_set_extent_refs(l, item, refs + 1);
 659         btrfs_mark_buffer_dirty(path->nodes[0]);
 660
 661         btrfs_release_path(root->fs_info->extent_root, path);
 662
 663         path->reada = 0;
 664         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 665                                           path, bytenr, root_objectid,
 666                                           ref_generation, owner, owner_offset);
 667         BUG_ON(ret);
 668         finish_current_insert(trans, root->fs_info->extent_root);
 669         del_pending_extents(trans, root->fs_info->extent_root);
 670
 671         btrfs_free_path(path);
 672         return 0;
 673 }
 674
 675 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 676                          struct btrfs_root *root)
 677 {
 678         finish_current_insert(trans, root->fs_info->extent_root);
 679         del_pending_extents(trans, root->fs_info->extent_root);
 680         return 0;
 681 }
 682
 683 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 684                              struct btrfs_root *root, u64 bytenr,
 685                              u64 num_bytes, u32 *refs)
 686 {
 687         struct btrfs_path *path;
 688         int ret;
 689         struct btrfs_key key;
 690         struct extent_buffer *l;
 691         struct btrfs_extent_item *item;
 692
 693         WARN_ON(num_bytes < root->sectorsize);
 694         path = btrfs_alloc_path();
 695         path->reada = 0;
 696         key.objectid = bytenr;
 697         key.offset = num_bytes;
 698         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 699         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 700                                 0, 0);
 701         if (ret < 0)
 702                 goto out;
 703         if (ret != 0) {
 704                 btrfs_print_leaf(root, path->nodes[0]);
 705                 printk("failed to find block number %Lu\n", bytenr);
 706                 BUG();
 707         }
 708         l = path->nodes[0];
 709         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 710         *refs = btrfs_extent_refs(l, item);
 711 out:
 712         btrfs_free_path(path);
 713         return 0;
 714 }
 715
 716 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 717                                   struct btrfs_path *count_path,
 718                                   u64 first_extent)
 719 {
 720         struct btrfs_root *extent_root = root->fs_info->extent_root;
 721         struct btrfs_path *path;
 722         u64 bytenr;
 723         u64 found_objectid;
 724         u64 root_objectid = root->root_key.objectid;
 725         u32 total_count = 0;
 726         u32 cur_count;
 727         u32 nritems;
 728         int ret;
 729         struct btrfs_key key;
 730         struct btrfs_key found_key;
 731         struct extent_buffer *l;
 732         struct btrfs_extent_item *item;
 733         struct btrfs_extent_ref *ref_item;
 734         int level = -1;
 735
 736         path = btrfs_alloc_path();
 737 again:
 738         if (level == -1)
 739                 bytenr = first_extent;
 740         else
 741                 bytenr = count_path->nodes[level]->start;
 742
 743         cur_count = 0;
 744         key.objectid = bytenr;
 745         key.offset = 0;
 746
 747         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 748         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 749         if (ret < 0)
 750                 goto out;
 751         BUG_ON(ret == 0);
 752
 753         l = path->nodes[0];
 754         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 755
 756         if (found_key.objectid != bytenr ||
 757             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 758                 goto out;
 759         }
 760
 761         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 762         while (1) {
 763                 l = path->nodes[0];
 764                 nritems = btrfs_header_nritems(l);
 765                 if (path->slots[0] >= nritems) {
 766                         ret = btrfs_next_leaf(extent_root, path);
 767                         if (ret == 0)
 768                                 continue;
 769                         break;
 770                 }
 771                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 772                 if (found_key.objectid != bytenr)
 773                         break;
 774
 775                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 776                         path->slots[0]++;
 777                         continue;
 778                 }
 779
 780                 cur_count++;
 781                 ref_item = btrfs_item_ptr(l, path->slots[0],
 782                                           struct btrfs_extent_ref);
 783                 found_objectid = btrfs_ref_root(l, ref_item);
 784
 785                 if (found_objectid != root_objectid) {
 786                         total_count = 2;
 787                         goto out;
 788                 }
 789                 total_count = 1;
 790                 path->slots[0]++;
 791         }
 792         if (cur_count == 0) {
 793                 total_count = 0;
 794                 goto out;
 795         }
 796         if (level >= 0 && root->node == count_path->nodes[level])
 797                 goto out;
 798         level++;
 799         btrfs_release_path(root, path);
 800         goto again;
 801
 802 out:
 803         btrfs_free_path(path);
 804         return total_count;
 805 }
 806 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 807                        struct btrfs_root *root, u64 owner_objectid)
 808 {
 809         u64 generation;
 810         u64 key_objectid;
 811         u64 level;
 812         u32 nritems;
 813         struct btrfs_disk_key disk_key;
 814
 815         level = btrfs_header_level(root->node);
 816         generation = trans->transid;
 817         nritems = btrfs_header_nritems(root->node);
 818         if (nritems > 0) {
 819                 if (level == 0)
 820                         btrfs_item_key(root->node, &disk_key, 0);
 821                 else
 822                         btrfs_node_key(root->node, &disk_key, 0);
 823                 key_objectid = btrfs_disk_key_objectid(&disk_key);
 824         } else {
 825                 key_objectid = 0;
 826         }
 827         return btrfs_inc_extent_ref(trans, root, root->node->start,
 828                                     root->node->len, owner_objectid,
 829                                     generation, level, key_objectid);
 830 }
 831
 832 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 833                   struct extent_buffer *buf)
 834 {
 835         u64 bytenr;
 836         u32 nritems;
 837         struct btrfs_key key;
 838         struct btrfs_file_extent_item *fi;
 839         int i;
 840         int level;
 841         int ret;
 842         int faili;
 843
 844         if (!root->ref_cows)
 845                 return 0;
 846
 847         level = btrfs_header_level(buf);
 848         nritems = btrfs_header_nritems(buf);
 849         for (i = 0; i < nritems; i++) {
 850                 if (level == 0) {
 851                         u64 disk_bytenr;
 852                         btrfs_item_key_to_cpu(buf, &key, i);
 853                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 854                                 continue;
 855                         fi = btrfs_item_ptr(buf, i,
 856                                             struct btrfs_file_extent_item);
 857                         if (btrfs_file_extent_type(buf, fi) ==
 858                             BTRFS_FILE_EXTENT_INLINE)
 859                                 continue;
 860                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 861                         if (disk_bytenr == 0)
 862                                 continue;
 863                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
 864                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 865                                     root->root_key.objectid, trans->transid,
 866                                     key.objectid, key.offset);
 867                         if (ret) {
 868                                 faili = i;
 869                                 goto fail;
 870                         }
 871                 } else {
 872                         bytenr = btrfs_node_blockptr(buf, i);
 873                         btrfs_node_key_to_cpu(buf, &key, i);
 874                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
 875                                            btrfs_level_size(root, level - 1),
 876                                            root->root_key.objectid,
 877                                            trans->transid,
 878                                            level - 1, key.objectid);
 879                         if (ret) {
 880                                 faili = i;
 881                                 goto fail;
 882                         }
 883                 }
 884         }
 885         return 0;
 886 fail:
 887         WARN_ON(1);
 888 #if 0
 889         for (i =0; i < faili; i++) {
 890                 if (level == 0) {
 891                         u64 disk_bytenr;
 892                         btrfs_item_key_to_cpu(buf, &key, i);
 893                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 894                                 continue;
 895                         fi = btrfs_item_ptr(buf, i,
 896                                             struct btrfs_file_extent_item);
 897                         if (btrfs_file_extent_type(buf, fi) ==
 898                             BTRFS_FILE_EXTENT_INLINE)
 899                                 continue;
 900                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 901                         if (disk_bytenr == 0)
 902                                 continue;
 903                         err = btrfs_free_extent(trans, root, disk_bytenr,
 904                                     btrfs_file_extent_disk_num_bytes(buf,
 905                                                                       fi), 0);
 906                         BUG_ON(err);
 907                 } else {
 908                         bytenr = btrfs_node_blockptr(buf, i);
 909                         err = btrfs_free_extent(trans, root, bytenr,
 910                                         btrfs_level_size(root, level - 1), 0);
 911                         BUG_ON(err);
 912                 }
 913         }
 914 #endif
 915         return ret;
 916 }
 917
 918 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 919                                  struct btrfs_root *root,
 920                                  struct btrfs_path *path,
 921                                  struct btrfs_block_group_cache *cache)
 922 {
 923         int ret;
 924         int pending_ret;
 925         struct btrfs_root *extent_root = root->fs_info->extent_root;
 926         unsigned long bi;
 927         struct extent_buffer *leaf;
 928
 929         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 930         if (ret < 0)
 931                 goto fail;
 932         BUG_ON(ret);
 933
 934         leaf = path->nodes[0];
 935         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 936         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 937         btrfs_mark_buffer_dirty(leaf);
 938         btrfs_release_path(extent_root, path);
 939 fail:
 940         finish_current_insert(trans, extent_root);
 941         pending_ret = del_pending_extents(trans, extent_root);
 942         if (ret)
 943                 return ret;
 944         if (pending_ret)
 945                 return pending_ret;
 946         return 0;
 947
 948 }
 949
 950 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 951                                    struct btrfs_root *root)
 952 {
 953         struct extent_io_tree *block_group_cache;
 954         struct btrfs_block_group_cache *cache;
 955         int ret;
 956         int err = 0;
 957         int werr = 0;
 958         struct btrfs_path *path;
 959         u64 last = 0;
 960         u64 start;
 961         u64 end;
 962         u64 ptr;
 963
 964         block_group_cache = &root->fs_info->block_group_cache;
 965         path = btrfs_alloc_path();
 966         if (!path)
 967                 return -ENOMEM;
 968
 969         while(1) {
 970                 ret = find_first_extent_bit(block_group_cache, last,
 971                                             &start, &end, BLOCK_GROUP_DIRTY);
 972                 if (ret)
 973                         break;
 974
 975                 last = end + 1;
 976                 ret = get_state_private(block_group_cache, start, &ptr);
 977                 if (ret)
 978                         break;
 979
 980                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 981                 err = write_one_cache_group(trans, root,
 982                                             path, cache);
 983                 /*
 984                  * if we fail to write the cache group, we want
 985                  * to keep it marked dirty in hopes that a later
 986                  * write will work
 987                  */
 988                 if (err) {
 989                         werr = err;
 990                         continue;
 991                 }
 992                 clear_extent_bits(block_group_cache, start, end,
 993                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 994         }
 995         btrfs_free_path(path);
 996         return werr;
 997 }
 998
 999 static int update_block_group(struct btrfs_trans_handle *trans,
1000                               struct btrfs_root *root,
1001                               u64 bytenr, u64 num_bytes, int alloc,
1002                               int mark_free, int data)
1003 {
1004         struct btrfs_block_group_cache *cache;
1005         struct btrfs_fs_info *info = root->fs_info;
1006         u64 total = num_bytes;
1007         u64 old_val;
1008         u64 byte_in_group;
1009         u64 start;
1010         u64 end;
1011
1012         while(total) {
1013                 cache = btrfs_lookup_block_group(info, bytenr);
1014                 if (!cache) {
1015                         return -1;
1016                 }
1017                 byte_in_group = bytenr - cache->key.objectid;
1018                 WARN_ON(byte_in_group > cache->key.offset);
1019                 start = cache->key.objectid;
1020                 end = start + cache->key.offset - 1;
1021                 set_extent_bits(&info->block_group_cache, start, end,
1022                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1023
1024                 old_val = btrfs_block_group_used(&cache->item);
1025                 num_bytes = min(total, cache->key.offset - byte_in_group);
1026                 if (alloc) {
1027                         if (cache->data != data &&
1028                             old_val < (cache->key.offset >> 1)) {
1029                                 int bit_to_clear;
1030                                 int bit_to_set;
1031                                 cache->data = data;
1032                                 if (data) {
1033                                         bit_to_clear = BLOCK_GROUP_METADATA;
1034                                         bit_to_set = BLOCK_GROUP_DATA;
1035                                         cache->item.flags &=
1036                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1037                                         cache->item.flags |=
1038                                                 BTRFS_BLOCK_GROUP_DATA;
1039                                 } else {
1040                                         bit_to_clear = BLOCK_GROUP_DATA;
1041                                         bit_to_set = BLOCK_GROUP_METADATA;
1042                                         cache->item.flags &=
1043                                                 ~BTRFS_BLOCK_GROUP_MIXED;
1044                                         cache->item.flags &=
1045                                                 ~BTRFS_BLOCK_GROUP_DATA;
1046                                 }
1047                                 clear_extent_bits(&info->block_group_cache,
1048                                                   start, end, bit_to_clear,
1049                                                   GFP_NOFS);
1050                                 set_extent_bits(&info->block_group_cache,
1051                                                 start, end, bit_to_set,
1052                                                 GFP_NOFS);
1053                         } else if (cache->data != data &&
1054                                    cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1055                                 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1056                                 set_extent_bits(&info->block_group_cache,
1057                                                 start, end,
1058                                                 BLOCK_GROUP_DATA |
1059                                                 BLOCK_GROUP_METADATA,
1060                                                 GFP_NOFS);
1061                         }
1062                         old_val += num_bytes;
1063                 } else {
1064                         old_val -= num_bytes;
1065                         if (mark_free) {
1066                                 set_extent_dirty(&info->free_space_cache,
1067                                                  bytenr, bytenr + num_bytes - 1,
1068                                                  GFP_NOFS);
1069                         }
1070                 }
1071                 btrfs_set_block_group_used(&cache->item, old_val);
1072                 total -= num_bytes;
1073                 bytenr += num_bytes;
1074         }
1075         return 0;
1076 }
1077 static int update_pinned_extents(struct btrfs_root *root,
1078                                 u64 bytenr, u64 num, int pin)
1079 {
1080         u64 len;
1081         struct btrfs_block_group_cache *cache;
1082         struct btrfs_fs_info *fs_info = root->fs_info;
1083
1084         if (pin) {
1085                 set_extent_dirty(&fs_info->pinned_extents,
1086                                 bytenr, bytenr + num - 1, GFP_NOFS);
1087         } else {
1088                 clear_extent_dirty(&fs_info->pinned_extents,
1089                                 bytenr, bytenr + num - 1, GFP_NOFS);
1090         }
1091         while (num > 0) {
1092                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1093                 WARN_ON(!cache);
1094                 len = min(num, cache->key.offset -
1095                           (bytenr - cache->key.objectid));
1096                 if (pin) {
1097                         cache->pinned += len;
1098                         fs_info->total_pinned += len;
1099                 } else {
1100                         cache->pinned -= len;
1101                         fs_info->total_pinned -= len;
1102                 }
1103                 bytenr += len;
1104                 num -= len;
1105         }
1106         return 0;
1107 }
1108
1109 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1110 {
1111         u64 last = 0;
1112         u64 start;
1113         u64 end;
1114         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1115         int ret;
1116
1117         while(1) {
1118                 ret = find_first_extent_bit(pinned_extents, last,
1119                                             &start, &end, EXTENT_DIRTY);
1120                 if (ret)
1121                         break;
1122                 set_extent_dirty(copy, start, end, GFP_NOFS);
1123                 last = end + 1;
1124         }
1125         return 0;
1126 }
1127
1128 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1129                                struct btrfs_root *root,
1130                                struct extent_io_tree *unpin)
1131 {
1132         u64 start;
1133         u64 end;
1134         int ret;
1135         struct extent_io_tree *free_space_cache;
1136         free_space_cache = &root->fs_info->free_space_cache;
1137
1138         while(1) {
1139                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1140                                             EXTENT_DIRTY);
1141                 if (ret)
1142                         break;
1143                 update_pinned_extents(root, start, end + 1 - start, 0);
1144                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1145                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1146         }
1147         return 0;
1148 }
1149
1150 static int finish_current_insert(struct btrfs_trans_handle *trans,
1151                                  struct btrfs_root *extent_root)
1152 {
1153         u64 start;
1154         u64 end;
1155         struct btrfs_fs_info *info = extent_root->fs_info;
1156         struct extent_buffer *eb;
1157         struct btrfs_path *path;
1158         struct btrfs_key ins;
1159         struct btrfs_disk_key first;
1160         struct btrfs_extent_item extent_item;
1161         int ret;
1162         int level;
1163         int err = 0;
1164
1165         btrfs_set_stack_extent_refs(&extent_item, 1);
1166         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1167         path = btrfs_alloc_path();
1168
1169         while(1) {
1170                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1171                                             &end, EXTENT_LOCKED);
1172                 if (ret)
1173                         break;
1174
1175                 ins.objectid = start;
1176                 ins.offset = end + 1 - start;
1177                 err = btrfs_insert_item(trans, extent_root, &ins,
1178                                         &extent_item, sizeof(extent_item));
1179                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1180                                   GFP_NOFS);
1181                 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1182                 level = btrfs_header_level(eb);
1183                 if (level == 0) {
1184                         btrfs_item_key(eb, &first, 0);
1185                 } else {
1186                         btrfs_node_key(eb, &first, 0);
1187                 }
1188                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1189                                           start, extent_root->root_key.objectid,
1190                                           0, level,
1191                                           btrfs_disk_key_objectid(&first));
1192                 BUG_ON(err);
1193                 free_extent_buffer(eb);
1194         }
1195         btrfs_free_path(path);
1196         return 0;
1197 }
1198
1199 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1200                           int pending)
1201 {
1202         int err = 0;
1203         struct extent_buffer *buf;
1204
1205         if (!pending) {
1206                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1207                 if (buf) {
1208                         if (btrfs_buffer_uptodate(buf)) {
1209                                 u64 transid =
1210                                     root->fs_info->running_transaction->transid;
1211                                 u64 header_transid =
1212                                         btrfs_header_generation(buf);
1213                                 if (header_transid == transid) {
1214                                         clean_tree_block(NULL, root, buf);
1215                                         free_extent_buffer(buf);
1216                                         return 1;
1217                                 }
1218                         }
1219                         free_extent_buffer(buf);
1220                 }
1221                 update_pinned_extents(root, bytenr, num_bytes, 1);
1222         } else {
1223                 set_extent_bits(&root->fs_info->pending_del,
1224                                 bytenr, bytenr + num_bytes - 1,
1225                                 EXTENT_LOCKED, GFP_NOFS);
1226         }
1227         BUG_ON(err < 0);
1228         return 0;
1229 }
1230
1231 /*
1232  * remove an extent from the root, returns 0 on success
1233  */
1234 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1235                          *root, u64 bytenr, u64 num_bytes,
1236                          u64 root_objectid, u64 ref_generation,
1237                          u64 owner_objectid, u64 owner_offset, int pin,
1238                          int mark_free)
1239 {
1240         struct btrfs_path *path;
1241         struct btrfs_key key;
1242         struct btrfs_fs_info *info = root->fs_info;
1243         struct btrfs_root *extent_root = info->extent_root;
1244         struct extent_buffer *leaf;
1245         int ret;
1246         struct btrfs_extent_item *ei;
1247         u32 refs;
1248
1249         key.objectid = bytenr;
1250         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1251         key.offset = num_bytes;
1252         path = btrfs_alloc_path();
1253         if (!path)
1254                 return -ENOMEM;
1255
1256         path->reada = 0;
1257         ret = lookup_extent_backref(trans, extent_root, path,
1258                                     bytenr, root_objectid,
1259                                     ref_generation,
1260                                     owner_objectid, owner_offset, 1);
1261         if (ret == 0) {
1262                 ret = btrfs_del_item(trans, extent_root, path);
1263         } else {
1264                 btrfs_print_leaf(extent_root, path->nodes[0]);
1265                 WARN_ON(1);
1266                 printk("Unable to find ref byte nr %Lu root %Lu "
1267                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1268                        root_objectid, ref_generation, owner_objectid,
1269                        owner_offset);
1270         }
1271         btrfs_release_path(extent_root, path);
1272         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1273         if (ret < 0)
1274                 return ret;
1275         BUG_ON(ret);
1276
1277         leaf = path->nodes[0];
1278         ei = btrfs_item_ptr(leaf, path->slots[0],
1279                             struct btrfs_extent_item);
1280         refs = btrfs_extent_refs(leaf, ei);
1281         BUG_ON(refs == 0);
1282         refs -= 1;
1283         btrfs_set_extent_refs(leaf, ei, refs);
1284         btrfs_mark_buffer_dirty(leaf);
1285
1286         if (refs == 0) {
1287                 u64 super_used;
1288                 u64 root_used;
1289
1290                 if (pin) {
1291                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1292                         if (ret > 0)
1293                                 mark_free = 1;
1294                         BUG_ON(ret < 0);
1295                 }
1296
1297                 /* block accounting for super block */
1298                 super_used = btrfs_super_bytes_used(&info->super_copy);
1299                 btrfs_set_super_bytes_used(&info->super_copy,
1300                                            super_used - num_bytes);
1301
1302                 /* block accounting for root item */
1303                 root_used = btrfs_root_used(&root->root_item);
1304                 btrfs_set_root_used(&root->root_item,
1305                                            root_used - num_bytes);
1306
1307                 ret = btrfs_del_item(trans, extent_root, path);
1308                 if (ret) {
1309                         return ret;
1310                 }
1311                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1312                                          mark_free, 0);
1313                 BUG_ON(ret);
1314         }
1315         btrfs_free_path(path);
1316         finish_current_insert(trans, extent_root);
1317         return ret;
1318 }
1319
1320 /*
1321  * find all the blocks marked as pending in the radix tree and remove
1322  * them from the extent map
1323  */
1324 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1325                                btrfs_root *extent_root)
1326 {
1327         int ret;
1328         int err = 0;
1329         u64 start;
1330         u64 end;
1331         struct extent_io_tree *pending_del;
1332         struct extent_io_tree *pinned_extents;
1333
1334         pending_del = &extent_root->fs_info->pending_del;
1335         pinned_extents = &extent_root->fs_info->pinned_extents;
1336
1337         while(1) {
1338                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1339                                             EXTENT_LOCKED);
1340                 if (ret)
1341                         break;
1342                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1343                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1344                                   GFP_NOFS);
1345                 ret = __free_extent(trans, extent_root,
1346                                      start, end + 1 - start,
1347                                      extent_root->root_key.objectid,
1348                                      0, 0, 0, 0, 0);
1349                 if (ret)
1350                         err = ret;
1351         }
1352         return err;
1353 }
1354
1355 /*
1356  * remove an extent from the root, returns 0 on success
1357  */
1358 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1359                       *root, u64 bytenr, u64 num_bytes,
1360                       u64 root_objectid, u64 ref_generation,
1361                       u64 owner_objectid, u64 owner_offset, int pin)
1362 {
1363         struct btrfs_root *extent_root = root->fs_info->extent_root;
1364         int pending_ret;
1365         int ret;
1366
1367         WARN_ON(num_bytes < root->sectorsize);
1368         if (!root->ref_cows)
1369                 ref_generation = 0;
1370
1371         if (root == extent_root) {
1372                 pin_down_bytes(root, bytenr, num_bytes, 1);
1373                 return 0;
1374         }
1375         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1376                             ref_generation, owner_objectid, owner_offset,
1377                             pin, pin == 0);
1378         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1379         return ret ? ret : pending_ret;
1380 }
1381
1382 static u64 stripe_align(struct btrfs_root *root, u64 val)
1383 {
1384         u64 mask = ((u64)root->stripesize - 1);
1385         u64 ret = (val + mask) & ~mask;
1386         return ret;
1387 }
1388
1389 /*
1390  * walks the btree of allocated extents and find a hole of a given size.
1391  * The key ins is changed to record the hole:
1392  * ins->objectid == block start
1393  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1394  * ins->offset == number of blocks
1395  * Any available blocks before search_start are skipped.
1396  */
1397 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1398                                      struct btrfs_root *orig_root,
1399                                      u64 num_bytes, u64 empty_size,
1400                                      u64 search_start, u64 search_end,
1401                                      u64 hint_byte, struct btrfs_key *ins,
1402                                      u64 exclude_start, u64 exclude_nr,
1403                                      int data)
1404 {
1405         struct btrfs_path *path;
1406         struct btrfs_key key;
1407         u64 hole_size = 0;
1408         u64 aligned;
1409         int ret;
1410         int slot = 0;
1411         u64 last_byte = 0;
1412         u64 *last_ptr = NULL;
1413         u64 orig_search_start = search_start;
1414         int start_found;
1415         struct extent_buffer *l;
1416         struct btrfs_root * root = orig_root->fs_info->extent_root;
1417         struct btrfs_fs_info *info = root->fs_info;
1418         u64 total_needed = num_bytes;
1419         int level;
1420         struct btrfs_block_group_cache *block_group;
1421         int full_scan = 0;
1422         int wrapped = 0;
1423         u64 cached_start;
1424
1425         WARN_ON(num_bytes < root->sectorsize);
1426         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1427
1428         level = btrfs_header_level(root->node);
1429
1430         if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1431                 data = BTRFS_BLOCK_GROUP_MIXED;
1432         }
1433
1434         if (!data)
1435                 last_ptr = &root->fs_info->last_alloc;
1436
1437         if (data && btrfs_test_opt(root, SSD))
1438                 last_ptr = &root->fs_info->last_data_alloc;
1439
1440         if (last_ptr) {
1441                 if (*last_ptr)
1442                         hint_byte = *last_ptr;
1443                 else {
1444                         hint_byte = hint_byte &
1445                                 ~((u64)BTRFS_BLOCK_GROUP_SIZE - 1);
1446                         empty_size += 2 * 1024 * 1024;
1447                 }
1448         }
1449
1450         search_end = min(search_end,
1451                          btrfs_super_total_bytes(&info->super_copy));
1452         if (hint_byte) {
1453                 block_group = btrfs_lookup_block_group(info, hint_byte);
1454                 if (!block_group)
1455                         hint_byte = search_start;
1456                 block_group = btrfs_find_block_group(root, block_group,
1457                                                      hint_byte, data, 1);
1458         } else {
1459                 block_group = btrfs_find_block_group(root,
1460                                                      trans->block_group,
1461                                                      search_start, data, 1);
1462         }
1463
1464         total_needed += empty_size;
1465         path = btrfs_alloc_path();
1466 check_failed:
1467         if (!block_group) {
1468                 block_group = btrfs_lookup_block_group(info, search_start);
1469                 if (!block_group)
1470                         block_group = btrfs_lookup_block_group(info,
1471                                                        orig_search_start);
1472         }
1473         search_start = find_search_start(root, &block_group, search_start,
1474                                          total_needed, data);
1475
1476         if (last_ptr && *last_ptr && search_start != *last_ptr) {
1477                 *last_ptr = 0;
1478                 if (!empty_size) {
1479                         empty_size += 16 * 1024 * 1024;
1480                         total_needed += empty_size;
1481                 }
1482                 search_start = find_search_start(root, &block_group,
1483                                                  search_start, total_needed,
1484                                                  data);
1485         }
1486
1487         search_start = stripe_align(root, search_start);
1488         cached_start = search_start;
1489         btrfs_init_path(path);
1490         ins->objectid = search_start;
1491         ins->offset = 0;
1492         start_found = 0;
1493         path->reada = 2;
1494
1495         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1496         if (ret < 0)
1497                 goto error;
1498         ret = find_previous_extent(root, path);
1499         if (ret < 0)
1500                 goto error;
1501         l = path->nodes[0];
1502         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1503         while (1) {
1504                 l = path->nodes[0];
1505                 slot = path->slots[0];
1506                 if (slot >= btrfs_header_nritems(l)) {
1507                         ret = btrfs_next_leaf(root, path);
1508                         if (ret == 0)
1509                                 continue;
1510                         if (ret < 0)
1511                                 goto error;
1512
1513                         search_start = max(search_start,
1514                                            block_group->key.objectid);
1515                         if (!start_found) {
1516                                 aligned = stripe_align(root, search_start);
1517                                 ins->objectid = aligned;
1518                                 if (aligned >= search_end) {
1519                                         ret = -ENOSPC;
1520                                         goto error;
1521                                 }
1522                                 ins->offset = search_end - aligned;
1523                                 start_found = 1;
1524                                 goto check_pending;
1525                         }
1526                         ins->objectid = stripe_align(root,
1527                                                      last_byte > search_start ?
1528                                                      last_byte : search_start);
1529                         if (search_end <= ins->objectid) {
1530                                 ret = -ENOSPC;
1531                                 goto error;
1532                         }
1533                         ins->offset = search_end - ins->objectid;
1534                         BUG_ON(ins->objectid >= search_end);
1535                         goto check_pending;
1536                 }
1537                 btrfs_item_key_to_cpu(l, &key, slot);
1538
1539                 if (key.objectid >= search_start && key.objectid > last_byte &&
1540                     start_found) {
1541                         if (last_byte < search_start)
1542                                 last_byte = search_start;
1543                         aligned = stripe_align(root, last_byte);
1544                         hole_size = key.objectid - aligned;
1545                         if (key.objectid > aligned && hole_size >= num_bytes) {
1546                                 ins->objectid = aligned;
1547                                 ins->offset = hole_size;
1548                                 goto check_pending;
1549                         }
1550                 }
1551                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1552                         if (!start_found && btrfs_key_type(&key) ==
1553                             BTRFS_BLOCK_GROUP_ITEM_KEY) {
1554                                 last_byte = key.objectid;
1555                                 start_found = 1;
1556                         }
1557                         goto next;
1558                 }
1559
1560
1561                 start_found = 1;
1562                 last_byte = key.objectid + key.offset;
1563
1564                 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1565                     last_byte >= block_group->key.objectid +
1566                     block_group->key.offset) {
1567                         btrfs_release_path(root, path);
1568                         search_start = block_group->key.objectid +
1569                                 block_group->key.offset;
1570                         goto new_group;
1571                 }
1572 next:
1573                 path->slots[0]++;
1574                 cond_resched();
1575         }
1576 check_pending:
1577         /* we have to make sure we didn't find an extent that has already
1578          * been allocated by the map tree or the original allocation
1579          */
1580         btrfs_release_path(root, path);
1581         BUG_ON(ins->objectid < search_start);
1582
1583         if (ins->objectid + num_bytes >= search_end)
1584                 goto enospc;
1585         if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1586             ins->objectid + num_bytes > block_group->
1587             key.objectid + block_group->key.offset) {
1588                 search_start = block_group->key.objectid +
1589                         block_group->key.offset;
1590                 goto new_group;
1591         }
1592         if (test_range_bit(&info->extent_ins, ins->objectid,
1593                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1594                 search_start = ins->objectid + num_bytes;
1595                 goto new_group;
1596         }
1597         if (test_range_bit(&info->pinned_extents, ins->objectid,
1598                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1599                 search_start = ins->objectid + num_bytes;
1600                 goto new_group;
1601         }
1602         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1603             ins->objectid < exclude_start + exclude_nr)) {
1604                 search_start = exclude_start + exclude_nr;
1605                 goto new_group;
1606         }
1607         if (!data) {
1608                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1609                 if (block_group)
1610                         trans->block_group = block_group;
1611         }
1612         ins->offset = num_bytes;
1613         btrfs_free_path(path);
1614         if (last_ptr) {
1615                 *last_ptr = ins->objectid + ins->offset;
1616                 if (*last_ptr ==
1617                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1618                         *last_ptr = 0;
1619                 }
1620         }
1621         return 0;
1622
1623 new_group:
1624         if (search_start + num_bytes >= search_end) {
1625 enospc:
1626                 search_start = orig_search_start;
1627                 if (full_scan) {
1628                         ret = -ENOSPC;
1629                         goto error;
1630                 }
1631                 if (wrapped) {
1632                         if (!full_scan)
1633                                 total_needed -= empty_size;
1634                         full_scan = 1;
1635                         data = BTRFS_BLOCK_GROUP_MIXED;
1636                 } else
1637                         wrapped = 1;
1638         }
1639         block_group = btrfs_lookup_block_group(info, search_start);
1640         cond_resched();
1641         block_group = btrfs_find_block_group(root, block_group,
1642                                              search_start, data, 0);
1643         goto check_failed;
1644
1645 error:
1646         btrfs_release_path(root, path);
1647         btrfs_free_path(path);
1648         return ret;
1649 }
1650 /*
1651  * finds a free extent and does all the dirty work required for allocation
1652  * returns the key for the extent through ins, and a tree buffer for
1653  * the first block of the extent through buf.
1654  *
1655  * returns 0 if everything worked, non-zero otherwise.
1656  */
1657 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1658                        struct btrfs_root *root,
1659                        u64 num_bytes, u64 root_objectid, u64 ref_generation,
1660                        u64 owner, u64 owner_offset,
1661                        u64 empty_size, u64 hint_byte,
1662                        u64 search_end, struct btrfs_key *ins, int data)
1663 {
1664         int ret;
1665         int pending_ret;
1666         u64 super_used;
1667         u64 root_used;
1668         u64 search_start = 0;
1669         u64 new_hint;
1670         u32 sizes[2];
1671         struct btrfs_fs_info *info = root->fs_info;
1672         struct btrfs_root *extent_root = info->extent_root;
1673         struct btrfs_extent_item *extent_item;
1674         struct btrfs_extent_ref *ref;
1675         struct btrfs_path *path;
1676         struct btrfs_key keys[2];
1677
1678         new_hint = max(hint_byte, root->fs_info->alloc_start);
1679         if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1680                 hint_byte = new_hint;
1681
1682         WARN_ON(num_bytes < root->sectorsize);
1683         ret = find_free_extent(trans, root, num_bytes, empty_size,
1684                                search_start, search_end, hint_byte, ins,
1685                                trans->alloc_exclude_start,
1686                                trans->alloc_exclude_nr, data);
1687         BUG_ON(ret);
1688         if (ret)
1689                 return ret;
1690
1691         /* block accounting for super block */
1692         super_used = btrfs_super_bytes_used(&info->super_copy);
1693         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1694
1695         /* block accounting for root item */
1696         root_used = btrfs_root_used(&root->root_item);
1697         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1698
1699         clear_extent_dirty(&root->fs_info->free_space_cache,
1700                            ins->objectid, ins->objectid + ins->offset - 1,
1701                            GFP_NOFS);
1702
1703         if (root == extent_root) {
1704                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1705                                 ins->objectid + ins->offset - 1,
1706                                 EXTENT_LOCKED, GFP_NOFS);
1707                 WARN_ON(data == 1);
1708                 goto update_block;
1709         }
1710
1711         WARN_ON(trans->alloc_exclude_nr);
1712         trans->alloc_exclude_start = ins->objectid;
1713         trans->alloc_exclude_nr = ins->offset;
1714
1715         memcpy(&keys[0], ins, sizeof(*ins));
1716         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1717                                          owner, owner_offset);
1718         keys[1].objectid = ins->objectid;
1719         keys[1].type = BTRFS_EXTENT_REF_KEY;
1720         sizes[0] = sizeof(*extent_item);
1721         sizes[1] = sizeof(*ref);
1722
1723         path = btrfs_alloc_path();
1724         BUG_ON(!path);
1725
1726         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1727                                        sizes, 2);
1728
1729         BUG_ON(ret);
1730         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1731                                      struct btrfs_extent_item);
1732         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1733         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1734                              struct btrfs_extent_ref);
1735
1736         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1737         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1738         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1739         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1740
1741         btrfs_mark_buffer_dirty(path->nodes[0]);
1742
1743         trans->alloc_exclude_start = 0;
1744         trans->alloc_exclude_nr = 0;
1745         btrfs_free_path(path);
1746         finish_current_insert(trans, extent_root);
1747         pending_ret = del_pending_extents(trans, extent_root);
1748
1749         if (ret) {
1750                 return ret;
1751         }
1752         if (pending_ret) {
1753                 return pending_ret;
1754         }
1755
1756 update_block:
1757         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1758                                  data);
1759         if (ret) {
1760                 printk("update block group failed for %Lu %Lu\n",
1761                        ins->objectid, ins->offset);
1762                 BUG();
1763         }
1764         return 0;
1765 }
1766
1767 /*
1768  * helper function to allocate a block for a given tree
1769  * returns the tree buffer or NULL.
1770  */
1771 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1772                                              struct btrfs_root *root,
1773                                              u32 blocksize,
1774                                              u64 root_objectid, u64 hint,
1775                                              u64 empty_size)
1776 {
1777         u64 ref_generation;
1778
1779         if (root->ref_cows)
1780                 ref_generation = trans->transid;
1781         else
1782                 ref_generation = 0;
1783
1784
1785         return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1786                                         ref_generation, 0, 0, hint, empty_size);
1787 }
1788
1789 /*
1790  * helper function to allocate a block for a given tree
1791  * returns the tree buffer or NULL.
1792  */
1793 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1794                                              struct btrfs_root *root,
1795                                              u32 blocksize,
1796                                              u64 root_objectid,
1797                                              u64 ref_generation,
1798                                              u64 first_objectid,
1799                                              int level,
1800                                              u64 hint,
1801                                              u64 empty_size)
1802 {
1803         struct btrfs_key ins;
1804         int ret;
1805         struct extent_buffer *buf;
1806
1807         ret = btrfs_alloc_extent(trans, root, blocksize,
1808                                  root_objectid, ref_generation,
1809                                  level, first_objectid, empty_size, hint,
1810                                  (u64)-1, &ins, 0);
1811         if (ret) {
1812                 BUG_ON(ret > 0);
1813                 return ERR_PTR(ret);
1814         }
1815         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1816         if (!buf) {
1817                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1818                                   root->root_key.objectid, ref_generation,
1819                                   0, 0, 0);
1820                 return ERR_PTR(-ENOMEM);
1821         }
1822         btrfs_set_header_generation(buf, trans->transid);
1823         clean_tree_block(trans, root, buf);
1824         wait_on_tree_block_writeback(root, buf);
1825         btrfs_set_buffer_uptodate(buf);
1826
1827         if (PageDirty(buf->first_page)) {
1828                 printk("page %lu dirty\n", buf->first_page->index);
1829                 WARN_ON(1);
1830         }
1831
1832         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1833                          buf->start + buf->len - 1, GFP_NOFS);
1834         set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1835                         buf->start, buf->start + buf->len - 1,
1836                         EXTENT_CSUM, GFP_NOFS);
1837         buf->flags |= EXTENT_CSUM;
1838         btrfs_set_buffer_defrag(buf);
1839         trans->blocks_used++;
1840         return buf;
1841 }
1842
1843 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1844                                   struct btrfs_root *root,
1845                                   struct extent_buffer *leaf)
1846 {
1847         u64 leaf_owner;
1848         u64 leaf_generation;
1849         struct btrfs_key key;
1850         struct btrfs_file_extent_item *fi;
1851         int i;
1852         int nritems;
1853         int ret;
1854
1855         BUG_ON(!btrfs_is_leaf(leaf));
1856         nritems = btrfs_header_nritems(leaf);
1857         leaf_owner = btrfs_header_owner(leaf);
1858         leaf_generation = btrfs_header_generation(leaf);
1859
1860         for (i = 0; i < nritems; i++) {
1861                 u64 disk_bytenr;
1862
1863                 btrfs_item_key_to_cpu(leaf, &key, i);
1864                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1865                         continue;
1866                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1867                 if (btrfs_file_extent_type(leaf, fi) ==
1868                     BTRFS_FILE_EXTENT_INLINE)
1869                         continue;
1870                 /*
1871                  * FIXME make sure to insert a trans record that
1872                  * repeats the snapshot del on crash
1873                  */
1874                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1875                 if (disk_bytenr == 0)
1876                         continue;
1877                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1878                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
1879                                 leaf_owner, leaf_generation,
1880                                 key.objectid, key.offset, 0);
1881                 BUG_ON(ret);
1882         }
1883         return 0;
1884 }
1885
1886 static void noinline reada_walk_down(struct btrfs_root *root,
1887                                      struct extent_buffer *node,
1888                                      int slot)
1889 {
1890         u64 bytenr;
1891         u64 last = 0;
1892         u32 nritems;
1893         u32 refs;
1894         u32 blocksize;
1895         int ret;
1896         int i;
1897         int level;
1898         int skipped = 0;
1899
1900         nritems = btrfs_header_nritems(node);
1901         level = btrfs_header_level(node);
1902         if (level)
1903                 return;
1904
1905         for (i = slot; i < nritems && skipped < 32; i++) {
1906                 bytenr = btrfs_node_blockptr(node, i);
1907                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1908                              (last > bytenr && last - bytenr > 32 * 1024))) {
1909                         skipped++;
1910                         continue;
1911                 }
1912                 blocksize = btrfs_level_size(root, level - 1);
1913                 if (i != slot) {
1914                         ret = lookup_extent_ref(NULL, root, bytenr,
1915                                                 blocksize, &refs);
1916                         BUG_ON(ret);
1917                         if (refs != 1) {
1918                                 skipped++;
1919                                 continue;
1920                         }
1921                 }
1922                 mutex_unlock(&root->fs_info->fs_mutex);
1923                 ret = readahead_tree_block(root, bytenr, blocksize);
1924                 last = bytenr + blocksize;
1925                 cond_resched();
1926                 mutex_lock(&root->fs_info->fs_mutex);
1927                 if (ret)
1928                         break;
1929         }
1930 }
1931
1932 /*
1933  * helper function for drop_snapshot, this walks down the tree dropping ref
1934  * counts as it goes.
1935  */
1936 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1937                                    struct btrfs_root *root,
1938                                    struct btrfs_path *path, int *level)
1939 {
1940         u64 root_owner;
1941         u64 root_gen;
1942         u64 bytenr;
1943         struct extent_buffer *next;
1944         struct extent_buffer *cur;
1945         struct extent_buffer *parent;
1946         u32 blocksize;
1947         int ret;
1948         u32 refs;
1949
1950         WARN_ON(*level < 0);
1951         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1952         ret = lookup_extent_ref(trans, root,
1953                                 path->nodes[*level]->start,
1954                                 path->nodes[*level]->len, &refs);
1955         BUG_ON(ret);
1956         if (refs > 1)
1957                 goto out;
1958
1959         /*
1960          * walk down to the last node level and free all the leaves
1961          */
1962         while(*level >= 0) {
1963                 WARN_ON(*level < 0);
1964                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1965                 cur = path->nodes[*level];
1966
1967                 if (btrfs_header_level(cur) != *level)
1968                         WARN_ON(1);
1969
1970                 if (path->slots[*level] >=
1971                     btrfs_header_nritems(cur))
1972                         break;
1973                 if (*level == 0) {
1974                         ret = drop_leaf_ref(trans, root, cur);
1975                         BUG_ON(ret);
1976                         break;
1977                 }
1978                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1979                 blocksize = btrfs_level_size(root, *level - 1);
1980                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1981                 BUG_ON(ret);
1982                 if (refs != 1) {
1983                         parent = path->nodes[*level];
1984                         root_owner = btrfs_header_owner(parent);
1985                         root_gen = btrfs_header_generation(parent);
1986                         path->slots[*level]++;
1987                         ret = btrfs_free_extent(trans, root, bytenr,
1988                                                 blocksize, root_owner,
1989                                                 root_gen, 0, 0, 1);
1990                         BUG_ON(ret);
1991                         continue;
1992                 }
1993                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1994                 if (!next || !btrfs_buffer_uptodate(next)) {
1995                         free_extent_buffer(next);
1996                         reada_walk_down(root, cur, path->slots[*level]);
1997                         mutex_unlock(&root->fs_info->fs_mutex);
1998                         next = read_tree_block(root, bytenr, blocksize);
1999                         mutex_lock(&root->fs_info->fs_mutex);
2000
2001                         /* we dropped the lock, check one more time */
2002                         ret = lookup_extent_ref(trans, root, bytenr,
2003                                                 blocksize, &refs);
2004                         BUG_ON(ret);
2005                         if (refs != 1) {
2006                                 parent = path->nodes[*level];
2007                                 root_owner = btrfs_header_owner(parent);
2008                                 root_gen = btrfs_header_generation(parent);
2009
2010                                 path->slots[*level]++;
2011                                 free_extent_buffer(next);
2012                                 ret = btrfs_free_extent(trans, root, bytenr,
2013                                                         blocksize,
2014                                                         root_owner,
2015                                                         root_gen, 0, 0, 1);
2016                                 BUG_ON(ret);
2017                                 continue;
2018                         }
2019                 }
2020                 WARN_ON(*level <= 0);
2021                 if (path->nodes[*level-1])
2022                         free_extent_buffer(path->nodes[*level-1]);
2023                 path->nodes[*level-1] = next;
2024                 *level = btrfs_header_level(next);
2025                 path->slots[*level] = 0;
2026         }
2027 out:
2028         WARN_ON(*level < 0);
2029         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2030
2031         if (path->nodes[*level] == root->node) {
2032                 root_owner = root->root_key.objectid;
2033                 parent = path->nodes[*level];
2034         } else {
2035                 parent = path->nodes[*level + 1];
2036                 root_owner = btrfs_header_owner(parent);
2037         }
2038
2039         root_gen = btrfs_header_generation(parent);
2040         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2041                                 path->nodes[*level]->len,
2042                                 root_owner, root_gen, 0, 0, 1);
2043         free_extent_buffer(path->nodes[*level]);
2044         path->nodes[*level] = NULL;
2045         *level += 1;
2046         BUG_ON(ret);
2047         return 0;
2048 }
2049
2050 /*
2051  * helper for dropping snapshots.  This walks back up the tree in the path
2052  * to find the first node higher up where we haven't yet gone through
2053  * all the slots
2054  */
2055 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2056                                  struct btrfs_root *root,
2057                                  struct btrfs_path *path, int *level)
2058 {
2059         u64 root_owner;
2060         u64 root_gen;
2061         struct btrfs_root_item *root_item = &root->root_item;
2062         int i;
2063         int slot;
2064         int ret;
2065
2066         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2067                 slot = path->slots[i];
2068                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2069                         struct extent_buffer *node;
2070                         struct btrfs_disk_key disk_key;
2071                         node = path->nodes[i];
2072                         path->slots[i]++;
2073                         *level = i;
2074                         WARN_ON(*level == 0);
2075                         btrfs_node_key(node, &disk_key, path->slots[i]);
2076                         memcpy(&root_item->drop_progress,
2077                                &disk_key, sizeof(disk_key));
2078                         root_item->drop_level = i;
2079                         return 0;
2080                 } else {
2081                         if (path->nodes[*level] == root->node) {
2082                                 root_owner = root->root_key.objectid;
2083                                 root_gen =
2084                                    btrfs_header_generation(path->nodes[*level]);
2085                         } else {
2086                                 struct extent_buffer *node;
2087                                 node = path->nodes[*level + 1];
2088                                 root_owner = btrfs_header_owner(node);
2089                                 root_gen = btrfs_header_generation(node);
2090                         }
2091                         ret = btrfs_free_extent(trans, root,
2092                                                 path->nodes[*level]->start,
2093                                                 path->nodes[*level]->len,
2094                                                 root_owner, root_gen, 0, 0, 1);
2095                         BUG_ON(ret);
2096                         free_extent_buffer(path->nodes[*level]);
2097                         path->nodes[*level] = NULL;
2098                         *level = i + 1;
2099                 }
2100         }
2101         return 1;
2102 }
2103
2104 /*
2105  * drop the reference count on the tree rooted at 'snap'.  This traverses
2106  * the tree freeing any blocks that have a ref count of zero after being
2107  * decremented.
2108  */
2109 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2110                         *root)
2111 {
2112         int ret = 0;
2113         int wret;
2114         int level;
2115         struct btrfs_path *path;
2116         int i;
2117         int orig_level;
2118         struct btrfs_root_item *root_item = &root->root_item;
2119
2120         path = btrfs_alloc_path();
2121         BUG_ON(!path);
2122
2123         level = btrfs_header_level(root->node);
2124         orig_level = level;
2125         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2126                 path->nodes[level] = root->node;
2127                 extent_buffer_get(root->node);
2128                 path->slots[level] = 0;
2129         } else {
2130                 struct btrfs_key key;
2131                 struct btrfs_disk_key found_key;
2132                 struct extent_buffer *node;
2133
2134                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2135                 level = root_item->drop_level;
2136                 path->lowest_level = level;
2137                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2138                 if (wret < 0) {
2139                         ret = wret;
2140                         goto out;
2141                 }
2142                 node = path->nodes[level];
2143                 btrfs_node_key(node, &found_key, path->slots[level]);
2144                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2145                                sizeof(found_key)));
2146         }
2147         while(1) {
2148                 wret = walk_down_tree(trans, root, path, &level);
2149                 if (wret > 0)
2150                         break;
2151                 if (wret < 0)
2152                         ret = wret;
2153
2154                 wret = walk_up_tree(trans, root, path, &level);
2155                 if (wret > 0)
2156                         break;
2157                 if (wret < 0)
2158                         ret = wret;
2159                 ret = -EAGAIN;
2160                 break;
2161         }
2162         for (i = 0; i <= orig_level; i++) {
2163                 if (path->nodes[i]) {
2164                         free_extent_buffer(path->nodes[i]);
2165                         path->nodes[i] = NULL;
2166                 }
2167         }
2168 out:
2169         btrfs_free_path(path);
2170         return ret;
2171 }
2172
2173 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2174 {
2175         u64 start;
2176         u64 end;
2177         u64 ptr;
2178         int ret;
2179         while(1) {
2180                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2181                                             &start, &end, (unsigned int)-1);
2182                 if (ret)
2183                         break;
2184                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2185                 if (!ret)
2186                         kfree((void *)(unsigned long)ptr);
2187                 clear_extent_bits(&info->block_group_cache, start,
2188                                   end, (unsigned int)-1, GFP_NOFS);
2189         }
2190         while(1) {
2191                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2192                                             &start, &end, EXTENT_DIRTY);
2193                 if (ret)
2194                         break;
2195                 clear_extent_dirty(&info->free_space_cache, start,
2196                                    end, GFP_NOFS);
2197         }
2198         return 0;
2199 }
2200
2201 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2202                                          u64 len)
2203 {
2204         u64 page_start;
2205         u64 page_end;
2206         u64 delalloc_start;
2207         u64 existing_delalloc;
2208         unsigned long last_index;
2209         unsigned long i;
2210         struct page *page;
2211         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2212         struct file_ra_state *ra;
2213
2214         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2215
2216         mutex_lock(&inode->i_mutex);
2217         i = start >> PAGE_CACHE_SHIFT;
2218         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2219
2220         file_ra_state_init(ra, inode->i_mapping);
2221         btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2222         kfree(ra);
2223
2224         for (; i <= last_index; i++) {
2225                 page = grab_cache_page(inode->i_mapping, i);
2226                 if (!page)
2227                         goto out_unlock;
2228                 if (!PageUptodate(page)) {
2229                         btrfs_readpage(NULL, page);
2230                         lock_page(page);
2231                         if (!PageUptodate(page)) {
2232                                 unlock_page(page);
2233                                 page_cache_release(page);
2234                                 goto out_unlock;
2235                         }
2236                 }
2237                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2238                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2239
2240                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2241
2242                 delalloc_start = page_start;
2243                 existing_delalloc = count_range_bits(io_tree,
2244                                              &delalloc_start, page_end,
2245                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2246
2247                 set_extent_delalloc(io_tree, page_start,
2248                                     page_end, GFP_NOFS);
2249
2250                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2251                 set_page_dirty(page);
2252                 unlock_page(page);
2253                 page_cache_release(page);
2254         }
2255
2256 out_unlock:
2257         mutex_unlock(&inode->i_mutex);
2258         return 0;
2259 }
2260
2261 /*
2262  * note, this releases the path
2263  */
2264 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2265                                   struct btrfs_path *path,
2266                                   struct btrfs_key *extent_key)
2267 {
2268         struct inode *inode;
2269         struct btrfs_root *found_root;
2270         struct btrfs_key *root_location;
2271         struct btrfs_extent_ref *ref;
2272         u64 ref_root;
2273         u64 ref_gen;
2274         u64 ref_objectid;
2275         u64 ref_offset;
2276         int ret;
2277
2278         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2279                              struct btrfs_extent_ref);
2280         ref_root = btrfs_ref_root(path->nodes[0], ref);
2281         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2282         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2283         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2284         btrfs_release_path(extent_root, path);
2285
2286         root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2287         root_location->objectid = ref_root;
2288         if (ref_gen == 0)
2289                 root_location->offset = 0;
2290         else
2291                 root_location->offset = (u64)-1;
2292         root_location->type = BTRFS_ROOT_ITEM_KEY;
2293
2294         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2295                                                 root_location);
2296         BUG_ON(!found_root);
2297         kfree(root_location);
2298
2299         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2300                 mutex_unlock(&extent_root->fs_info->fs_mutex);
2301                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2302                                           ref_objectid, found_root);
2303                 if (inode->i_state & I_NEW) {
2304                         /* the inode and parent dir are two different roots */
2305                         BTRFS_I(inode)->root = found_root;
2306                         BTRFS_I(inode)->location.objectid = ref_objectid;
2307                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2308                         BTRFS_I(inode)->location.offset = 0;
2309                         btrfs_read_locked_inode(inode);
2310                         unlock_new_inode(inode);
2311
2312                 }
2313                 /* this can happen if the reference is not against
2314                  * the latest version of the tree root
2315                  */
2316                 if (is_bad_inode(inode)) {
2317                         mutex_lock(&extent_root->fs_info->fs_mutex);
2318                         goto out;
2319                 }
2320                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2321                 /* FIXME, data=ordered will help get rid of this */
2322                 filemap_fdatawrite(inode->i_mapping);
2323                 iput(inode);
2324                 mutex_lock(&extent_root->fs_info->fs_mutex);
2325         } else {
2326                 struct btrfs_trans_handle *trans;
2327                 struct btrfs_key found_key;
2328                 struct extent_buffer *eb;
2329                 int level;
2330                 int i;
2331
2332                 trans = btrfs_start_transaction(found_root, 1);
2333                 eb = read_tree_block(found_root, extent_key->objectid,
2334                                      extent_key->offset);
2335                 level = btrfs_header_level(eb);
2336
2337                 if (level == 0)
2338                         btrfs_item_key_to_cpu(eb, &found_key, 0);
2339                 else
2340                         btrfs_node_key_to_cpu(eb, &found_key, 0);
2341
2342                 free_extent_buffer(eb);
2343
2344                 path->lowest_level = level;
2345                 path->reada = 2;
2346                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2347                                         0, 1);
2348                 path->lowest_level = 0;
2349                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2350                         if (!path->nodes[i])
2351                                 break;
2352                         free_extent_buffer(path->nodes[i]);
2353                         path->nodes[i] = NULL;
2354                 }
2355                 btrfs_release_path(found_root, path);
2356                 btrfs_end_transaction(trans, found_root);
2357         }
2358
2359 out:
2360         return 0;
2361 }
2362
2363 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2364                                         struct btrfs_path *path,
2365                                         struct btrfs_key *extent_key)
2366 {
2367         struct btrfs_key key;
2368         struct btrfs_key found_key;
2369         struct extent_buffer *leaf;
2370         u32 nritems;
2371         u32 item_size;
2372         int ret = 0;
2373
2374         key.objectid = extent_key->objectid;
2375         key.type = BTRFS_EXTENT_REF_KEY;
2376         key.offset = 0;
2377
2378         while(1) {
2379                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2380
2381                 if (ret < 0)
2382                         goto out;
2383
2384                 ret = 0;
2385                 leaf = path->nodes[0];
2386                 nritems = btrfs_header_nritems(leaf);
2387                 if (path->slots[0] == nritems)
2388                         goto out;
2389
2390                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2391                 if (found_key.objectid != extent_key->objectid)
2392                         break;
2393
2394                 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2395                         break;
2396
2397                 key.offset = found_key.offset + 1;
2398                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2399
2400                 ret = relocate_one_reference(extent_root, path, extent_key);
2401                 if (ret)
2402                         goto out;
2403         }
2404         ret = 0;
2405 out:
2406         btrfs_release_path(extent_root, path);
2407         return ret;
2408 }
2409
2410 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2411 {
2412         struct btrfs_trans_handle *trans;
2413         struct btrfs_root *tree_root = root->fs_info->tree_root;
2414         struct btrfs_path *path;
2415         u64 cur_byte;
2416         u64 total_found;
2417         struct btrfs_fs_info *info = root->fs_info;
2418         struct extent_io_tree *block_group_cache;
2419         struct btrfs_key key;
2420         struct btrfs_key found_key;
2421         struct extent_buffer *leaf;
2422         u32 nritems;
2423         int ret;
2424         int progress = 0;
2425
2426         btrfs_set_super_total_bytes(&info->super_copy, new_size);
2427         clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2428                            GFP_NOFS);
2429         block_group_cache = &info->block_group_cache;
2430         path = btrfs_alloc_path();
2431         root = root->fs_info->extent_root;
2432         path->reada = 2;
2433
2434 again:
2435         total_found = 0;
2436         key.objectid = new_size;
2437         key.offset = 0;
2438         key.type = 0;
2439         cur_byte = key.objectid;
2440
2441         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2442         if (ret < 0)
2443                 goto out;
2444
2445         ret = find_previous_extent(root, path);
2446         if (ret < 0)
2447                 goto out;
2448         if (ret == 0) {
2449                 leaf = path->nodes[0];
2450                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2451                 if (found_key.objectid + found_key.offset > new_size) {
2452                         cur_byte = found_key.objectid;
2453                         key.objectid = cur_byte;
2454                 }
2455         }
2456         btrfs_release_path(root, path);
2457
2458         while(1) {
2459                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2460                 if (ret < 0)
2461                         goto out;
2462
2463                 leaf = path->nodes[0];
2464                 nritems = btrfs_header_nritems(leaf);
2465 next:
2466                 if (path->slots[0] >= nritems) {
2467                         ret = btrfs_next_leaf(root, path);
2468                         if (ret < 0)
2469                                 goto out;
2470                         if (ret == 1) {
2471                                 ret = 0;
2472                                 break;
2473                         }
2474                         leaf = path->nodes[0];
2475                         nritems = btrfs_header_nritems(leaf);
2476                 }
2477
2478                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2479
2480                 if (progress && need_resched()) {
2481                         memcpy(&key, &found_key, sizeof(key));
2482                         mutex_unlock(&root->fs_info->fs_mutex);
2483                         cond_resched();
2484                         mutex_lock(&root->fs_info->fs_mutex);
2485                         btrfs_release_path(root, path);
2486                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
2487                         progress = 0;
2488                         goto next;
2489                 }
2490                 progress = 1;
2491
2492                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2493                     found_key.objectid + found_key.offset <= cur_byte) {
2494                         path->slots[0]++;
2495                         goto next;
2496                 }
2497
2498                 total_found++;
2499                 cur_byte = found_key.objectid + found_key.offset;
2500                 key.objectid = cur_byte;
2501                 btrfs_release_path(root, path);
2502                 ret = relocate_one_extent(root, path, &found_key);
2503         }
2504
2505         btrfs_release_path(root, path);
2506
2507         if (total_found > 0) {
2508                 trans = btrfs_start_transaction(tree_root, 1);
2509                 btrfs_commit_transaction(trans, tree_root);
2510
2511                 mutex_unlock(&root->fs_info->fs_mutex);
2512                 btrfs_clean_old_snapshots(tree_root);
2513                 mutex_lock(&root->fs_info->fs_mutex);
2514
2515                 trans = btrfs_start_transaction(tree_root, 1);
2516                 btrfs_commit_transaction(trans, tree_root);
2517                 goto again;
2518         }
2519
2520         trans = btrfs_start_transaction(root, 1);
2521         key.objectid = new_size;
2522         key.offset = 0;
2523         key.type = 0;
2524         while(1) {
2525                 u64 ptr;
2526
2527                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2528                 if (ret < 0)
2529                         goto out;
2530
2531                 leaf = path->nodes[0];
2532                 nritems = btrfs_header_nritems(leaf);
2533 bg_next:
2534                 if (path->slots[0] >= nritems) {
2535                         ret = btrfs_next_leaf(root, path);
2536                         if (ret < 0)
2537                                 break;
2538                         if (ret == 1) {
2539                                 ret = 0;
2540                                 break;
2541                         }
2542                         leaf = path->nodes[0];
2543                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2544
2545                         /*
2546                          * btrfs_next_leaf doesn't cow buffers, we have to
2547                          * do the search again
2548                          */
2549                         memcpy(&key, &found_key, sizeof(key));
2550                         btrfs_release_path(root, path);
2551                         goto resched_check;
2552                 }
2553
2554                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2555                 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2556                         printk("shrinker found key %Lu %u %Lu\n",
2557                                 found_key.objectid, found_key.type,
2558                                 found_key.offset);
2559                         path->slots[0]++;
2560                         goto bg_next;
2561                 }
2562                 ret = get_state_private(&info->block_group_cache,
2563                                         found_key.objectid, &ptr);
2564                 if (!ret)
2565                         kfree((void *)(unsigned long)ptr);
2566
2567                 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2568                                   found_key.objectid + found_key.offset - 1,
2569                                   (unsigned int)-1, GFP_NOFS);
2570
2571                 key.objectid = found_key.objectid + 1;
2572                 btrfs_del_item(trans, root, path);
2573                 btrfs_release_path(root, path);
2574 resched_check:
2575                 if (need_resched()) {
2576                         mutex_unlock(&root->fs_info->fs_mutex);
2577                         cond_resched();
2578                         mutex_lock(&root->fs_info->fs_mutex);
2579                 }
2580         }
2581         clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2582                            GFP_NOFS);
2583         btrfs_commit_transaction(trans, root);
2584 out:
2585         btrfs_free_path(path);
2586         return ret;
2587 }
2588
2589 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2590                            struct btrfs_root *root, u64 new_size)
2591 {
2592         struct btrfs_path *path;
2593         u64 nr = 0;
2594         u64 cur_byte;
2595         u64 old_size;
2596         unsigned long rem;
2597         struct btrfs_block_group_cache *cache;
2598         struct btrfs_block_group_item *item;
2599         struct btrfs_fs_info *info = root->fs_info;
2600         struct extent_io_tree *block_group_cache;
2601         struct btrfs_key key;
2602         struct extent_buffer *leaf;
2603         int ret;
2604         int bit;
2605
2606         old_size = btrfs_super_total_bytes(&info->super_copy);
2607         block_group_cache = &info->block_group_cache;
2608
2609         root = info->extent_root;
2610
2611         cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2612
2613         cur_byte = cache->key.objectid + cache->key.offset;
2614         if (cur_byte >= new_size)
2615                 goto set_size;
2616
2617         key.offset = BTRFS_BLOCK_GROUP_SIZE;
2618         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2619
2620         path = btrfs_alloc_path();
2621         if (!path)
2622                 return -ENOMEM;
2623
2624         while(cur_byte < new_size) {
2625                 key.objectid = cur_byte;
2626                 ret = btrfs_insert_empty_item(trans, root, path, &key,
2627                                         sizeof(struct btrfs_block_group_item));
2628                 BUG_ON(ret);
2629                 leaf = path->nodes[0];
2630                 item = btrfs_item_ptr(leaf, path->slots[0],
2631                                       struct btrfs_block_group_item);
2632
2633                 btrfs_set_disk_block_group_used(leaf, item, 0);
2634                 div_long_long_rem(nr, 3, &rem);
2635                 if (rem) {
2636                         btrfs_set_disk_block_group_flags(leaf, item,
2637                                                  BTRFS_BLOCK_GROUP_DATA);
2638                 } else {
2639                         btrfs_set_disk_block_group_flags(leaf, item, 0);
2640                 }
2641                 nr++;
2642
2643                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2644                 BUG_ON(!cache);
2645
2646                 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2647                                    sizeof(cache->item));
2648
2649                 memcpy(&cache->key, &key, sizeof(key));
2650                 cache->cached = 0;
2651                 cache->pinned = 0;
2652                 cur_byte = key.objectid + key.offset;
2653                 btrfs_release_path(root, path);
2654
2655                 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2656                         bit = BLOCK_GROUP_DATA;
2657                         cache->data = BTRFS_BLOCK_GROUP_DATA;
2658                 } else {
2659                         bit = BLOCK_GROUP_METADATA;
2660                         cache->data = 0;
2661                 }
2662
2663                 /* use EXTENT_LOCKED to prevent merging */
2664                 set_extent_bits(block_group_cache, key.objectid,
2665                                 key.objectid + key.offset - 1,
2666                                 bit | EXTENT_LOCKED, GFP_NOFS);
2667                 set_state_private(block_group_cache, key.objectid,
2668                                   (unsigned long)cache);
2669         }
2670         btrfs_free_path(path);
2671 set_size:
2672         btrfs_set_super_total_bytes(&info->super_copy, new_size);
2673         return 0;
2674 }
2675
2676 int btrfs_read_block_groups(struct btrfs_root *root)
2677 {
2678         struct btrfs_path *path;
2679         int ret;
2680         int err = 0;
2681         int bit;
2682         struct btrfs_block_group_cache *cache;
2683         struct btrfs_fs_info *info = root->fs_info;
2684         struct extent_io_tree *block_group_cache;
2685         struct btrfs_key key;
2686         struct btrfs_key found_key;
2687         struct extent_buffer *leaf;
2688
2689         block_group_cache = &info->block_group_cache;
2690
2691         root = info->extent_root;
2692         key.objectid = 0;
2693         key.offset = BTRFS_BLOCK_GROUP_SIZE;
2694         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2695
2696         path = btrfs_alloc_path();
2697         if (!path)
2698                 return -ENOMEM;
2699
2700         while(1) {
2701                 ret = btrfs_search_slot(NULL, info->extent_root,
2702                                         &key, path, 0, 0);
2703                 if (ret != 0) {
2704                         err = ret;
2705                         break;
2706                 }
2707                 leaf = path->nodes[0];
2708                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2709                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2710                 if (!cache) {
2711                         err = -1;
2712                         break;
2713                 }
2714
2715                 read_extent_buffer(leaf, &cache->item,
2716                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
2717                                    sizeof(cache->item));
2718                 memcpy(&cache->key, &found_key, sizeof(found_key));
2719                 cache->cached = 0;
2720                 cache->pinned = 0;
2721                 key.objectid = found_key.objectid + found_key.offset;
2722                 btrfs_release_path(root, path);
2723
2724                 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2725                         bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2726                         cache->data = BTRFS_BLOCK_GROUP_MIXED;
2727                 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2728                         bit = BLOCK_GROUP_DATA;
2729                         cache->data = BTRFS_BLOCK_GROUP_DATA;
2730                 } else {
2731                         bit = BLOCK_GROUP_METADATA;
2732                         cache->data = 0;
2733                 }
2734
2735                 /* use EXTENT_LOCKED to prevent merging */
2736                 set_extent_bits(block_group_cache, found_key.objectid,
2737                                 found_key.objectid + found_key.offset - 1,
2738                                 bit | EXTENT_LOCKED, GFP_NOFS);
2739                 set_state_private(block_group_cache, found_key.objectid,
2740                                   (unsigned long)cache);
2741
2742                 if (key.objectid >=
2743                     btrfs_super_total_bytes(&info->super_copy))
2744                         break;
2745         }
2746
2747         btrfs_free_path(path);
2748         return 0;
2749 }