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 "ctree.h"
  21 #include "disk-io.h"
  22 #include "print-tree.h"
  23 #include "transaction.h"
  24
  25 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
  26 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  27 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  28
  29 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  30                                  btrfs_root *extent_root);
  31 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  32                                btrfs_root *extent_root);
  33
  34 static int cache_block_group(struct btrfs_root *root,
  35                              struct btrfs_block_group_cache *block_group)
  36 {
  37         struct btrfs_path *path;
  38         int ret;
  39         struct btrfs_key key;
  40         struct extent_buffer *leaf;
  41         struct extent_map_tree *free_space_cache;
  42         int slot;
  43         u64 last = 0;
  44         u64 hole_size;
  45         u64 first_free;
  46         int found = 0;
  47
  48         root = root->fs_info->extent_root;
  49         free_space_cache = &root->fs_info->free_space_cache;
  50
  51         if (block_group->cached)
  52                 return 0;
  53
  54         path = btrfs_alloc_path();
  55         if (!path)
  56                 return -ENOMEM;
  57
  58         path->reada = 2;
  59         first_free = block_group->key.objectid;
  60         key.objectid = block_group->key.objectid;
  61         key.offset = 0;
  62
  63         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  64         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  65
  66         if (ret < 0)
  67                 return ret;
  68
  69         if (ret && path->slots[0] > 0)
  70                 path->slots[0]--;
  71
  72         while(1) {
  73                 leaf = path->nodes[0];
  74                 slot = path->slots[0];
  75                 if (slot >= btrfs_header_nritems(leaf)) {
  76                         ret = btrfs_next_leaf(root, path);
  77                         if (ret < 0)
  78                                 goto err;
  79                         if (ret == 0) {
  80                                 continue;
  81                         } else {
  82                                 break;
  83                         }
  84                 }
  85
  86                 btrfs_item_key_to_cpu(leaf, &key, slot);
  87                 if (key.objectid < block_group->key.objectid) {
  88                         if (key.objectid + key.offset > first_free)
  89                                 first_free = key.objectid + key.offset;
  90                         goto next;
  91                 }
  92
  93                 if (key.objectid >= block_group->key.objectid +
  94                     block_group->key.offset) {
  95                         break;
  96                 }
  97
  98                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
  99                         if (!found) {
 100                                 last = first_free;
 101                                 found = 1;
 102                         }
 103                         if (key.objectid > last) {
 104                                 hole_size = key.objectid - last;
 105                                 set_extent_dirty(free_space_cache, last,
 106                                                  last + hole_size - 1,
 107                                                  GFP_NOFS);
 108                         }
 109                         last = key.objectid + key.offset;
 110                 }
 111 next:
 112                 path->slots[0]++;
 113         }
 114
 115         if (!found)
 116                 last = first_free;
 117         if (block_group->key.objectid +
 118             block_group->key.offset > last) {
 119                 hole_size = block_group->key.objectid +
 120                         block_group->key.offset - last;
 121                 set_extent_dirty(free_space_cache, last,
 122                                  last + hole_size - 1, GFP_NOFS);
 123         }
 124         block_group->cached = 1;
 125 err:
 126         btrfs_free_path(path);
 127         return 0;
 128 }
 129
 130 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 131                                                          btrfs_fs_info *info,
 132                                                          u64 blocknr)
 133 {
 134         struct extent_map_tree *block_group_cache;
 135         struct btrfs_block_group_cache *block_group = NULL;
 136         u64 ptr;
 137         u64 start;
 138         u64 end;
 139         int ret;
 140
 141         block_group_cache = &info->block_group_cache;
 142         ret = find_first_extent_bit(block_group_cache,
 143                                     blocknr, &start, &end,
 144                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
 145         if (ret) {
 146                 return NULL;
 147         }
 148         ret = get_state_private(block_group_cache, start, &ptr);
 149         if (ret)
 150                 return NULL;
 151
 152         block_group = (struct btrfs_block_group_cache *)ptr;
 153
 154
 155         if (block_group->key.objectid <= blocknr && blocknr <=
 156             block_group->key.objectid + block_group->key.offset)
 157                 return block_group;
 158
 159         return NULL;
 160 }
 161
 162 static u64 find_search_start(struct btrfs_root *root,
 163                              struct btrfs_block_group_cache **cache_ret,
 164                              u64 search_start, int num, int data)
 165 {
 166         int ret;
 167         struct btrfs_block_group_cache *cache = *cache_ret;
 168         u64 last = max(search_start, cache->key.objectid);
 169         u64 start = 0;
 170         u64 end = 0;
 171
 172 again:
 173         ret = cache_block_group(root, cache);
 174         if (ret)
 175                 goto out;
 176         while(1) {
 177                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 178                                             last, &start, &end, EXTENT_DIRTY);
 179                 if (ret)
 180                         goto out;
 181
 182                 start = max(last, start);
 183                 last = end + 1;
 184                 if (end + 1 - start < num)
 185                         continue;
 186                 if (start + num >= cache->key.objectid + cache->key.offset)
 187                         goto new_group;
 188                 return start;
 189         }
 190 out:
 191         return search_start;
 192
 193 new_group:
 194         cache = btrfs_lookup_block_group(root->fs_info,
 195                                          last + cache->key.offset - 1);
 196         if (!cache) {
 197                 return search_start;
 198         }
 199         cache = btrfs_find_block_group(root, cache,
 200                                        last + cache->key.offset - 1, data, 0);
 201         *cache_ret = cache;
 202         last = min(cache->key.objectid, last);
 203         goto again;
 204 }
 205
 206 static u64 div_factor(u64 num, int factor)
 207 {
 208         num *= factor;
 209         do_div(num, 10);
 210         return num;
 211 }
 212
 213 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 214                                                  struct btrfs_block_group_cache
 215                                                  *hint, u64 search_start,
 216                                                  int data, int owner)
 217 {
 218         struct btrfs_block_group_cache *cache;
 219         struct extent_map_tree *block_group_cache;
 220         struct btrfs_block_group_cache *found_group = NULL;
 221         struct btrfs_fs_info *info = root->fs_info;
 222         u64 used;
 223         u64 last = 0;
 224         u64 hint_last;
 225         u64 start;
 226         u64 end;
 227         u64 free_check;
 228         u64 ptr;
 229         int bit;
 230         int ret;
 231         int full_search = 0;
 232         int factor = 8;
 233         int data_swap = 0;
 234
 235         block_group_cache = &info->block_group_cache;
 236
 237         if (!owner)
 238                 factor = 5;
 239
 240         if (data)
 241                 bit = BLOCK_GROUP_DATA;
 242         else
 243                 bit = BLOCK_GROUP_METADATA;
 244
 245         if (search_start) {
 246                 struct btrfs_block_group_cache *shint;
 247                 shint = btrfs_lookup_block_group(info, search_start);
 248                 if (shint && shint->data == data) {
 249                         used = btrfs_block_group_used(&shint->item);
 250                         if (used < div_factor(shint->key.offset, factor)) {
 251                                 return shint;
 252                         }
 253                 }
 254         }
 255         if (hint && hint->data == data) {
 256                 used = btrfs_block_group_used(&hint->item);
 257                 if (used < div_factor(hint->key.offset, factor)) {
 258                         return hint;
 259                 }
 260                 last = hint->key.offset * 3;
 261                 if (hint->key.objectid >= last)
 262                         last = max(search_start + hint->key.offset - 1,
 263                                    hint->key.objectid - last);
 264                 else
 265                         last = hint->key.objectid + hint->key.offset;
 266                 hint_last = last;
 267         } else {
 268                 if (hint)
 269                         hint_last = max(hint->key.objectid, search_start);
 270                 else
 271                         hint_last = search_start;
 272
 273                 last = hint_last;
 274         }
 275 again:
 276         while(1) {
 277                 ret = find_first_extent_bit(block_group_cache, last,
 278                                             &start, &end, bit);
 279                 if (ret)
 280                         break;
 281
 282                 ret = get_state_private(block_group_cache, start, &ptr);
 283                 if (ret)
 284                         break;
 285
 286                 cache = (struct btrfs_block_group_cache *)ptr;
 287                 last = cache->key.objectid + cache->key.offset;
 288                 used = btrfs_block_group_used(&cache->item);
 289
 290                 if (full_search)
 291                         free_check = cache->key.offset;
 292                 else
 293                         free_check = div_factor(cache->key.offset, factor);
 294
 295                 if (used < free_check) {
 296                         found_group = cache;
 297                         goto found;
 298                 }
 299                 cond_resched();
 300         }
 301         if (!full_search) {
 302                 last = search_start;
 303                 full_search = 1;
 304                 goto again;
 305         }
 306         if (!data_swap) {
 307                 data_swap = 1;
 308                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 309                 last = search_start;
 310                 goto again;
 311         }
 312 found:
 313         return found_group;
 314 }
 315
 316 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 317                                 struct btrfs_root *root,
 318                                 u64 blocknr, u64 num_blocks)
 319 {
 320         struct btrfs_path *path;
 321         int ret;
 322         struct btrfs_key key;
 323         struct extent_buffer *l;
 324         struct btrfs_extent_item *item;
 325         u32 refs;
 326
 327         path = btrfs_alloc_path();
 328         if (!path)
 329                 return -ENOMEM;
 330
 331         key.objectid = blocknr;
 332         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 333         key.offset = num_blocks;
 334         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 335                                 0, 1);
 336         if (ret < 0)
 337                 return ret;
 338         if (ret != 0) {
 339                 BUG();
 340         }
 341         BUG_ON(ret != 0);
 342         l = path->nodes[0];
 343         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 344         refs = btrfs_extent_refs(l, item);
 345         btrfs_set_extent_refs(l, item, refs + 1);
 346         btrfs_mark_buffer_dirty(path->nodes[0]);
 347
 348         btrfs_release_path(root->fs_info->extent_root, path);
 349         btrfs_free_path(path);
 350         finish_current_insert(trans, root->fs_info->extent_root);
 351         del_pending_extents(trans, root->fs_info->extent_root);
 352         return 0;
 353 }
 354
 355 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 356                          struct btrfs_root *root)
 357 {
 358         finish_current_insert(trans, root->fs_info->extent_root);
 359         del_pending_extents(trans, root->fs_info->extent_root);
 360         return 0;
 361 }
 362
 363 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 364                              struct btrfs_root *root, u64 blocknr,
 365                              u64 num_blocks, u32 *refs)
 366 {
 367         struct btrfs_path *path;
 368         int ret;
 369         struct btrfs_key key;
 370         struct extent_buffer *l;
 371         struct btrfs_extent_item *item;
 372
 373         path = btrfs_alloc_path();
 374         key.objectid = blocknr;
 375         key.offset = num_blocks;
 376         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 377         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 378                                 0, 0);
 379         if (ret < 0)
 380                 goto out;
 381         if (ret != 0) {
 382                 btrfs_print_leaf(root, path->nodes[0]);
 383                 printk("failed to find block number %Lu\n", blocknr);
 384                 BUG();
 385         }
 386         l = path->nodes[0];
 387         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 388         *refs = btrfs_extent_refs(l, item);
 389 out:
 390         btrfs_free_path(path);
 391         return 0;
 392 }
 393
 394 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 395                        struct btrfs_root *root)
 396 {
 397         return btrfs_inc_extent_ref(trans, root,
 398                                     extent_buffer_blocknr(root->node), 1);
 399 }
 400
 401 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 402                   struct extent_buffer *buf)
 403 {
 404         u64 blocknr;
 405         u32 nritems;
 406         struct btrfs_key key;
 407         struct btrfs_file_extent_item *fi;
 408         int i;
 409         int leaf;
 410         int ret;
 411         int faili;
 412         int err;
 413
 414         if (!root->ref_cows)
 415                 return 0;
 416
 417         leaf = btrfs_is_leaf(buf);
 418         nritems = btrfs_header_nritems(buf);
 419         for (i = 0; i < nritems; i++) {
 420                 if (leaf) {
 421                         u64 disk_blocknr;
 422                         btrfs_item_key_to_cpu(buf, &key, i);
 423                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 424                                 continue;
 425                         fi = btrfs_item_ptr(buf, i,
 426                                             struct btrfs_file_extent_item);
 427                         if (btrfs_file_extent_type(buf, fi) ==
 428                             BTRFS_FILE_EXTENT_INLINE)
 429                                 continue;
 430                         disk_blocknr = btrfs_file_extent_disk_blocknr(buf, fi);
 431                         if (disk_blocknr == 0)
 432                                 continue;
 433                         ret = btrfs_inc_extent_ref(trans, root, disk_blocknr,
 434                                     btrfs_file_extent_disk_num_blocks(buf, fi));
 435                         if (ret) {
 436                                 faili = i;
 437                                 goto fail;
 438                         }
 439                 } else {
 440                         blocknr = btrfs_node_blockptr(buf, i);
 441                         ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
 442                         if (ret) {
 443                                 faili = i;
 444                                 goto fail;
 445                         }
 446                 }
 447         }
 448         return 0;
 449 fail:
 450         WARN_ON(1);
 451         for (i =0; i < faili; i++) {
 452                 if (leaf) {
 453                         u64 disk_blocknr;
 454                         btrfs_item_key_to_cpu(buf, &key, i);
 455                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 456                                 continue;
 457                         fi = btrfs_item_ptr(buf, i,
 458                                             struct btrfs_file_extent_item);
 459                         if (btrfs_file_extent_type(buf, fi) ==
 460                             BTRFS_FILE_EXTENT_INLINE)
 461                                 continue;
 462                         disk_blocknr = btrfs_file_extent_disk_blocknr(buf, fi);
 463                         if (disk_blocknr == 0)
 464                                 continue;
 465                         err = btrfs_free_extent(trans, root, disk_blocknr,
 466                                     btrfs_file_extent_disk_num_blocks(buf,
 467                                                                       fi), 0);
 468                         BUG_ON(err);
 469                 } else {
 470                         blocknr = btrfs_node_blockptr(buf, i);
 471                         err = btrfs_free_extent(trans, root, blocknr, 1, 0);
 472                         BUG_ON(err);
 473                 }
 474         }
 475         return ret;
 476 }
 477
 478 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 479                                  struct btrfs_root *root,
 480                                  struct btrfs_path *path,
 481                                  struct btrfs_block_group_cache *cache)
 482 {
 483         int ret;
 484         int pending_ret;
 485         struct btrfs_root *extent_root = root->fs_info->extent_root;
 486         unsigned long bi;
 487         struct extent_buffer *leaf;
 488
 489         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 490         if (ret < 0)
 491                 goto fail;
 492         BUG_ON(ret);
 493
 494         leaf = path->nodes[0];
 495         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 496         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 497         btrfs_mark_buffer_dirty(leaf);
 498         btrfs_release_path(extent_root, path);
 499 fail:
 500         finish_current_insert(trans, extent_root);
 501         pending_ret = del_pending_extents(trans, extent_root);
 502         if (ret)
 503                 return ret;
 504         if (pending_ret)
 505                 return pending_ret;
 506         return 0;
 507
 508 }
 509
 510 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 511                                    struct btrfs_root *root)
 512 {
 513         struct extent_map_tree *block_group_cache;
 514         struct btrfs_block_group_cache *cache;
 515         int ret;
 516         int err = 0;
 517         int werr = 0;
 518         struct btrfs_path *path;
 519         u64 last = 0;
 520         u64 start;
 521         u64 end;
 522         u64 ptr;
 523
 524         block_group_cache = &root->fs_info->block_group_cache;
 525         path = btrfs_alloc_path();
 526         if (!path)
 527                 return -ENOMEM;
 528
 529         while(1) {
 530                 ret = find_first_extent_bit(block_group_cache, last,
 531                                             &start, &end, BLOCK_GROUP_DIRTY);
 532                 if (ret)
 533                         break;
 534
 535                 last = end + 1;
 536                 ret = get_state_private(block_group_cache, start, &ptr);
 537                 if (ret)
 538                         break;
 539
 540                 cache = (struct btrfs_block_group_cache *)ptr;
 541                 err = write_one_cache_group(trans, root,
 542                                             path, cache);
 543                 /*
 544                  * if we fail to write the cache group, we want
 545                  * to keep it marked dirty in hopes that a later
 546                  * write will work
 547                  */
 548                 if (err) {
 549                         werr = err;
 550                         continue;
 551                 }
 552                 clear_extent_bits(block_group_cache, start, end,
 553                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 554         }
 555         btrfs_free_path(path);
 556         return werr;
 557 }
 558
 559 static int update_block_group(struct btrfs_trans_handle *trans,
 560                               struct btrfs_root *root,
 561                               u64 blocknr, u64 num, int alloc, int mark_free,
 562                               int data)
 563 {
 564         struct btrfs_block_group_cache *cache;
 565         struct btrfs_fs_info *info = root->fs_info;
 566         u64 total = num;
 567         u64 old_val;
 568         u64 block_in_group;
 569         u64 start;
 570         u64 end;
 571
 572         while(total) {
 573                 cache = btrfs_lookup_block_group(info, blocknr);
 574                 if (!cache) {
 575                         return -1;
 576                 }
 577                 block_in_group = blocknr - cache->key.objectid;
 578                 WARN_ON(block_in_group > cache->key.offset);
 579                 start = cache->key.objectid;
 580                 end = start + cache->key.offset - 1;
 581                 set_extent_bits(&info->block_group_cache, start, end,
 582                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
 583
 584                 old_val = btrfs_block_group_used(&cache->item);
 585                 num = min(total, cache->key.offset - block_in_group);
 586                 if (alloc) {
 587                         if (cache->data != data &&
 588                             old_val < (cache->key.offset >> 1)) {
 589                                 int bit_to_clear;
 590                                 int bit_to_set;
 591
 592                                 cache->data = data;
 593                                 if (data) {
 594                                         bit_to_clear = BLOCK_GROUP_DATA;
 595                                         bit_to_set = BLOCK_GROUP_METADATA;
 596                                         cache->item.flags |=
 597                                                 BTRFS_BLOCK_GROUP_DATA;
 598                                 } else {
 599                                         bit_to_clear = BLOCK_GROUP_METADATA;
 600                                         bit_to_set = BLOCK_GROUP_DATA;
 601                                         cache->item.flags &=
 602                                                 ~BTRFS_BLOCK_GROUP_DATA;
 603                                 }
 604                                 clear_extent_bits(&info->block_group_cache,
 605                                                   start, end, bit_to_clear,
 606                                                   GFP_NOFS);
 607                                 set_extent_bits(&info->block_group_cache,
 608                                                 start, end, bit_to_set,
 609                                                 GFP_NOFS);
 610                         }
 611                         old_val += num;
 612                 } else {
 613                         old_val -= num;
 614                         if (mark_free) {
 615                                 set_extent_dirty(&info->free_space_cache,
 616                                                  blocknr, blocknr + num - 1,
 617                                                  GFP_NOFS);
 618                         }
 619                 }
 620                 btrfs_set_block_group_used(&cache->item, old_val);
 621                 total -= num;
 622                 blocknr += num;
 623         }
 624         return 0;
 625 }
 626
 627 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
 628 {
 629         u64 last = 0;
 630         u64 start;
 631         u64 end;
 632         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
 633         int ret;
 634
 635         while(1) {
 636                 ret = find_first_extent_bit(pinned_extents, last,
 637                                             &start, &end, EXTENT_DIRTY);
 638                 if (ret)
 639                         break;
 640                 set_extent_dirty(copy, start, end, GFP_NOFS);
 641                 last = end + 1;
 642         }
 643         return 0;
 644 }
 645
 646 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 647                                struct btrfs_root *root,
 648                                struct extent_map_tree *unpin)
 649 {
 650         u64 start;
 651         u64 end;
 652         int ret;
 653         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
 654         struct extent_map_tree *free_space_cache;
 655
 656         free_space_cache = &root->fs_info->free_space_cache;
 657
 658         while(1) {
 659                 ret = find_first_extent_bit(unpin, 0, &start, &end,
 660                                             EXTENT_DIRTY);
 661                 if (ret)
 662                         break;
 663
 664                 clear_extent_dirty(pinned_extents, start, end,
 665                                    GFP_NOFS);
 666                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
 667                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
 668         }
 669         return 0;
 670 }
 671
 672 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 673                                  btrfs_root *extent_root)
 674 {
 675         struct btrfs_key ins;
 676         struct btrfs_extent_item extent_item;
 677         int ret;
 678         int err = 0;
 679         u64 start;
 680         u64 end;
 681         struct btrfs_fs_info *info = extent_root->fs_info;
 682
 683         btrfs_set_stack_extent_refs(&extent_item, 1);
 684         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
 685         btrfs_set_stack_extent_owner(&extent_item,
 686                                      extent_root->root_key.objectid);
 687
 688         while(1) {
 689                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
 690                                             &end, EXTENT_LOCKED);
 691                 if (ret)
 692                         break;
 693
 694                 ins.objectid = start;
 695                 ins.offset = end + 1 - start;
 696                 err = btrfs_insert_item(trans, extent_root, &ins,
 697                                         &extent_item, sizeof(extent_item));
 698                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
 699                                   GFP_NOFS);
 700         }
 701         return 0;
 702 }
 703
 704 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
 705 {
 706         int err = 0;
 707         struct extent_buffer *buf;
 708
 709         if (!pending) {
 710                 buf = btrfs_find_tree_block(root, blocknr);
 711                 if (buf) {
 712                         if (btrfs_buffer_uptodate(buf)) {
 713                                 u64 transid =
 714                                     root->fs_info->running_transaction->transid;
 715                                 if (btrfs_header_generation(buf) == transid) {
 716                                         free_extent_buffer(buf);
 717                                         return 0;
 718                                 }
 719                         }
 720                         free_extent_buffer(buf);
 721                 }
 722                 set_extent_dirty(&root->fs_info->pinned_extents,
 723                                  blocknr, blocknr, GFP_NOFS);
 724         } else {
 725                 set_extent_bits(&root->fs_info->pending_del,
 726                                   blocknr, blocknr, EXTENT_LOCKED, GFP_NOFS);
 727         }
 728         BUG_ON(err < 0);
 729         return 0;
 730 }
 731
 732 /*
 733  * remove an extent from the root, returns 0 on success
 734  */
 735 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 736                          *root, u64 blocknr, u64 num_blocks, int pin,
 737                          int mark_free)
 738 {
 739         struct btrfs_path *path;
 740         struct btrfs_key key;
 741         struct btrfs_fs_info *info = root->fs_info;
 742         struct btrfs_root *extent_root = info->extent_root;
 743         struct extent_buffer *leaf;
 744         int ret;
 745         struct btrfs_extent_item *ei;
 746         u32 refs;
 747
 748         key.objectid = blocknr;
 749         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 750         key.offset = num_blocks;
 751
 752         path = btrfs_alloc_path();
 753         if (!path)
 754                 return -ENOMEM;
 755
 756         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 757         if (ret < 0)
 758                 return ret;
 759         BUG_ON(ret);
 760
 761         leaf = path->nodes[0];
 762         ei = btrfs_item_ptr(leaf, path->slots[0],
 763                             struct btrfs_extent_item);
 764         refs = btrfs_extent_refs(leaf, ei);
 765         BUG_ON(refs == 0);
 766         refs -= 1;
 767         btrfs_set_extent_refs(leaf, ei, refs);
 768         btrfs_mark_buffer_dirty(leaf);
 769
 770         if (refs == 0) {
 771                 u64 super_blocks_used, root_blocks_used;
 772
 773                 if (pin) {
 774                         ret = pin_down_block(root, blocknr, 0);
 775                         BUG_ON(ret);
 776                 }
 777
 778                 /* block accounting for super block */
 779                 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
 780                 btrfs_set_super_blocks_used(&info->super_copy,
 781                                             super_blocks_used - num_blocks);
 782
 783                 /* block accounting for root item */
 784                 root_blocks_used = btrfs_root_used(&root->root_item);
 785                 btrfs_set_root_used(&root->root_item,
 786                                            root_blocks_used - num_blocks);
 787
 788                 ret = btrfs_del_item(trans, extent_root, path);
 789                 if (ret) {
 790                         return ret;
 791                 }
 792                 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
 793                                          mark_free, 0);
 794                 BUG_ON(ret);
 795         }
 796         btrfs_free_path(path);
 797         finish_current_insert(trans, extent_root);
 798         return ret;
 799 }
 800
 801 /*
 802  * find all the blocks marked as pending in the radix tree and remove
 803  * them from the extent map
 804  */
 805 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 806                                btrfs_root *extent_root)
 807 {
 808         int ret;
 809         int err = 0;
 810         u64 start;
 811         u64 end;
 812         struct extent_map_tree *pending_del;
 813         struct extent_map_tree *pinned_extents;
 814
 815         pending_del = &extent_root->fs_info->pending_del;
 816         pinned_extents = &extent_root->fs_info->pinned_extents;
 817
 818         while(1) {
 819                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
 820                                             EXTENT_LOCKED);
 821                 if (ret)
 822                         break;
 823
 824                 set_extent_dirty(pinned_extents, start, end, GFP_NOFS);
 825                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
 826                                   GFP_NOFS);
 827                 ret = __free_extent(trans, extent_root,
 828                                      start, end + 1 - start, 0, 0);
 829                 if (ret)
 830                         err = ret;
 831         }
 832         return err;
 833 }
 834
 835 /*
 836  * remove an extent from the root, returns 0 on success
 837  */
 838 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 839                       *root, u64 blocknr, u64 num_blocks, int pin)
 840 {
 841         struct btrfs_root *extent_root = root->fs_info->extent_root;
 842         int pending_ret;
 843         int ret;
 844
 845         if (root == extent_root) {
 846                 pin_down_block(root, blocknr, 1);
 847                 return 0;
 848         }
 849         ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
 850         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
 851         return ret ? ret : pending_ret;
 852 }
 853
 854 /*
 855  * walks the btree of allocated extents and find a hole of a given size.
 856  * The key ins is changed to record the hole:
 857  * ins->objectid == block start
 858  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 859  * ins->offset == number of blocks
 860  * Any available blocks before search_start are skipped.
 861  */
 862 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 863                             *orig_root, u64 num_blocks, u64 empty_size,
 864                             u64 search_start, u64 search_end, u64 hint_block,
 865                             struct btrfs_key *ins, u64 exclude_start,
 866                             u64 exclude_nr, int data)
 867 {
 868         struct btrfs_path *path;
 869         struct btrfs_key key;
 870         int ret;
 871         u64 hole_size = 0;
 872         int slot = 0;
 873         u64 last_block = 0;
 874         u64 orig_search_start = search_start;
 875         int start_found;
 876         struct extent_buffer *l;
 877         struct btrfs_root * root = orig_root->fs_info->extent_root;
 878         struct btrfs_fs_info *info = root->fs_info;
 879         int total_needed = num_blocks;
 880         int level;
 881         struct btrfs_block_group_cache *block_group;
 882         int full_scan = 0;
 883         int wrapped = 0;
 884         u64 cached_search_start = 0;
 885
 886         WARN_ON(num_blocks < 1);
 887         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 888
 889         level = btrfs_header_level(root->node);
 890
 891         if (search_end == (u64)-1)
 892                 search_end = btrfs_super_total_blocks(&info->super_copy);
 893         if (hint_block) {
 894                 block_group = btrfs_lookup_block_group(info, hint_block);
 895                 block_group = btrfs_find_block_group(root, block_group,
 896                                                      hint_block, data, 1);
 897         } else {
 898                 block_group = btrfs_find_block_group(root,
 899                                                      trans->block_group, 0,
 900                                                      data, 1);
 901         }
 902
 903         total_needed += empty_size;
 904         path = btrfs_alloc_path();
 905
 906 check_failed:
 907         search_start = find_search_start(root, &block_group,
 908                                          search_start, total_needed, data);
 909         cached_search_start = search_start;
 910
 911         btrfs_init_path(path);
 912         ins->objectid = search_start;
 913         ins->offset = 0;
 914         start_found = 0;
 915         path->reada = 2;
 916
 917         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 918         if (ret < 0)
 919                 goto error;
 920
 921         if (path->slots[0] > 0) {
 922                 path->slots[0]--;
 923         }
 924
 925         l = path->nodes[0];
 926         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 927
 928         /*
 929          * a rare case, go back one key if we hit a block group item
 930          * instead of an extent item
 931          */
 932         if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
 933             key.objectid + key.offset >= search_start) {
 934                 ins->objectid = key.objectid;
 935                 ins->offset = key.offset - 1;
 936                 btrfs_release_path(root, path);
 937                 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 938                 if (ret < 0)
 939                         goto error;
 940
 941                 if (path->slots[0] > 0) {
 942                         path->slots[0]--;
 943                 }
 944         }
 945
 946         while (1) {
 947                 l = path->nodes[0];
 948                 slot = path->slots[0];
 949                 if (slot >= btrfs_header_nritems(l)) {
 950                         ret = btrfs_next_leaf(root, path);
 951                         if (ret == 0)
 952                                 continue;
 953                         if (ret < 0)
 954                                 goto error;
 955                         if (!start_found) {
 956                                 ins->objectid = search_start;
 957                                 ins->offset = search_end - search_start;
 958                                 start_found = 1;
 959                                 goto check_pending;
 960                         }
 961                         ins->objectid = last_block > search_start ?
 962                                         last_block : search_start;
 963                         ins->offset = search_end - ins->objectid;
 964                         goto check_pending;
 965                 }
 966                 btrfs_item_key_to_cpu(l, &key, slot);
 967
 968                 if (key.objectid >= search_start && key.objectid > last_block &&
 969                     start_found) {
 970                         if (last_block < search_start)
 971                                 last_block = search_start;
 972                         hole_size = key.objectid - last_block;
 973                         if (hole_size >= num_blocks) {
 974                                 ins->objectid = last_block;
 975                                 ins->offset = hole_size;
 976                                 goto check_pending;
 977                         }
 978                 }
 979                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
 980                         if (!start_found) {
 981                                 last_block = key.objectid;
 982                                 start_found = 1;
 983                         }
 984                         goto next;
 985                 }
 986
 987
 988                 start_found = 1;
 989                 last_block = key.objectid + key.offset;
 990
 991                 if (!full_scan && last_block >= block_group->key.objectid +
 992                     block_group->key.offset) {
 993                         btrfs_release_path(root, path);
 994                         search_start = block_group->key.objectid +
 995                                 block_group->key.offset * 2;
 996                         goto new_group;
 997                 }
 998 next:
 999                 path->slots[0]++;
1000                 cond_resched();
1001         }
1002 check_pending:
1003         /* we have to make sure we didn't find an extent that has already
1004          * been allocated by the map tree or the original allocation
1005          */
1006         btrfs_release_path(root, path);
1007         BUG_ON(ins->objectid < search_start);
1008
1009         if (ins->objectid + num_blocks >= search_end)
1010                 goto enospc;
1011
1012         if (test_range_bit(&info->extent_ins, ins->objectid,
1013                            ins->objectid + num_blocks -1, EXTENT_LOCKED, 0)) {
1014                 search_start = ins->objectid + num_blocks;
1015                 goto new_group;
1016         }
1017         if (test_range_bit(&info->pinned_extents, ins->objectid,
1018                            ins->objectid + num_blocks -1, EXTENT_DIRTY, 0)) {
1019                 search_start = ins->objectid + num_blocks;
1020                 goto new_group;
1021         }
1022         if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1023             ins->objectid < exclude_start + exclude_nr)) {
1024                 search_start = exclude_start + exclude_nr;
1025                 goto new_group;
1026         }
1027         if (!data) {
1028                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1029                 if (block_group)
1030                         trans->block_group = block_group;
1031         }
1032         ins->offset = num_blocks;
1033         btrfs_free_path(path);
1034         return 0;
1035
1036 new_group:
1037         if (search_start + num_blocks >= search_end) {
1038 enospc:
1039                 search_start = orig_search_start;
1040                 if (full_scan) {
1041                         ret = -ENOSPC;
1042                         goto error;
1043                 }
1044                 if (wrapped) {
1045                         if (!full_scan)
1046                                 total_needed -= empty_size;
1047                         full_scan = 1;
1048                 } else
1049                         wrapped = 1;
1050         }
1051         block_group = btrfs_lookup_block_group(info, search_start);
1052         cond_resched();
1053         if (!full_scan)
1054                 block_group = btrfs_find_block_group(root, block_group,
1055                                                      search_start, data, 0);
1056         goto check_failed;
1057
1058 error:
1059         btrfs_release_path(root, path);
1060         btrfs_free_path(path);
1061         return ret;
1062 }
1063 /*
1064  * finds a free extent and does all the dirty work required for allocation
1065  * returns the key for the extent through ins, and a tree buffer for
1066  * the first block of the extent through buf.
1067  *
1068  * returns 0 if everything worked, non-zero otherwise.
1069  */
1070 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1071                        struct btrfs_root *root, u64 owner,
1072                        u64 num_blocks, u64 empty_size, u64 hint_block,
1073                        u64 search_end, struct btrfs_key *ins, int data)
1074 {
1075         int ret;
1076         int pending_ret;
1077         u64 super_blocks_used, root_blocks_used;
1078         u64 search_start = 0;
1079         struct btrfs_fs_info *info = root->fs_info;
1080         struct btrfs_root *extent_root = info->extent_root;
1081         struct btrfs_extent_item extent_item;
1082
1083         btrfs_set_stack_extent_refs(&extent_item, 1);
1084         btrfs_set_stack_extent_owner(&extent_item, owner);
1085
1086         WARN_ON(num_blocks < 1);
1087         ret = find_free_extent(trans, root, num_blocks, empty_size,
1088                                search_start, search_end, hint_block, ins,
1089                                trans->alloc_exclude_start,
1090                                trans->alloc_exclude_nr, data);
1091         BUG_ON(ret);
1092         if (ret)
1093                 return ret;
1094
1095         /* block accounting for super block */
1096         super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1097         btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1098                                     num_blocks);
1099
1100         /* block accounting for root item */
1101         root_blocks_used = btrfs_root_used(&root->root_item);
1102         btrfs_set_root_used(&root->root_item, root_blocks_used +
1103                                    num_blocks);
1104
1105         clear_extent_dirty(&root->fs_info->free_space_cache,
1106                            ins->objectid, ins->objectid + ins->offset - 1,
1107                            GFP_NOFS);
1108
1109         if (root == extent_root) {
1110                 BUG_ON(num_blocks != 1);
1111                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1112                                 ins->objectid + ins->offset - 1,
1113                                 EXTENT_LOCKED, GFP_NOFS);
1114                 goto update_block;
1115         }
1116
1117         WARN_ON(trans->alloc_exclude_nr);
1118         trans->alloc_exclude_start = ins->objectid;
1119         trans->alloc_exclude_nr = ins->offset;
1120         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1121                                 sizeof(extent_item));
1122
1123         trans->alloc_exclude_start = 0;
1124         trans->alloc_exclude_nr = 0;
1125
1126         BUG_ON(ret);
1127         finish_current_insert(trans, extent_root);
1128         pending_ret = del_pending_extents(trans, extent_root);
1129
1130         if (ret) {
1131                 return ret;
1132         }
1133         if (pending_ret) {
1134                 return pending_ret;
1135         }
1136
1137 update_block:
1138         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1139                                  data);
1140         BUG_ON(ret);
1141         return 0;
1142 }
1143
1144 /*
1145  * helper function to allocate a block for a given tree
1146  * returns the tree buffer or NULL.
1147  */
1148 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1149                                              struct btrfs_root *root, u64 hint,
1150                                              u64 empty_size)
1151 {
1152         struct btrfs_key ins;
1153         int ret;
1154         struct extent_buffer *buf;
1155
1156         ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1157                                  1, empty_size, hint, (u64)-1, &ins, 0);
1158         if (ret) {
1159                 BUG_ON(ret > 0);
1160                 return ERR_PTR(ret);
1161         }
1162         buf = btrfs_find_create_tree_block(root, ins.objectid);
1163         if (!buf) {
1164                 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1165                 return ERR_PTR(-ENOMEM);
1166         }
1167         btrfs_set_buffer_uptodate(buf);
1168         buf->alloc_addr = (unsigned long)__builtin_return_address(0);
1169         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1170                          buf->start + buf->len - 1, GFP_NOFS);
1171         /*
1172         set_buffer_checked(buf);
1173         set_buffer_defrag(buf);
1174         */
1175         /* FIXME!!!!!!!!!!!!!!!!
1176         set_radix_bit(&trans->transaction->dirty_pages, buf->pages[0]->index);
1177         */
1178         trans->blocks_used++;
1179         return buf;
1180 }
1181
1182 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1183                          struct btrfs_root *root, struct extent_buffer *leaf)
1184 {
1185         struct btrfs_key key;
1186         struct btrfs_file_extent_item *fi;
1187         int i;
1188         int nritems;
1189         int ret;
1190
1191         BUG_ON(!btrfs_is_leaf(leaf));
1192         nritems = btrfs_header_nritems(leaf);
1193         for (i = 0; i < nritems; i++) {
1194                 u64 disk_blocknr;
1195
1196                 btrfs_item_key_to_cpu(leaf, &key, i);
1197                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1198                         continue;
1199                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1200                 if (btrfs_file_extent_type(leaf, fi) ==
1201                     BTRFS_FILE_EXTENT_INLINE)
1202                         continue;
1203                 /*
1204                  * FIXME make sure to insert a trans record that
1205                  * repeats the snapshot del on crash
1206                  */
1207                 disk_blocknr = btrfs_file_extent_disk_blocknr(leaf, fi);
1208                 if (disk_blocknr == 0)
1209                         continue;
1210                 ret = btrfs_free_extent(trans, root, disk_blocknr,
1211                                 btrfs_file_extent_disk_num_blocks(leaf, fi), 0);
1212                 BUG_ON(ret);
1213         }
1214         return 0;
1215 }
1216
1217 static void reada_walk_down(struct btrfs_root *root,
1218                             struct extent_buffer *node)
1219 {
1220         int i;
1221         u32 nritems;
1222         u64 blocknr;
1223         int ret;
1224         u32 refs;
1225
1226         nritems = btrfs_header_nritems(node);
1227         for (i = 0; i < nritems; i++) {
1228                 blocknr = btrfs_node_blockptr(node, i);
1229                 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1230                 BUG_ON(ret);
1231                 if (refs != 1)
1232                         continue;
1233                 mutex_unlock(&root->fs_info->fs_mutex);
1234                 ret = readahead_tree_block(root, blocknr);
1235                 cond_resched();
1236                 mutex_lock(&root->fs_info->fs_mutex);
1237                 if (ret)
1238                         break;
1239         }
1240 }
1241
1242 /*
1243  * helper function for drop_snapshot, this walks down the tree dropping ref
1244  * counts as it goes.
1245  */
1246 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1247                           *root, struct btrfs_path *path, int *level)
1248 {
1249         struct extent_buffer *next;
1250         struct extent_buffer *cur;
1251         u64 blocknr;
1252         int ret;
1253         u32 refs;
1254
1255         WARN_ON(*level < 0);
1256         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1257         ret = lookup_extent_ref(trans, root,
1258                                 extent_buffer_blocknr(path->nodes[*level]),
1259                                 1, &refs);
1260         BUG_ON(ret);
1261         if (refs > 1)
1262                 goto out;
1263
1264         /*
1265          * walk down to the last node level and free all the leaves
1266          */
1267         while(*level >= 0) {
1268                 WARN_ON(*level < 0);
1269                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1270                 cur = path->nodes[*level];
1271
1272                 if (*level > 0 && path->slots[*level] == 0)
1273                         reada_walk_down(root, cur);
1274
1275                 if (btrfs_header_level(cur) != *level)
1276                         WARN_ON(1);
1277
1278                 if (path->slots[*level] >=
1279                     btrfs_header_nritems(cur))
1280                         break;
1281                 if (*level == 0) {
1282                         ret = drop_leaf_ref(trans, root, cur);
1283                         BUG_ON(ret);
1284                         break;
1285                 }
1286                 blocknr = btrfs_node_blockptr(cur, path->slots[*level]);
1287                 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1288                 BUG_ON(ret);
1289                 if (refs != 1) {
1290                         path->slots[*level]++;
1291                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1292                         BUG_ON(ret);
1293                         continue;
1294                 }
1295                 next = btrfs_find_tree_block(root, blocknr);
1296                 if (!next || !btrfs_buffer_uptodate(next)) {
1297                         free_extent_buffer(next);
1298                         mutex_unlock(&root->fs_info->fs_mutex);
1299                         next = read_tree_block(root, blocknr);
1300                         mutex_lock(&root->fs_info->fs_mutex);
1301
1302                         /* we dropped the lock, check one more time */
1303                         ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1304                         BUG_ON(ret);
1305                         if (refs != 1) {
1306                                 path->slots[*level]++;
1307                                 free_extent_buffer(next);
1308                                 ret = btrfs_free_extent(trans, root,
1309                                                         blocknr, 1, 1);
1310                                 BUG_ON(ret);
1311                                 continue;
1312                         }
1313                 }
1314                 WARN_ON(*level <= 0);
1315                 if (path->nodes[*level-1])
1316                         free_extent_buffer(path->nodes[*level-1]);
1317                 path->nodes[*level-1] = next;
1318                 *level = btrfs_header_level(next);
1319                 path->slots[*level] = 0;
1320         }
1321 out:
1322         WARN_ON(*level < 0);
1323         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1324         ret = btrfs_free_extent(trans, root,
1325                         extent_buffer_blocknr(path->nodes[*level]), 1, 1);
1326         free_extent_buffer(path->nodes[*level]);
1327         path->nodes[*level] = NULL;
1328         *level += 1;
1329         BUG_ON(ret);
1330         return 0;
1331 }
1332
1333 /*
1334  * helper for dropping snapshots.  This walks back up the tree in the path
1335  * to find the first node higher up where we haven't yet gone through
1336  * all the slots
1337  */
1338 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1339                         *root, struct btrfs_path *path, int *level)
1340 {
1341         int i;
1342         int slot;
1343         int ret;
1344         struct btrfs_root_item *root_item = &root->root_item;
1345
1346         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1347                 slot = path->slots[i];
1348                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1349                         struct extent_buffer *node;
1350                         struct btrfs_disk_key disk_key;
1351                         node = path->nodes[i];
1352                         path->slots[i]++;
1353                         *level = i;
1354                         WARN_ON(*level == 0);
1355                         btrfs_node_key(node, &disk_key, path->slots[i]);
1356                         memcpy(&root_item->drop_progress,
1357                                &disk_key, sizeof(disk_key));
1358                         root_item->drop_level = i;
1359                         return 0;
1360                 } else {
1361                         ret = btrfs_free_extent(trans, root,
1362                                     extent_buffer_blocknr(path->nodes[*level]),
1363                                     1, 1);
1364                         BUG_ON(ret);
1365                         free_extent_buffer(path->nodes[*level]);
1366                         path->nodes[*level] = NULL;
1367                         *level = i + 1;
1368                 }
1369         }
1370         return 1;
1371 }
1372
1373 /*
1374  * drop the reference count on the tree rooted at 'snap'.  This traverses
1375  * the tree freeing any blocks that have a ref count of zero after being
1376  * decremented.
1377  */
1378 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1379                         *root)
1380 {
1381         int ret = 0;
1382         int wret;
1383         int level;
1384         struct btrfs_path *path;
1385         int i;
1386         int orig_level;
1387         struct btrfs_root_item *root_item = &root->root_item;
1388
1389         path = btrfs_alloc_path();
1390         BUG_ON(!path);
1391
1392         level = btrfs_header_level(root->node);
1393         orig_level = level;
1394         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1395                 path->nodes[level] = root->node;
1396                 extent_buffer_get(root->node);
1397                 path->slots[level] = 0;
1398         } else {
1399                 struct btrfs_key key;
1400                 struct btrfs_disk_key found_key;
1401                 struct extent_buffer *node;
1402
1403                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1404                 level = root_item->drop_level;
1405                 path->lowest_level = level;
1406                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1407                 if (wret < 0) {
1408                         ret = wret;
1409                         goto out;
1410                 }
1411                 node = path->nodes[level];
1412                 btrfs_node_key(node, &found_key, path->slots[level]);
1413                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1414                                sizeof(found_key)));
1415         }
1416         while(1) {
1417                 wret = walk_down_tree(trans, root, path, &level);
1418                 if (wret > 0)
1419                         break;
1420                 if (wret < 0)
1421                         ret = wret;
1422
1423                 wret = walk_up_tree(trans, root, path, &level);
1424                 if (wret > 0)
1425                         break;
1426                 if (wret < 0)
1427                         ret = wret;
1428                 ret = -EAGAIN;
1429                 break;
1430         }
1431         for (i = 0; i <= orig_level; i++) {
1432                 if (path->nodes[i]) {
1433                         free_extent_buffer(path->nodes[i]);
1434                         path->nodes[i] = 0;
1435                 }
1436         }
1437 out:
1438         btrfs_free_path(path);
1439         return ret;
1440 }
1441
1442 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1443 {
1444         u64 start;
1445         u64 end;
1446         int ret;
1447
1448         while(1) {
1449                 ret = find_first_extent_bit(&info->block_group_cache, 0,
1450                                             &start, &end, (unsigned int)-1);
1451                 if (ret)
1452                         break;
1453                 clear_extent_bits(&info->block_group_cache, start,
1454                                   end, (unsigned int)-1, GFP_NOFS);
1455         }
1456         while(1) {
1457                 ret = find_first_extent_bit(&info->free_space_cache, 0,
1458                                             &start, &end, EXTENT_DIRTY);
1459                 if (ret)
1460                         break;
1461                 clear_extent_dirty(&info->free_space_cache, start,
1462                                    end, GFP_NOFS);
1463         }
1464         return 0;
1465 }
1466
1467 int btrfs_read_block_groups(struct btrfs_root *root)
1468 {
1469         struct btrfs_path *path;
1470         int ret;
1471         int err = 0;
1472         int bit;
1473         struct btrfs_block_group_cache *cache;
1474         struct btrfs_fs_info *info = root->fs_info;
1475         struct extent_map_tree *block_group_cache;
1476         struct btrfs_key key;
1477         struct btrfs_key found_key;
1478         struct extent_buffer *leaf;
1479         u64 group_size_blocks;
1480
1481         block_group_cache = &info->block_group_cache;
1482
1483         group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1484                 info->sb->s_blocksize_bits;
1485
1486         root = info->extent_root;
1487         key.objectid = 0;
1488         key.offset = group_size_blocks;
1489         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1490
1491         path = btrfs_alloc_path();
1492         if (!path)
1493                 return -ENOMEM;
1494
1495         while(1) {
1496                 ret = btrfs_search_slot(NULL, info->extent_root,
1497                                         &key, path, 0, 0);
1498                 if (ret != 0) {
1499                         err = ret;
1500                         break;
1501                 }
1502                 leaf = path->nodes[0];
1503                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1504                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1505                 if (!cache) {
1506                         err = -1;
1507                         break;
1508                 }
1509
1510                 read_extent_buffer(leaf, &cache->item,
1511                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
1512                                    sizeof(cache->item));
1513                 memcpy(&cache->key, &found_key, sizeof(found_key));
1514                 cache->cached = 0;
1515
1516                 key.objectid = found_key.objectid + found_key.offset;
1517                 btrfs_release_path(root, path);
1518
1519                 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
1520                         bit = BLOCK_GROUP_DATA;
1521                         cache->data = 1;
1522                 } else {
1523                         bit = BLOCK_GROUP_METADATA;
1524                         cache->data = 0;
1525                 }
1526
1527                 /* use EXTENT_LOCKED to prevent merging */
1528                 set_extent_bits(block_group_cache, found_key.objectid,
1529                                 found_key.objectid + found_key.offset - 1,
1530                                 bit | EXTENT_LOCKED, GFP_NOFS);
1531                 set_state_private(block_group_cache, found_key.objectid,
1532                                   (u64)cache);
1533
1534                 if (key.objectid >=
1535                     btrfs_super_total_blocks(&info->super_copy))
1536                         break;
1537         }
1538
1539         btrfs_free_path(path);
1540         return 0;
1541 }