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 max(cache->last_alloc, 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 max((*cache_ret)->last_alloc, 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 + shint->pinned <
 251                             div_factor(shint->key.offset, factor)) {
 252                                 return shint;
 253                         }
 254                 }
 255         }
 256         if (hint && hint->data == data) {
 257                 used = btrfs_block_group_used(&hint->item);
 258                 if (used + hint->pinned <
 259                     div_factor(hint->key.offset, factor)) {
 260                         return hint;
 261                 }
 262                 last = hint->key.offset * 3;
 263                 if (hint->key.objectid >= last)
 264                         last = max(search_start + hint->key.offset - 1,
 265                                    hint->key.objectid - last);
 266                 else
 267                         last = hint->key.objectid + hint->key.offset;
 268                 hint_last = last;
 269         } else {
 270                 if (hint)
 271                         hint_last = max(hint->key.objectid, search_start);
 272                 else
 273                         hint_last = search_start;
 274
 275                 last = hint_last;
 276         }
 277 again:
 278         while(1) {
 279                 ret = find_first_extent_bit(block_group_cache, last,
 280                                             &start, &end, bit);
 281                 if (ret)
 282                         break;
 283
 284                 ret = get_state_private(block_group_cache, start, &ptr);
 285                 if (ret)
 286                         break;
 287
 288                 cache = (struct btrfs_block_group_cache *)ptr;
 289                 last = cache->key.objectid + cache->key.offset;
 290                 used = btrfs_block_group_used(&cache->item);
 291
 292                 if (full_search)
 293                         free_check = cache->key.offset;
 294                 else
 295                         free_check = div_factor(cache->key.offset, factor);
 296
 297                 if (used + cache->pinned < free_check) {
 298                         found_group = cache;
 299                         goto found;
 300                 }
 301                 cond_resched();
 302         }
 303         if (!full_search) {
 304                 last = search_start;
 305                 full_search = 1;
 306                 goto again;
 307         }
 308         if (!data_swap) {
 309                 data_swap = 1;
 310                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 311                 last = search_start;
 312                 goto again;
 313         }
 314 found:
 315         return found_group;
 316 }
 317
 318 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 319                                 struct btrfs_root *root,
 320                                 u64 blocknr, u64 num_blocks)
 321 {
 322         struct btrfs_path *path;
 323         int ret;
 324         struct btrfs_key key;
 325         struct extent_buffer *l;
 326         struct btrfs_extent_item *item;
 327         u32 refs;
 328
 329         path = btrfs_alloc_path();
 330         if (!path)
 331                 return -ENOMEM;
 332
 333         key.objectid = blocknr;
 334         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 335         key.offset = num_blocks;
 336         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 337                                 0, 1);
 338         if (ret < 0)
 339                 return ret;
 340         if (ret != 0) {
 341                 BUG();
 342         }
 343         BUG_ON(ret != 0);
 344         l = path->nodes[0];
 345         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 346         refs = btrfs_extent_refs(l, item);
 347         btrfs_set_extent_refs(l, item, refs + 1);
 348         btrfs_mark_buffer_dirty(path->nodes[0]);
 349
 350         btrfs_release_path(root->fs_info->extent_root, path);
 351         btrfs_free_path(path);
 352         finish_current_insert(trans, root->fs_info->extent_root);
 353         del_pending_extents(trans, root->fs_info->extent_root);
 354         return 0;
 355 }
 356
 357 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 358                          struct btrfs_root *root)
 359 {
 360         finish_current_insert(trans, root->fs_info->extent_root);
 361         del_pending_extents(trans, root->fs_info->extent_root);
 362         return 0;
 363 }
 364
 365 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 366                              struct btrfs_root *root, u64 blocknr,
 367                              u64 num_blocks, u32 *refs)
 368 {
 369         struct btrfs_path *path;
 370         int ret;
 371         struct btrfs_key key;
 372         struct extent_buffer *l;
 373         struct btrfs_extent_item *item;
 374
 375         path = btrfs_alloc_path();
 376         key.objectid = blocknr;
 377         key.offset = num_blocks;
 378         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 379         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 380                                 0, 0);
 381         if (ret < 0)
 382                 goto out;
 383         if (ret != 0) {
 384                 btrfs_print_leaf(root, path->nodes[0]);
 385                 printk("failed to find block number %Lu\n", blocknr);
 386                 BUG();
 387         }
 388         l = path->nodes[0];
 389         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 390         *refs = btrfs_extent_refs(l, item);
 391 out:
 392         btrfs_free_path(path);
 393         return 0;
 394 }
 395
 396 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 397                        struct btrfs_root *root)
 398 {
 399         return btrfs_inc_extent_ref(trans, root,
 400                                     extent_buffer_blocknr(root->node), 1);
 401 }
 402
 403 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 404                   struct extent_buffer *buf)
 405 {
 406         u64 blocknr;
 407         u32 nritems;
 408         struct btrfs_key key;
 409         struct btrfs_file_extent_item *fi;
 410         int i;
 411         int leaf;
 412         int ret;
 413         int faili;
 414         int err;
 415
 416         if (!root->ref_cows)
 417                 return 0;
 418
 419         leaf = btrfs_is_leaf(buf);
 420         nritems = btrfs_header_nritems(buf);
 421         for (i = 0; i < nritems; i++) {
 422                 if (leaf) {
 423                         u64 disk_blocknr;
 424                         btrfs_item_key_to_cpu(buf, &key, i);
 425                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 426                                 continue;
 427                         fi = btrfs_item_ptr(buf, i,
 428                                             struct btrfs_file_extent_item);
 429                         if (btrfs_file_extent_type(buf, fi) ==
 430                             BTRFS_FILE_EXTENT_INLINE)
 431                                 continue;
 432                         disk_blocknr = btrfs_file_extent_disk_blocknr(buf, fi);
 433                         if (disk_blocknr == 0)
 434                                 continue;
 435                         ret = btrfs_inc_extent_ref(trans, root, disk_blocknr,
 436                                     btrfs_file_extent_disk_num_blocks(buf, fi));
 437                         if (ret) {
 438                                 faili = i;
 439                                 goto fail;
 440                         }
 441                 } else {
 442                         blocknr = btrfs_node_blockptr(buf, i);
 443                         ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
 444                         if (ret) {
 445                                 faili = i;
 446                                 goto fail;
 447                         }
 448                 }
 449         }
 450         return 0;
 451 fail:
 452         WARN_ON(1);
 453         for (i =0; i < faili; i++) {
 454                 if (leaf) {
 455                         u64 disk_blocknr;
 456                         btrfs_item_key_to_cpu(buf, &key, i);
 457                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 458                                 continue;
 459                         fi = btrfs_item_ptr(buf, i,
 460                                             struct btrfs_file_extent_item);
 461                         if (btrfs_file_extent_type(buf, fi) ==
 462                             BTRFS_FILE_EXTENT_INLINE)
 463                                 continue;
 464                         disk_blocknr = btrfs_file_extent_disk_blocknr(buf, fi);
 465                         if (disk_blocknr == 0)
 466                                 continue;
 467                         err = btrfs_free_extent(trans, root, disk_blocknr,
 468                                     btrfs_file_extent_disk_num_blocks(buf,
 469                                                                       fi), 0);
 470                         BUG_ON(err);
 471                 } else {
 472                         blocknr = btrfs_node_blockptr(buf, i);
 473                         err = btrfs_free_extent(trans, root, blocknr, 1, 0);
 474                         BUG_ON(err);
 475                 }
 476         }
 477         return ret;
 478 }
 479
 480 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 481                                  struct btrfs_root *root,
 482                                  struct btrfs_path *path,
 483                                  struct btrfs_block_group_cache *cache)
 484 {
 485         int ret;
 486         int pending_ret;
 487         struct btrfs_root *extent_root = root->fs_info->extent_root;
 488         unsigned long bi;
 489         struct extent_buffer *leaf;
 490
 491         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 492         if (ret < 0)
 493                 goto fail;
 494         BUG_ON(ret);
 495
 496         leaf = path->nodes[0];
 497         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 498         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 499         btrfs_mark_buffer_dirty(leaf);
 500         btrfs_release_path(extent_root, path);
 501 fail:
 502         finish_current_insert(trans, extent_root);
 503         pending_ret = del_pending_extents(trans, extent_root);
 504         if (ret)
 505                 return ret;
 506         if (pending_ret)
 507                 return pending_ret;
 508         if (cache->data)
 509                 cache->last_alloc = cache->first_free;
 510         return 0;
 511
 512 }
 513
 514 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 515                                    struct btrfs_root *root)
 516 {
 517         struct extent_map_tree *block_group_cache;
 518         struct btrfs_block_group_cache *cache;
 519         int ret;
 520         int err = 0;
 521         int werr = 0;
 522         struct btrfs_path *path;
 523         u64 last = 0;
 524         u64 start;
 525         u64 end;
 526         u64 ptr;
 527
 528         block_group_cache = &root->fs_info->block_group_cache;
 529         path = btrfs_alloc_path();
 530         if (!path)
 531                 return -ENOMEM;
 532
 533         while(1) {
 534                 ret = find_first_extent_bit(block_group_cache, last,
 535                                             &start, &end, BLOCK_GROUP_DIRTY);
 536                 if (ret)
 537                         break;
 538
 539                 last = end + 1;
 540                 ret = get_state_private(block_group_cache, start, &ptr);
 541                 if (ret)
 542                         break;
 543
 544                 cache = (struct btrfs_block_group_cache *)ptr;
 545                 err = write_one_cache_group(trans, root,
 546                                             path, cache);
 547                 /*
 548                  * if we fail to write the cache group, we want
 549                  * to keep it marked dirty in hopes that a later
 550                  * write will work
 551                  */
 552                 if (err) {
 553                         werr = err;
 554                         continue;
 555                 }
 556                 clear_extent_bits(block_group_cache, start, end,
 557                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 558         }
 559         btrfs_free_path(path);
 560         return werr;
 561 }
 562
 563 static int update_block_group(struct btrfs_trans_handle *trans,
 564                               struct btrfs_root *root,
 565                               u64 blocknr, u64 num, int alloc, int mark_free,
 566                               int data)
 567 {
 568         struct btrfs_block_group_cache *cache;
 569         struct btrfs_fs_info *info = root->fs_info;
 570         u64 total = num;
 571         u64 old_val;
 572         u64 block_in_group;
 573         u64 start;
 574         u64 end;
 575
 576         while(total) {
 577                 cache = btrfs_lookup_block_group(info, blocknr);
 578                 if (!cache) {
 579                         return -1;
 580                 }
 581                 block_in_group = blocknr - cache->key.objectid;
 582                 WARN_ON(block_in_group > cache->key.offset);
 583                 start = cache->key.objectid;
 584                 end = start + cache->key.offset - 1;
 585                 set_extent_bits(&info->block_group_cache, start, end,
 586                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
 587
 588                 old_val = btrfs_block_group_used(&cache->item);
 589                 num = min(total, cache->key.offset - block_in_group);
 590                 if (alloc) {
 591                         if (blocknr > cache->last_alloc)
 592                                 cache->last_alloc = blocknr;
 593                         if (cache->data != data &&
 594                             old_val < (cache->key.offset >> 1)) {
 595                                 int bit_to_clear;
 596                                 int bit_to_set;
 597
 598                                 cache->data = data;
 599                                 if (data) {
 600                                         bit_to_clear = BLOCK_GROUP_DATA;
 601                                         bit_to_set = BLOCK_GROUP_METADATA;
 602                                         cache->item.flags |=
 603                                                 BTRFS_BLOCK_GROUP_DATA;
 604                                 } else {
 605                                         bit_to_clear = BLOCK_GROUP_METADATA;
 606                                         bit_to_set = BLOCK_GROUP_DATA;
 607                                         cache->item.flags &=
 608                                                 ~BTRFS_BLOCK_GROUP_DATA;
 609                                 }
 610                                 clear_extent_bits(&info->block_group_cache,
 611                                                   start, end, bit_to_clear,
 612                                                   GFP_NOFS);
 613                                 set_extent_bits(&info->block_group_cache,
 614                                                 start, end, bit_to_set,
 615                                                 GFP_NOFS);
 616                         }
 617                         old_val += num;
 618                 } else {
 619                         old_val -= num;
 620                         if (blocknr < cache->first_free)
 621                                 cache->first_free = blocknr;
 622                         if (mark_free) {
 623                                 set_extent_dirty(&info->free_space_cache,
 624                                                  blocknr, blocknr + num - 1,
 625                                                  GFP_NOFS);
 626                         }
 627                 }
 628                 btrfs_set_block_group_used(&cache->item, old_val);
 629                 total -= num;
 630                 blocknr += num;
 631         }
 632         return 0;
 633 }
 634
 635 int btrfs_copy_pinned(struct btrfs_root *root, struct radix_tree_root *copy)
 636 {
 637         unsigned long gang[8];
 638         u64 last = 0;
 639         struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
 640         int ret;
 641         int i;
 642
 643         while(1) {
 644                 ret = find_first_radix_bit(pinned_radix, gang, last,
 645                                            ARRAY_SIZE(gang));
 646                 if (!ret)
 647                         break;
 648                 for (i = 0 ; i < ret; i++) {
 649                         set_radix_bit(copy, gang[i]);
 650                         last = gang[i] + 1;
 651                 }
 652         }
 653         ret = find_first_radix_bit(&root->fs_info->extent_ins_radix, gang, 0,
 654                                    ARRAY_SIZE(gang));
 655         WARN_ON(ret);
 656         return 0;
 657 }
 658
 659 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 660                                struct btrfs_root *root,
 661                                struct radix_tree_root *unpin_radix)
 662 {
 663         unsigned long gang[8];
 664         struct btrfs_block_group_cache *block_group;
 665         u64 first = 0;
 666         int ret;
 667         int i;
 668         struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
 669         struct extent_map_tree *free_space_cache;
 670
 671         free_space_cache = &root->fs_info->free_space_cache;
 672
 673         while(1) {
 674                 ret = find_first_radix_bit(unpin_radix, gang, 0,
 675                                            ARRAY_SIZE(gang));
 676                 if (!ret)
 677                         break;
 678                 if (!first)
 679                         first = gang[0];
 680                 for (i = 0; i < ret; i++) {
 681                         clear_radix_bit(pinned_radix, gang[i]);
 682                         clear_radix_bit(unpin_radix, gang[i]);
 683                         block_group = btrfs_lookup_block_group(root->fs_info,
 684                                                                gang[i]);
 685                         if (block_group) {
 686                                 WARN_ON(block_group->pinned == 0);
 687                                 block_group->pinned--;
 688                                 if (gang[i] < block_group->last_alloc)
 689                                         block_group->last_alloc = gang[i];
 690                                 set_extent_dirty(free_space_cache,
 691                                                  gang[i], gang[i], GFP_NOFS);
 692                         }
 693                 }
 694         }
 695         return 0;
 696 }
 697
 698 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 699                                  btrfs_root *extent_root)
 700 {
 701         struct btrfs_key ins;
 702         struct btrfs_extent_item extent_item;
 703         int i;
 704         int ret;
 705         int err;
 706         unsigned long gang[8];
 707         struct btrfs_fs_info *info = extent_root->fs_info;
 708
 709         btrfs_set_stack_extent_refs(&extent_item, 1);
 710         ins.offset = 1;
 711         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
 712         btrfs_set_stack_extent_owner(&extent_item,
 713                                      extent_root->root_key.objectid);
 714
 715         while(1) {
 716                 ret = find_first_radix_bit(&info->extent_ins_radix, gang, 0,
 717                                            ARRAY_SIZE(gang));
 718                 if (!ret)
 719                         break;
 720
 721                 for (i = 0; i < ret; i++) {
 722                         ins.objectid = gang[i];
 723                         err = btrfs_insert_item(trans, extent_root, &ins,
 724                                                 &extent_item,
 725                                                 sizeof(extent_item));
 726                         clear_radix_bit(&info->extent_ins_radix, gang[i]);
 727                         WARN_ON(err);
 728                 }
 729         }
 730         return 0;
 731 }
 732
 733 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
 734 {
 735         int err;
 736         struct extent_buffer *buf;
 737
 738         if (!pending) {
 739                 buf = btrfs_find_tree_block(root, blocknr);
 740                 if (buf) {
 741                         if (btrfs_buffer_uptodate(buf)) {
 742                                 u64 transid =
 743                                     root->fs_info->running_transaction->transid;
 744                                 if (btrfs_header_generation(buf) == transid) {
 745                                         free_extent_buffer(buf);
 746                                         return 0;
 747                                 }
 748                         }
 749                         free_extent_buffer(buf);
 750                 }
 751                 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
 752                 if (!err) {
 753                         struct btrfs_block_group_cache *cache;
 754                         cache = btrfs_lookup_block_group(root->fs_info,
 755                                                          blocknr);
 756                         if (cache)
 757                                 cache->pinned++;
 758                 }
 759         } else {
 760                 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
 761         }
 762         BUG_ON(err < 0);
 763         return 0;
 764 }
 765
 766 /*
 767  * remove an extent from the root, returns 0 on success
 768  */
 769 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 770                          *root, u64 blocknr, u64 num_blocks, int pin,
 771                          int mark_free)
 772 {
 773         struct btrfs_path *path;
 774         struct btrfs_key key;
 775         struct btrfs_fs_info *info = root->fs_info;
 776         struct btrfs_root *extent_root = info->extent_root;
 777         struct extent_buffer *leaf;
 778         int ret;
 779         struct btrfs_extent_item *ei;
 780         u32 refs;
 781
 782         key.objectid = blocknr;
 783         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 784         key.offset = num_blocks;
 785
 786         path = btrfs_alloc_path();
 787         if (!path)
 788                 return -ENOMEM;
 789
 790         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 791         if (ret < 0)
 792                 return ret;
 793         BUG_ON(ret);
 794
 795         leaf = path->nodes[0];
 796         ei = btrfs_item_ptr(leaf, path->slots[0],
 797                             struct btrfs_extent_item);
 798         refs = btrfs_extent_refs(leaf, ei);
 799         BUG_ON(refs == 0);
 800         refs -= 1;
 801         btrfs_set_extent_refs(leaf, ei, refs);
 802         btrfs_mark_buffer_dirty(leaf);
 803
 804         if (refs == 0) {
 805                 u64 super_blocks_used, root_blocks_used;
 806
 807                 if (pin) {
 808                         ret = pin_down_block(root, blocknr, 0);
 809                         BUG_ON(ret);
 810                 }
 811
 812                 /* block accounting for super block */
 813                 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
 814                 btrfs_set_super_blocks_used(&info->super_copy,
 815                                             super_blocks_used - num_blocks);
 816
 817                 /* block accounting for root item */
 818                 root_blocks_used = btrfs_root_used(&root->root_item);
 819                 btrfs_set_root_used(&root->root_item,
 820                                            root_blocks_used - num_blocks);
 821
 822                 ret = btrfs_del_item(trans, extent_root, path);
 823                 if (ret) {
 824                         return ret;
 825                 }
 826                 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
 827                                          mark_free, 0);
 828                 BUG_ON(ret);
 829         }
 830         btrfs_free_path(path);
 831         finish_current_insert(trans, extent_root);
 832         return ret;
 833 }
 834
 835 /*
 836  * find all the blocks marked as pending in the radix tree and remove
 837  * them from the extent map
 838  */
 839 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 840                                btrfs_root *extent_root)
 841 {
 842         int ret;
 843         int wret;
 844         int err = 0;
 845         unsigned long gang[4];
 846         int i;
 847         struct radix_tree_root *pending_radix;
 848         struct radix_tree_root *pinned_radix;
 849         struct btrfs_block_group_cache *cache;
 850
 851         pending_radix = &extent_root->fs_info->pending_del_radix;
 852         pinned_radix = &extent_root->fs_info->pinned_radix;
 853
 854         while(1) {
 855                 ret = find_first_radix_bit(pending_radix, gang, 0,
 856                                            ARRAY_SIZE(gang));
 857                 if (!ret)
 858                         break;
 859                 for (i = 0; i < ret; i++) {
 860                         wret = set_radix_bit(pinned_radix, gang[i]);
 861                         if (wret == 0) {
 862                                 cache =
 863                                   btrfs_lookup_block_group(extent_root->fs_info,
 864                                                            gang[i]);
 865                                 if (cache)
 866                                         cache->pinned++;
 867                         }
 868                         if (wret < 0) {
 869                                 printk(KERN_CRIT "set_radix_bit, err %d\n",
 870                                        wret);
 871                                 BUG_ON(wret < 0);
 872                         }
 873                         wret = clear_radix_bit(pending_radix, gang[i]);
 874                         BUG_ON(wret);
 875                         wret = __free_extent(trans, extent_root,
 876                                              gang[i], 1, 0, 0);
 877                         if (wret)
 878                                 err = wret;
 879                 }
 880         }
 881         return err;
 882 }
 883
 884 /*
 885  * remove an extent from the root, returns 0 on success
 886  */
 887 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 888                       *root, u64 blocknr, u64 num_blocks, int pin)
 889 {
 890         struct btrfs_root *extent_root = root->fs_info->extent_root;
 891         int pending_ret;
 892         int ret;
 893
 894         if (root == extent_root) {
 895                 pin_down_block(root, blocknr, 1);
 896                 return 0;
 897         }
 898         ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
 899         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
 900         return ret ? ret : pending_ret;
 901 }
 902
 903 /*
 904  * walks the btree of allocated extents and find a hole of a given size.
 905  * The key ins is changed to record the hole:
 906  * ins->objectid == block start
 907  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 908  * ins->offset == number of blocks
 909  * Any available blocks before search_start are skipped.
 910  */
 911 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 912                             *orig_root, u64 num_blocks, u64 empty_size,
 913                             u64 search_start, u64 search_end, u64 hint_block,
 914                             struct btrfs_key *ins, u64 exclude_start,
 915                             u64 exclude_nr, int data)
 916 {
 917         struct btrfs_path *path;
 918         struct btrfs_key key;
 919         int ret;
 920         u64 hole_size = 0;
 921         int slot = 0;
 922         u64 last_block = 0;
 923         u64 test_block;
 924         u64 orig_search_start = search_start;
 925         int start_found;
 926         struct extent_buffer *l;
 927         struct btrfs_root * root = orig_root->fs_info->extent_root;
 928         struct btrfs_fs_info *info = root->fs_info;
 929         int total_needed = num_blocks;
 930         int level;
 931         struct btrfs_block_group_cache *block_group;
 932         int full_scan = 0;
 933         int wrapped = 0;
 934         u64 cached_search_start = 0;
 935
 936         WARN_ON(num_blocks < 1);
 937         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 938
 939         level = btrfs_header_level(root->node);
 940
 941         if (search_end == (u64)-1)
 942                 search_end = btrfs_super_total_blocks(&info->super_copy);
 943         if (hint_block) {
 944                 block_group = btrfs_lookup_block_group(info, hint_block);
 945                 block_group = btrfs_find_block_group(root, block_group,
 946                                                      hint_block, data, 1);
 947         } else {
 948                 block_group = btrfs_find_block_group(root,
 949                                                      trans->block_group, 0,
 950                                                      data, 1);
 951         }
 952
 953         total_needed += empty_size;
 954         path = btrfs_alloc_path();
 955
 956 check_failed:
 957         search_start = find_search_start(root, &block_group,
 958                                          search_start, total_needed, data);
 959         cached_search_start = search_start;
 960
 961         btrfs_init_path(path);
 962         ins->objectid = search_start;
 963         ins->offset = 0;
 964         start_found = 0;
 965         path->reada = 2;
 966
 967         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 968         if (ret < 0)
 969                 goto error;
 970
 971         if (path->slots[0] > 0) {
 972                 path->slots[0]--;
 973         }
 974
 975         l = path->nodes[0];
 976         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 977
 978         /*
 979          * a rare case, go back one key if we hit a block group item
 980          * instead of an extent item
 981          */
 982         if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
 983             key.objectid + key.offset >= search_start) {
 984                 ins->objectid = key.objectid;
 985                 ins->offset = key.offset - 1;
 986                 btrfs_release_path(root, path);
 987                 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 988                 if (ret < 0)
 989                         goto error;
 990
 991                 if (path->slots[0] > 0) {
 992                         path->slots[0]--;
 993                 }
 994         }
 995
 996         while (1) {
 997                 l = path->nodes[0];
 998                 slot = path->slots[0];
 999                 if (slot >= btrfs_header_nritems(l)) {
1000                         ret = btrfs_next_leaf(root, path);
1001                         if (ret == 0)
1002                                 continue;
1003                         if (ret < 0)
1004                                 goto error;
1005                         if (!start_found) {
1006                                 ins->objectid = search_start;
1007                                 ins->offset = search_end - search_start;
1008                                 start_found = 1;
1009                                 goto check_pending;
1010                         }
1011                         ins->objectid = last_block > search_start ?
1012                                         last_block : search_start;
1013                         ins->offset = search_end - ins->objectid;
1014                         goto check_pending;
1015                 }
1016                 btrfs_item_key_to_cpu(l, &key, slot);
1017
1018                 if (key.objectid >= search_start && key.objectid > last_block &&
1019                     start_found) {
1020                         if (last_block < search_start)
1021                                 last_block = search_start;
1022                         hole_size = key.objectid - last_block;
1023                         if (hole_size >= num_blocks) {
1024                                 ins->objectid = last_block;
1025                                 ins->offset = hole_size;
1026                                 goto check_pending;
1027                         }
1028                 }
1029                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1030                         if (!start_found) {
1031                                 last_block = key.objectid;
1032                                 start_found = 1;
1033                         }
1034                         goto next;
1035                 }
1036
1037
1038                 start_found = 1;
1039                 last_block = key.objectid + key.offset;
1040
1041                 if (!full_scan && last_block >= block_group->key.objectid +
1042                     block_group->key.offset) {
1043                         btrfs_release_path(root, path);
1044                         search_start = block_group->key.objectid +
1045                                 block_group->key.offset * 2;
1046                         goto new_group;
1047                 }
1048 next:
1049                 path->slots[0]++;
1050                 cond_resched();
1051         }
1052 check_pending:
1053         /* we have to make sure we didn't find an extent that has already
1054          * been allocated by the map tree or the original allocation
1055          */
1056         btrfs_release_path(root, path);
1057         BUG_ON(ins->objectid < search_start);
1058
1059         if (ins->objectid + num_blocks >= search_end)
1060                 goto enospc;
1061
1062         for (test_block = ins->objectid;
1063              test_block < ins->objectid + num_blocks; test_block++) {
1064                 if (test_radix_bit(&info->pinned_radix, test_block) ||
1065                     test_radix_bit(&info->extent_ins_radix, test_block)) {
1066                         search_start = test_block + 1;
1067                         goto new_group;
1068                 }
1069         }
1070         if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1071             ins->objectid < exclude_start + exclude_nr)) {
1072                 search_start = exclude_start + exclude_nr;
1073                 goto new_group;
1074         }
1075         if (!data) {
1076                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1077                 if (block_group)
1078                         trans->block_group = block_group;
1079         }
1080         ins->offset = num_blocks;
1081         btrfs_free_path(path);
1082         return 0;
1083
1084 new_group:
1085         if (search_start + num_blocks >= search_end) {
1086 enospc:
1087                 search_start = orig_search_start;
1088                 if (full_scan) {
1089                         ret = -ENOSPC;
1090                         goto error;
1091                 }
1092                 if (wrapped) {
1093                         if (!full_scan)
1094                                 total_needed -= empty_size;
1095                         full_scan = 1;
1096                 } else
1097                         wrapped = 1;
1098         }
1099         block_group = btrfs_lookup_block_group(info, search_start);
1100         cond_resched();
1101         if (!full_scan)
1102                 block_group = btrfs_find_block_group(root, block_group,
1103                                                      search_start, data, 0);
1104         goto check_failed;
1105
1106 error:
1107         btrfs_release_path(root, path);
1108         btrfs_free_path(path);
1109         return ret;
1110 }
1111 /*
1112  * finds a free extent and does all the dirty work required for allocation
1113  * returns the key for the extent through ins, and a tree buffer for
1114  * the first block of the extent through buf.
1115  *
1116  * returns 0 if everything worked, non-zero otherwise.
1117  */
1118 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1119                        struct btrfs_root *root, u64 owner,
1120                        u64 num_blocks, u64 empty_size, u64 hint_block,
1121                        u64 search_end, struct btrfs_key *ins, int data)
1122 {
1123         int ret;
1124         int pending_ret;
1125         u64 super_blocks_used, root_blocks_used;
1126         u64 search_start = 0;
1127         struct btrfs_fs_info *info = root->fs_info;
1128         struct btrfs_root *extent_root = info->extent_root;
1129         struct btrfs_extent_item extent_item;
1130
1131         btrfs_set_stack_extent_refs(&extent_item, 1);
1132         btrfs_set_stack_extent_owner(&extent_item, owner);
1133
1134         WARN_ON(num_blocks < 1);
1135         ret = find_free_extent(trans, root, num_blocks, empty_size,
1136                                search_start, search_end, hint_block, ins,
1137                                trans->alloc_exclude_start,
1138                                trans->alloc_exclude_nr, data);
1139         BUG_ON(ret);
1140         if (ret)
1141                 return ret;
1142
1143         /* block accounting for super block */
1144         super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1145         btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1146                                     num_blocks);
1147
1148         /* block accounting for root item */
1149         root_blocks_used = btrfs_root_used(&root->root_item);
1150         btrfs_set_root_used(&root->root_item, root_blocks_used +
1151                                    num_blocks);
1152
1153         clear_extent_dirty(&root->fs_info->free_space_cache,
1154                            ins->objectid, ins->objectid + ins->offset - 1,
1155                            GFP_NOFS);
1156
1157         if (root == extent_root) {
1158                 BUG_ON(num_blocks != 1);
1159                 set_radix_bit(&root->fs_info->extent_ins_radix, ins->objectid);
1160                 goto update_block;
1161         }
1162
1163         WARN_ON(trans->alloc_exclude_nr);
1164         trans->alloc_exclude_start = ins->objectid;
1165         trans->alloc_exclude_nr = ins->offset;
1166         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1167                                 sizeof(extent_item));
1168
1169         trans->alloc_exclude_start = 0;
1170         trans->alloc_exclude_nr = 0;
1171
1172         BUG_ON(ret);
1173         finish_current_insert(trans, extent_root);
1174         pending_ret = del_pending_extents(trans, extent_root);
1175
1176         if (ret) {
1177                 return ret;
1178         }
1179         if (pending_ret) {
1180                 return pending_ret;
1181         }
1182
1183 update_block:
1184         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1185                                  data);
1186         BUG_ON(ret);
1187         return 0;
1188 }
1189
1190 /*
1191  * helper function to allocate a block for a given tree
1192  * returns the tree buffer or NULL.
1193  */
1194 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1195                                              struct btrfs_root *root, u64 hint,
1196                                              u64 empty_size)
1197 {
1198         struct btrfs_key ins;
1199         int ret;
1200         struct extent_buffer *buf;
1201
1202         ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1203                                  1, empty_size, hint, (u64)-1, &ins, 0);
1204         if (ret) {
1205                 BUG_ON(ret > 0);
1206                 return ERR_PTR(ret);
1207         }
1208         buf = btrfs_find_create_tree_block(root, ins.objectid);
1209         if (!buf) {
1210                 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1211                 return ERR_PTR(-ENOMEM);
1212         }
1213         btrfs_set_buffer_uptodate(buf);
1214         buf->alloc_addr = (unsigned long)__builtin_return_address(0);
1215         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1216                          buf->start + buf->len - 1, GFP_NOFS);
1217         /*
1218         set_buffer_checked(buf);
1219         set_buffer_defrag(buf);
1220         */
1221         /* FIXME!!!!!!!!!!!!!!!!
1222         set_radix_bit(&trans->transaction->dirty_pages, buf->pages[0]->index);
1223         */
1224         trans->blocks_used++;
1225         return buf;
1226 }
1227
1228 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1229                          struct btrfs_root *root, struct extent_buffer *leaf)
1230 {
1231         struct btrfs_key key;
1232         struct btrfs_file_extent_item *fi;
1233         int i;
1234         int nritems;
1235         int ret;
1236
1237         BUG_ON(!btrfs_is_leaf(leaf));
1238         nritems = btrfs_header_nritems(leaf);
1239         for (i = 0; i < nritems; i++) {
1240                 u64 disk_blocknr;
1241
1242                 btrfs_item_key_to_cpu(leaf, &key, i);
1243                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1244                         continue;
1245                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1246                 if (btrfs_file_extent_type(leaf, fi) ==
1247                     BTRFS_FILE_EXTENT_INLINE)
1248                         continue;
1249                 /*
1250                  * FIXME make sure to insert a trans record that
1251                  * repeats the snapshot del on crash
1252                  */
1253                 disk_blocknr = btrfs_file_extent_disk_blocknr(leaf, fi);
1254                 if (disk_blocknr == 0)
1255                         continue;
1256                 ret = btrfs_free_extent(trans, root, disk_blocknr,
1257                                 btrfs_file_extent_disk_num_blocks(leaf, fi), 0);
1258                 BUG_ON(ret);
1259         }
1260         return 0;
1261 }
1262
1263 static void reada_walk_down(struct btrfs_root *root,
1264                             struct extent_buffer *node)
1265 {
1266         int i;
1267         u32 nritems;
1268         u64 blocknr;
1269         int ret;
1270         u32 refs;
1271
1272         nritems = btrfs_header_nritems(node);
1273         for (i = 0; i < nritems; i++) {
1274                 blocknr = btrfs_node_blockptr(node, i);
1275                 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1276                 BUG_ON(ret);
1277                 if (refs != 1)
1278                         continue;
1279                 mutex_unlock(&root->fs_info->fs_mutex);
1280                 ret = readahead_tree_block(root, blocknr);
1281                 cond_resched();
1282                 mutex_lock(&root->fs_info->fs_mutex);
1283                 if (ret)
1284                         break;
1285         }
1286 }
1287
1288 /*
1289  * helper function for drop_snapshot, this walks down the tree dropping ref
1290  * counts as it goes.
1291  */
1292 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1293                           *root, struct btrfs_path *path, int *level)
1294 {
1295         struct extent_buffer *next;
1296         struct extent_buffer *cur;
1297         u64 blocknr;
1298         int ret;
1299         u32 refs;
1300
1301         WARN_ON(*level < 0);
1302         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1303         ret = lookup_extent_ref(trans, root,
1304                                 extent_buffer_blocknr(path->nodes[*level]),
1305                                 1, &refs);
1306         BUG_ON(ret);
1307         if (refs > 1)
1308                 goto out;
1309
1310         /*
1311          * walk down to the last node level and free all the leaves
1312          */
1313         while(*level >= 0) {
1314                 WARN_ON(*level < 0);
1315                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1316                 cur = path->nodes[*level];
1317
1318                 if (*level > 0 && path->slots[*level] == 0)
1319                         reada_walk_down(root, cur);
1320
1321                 if (btrfs_header_level(cur) != *level)
1322                         WARN_ON(1);
1323
1324                 if (path->slots[*level] >=
1325                     btrfs_header_nritems(cur))
1326                         break;
1327                 if (*level == 0) {
1328                         ret = drop_leaf_ref(trans, root, cur);
1329                         BUG_ON(ret);
1330                         break;
1331                 }
1332                 blocknr = btrfs_node_blockptr(cur, path->slots[*level]);
1333                 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1334                 BUG_ON(ret);
1335                 if (refs != 1) {
1336                         path->slots[*level]++;
1337                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1338                         BUG_ON(ret);
1339                         continue;
1340                 }
1341                 next = btrfs_find_tree_block(root, blocknr);
1342                 if (!next || !btrfs_buffer_uptodate(next)) {
1343                         free_extent_buffer(next);
1344                         mutex_unlock(&root->fs_info->fs_mutex);
1345                         next = read_tree_block(root, blocknr);
1346                         mutex_lock(&root->fs_info->fs_mutex);
1347
1348                         /* we dropped the lock, check one more time */
1349                         ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1350                         BUG_ON(ret);
1351                         if (refs != 1) {
1352                                 path->slots[*level]++;
1353                                 free_extent_buffer(next);
1354                                 ret = btrfs_free_extent(trans, root,
1355                                                         blocknr, 1, 1);
1356                                 BUG_ON(ret);
1357                                 continue;
1358                         }
1359                 }
1360                 WARN_ON(*level <= 0);
1361                 if (path->nodes[*level-1])
1362                         free_extent_buffer(path->nodes[*level-1]);
1363                 path->nodes[*level-1] = next;
1364                 *level = btrfs_header_level(next);
1365                 path->slots[*level] = 0;
1366         }
1367 out:
1368         WARN_ON(*level < 0);
1369         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1370         ret = btrfs_free_extent(trans, root,
1371                         extent_buffer_blocknr(path->nodes[*level]), 1, 1);
1372         free_extent_buffer(path->nodes[*level]);
1373         path->nodes[*level] = NULL;
1374         *level += 1;
1375         BUG_ON(ret);
1376         return 0;
1377 }
1378
1379 /*
1380  * helper for dropping snapshots.  This walks back up the tree in the path
1381  * to find the first node higher up where we haven't yet gone through
1382  * all the slots
1383  */
1384 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1385                         *root, struct btrfs_path *path, int *level)
1386 {
1387         int i;
1388         int slot;
1389         int ret;
1390         struct btrfs_root_item *root_item = &root->root_item;
1391
1392         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1393                 slot = path->slots[i];
1394                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1395                         struct extent_buffer *node;
1396                         struct btrfs_disk_key disk_key;
1397                         node = path->nodes[i];
1398                         path->slots[i]++;
1399                         *level = i;
1400                         WARN_ON(*level == 0);
1401                         btrfs_node_key(node, &disk_key, path->slots[i]);
1402                         memcpy(&root_item->drop_progress,
1403                                &disk_key, sizeof(disk_key));
1404                         root_item->drop_level = i;
1405                         return 0;
1406                 } else {
1407                         ret = btrfs_free_extent(trans, root,
1408                                     extent_buffer_blocknr(path->nodes[*level]),
1409                                     1, 1);
1410                         BUG_ON(ret);
1411                         free_extent_buffer(path->nodes[*level]);
1412                         path->nodes[*level] = NULL;
1413                         *level = i + 1;
1414                 }
1415         }
1416         return 1;
1417 }
1418
1419 /*
1420  * drop the reference count on the tree rooted at 'snap'.  This traverses
1421  * the tree freeing any blocks that have a ref count of zero after being
1422  * decremented.
1423  */
1424 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1425                         *root)
1426 {
1427         int ret = 0;
1428         int wret;
1429         int level;
1430         struct btrfs_path *path;
1431         int i;
1432         int orig_level;
1433         struct btrfs_root_item *root_item = &root->root_item;
1434
1435         path = btrfs_alloc_path();
1436         BUG_ON(!path);
1437
1438         level = btrfs_header_level(root->node);
1439         orig_level = level;
1440         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1441                 path->nodes[level] = root->node;
1442                 extent_buffer_get(root->node);
1443                 path->slots[level] = 0;
1444         } else {
1445                 struct btrfs_key key;
1446                 struct btrfs_disk_key found_key;
1447                 struct extent_buffer *node;
1448
1449                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1450                 level = root_item->drop_level;
1451                 path->lowest_level = level;
1452                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1453                 if (wret < 0) {
1454                         ret = wret;
1455                         goto out;
1456                 }
1457                 node = path->nodes[level];
1458                 btrfs_node_key(node, &found_key, path->slots[level]);
1459                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1460                                sizeof(found_key)));
1461         }
1462         while(1) {
1463                 wret = walk_down_tree(trans, root, path, &level);
1464                 if (wret > 0)
1465                         break;
1466                 if (wret < 0)
1467                         ret = wret;
1468
1469                 wret = walk_up_tree(trans, root, path, &level);
1470                 if (wret > 0)
1471                         break;
1472                 if (wret < 0)
1473                         ret = wret;
1474                 ret = -EAGAIN;
1475                 break;
1476         }
1477         for (i = 0; i <= orig_level; i++) {
1478                 if (path->nodes[i]) {
1479                         free_extent_buffer(path->nodes[i]);
1480                         path->nodes[i] = 0;
1481                 }
1482         }
1483 out:
1484         btrfs_free_path(path);
1485         return ret;
1486 }
1487
1488 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1489 {
1490         u64 start;
1491         u64 end;
1492         int ret;
1493
1494         while(1) {
1495                 ret = find_first_extent_bit(&info->block_group_cache, 0,
1496                                             &start, &end, (unsigned int)-1);
1497                 if (ret)
1498                         break;
1499                 clear_extent_bits(&info->block_group_cache, start,
1500                                   end, (unsigned int)-1, GFP_NOFS);
1501         }
1502         while(1) {
1503                 ret = find_first_extent_bit(&info->free_space_cache, 0,
1504                                             &start, &end, EXTENT_DIRTY);
1505                 if (ret)
1506                         break;
1507                 clear_extent_dirty(&info->free_space_cache, start,
1508                                    end, GFP_NOFS);
1509         }
1510         return 0;
1511 }
1512
1513 int btrfs_read_block_groups(struct btrfs_root *root)
1514 {
1515         struct btrfs_path *path;
1516         int ret;
1517         int err = 0;
1518         int bit;
1519         struct btrfs_block_group_cache *cache;
1520         struct btrfs_fs_info *info = root->fs_info;
1521         struct extent_map_tree *block_group_cache;
1522         struct btrfs_key key;
1523         struct btrfs_key found_key;
1524         struct extent_buffer *leaf;
1525         u64 group_size_blocks;
1526
1527         block_group_cache = &info->block_group_cache;
1528
1529         group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1530                 info->sb->s_blocksize_bits;
1531
1532         root = info->extent_root;
1533         key.objectid = 0;
1534         key.offset = group_size_blocks;
1535         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1536
1537         path = btrfs_alloc_path();
1538         if (!path)
1539                 return -ENOMEM;
1540
1541         while(1) {
1542                 ret = btrfs_search_slot(NULL, info->extent_root,
1543                                         &key, path, 0, 0);
1544                 if (ret != 0) {
1545                         err = ret;
1546                         break;
1547                 }
1548                 leaf = path->nodes[0];
1549                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1550                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1551                 if (!cache) {
1552                         err = -1;
1553                         break;
1554                 }
1555
1556                 read_extent_buffer(leaf, &cache->item,
1557                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
1558                                    sizeof(cache->item));
1559                 memcpy(&cache->key, &found_key, sizeof(found_key));
1560                 cache->last_alloc = cache->key.objectid;
1561                 cache->first_free = cache->key.objectid;
1562                 cache->pinned = 0;
1563                 cache->cached = 0;
1564
1565                 key.objectid = found_key.objectid + found_key.offset;
1566                 btrfs_release_path(root, path);
1567
1568                 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
1569                         bit = BLOCK_GROUP_DATA;
1570                         cache->data = 1;
1571                 } else {
1572                         bit = BLOCK_GROUP_METADATA;
1573                         cache->data = 0;
1574                 }
1575
1576                 /* use EXTENT_LOCKED to prevent merging */
1577                 set_extent_bits(block_group_cache, found_key.objectid,
1578                                 found_key.objectid + found_key.offset - 1,
1579                                 bit | EXTENT_LOCKED, GFP_NOFS);
1580                 set_state_private(block_group_cache, found_key.objectid,
1581                                   (u64)cache);
1582
1583                 if (key.objectid >=
1584                     btrfs_super_total_blocks(&info->super_copy))
1585                         break;
1586         }
1587
1588         btrfs_free_path(path);
1589         return 0;
1590 }