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 bytenr)
 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                                     bytenr, &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 *)(unsigned long)ptr;
 153
 154
 155         if (block_group->key.objectid <= bytenr && bytenr <=
 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;
 169         u64 start = 0;
 170         u64 end = 0;
 171         int wrapped = 0;
 172
 173 again:
 174         ret = cache_block_group(root, cache);
 175         if (ret)
 176                 goto out;
 177
 178         last = max(search_start, cache->key.objectid);
 179
 180         while(1) {
 181                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 182                                             last, &start, &end, EXTENT_DIRTY);
 183                 if (ret) {
 184                         goto new_group;
 185                 }
 186
 187                 start = max(last, start);
 188                 last = end + 1;
 189                 if (end + 1 - start < num)
 190                         continue;
 191                 if (start + num >= cache->key.objectid + cache->key.offset)
 192                         goto new_group;
 193                 return start;
 194         }
 195 out:
 196         return search_start;
 197
 198 new_group:
 199         last = cache->key.objectid + cache->key.offset;
 200 wrapped:
 201         cache = btrfs_lookup_block_group(root->fs_info, last);
 202         if (!cache) {
 203                 if (!wrapped) {
 204                         wrapped = 1;
 205                         last = search_start;
 206                         data = BTRFS_BLOCK_GROUP_MIXED;
 207                         goto wrapped;
 208                 }
 209                 return search_start;
 210         }
 211         cache = btrfs_find_block_group(root, cache, last, data, 0);
 212         *cache_ret = cache;
 213         goto again;
 214 }
 215
 216 static u64 div_factor(u64 num, int factor)
 217 {
 218         num *= factor;
 219         do_div(num, 10);
 220         return num;
 221 }
 222
 223 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 224                                                  struct btrfs_block_group_cache
 225                                                  *hint, u64 search_start,
 226                                                  int data, int owner)
 227 {
 228         struct btrfs_block_group_cache *cache;
 229         struct extent_map_tree *block_group_cache;
 230         struct btrfs_block_group_cache *found_group = NULL;
 231         struct btrfs_fs_info *info = root->fs_info;
 232         u64 used;
 233         u64 last = 0;
 234         u64 hint_last;
 235         u64 start;
 236         u64 end;
 237         u64 free_check;
 238         u64 ptr;
 239         int bit;
 240         int ret;
 241         int full_search = 0;
 242         int factor = 8;
 243         int data_swap = 0;
 244
 245         block_group_cache = &info->block_group_cache;
 246
 247         if (!owner)
 248                 factor = 8;
 249
 250         if (data == BTRFS_BLOCK_GROUP_MIXED)
 251                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 252         else if (data)
 253                 bit = BLOCK_GROUP_DATA;
 254         else
 255                 bit = BLOCK_GROUP_METADATA;
 256
 257         if (search_start) {
 258                 struct btrfs_block_group_cache *shint;
 259                 shint = btrfs_lookup_block_group(info, search_start);
 260                 if (shint && (shint->data == data ||
 261                               shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 262                         used = btrfs_block_group_used(&shint->item);
 263                         if (used < div_factor(shint->key.offset, factor)) {
 264                                 return shint;
 265                         }
 266                 }
 267         }
 268         if (hint && (hint->data == data ||
 269                      hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
 270                 used = btrfs_block_group_used(&hint->item);
 271                 if (used < div_factor(hint->key.offset, factor)) {
 272                         return hint;
 273                 }
 274                 last = hint->key.objectid + hint->key.offset;
 275                 hint_last = last;
 276         } else {
 277                 if (hint)
 278                         hint_last = max(hint->key.objectid, search_start);
 279                 else
 280                         hint_last = search_start;
 281
 282                 last = hint_last;
 283         }
 284 again:
 285         while(1) {
 286                 ret = find_first_extent_bit(block_group_cache, last,
 287                                             &start, &end, bit);
 288                 if (ret)
 289                         break;
 290
 291                 ret = get_state_private(block_group_cache, start, &ptr);
 292                 if (ret)
 293                         break;
 294
 295                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 296                 last = cache->key.objectid + cache->key.offset;
 297                 used = btrfs_block_group_used(&cache->item);
 298
 299                 if (full_search)
 300                         free_check = cache->key.offset;
 301                 else
 302                         free_check = div_factor(cache->key.offset, factor);
 303
 304                 if (used < free_check) {
 305                         found_group = cache;
 306                         goto found;
 307                 }
 308                 cond_resched();
 309         }
 310         if (!full_search) {
 311                 last = search_start;
 312                 full_search = 1;
 313                 goto again;
 314         }
 315         if (!data_swap) {
 316                 data_swap = 1;
 317                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
 318                 last = search_start;
 319                 goto again;
 320         }
 321 found:
 322         return found_group;
 323 }
 324
 325 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 326                                 struct btrfs_root *root,
 327                                 u64 bytenr, u64 num_bytes)
 328 {
 329         struct btrfs_path *path;
 330         int ret;
 331         struct btrfs_key key;
 332         struct extent_buffer *l;
 333         struct btrfs_extent_item *item;
 334         u32 refs;
 335
 336         WARN_ON(num_bytes < root->sectorsize);
 337         path = btrfs_alloc_path();
 338         if (!path)
 339                 return -ENOMEM;
 340
 341         key.objectid = bytenr;
 342         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 343         key.offset = num_bytes;
 344         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 345                                 0, 1);
 346         if (ret < 0)
 347                 return ret;
 348         if (ret != 0) {
 349                 BUG();
 350         }
 351         BUG_ON(ret != 0);
 352         l = path->nodes[0];
 353         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 354         refs = btrfs_extent_refs(l, item);
 355         btrfs_set_extent_refs(l, item, refs + 1);
 356         btrfs_mark_buffer_dirty(path->nodes[0]);
 357
 358         btrfs_release_path(root->fs_info->extent_root, path);
 359         btrfs_free_path(path);
 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 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 366                          struct btrfs_root *root)
 367 {
 368         finish_current_insert(trans, root->fs_info->extent_root);
 369         del_pending_extents(trans, root->fs_info->extent_root);
 370         return 0;
 371 }
 372
 373 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 374                              struct btrfs_root *root, u64 bytenr,
 375                              u64 num_bytes, u32 *refs)
 376 {
 377         struct btrfs_path *path;
 378         int ret;
 379         struct btrfs_key key;
 380         struct extent_buffer *l;
 381         struct btrfs_extent_item *item;
 382
 383         WARN_ON(num_bytes < root->sectorsize);
 384         path = btrfs_alloc_path();
 385         key.objectid = bytenr;
 386         key.offset = num_bytes;
 387         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 388         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 389                                 0, 0);
 390         if (ret < 0)
 391                 goto out;
 392         if (ret != 0) {
 393                 btrfs_print_leaf(root, path->nodes[0]);
 394                 printk("failed to find block number %Lu\n", bytenr);
 395                 BUG();
 396         }
 397         l = path->nodes[0];
 398         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 399         *refs = btrfs_extent_refs(l, item);
 400 out:
 401         btrfs_free_path(path);
 402         return 0;
 403 }
 404
 405 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 406                        struct btrfs_root *root)
 407 {
 408         return btrfs_inc_extent_ref(trans, root, root->node->start,
 409                                     root->node->len);
 410 }
 411
 412 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 413                   struct extent_buffer *buf)
 414 {
 415         u64 bytenr;
 416         u32 nritems;
 417         struct btrfs_key key;
 418         struct btrfs_file_extent_item *fi;
 419         int i;
 420         int level;
 421         int ret;
 422         int faili;
 423         int err;
 424
 425         if (!root->ref_cows)
 426                 return 0;
 427
 428         level = btrfs_header_level(buf);
 429         nritems = btrfs_header_nritems(buf);
 430         for (i = 0; i < nritems; i++) {
 431                 if (level == 0) {
 432                         u64 disk_bytenr;
 433                         btrfs_item_key_to_cpu(buf, &key, i);
 434                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 435                                 continue;
 436                         fi = btrfs_item_ptr(buf, i,
 437                                             struct btrfs_file_extent_item);
 438                         if (btrfs_file_extent_type(buf, fi) ==
 439                             BTRFS_FILE_EXTENT_INLINE)
 440                                 continue;
 441                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 442                         if (disk_bytenr == 0)
 443                                 continue;
 444                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
 445                                     btrfs_file_extent_disk_num_bytes(buf, fi));
 446                         if (ret) {
 447                                 faili = i;
 448                                 goto fail;
 449                         }
 450                 } else {
 451                         bytenr = btrfs_node_blockptr(buf, i);
 452                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
 453                                            btrfs_level_size(root, level - 1));
 454                         if (ret) {
 455                                 faili = i;
 456                                 goto fail;
 457                         }
 458                 }
 459         }
 460         return 0;
 461 fail:
 462         WARN_ON(1);
 463         for (i =0; i < faili; i++) {
 464                 if (level == 0) {
 465                         u64 disk_bytenr;
 466                         btrfs_item_key_to_cpu(buf, &key, i);
 467                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 468                                 continue;
 469                         fi = btrfs_item_ptr(buf, i,
 470                                             struct btrfs_file_extent_item);
 471                         if (btrfs_file_extent_type(buf, fi) ==
 472                             BTRFS_FILE_EXTENT_INLINE)
 473                                 continue;
 474                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 475                         if (disk_bytenr == 0)
 476                                 continue;
 477                         err = btrfs_free_extent(trans, root, disk_bytenr,
 478                                     btrfs_file_extent_disk_num_bytes(buf,
 479                                                                       fi), 0);
 480                         BUG_ON(err);
 481                 } else {
 482                         bytenr = btrfs_node_blockptr(buf, i);
 483                         err = btrfs_free_extent(trans, root, bytenr,
 484                                         btrfs_level_size(root, level - 1), 0);
 485                         BUG_ON(err);
 486                 }
 487         }
 488         return ret;
 489 }
 490
 491 static int write_one_cache_group(struct btrfs_trans_handle *trans,
 492                                  struct btrfs_root *root,
 493                                  struct btrfs_path *path,
 494                                  struct btrfs_block_group_cache *cache)
 495 {
 496         int ret;
 497         int pending_ret;
 498         struct btrfs_root *extent_root = root->fs_info->extent_root;
 499         unsigned long bi;
 500         struct extent_buffer *leaf;
 501
 502         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 503         if (ret < 0)
 504                 goto fail;
 505         BUG_ON(ret);
 506
 507         leaf = path->nodes[0];
 508         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
 509         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
 510         btrfs_mark_buffer_dirty(leaf);
 511         btrfs_release_path(extent_root, path);
 512 fail:
 513         finish_current_insert(trans, extent_root);
 514         pending_ret = del_pending_extents(trans, extent_root);
 515         if (ret)
 516                 return ret;
 517         if (pending_ret)
 518                 return pending_ret;
 519         return 0;
 520
 521 }
 522
 523 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 524                                    struct btrfs_root *root)
 525 {
 526         struct extent_map_tree *block_group_cache;
 527         struct btrfs_block_group_cache *cache;
 528         int ret;
 529         int err = 0;
 530         int werr = 0;
 531         struct btrfs_path *path;
 532         u64 last = 0;
 533         u64 start;
 534         u64 end;
 535         u64 ptr;
 536
 537         block_group_cache = &root->fs_info->block_group_cache;
 538         path = btrfs_alloc_path();
 539         if (!path)
 540                 return -ENOMEM;
 541
 542         while(1) {
 543                 ret = find_first_extent_bit(block_group_cache, last,
 544                                             &start, &end, BLOCK_GROUP_DIRTY);
 545                 if (ret)
 546                         break;
 547
 548                 last = end + 1;
 549                 ret = get_state_private(block_group_cache, start, &ptr);
 550                 if (ret)
 551                         break;
 552
 553                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 554                 err = write_one_cache_group(trans, root,
 555                                             path, cache);
 556                 /*
 557                  * if we fail to write the cache group, we want
 558                  * to keep it marked dirty in hopes that a later
 559                  * write will work
 560                  */
 561                 if (err) {
 562                         werr = err;
 563                         continue;
 564                 }
 565                 clear_extent_bits(block_group_cache, start, end,
 566                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
 567         }
 568         btrfs_free_path(path);
 569         return werr;
 570 }
 571
 572 static int update_block_group(struct btrfs_trans_handle *trans,
 573                               struct btrfs_root *root,
 574                               u64 bytenr, u64 num_bytes, int alloc,
 575                               int mark_free, int data)
 576 {
 577         struct btrfs_block_group_cache *cache;
 578         struct btrfs_fs_info *info = root->fs_info;
 579         u64 total = num_bytes;
 580         u64 old_val;
 581         u64 byte_in_group;
 582         u64 start;
 583         u64 end;
 584
 585         while(total) {
 586                 cache = btrfs_lookup_block_group(info, bytenr);
 587                 if (!cache) {
 588                         return -1;
 589                 }
 590                 byte_in_group = bytenr - cache->key.objectid;
 591                 WARN_ON(byte_in_group > cache->key.offset);
 592                 start = cache->key.objectid;
 593                 end = start + cache->key.offset - 1;
 594                 set_extent_bits(&info->block_group_cache, start, end,
 595                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
 596
 597                 old_val = btrfs_block_group_used(&cache->item);
 598                 num_bytes = min(total, cache->key.offset - byte_in_group);
 599                 if (alloc) {
 600                         if (cache->data != data &&
 601                             old_val < (cache->key.offset >> 1)) {
 602                                 int bit_to_clear;
 603                                 int bit_to_set;
 604                                 cache->data = data;
 605                                 if (data) {
 606                                         bit_to_clear = BLOCK_GROUP_METADATA;
 607                                         bit_to_set = BLOCK_GROUP_DATA;
 608                                         cache->item.flags &=
 609                                                 ~BTRFS_BLOCK_GROUP_MIXED;
 610                                         cache->item.flags |=
 611                                                 BTRFS_BLOCK_GROUP_DATA;
 612                                 } else {
 613                                         bit_to_clear = BLOCK_GROUP_DATA;
 614                                         bit_to_set = BLOCK_GROUP_METADATA;
 615                                         cache->item.flags &=
 616                                                 ~BTRFS_BLOCK_GROUP_MIXED;
 617                                         cache->item.flags &=
 618                                                 ~BTRFS_BLOCK_GROUP_DATA;
 619                                 }
 620                                 clear_extent_bits(&info->block_group_cache,
 621                                                   start, end, bit_to_clear,
 622                                                   GFP_NOFS);
 623                                 set_extent_bits(&info->block_group_cache,
 624                                                 start, end, bit_to_set,
 625                                                 GFP_NOFS);
 626                         } else if (cache->data != data &&
 627                                    cache->data != BTRFS_BLOCK_GROUP_MIXED) {
 628                                 cache->data = BTRFS_BLOCK_GROUP_MIXED;
 629                                 set_extent_bits(&info->block_group_cache,
 630                                                 start, end,
 631                                                 BLOCK_GROUP_DATA |
 632                                                 BLOCK_GROUP_METADATA,
 633                                                 GFP_NOFS);
 634                         }
 635                         old_val += num_bytes;
 636                 } else {
 637                         old_val -= num_bytes;
 638                         if (mark_free) {
 639                                 set_extent_dirty(&info->free_space_cache,
 640                                                  bytenr, bytenr + num_bytes - 1,
 641                                                  GFP_NOFS);
 642                         }
 643                 }
 644                 btrfs_set_block_group_used(&cache->item, old_val);
 645                 total -= num_bytes;
 646                 bytenr += num_bytes;
 647         }
 648         return 0;
 649 }
 650
 651 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
 652 {
 653         u64 last = 0;
 654         u64 start;
 655         u64 end;
 656         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
 657         int ret;
 658
 659         while(1) {
 660                 ret = find_first_extent_bit(pinned_extents, last,
 661                                             &start, &end, EXTENT_DIRTY);
 662                 if (ret)
 663                         break;
 664                 set_extent_dirty(copy, start, end, GFP_NOFS);
 665                 last = end + 1;
 666         }
 667         return 0;
 668 }
 669
 670 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 671                                struct btrfs_root *root,
 672                                struct extent_map_tree *unpin)
 673 {
 674         u64 start;
 675         u64 end;
 676         int ret;
 677         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
 678         struct extent_map_tree *free_space_cache;
 679
 680         free_space_cache = &root->fs_info->free_space_cache;
 681
 682         while(1) {
 683                 ret = find_first_extent_bit(unpin, 0, &start, &end,
 684                                             EXTENT_DIRTY);
 685                 if (ret)
 686                         break;
 687
 688                 clear_extent_dirty(pinned_extents, start, end,
 689                                    GFP_NOFS);
 690                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
 691                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
 692         }
 693         return 0;
 694 }
 695
 696 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 697                                  btrfs_root *extent_root)
 698 {
 699         struct btrfs_key ins;
 700         struct btrfs_extent_item extent_item;
 701         int ret;
 702         int err = 0;
 703         u64 start;
 704         u64 end;
 705         struct btrfs_fs_info *info = extent_root->fs_info;
 706
 707         btrfs_set_stack_extent_refs(&extent_item, 1);
 708         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
 709         btrfs_set_stack_extent_owner(&extent_item,
 710                                      extent_root->root_key.objectid);
 711
 712         while(1) {
 713                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
 714                                             &end, EXTENT_LOCKED);
 715                 if (ret)
 716                         break;
 717
 718                 ins.objectid = start;
 719                 ins.offset = end + 1 - start;
 720                 err = btrfs_insert_item(trans, extent_root, &ins,
 721                                         &extent_item, sizeof(extent_item));
 722                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
 723                                   GFP_NOFS);
 724         }
 725         return 0;
 726 }
 727
 728 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
 729                           int pending)
 730 {
 731         int err = 0;
 732         struct extent_buffer *buf;
 733
 734         if (!pending) {
 735                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
 736                 if (buf) {
 737                         if (btrfs_buffer_uptodate(buf)) {
 738                                 u64 transid =
 739                                     root->fs_info->running_transaction->transid;
 740                                 if (btrfs_header_generation(buf) == transid) {
 741                                         free_extent_buffer(buf);
 742                                         return 1;
 743                                 }
 744                         }
 745                         free_extent_buffer(buf);
 746                 }
 747                 set_extent_dirty(&root->fs_info->pinned_extents,
 748                                  bytenr, bytenr + num_bytes - 1, GFP_NOFS);
 749         } else {
 750                 set_extent_bits(&root->fs_info->pending_del,
 751                                 bytenr, bytenr + num_bytes - 1,
 752                                 EXTENT_LOCKED, GFP_NOFS);
 753         }
 754         BUG_ON(err < 0);
 755         return 0;
 756 }
 757
 758 /*
 759  * remove an extent from the root, returns 0 on success
 760  */
 761 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 762                          *root, u64 bytenr, u64 num_bytes, int pin,
 763                          int mark_free)
 764 {
 765         struct btrfs_path *path;
 766         struct btrfs_key key;
 767         struct btrfs_fs_info *info = root->fs_info;
 768         struct btrfs_root *extent_root = info->extent_root;
 769         struct extent_buffer *leaf;
 770         int ret;
 771         struct btrfs_extent_item *ei;
 772         u32 refs;
 773
 774         key.objectid = bytenr;
 775         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 776         key.offset = num_bytes;
 777
 778         path = btrfs_alloc_path();
 779         if (!path)
 780                 return -ENOMEM;
 781
 782         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 783         if (ret < 0)
 784                 return ret;
 785         BUG_ON(ret);
 786
 787         leaf = path->nodes[0];
 788         ei = btrfs_item_ptr(leaf, path->slots[0],
 789                             struct btrfs_extent_item);
 790         refs = btrfs_extent_refs(leaf, ei);
 791         BUG_ON(refs == 0);
 792         refs -= 1;
 793         btrfs_set_extent_refs(leaf, ei, refs);
 794         btrfs_mark_buffer_dirty(leaf);
 795
 796         if (refs == 0) {
 797                 u64 super_used;
 798                 u64 root_used;
 799
 800                 if (pin) {
 801                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
 802                         if (ret > 0)
 803                                 mark_free = 1;
 804                         BUG_ON(ret < 0);
 805                 }
 806
 807                 /* block accounting for super block */
 808                 super_used = btrfs_super_bytes_used(&info->super_copy);
 809                 btrfs_set_super_bytes_used(&info->super_copy,
 810                                            super_used - num_bytes);
 811
 812                 /* block accounting for root item */
 813                 root_used = btrfs_root_used(&root->root_item);
 814                 btrfs_set_root_used(&root->root_item,
 815                                            root_used - num_bytes);
 816
 817                 ret = btrfs_del_item(trans, extent_root, path);
 818                 if (ret) {
 819                         return ret;
 820                 }
 821                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
 822                                          mark_free, 0);
 823                 BUG_ON(ret);
 824         }
 825         btrfs_free_path(path);
 826         finish_current_insert(trans, extent_root);
 827         return ret;
 828 }
 829
 830 /*
 831  * find all the blocks marked as pending in the radix tree and remove
 832  * them from the extent map
 833  */
 834 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 835                                btrfs_root *extent_root)
 836 {
 837         int ret;
 838         int err = 0;
 839         u64 start;
 840         u64 end;
 841         struct extent_map_tree *pending_del;
 842         struct extent_map_tree *pinned_extents;
 843
 844         pending_del = &extent_root->fs_info->pending_del;
 845         pinned_extents = &extent_root->fs_info->pinned_extents;
 846
 847         while(1) {
 848                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
 849                                             EXTENT_LOCKED);
 850                 if (ret)
 851                         break;
 852
 853                 set_extent_dirty(pinned_extents, start, end, GFP_NOFS);
 854                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
 855                                   GFP_NOFS);
 856                 ret = __free_extent(trans, extent_root,
 857                                      start, end + 1 - start, 0, 0);
 858                 if (ret)
 859                         err = ret;
 860         }
 861         return err;
 862 }
 863
 864 /*
 865  * remove an extent from the root, returns 0 on success
 866  */
 867 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 868                       *root, u64 bytenr, u64 num_bytes, int pin)
 869 {
 870         struct btrfs_root *extent_root = root->fs_info->extent_root;
 871         int pending_ret;
 872         int ret;
 873
 874         WARN_ON(num_bytes < root->sectorsize);
 875         if (root == extent_root) {
 876                 pin_down_bytes(root, bytenr, num_bytes, 1);
 877                 return 0;
 878         }
 879         ret = __free_extent(trans, root, bytenr, num_bytes, pin, pin == 0);
 880         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
 881         return ret ? ret : pending_ret;
 882 }
 883
 884 /*
 885  * walks the btree of allocated extents and find a hole of a given size.
 886  * The key ins is changed to record the hole:
 887  * ins->objectid == block start
 888  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 889  * ins->offset == number of blocks
 890  * Any available blocks before search_start are skipped.
 891  */
 892 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 893                             *orig_root, u64 num_bytes, u64 empty_size,
 894                             u64 search_start, u64 search_end, u64 hint_byte,
 895                             struct btrfs_key *ins, u64 exclude_start,
 896                             u64 exclude_nr, int data)
 897 {
 898         struct btrfs_path *path;
 899         struct btrfs_key key;
 900         int ret;
 901         u64 hole_size = 0;
 902         int slot = 0;
 903         u64 last_byte = 0;
 904         u64 orig_search_start = search_start;
 905         int start_found;
 906         struct extent_buffer *l;
 907         struct btrfs_root * root = orig_root->fs_info->extent_root;
 908         struct btrfs_fs_info *info = root->fs_info;
 909         u64 total_needed = num_bytes;
 910         int level;
 911         struct btrfs_block_group_cache *block_group;
 912         int full_scan = 0;
 913         int wrapped = 0;
 914         u64 cached_start;
 915
 916         WARN_ON(num_bytes < root->sectorsize);
 917         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 918
 919         level = btrfs_header_level(root->node);
 920
 921         if (search_end == (u64)-1)
 922                 search_end = btrfs_super_total_bytes(&info->super_copy);
 923         if (hint_byte) {
 924                 block_group = btrfs_lookup_block_group(info, hint_byte);
 925                 block_group = btrfs_find_block_group(root, block_group,
 926                                                      hint_byte, data, 1);
 927         } else {
 928                 block_group = btrfs_find_block_group(root,
 929                                                      trans->block_group, 0,
 930                                                      data, 1);
 931         }
 932
 933         total_needed += empty_size;
 934         path = btrfs_alloc_path();
 935
 936 check_failed:
 937         search_start = find_search_start(root, &block_group,
 938                                          search_start, total_needed, data);
 939         cached_start = search_start;
 940         btrfs_init_path(path);
 941         ins->objectid = search_start;
 942         ins->offset = 0;
 943         start_found = 0;
 944         path->reada = 2;
 945
 946         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 947         if (ret < 0)
 948                 goto error;
 949
 950         if (path->slots[0] > 0) {
 951                 path->slots[0]--;
 952         }
 953
 954         l = path->nodes[0];
 955         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 956
 957         /*
 958          * a rare case, go back one key if we hit a block group item
 959          * instead of an extent item
 960          */
 961         if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
 962             key.objectid + key.offset >= search_start) {
 963                 ins->objectid = key.objectid;
 964                 ins->offset = key.offset - 1;
 965                 btrfs_release_path(root, path);
 966                 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 967                 if (ret < 0)
 968                         goto error;
 969
 970                 if (path->slots[0] > 0) {
 971                         path->slots[0]--;
 972                 }
 973         }
 974
 975         while (1) {
 976                 l = path->nodes[0];
 977                 slot = path->slots[0];
 978                 if (slot >= btrfs_header_nritems(l)) {
 979                         ret = btrfs_next_leaf(root, path);
 980                         if (ret == 0)
 981                                 continue;
 982                         if (ret < 0)
 983                                 goto error;
 984
 985                         search_start = max(search_start,
 986                                            block_group->key.objectid);
 987                         if (!start_found) {
 988                                 ins->objectid = search_start;
 989                                 ins->offset = search_end - search_start;
 990                                 start_found = 1;
 991                                 goto check_pending;
 992                         }
 993                         ins->objectid = last_byte > search_start ?
 994                                         last_byte : search_start;
 995                         ins->offset = search_end - ins->objectid;
 996                         BUG_ON(ins->objectid >= search_end);
 997                         goto check_pending;
 998                 }
 999                 btrfs_item_key_to_cpu(l, &key, slot);
1000
1001                 if (key.objectid >= search_start && key.objectid > last_byte &&
1002                     start_found) {
1003                         if (last_byte < search_start)
1004                                 last_byte = search_start;
1005                         hole_size = key.objectid - last_byte;
1006                         if (hole_size >= num_bytes) {
1007                                 ins->objectid = last_byte;
1008                                 ins->offset = hole_size;
1009                                 goto check_pending;
1010                         }
1011                 }
1012                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1013                         if (!start_found) {
1014                                 last_byte = key.objectid;
1015                                 start_found = 1;
1016                         }
1017                         goto next;
1018                 }
1019
1020
1021                 start_found = 1;
1022                 last_byte = key.objectid + key.offset;
1023
1024                 if (!full_scan && last_byte >= block_group->key.objectid +
1025                     block_group->key.offset) {
1026                         btrfs_release_path(root, path);
1027                         search_start = block_group->key.objectid +
1028                                 block_group->key.offset;
1029                         goto new_group;
1030                 }
1031 next:
1032                 path->slots[0]++;
1033                 cond_resched();
1034         }
1035 check_pending:
1036         /* we have to make sure we didn't find an extent that has already
1037          * been allocated by the map tree or the original allocation
1038          */
1039         btrfs_release_path(root, path);
1040         BUG_ON(ins->objectid < search_start);
1041
1042         if (ins->objectid + num_bytes >= search_end)
1043                 goto enospc;
1044
1045         if (!full_scan && ins->objectid + num_bytes >= block_group->
1046             key.objectid + block_group->key.offset) {
1047                 search_start = block_group->key.objectid +
1048                         block_group->key.offset;
1049                 goto new_group;
1050         }
1051         if (test_range_bit(&info->extent_ins, ins->objectid,
1052                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1053                 search_start = ins->objectid + num_bytes;
1054                 goto new_group;
1055         }
1056         if (test_range_bit(&info->pinned_extents, ins->objectid,
1057                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1058                 search_start = ins->objectid + num_bytes;
1059                 goto new_group;
1060         }
1061         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1062             ins->objectid < exclude_start + exclude_nr)) {
1063                 search_start = exclude_start + exclude_nr;
1064                 goto new_group;
1065         }
1066         if (!data) {
1067                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1068                 if (block_group)
1069                         trans->block_group = block_group;
1070         }
1071         ins->offset = num_bytes;
1072         btrfs_free_path(path);
1073         return 0;
1074
1075 new_group:
1076         if (search_start + num_bytes >= search_end) {
1077 enospc:
1078                 search_start = orig_search_start;
1079                 if (full_scan) {
1080                         ret = -ENOSPC;
1081                         goto error;
1082                 }
1083                 if (wrapped) {
1084                         if (!full_scan)
1085                                 total_needed -= empty_size;
1086                         full_scan = 1;
1087                 } else
1088                         wrapped = 1;
1089         }
1090         block_group = btrfs_lookup_block_group(info, search_start);
1091         cond_resched();
1092         if (!full_scan)
1093                 block_group = btrfs_find_block_group(root, block_group,
1094                                                      search_start, data, 0);
1095         goto check_failed;
1096
1097 error:
1098         btrfs_release_path(root, path);
1099         btrfs_free_path(path);
1100         return ret;
1101 }
1102 /*
1103  * finds a free extent and does all the dirty work required for allocation
1104  * returns the key for the extent through ins, and a tree buffer for
1105  * the first block of the extent through buf.
1106  *
1107  * returns 0 if everything worked, non-zero otherwise.
1108  */
1109 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1110                        struct btrfs_root *root, u64 owner,
1111                        u64 num_bytes, u64 empty_size, u64 hint_byte,
1112                        u64 search_end, struct btrfs_key *ins, int data)
1113 {
1114         int ret;
1115         int pending_ret;
1116         u64 super_used, root_used;
1117         u64 search_start = 0;
1118         struct btrfs_fs_info *info = root->fs_info;
1119         struct btrfs_root *extent_root = info->extent_root;
1120         struct btrfs_extent_item extent_item;
1121
1122         btrfs_set_stack_extent_refs(&extent_item, 1);
1123         btrfs_set_stack_extent_owner(&extent_item, owner);
1124
1125         WARN_ON(num_bytes < root->sectorsize);
1126         ret = find_free_extent(trans, root, num_bytes, empty_size,
1127                                search_start, search_end, hint_byte, ins,
1128                                trans->alloc_exclude_start,
1129                                trans->alloc_exclude_nr, data);
1130         BUG_ON(ret);
1131         if (ret)
1132                 return ret;
1133
1134         /* block accounting for super block */
1135         super_used = btrfs_super_bytes_used(&info->super_copy);
1136         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1137
1138         /* block accounting for root item */
1139         root_used = btrfs_root_used(&root->root_item);
1140         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1141
1142         clear_extent_dirty(&root->fs_info->free_space_cache,
1143                            ins->objectid, ins->objectid + ins->offset - 1,
1144                            GFP_NOFS);
1145
1146         if (root == extent_root) {
1147                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1148                                 ins->objectid + ins->offset - 1,
1149                                 EXTENT_LOCKED, GFP_NOFS);
1150                 WARN_ON(data == 1);
1151                 goto update_block;
1152         }
1153
1154         WARN_ON(trans->alloc_exclude_nr);
1155         trans->alloc_exclude_start = ins->objectid;
1156         trans->alloc_exclude_nr = ins->offset;
1157         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1158                                 sizeof(extent_item));
1159
1160         trans->alloc_exclude_start = 0;
1161         trans->alloc_exclude_nr = 0;
1162
1163         BUG_ON(ret);
1164         finish_current_insert(trans, extent_root);
1165         pending_ret = del_pending_extents(trans, extent_root);
1166
1167         if (ret) {
1168                 return ret;
1169         }
1170         if (pending_ret) {
1171                 return pending_ret;
1172         }
1173
1174 update_block:
1175         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1176                                  data);
1177         BUG_ON(ret);
1178         return 0;
1179 }
1180
1181 /*
1182  * helper function to allocate a block for a given tree
1183  * returns the tree buffer or NULL.
1184  */
1185 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1186                                              struct btrfs_root *root,
1187                                              u32 blocksize, u64 hint,
1188                                              u64 empty_size)
1189 {
1190         struct btrfs_key ins;
1191         int ret;
1192         struct extent_buffer *buf;
1193
1194         ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1195                                  blocksize, empty_size, hint,
1196                                  (u64)-1, &ins, 0);
1197         if (ret) {
1198                 BUG_ON(ret > 0);
1199                 return ERR_PTR(ret);
1200         }
1201         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1202         if (!buf) {
1203                 btrfs_free_extent(trans, root, ins.objectid, blocksize, 0);
1204                 return ERR_PTR(-ENOMEM);
1205         }
1206         btrfs_set_buffer_uptodate(buf);
1207         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1208                          buf->start + buf->len - 1, GFP_NOFS);
1209         set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1210                         buf->start, buf->start + buf->len - 1,
1211                         EXTENT_CSUM, GFP_NOFS);
1212         buf->flags |= EXTENT_CSUM;
1213         btrfs_set_buffer_defrag(buf);
1214         trans->blocks_used++;
1215         return buf;
1216 }
1217
1218 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1219                          struct btrfs_root *root, struct extent_buffer *leaf)
1220 {
1221         struct btrfs_key key;
1222         struct btrfs_file_extent_item *fi;
1223         int i;
1224         int nritems;
1225         int ret;
1226
1227         BUG_ON(!btrfs_is_leaf(leaf));
1228         nritems = btrfs_header_nritems(leaf);
1229         for (i = 0; i < nritems; i++) {
1230                 u64 disk_bytenr;
1231
1232                 btrfs_item_key_to_cpu(leaf, &key, i);
1233                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1234                         continue;
1235                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1236                 if (btrfs_file_extent_type(leaf, fi) ==
1237                     BTRFS_FILE_EXTENT_INLINE)
1238                         continue;
1239                 /*
1240                  * FIXME make sure to insert a trans record that
1241                  * repeats the snapshot del on crash
1242                  */
1243                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1244                 if (disk_bytenr == 0)
1245                         continue;
1246                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1247                                 btrfs_file_extent_disk_num_bytes(leaf, fi), 0);
1248                 BUG_ON(ret);
1249         }
1250         return 0;
1251 }
1252
1253 static void reada_walk_down(struct btrfs_root *root,
1254                             struct extent_buffer *node)
1255 {
1256         int i;
1257         u32 nritems;
1258         u64 bytenr;
1259         int ret;
1260         u32 refs;
1261         int level;
1262         u32 blocksize;
1263
1264         nritems = btrfs_header_nritems(node);
1265         level = btrfs_header_level(node);
1266         for (i = 0; i < nritems; i++) {
1267                 bytenr = btrfs_node_blockptr(node, i);
1268                 blocksize = btrfs_level_size(root, level - 1);
1269                 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1270                 BUG_ON(ret);
1271                 if (refs != 1)
1272                         continue;
1273                 mutex_unlock(&root->fs_info->fs_mutex);
1274                 ret = readahead_tree_block(root, bytenr, blocksize);
1275                 cond_resched();
1276                 mutex_lock(&root->fs_info->fs_mutex);
1277                 if (ret)
1278                         break;
1279         }
1280 }
1281
1282 /*
1283  * helper function for drop_snapshot, this walks down the tree dropping ref
1284  * counts as it goes.
1285  */
1286 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1287                           *root, struct btrfs_path *path, int *level)
1288 {
1289         struct extent_buffer *next;
1290         struct extent_buffer *cur;
1291         u64 bytenr;
1292         u32 blocksize;
1293         int ret;
1294         u32 refs;
1295
1296         WARN_ON(*level < 0);
1297         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1298         ret = lookup_extent_ref(trans, root,
1299                                 path->nodes[*level]->start,
1300                                 path->nodes[*level]->len, &refs);
1301         BUG_ON(ret);
1302         if (refs > 1)
1303                 goto out;
1304
1305         /*
1306          * walk down to the last node level and free all the leaves
1307          */
1308         while(*level >= 0) {
1309                 WARN_ON(*level < 0);
1310                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1311                 cur = path->nodes[*level];
1312
1313                 if (*level > 0 && path->slots[*level] == 0)
1314                         reada_walk_down(root, cur);
1315
1316                 if (btrfs_header_level(cur) != *level)
1317                         WARN_ON(1);
1318
1319                 if (path->slots[*level] >=
1320                     btrfs_header_nritems(cur))
1321                         break;
1322                 if (*level == 0) {
1323                         ret = drop_leaf_ref(trans, root, cur);
1324                         BUG_ON(ret);
1325                         break;
1326                 }
1327                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1328                 blocksize = btrfs_level_size(root, *level - 1);
1329                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1330                 BUG_ON(ret);
1331                 if (refs != 1) {
1332                         path->slots[*level]++;
1333                         ret = btrfs_free_extent(trans, root, bytenr,
1334                                                 blocksize, 1);
1335                         BUG_ON(ret);
1336                         continue;
1337                 }
1338                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1339                 if (!next || !btrfs_buffer_uptodate(next)) {
1340                         free_extent_buffer(next);
1341                         mutex_unlock(&root->fs_info->fs_mutex);
1342                         next = read_tree_block(root, bytenr, blocksize);
1343                         mutex_lock(&root->fs_info->fs_mutex);
1344
1345                         /* we dropped the lock, check one more time */
1346                         ret = lookup_extent_ref(trans, root, bytenr,
1347                                                 blocksize, &refs);
1348                         BUG_ON(ret);
1349                         if (refs != 1) {
1350                                 path->slots[*level]++;
1351                                 free_extent_buffer(next);
1352                                 ret = btrfs_free_extent(trans, root,
1353                                                         bytenr, blocksize, 1);
1354                                 BUG_ON(ret);
1355                                 continue;
1356                         }
1357                 }
1358                 WARN_ON(*level <= 0);
1359                 if (path->nodes[*level-1])
1360                         free_extent_buffer(path->nodes[*level-1]);
1361                 path->nodes[*level-1] = next;
1362                 *level = btrfs_header_level(next);
1363                 path->slots[*level] = 0;
1364         }
1365 out:
1366         WARN_ON(*level < 0);
1367         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1368         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1369                                 path->nodes[*level]->len, 1);
1370         free_extent_buffer(path->nodes[*level]);
1371         path->nodes[*level] = NULL;
1372         *level += 1;
1373         BUG_ON(ret);
1374         return 0;
1375 }
1376
1377 /*
1378  * helper for dropping snapshots.  This walks back up the tree in the path
1379  * to find the first node higher up where we haven't yet gone through
1380  * all the slots
1381  */
1382 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1383                         *root, struct btrfs_path *path, int *level)
1384 {
1385         int i;
1386         int slot;
1387         int ret;
1388         struct btrfs_root_item *root_item = &root->root_item;
1389
1390         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1391                 slot = path->slots[i];
1392                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1393                         struct extent_buffer *node;
1394                         struct btrfs_disk_key disk_key;
1395                         node = path->nodes[i];
1396                         path->slots[i]++;
1397                         *level = i;
1398                         WARN_ON(*level == 0);
1399                         btrfs_node_key(node, &disk_key, path->slots[i]);
1400                         memcpy(&root_item->drop_progress,
1401                                &disk_key, sizeof(disk_key));
1402                         root_item->drop_level = i;
1403                         return 0;
1404                 } else {
1405                         ret = btrfs_free_extent(trans, root,
1406                                                 path->nodes[*level]->start,
1407                                                 path->nodes[*level]->len, 1);
1408                         BUG_ON(ret);
1409                         free_extent_buffer(path->nodes[*level]);
1410                         path->nodes[*level] = NULL;
1411                         *level = i + 1;
1412                 }
1413         }
1414         return 1;
1415 }
1416
1417 /*
1418  * drop the reference count on the tree rooted at 'snap'.  This traverses
1419  * the tree freeing any blocks that have a ref count of zero after being
1420  * decremented.
1421  */
1422 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1423                         *root)
1424 {
1425         int ret = 0;
1426         int wret;
1427         int level;
1428         struct btrfs_path *path;
1429         int i;
1430         int orig_level;
1431         struct btrfs_root_item *root_item = &root->root_item;
1432
1433         path = btrfs_alloc_path();
1434         BUG_ON(!path);
1435
1436         level = btrfs_header_level(root->node);
1437         orig_level = level;
1438         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1439                 path->nodes[level] = root->node;
1440                 extent_buffer_get(root->node);
1441                 path->slots[level] = 0;
1442         } else {
1443                 struct btrfs_key key;
1444                 struct btrfs_disk_key found_key;
1445                 struct extent_buffer *node;
1446
1447                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1448                 level = root_item->drop_level;
1449                 path->lowest_level = level;
1450                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1451                 if (wret < 0) {
1452                         ret = wret;
1453                         goto out;
1454                 }
1455                 node = path->nodes[level];
1456                 btrfs_node_key(node, &found_key, path->slots[level]);
1457                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1458                                sizeof(found_key)));
1459         }
1460         while(1) {
1461                 wret = walk_down_tree(trans, root, path, &level);
1462                 if (wret > 0)
1463                         break;
1464                 if (wret < 0)
1465                         ret = wret;
1466
1467                 wret = walk_up_tree(trans, root, path, &level);
1468                 if (wret > 0)
1469                         break;
1470                 if (wret < 0)
1471                         ret = wret;
1472                 ret = -EAGAIN;
1473                 break;
1474         }
1475         for (i = 0; i <= orig_level; i++) {
1476                 if (path->nodes[i]) {
1477                         free_extent_buffer(path->nodes[i]);
1478                         path->nodes[i] = NULL;
1479                 }
1480         }
1481 out:
1482         btrfs_free_path(path);
1483         return ret;
1484 }
1485
1486 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1487 {
1488         u64 start;
1489         u64 end;
1490         u64 ptr;
1491         int ret;
1492         while(1) {
1493                 ret = find_first_extent_bit(&info->block_group_cache, 0,
1494                                             &start, &end, (unsigned int)-1);
1495                 if (ret)
1496                         break;
1497                 ret = get_state_private(&info->block_group_cache, start, &ptr);
1498                 if (!ret)
1499                         kfree((void *)(unsigned long)ptr);
1500                 clear_extent_bits(&info->block_group_cache, start,
1501                                   end, (unsigned int)-1, GFP_NOFS);
1502         }
1503         while(1) {
1504                 ret = find_first_extent_bit(&info->free_space_cache, 0,
1505                                             &start, &end, EXTENT_DIRTY);
1506                 if (ret)
1507                         break;
1508                 clear_extent_dirty(&info->free_space_cache, start,
1509                                    end, GFP_NOFS);
1510         }
1511         return 0;
1512 }
1513
1514 int btrfs_read_block_groups(struct btrfs_root *root)
1515 {
1516         struct btrfs_path *path;
1517         int ret;
1518         int err = 0;
1519         int bit;
1520         struct btrfs_block_group_cache *cache;
1521         struct btrfs_fs_info *info = root->fs_info;
1522         struct extent_map_tree *block_group_cache;
1523         struct btrfs_key key;
1524         struct btrfs_key found_key;
1525         struct extent_buffer *leaf;
1526
1527         block_group_cache = &info->block_group_cache;
1528
1529         root = info->extent_root;
1530         key.objectid = 0;
1531         key.offset = BTRFS_BLOCK_GROUP_SIZE;
1532         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1533
1534         path = btrfs_alloc_path();
1535         if (!path)
1536                 return -ENOMEM;
1537
1538         while(1) {
1539                 ret = btrfs_search_slot(NULL, info->extent_root,
1540                                         &key, path, 0, 0);
1541                 if (ret != 0) {
1542                         err = ret;
1543                         break;
1544                 }
1545                 leaf = path->nodes[0];
1546                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1547                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1548                 if (!cache) {
1549                         err = -1;
1550                         break;
1551                 }
1552
1553                 read_extent_buffer(leaf, &cache->item,
1554                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
1555                                    sizeof(cache->item));
1556                 memcpy(&cache->key, &found_key, sizeof(found_key));
1557                 cache->cached = 0;
1558
1559                 key.objectid = found_key.objectid + found_key.offset;
1560                 btrfs_release_path(root, path);
1561
1562                 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
1563                         bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
1564                         cache->data = BTRFS_BLOCK_GROUP_MIXED;
1565                 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
1566                         bit = BLOCK_GROUP_DATA;
1567                         cache->data = BTRFS_BLOCK_GROUP_DATA;
1568                 } else {
1569                         bit = BLOCK_GROUP_METADATA;
1570                         cache->data = 0;
1571                 }
1572
1573                 /* use EXTENT_LOCKED to prevent merging */
1574                 set_extent_bits(block_group_cache, found_key.objectid,
1575                                 found_key.objectid + found_key.offset - 1,
1576                                 bit | EXTENT_LOCKED, GFP_NOFS);
1577                 set_state_private(block_group_cache, found_key.objectid,
1578                                   (unsigned long)cache);
1579
1580                 if (key.objectid >=
1581                     btrfs_super_total_bytes(&info->super_copy))
1582                         break;
1583         }
1584
1585         btrfs_free_path(path);
1586         return 0;
1587 }