extent-tree.c

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