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 <stdint.h>
  22 #include "kerncompat.h"
  23 #include "radix-tree.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "crc32c.h"
  29 #include "volumes.h"
  30 #include "free-space-cache.h"
  31 #include "math.h"
  32
  33 #define PENDING_EXTENT_INSERT 0
  34 #define PENDING_EXTENT_DELETE 1
  35 #define PENDING_BACKREF_UPDATE 2
  36
  37 struct pending_extent_op {
  38         int type;
  39         u64 bytenr;
  40         u64 num_bytes;
  41         u64 flags;
  42         struct btrfs_disk_key key;
  43         int level;
  44 };
  45
  46 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  47                                      struct btrfs_root *root,
  48                                      u64 root_objectid, u64 generation,
  49                                      u64 flags, struct btrfs_disk_key *key,
  50                                      int level, struct btrfs_key *ins);
  51 static int __free_extent(struct btrfs_trans_handle *trans,
  52                          struct btrfs_root *root,
  53                          u64 bytenr, u64 num_bytes, u64 parent,
  54                          u64 root_objectid, u64 owner_objectid,
  55                          u64 owner_offset, int refs_to_drop);
  56 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  57                                  btrfs_root *extent_root);
  58 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  59                                btrfs_root *extent_root);
  60
  61 static int remove_sb_from_cache(struct btrfs_root *root,
  62                                 struct btrfs_block_group_cache *cache)
  63 {
  64         u64 bytenr;
  65         u64 *logical;
  66         int stripe_len;
  67         int i, nr, ret;
  68         struct extent_io_tree *free_space_cache;
  69
  70         free_space_cache = &root->fs_info->free_space_cache;
  71         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
  72                 bytenr = btrfs_sb_offset(i);
  73                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
  74                                        cache->key.objectid, bytenr, 0,
  75                                        &logical, &nr, &stripe_len);
  76                 BUG_ON(ret);
  77                 while (nr--) {
  78                         clear_extent_dirty(free_space_cache, logical[nr],
  79                                 logical[nr] + stripe_len - 1, GFP_NOFS);
  80                 }
  81                 kfree(logical);
  82         }
  83         return 0;
  84 }
  85
  86 static int cache_block_group(struct btrfs_root *root,
  87                              struct btrfs_block_group_cache *block_group)
  88 {
  89         struct btrfs_path *path;
  90         int ret;
  91         struct btrfs_key key;
  92         struct extent_buffer *leaf;
  93         struct extent_io_tree *free_space_cache;
  94         int slot;
  95         u64 last;
  96         u64 hole_size;
  97
  98         if (!block_group)
  99                 return 0;
 100
 101         root = root->fs_info->extent_root;
 102         free_space_cache = &root->fs_info->free_space_cache;
 103
 104         if (block_group->cached)
 105                 return 0;
 106
 107         path = btrfs_alloc_path();
 108         if (!path)
 109                 return -ENOMEM;
 110
 111         path->reada = 2;
 112         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 113         key.objectid = last;
 114         key.offset = 0;
 115         key.type = 0;
 116
 117         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 118         if (ret < 0)
 119                 goto err;
 120
 121         while(1) {
 122                 leaf = path->nodes[0];
 123                 slot = path->slots[0];
 124                 if (slot >= btrfs_header_nritems(leaf)) {
 125                         ret = btrfs_next_leaf(root, path);
 126                         if (ret < 0)
 127                                 goto err;
 128                         if (ret == 0) {
 129                                 continue;
 130                         } else {
 131                                 break;
 132                         }
 133                 }
 134                 btrfs_item_key_to_cpu(leaf, &key, slot);
 135                 if (key.objectid < block_group->key.objectid) {
 136                         goto next;
 137                 }
 138                 if (key.objectid >= block_group->key.objectid +
 139                     block_group->key.offset) {
 140                         break;
 141                 }
 142
 143                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 144                     key.type == BTRFS_METADATA_ITEM_KEY) {
 145                         if (key.objectid > last) {
 146                                 hole_size = key.objectid - last;
 147                                 set_extent_dirty(free_space_cache, last,
 148                                                  last + hole_size - 1,
 149                                                  GFP_NOFS);
 150                         }
 151                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 152                                 last = key.objectid + root->leafsize;
 153                         else
 154                                 last = key.objectid + key.offset;
 155                 }
 156 next:
 157                 path->slots[0]++;
 158         }
 159
 160         if (block_group->key.objectid +
 161             block_group->key.offset > last) {
 162                 hole_size = block_group->key.objectid +
 163                         block_group->key.offset - last;
 164                 set_extent_dirty(free_space_cache, last,
 165                                  last + hole_size - 1, GFP_NOFS);
 166         }
 167         remove_sb_from_cache(root, block_group);
 168         block_group->cached = 1;
 169 err:
 170         btrfs_free_path(path);
 171         return 0;
 172 }
 173
 174 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 175                                                        btrfs_fs_info *info,
 176                                                        u64 bytenr)
 177 {
 178         struct extent_io_tree *block_group_cache;
 179         struct btrfs_block_group_cache *block_group = NULL;
 180         u64 ptr;
 181         u64 start;
 182         u64 end;
 183         int ret;
 184
 185         bytenr = max_t(u64, bytenr,
 186                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 187         block_group_cache = &info->block_group_cache;
 188         ret = find_first_extent_bit(block_group_cache,
 189                                     bytenr, &start, &end,
 190                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 191                                     BLOCK_GROUP_SYSTEM);
 192         if (ret) {
 193                 return NULL;
 194         }
 195         ret = get_state_private(block_group_cache, start, &ptr);
 196         if (ret)
 197                 return NULL;
 198
 199         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 200         return block_group;
 201 }
 202
 203 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 204                                                          btrfs_fs_info *info,
 205                                                          u64 bytenr)
 206 {
 207         struct extent_io_tree *block_group_cache;
 208         struct btrfs_block_group_cache *block_group = NULL;
 209         u64 ptr;
 210         u64 start;
 211         u64 end;
 212         int ret;
 213
 214         block_group_cache = &info->block_group_cache;
 215         ret = find_first_extent_bit(block_group_cache,
 216                                     bytenr, &start, &end,
 217                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 218                                     BLOCK_GROUP_SYSTEM);
 219         if (ret) {
 220                 return NULL;
 221         }
 222         ret = get_state_private(block_group_cache, start, &ptr);
 223         if (ret)
 224                 return NULL;
 225
 226         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 227         if (block_group->key.objectid <= bytenr && bytenr <
 228             block_group->key.objectid + block_group->key.offset)
 229                 return block_group;
 230         return NULL;
 231 }
 232
 233 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 234 {
 235         return (cache->flags & bits) == bits;
 236 }
 237
 238 static int noinline find_search_start(struct btrfs_root *root,
 239                               struct btrfs_block_group_cache **cache_ret,
 240                               u64 *start_ret, int num, int data)
 241 {
 242         int ret;
 243         struct btrfs_block_group_cache *cache = *cache_ret;
 244         u64 last = *start_ret;
 245         u64 start = 0;
 246         u64 end = 0;
 247         u64 search_start = *start_ret;
 248         int wrapped = 0;
 249
 250         if (!cache)
 251                 goto out;
 252 again:
 253         ret = cache_block_group(root, cache);
 254         if (ret)
 255                 goto out;
 256
 257         last = max(search_start, cache->key.objectid);
 258         if (cache->ro || !block_group_bits(cache, data))
 259                 goto new_group;
 260
 261         while(1) {
 262                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 263                                             last, &start, &end, EXTENT_DIRTY);
 264                 if (ret) {
 265                         goto new_group;
 266                 }
 267
 268                 start = max(last, start);
 269                 last = end + 1;
 270                 if (last - start < num) {
 271                         continue;
 272                 }
 273                 if (start + num > cache->key.objectid + cache->key.offset) {
 274                         goto new_group;
 275                 }
 276                 *start_ret = start;
 277                 return 0;
 278         }
 279 out:
 280         *start_ret = last;
 281         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 282         if (!cache) {
 283                 printk("Unable to find block group for %llu\n",
 284                         (unsigned long long)search_start);
 285                 WARN_ON(1);
 286         }
 287         return -ENOSPC;
 288
 289 new_group:
 290         last = cache->key.objectid + cache->key.offset;
 291 wrapped:
 292         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 293         if (!cache) {
 294                 if (!wrapped) {
 295                         wrapped = 1;
 296                         last = search_start;
 297                         goto wrapped;
 298                 }
 299                 goto out;
 300         }
 301         *cache_ret = cache;
 302         goto again;
 303 }
 304
 305 static int block_group_state_bits(u64 flags)
 306 {
 307         int bits = 0;
 308         if (flags & BTRFS_BLOCK_GROUP_DATA)
 309                 bits |= BLOCK_GROUP_DATA;
 310         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 311                 bits |= BLOCK_GROUP_METADATA;
 312         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 313                 bits |= BLOCK_GROUP_SYSTEM;
 314         return bits;
 315 }
 316
 317 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 318                                                  struct btrfs_block_group_cache
 319                                                  *hint, u64 search_start,
 320                                                  int data, int owner)
 321 {
 322         struct btrfs_block_group_cache *cache;
 323         struct extent_io_tree *block_group_cache;
 324         struct btrfs_block_group_cache *found_group = NULL;
 325         struct btrfs_fs_info *info = root->fs_info;
 326         u64 used;
 327         u64 last = 0;
 328         u64 hint_last;
 329         u64 start;
 330         u64 end;
 331         u64 free_check;
 332         u64 ptr;
 333         int bit;
 334         int ret;
 335         int full_search = 0;
 336         int factor = 10;
 337
 338         block_group_cache = &info->block_group_cache;
 339
 340         if (!owner)
 341                 factor = 10;
 342
 343         bit = block_group_state_bits(data);
 344
 345         if (search_start) {
 346                 struct btrfs_block_group_cache *shint;
 347                 shint = btrfs_lookup_block_group(info, search_start);
 348                 if (shint && !shint->ro && block_group_bits(shint, data)) {
 349                         used = btrfs_block_group_used(&shint->item);
 350                         if (used + shint->pinned <
 351                             div_factor(shint->key.offset, factor)) {
 352                                 return shint;
 353                         }
 354                 }
 355         }
 356         if (hint && !hint->ro && block_group_bits(hint, data)) {
 357                 used = btrfs_block_group_used(&hint->item);
 358                 if (used + hint->pinned <
 359                     div_factor(hint->key.offset, factor)) {
 360                         return hint;
 361                 }
 362                 last = hint->key.objectid + hint->key.offset;
 363                 hint_last = last;
 364         } else {
 365                 if (hint)
 366                         hint_last = max(hint->key.objectid, search_start);
 367                 else
 368                         hint_last = search_start;
 369
 370                 last = hint_last;
 371         }
 372 again:
 373         while(1) {
 374                 ret = find_first_extent_bit(block_group_cache, last,
 375                                             &start, &end, bit);
 376                 if (ret)
 377                         break;
 378
 379                 ret = get_state_private(block_group_cache, start, &ptr);
 380                 if (ret)
 381                         break;
 382
 383                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 384                 last = cache->key.objectid + cache->key.offset;
 385                 used = btrfs_block_group_used(&cache->item);
 386
 387                 if (!cache->ro && block_group_bits(cache, data)) {
 388                         if (full_search)
 389                                 free_check = cache->key.offset;
 390                         else
 391                                 free_check = div_factor(cache->key.offset,
 392                                                         factor);
 393
 394                         if (used + cache->pinned < free_check) {
 395                                 found_group = cache;
 396                                 goto found;
 397                         }
 398                 }
 399                 cond_resched();
 400         }
 401         if (!full_search) {
 402                 last = search_start;
 403                 full_search = 1;
 404                 goto again;
 405         }
 406 found:
 407         return found_group;
 408 }
 409
 410 /*
 411  * Back reference rules.  Back refs have three main goals:
 412  *
 413  * 1) differentiate between all holders of references to an extent so that
 414  *    when a reference is dropped we can make sure it was a valid reference
 415  *    before freeing the extent.
 416  *
 417  * 2) Provide enough information to quickly find the holders of an extent
 418  *    if we notice a given block is corrupted or bad.
 419  *
 420  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 421  *    maintenance.  This is actually the same as #2, but with a slightly
 422  *    different use case.
 423  *
 424  * There are two kinds of back refs. The implicit back refs is optimized
 425  * for pointers in non-shared tree blocks. For a given pointer in a block,
 426  * back refs of this kind provide information about the block's owner tree
 427  * and the pointer's key. These information allow us to find the block by
 428  * b-tree searching. The full back refs is for pointers in tree blocks not
 429  * referenced by their owner trees. The location of tree block is recorded
 430  * in the back refs. Actually the full back refs is generic, and can be
 431  * used in all cases the implicit back refs is used. The major shortcoming
 432  * of the full back refs is its overhead. Every time a tree block gets
 433  * COWed, we have to update back refs entry for all pointers in it.
 434  *
 435  * For a newly allocated tree block, we use implicit back refs for
 436  * pointers in it. This means most tree related operations only involve
 437  * implicit back refs. For a tree block created in old transaction, the
 438  * only way to drop a reference to it is COW it. So we can detect the
 439  * event that tree block loses its owner tree's reference and do the
 440  * back refs conversion.
 441  *
 442  * When a tree block is COW'd through a tree, there are four cases:
 443  *
 444  * The reference count of the block is one and the tree is the block's
 445  * owner tree. Nothing to do in this case.
 446  *
 447  * The reference count of the block is one and the tree is not the
 448  * block's owner tree. In this case, full back refs is used for pointers
 449  * in the block. Remove these full back refs, add implicit back refs for
 450  * every pointers in the new block.
 451  *
 452  * The reference count of the block is greater than one and the tree is
 453  * the block's owner tree. In this case, implicit back refs is used for
 454  * pointers in the block. Add full back refs for every pointers in the
 455  * block, increase lower level extents' reference counts. The original
 456  * implicit back refs are entailed to the new block.
 457  *
 458  * The reference count of the block is greater than one and the tree is
 459  * not the block's owner tree. Add implicit back refs for every pointer in
 460  * the new block, increase lower level extents' reference count.
 461  *
 462  * Back Reference Key composing:
 463  *
 464  * The key objectid corresponds to the first byte in the extent,
 465  * The key type is used to differentiate between types of back refs.
 466  * There are different meanings of the key offset for different types
 467  * of back refs.
 468  *
 469  * File extents can be referenced by:
 470  *
 471  * - multiple snapshots, subvolumes, or different generations in one subvol
 472  * - different files inside a single subvolume
 473  * - different offsets inside a file (bookend extents in file.c)
 474  *
 475  * The extent ref structure for the implicit back refs has fields for:
 476  *
 477  * - Objectid of the subvolume root
 478  * - objectid of the file holding the reference
 479  * - original offset in the file
 480  * - how many bookend extents
 481  *
 482  * The key offset for the implicit back refs is hash of the first
 483  * three fields.
 484  *
 485  * The extent ref structure for the full back refs has field for:
 486  *
 487  * - number of pointers in the tree leaf
 488  *
 489  * The key offset for the implicit back refs is the first byte of
 490  * the tree leaf
 491  *
 492  * When a file extent is allocated, The implicit back refs is used.
 493  * the fields are filled in:
 494  *
 495  *     (root_key.objectid, inode objectid, offset in file, 1)
 496  *
 497  * When a file extent is removed file truncation, we find the
 498  * corresponding implicit back refs and check the following fields:
 499  *
 500  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 501  *
 502  * Btree extents can be referenced by:
 503  *
 504  * - Different subvolumes
 505  *
 506  * Both the implicit back refs and the full back refs for tree blocks
 507  * only consist of key. The key offset for the implicit back refs is
 508  * objectid of block's owner tree. The key offset for the full back refs
 509  * is the first byte of parent block.
 510  *
 511  * When implicit back refs is used, information about the lowest key and
 512  * level of the tree block are required. These information are stored in
 513  * tree block info structure.
 514  */
 515
 516 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 517 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 518                                   struct btrfs_root *root,
 519                                   struct btrfs_path *path,
 520                                   u64 owner, u32 extra_size)
 521 {
 522         struct btrfs_extent_item *item;
 523         struct btrfs_extent_item_v0 *ei0;
 524         struct btrfs_extent_ref_v0 *ref0;
 525         struct btrfs_tree_block_info *bi;
 526         struct extent_buffer *leaf;
 527         struct btrfs_key key;
 528         struct btrfs_key found_key;
 529         u32 new_size = sizeof(*item);
 530         u64 refs;
 531         int ret;
 532
 533         leaf = path->nodes[0];
 534         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 535
 536         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 537         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 538                              struct btrfs_extent_item_v0);
 539         refs = btrfs_extent_refs_v0(leaf, ei0);
 540
 541         if (owner == (u64)-1) {
 542                 while (1) {
 543                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 544                                 ret = btrfs_next_leaf(root, path);
 545                                 if (ret < 0)
 546                                         return ret;
 547                                 BUG_ON(ret > 0);
 548                                 leaf = path->nodes[0];
 549                         }
 550                         btrfs_item_key_to_cpu(leaf, &found_key,
 551                                               path->slots[0]);
 552                         BUG_ON(key.objectid != found_key.objectid);
 553                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 554                                 path->slots[0]++;
 555                                 continue;
 556                         }
 557                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 558                                               struct btrfs_extent_ref_v0);
 559                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 560                         break;
 561                 }
 562         }
 563         btrfs_release_path(root, path);
 564
 565         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 566                 new_size += sizeof(*bi);
 567
 568         new_size -= sizeof(*ei0);
 569         ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
 570         if (ret < 0)
 571                 return ret;
 572         BUG_ON(ret);
 573
 574         ret = btrfs_extend_item(trans, root, path, new_size);
 575         BUG_ON(ret);
 576
 577         leaf = path->nodes[0];
 578         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 579         btrfs_set_extent_refs(leaf, item, refs);
 580         /* FIXME: get real generation */
 581         btrfs_set_extent_generation(leaf, item, 0);
 582         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 583                 btrfs_set_extent_flags(leaf, item,
 584                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 585                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 586                 bi = (struct btrfs_tree_block_info *)(item + 1);
 587                 /* FIXME: get first key of the block */
 588                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 589                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 590         } else {
 591                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 592         }
 593         btrfs_mark_buffer_dirty(leaf);
 594         return 0;
 595 }
 596 #endif
 597
 598 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 599 {
 600         u32 high_crc = ~(u32)0;
 601         u32 low_crc = ~(u32)0;
 602         __le64 lenum;
 603
 604         lenum = cpu_to_le64(root_objectid);
 605         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 606         lenum = cpu_to_le64(owner);
 607         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 608         lenum = cpu_to_le64(offset);
 609         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 610
 611         return ((u64)high_crc << 31) ^ (u64)low_crc;
 612 }
 613
 614 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 615                                      struct btrfs_extent_data_ref *ref)
 616 {
 617         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 618                                     btrfs_extent_data_ref_objectid(leaf, ref),
 619                                     btrfs_extent_data_ref_offset(leaf, ref));
 620 }
 621
 622 static int match_extent_data_ref(struct extent_buffer *leaf,
 623                                  struct btrfs_extent_data_ref *ref,
 624                                  u64 root_objectid, u64 owner, u64 offset)
 625 {
 626         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 627             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 628             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 629                 return 0;
 630         return 1;
 631 }
 632
 633 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 634                                            struct btrfs_root *root,
 635                                            struct btrfs_path *path,
 636                                            u64 bytenr, u64 parent,
 637                                            u64 root_objectid,
 638                                            u64 owner, u64 offset)
 639 {
 640         struct btrfs_key key;
 641         struct btrfs_extent_data_ref *ref;
 642         struct extent_buffer *leaf;
 643         u32 nritems;
 644         int ret;
 645         int recow;
 646         int err = -ENOENT;
 647
 648         key.objectid = bytenr;
 649         if (parent) {
 650                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 651                 key.offset = parent;
 652         } else {
 653                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 654                 key.offset = hash_extent_data_ref(root_objectid,
 655                                                   owner, offset);
 656         }
 657 again:
 658         recow = 0;
 659         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 660         if (ret < 0) {
 661                 err = ret;
 662                 goto fail;
 663         }
 664
 665         if (parent) {
 666                 if (!ret)
 667                         return 0;
 668 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 669                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 670                 btrfs_release_path(root, path);
 671                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 672                 if (ret < 0) {
 673                         err = ret;
 674                         goto fail;
 675                 }
 676                 if (!ret)
 677                         return 0;
 678 #endif
 679                 goto fail;
 680         }
 681
 682         leaf = path->nodes[0];
 683         nritems = btrfs_header_nritems(leaf);
 684         while (1) {
 685                 if (path->slots[0] >= nritems) {
 686                         ret = btrfs_next_leaf(root, path);
 687                         if (ret < 0)
 688                                 err = ret;
 689                         if (ret)
 690                                 goto fail;
 691
 692                         leaf = path->nodes[0];
 693                         nritems = btrfs_header_nritems(leaf);
 694                         recow = 1;
 695                 }
 696
 697                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 698                 if (key.objectid != bytenr ||
 699                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 700                         goto fail;
 701
 702                 ref = btrfs_item_ptr(leaf, path->slots[0],
 703                                      struct btrfs_extent_data_ref);
 704
 705                 if (match_extent_data_ref(leaf, ref, root_objectid,
 706                                           owner, offset)) {
 707                         if (recow) {
 708                                 btrfs_release_path(root, path);
 709                                 goto again;
 710                         }
 711                         err = 0;
 712                         break;
 713                 }
 714                 path->slots[0]++;
 715         }
 716 fail:
 717         return err;
 718 }
 719
 720 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 721                                            struct btrfs_root *root,
 722                                            struct btrfs_path *path,
 723                                            u64 bytenr, u64 parent,
 724                                            u64 root_objectid, u64 owner,
 725                                            u64 offset, int refs_to_add)
 726 {
 727         struct btrfs_key key;
 728         struct extent_buffer *leaf;
 729         u32 size;
 730         u32 num_refs;
 731         int ret;
 732
 733         key.objectid = bytenr;
 734         if (parent) {
 735                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 736                 key.offset = parent;
 737                 size = sizeof(struct btrfs_shared_data_ref);
 738         } else {
 739                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 740                 key.offset = hash_extent_data_ref(root_objectid,
 741                                                   owner, offset);
 742                 size = sizeof(struct btrfs_extent_data_ref);
 743         }
 744
 745         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 746         if (ret && ret != -EEXIST)
 747                 goto fail;
 748
 749         leaf = path->nodes[0];
 750         if (parent) {
 751                 struct btrfs_shared_data_ref *ref;
 752                 ref = btrfs_item_ptr(leaf, path->slots[0],
 753                                      struct btrfs_shared_data_ref);
 754                 if (ret == 0) {
 755                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 756                 } else {
 757                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 758                         num_refs += refs_to_add;
 759                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 760                 }
 761         } else {
 762                 struct btrfs_extent_data_ref *ref;
 763                 while (ret == -EEXIST) {
 764                         ref = btrfs_item_ptr(leaf, path->slots[0],
 765                                              struct btrfs_extent_data_ref);
 766                         if (match_extent_data_ref(leaf, ref, root_objectid,
 767                                                   owner, offset))
 768                                 break;
 769                         btrfs_release_path(root, path);
 770
 771                         key.offset++;
 772                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 773                                                       size);
 774                         if (ret && ret != -EEXIST)
 775                                 goto fail;
 776
 777                         leaf = path->nodes[0];
 778                 }
 779                 ref = btrfs_item_ptr(leaf, path->slots[0],
 780                                      struct btrfs_extent_data_ref);
 781                 if (ret == 0) {
 782                         btrfs_set_extent_data_ref_root(leaf, ref,
 783                                                        root_objectid);
 784                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 785                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 786                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 787                 } else {
 788                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 789                         num_refs += refs_to_add;
 790                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 791                 }
 792         }
 793         btrfs_mark_buffer_dirty(leaf);
 794         ret = 0;
 795 fail:
 796         btrfs_release_path(root, path);
 797         return ret;
 798 }
 799
 800 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 801                                            struct btrfs_root *root,
 802                                            struct btrfs_path *path,
 803                                            int refs_to_drop)
 804 {
 805         struct btrfs_key key;
 806         struct btrfs_extent_data_ref *ref1 = NULL;
 807         struct btrfs_shared_data_ref *ref2 = NULL;
 808         struct extent_buffer *leaf;
 809         u32 num_refs = 0;
 810         int ret = 0;
 811
 812         leaf = path->nodes[0];
 813         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 814
 815         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 816                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 817                                       struct btrfs_extent_data_ref);
 818                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 819         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 820                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 821                                       struct btrfs_shared_data_ref);
 822                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 823 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 824         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 825                 struct btrfs_extent_ref_v0 *ref0;
 826                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 827                                       struct btrfs_extent_ref_v0);
 828                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 829 #endif
 830         } else {
 831                 BUG();
 832         }
 833
 834         BUG_ON(num_refs < refs_to_drop);
 835         num_refs -= refs_to_drop;
 836
 837         if (num_refs == 0) {
 838                 ret = btrfs_del_item(trans, root, path);
 839         } else {
 840                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 841                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 842                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 843                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 844 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 845                 else {
 846                         struct btrfs_extent_ref_v0 *ref0;
 847                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 848                                         struct btrfs_extent_ref_v0);
 849                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
 850                 }
 851 #endif
 852                 btrfs_mark_buffer_dirty(leaf);
 853         }
 854         return ret;
 855 }
 856
 857 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
 858                                           struct btrfs_path *path,
 859                                           struct btrfs_extent_inline_ref *iref)
 860 {
 861         struct btrfs_key key;
 862         struct extent_buffer *leaf;
 863         struct btrfs_extent_data_ref *ref1;
 864         struct btrfs_shared_data_ref *ref2;
 865         u32 num_refs = 0;
 866
 867         leaf = path->nodes[0];
 868         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 869         if (iref) {
 870                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
 871                     BTRFS_EXTENT_DATA_REF_KEY) {
 872                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 873                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 874                 } else {
 875                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 876                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 877                 }
 878         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 879                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 880                                       struct btrfs_extent_data_ref);
 881                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 882         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 883                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 884                                       struct btrfs_shared_data_ref);
 885                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 886 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 887         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 888                 struct btrfs_extent_ref_v0 *ref0;
 889                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 890                                       struct btrfs_extent_ref_v0);
 891                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 892 #endif
 893         } else {
 894                 BUG();
 895         }
 896         return num_refs;
 897 }
 898
 899 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 900                                           struct btrfs_root *root,
 901                                           struct btrfs_path *path,
 902                                           u64 bytenr, u64 parent,
 903                                           u64 root_objectid)
 904 {
 905         struct btrfs_key key;
 906         int ret;
 907
 908         key.objectid = bytenr;
 909         if (parent) {
 910                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 911                 key.offset = parent;
 912         } else {
 913                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 914                 key.offset = root_objectid;
 915         }
 916
 917         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 918         if (ret > 0)
 919                 ret = -ENOENT;
 920 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 921         if (ret == -ENOENT && parent) {
 922                 btrfs_release_path(root, path);
 923                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 924                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 925                 if (ret > 0)
 926                         ret = -ENOENT;
 927         }
 928 #endif
 929         return ret;
 930 }
 931
 932 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 933                                           struct btrfs_root *root,
 934                                           struct btrfs_path *path,
 935                                           u64 bytenr, u64 parent,
 936                                           u64 root_objectid)
 937 {
 938         struct btrfs_key key;
 939         int ret;
 940
 941         key.objectid = bytenr;
 942         if (parent) {
 943                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 944                 key.offset = parent;
 945         } else {
 946                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 947                 key.offset = root_objectid;
 948         }
 949
 950         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 951
 952         btrfs_release_path(root, path);
 953         return ret;
 954 }
 955
 956 static inline int extent_ref_type(u64 parent, u64 owner)
 957 {
 958         int type;
 959         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 960                 if (parent > 0)
 961                         type = BTRFS_SHARED_BLOCK_REF_KEY;
 962                 else
 963                         type = BTRFS_TREE_BLOCK_REF_KEY;
 964         } else {
 965                 if (parent > 0)
 966                         type = BTRFS_SHARED_DATA_REF_KEY;
 967                 else
 968                         type = BTRFS_EXTENT_DATA_REF_KEY;
 969         }
 970         return type;
 971 }
 972
 973 static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
 974
 975 {
 976         int level;
 977         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 978                 if (!path->nodes[level])
 979                         break;
 980                 if (path->slots[level] + 1 >=
 981                     btrfs_header_nritems(path->nodes[level]))
 982                         continue;
 983                 if (level == 0)
 984                         btrfs_item_key_to_cpu(path->nodes[level], key,
 985                                               path->slots[level] + 1);
 986                 else
 987                         btrfs_node_key_to_cpu(path->nodes[level], key,
 988                                               path->slots[level] + 1);
 989                 return 0;
 990         }
 991         return 1;
 992 }
 993
 994 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 995                                  struct btrfs_root *root,
 996                                  struct btrfs_path *path,
 997                                  struct btrfs_extent_inline_ref **ref_ret,
 998                                  u64 bytenr, u64 num_bytes,
 999                                  u64 parent, u64 root_objectid,
1000                                  u64 owner, u64 offset, int insert)
1001 {
1002         struct btrfs_key key;
1003         struct extent_buffer *leaf;
1004         struct btrfs_extent_item *ei;
1005         struct btrfs_extent_inline_ref *iref;
1006         u64 flags;
1007         u32 item_size;
1008         unsigned long ptr;
1009         unsigned long end;
1010         int extra_size;
1011         int type;
1012         int want;
1013         int ret;
1014         int err = 0;
1015         int skinny_metadata =
1016                 btrfs_fs_incompat(root->fs_info,
1017                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1018
1019         key.objectid = bytenr;
1020         key.type = BTRFS_EXTENT_ITEM_KEY;
1021         key.offset = num_bytes;
1022
1023         want = extent_ref_type(parent, owner);
1024         if (insert)
1025                 extra_size = btrfs_extent_inline_ref_size(want);
1026         else
1027                 extra_size = -1;
1028
1029         if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1030                 skinny_metadata = 1;
1031                 key.type = BTRFS_METADATA_ITEM_KEY;
1032                 key.offset = owner;
1033         } else if (skinny_metadata) {
1034                 skinny_metadata = 0;
1035         }
1036
1037 again:
1038         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1039         if (ret < 0) {
1040                 err = ret;
1041                 goto out;
1042         }
1043
1044         /*
1045          * We may be a newly converted file system which still has the old fat
1046          * extent entries for metadata, so try and see if we have one of those.
1047          */
1048         if (ret > 0 && skinny_metadata) {
1049                 skinny_metadata = 0;
1050                 if (path->slots[0]) {
1051                         path->slots[0]--;
1052                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1053                                               path->slots[0]);
1054                         if (key.objectid == bytenr &&
1055                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1056                             key.offset == num_bytes)
1057                                 ret = 0;
1058                 }
1059                 if (ret) {
1060                         key.type = BTRFS_EXTENT_ITEM_KEY;
1061                         key.offset = num_bytes;
1062                         goto again;
1063                 }
1064         }
1065
1066         if (ret) {
1067                 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1068                 return -ENOENT;
1069         }
1070
1071         BUG_ON(ret);
1072
1073         leaf = path->nodes[0];
1074         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1075 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1076         if (item_size < sizeof(*ei)) {
1077                 if (!insert) {
1078                         err = -ENOENT;
1079                         goto out;
1080                 }
1081                 ret = convert_extent_item_v0(trans, root, path, owner,
1082                                              extra_size);
1083                 if (ret < 0) {
1084                         err = ret;
1085                         goto out;
1086                 }
1087                 leaf = path->nodes[0];
1088                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1089         }
1090 #endif
1091         if (item_size < sizeof(*ei)) {
1092                 printf("Size is %u, needs to be %u, slot %d\n",
1093                        (unsigned)item_size,
1094                        (unsigned)sizeof(*ei), path->slots[0]);
1095                 btrfs_print_leaf(root, leaf);
1096                 return -EINVAL;
1097         }
1098         BUG_ON(item_size < sizeof(*ei));
1099
1100         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1101         flags = btrfs_extent_flags(leaf, ei);
1102
1103         ptr = (unsigned long)(ei + 1);
1104         end = (unsigned long)ei + item_size;
1105
1106         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1107                 ptr += sizeof(struct btrfs_tree_block_info);
1108                 BUG_ON(ptr > end);
1109         } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1110                 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1111                         return -EIO;
1112                 }
1113         }
1114
1115         err = -ENOENT;
1116         while (1) {
1117                 if (ptr >= end) {
1118                         WARN_ON(ptr > end);
1119                         break;
1120                 }
1121                 iref = (struct btrfs_extent_inline_ref *)ptr;
1122                 type = btrfs_extent_inline_ref_type(leaf, iref);
1123                 if (want < type)
1124                         break;
1125                 if (want > type) {
1126                         ptr += btrfs_extent_inline_ref_size(type);
1127                         continue;
1128                 }
1129
1130                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1131                         struct btrfs_extent_data_ref *dref;
1132                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1133                         if (match_extent_data_ref(leaf, dref, root_objectid,
1134                                                   owner, offset)) {
1135                                 err = 0;
1136                                 break;
1137                         }
1138                         if (hash_extent_data_ref_item(leaf, dref) <
1139                             hash_extent_data_ref(root_objectid, owner, offset))
1140                                 break;
1141                 } else {
1142                         u64 ref_offset;
1143                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1144                         if (parent > 0) {
1145                                 if (parent == ref_offset) {
1146                                         err = 0;
1147                                         break;
1148                                 }
1149                                 if (ref_offset < parent)
1150                                         break;
1151                         } else {
1152                                 if (root_objectid == ref_offset) {
1153                                         err = 0;
1154                                         break;
1155                                 }
1156                                 if (ref_offset < root_objectid)
1157                                         break;
1158                         }
1159                 }
1160                 ptr += btrfs_extent_inline_ref_size(type);
1161         }
1162         if (err == -ENOENT && insert) {
1163                 if (item_size + extra_size >=
1164                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1165                         err = -EAGAIN;
1166                         goto out;
1167                 }
1168                 /*
1169                  * To add new inline back ref, we have to make sure
1170                  * there is no corresponding back ref item.
1171                  * For simplicity, we just do not add new inline back
1172                  * ref if there is any back ref item.
1173                  */
1174                 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1175                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1176                         err = -EAGAIN;
1177                         goto out;
1178                 }
1179         }
1180         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1181 out:
1182         return err;
1183 }
1184
1185 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1186                                 struct btrfs_root *root,
1187                                 struct btrfs_path *path,
1188                                 struct btrfs_extent_inline_ref *iref,
1189                                 u64 parent, u64 root_objectid,
1190                                 u64 owner, u64 offset, int refs_to_add)
1191 {
1192         struct extent_buffer *leaf;
1193         struct btrfs_extent_item *ei;
1194         unsigned long ptr;
1195         unsigned long end;
1196         unsigned long item_offset;
1197         u64 refs;
1198         int size;
1199         int type;
1200         int ret;
1201
1202         leaf = path->nodes[0];
1203         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1204         item_offset = (unsigned long)iref - (unsigned long)ei;
1205
1206         type = extent_ref_type(parent, owner);
1207         size = btrfs_extent_inline_ref_size(type);
1208
1209         ret = btrfs_extend_item(trans, root, path, size);
1210         BUG_ON(ret);
1211
1212         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1213         refs = btrfs_extent_refs(leaf, ei);
1214         refs += refs_to_add;
1215         btrfs_set_extent_refs(leaf, ei, refs);
1216
1217         ptr = (unsigned long)ei + item_offset;
1218         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1219         if (ptr < end - size)
1220                 memmove_extent_buffer(leaf, ptr + size, ptr,
1221                                       end - size - ptr);
1222
1223         iref = (struct btrfs_extent_inline_ref *)ptr;
1224         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1225         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1226                 struct btrfs_extent_data_ref *dref;
1227                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1228                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1229                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1230                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1231                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1232         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1233                 struct btrfs_shared_data_ref *sref;
1234                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1235                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1236                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1237         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1238                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1239         } else {
1240                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1241         }
1242         btrfs_mark_buffer_dirty(leaf);
1243         return 0;
1244 }
1245
1246 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1247                                  struct btrfs_root *root,
1248                                  struct btrfs_path *path,
1249                                  struct btrfs_extent_inline_ref **ref_ret,
1250                                  u64 bytenr, u64 num_bytes, u64 parent,
1251                                  u64 root_objectid, u64 owner, u64 offset)
1252 {
1253         int ret;
1254
1255         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1256                                            bytenr, num_bytes, parent,
1257                                            root_objectid, owner, offset, 0);
1258         if (ret != -ENOENT)
1259                 return ret;
1260
1261         btrfs_release_path(root, path);
1262         *ref_ret = NULL;
1263
1264         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1265                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1266                                             root_objectid);
1267         } else {
1268                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1269                                              root_objectid, owner, offset);
1270         }
1271         return ret;
1272 }
1273
1274 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1275                                  struct btrfs_root *root,
1276                                  struct btrfs_path *path,
1277                                  struct btrfs_extent_inline_ref *iref,
1278                                  int refs_to_mod)
1279 {
1280         struct extent_buffer *leaf;
1281         struct btrfs_extent_item *ei;
1282         struct btrfs_extent_data_ref *dref = NULL;
1283         struct btrfs_shared_data_ref *sref = NULL;
1284         unsigned long ptr;
1285         unsigned long end;
1286         u32 item_size;
1287         int size;
1288         int type;
1289         int ret;
1290         u64 refs;
1291
1292         leaf = path->nodes[0];
1293         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1294         refs = btrfs_extent_refs(leaf, ei);
1295         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1296         refs += refs_to_mod;
1297         btrfs_set_extent_refs(leaf, ei, refs);
1298
1299         type = btrfs_extent_inline_ref_type(leaf, iref);
1300
1301         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1302                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1303                 refs = btrfs_extent_data_ref_count(leaf, dref);
1304         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1305                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1306                 refs = btrfs_shared_data_ref_count(leaf, sref);
1307         } else {
1308                 refs = 1;
1309                 BUG_ON(refs_to_mod != -1);
1310         }
1311
1312         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1313         refs += refs_to_mod;
1314
1315         if (refs > 0) {
1316                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1317                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1318                 else
1319                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1320         } else {
1321                 size =  btrfs_extent_inline_ref_size(type);
1322                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1323                 ptr = (unsigned long)iref;
1324                 end = (unsigned long)ei + item_size;
1325                 if (ptr + size < end)
1326                         memmove_extent_buffer(leaf, ptr, ptr + size,
1327                                               end - ptr - size);
1328                 item_size -= size;
1329                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1330                 BUG_ON(ret);
1331         }
1332         btrfs_mark_buffer_dirty(leaf);
1333         return 0;
1334 }
1335
1336 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1337                                  struct btrfs_root *root,
1338                                  struct btrfs_path *path,
1339                                  u64 bytenr, u64 num_bytes, u64 parent,
1340                                  u64 root_objectid, u64 owner,
1341                                  u64 offset, int refs_to_add)
1342 {
1343         struct btrfs_extent_inline_ref *iref;
1344         int ret;
1345
1346         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1347                                            bytenr, num_bytes, parent,
1348                                            root_objectid, owner, offset, 1);
1349         if (ret == 0) {
1350                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1351                 ret = update_inline_extent_backref(trans, root, path, iref,
1352                                                    refs_to_add);
1353         } else if (ret == -ENOENT) {
1354                 ret = setup_inline_extent_backref(trans, root, path, iref,
1355                                                   parent, root_objectid,
1356                                                   owner, offset, refs_to_add);
1357         }
1358         return ret;
1359 }
1360
1361 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1362                                  struct btrfs_root *root,
1363                                  struct btrfs_path *path,
1364                                  u64 bytenr, u64 parent, u64 root_objectid,
1365                                  u64 owner, u64 offset, int refs_to_add)
1366 {
1367         int ret;
1368
1369         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1370                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1371                                              parent, root_objectid,
1372                                              owner, offset, refs_to_add);
1373         } else {
1374                 BUG_ON(refs_to_add != 1);
1375                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1376                                             parent, root_objectid);
1377         }
1378         return ret;
1379 }
1380
1381 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1382                                  struct btrfs_root *root,
1383                                  struct btrfs_path *path,
1384                                  struct btrfs_extent_inline_ref *iref,
1385                                  int refs_to_drop, int is_data)
1386 {
1387         int ret;
1388
1389         BUG_ON(!is_data && refs_to_drop != 1);
1390         if (iref) {
1391                 ret = update_inline_extent_backref(trans, root, path, iref,
1392                                                    -refs_to_drop);
1393         } else if (is_data) {
1394                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1395         } else {
1396                 ret = btrfs_del_item(trans, root, path);
1397         }
1398         return ret;
1399 }
1400
1401 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1402                          struct btrfs_root *root,
1403                          u64 bytenr, u64 num_bytes, u64 parent,
1404                          u64 root_objectid, u64 owner, u64 offset)
1405 {
1406         struct btrfs_path *path;
1407         struct extent_buffer *leaf;
1408         struct btrfs_extent_item *item;
1409         u64 refs;
1410         int ret;
1411         int err = 0;
1412
1413         path = btrfs_alloc_path();
1414         if (!path)
1415                 return -ENOMEM;
1416
1417         path->reada = 1;
1418         path->leave_spinning = 1;
1419
1420         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1421                                            path, bytenr, num_bytes, parent,
1422                                            root_objectid, owner, offset, 1);
1423         if (ret == 0)
1424                 goto out;
1425
1426         if (ret != -EAGAIN) {
1427                 err = ret;
1428                 goto out;
1429         }
1430
1431         leaf = path->nodes[0];
1432         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1433         refs = btrfs_extent_refs(leaf, item);
1434         btrfs_set_extent_refs(leaf, item, refs + 1);
1435
1436         btrfs_mark_buffer_dirty(leaf);
1437         btrfs_release_path(root->fs_info->extent_root, path);
1438
1439         path->reada = 1;
1440         path->leave_spinning = 1;
1441
1442         /* now insert the actual backref */
1443         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1444                                     path, bytenr, parent, root_objectid,
1445                                     owner, offset, 1);
1446         if (ret)
1447                 err = ret;
1448 out:
1449         btrfs_free_path(path);
1450         finish_current_insert(trans, root->fs_info->extent_root);
1451         del_pending_extents(trans, root->fs_info->extent_root);
1452         BUG_ON(err);
1453         return err;
1454 }
1455
1456 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1457                          struct btrfs_root *root)
1458 {
1459         finish_current_insert(trans, root->fs_info->extent_root);
1460         del_pending_extents(trans, root->fs_info->extent_root);
1461         return 0;
1462 }
1463
1464 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1465                              struct btrfs_root *root, u64 bytenr,
1466                              u64 offset, int metadata, u64 *refs, u64 *flags)
1467 {
1468         struct btrfs_path *path;
1469         int ret;
1470         struct btrfs_key key;
1471         struct extent_buffer *l;
1472         struct btrfs_extent_item *item;
1473         u32 item_size;
1474         u64 num_refs;
1475         u64 extent_flags;
1476
1477         if (metadata &&
1478             !btrfs_fs_incompat(root->fs_info,
1479                                BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
1480                 offset = root->leafsize;
1481                 metadata = 0;
1482         }
1483
1484         path = btrfs_alloc_path();
1485         path->reada = 1;
1486
1487         key.objectid = bytenr;
1488         key.offset = offset;
1489         if (metadata)
1490                 key.type = BTRFS_METADATA_ITEM_KEY;
1491         else
1492                 key.type = BTRFS_EXTENT_ITEM_KEY;
1493
1494 again:
1495         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1496                                 0, 0);
1497         if (ret < 0)
1498                 goto out;
1499
1500         /*
1501          * Deal with the fact that we may have mixed SKINNY and normal refs.  If
1502          * we didn't find what we wanted check and see if we have a normal ref
1503          * right next to us, or re-search if we are on the edge of the leaf just
1504          * to make sure.
1505          */
1506         if (ret > 0 && metadata) {
1507                 if (path->slots) {
1508                         path->slots[0]--;
1509                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1510                                               path->slots[0]);
1511                         if (key.objectid == bytenr &&
1512                             key.type == BTRFS_METADATA_ITEM_KEY)
1513                                 ret = 0;
1514                 }
1515
1516                 if (ret) {
1517                         btrfs_release_path(root, path);
1518                         key.type = BTRFS_EXTENT_ITEM_KEY;
1519                         key.offset = root->leafsize;
1520                         metadata = 0;
1521                         goto again;
1522                 }
1523         }
1524
1525         if (ret != 0) {
1526                 ret = -EIO;
1527                 goto out;
1528         }
1529
1530         l = path->nodes[0];
1531         item_size = btrfs_item_size_nr(l, path->slots[0]);
1532         if (item_size >= sizeof(*item)) {
1533                 item = btrfs_item_ptr(l, path->slots[0],
1534                                       struct btrfs_extent_item);
1535                 num_refs = btrfs_extent_refs(l, item);
1536                 extent_flags = btrfs_extent_flags(l, item);
1537         } else {
1538 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1539                         struct btrfs_extent_item_v0 *ei0;
1540                         BUG_ON(item_size != sizeof(*ei0));
1541                         ei0 = btrfs_item_ptr(l, path->slots[0],
1542                                              struct btrfs_extent_item_v0);
1543                         num_refs = btrfs_extent_refs_v0(l, ei0);
1544                         /* FIXME: this isn't correct for data */
1545                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1546 #else
1547                         BUG();
1548 #endif
1549         }
1550         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1551         if (refs)
1552                 *refs = num_refs;
1553         if (flags)
1554                 *flags = extent_flags;
1555 out:
1556         btrfs_free_path(path);
1557         return 0;
1558 }
1559
1560 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1561                           struct btrfs_root *root,
1562                           u64 bytenr, int level, u64 flags)
1563 {
1564         struct btrfs_path *path;
1565         int ret;
1566         struct btrfs_key key;
1567         struct extent_buffer *l;
1568         struct btrfs_extent_item *item;
1569         u32 item_size;
1570         int skinny_metadata =
1571                 btrfs_fs_incompat(root->fs_info,
1572                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1573
1574         path = btrfs_alloc_path();
1575         path->reada = 1;
1576
1577         key.objectid = bytenr;
1578         if (skinny_metadata) {
1579                 key.offset = level;
1580                 key.type = BTRFS_METADATA_ITEM_KEY;
1581         } else {
1582                 key.offset = root->leafsize;
1583                 key.type = BTRFS_EXTENT_ITEM_KEY;
1584         }
1585
1586 again:
1587         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1588                                 0, 0);
1589         if (ret < 0)
1590                 goto out;
1591
1592         if (ret > 0 && skinny_metadata) {
1593                 skinny_metadata = 0;
1594                 if (path->slots[0]--) {
1595                         path->slots[0]--;
1596                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1597                                               path->slots[0]);
1598                         if (key.objectid == bytenr &&
1599                             key.offset == root->leafsize &&
1600                             key.type == BTRFS_EXTENT_ITEM_KEY)
1601                                 ret = 0;
1602                 }
1603                 if (ret) {
1604                         btrfs_release_path(root, path);
1605                         key.offset = root->leafsize;
1606                         key.type = BTRFS_EXTENT_ITEM_KEY;
1607                         goto again;
1608                 }
1609         }
1610
1611         if (ret != 0) {
1612                 btrfs_print_leaf(root, path->nodes[0]);
1613                 printk("failed to find block number %Lu\n",
1614                         (unsigned long long)bytenr);
1615                 BUG();
1616         }
1617         l = path->nodes[0];
1618         item_size = btrfs_item_size_nr(l, path->slots[0]);
1619 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1620         if (item_size < sizeof(*item)) {
1621                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1622                                              path, (u64)-1, 0);
1623                 if (ret < 0)
1624                         goto out;
1625
1626                 l = path->nodes[0];
1627                 item_size = btrfs_item_size_nr(l, path->slots[0]);
1628         }
1629 #endif
1630         BUG_ON(item_size < sizeof(*item));
1631         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1632         flags |= btrfs_extent_flags(l, item);
1633         btrfs_set_extent_flags(l, item, flags);
1634 out:
1635         btrfs_free_path(path);
1636         finish_current_insert(trans, root->fs_info->extent_root);
1637         del_pending_extents(trans, root->fs_info->extent_root);
1638         return ret;
1639 }
1640
1641 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1642                            struct btrfs_root *root,
1643                            struct extent_buffer *buf,
1644                            int record_parent, int inc)
1645 {
1646         u64 bytenr;
1647         u64 num_bytes;
1648         u64 parent;
1649         u64 ref_root;
1650         u32 nritems;
1651         struct btrfs_key key;
1652         struct btrfs_file_extent_item *fi;
1653         int i;
1654         int level;
1655         int ret = 0;
1656         int (*process_func)(struct btrfs_trans_handle *trans,
1657                             struct btrfs_root *root,
1658                             u64, u64, u64, u64, u64, u64);
1659
1660         ref_root = btrfs_header_owner(buf);
1661         nritems = btrfs_header_nritems(buf);
1662         level = btrfs_header_level(buf);
1663
1664         if (!root->ref_cows && level == 0)
1665                 return 0;
1666
1667         if (inc)
1668                 process_func = btrfs_inc_extent_ref;
1669         else
1670                 process_func = btrfs_free_extent;
1671
1672         if (record_parent)
1673                 parent = buf->start;
1674         else
1675                 parent = 0;
1676
1677         for (i = 0; i < nritems; i++) {
1678                 cond_resched();
1679                 if (level == 0) {
1680                         btrfs_item_key_to_cpu(buf, &key, i);
1681                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1682                                 continue;
1683                         fi = btrfs_item_ptr(buf, i,
1684                                             struct btrfs_file_extent_item);
1685                         if (btrfs_file_extent_type(buf, fi) ==
1686                             BTRFS_FILE_EXTENT_INLINE)
1687                                 continue;
1688                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1689                         if (bytenr == 0)
1690                                 continue;
1691
1692                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1693                         key.offset -= btrfs_file_extent_offset(buf, fi);
1694                         ret = process_func(trans, root, bytenr, num_bytes,
1695                                            parent, ref_root, key.objectid,
1696                                            key.offset);
1697                         if (ret) {
1698                                 WARN_ON(1);
1699                                 goto fail;
1700                         }
1701                 } else {
1702                         bytenr = btrfs_node_blockptr(buf, i);
1703                         num_bytes = btrfs_level_size(root, level - 1);
1704                         ret = process_func(trans, root, bytenr, num_bytes,
1705                                            parent, ref_root, level - 1, 0);
1706                         if (ret) {
1707                                 WARN_ON(1);
1708                                 goto fail;
1709                         }
1710                 }
1711         }
1712         return 0;
1713 fail:
1714         WARN_ON(1);
1715         return ret;
1716 }
1717
1718 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1719                   struct extent_buffer *buf, int record_parent)
1720 {
1721         return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1722 }
1723
1724 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1725                   struct extent_buffer *buf, int record_parent)
1726 {
1727         return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1728 }
1729
1730 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1731                                  struct btrfs_root *root,
1732                                  struct btrfs_path *path,
1733                                  struct btrfs_block_group_cache *cache)
1734 {
1735         int ret;
1736         int pending_ret;
1737         struct btrfs_root *extent_root = root->fs_info->extent_root;
1738         unsigned long bi;
1739         struct extent_buffer *leaf;
1740
1741         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1742         if (ret < 0)
1743                 goto fail;
1744         BUG_ON(ret);
1745
1746         leaf = path->nodes[0];
1747         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1748         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1749         btrfs_mark_buffer_dirty(leaf);
1750         btrfs_release_path(extent_root, path);
1751 fail:
1752         finish_current_insert(trans, extent_root);
1753         pending_ret = del_pending_extents(trans, extent_root);
1754         if (ret)
1755                 return ret;
1756         if (pending_ret)
1757                 return pending_ret;
1758         return 0;
1759
1760 }
1761
1762 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1763                                    struct btrfs_root *root)
1764 {
1765         struct extent_io_tree *block_group_cache;
1766         struct btrfs_block_group_cache *cache;
1767         int ret;
1768         struct btrfs_path *path;
1769         u64 last = 0;
1770         u64 start;
1771         u64 end;
1772         u64 ptr;
1773
1774         block_group_cache = &root->fs_info->block_group_cache;
1775         path = btrfs_alloc_path();
1776         if (!path)
1777                 return -ENOMEM;
1778
1779         while(1) {
1780                 ret = find_first_extent_bit(block_group_cache, last,
1781                                             &start, &end, BLOCK_GROUP_DIRTY);
1782                 if (ret) {
1783                         if (last == 0)
1784                                 break;
1785                         last = 0;
1786                         continue;
1787                 }
1788
1789                 last = end + 1;
1790                 ret = get_state_private(block_group_cache, start, &ptr);
1791                 BUG_ON(ret);
1792
1793                 clear_extent_bits(block_group_cache, start, end,
1794                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1795
1796                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1797                 ret = write_one_cache_group(trans, root, path, cache);
1798         }
1799         btrfs_free_path(path);
1800         return 0;
1801 }
1802
1803 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1804                                                   u64 flags)
1805 {
1806         struct list_head *head = &info->space_info;
1807         struct list_head *cur;
1808         struct btrfs_space_info *found;
1809         list_for_each(cur, head) {
1810                 found = list_entry(cur, struct btrfs_space_info, list);
1811                 if (found->flags & flags)
1812                         return found;
1813         }
1814         return NULL;
1815
1816 }
1817
1818 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1819                              u64 total_bytes, u64 bytes_used,
1820                              struct btrfs_space_info **space_info)
1821 {
1822         struct btrfs_space_info *found;
1823
1824         found = __find_space_info(info, flags);
1825         if (found) {
1826                 found->total_bytes += total_bytes;
1827                 found->bytes_used += bytes_used;
1828                 if (found->total_bytes < found->bytes_used) {
1829                         fprintf(stderr, "warning, bad space info total_bytes "
1830                                 "%llu used %llu\n",
1831                                (unsigned long long)found->total_bytes,
1832                                (unsigned long long)found->bytes_used);
1833                 }
1834                 *space_info = found;
1835                 return 0;
1836         }
1837         found = kmalloc(sizeof(*found), GFP_NOFS);
1838         if (!found)
1839                 return -ENOMEM;
1840
1841         list_add(&found->list, &info->space_info);
1842         found->flags = flags;
1843         found->total_bytes = total_bytes;
1844         found->bytes_used = bytes_used;
1845         found->bytes_pinned = 0;
1846         found->full = 0;
1847         *space_info = found;
1848         return 0;
1849 }
1850
1851
1852 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1853 {
1854         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1855                                    BTRFS_BLOCK_GROUP_RAID1 |
1856                                    BTRFS_BLOCK_GROUP_RAID10 |
1857                                    BTRFS_BLOCK_GROUP_RAID5 |
1858                                    BTRFS_BLOCK_GROUP_RAID6 |
1859                                    BTRFS_BLOCK_GROUP_DUP);
1860         if (extra_flags) {
1861                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1862                         fs_info->avail_data_alloc_bits |= extra_flags;
1863                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1864                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1865                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1866                         fs_info->avail_system_alloc_bits |= extra_flags;
1867         }
1868 }
1869
1870 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1871                           struct btrfs_root *extent_root, u64 alloc_bytes,
1872                           u64 flags)
1873 {
1874         struct btrfs_space_info *space_info;
1875         u64 thresh;
1876         u64 start;
1877         u64 num_bytes;
1878         int ret;
1879
1880         space_info = __find_space_info(extent_root->fs_info, flags);
1881         if (!space_info) {
1882                 ret = update_space_info(extent_root->fs_info, flags,
1883                                         0, 0, &space_info);
1884                 BUG_ON(ret);
1885         }
1886         BUG_ON(!space_info);
1887
1888         if (space_info->full)
1889                 return 0;
1890
1891         thresh = div_factor(space_info->total_bytes, 7);
1892         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1893             thresh)
1894                 return 0;
1895
1896         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1897                                 space_info->flags);
1898         if (ret == -ENOSPC) {
1899                 space_info->full = 1;
1900                 return 0;
1901         }
1902
1903         BUG_ON(ret);
1904
1905         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1906                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1907         BUG_ON(ret);
1908         return 0;
1909 }
1910
1911 static int update_block_group(struct btrfs_trans_handle *trans,
1912                               struct btrfs_root *root,
1913                               u64 bytenr, u64 num_bytes, int alloc,
1914                               int mark_free)
1915 {
1916         struct btrfs_block_group_cache *cache;
1917         struct btrfs_fs_info *info = root->fs_info;
1918         u64 total = num_bytes;
1919         u64 old_val;
1920         u64 byte_in_group;
1921         u64 start;
1922         u64 end;
1923
1924         /* block accounting for super block */
1925         old_val = btrfs_super_bytes_used(info->super_copy);
1926         if (alloc)
1927                 old_val += num_bytes;
1928         else
1929                 old_val -= num_bytes;
1930         btrfs_set_super_bytes_used(info->super_copy, old_val);
1931
1932         /* block accounting for root item */
1933         old_val = btrfs_root_used(&root->root_item);
1934         if (alloc)
1935                 old_val += num_bytes;
1936         else
1937                 old_val -= num_bytes;
1938         btrfs_set_root_used(&root->root_item, old_val);
1939
1940         while(total) {
1941                 cache = btrfs_lookup_block_group(info, bytenr);
1942                 if (!cache) {
1943                         return -1;
1944                 }
1945                 byte_in_group = bytenr - cache->key.objectid;
1946                 WARN_ON(byte_in_group > cache->key.offset);
1947                 start = cache->key.objectid;
1948                 end = start + cache->key.offset - 1;
1949                 set_extent_bits(&info->block_group_cache, start, end,
1950                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1951
1952                 old_val = btrfs_block_group_used(&cache->item);
1953                 num_bytes = min(total, cache->key.offset - byte_in_group);
1954
1955                 if (alloc) {
1956                         old_val += num_bytes;
1957                         cache->space_info->bytes_used += num_bytes;
1958                 } else {
1959                         old_val -= num_bytes;
1960                         cache->space_info->bytes_used -= num_bytes;
1961                         if (mark_free) {
1962                                 set_extent_dirty(&info->free_space_cache,
1963                                                  bytenr, bytenr + num_bytes - 1,
1964                                                  GFP_NOFS);
1965                         }
1966                 }
1967                 btrfs_set_block_group_used(&cache->item, old_val);
1968                 total -= num_bytes;
1969                 bytenr += num_bytes;
1970         }
1971         return 0;
1972 }
1973
1974 static int update_pinned_extents(struct btrfs_root *root,
1975                                 u64 bytenr, u64 num, int pin)
1976 {
1977         u64 len;
1978         struct btrfs_block_group_cache *cache;
1979         struct btrfs_fs_info *fs_info = root->fs_info;
1980
1981         if (pin) {
1982                 set_extent_dirty(&fs_info->pinned_extents,
1983                                 bytenr, bytenr + num - 1, GFP_NOFS);
1984         } else {
1985                 clear_extent_dirty(&fs_info->pinned_extents,
1986                                 bytenr, bytenr + num - 1, GFP_NOFS);
1987         }
1988         while (num > 0) {
1989                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1990                 if (!cache) {
1991                         len = min((u64)root->sectorsize, num);
1992                         goto next;
1993                 }
1994                 WARN_ON(!cache);
1995                 len = min(num, cache->key.offset -
1996                           (bytenr - cache->key.objectid));
1997                 if (pin) {
1998                         cache->pinned += len;
1999                         cache->space_info->bytes_pinned += len;
2000                         fs_info->total_pinned += len;
2001                 } else {
2002                         cache->pinned -= len;
2003                         cache->space_info->bytes_pinned -= len;
2004                         fs_info->total_pinned -= len;
2005                 }
2006 next:
2007                 bytenr += len;
2008                 num -= len;
2009         }
2010         return 0;
2011 }
2012
2013 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2014 {
2015         u64 last = 0;
2016         u64 start;
2017         u64 end;
2018         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2019         int ret;
2020
2021         while(1) {
2022                 ret = find_first_extent_bit(pinned_extents, last,
2023                                             &start, &end, EXTENT_DIRTY);
2024                 if (ret)
2025                         break;
2026                 set_extent_dirty(copy, start, end, GFP_NOFS);
2027                 last = end + 1;
2028         }
2029         return 0;
2030 }
2031
2032 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2033                                struct btrfs_root *root,
2034                                struct extent_io_tree *unpin)
2035 {
2036         u64 start;
2037         u64 end;
2038         int ret;
2039         struct extent_io_tree *free_space_cache;
2040         free_space_cache = &root->fs_info->free_space_cache;
2041
2042         while(1) {
2043                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2044                                             EXTENT_DIRTY);
2045                 if (ret)
2046                         break;
2047                 update_pinned_extents(root, start, end + 1 - start, 0);
2048                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2049                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2050         }
2051         return 0;
2052 }
2053
2054 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2055 {
2056         u64 start;
2057         u64 end;
2058         int ret;
2059
2060         ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2061                                     &end, EXTENT_LOCKED);
2062         if (!ret) {
2063                 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2064                                             EXTENT_LOCKED);
2065         }
2066         return ret == 0;
2067
2068 }
2069 static int finish_current_insert(struct btrfs_trans_handle *trans,
2070                                  struct btrfs_root *extent_root)
2071 {
2072         u64 start;
2073         u64 end;
2074         u64 priv;
2075         struct btrfs_fs_info *info = extent_root->fs_info;
2076         struct btrfs_path *path;
2077         struct pending_extent_op *extent_op;
2078         struct btrfs_key key;
2079         int ret;
2080         int skinny_metadata =
2081                 btrfs_fs_incompat(extent_root->fs_info,
2082                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2083
2084         path = btrfs_alloc_path();
2085
2086         while(1) {
2087                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2088                                             &end, EXTENT_LOCKED);
2089                 if (ret)
2090                         break;
2091
2092                 ret = get_state_private(&info->extent_ins, start, &priv);
2093                 BUG_ON(ret);
2094                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2095
2096                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2097                         key.objectid = start;
2098                         if (skinny_metadata) {
2099                                 key.offset = extent_op->level;
2100                                 key.type = BTRFS_METADATA_ITEM_KEY;
2101                         } else {
2102                                 key.offset = extent_op->num_bytes;
2103                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2104                         }
2105                         ret = alloc_reserved_tree_block(trans, extent_root,
2106                                                 extent_root->root_key.objectid,
2107                                                 trans->transid,
2108                                                 extent_op->flags,
2109                                                 &extent_op->key,
2110                                                 extent_op->level, &key);
2111                 } else {
2112                         BUG_ON(1);
2113                 }
2114
2115                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2116                                   GFP_NOFS);
2117                 kfree(extent_op);
2118         }
2119         btrfs_free_path(path);
2120         return 0;
2121 }
2122
2123 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2124                           struct btrfs_root *root,
2125                           u64 bytenr, u64 num_bytes, int is_data)
2126 {
2127         int err = 0;
2128         struct extent_buffer *buf;
2129
2130         if (is_data)
2131                 goto pinit;
2132
2133         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2134         if (!buf)
2135                 goto pinit;
2136
2137         /* we can reuse a block if it hasn't been written
2138          * and it is from this transaction.  We can't
2139          * reuse anything from the tree log root because
2140          * it has tiny sub-transactions.
2141          */
2142         if (btrfs_buffer_uptodate(buf, 0)) {
2143                 u64 header_owner = btrfs_header_owner(buf);
2144                 u64 header_transid = btrfs_header_generation(buf);
2145                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2146                     header_transid == trans->transid &&
2147                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2148                         clean_tree_block(NULL, root, buf);
2149                         free_extent_buffer(buf);
2150                         return 1;
2151                 }
2152         }
2153         free_extent_buffer(buf);
2154 pinit:
2155         update_pinned_extents(root, bytenr, num_bytes, 1);
2156
2157         BUG_ON(err < 0);
2158         return 0;
2159 }
2160
2161 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2162                        u64 bytenr, u64 num_bytes)
2163 {
2164         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2165 }
2166
2167 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2168                         u64 bytenr, u64 num_bytes)
2169 {
2170         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2171 }
2172
2173 /*
2174  * remove an extent from the root, returns 0 on success
2175  */
2176 static int __free_extent(struct btrfs_trans_handle *trans,
2177                          struct btrfs_root *root,
2178                          u64 bytenr, u64 num_bytes, u64 parent,
2179                          u64 root_objectid, u64 owner_objectid,
2180                          u64 owner_offset, int refs_to_drop)
2181 {
2182
2183         struct btrfs_key key;
2184         struct btrfs_path *path;
2185         struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2186         struct btrfs_root *extent_root = root->fs_info->extent_root;
2187         struct extent_buffer *leaf;
2188         struct btrfs_extent_item *ei;
2189         struct btrfs_extent_inline_ref *iref;
2190         int ret;
2191         int is_data;
2192         int extent_slot = 0;
2193         int found_extent = 0;
2194         int num_to_del = 1;
2195         u32 item_size;
2196         u64 refs;
2197         int skinny_metadata =
2198                 btrfs_fs_incompat(extent_root->fs_info,
2199                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2200
2201         if (root->fs_info->free_extent_hook) {
2202                 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2203                                                 parent, root_objectid, owner_objectid,
2204                                                 owner_offset, refs_to_drop);
2205
2206         }
2207         path = btrfs_alloc_path();
2208         if (!path)
2209                 return -ENOMEM;
2210
2211         path->reada = 1;
2212         path->leave_spinning = 1;
2213
2214         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2215         if (is_data)
2216                 skinny_metadata = 0;
2217         BUG_ON(!is_data && refs_to_drop != 1);
2218
2219         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2220                                     bytenr, num_bytes, parent,
2221                                     root_objectid, owner_objectid,
2222                                     owner_offset);
2223         if (ret == 0) {
2224                 extent_slot = path->slots[0];
2225                 while (extent_slot >= 0) {
2226                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2227                                               extent_slot);
2228                         if (key.objectid != bytenr)
2229                                 break;
2230                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2231                             key.offset == num_bytes) {
2232                                 found_extent = 1;
2233                                 break;
2234                         }
2235                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
2236                             key.offset == owner_objectid) {
2237                                 found_extent = 1;
2238                                 break;
2239                         }
2240                         if (path->slots[0] - extent_slot > 5)
2241                                 break;
2242                         extent_slot--;
2243                 }
2244 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2245                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2246                 if (found_extent && item_size < sizeof(*ei))
2247                         found_extent = 0;
2248 #endif
2249                 if (!found_extent) {
2250                         BUG_ON(iref);
2251                         ret = remove_extent_backref(trans, extent_root, path,
2252                                                     NULL, refs_to_drop,
2253                                                     is_data);
2254                         BUG_ON(ret);
2255                         btrfs_release_path(extent_root, path);
2256                         path->leave_spinning = 1;
2257
2258                         key.objectid = bytenr;
2259
2260                         if (skinny_metadata) {
2261                                 key.type = BTRFS_METADATA_ITEM_KEY;
2262                                 key.offset = owner_objectid;
2263                         } else {
2264                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2265                                 key.offset = num_bytes;
2266                         }
2267
2268                         ret = btrfs_search_slot(trans, extent_root,
2269                                                 &key, path, -1, 1);
2270                         if (ret > 0 && skinny_metadata && path->slots[0]) {
2271                                 path->slots[0]--;
2272                                 btrfs_item_key_to_cpu(path->nodes[0],
2273                                                       &key,
2274                                                       path->slots[0]);
2275                                 if (key.objectid == bytenr &&
2276                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2277                                     key.offset == num_bytes)
2278                                         ret = 0;
2279                         }
2280
2281                         if (ret > 0 && skinny_metadata) {
2282                                 skinny_metadata = 0;
2283                                 btrfs_release_path(extent_root, path);
2284                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2285                                 key.offset = num_bytes;
2286                                 ret = btrfs_search_slot(trans, extent_root,
2287                                                         &key, path, -1, 1);
2288                         }
2289
2290                         if (ret) {
2291                                 printk(KERN_ERR "umm, got %d back from search"
2292                                        ", was looking for %llu\n", ret,
2293                                        (unsigned long long)bytenr);
2294                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2295                         }
2296                         BUG_ON(ret);
2297                         extent_slot = path->slots[0];
2298                 }
2299         } else {
2300                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2301                        "parent %llu root %llu  owner %llu offset %llu\n",
2302                        (unsigned long long)bytenr,
2303                        (unsigned long long)parent,
2304                        (unsigned long long)root_objectid,
2305                        (unsigned long long)owner_objectid,
2306                        (unsigned long long)owner_offset);
2307                 ret = -EIO;
2308                 goto fail;
2309         }
2310
2311         leaf = path->nodes[0];
2312         item_size = btrfs_item_size_nr(leaf, extent_slot);
2313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2314         if (item_size < sizeof(*ei)) {
2315                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2316                 ret = convert_extent_item_v0(trans, extent_root, path,
2317                                              owner_objectid, 0);
2318                 BUG_ON(ret < 0);
2319
2320                 btrfs_release_path(extent_root, path);
2321                 path->leave_spinning = 1;
2322
2323                 key.objectid = bytenr;
2324                 key.type = BTRFS_EXTENT_ITEM_KEY;
2325                 key.offset = num_bytes;
2326
2327                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2328                                         -1, 1);
2329                 if (ret) {
2330                         printk(KERN_ERR "umm, got %d back from search"
2331                                ", was looking for %llu\n", ret,
2332                                (unsigned long long)bytenr);
2333                         btrfs_print_leaf(extent_root, path->nodes[0]);
2334                 }
2335                 BUG_ON(ret);
2336                 extent_slot = path->slots[0];
2337                 leaf = path->nodes[0];
2338                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2339         }
2340 #endif
2341         BUG_ON(item_size < sizeof(*ei));
2342         ei = btrfs_item_ptr(leaf, extent_slot,
2343                             struct btrfs_extent_item);
2344         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2345             key.type == BTRFS_EXTENT_ITEM_KEY) {
2346                 struct btrfs_tree_block_info *bi;
2347                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2348                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2349                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2350         }
2351
2352         refs = btrfs_extent_refs(leaf, ei);
2353         BUG_ON(refs < refs_to_drop);
2354         refs -= refs_to_drop;
2355
2356         if (refs > 0) {
2357                 /*
2358                  * In the case of inline back ref, reference count will
2359                  * be updated by remove_extent_backref
2360                  */
2361                 if (iref) {
2362                         BUG_ON(!found_extent);
2363                 } else {
2364                         btrfs_set_extent_refs(leaf, ei, refs);
2365                         btrfs_mark_buffer_dirty(leaf);
2366                 }
2367                 if (found_extent) {
2368                         ret = remove_extent_backref(trans, extent_root, path,
2369                                                     iref, refs_to_drop,
2370                                                     is_data);
2371                         BUG_ON(ret);
2372                 }
2373         } else {
2374                 int mark_free = 0;
2375                 int pin = 1;
2376
2377                 if (found_extent) {
2378                         BUG_ON(is_data && refs_to_drop !=
2379                                extent_data_ref_count(root, path, iref));
2380                         if (iref) {
2381                                 BUG_ON(path->slots[0] != extent_slot);
2382                         } else {
2383                                 BUG_ON(path->slots[0] != extent_slot + 1);
2384                                 path->slots[0] = extent_slot;
2385                                 num_to_del = 2;
2386                         }
2387                 }
2388
2389                 if (ops && ops->free_extent) {
2390                         ret = ops->free_extent(root, bytenr, num_bytes);
2391                         if (ret > 0) {
2392                                 pin = 0;
2393                                 mark_free = 0;
2394                         }
2395                 }
2396
2397                 if (pin) {
2398                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2399                                              is_data);
2400                         if (ret > 0)
2401                                 mark_free = 1;
2402                         BUG_ON(ret < 0);
2403                 }
2404
2405                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2406                                       num_to_del);
2407                 BUG_ON(ret);
2408                 btrfs_release_path(extent_root, path);
2409
2410                 if (is_data) {
2411                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2412                         BUG_ON(ret);
2413                 }
2414
2415                 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2416         }
2417 fail:
2418         btrfs_free_path(path);
2419         finish_current_insert(trans, extent_root);
2420         return ret;
2421 }
2422
2423 /*
2424  * find all the blocks marked as pending in the radix tree and remove
2425  * them from the extent map
2426  */
2427 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2428                                btrfs_root *extent_root)
2429 {
2430         int ret;
2431         int err = 0;
2432         u64 start;
2433         u64 end;
2434         u64 priv;
2435         struct extent_io_tree *pending_del;
2436         struct extent_io_tree *extent_ins;
2437         struct pending_extent_op *extent_op;
2438
2439         extent_ins = &extent_root->fs_info->extent_ins;
2440         pending_del = &extent_root->fs_info->pending_del;
2441
2442         while(1) {
2443                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2444                                             EXTENT_LOCKED);
2445                 if (ret)
2446                         break;
2447
2448                 ret = get_state_private(pending_del, start, &priv);
2449                 BUG_ON(ret);
2450                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2451
2452                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2453                                   GFP_NOFS);
2454
2455                 if (!test_range_bit(extent_ins, start, end,
2456                                     EXTENT_LOCKED, 0)) {
2457                         ret = __free_extent(trans, extent_root,
2458                                             start, end + 1 - start, 0,
2459                                             extent_root->root_key.objectid,
2460                                             extent_op->level, 0, 1);
2461                         kfree(extent_op);
2462                 } else {
2463                         kfree(extent_op);
2464                         ret = get_state_private(extent_ins, start, &priv);
2465                         BUG_ON(ret);
2466                         extent_op = (struct pending_extent_op *)
2467                                                         (unsigned long)priv;
2468
2469                         clear_extent_bits(extent_ins, start, end,
2470                                           EXTENT_LOCKED, GFP_NOFS);
2471
2472                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2473                                 BUG_ON(1);
2474
2475                         kfree(extent_op);
2476                 }
2477                 if (ret)
2478                         err = ret;
2479         }
2480         return err;
2481 }
2482
2483 /*
2484  * remove an extent from the root, returns 0 on success
2485  */
2486
2487 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2488                       struct btrfs_root *root,
2489                       u64 bytenr, u64 num_bytes, u64 parent,
2490                       u64 root_objectid, u64 owner, u64 offset)
2491 {
2492         struct btrfs_root *extent_root = root->fs_info->extent_root;
2493         int pending_ret;
2494         int ret;
2495
2496         WARN_ON(num_bytes < root->sectorsize);
2497         if (root == extent_root) {
2498                 struct pending_extent_op *extent_op;
2499
2500                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2501                 BUG_ON(!extent_op);
2502
2503                 extent_op->type = PENDING_EXTENT_DELETE;
2504                 extent_op->bytenr = bytenr;
2505                 extent_op->num_bytes = num_bytes;
2506                 extent_op->level = (int)owner;
2507
2508                 set_extent_bits(&root->fs_info->pending_del,
2509                                 bytenr, bytenr + num_bytes - 1,
2510                                 EXTENT_LOCKED, GFP_NOFS);
2511                 set_state_private(&root->fs_info->pending_del,
2512                                   bytenr, (unsigned long)extent_op);
2513                 return 0;
2514         }
2515         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2516                             root_objectid, owner, offset, 1);
2517         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2518         return ret ? ret : pending_ret;
2519 }
2520
2521 static u64 stripe_align(struct btrfs_root *root, u64 val)
2522 {
2523         u64 mask = ((u64)root->stripesize - 1);
2524         u64 ret = (val + mask) & ~mask;
2525         return ret;
2526 }
2527
2528 /*
2529  * walks the btree of allocated extents and find a hole of a given size.
2530  * The key ins is changed to record the hole:
2531  * ins->objectid == block start
2532  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2533  * ins->offset == number of blocks
2534  * Any available blocks before search_start are skipped.
2535  */
2536 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2537                                      struct btrfs_root *orig_root,
2538                                      u64 num_bytes, u64 empty_size,
2539                                      u64 search_start, u64 search_end,
2540                                      u64 hint_byte, struct btrfs_key *ins,
2541                                      u64 exclude_start, u64 exclude_nr,
2542                                      int data)
2543 {
2544         int ret;
2545         u64 orig_search_start = search_start;
2546         struct btrfs_root * root = orig_root->fs_info->extent_root;
2547         struct btrfs_fs_info *info = root->fs_info;
2548         u64 total_needed = num_bytes;
2549         struct btrfs_block_group_cache *block_group;
2550         int full_scan = 0;
2551         int wrapped = 0;
2552
2553         WARN_ON(num_bytes < root->sectorsize);
2554         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2555
2556         search_start = stripe_align(root, search_start);
2557
2558         if (hint_byte) {
2559                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2560                 if (!block_group)
2561                         hint_byte = search_start;
2562                 block_group = btrfs_find_block_group(root, block_group,
2563                                                      hint_byte, data, 1);
2564         } else {
2565                 block_group = btrfs_find_block_group(root,
2566                                                      trans->block_group,
2567                                                      search_start, data, 1);
2568         }
2569
2570         total_needed += empty_size;
2571
2572 check_failed:
2573         search_start = stripe_align(root, search_start);
2574         if (!block_group) {
2575                 block_group = btrfs_lookup_first_block_group(info,
2576                                                              search_start);
2577                 if (!block_group)
2578                         block_group = btrfs_lookup_first_block_group(info,
2579                                                        orig_search_start);
2580         }
2581         ret = find_search_start(root, &block_group, &search_start,
2582                                 total_needed, data);
2583         if (ret)
2584                 goto new_group;
2585
2586         ins->objectid = search_start;
2587         ins->offset = num_bytes;
2588
2589         if (ins->objectid + num_bytes >
2590             block_group->key.objectid + block_group->key.offset) {
2591                 search_start = block_group->key.objectid +
2592                         block_group->key.offset;
2593                 goto new_group;
2594         }
2595
2596         if (test_range_bit(&info->extent_ins, ins->objectid,
2597                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2598                 search_start = ins->objectid + num_bytes;
2599                 goto new_group;
2600         }
2601
2602         if (test_range_bit(&info->pinned_extents, ins->objectid,
2603                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2604                 search_start = ins->objectid + num_bytes;
2605                 goto new_group;
2606         }
2607
2608         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2609             ins->objectid < exclude_start + exclude_nr)) {
2610                 search_start = exclude_start + exclude_nr;
2611                 goto new_group;
2612         }
2613
2614         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2615                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2616                 if (block_group)
2617                         trans->block_group = block_group;
2618         }
2619         ins->offset = num_bytes;
2620         return 0;
2621
2622 new_group:
2623         block_group = btrfs_lookup_first_block_group(info, search_start);
2624         if (!block_group) {
2625                 search_start = orig_search_start;
2626                 if (full_scan) {
2627                         ret = -ENOSPC;
2628                         goto error;
2629                 }
2630                 if (wrapped) {
2631                         if (!full_scan)
2632                                 total_needed -= empty_size;
2633                         full_scan = 1;
2634                 } else
2635                         wrapped = 1;
2636         }
2637         cond_resched();
2638         block_group = btrfs_find_block_group(root, block_group,
2639                                              search_start, data, 0);
2640         goto check_failed;
2641
2642 error:
2643         return ret;
2644 }
2645
2646 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2647                                 struct btrfs_root *root,
2648                                 u64 num_bytes, u64 empty_size,
2649                                 u64 hint_byte, u64 search_end,
2650                                 struct btrfs_key *ins, int data)
2651 {
2652         int ret;
2653         u64 search_start = 0;
2654         u64 alloc_profile;
2655         struct btrfs_fs_info *info = root->fs_info;
2656
2657         if (info->extent_ops) {
2658                 struct btrfs_extent_ops *ops = info->extent_ops;
2659                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2660                 BUG_ON(ret);
2661                 goto found;
2662         }
2663
2664         if (data) {
2665                 alloc_profile = info->avail_data_alloc_bits &
2666                                 info->data_alloc_profile;
2667                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2668         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2669                    info->system_allocs >= 0) {
2670                 alloc_profile = info->avail_system_alloc_bits &
2671                                 info->system_alloc_profile;
2672                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2673         } else {
2674                 alloc_profile = info->avail_metadata_alloc_bits &
2675                                 info->metadata_alloc_profile;
2676                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2677         }
2678
2679         if (root->ref_cows) {
2680                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2681                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2682                                              num_bytes,
2683                                              BTRFS_BLOCK_GROUP_METADATA);
2684                         BUG_ON(ret);
2685                 }
2686                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2687                                      num_bytes + 2 * 1024 * 1024, data);
2688                 BUG_ON(ret);
2689         }
2690
2691         WARN_ON(num_bytes < root->sectorsize);
2692         ret = find_free_extent(trans, root, num_bytes, empty_size,
2693                                search_start, search_end, hint_byte, ins,
2694                                trans->alloc_exclude_start,
2695                                trans->alloc_exclude_nr, data);
2696         BUG_ON(ret);
2697 found:
2698         clear_extent_dirty(&root->fs_info->free_space_cache,
2699                            ins->objectid, ins->objectid + ins->offset - 1,
2700                            GFP_NOFS);
2701         return ret;
2702 }
2703
2704 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2705                                      struct btrfs_root *root,
2706                                      u64 root_objectid, u64 generation,
2707                                      u64 flags, struct btrfs_disk_key *key,
2708                                      int level, struct btrfs_key *ins)
2709 {
2710         int ret;
2711         struct btrfs_fs_info *fs_info = root->fs_info;
2712         struct btrfs_extent_item *extent_item;
2713         struct btrfs_tree_block_info *block_info;
2714         struct btrfs_extent_inline_ref *iref;
2715         struct btrfs_path *path;
2716         struct extent_buffer *leaf;
2717         u32 size = sizeof(*extent_item) + sizeof(*iref);
2718         int skinny_metadata =
2719                 btrfs_fs_incompat(fs_info,
2720                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2721
2722         if (!skinny_metadata)
2723                 size += sizeof(*block_info);
2724
2725         path = btrfs_alloc_path();
2726         BUG_ON(!path);
2727
2728         path->leave_spinning = 1;
2729         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2730                                       ins, size);
2731         BUG_ON(ret);
2732
2733         leaf = path->nodes[0];
2734         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2735                                      struct btrfs_extent_item);
2736         btrfs_set_extent_refs(leaf, extent_item, 1);
2737         btrfs_set_extent_generation(leaf, extent_item, generation);
2738         btrfs_set_extent_flags(leaf, extent_item,
2739                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2740
2741         if (skinny_metadata) {
2742                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2743         } else {
2744                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2745                 btrfs_set_tree_block_key(leaf, block_info, key);
2746                 btrfs_set_tree_block_level(leaf, block_info, level);
2747                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2748         }
2749
2750         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2751         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2752
2753         btrfs_mark_buffer_dirty(leaf);
2754         btrfs_free_path(path);
2755
2756         ret = update_block_group(trans, root, ins->objectid, root->leafsize,
2757                                  1, 0);
2758         return 0;
2759 }
2760
2761 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2762                             struct btrfs_root *root, u64 num_bytes,
2763                             u64 root_objectid, u64 generation,
2764                             u64 flags, struct btrfs_disk_key *key,
2765                             int level, u64 empty_size, u64 hint_byte,
2766                             u64 search_end, struct btrfs_key *ins)
2767 {
2768         int ret;
2769         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2770                                    hint_byte, search_end, ins, 0);
2771         BUG_ON(ret);
2772
2773         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2774                 struct pending_extent_op *extent_op;
2775
2776                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2777                 BUG_ON(!extent_op);
2778
2779                 extent_op->type = PENDING_EXTENT_INSERT;
2780                 extent_op->bytenr = ins->objectid;
2781                 extent_op->num_bytes = ins->offset;
2782                 extent_op->level = level;
2783                 extent_op->flags = flags;
2784                 memcpy(&extent_op->key, key, sizeof(*key));
2785
2786                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2787                                 ins->objectid + ins->offset - 1,
2788                                 EXTENT_LOCKED, GFP_NOFS);
2789                 set_state_private(&root->fs_info->extent_ins,
2790                                   ins->objectid, (unsigned long)extent_op);
2791         } else {
2792                 if (btrfs_fs_incompat(root->fs_info,
2793                                 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
2794                         ins->offset = level;
2795                         ins->type = BTRFS_METADATA_ITEM_KEY;
2796                 }
2797                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2798                                                 generation, flags,
2799                                                 key, level, ins);
2800                 finish_current_insert(trans, root->fs_info->extent_root);
2801                 del_pending_extents(trans, root->fs_info->extent_root);
2802         }
2803         return ret;
2804 }
2805
2806 /*
2807  * helper function to allocate a block for a given tree
2808  * returns the tree buffer or NULL.
2809  */
2810 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2811                                         struct btrfs_root *root,
2812                                         u32 blocksize, u64 root_objectid,
2813                                         struct btrfs_disk_key *key, int level,
2814                                         u64 hint, u64 empty_size)
2815 {
2816         struct btrfs_key ins;
2817         int ret;
2818         struct extent_buffer *buf;
2819
2820         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2821                                trans->transid, 0, key, level,
2822                                empty_size, hint, (u64)-1, &ins);
2823         if (ret) {
2824                 BUG_ON(ret > 0);
2825                 return ERR_PTR(ret);
2826         }
2827
2828         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2829         if (!buf) {
2830                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2831                                   0, root->root_key.objectid, level, 0);
2832                 BUG_ON(1);
2833                 return ERR_PTR(-ENOMEM);
2834         }
2835         btrfs_set_buffer_uptodate(buf);
2836         trans->blocks_used++;
2837
2838         return buf;
2839 }
2840
2841 #if 0
2842
2843 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2844                                   struct btrfs_root *root,
2845                                   struct extent_buffer *leaf)
2846 {
2847         u64 leaf_owner;
2848         u64 leaf_generation;
2849         struct btrfs_key key;
2850         struct btrfs_file_extent_item *fi;
2851         int i;
2852         int nritems;
2853         int ret;
2854
2855         BUG_ON(!btrfs_is_leaf(leaf));
2856         nritems = btrfs_header_nritems(leaf);
2857         leaf_owner = btrfs_header_owner(leaf);
2858         leaf_generation = btrfs_header_generation(leaf);
2859
2860         for (i = 0; i < nritems; i++) {
2861                 u64 disk_bytenr;
2862
2863                 btrfs_item_key_to_cpu(leaf, &key, i);
2864                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2865                         continue;
2866                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2867                 if (btrfs_file_extent_type(leaf, fi) ==
2868                     BTRFS_FILE_EXTENT_INLINE)
2869                         continue;
2870                 /*
2871                  * FIXME make sure to insert a trans record that
2872                  * repeats the snapshot del on crash
2873                  */
2874                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2875                 if (disk_bytenr == 0)
2876                         continue;
2877                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2878                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2879                                 leaf->start, leaf_owner, leaf_generation,
2880                                 key.objectid, 0);
2881                 BUG_ON(ret);
2882         }
2883         return 0;
2884 }
2885
2886 static void noinline reada_walk_down(struct btrfs_root *root,
2887                                      struct extent_buffer *node,
2888                                      int slot)
2889 {
2890         u64 bytenr;
2891         u64 last = 0;
2892         u32 nritems;
2893         u32 refs;
2894         u32 blocksize;
2895         int ret;
2896         int i;
2897         int level;
2898         int skipped = 0;
2899
2900         nritems = btrfs_header_nritems(node);
2901         level = btrfs_header_level(node);
2902         if (level)
2903                 return;
2904
2905         for (i = slot; i < nritems && skipped < 32; i++) {
2906                 bytenr = btrfs_node_blockptr(node, i);
2907                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2908                              (last > bytenr && last - bytenr > 32 * 1024))) {
2909                         skipped++;
2910                         continue;
2911                 }
2912                 blocksize = btrfs_level_size(root, level - 1);
2913                 if (i != slot) {
2914                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2915                                                       blocksize, &refs);
2916                         BUG_ON(ret);
2917                         if (refs != 1) {
2918                                 skipped++;
2919                                 continue;
2920                         }
2921                 }
2922                 mutex_unlock(&root->fs_info->fs_mutex);
2923                 ret = readahead_tree_block(root, bytenr, blocksize,
2924                                            btrfs_node_ptr_generation(node, i));
2925                 last = bytenr + blocksize;
2926                 cond_resched();
2927                 mutex_lock(&root->fs_info->fs_mutex);
2928                 if (ret)
2929                         break;
2930         }
2931 }
2932
2933 /*
2934  * helper function for drop_snapshot, this walks down the tree dropping ref
2935  * counts as it goes.
2936  */
2937 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2938                                    struct btrfs_root *root,
2939                                    struct btrfs_path *path, int *level)
2940 {
2941         u64 root_owner;
2942         u64 root_gen;
2943         u64 bytenr;
2944         u64 ptr_gen;
2945         struct extent_buffer *next;
2946         struct extent_buffer *cur;
2947         struct extent_buffer *parent;
2948         u32 blocksize;
2949         int ret;
2950         u32 refs;
2951
2952         WARN_ON(*level < 0);
2953         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2954         ret = btrfs_lookup_extent_ref(trans, root,
2955                                       path->nodes[*level]->start,
2956                                       path->nodes[*level]->len, &refs);
2957         BUG_ON(ret);
2958         if (refs > 1)
2959                 goto out;
2960
2961         /*
2962          * walk down to the last node level and free all the leaves
2963          */
2964         while(*level >= 0) {
2965                 WARN_ON(*level < 0);
2966                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2967                 cur = path->nodes[*level];
2968
2969                 if (btrfs_header_level(cur) != *level)
2970                         WARN_ON(1);
2971
2972                 if (path->slots[*level] >=
2973                     btrfs_header_nritems(cur))
2974                         break;
2975                 if (*level == 0) {
2976                         ret = drop_leaf_ref(trans, root, cur);
2977                         BUG_ON(ret);
2978                         break;
2979                 }
2980                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2981                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2982                 blocksize = btrfs_level_size(root, *level - 1);
2983                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2984                                               &refs);
2985                 BUG_ON(ret);
2986                 if (refs != 1) {
2987                         parent = path->nodes[*level];
2988                         root_owner = btrfs_header_owner(parent);
2989                         root_gen = btrfs_header_generation(parent);
2990                         path->slots[*level]++;
2991                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2992                                                 parent->start, root_owner,
2993                                                 root_gen, *level - 1, 1);
2994                         BUG_ON(ret);
2995                         continue;
2996                 }
2997                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2998                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2999                         free_extent_buffer(next);
3000                         reada_walk_down(root, cur, path->slots[*level]);
3001                         mutex_unlock(&root->fs_info->fs_mutex);
3002                         next = read_tree_block(root, bytenr, blocksize,
3003                                                ptr_gen);
3004                         mutex_lock(&root->fs_info->fs_mutex);
3005                 }
3006                 WARN_ON(*level <= 0);
3007                 if (path->nodes[*level-1])
3008                         free_extent_buffer(path->nodes[*level-1]);
3009                 path->nodes[*level-1] = next;
3010                 *level = btrfs_header_level(next);
3011                 path->slots[*level] = 0;
3012         }
3013 out:
3014         WARN_ON(*level < 0);
3015         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3016
3017         if (path->nodes[*level] == root->node) {
3018                 root_owner = root->root_key.objectid;
3019                 parent = path->nodes[*level];
3020         } else {
3021                 parent = path->nodes[*level + 1];
3022                 root_owner = btrfs_header_owner(parent);
3023         }
3024
3025         root_gen = btrfs_header_generation(parent);
3026         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3027                                 path->nodes[*level]->len, parent->start,
3028                                 root_owner, root_gen, *level, 1);
3029         free_extent_buffer(path->nodes[*level]);
3030         path->nodes[*level] = NULL;
3031         *level += 1;
3032         BUG_ON(ret);
3033         return 0;
3034 }
3035
3036 /*
3037  * helper for dropping snapshots.  This walks back up the tree in the path
3038  * to find the first node higher up where we haven't yet gone through
3039  * all the slots
3040  */
3041 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3042                                  struct btrfs_root *root,
3043                                  struct btrfs_path *path, int *level)
3044 {
3045         u64 root_owner;
3046         u64 root_gen;
3047         struct btrfs_root_item *root_item = &root->root_item;
3048         int i;
3049         int slot;
3050         int ret;
3051
3052         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3053                 slot = path->slots[i];
3054                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3055                         struct extent_buffer *node;
3056                         struct btrfs_disk_key disk_key;
3057                         node = path->nodes[i];
3058                         path->slots[i]++;
3059                         *level = i;
3060                         WARN_ON(*level == 0);
3061                         btrfs_node_key(node, &disk_key, path->slots[i]);
3062                         memcpy(&root_item->drop_progress,
3063                                &disk_key, sizeof(disk_key));
3064                         root_item->drop_level = i;
3065                         return 0;
3066                 } else {
3067                         struct extent_buffer *parent;
3068                         if (path->nodes[*level] == root->node)
3069                                 parent = path->nodes[*level];
3070                         else
3071                                 parent = path->nodes[*level + 1];
3072
3073                         root_owner = btrfs_header_owner(parent);
3074                         root_gen = btrfs_header_generation(parent);
3075                         ret = btrfs_free_extent(trans, root,
3076                                                 path->nodes[*level]->start,
3077                                                 path->nodes[*level]->len,
3078                                                 parent->start, root_owner,
3079                                                 root_gen, *level, 1);
3080                         BUG_ON(ret);
3081                         free_extent_buffer(path->nodes[*level]);
3082                         path->nodes[*level] = NULL;
3083                         *level = i + 1;
3084                 }
3085         }
3086         return 1;
3087 }
3088
3089 #endif
3090
3091 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3092 {
3093         struct btrfs_space_info *sinfo;
3094         struct btrfs_block_group_cache *cache;
3095         u64 start;
3096         u64 end;
3097         u64 ptr;
3098         int ret;
3099
3100         while(1) {
3101                 ret = find_first_extent_bit(&info->block_group_cache, 0,
3102                                             &start, &end, (unsigned int)-1);
3103                 if (ret)
3104                         break;
3105                 ret = get_state_private(&info->block_group_cache, start, &ptr);
3106                 if (!ret) {
3107                         cache = (struct btrfs_block_group_cache *)
3108                                         (uintptr_t)ptr;
3109                         if (cache->free_space_ctl) {
3110                                 btrfs_remove_free_space_cache(cache);
3111                                 kfree(cache->free_space_ctl);
3112                         }
3113                         kfree(cache);
3114                 }
3115                 clear_extent_bits(&info->block_group_cache, start,
3116                                   end, (unsigned int)-1, GFP_NOFS);
3117         }
3118         while(1) {
3119                 ret = find_first_extent_bit(&info->free_space_cache, 0,
3120                                             &start, &end, EXTENT_DIRTY);
3121                 if (ret)
3122                         break;
3123                 clear_extent_dirty(&info->free_space_cache, start,
3124                                    end, GFP_NOFS);
3125         }
3126
3127         while (!list_empty(&info->space_info)) {
3128                 sinfo = list_entry(info->space_info.next,
3129                                    struct btrfs_space_info, list);
3130                 list_del_init(&sinfo->list);
3131                 kfree(sinfo);
3132         }
3133         return 0;
3134 }
3135
3136 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3137                            struct btrfs_key *key)
3138 {
3139         int ret;
3140         struct btrfs_key found_key;
3141         struct extent_buffer *leaf;
3142         int slot;
3143
3144         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3145         if (ret < 0)
3146                 return ret;
3147         while(1) {
3148                 slot = path->slots[0];
3149                 leaf = path->nodes[0];
3150                 if (slot >= btrfs_header_nritems(leaf)) {
3151                         ret = btrfs_next_leaf(root, path);
3152                         if (ret == 0)
3153                                 continue;
3154                         if (ret < 0)
3155                                 goto error;
3156                         break;
3157                 }
3158                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3159
3160                 if (found_key.objectid >= key->objectid &&
3161                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3162                         return 0;
3163                 path->slots[0]++;
3164         }
3165         ret = -ENOENT;
3166 error:
3167         return ret;
3168 }
3169
3170 int btrfs_read_block_groups(struct btrfs_root *root)
3171 {
3172         struct btrfs_path *path;
3173         int ret;
3174         int bit;
3175         struct btrfs_block_group_cache *cache;
3176         struct btrfs_fs_info *info = root->fs_info;
3177         struct btrfs_space_info *space_info;
3178         struct extent_io_tree *block_group_cache;
3179         struct btrfs_key key;
3180         struct btrfs_key found_key;
3181         struct extent_buffer *leaf;
3182
3183         block_group_cache = &info->block_group_cache;
3184
3185         root = info->extent_root;
3186         key.objectid = 0;
3187         key.offset = 0;
3188         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3189         path = btrfs_alloc_path();
3190         if (!path)
3191                 return -ENOMEM;
3192
3193         while(1) {
3194                 ret = find_first_block_group(root, path, &key);
3195                 if (ret > 0) {
3196                         ret = 0;
3197                         goto error;
3198                 }
3199                 if (ret != 0) {
3200                         goto error;
3201                 }
3202                 leaf = path->nodes[0];
3203                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3204                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3205                 if (!cache) {
3206                         ret = -ENOMEM;
3207                         break;
3208                 }
3209
3210                 read_extent_buffer(leaf, &cache->item,
3211                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3212                                    sizeof(cache->item));
3213                 memcpy(&cache->key, &found_key, sizeof(found_key));
3214                 cache->cached = 0;
3215                 cache->pinned = 0;
3216                 key.objectid = found_key.objectid + found_key.offset;
3217                 btrfs_release_path(root, path);
3218                 cache->flags = btrfs_block_group_flags(&cache->item);
3219                 bit = 0;
3220                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3221                         bit = BLOCK_GROUP_DATA;
3222                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3223                         bit = BLOCK_GROUP_SYSTEM;
3224                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3225                         bit = BLOCK_GROUP_METADATA;
3226                 }
3227                 set_avail_alloc_bits(info, cache->flags);
3228                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3229                         cache->ro = 1;
3230
3231                 ret = update_space_info(info, cache->flags, found_key.offset,
3232                                         btrfs_block_group_used(&cache->item),
3233                                         &space_info);
3234                 BUG_ON(ret);
3235                 cache->space_info = space_info;
3236
3237                 /* use EXTENT_LOCKED to prevent merging */
3238                 set_extent_bits(block_group_cache, found_key.objectid,
3239                                 found_key.objectid + found_key.offset - 1,
3240                                 bit | EXTENT_LOCKED, GFP_NOFS);
3241                 set_state_private(block_group_cache, found_key.objectid,
3242                                   (unsigned long)cache);
3243         }
3244         ret = 0;
3245 error:
3246         btrfs_free_path(path);
3247         return ret;
3248 }
3249
3250 struct btrfs_block_group_cache *
3251 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3252                       u64 chunk_objectid, u64 chunk_offset, u64 size)
3253 {
3254         int ret;
3255         int bit = 0;
3256         struct btrfs_block_group_cache *cache;
3257         struct extent_io_tree *block_group_cache;
3258
3259         block_group_cache = &fs_info->block_group_cache;
3260
3261         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3262         BUG_ON(!cache);
3263         cache->key.objectid = chunk_offset;
3264         cache->key.offset = size;
3265
3266         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3267         btrfs_set_block_group_used(&cache->item, bytes_used);
3268         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3269         cache->flags = type;
3270         btrfs_set_block_group_flags(&cache->item, type);
3271
3272         ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3273                                 &cache->space_info);
3274         BUG_ON(ret);
3275
3276         bit = block_group_state_bits(type);
3277         set_extent_bits(block_group_cache, chunk_offset,
3278                         chunk_offset + size - 1,
3279                         bit | EXTENT_LOCKED, GFP_NOFS);
3280
3281         set_state_private(block_group_cache, chunk_offset,
3282                           (unsigned long)cache);
3283         set_avail_alloc_bits(fs_info, type);
3284
3285         return cache;
3286 }
3287
3288 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3289                            struct btrfs_root *root, u64 bytes_used,
3290                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3291                            u64 size)
3292 {
3293         int ret;
3294         struct btrfs_root *extent_root;
3295         struct btrfs_block_group_cache *cache;
3296
3297         cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3298                                       chunk_objectid, chunk_offset, size);
3299         extent_root = root->fs_info->extent_root;
3300         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3301                                 sizeof(cache->item));
3302         BUG_ON(ret);
3303
3304         finish_current_insert(trans, extent_root);
3305         ret = del_pending_extents(trans, extent_root);
3306         return 0;
3307 }
3308
3309 /*
3310  * This is for converter use only.
3311  *
3312  * In that case, we don't know where are free blocks located.
3313  * Therefore all block group cache entries must be setup properly
3314  * before doing any block allocation.
3315  */
3316 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3317                             struct btrfs_root *root)
3318 {
3319         u64 total_bytes;
3320         u64 cur_start;
3321         u64 group_type;
3322         u64 group_size;
3323         u64 group_align;
3324         u64 total_data = 0;
3325         u64 total_metadata = 0;
3326         u64 chunk_objectid;
3327         int ret;
3328         int bit;
3329         struct btrfs_root *extent_root;
3330         struct btrfs_block_group_cache *cache;
3331         struct extent_io_tree *block_group_cache;
3332
3333         extent_root = root->fs_info->extent_root;
3334         block_group_cache = &root->fs_info->block_group_cache;
3335         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3336         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3337         group_align = 64 * root->sectorsize;
3338
3339         cur_start = 0;
3340         while (cur_start < total_bytes) {
3341                 group_size = total_bytes / 12;
3342                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3343                 if (cur_start == 0) {
3344                         bit = BLOCK_GROUP_SYSTEM;
3345                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3346                         group_size /= 4;
3347                         group_size &= ~(group_align - 1);
3348                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3349                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3350                 } else {
3351                         group_size &= ~(group_align - 1);
3352                         if (total_data >= total_metadata * 2) {
3353                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3354                                 group_size = min_t(u64, group_size,
3355                                                    1ULL * 1024 * 1024 * 1024);
3356                                 total_metadata += group_size;
3357                         } else {
3358                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3359                                 group_size = min_t(u64, group_size,
3360                                                    5ULL * 1024 * 1024 * 1024);
3361                                 total_data += group_size;
3362                         }
3363                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3364                                 group_size = total_bytes - cur_start;
3365                 }
3366
3367                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3368                 BUG_ON(!cache);
3369
3370                 cache->key.objectid = cur_start;
3371                 cache->key.offset = group_size;
3372                 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3373
3374                 btrfs_set_block_group_used(&cache->item, 0);
3375                 btrfs_set_block_group_chunk_objectid(&cache->item,
3376                                                      chunk_objectid);
3377                 btrfs_set_block_group_flags(&cache->item, group_type);
3378
3379                 cache->flags = group_type;
3380
3381                 ret = update_space_info(root->fs_info, group_type, group_size,
3382                                         0, &cache->space_info);
3383                 BUG_ON(ret);
3384                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3385
3386                 set_extent_bits(block_group_cache, cur_start,
3387                                 cur_start + group_size - 1,
3388                                 bit | EXTENT_LOCKED, GFP_NOFS);
3389                 set_state_private(block_group_cache, cur_start,
3390                                   (unsigned long)cache);
3391                 cur_start += group_size;
3392         }
3393         /* then insert all the items */
3394         cur_start = 0;
3395         while(cur_start < total_bytes) {
3396                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3397                 BUG_ON(!cache);
3398
3399                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3400                                         sizeof(cache->item));
3401                 BUG_ON(ret);
3402
3403                 finish_current_insert(trans, extent_root);
3404                 ret = del_pending_extents(trans, extent_root);
3405                 BUG_ON(ret);
3406
3407                 cur_start = cache->key.objectid + cache->key.offset;
3408         }
3409         return 0;
3410 }
3411
3412 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3413                              struct btrfs_root *root,
3414                              u64 bytenr, u64 num_bytes, int alloc,
3415                              int mark_free)
3416 {
3417         return update_block_group(trans, root, bytenr, num_bytes,
3418                                   alloc, mark_free);
3419 }
3420
3421 static int btrfs_count_extents_in_block_group(struct btrfs_root *root,
3422                                               struct btrfs_path *path, u64 start,
3423                                               u64 len,
3424                                               u64 *total)
3425 {
3426         struct btrfs_key key;
3427         struct extent_buffer *leaf;
3428         u64 bytes_used = 0;
3429         int ret;
3430         int slot;
3431
3432         key.offset = 0;
3433         key.objectid = start;
3434         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
3435         ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
3436                                 &key, path, 0, 0);
3437         if (ret < 0)
3438                 return ret;
3439         while(1) {
3440                 leaf = path->nodes[0];
3441                 slot = path->slots[0];
3442                 if (slot >= btrfs_header_nritems(leaf)) {
3443                         ret = btrfs_next_leaf(root, path);
3444                         if (ret < 0)
3445                                 return ret;
3446                         if (ret > 0)
3447                                 break;
3448                         leaf = path->nodes[0];
3449                         slot = path->slots[0];
3450                 }
3451                 btrfs_item_key_to_cpu(leaf, &key, slot);
3452                 if (key.objectid > start + len)
3453                         break;
3454                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3455                         bytes_used += key.offset;
3456                 if (key.type == BTRFS_METADATA_ITEM_KEY)
3457                         bytes_used += root->leafsize;
3458                 path->slots[0]++;
3459         }
3460         *total = bytes_used;
3461         btrfs_release_path(root, path);
3462         return 0;
3463 }
3464
3465 int btrfs_check_block_accounting(struct btrfs_root *root)
3466 {
3467         int ret;
3468         u64 start = 0;
3469         u64 bytes_used = 0;
3470         struct btrfs_path path;
3471         struct btrfs_block_group_cache *cache;
3472         struct btrfs_fs_info *fs_info = root->fs_info;
3473
3474         btrfs_init_path(&path);
3475
3476         while(1) {
3477                 cache = btrfs_lookup_block_group(fs_info, start);
3478                 if (!cache)
3479                         break;
3480
3481                 ret = btrfs_count_extents_in_block_group(root, &path,
3482                                                          cache->key.objectid,
3483                                                          cache->key.offset,
3484                                                          &bytes_used);
3485
3486                 if (ret == 0) {
3487                         u64 on_disk = btrfs_block_group_used(&cache->item);
3488                         if (on_disk != bytes_used) {
3489                                 fprintf(stderr, "bad block group accounting found %llu "
3490                                         "expected %llu block group %llu\n",
3491                                         (unsigned long long)bytes_used,
3492                                         (unsigned long long)on_disk,
3493                                         (unsigned long long)cache->key.objectid);
3494                         }
3495                 }
3496                 start = cache->key.objectid + cache->key.offset;
3497
3498                 cache->space_info->bytes_used = 0;
3499         }
3500         return 0;
3501 }
3502
3503 /*
3504  * Fixup block accounting. The initial block accounting created by
3505  * make_block_groups isn't accuracy in this case.
3506  */
3507 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3508                                struct btrfs_root *root)
3509 {
3510         int ret;
3511         int slot;
3512         u64 start = 0;
3513         u64 bytes_used = 0;
3514         struct btrfs_path path;
3515         struct btrfs_key key;
3516         struct extent_buffer *leaf;
3517         struct btrfs_block_group_cache *cache;
3518         struct btrfs_fs_info *fs_info = root->fs_info;
3519
3520         root = root->fs_info->extent_root;
3521
3522         while(extent_root_pending_ops(fs_info)) {
3523                 ret = finish_current_insert(trans, root);
3524                 if (ret)
3525                         return ret;
3526                 ret = del_pending_extents(trans, root);
3527                 if (ret)
3528                         return ret;
3529         }
3530
3531         while(1) {
3532                 cache = btrfs_lookup_first_block_group(fs_info, start);
3533                 if (!cache)
3534                         break;
3535                 start = cache->key.objectid + cache->key.offset;
3536                 btrfs_set_block_group_used(&cache->item, 0);
3537                 cache->space_info->bytes_used = 0;
3538                 set_extent_bits(&root->fs_info->block_group_cache,
3539                                 cache->key.objectid,
3540                                 cache->key.objectid + cache->key.offset -1,
3541                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
3542         }
3543
3544         btrfs_init_path(&path);
3545         key.offset = 0;
3546         key.objectid = 0;
3547         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
3548         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3549                                 &key, &path, 0, 0);
3550         if (ret < 0)
3551                 return ret;
3552         while(1) {
3553                 leaf = path.nodes[0];
3554                 slot = path.slots[0];
3555                 if (slot >= btrfs_header_nritems(leaf)) {
3556                         ret = btrfs_next_leaf(root, &path);
3557                         if (ret < 0)
3558                                 return ret;
3559                         if (ret > 0)
3560                                 break;
3561                         leaf = path.nodes[0];
3562                         slot = path.slots[0];
3563                 }
3564                 btrfs_item_key_to_cpu(leaf, &key, slot);
3565                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3566                         bytes_used += key.offset;
3567                         ret = btrfs_update_block_group(trans, root,
3568                                   key.objectid, key.offset, 1, 0);
3569                         BUG_ON(ret);
3570                 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3571                         bytes_used += root->leafsize;
3572                         ret = btrfs_update_block_group(trans, root,
3573                                   key.objectid, root->leafsize, 1, 0);
3574                         BUG_ON(ret);
3575                 }
3576                 path.slots[0]++;
3577         }
3578         btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3579         btrfs_release_path(root, &path);
3580         return 0;
3581 }