extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include "kerncompat.h"
  22 #include "radix-tree.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27 #include "crc32c.h"
  28 #include "volumes.h"
  29
  30 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  32 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  33
  34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  35
  36 #define PENDING_EXTENT_INSERT 0
  37 #define PENDING_EXTENT_DELETE 1
  38 #define PENDING_BACKREF_UPDATE 2
  39
  40 struct pending_extent_op {
  41         int type;
  42         u64 bytenr;
  43         u64 num_bytes;
  44         u64 flags;
  45         struct btrfs_disk_key key;
  46         int level;
  47 };
  48
  49 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  50                                      struct btrfs_root *root,
  51                                      u64 root_objectid, u64 generation,
  52                                      u64 flags, struct btrfs_disk_key *key,
  53                                      int level, struct btrfs_key *ins);
  54 static int __free_extent(struct btrfs_trans_handle *trans,
  55                          struct btrfs_root *root,
  56                          u64 bytenr, u64 num_bytes, u64 parent,
  57                          u64 root_objectid, u64 owner_objectid,
  58                          u64 owner_offset, int refs_to_drop);
  59 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  60                                  btrfs_root *extent_root);
  61 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  62                                btrfs_root *extent_root);
  63
  64 static int remove_sb_from_cache(struct btrfs_root *root,
  65                                 struct btrfs_block_group_cache *cache)
  66 {
  67         u64 bytenr;
  68         u64 *logical;
  69         int stripe_len;
  70         int i, nr, ret;
  71         struct extent_io_tree *free_space_cache;
  72
  73         free_space_cache = &root->fs_info->free_space_cache;
  74         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
  75                 bytenr = btrfs_sb_offset(i);
  76                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
  77                                        cache->key.objectid, bytenr, 0,
  78                                        &logical, &nr, &stripe_len);
  79                 BUG_ON(ret);
  80                 while (nr--) {
  81                         clear_extent_dirty(free_space_cache, logical[nr],
  82                                 logical[nr] + stripe_len - 1, GFP_NOFS);
  83                 }
  84                 kfree(logical);
  85         }
  86         return 0;
  87 }
  88
  89 static int cache_block_group(struct btrfs_root *root,
  90                              struct btrfs_block_group_cache *block_group)
  91 {
  92         struct btrfs_path *path;
  93         int ret;
  94         struct btrfs_key key;
  95         struct extent_buffer *leaf;
  96         struct extent_io_tree *free_space_cache;
  97         int slot;
  98         u64 last;
  99         u64 hole_size;
 100
 101         if (!block_group)
 102                 return 0;
 103
 104         root = root->fs_info->extent_root;
 105         free_space_cache = &root->fs_info->free_space_cache;
 106
 107         if (block_group->cached)
 108                 return 0;
 109
 110         path = btrfs_alloc_path();
 111         if (!path)
 112                 return -ENOMEM;
 113
 114         path->reada = 2;
 115         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 116         key.objectid = last;
 117         key.offset = 0;
 118         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 119         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 120         if (ret < 0)
 121                 goto err;
 122
 123         while(1) {
 124                 leaf = path->nodes[0];
 125                 slot = path->slots[0];
 126                 if (slot >= btrfs_header_nritems(leaf)) {
 127                         ret = btrfs_next_leaf(root, path);
 128                         if (ret < 0)
 129                                 goto err;
 130                         if (ret == 0) {
 131                                 continue;
 132                         } else {
 133                                 break;
 134                         }
 135                 }
 136                 btrfs_item_key_to_cpu(leaf, &key, slot);
 137                 if (key.objectid < block_group->key.objectid) {
 138                         goto next;
 139                 }
 140                 if (key.objectid >= block_group->key.objectid +
 141                     block_group->key.offset) {
 142                         break;
 143                 }
 144
 145                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 146                         if (key.objectid > last) {
 147                                 hole_size = key.objectid - last;
 148                                 set_extent_dirty(free_space_cache, last,
 149                                                  last + hole_size - 1,
 150                                                  GFP_NOFS);
 151                         }
 152                         last = key.objectid + key.offset;
 153                 }
 154 next:
 155                 path->slots[0]++;
 156         }
 157
 158         if (block_group->key.objectid +
 159             block_group->key.offset > last) {
 160                 hole_size = block_group->key.objectid +
 161                         block_group->key.offset - last;
 162                 set_extent_dirty(free_space_cache, last,
 163                                  last + hole_size - 1, GFP_NOFS);
 164         }
 165         remove_sb_from_cache(root, block_group);
 166         block_group->cached = 1;
 167 err:
 168         btrfs_free_path(path);
 169         return 0;
 170 }
 171
 172 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 173                                                        btrfs_fs_info *info,
 174                                                        u64 bytenr)
 175 {
 176         struct extent_io_tree *block_group_cache;
 177         struct btrfs_block_group_cache *block_group = NULL;
 178         u64 ptr;
 179         u64 start;
 180         u64 end;
 181         int ret;
 182
 183         bytenr = max_t(u64, bytenr,
 184                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 185         block_group_cache = &info->block_group_cache;
 186         ret = find_first_extent_bit(block_group_cache,
 187                                     bytenr, &start, &end,
 188                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 189                                     BLOCK_GROUP_SYSTEM);
 190         if (ret) {
 191                 return NULL;
 192         }
 193         ret = get_state_private(block_group_cache, start, &ptr);
 194         if (ret)
 195                 return NULL;
 196
 197         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 198         return block_group;
 199 }
 200
 201 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 202                                                          btrfs_fs_info *info,
 203                                                          u64 bytenr)
 204 {
 205         struct extent_io_tree *block_group_cache;
 206         struct btrfs_block_group_cache *block_group = NULL;
 207         u64 ptr;
 208         u64 start;
 209         u64 end;
 210         int ret;
 211
 212         block_group_cache = &info->block_group_cache;
 213         ret = find_first_extent_bit(block_group_cache,
 214                                     bytenr, &start, &end,
 215                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 216                                     BLOCK_GROUP_SYSTEM);
 217         if (ret) {
 218                 return NULL;
 219         }
 220         ret = get_state_private(block_group_cache, start, &ptr);
 221         if (ret)
 222                 return NULL;
 223
 224         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 225         if (block_group->key.objectid <= bytenr && bytenr <
 226             block_group->key.objectid + block_group->key.offset)
 227                 return block_group;
 228         return NULL;
 229 }
 230
 231 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 232 {
 233         return (cache->flags & bits) == bits;
 234 }
 235
 236 static int noinline find_search_start(struct btrfs_root *root,
 237                               struct btrfs_block_group_cache **cache_ret,
 238                               u64 *start_ret, int num, int data)
 239 {
 240         int ret;
 241         struct btrfs_block_group_cache *cache = *cache_ret;
 242         u64 last;
 243         u64 start = 0;
 244         u64 end = 0;
 245         u64 search_start = *start_ret;
 246         int wrapped = 0;
 247
 248         if (!cache) {
 249                 goto out;
 250         }
 251 again:
 252         ret = cache_block_group(root, cache);
 253         if (ret)
 254                 goto out;
 255
 256         last = max(search_start, cache->key.objectid);
 257         if (cache->ro || !block_group_bits(cache, data)) {
 258                 goto new_group;
 259         }
 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         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 281         if (!cache) {
 282                 printk("Unable to find block group for %llu\n",
 283                         (unsigned long long)search_start);
 284                 WARN_ON(1);
 285         }
 286         return -ENOSPC;
 287
 288 new_group:
 289         last = cache->key.objectid + cache->key.offset;
 290 wrapped:
 291         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 292         if (!cache) {
 293 no_cache:
 294                 if (!wrapped) {
 295                         wrapped = 1;
 296                         last = search_start;
 297                         goto wrapped;
 298                 }
 299                 goto out;
 300         }
 301         cache = btrfs_find_block_group(root, cache, last, data, 0);
 302         cache = btrfs_find_block_group(root, cache, last, data, 0);
 303         if (!cache)
 304                 goto no_cache;
 305
 306         *cache_ret = cache;
 307         goto again;
 308 }
 309
 310 static u64 div_factor(u64 num, int factor)
 311 {
 312         if (factor == 10)
 313                 return num;
 314         num *= factor;
 315         num /= 10;
 316         return num;
 317 }
 318
 319 static int block_group_state_bits(u64 flags)
 320 {
 321         int bits = 0;
 322         if (flags & BTRFS_BLOCK_GROUP_DATA)
 323                 bits |= BLOCK_GROUP_DATA;
 324         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 325                 bits |= BLOCK_GROUP_METADATA;
 326         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 327                 bits |= BLOCK_GROUP_SYSTEM;
 328         return bits;
 329 }
 330
 331 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 332                                                  struct btrfs_block_group_cache
 333                                                  *hint, u64 search_start,
 334                                                  int data, int owner)
 335 {
 336         struct btrfs_block_group_cache *cache;
 337         struct extent_io_tree *block_group_cache;
 338         struct btrfs_block_group_cache *found_group = NULL;
 339         struct btrfs_fs_info *info = root->fs_info;
 340         u64 used;
 341         u64 last = 0;
 342         u64 hint_last;
 343         u64 start;
 344         u64 end;
 345         u64 free_check;
 346         u64 ptr;
 347         int bit;
 348         int ret;
 349         int full_search = 0;
 350         int factor = 10;
 351
 352         block_group_cache = &info->block_group_cache;
 353
 354         if (!owner)
 355                 factor = 10;
 356
 357         bit = block_group_state_bits(data);
 358
 359         if (search_start) {
 360                 struct btrfs_block_group_cache *shint;
 361                 shint = btrfs_lookup_block_group(info, search_start);
 362                 if (shint && !shint->ro && block_group_bits(shint, data)) {
 363                         used = btrfs_block_group_used(&shint->item);
 364                         if (used + shint->pinned <
 365                             div_factor(shint->key.offset, factor)) {
 366                                 return shint;
 367                         }
 368                 }
 369         }
 370         if (hint && !hint->ro && block_group_bits(hint, data)) {
 371                 used = btrfs_block_group_used(&hint->item);
 372                 if (used + hint->pinned <
 373                     div_factor(hint->key.offset, factor)) {
 374                         return hint;
 375                 }
 376                 last = hint->key.objectid + hint->key.offset;
 377                 hint_last = last;
 378         } else {
 379                 if (hint)
 380                         hint_last = max(hint->key.objectid, search_start);
 381                 else
 382                         hint_last = search_start;
 383
 384                 last = hint_last;
 385         }
 386 again:
 387         while(1) {
 388                 ret = find_first_extent_bit(block_group_cache, last,
 389                                             &start, &end, bit);
 390                 if (ret)
 391                         break;
 392
 393                 ret = get_state_private(block_group_cache, start, &ptr);
 394                 if (ret)
 395                         break;
 396
 397                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 398                 last = cache->key.objectid + cache->key.offset;
 399                 used = btrfs_block_group_used(&cache->item);
 400
 401                 if (!cache->ro && block_group_bits(cache, data)) {
 402                         if (full_search)
 403                                 free_check = cache->key.offset;
 404                         else
 405                                 free_check = div_factor(cache->key.offset,
 406                                                         factor);
 407
 408                         if (used + cache->pinned < free_check) {
 409                                 found_group = cache;
 410                                 goto found;
 411                         }
 412                 }
 413                 cond_resched();
 414         }
 415         if (!full_search) {
 416                 last = search_start;
 417                 full_search = 1;
 418                 goto again;
 419         }
 420 found:
 421         return found_group;
 422 }
 423
 424 /*
 425  * Back reference rules.  Back refs have three main goals:
 426  *
 427  * 1) differentiate between all holders of references to an extent so that
 428  *    when a reference is dropped we can make sure it was a valid reference
 429  *    before freeing the extent.
 430  *
 431  * 2) Provide enough information to quickly find the holders of an extent
 432  *    if we notice a given block is corrupted or bad.
 433  *
 434  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 435  *    maintenance.  This is actually the same as #2, but with a slightly
 436  *    different use case.
 437  *
 438  * There are two kinds of back refs. The implicit back refs is optimized
 439  * for pointers in non-shared tree blocks. For a given pointer in a block,
 440  * back refs of this kind provide information about the block's owner tree
 441  * and the pointer's key. These information allow us to find the block by
 442  * b-tree searching. The full back refs is for pointers in tree blocks not
 443  * referenced by their owner trees. The location of tree block is recorded
 444  * in the back refs. Actually the full back refs is generic, and can be
 445  * used in all cases the implicit back refs is used. The major shortcoming
 446  * of the full back refs is its overhead. Every time a tree block gets
 447  * COWed, we have to update back refs entry for all pointers in it.
 448  *
 449  * For a newly allocated tree block, we use implicit back refs for
 450  * pointers in it. This means most tree related operations only involve
 451  * implicit back refs. For a tree block created in old transaction, the
 452  * only way to drop a reference to it is COW it. So we can detect the
 453  * event that tree block loses its owner tree's reference and do the
 454  * back refs conversion.
 455  *
 456  * When a tree block is COW'd through a tree, there are four cases:
 457  *
 458  * The reference count of the block is one and the tree is the block's
 459  * owner tree. Nothing to do in this case.
 460  *
 461  * The reference count of the block is one and the tree is not the
 462  * block's owner tree. In this case, full back refs is used for pointers
 463  * in the block. Remove these full back refs, add implicit back refs for
 464  * every pointers in the new block.
 465  *
 466  * The reference count of the block is greater than one and the tree is
 467  * the block's owner tree. In this case, implicit back refs is used for
 468  * pointers in the block. Add full back refs for every pointers in the
 469  * block, increase lower level extents' reference counts. The original
 470  * implicit back refs are entailed to the new block.
 471  *
 472  * The reference count of the block is greater than one and the tree is
 473  * not the block's owner tree. Add implicit back refs for every pointer in
 474  * the new block, increase lower level extents' reference count.
 475  *
 476  * Back Reference Key composing:
 477  *
 478  * The key objectid corresponds to the first byte in the extent,
 479  * The key type is used to differentiate between types of back refs.
 480  * There are different meanings of the key offset for different types
 481  * of back refs.
 482  *
 483  * File extents can be referenced by:
 484  *
 485  * - multiple snapshots, subvolumes, or different generations in one subvol
 486  * - different files inside a single subvolume
 487  * - different offsets inside a file (bookend extents in file.c)
 488  *
 489  * The extent ref structure for the implicit back refs has fields for:
 490  *
 491  * - Objectid of the subvolume root
 492  * - objectid of the file holding the reference
 493  * - original offset in the file
 494  * - how many bookend extents
 495  *
 496  * The key offset for the implicit back refs is hash of the first
 497  * three fields.
 498  *
 499  * The extent ref structure for the full back refs has field for:
 500  *
 501  * - number of pointers in the tree leaf
 502  *
 503  * The key offset for the implicit back refs is the first byte of
 504  * the tree leaf
 505  *
 506  * When a file extent is allocated, The implicit back refs is used.
 507  * the fields are filled in:
 508  *
 509  *     (root_key.objectid, inode objectid, offset in file, 1)
 510  *
 511  * When a file extent is removed file truncation, we find the
 512  * corresponding implicit back refs and check the following fields:
 513  *
 514  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 515  *
 516  * Btree extents can be referenced by:
 517  *
 518  * - Different subvolumes
 519  *
 520  * Both the implicit back refs and the full back refs for tree blocks
 521  * only consist of key. The key offset for the implicit back refs is
 522  * objectid of block's owner tree. The key offset for the full back refs
 523  * is the first byte of parent block.
 524  *
 525  * When implicit back refs is used, information about the lowest key and
 526  * level of the tree block are required. These information are stored in
 527  * tree block info structure.
 528  */
 529
 530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 531 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 532                                   struct btrfs_root *root,
 533                                   struct btrfs_path *path,
 534                                   u64 owner, u32 extra_size)
 535 {
 536         struct btrfs_extent_item *item;
 537         struct btrfs_extent_item_v0 *ei0;
 538         struct btrfs_extent_ref_v0 *ref0;
 539         struct btrfs_tree_block_info *bi;
 540         struct extent_buffer *leaf;
 541         struct btrfs_key key;
 542         struct btrfs_key found_key;
 543         u32 new_size = sizeof(*item);
 544         u64 refs;
 545         int ret;
 546
 547         leaf = path->nodes[0];
 548         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 549
 550         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 551         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 552                              struct btrfs_extent_item_v0);
 553         refs = btrfs_extent_refs_v0(leaf, ei0);
 554
 555         if (owner == (u64)-1) {
 556                 while (1) {
 557                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 558                                 ret = btrfs_next_leaf(root, path);
 559                                 if (ret < 0)
 560                                         return ret;
 561                                 BUG_ON(ret > 0);
 562                                 leaf = path->nodes[0];
 563                         }
 564                         btrfs_item_key_to_cpu(leaf, &found_key,
 565                                               path->slots[0]);
 566                         BUG_ON(key.objectid != found_key.objectid);
 567                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 568                                 path->slots[0]++;
 569                                 continue;
 570                         }
 571                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 572                                               struct btrfs_extent_ref_v0);
 573                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 574                         break;
 575                 }
 576         }
 577         btrfs_release_path(root, path);
 578
 579         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 580                 new_size += sizeof(*bi);
 581
 582         new_size -= sizeof(*ei0);
 583         ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
 584         if (ret < 0)
 585                 return ret;
 586         BUG_ON(ret);
 587
 588         ret = btrfs_extend_item(trans, root, path, new_size);
 589         BUG_ON(ret);
 590
 591         leaf = path->nodes[0];
 592         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 593         btrfs_set_extent_refs(leaf, item, refs);
 594         /* FIXME: get real generation */
 595         btrfs_set_extent_generation(leaf, item, 0);
 596         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 597                 btrfs_set_extent_flags(leaf, item,
 598                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 599                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 600                 bi = (struct btrfs_tree_block_info *)(item + 1);
 601                 /* FIXME: get first key of the block */
 602                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 603                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 604         } else {
 605                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 606         }
 607         btrfs_mark_buffer_dirty(leaf);
 608         return 0;
 609 }
 610 #endif
 611
 612 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 613 {
 614         u32 high_crc = ~(u32)0;
 615         u32 low_crc = ~(u32)0;
 616         __le64 lenum;
 617
 618         lenum = cpu_to_le64(root_objectid);
 619         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 620         lenum = cpu_to_le64(owner);
 621         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 622         lenum = cpu_to_le64(offset);
 623         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 624
 625         return ((u64)high_crc << 31) ^ (u64)low_crc;
 626 }
 627
 628 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 629                                      struct btrfs_extent_data_ref *ref)
 630 {
 631         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 632                                     btrfs_extent_data_ref_objectid(leaf, ref),
 633                                     btrfs_extent_data_ref_offset(leaf, ref));
 634 }
 635
 636 static int match_extent_data_ref(struct extent_buffer *leaf,
 637                                  struct btrfs_extent_data_ref *ref,
 638                                  u64 root_objectid, u64 owner, u64 offset)
 639 {
 640         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 641             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 642             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 643                 return 0;
 644         return 1;
 645 }
 646
 647 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 648                                            struct btrfs_root *root,
 649                                            struct btrfs_path *path,
 650                                            u64 bytenr, u64 parent,
 651                                            u64 root_objectid,
 652                                            u64 owner, u64 offset)
 653 {
 654         struct btrfs_key key;
 655         struct btrfs_extent_data_ref *ref;
 656         struct extent_buffer *leaf;
 657         u32 nritems;
 658         int ret;
 659         int recow;
 660         int err = -ENOENT;
 661
 662         key.objectid = bytenr;
 663         if (parent) {
 664                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 665                 key.offset = parent;
 666         } else {
 667                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 668                 key.offset = hash_extent_data_ref(root_objectid,
 669                                                   owner, offset);
 670         }
 671 again:
 672         recow = 0;
 673         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 674         if (ret < 0) {
 675                 err = ret;
 676                 goto fail;
 677         }
 678
 679         if (parent) {
 680                 if (!ret)
 681                         return 0;
 682 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 683                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 684                 btrfs_release_path(root, path);
 685                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 686                 if (ret < 0) {
 687                         err = ret;
 688                         goto fail;
 689                 }
 690                 if (!ret)
 691                         return 0;
 692 #endif
 693                 goto fail;
 694         }
 695
 696         leaf = path->nodes[0];
 697         nritems = btrfs_header_nritems(leaf);
 698         while (1) {
 699                 if (path->slots[0] >= nritems) {
 700                         ret = btrfs_next_leaf(root, path);
 701                         if (ret < 0)
 702                                 err = ret;
 703                         if (ret)
 704                                 goto fail;
 705
 706                         leaf = path->nodes[0];
 707                         nritems = btrfs_header_nritems(leaf);
 708                         recow = 1;
 709                 }
 710
 711                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 712                 if (key.objectid != bytenr ||
 713                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 714                         goto fail;
 715
 716                 ref = btrfs_item_ptr(leaf, path->slots[0],
 717                                      struct btrfs_extent_data_ref);
 718
 719                 if (match_extent_data_ref(leaf, ref, root_objectid,
 720                                           owner, offset)) {
 721                         if (recow) {
 722                                 btrfs_release_path(root, path);
 723                                 goto again;
 724                         }
 725                         err = 0;
 726                         break;
 727                 }
 728                 path->slots[0]++;
 729         }
 730 fail:
 731         return err;
 732 }
 733
 734 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 735                                            struct btrfs_root *root,
 736                                            struct btrfs_path *path,
 737                                            u64 bytenr, u64 parent,
 738                                            u64 root_objectid, u64 owner,
 739                                            u64 offset, int refs_to_add)
 740 {
 741         struct btrfs_key key;
 742         struct extent_buffer *leaf;
 743         u32 size;
 744         u32 num_refs;
 745         int ret;
 746
 747         key.objectid = bytenr;
 748         if (parent) {
 749                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 750                 key.offset = parent;
 751                 size = sizeof(struct btrfs_shared_data_ref);
 752         } else {
 753                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 754                 key.offset = hash_extent_data_ref(root_objectid,
 755                                                   owner, offset);
 756                 size = sizeof(struct btrfs_extent_data_ref);
 757         }
 758
 759         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 760         if (ret && ret != -EEXIST)
 761                 goto fail;
 762
 763         leaf = path->nodes[0];
 764         if (parent) {
 765                 struct btrfs_shared_data_ref *ref;
 766                 ref = btrfs_item_ptr(leaf, path->slots[0],
 767                                      struct btrfs_shared_data_ref);
 768                 if (ret == 0) {
 769                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 770                 } else {
 771                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 772                         num_refs += refs_to_add;
 773                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 774                 }
 775         } else {
 776                 struct btrfs_extent_data_ref *ref;
 777                 while (ret == -EEXIST) {
 778                         ref = btrfs_item_ptr(leaf, path->slots[0],
 779                                              struct btrfs_extent_data_ref);
 780                         if (match_extent_data_ref(leaf, ref, root_objectid,
 781                                                   owner, offset))
 782                                 break;
 783                         btrfs_release_path(root, path);
 784
 785                         key.offset++;
 786                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 787                                                       size);
 788                         if (ret && ret != -EEXIST)
 789                                 goto fail;
 790
 791                         leaf = path->nodes[0];
 792                 }
 793                 ref = btrfs_item_ptr(leaf, path->slots[0],
 794                                      struct btrfs_extent_data_ref);
 795                 if (ret == 0) {
 796                         btrfs_set_extent_data_ref_root(leaf, ref,
 797                                                        root_objectid);
 798                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 799                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 800                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 801                 } else {
 802                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 803                         num_refs += refs_to_add;
 804                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 805                 }
 806         }
 807         btrfs_mark_buffer_dirty(leaf);
 808         ret = 0;
 809 fail:
 810         btrfs_release_path(root, path);
 811         return ret;
 812 }
 813
 814 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 815                                            struct btrfs_root *root,
 816                                            struct btrfs_path *path,
 817                                            int refs_to_drop)
 818 {
 819         struct btrfs_key key;
 820         struct btrfs_extent_data_ref *ref1 = NULL;
 821         struct btrfs_shared_data_ref *ref2 = NULL;
 822         struct extent_buffer *leaf;
 823         u32 num_refs = 0;
 824         int ret = 0;
 825
 826         leaf = path->nodes[0];
 827         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 828
 829         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 830                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 831                                       struct btrfs_extent_data_ref);
 832                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 833         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 834                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 835                                       struct btrfs_shared_data_ref);
 836                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 837 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 838         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 839                 struct btrfs_extent_ref_v0 *ref0;
 840                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 841                                       struct btrfs_extent_ref_v0);
 842                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 843 #endif
 844         } else {
 845                 BUG();
 846         }
 847
 848         BUG_ON(num_refs < refs_to_drop);
 849         num_refs -= refs_to_drop;
 850
 851         if (num_refs == 0) {
 852                 ret = btrfs_del_item(trans, root, path);
 853         } else {
 854                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 855                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 856                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 857                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 858 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 859                 else {
 860                         struct btrfs_extent_ref_v0 *ref0;
 861                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 862                                         struct btrfs_extent_ref_v0);
 863                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
 864                 }
 865 #endif
 866                 btrfs_mark_buffer_dirty(leaf);
 867         }
 868         return ret;
 869 }
 870
 871 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
 872                                           struct btrfs_path *path,
 873                                           struct btrfs_extent_inline_ref *iref)
 874 {
 875         struct btrfs_key key;
 876         struct extent_buffer *leaf;
 877         struct btrfs_extent_data_ref *ref1;
 878         struct btrfs_shared_data_ref *ref2;
 879         u32 num_refs = 0;
 880
 881         leaf = path->nodes[0];
 882         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 883         if (iref) {
 884                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
 885                     BTRFS_EXTENT_DATA_REF_KEY) {
 886                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 887                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 888                 } else {
 889                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 890                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 891                 }
 892         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 893                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 894                                       struct btrfs_extent_data_ref);
 895                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 896         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 897                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 898                                       struct btrfs_shared_data_ref);
 899                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 900 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 901         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 902                 struct btrfs_extent_ref_v0 *ref0;
 903                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 904                                       struct btrfs_extent_ref_v0);
 905                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 906 #endif
 907         } else {
 908                 BUG();
 909         }
 910         return num_refs;
 911 }
 912
 913 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 914                                           struct btrfs_root *root,
 915                                           struct btrfs_path *path,
 916                                           u64 bytenr, u64 parent,
 917                                           u64 root_objectid)
 918 {
 919         struct btrfs_key key;
 920         int ret;
 921
 922         key.objectid = bytenr;
 923         if (parent) {
 924                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 925                 key.offset = parent;
 926         } else {
 927                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 928                 key.offset = root_objectid;
 929         }
 930
 931         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 932         if (ret > 0)
 933                 ret = -ENOENT;
 934 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 935         if (ret == -ENOENT && parent) {
 936                 btrfs_release_path(root, path);
 937                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 938                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 939                 if (ret > 0)
 940                         ret = -ENOENT;
 941         }
 942 #endif
 943         return ret;
 944 }
 945
 946 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 947                                           struct btrfs_root *root,
 948                                           struct btrfs_path *path,
 949                                           u64 bytenr, u64 parent,
 950                                           u64 root_objectid)
 951 {
 952         struct btrfs_key key;
 953         int ret;
 954
 955         key.objectid = bytenr;
 956         if (parent) {
 957                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 958                 key.offset = parent;
 959         } else {
 960                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 961                 key.offset = root_objectid;
 962         }
 963
 964         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 965
 966         btrfs_release_path(root, path);
 967         return ret;
 968 }
 969
 970 static inline int extent_ref_type(u64 parent, u64 owner)
 971 {
 972         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 973                 if (parent > 0)
 974                         return BTRFS_SHARED_BLOCK_REF_KEY;
 975                 else
 976                         return BTRFS_TREE_BLOCK_REF_KEY;
 977         } else {
 978                 if (parent > 0)
 979                         return BTRFS_SHARED_DATA_REF_KEY;
 980                 else
 981                         return BTRFS_EXTENT_DATA_REF_KEY;
 982         }
 983 }
 984
 985 static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
 986
 987 {
 988         int level;
 989         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 990                 if (!path->nodes[level])
 991                         break;
 992                 if (path->slots[level] + 1 >=
 993                     btrfs_header_nritems(path->nodes[level]))
 994                         continue;
 995                 if (level == 0)
 996                         btrfs_item_key_to_cpu(path->nodes[level], key,
 997                                               path->slots[level] + 1);
 998                 else
 999                         btrfs_node_key_to_cpu(path->nodes[level], key,
1000                                               path->slots[level] + 1);
1001                 return 0;
1002         }
1003         return 1;
1004 }
1005
1006 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1007                                  struct btrfs_root *root,
1008                                  struct btrfs_path *path,
1009                                  struct btrfs_extent_inline_ref **ref_ret,
1010                                  u64 bytenr, u64 num_bytes,
1011                                  u64 parent, u64 root_objectid,
1012                                  u64 owner, u64 offset, int insert)
1013 {
1014         struct btrfs_key key;
1015         struct extent_buffer *leaf;
1016         struct btrfs_extent_item *ei;
1017         struct btrfs_extent_inline_ref *iref;
1018         u64 flags;
1019         u32 item_size;
1020         unsigned long ptr;
1021         unsigned long end;
1022         int extra_size;
1023         int type;
1024         int want;
1025         int ret;
1026         int err = 0;
1027
1028         key.objectid = bytenr;
1029         key.type = BTRFS_EXTENT_ITEM_KEY;
1030         key.offset = num_bytes;
1031
1032         want = extent_ref_type(parent, owner);
1033         if (insert)
1034                 extra_size = btrfs_extent_inline_ref_size(want);
1035         else
1036                 extra_size = -1;
1037         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1038         if (ret < 0) {
1039                 err = ret;
1040                 goto out;
1041         }
1042         BUG_ON(ret);
1043
1044         leaf = path->nodes[0];
1045         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047         if (item_size < sizeof(*ei)) {
1048                 if (!insert) {
1049                         err = -ENOENT;
1050                         goto out;
1051                 }
1052                 ret = convert_extent_item_v0(trans, root, path, owner,
1053                                              extra_size);
1054                 if (ret < 0) {
1055                         err = ret;
1056                         goto out;
1057                 }
1058                 leaf = path->nodes[0];
1059                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1060         }
1061 #endif
1062         BUG_ON(item_size < sizeof(*ei));
1063
1064         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1065         flags = btrfs_extent_flags(leaf, ei);
1066
1067         ptr = (unsigned long)(ei + 1);
1068         end = (unsigned long)ei + item_size;
1069
1070         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1071                 ptr += sizeof(struct btrfs_tree_block_info);
1072                 BUG_ON(ptr > end);
1073         } else {
1074                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1075         }
1076
1077         err = -ENOENT;
1078         while (1) {
1079                 if (ptr >= end) {
1080                         WARN_ON(ptr > end);
1081                         break;
1082                 }
1083                 iref = (struct btrfs_extent_inline_ref *)ptr;
1084                 type = btrfs_extent_inline_ref_type(leaf, iref);
1085                 if (want < type)
1086                         break;
1087                 if (want > type) {
1088                         ptr += btrfs_extent_inline_ref_size(type);
1089                         continue;
1090                 }
1091
1092                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1093                         struct btrfs_extent_data_ref *dref;
1094                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1095                         if (match_extent_data_ref(leaf, dref, root_objectid,
1096                                                   owner, offset)) {
1097                                 err = 0;
1098                                 break;
1099                         }
1100                         if (hash_extent_data_ref_item(leaf, dref) <
1101                             hash_extent_data_ref(root_objectid, owner, offset))
1102                                 break;
1103                 } else {
1104                         u64 ref_offset;
1105                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1106                         if (parent > 0) {
1107                                 if (parent == ref_offset) {
1108                                         err = 0;
1109                                         break;
1110                                 }
1111                                 if (ref_offset < parent)
1112                                         break;
1113                         } else {
1114                                 if (root_objectid == ref_offset) {
1115                                         err = 0;
1116                                         break;
1117                                 }
1118                                 if (ref_offset < root_objectid)
1119                                         break;
1120                         }
1121                 }
1122                 ptr += btrfs_extent_inline_ref_size(type);
1123         }
1124         if (err == -ENOENT && insert) {
1125                 if (item_size + extra_size >=
1126                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1127                         err = -EAGAIN;
1128                         goto out;
1129                 }
1130                 /*
1131                  * To add new inline back ref, we have to make sure
1132                  * there is no corresponding back ref item.
1133                  * For simplicity, we just do not add new inline back
1134                  * ref if there is any back ref item.
1135                  */
1136                 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1137                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1138                         err = -EAGAIN;
1139                         goto out;
1140                 }
1141         }
1142         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1143 out:
1144         return err;
1145 }
1146
1147 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1148                                 struct btrfs_root *root,
1149                                 struct btrfs_path *path,
1150                                 struct btrfs_extent_inline_ref *iref,
1151                                 u64 parent, u64 root_objectid,
1152                                 u64 owner, u64 offset, int refs_to_add)
1153 {
1154         struct extent_buffer *leaf;
1155         struct btrfs_extent_item *ei;
1156         unsigned long ptr;
1157         unsigned long end;
1158         unsigned long item_offset;
1159         u64 refs;
1160         int size;
1161         int type;
1162         int ret;
1163
1164         leaf = path->nodes[0];
1165         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1166         item_offset = (unsigned long)iref - (unsigned long)ei;
1167
1168         type = extent_ref_type(parent, owner);
1169         size = btrfs_extent_inline_ref_size(type);
1170
1171         ret = btrfs_extend_item(trans, root, path, size);
1172         BUG_ON(ret);
1173
1174         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1175         refs = btrfs_extent_refs(leaf, ei);
1176         refs += refs_to_add;
1177         btrfs_set_extent_refs(leaf, ei, refs);
1178
1179         ptr = (unsigned long)ei + item_offset;
1180         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1181         if (ptr < end - size)
1182                 memmove_extent_buffer(leaf, ptr + size, ptr,
1183                                       end - size - ptr);
1184
1185         iref = (struct btrfs_extent_inline_ref *)ptr;
1186         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1187         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1188                 struct btrfs_extent_data_ref *dref;
1189                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1190                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1191                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1192                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1193                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1194         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1195                 struct btrfs_shared_data_ref *sref;
1196                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1197                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1198                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1199         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1200                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1201         } else {
1202                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1203         }
1204         btrfs_mark_buffer_dirty(leaf);
1205         return 0;
1206 }
1207
1208 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1209                                  struct btrfs_root *root,
1210                                  struct btrfs_path *path,
1211                                  struct btrfs_extent_inline_ref **ref_ret,
1212                                  u64 bytenr, u64 num_bytes, u64 parent,
1213                                  u64 root_objectid, u64 owner, u64 offset)
1214 {
1215         int ret;
1216
1217         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1218                                            bytenr, num_bytes, parent,
1219                                            root_objectid, owner, offset, 0);
1220         if (ret != -ENOENT)
1221                 return ret;
1222
1223         btrfs_release_path(root, path);
1224         *ref_ret = NULL;
1225
1226         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1227                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1228                                             root_objectid);
1229         } else {
1230                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1231                                              root_objectid, owner, offset);
1232         }
1233         return ret;
1234 }
1235
1236 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1237                                  struct btrfs_root *root,
1238                                  struct btrfs_path *path,
1239                                  struct btrfs_extent_inline_ref *iref,
1240                                  int refs_to_mod)
1241 {
1242         struct extent_buffer *leaf;
1243         struct btrfs_extent_item *ei;
1244         struct btrfs_extent_data_ref *dref = NULL;
1245         struct btrfs_shared_data_ref *sref = NULL;
1246         unsigned long ptr;
1247         unsigned long end;
1248         u32 item_size;
1249         int size;
1250         int type;
1251         int ret;
1252         u64 refs;
1253
1254         leaf = path->nodes[0];
1255         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1256         refs = btrfs_extent_refs(leaf, ei);
1257         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1258         refs += refs_to_mod;
1259         btrfs_set_extent_refs(leaf, ei, refs);
1260
1261         type = btrfs_extent_inline_ref_type(leaf, iref);
1262
1263         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1264                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1265                 refs = btrfs_extent_data_ref_count(leaf, dref);
1266         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1267                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1268                 refs = btrfs_shared_data_ref_count(leaf, sref);
1269         } else {
1270                 refs = 1;
1271                 BUG_ON(refs_to_mod != -1);
1272         }
1273
1274         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1275         refs += refs_to_mod;
1276
1277         if (refs > 0) {
1278                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1279                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1280                 else
1281                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1282         } else {
1283                 size =  btrfs_extent_inline_ref_size(type);
1284                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1285                 ptr = (unsigned long)iref;
1286                 end = (unsigned long)ei + item_size;
1287                 if (ptr + size < end)
1288                         memmove_extent_buffer(leaf, ptr, ptr + size,
1289                                               end - ptr - size);
1290                 item_size -= size;
1291                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1292                 BUG_ON(ret);
1293         }
1294         btrfs_mark_buffer_dirty(leaf);
1295         return 0;
1296 }
1297
1298 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1299                                  struct btrfs_root *root,
1300                                  struct btrfs_path *path,
1301                                  u64 bytenr, u64 num_bytes, u64 parent,
1302                                  u64 root_objectid, u64 owner,
1303                                  u64 offset, int refs_to_add)
1304 {
1305         struct btrfs_extent_inline_ref *iref;
1306         int ret;
1307
1308         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1309                                            bytenr, num_bytes, parent,
1310                                            root_objectid, owner, offset, 1);
1311         if (ret == 0) {
1312                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1313                 ret = update_inline_extent_backref(trans, root, path, iref,
1314                                                    refs_to_add);
1315         } else if (ret == -ENOENT) {
1316                 ret = setup_inline_extent_backref(trans, root, path, iref,
1317                                                   parent, root_objectid,
1318                                                   owner, offset, refs_to_add);
1319         }
1320         return ret;
1321 }
1322
1323 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1324                                  struct btrfs_root *root,
1325                                  struct btrfs_path *path,
1326                                  u64 bytenr, u64 parent, u64 root_objectid,
1327                                  u64 owner, u64 offset, int refs_to_add)
1328 {
1329         int ret;
1330
1331         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1332                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1333                                              parent, root_objectid,
1334                                              owner, offset, refs_to_add);
1335         } else {
1336                 BUG_ON(refs_to_add != 1);
1337                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1338                                             parent, root_objectid);
1339         }
1340         return ret;
1341 }
1342
1343 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1344                                  struct btrfs_root *root,
1345                                  struct btrfs_path *path,
1346                                  struct btrfs_extent_inline_ref *iref,
1347                                  int refs_to_drop, int is_data)
1348 {
1349         int ret;
1350
1351         BUG_ON(!is_data && refs_to_drop != 1);
1352         if (iref) {
1353                 ret = update_inline_extent_backref(trans, root, path, iref,
1354                                                    -refs_to_drop);
1355         } else if (is_data) {
1356                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1357         } else {
1358                 ret = btrfs_del_item(trans, root, path);
1359         }
1360         return ret;
1361 }
1362
1363 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1364                          struct btrfs_root *root,
1365                          u64 bytenr, u64 num_bytes, u64 parent,
1366                          u64 root_objectid, u64 owner, u64 offset)
1367 {
1368         struct btrfs_path *path;
1369         struct extent_buffer *leaf;
1370         struct btrfs_extent_item *item;
1371         u64 refs;
1372         int ret;
1373         int err = 0;
1374
1375         path = btrfs_alloc_path();
1376         if (!path)
1377                 return -ENOMEM;
1378
1379         path->reada = 1;
1380         path->leave_spinning = 1;
1381
1382         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1383                                            path, bytenr, num_bytes, parent,
1384                                            root_objectid, owner, offset, 1);
1385         if (ret == 0)
1386                 goto out;
1387
1388         if (ret != -EAGAIN) {
1389                 err = ret;
1390                 goto out;
1391         }
1392
1393         leaf = path->nodes[0];
1394         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1395         refs = btrfs_extent_refs(leaf, item);
1396         btrfs_set_extent_refs(leaf, item, refs + 1);
1397
1398         btrfs_mark_buffer_dirty(leaf);
1399         btrfs_release_path(root->fs_info->extent_root, path);
1400
1401         path->reada = 1;
1402         path->leave_spinning = 1;
1403
1404         /* now insert the actual backref */
1405         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1406                                     path, bytenr, parent, root_objectid,
1407                                     owner, offset, 1);
1408         if (ret)
1409                 err = ret;
1410 out:
1411         btrfs_free_path(path);
1412         finish_current_insert(trans, root->fs_info->extent_root);
1413         del_pending_extents(trans, root->fs_info->extent_root);
1414         BUG_ON(err);
1415         return err;
1416 }
1417
1418 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1419                          struct btrfs_root *root)
1420 {
1421         finish_current_insert(trans, root->fs_info->extent_root);
1422         del_pending_extents(trans, root->fs_info->extent_root);
1423         return 0;
1424 }
1425
1426 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1427                              struct btrfs_root *root, u64 bytenr,
1428                              u64 num_bytes, u64 *refs, u64 *flags)
1429 {
1430         struct btrfs_path *path;
1431         int ret;
1432         struct btrfs_key key;
1433         struct extent_buffer *l;
1434         struct btrfs_extent_item *item;
1435         u32 item_size;
1436         u64 num_refs;
1437         u64 extent_flags;
1438
1439         WARN_ON(num_bytes < root->sectorsize);
1440         path = btrfs_alloc_path();
1441         path->reada = 1;
1442         key.objectid = bytenr;
1443         key.offset = num_bytes;
1444         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1445         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1446                                 0, 0);
1447         if (ret < 0)
1448                 goto out;
1449         if (ret != 0) {
1450                 btrfs_print_leaf(root, path->nodes[0]);
1451                 printk("failed to find block number %Lu\n",
1452                        (unsigned long long)bytenr);
1453                 BUG();
1454         }
1455
1456         l = path->nodes[0];
1457         item_size = btrfs_item_size_nr(l, path->slots[0]);
1458         if (item_size >= sizeof(*item)) {
1459                 item = btrfs_item_ptr(l, path->slots[0],
1460                                       struct btrfs_extent_item);
1461                 num_refs = btrfs_extent_refs(l, item);
1462                 extent_flags = btrfs_extent_flags(l, item);
1463         } else {
1464 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1465                         struct btrfs_extent_item_v0 *ei0;
1466                         BUG_ON(item_size != sizeof(*ei0));
1467                         ei0 = btrfs_item_ptr(l, path->slots[0],
1468                                              struct btrfs_extent_item_v0);
1469                         num_refs = btrfs_extent_refs_v0(l, ei0);
1470                         /* FIXME: this isn't correct for data */
1471                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1472 #else
1473                         BUG();
1474 #endif
1475                 }
1476                 BUG_ON(num_refs == 0);
1477         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1478         if (refs)
1479                 *refs = num_refs;
1480         if (flags)
1481                 *flags = extent_flags;
1482 out:
1483         btrfs_free_path(path);
1484         return 0;
1485 }
1486
1487 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1488                           struct btrfs_root *root,
1489                           u64 bytenr, u64 num_bytes, u64 flags)
1490 {
1491         struct btrfs_path *path;
1492         int ret;
1493         struct btrfs_key key;
1494         struct extent_buffer *l;
1495         struct btrfs_extent_item *item;
1496         u32 item_size;
1497
1498         WARN_ON(num_bytes < root->sectorsize);
1499         path = btrfs_alloc_path();
1500         path->reada = 1;
1501         key.objectid = bytenr;
1502         key.offset = num_bytes;
1503         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1504         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1505                                 0, 0);
1506         if (ret < 0)
1507                 goto out;
1508         if (ret != 0) {
1509                 btrfs_print_leaf(root, path->nodes[0]);
1510                 printk("failed to find block number %Lu\n",
1511                         (unsigned long long)bytenr);
1512                 BUG();
1513         }
1514         l = path->nodes[0];
1515         item_size = btrfs_item_size_nr(l, path->slots[0]);
1516 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1517         if (item_size < sizeof(*item)) {
1518                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1519                                              path, (u64)-1, 0);
1520                 if (ret < 0)
1521                         goto out;
1522
1523                 l = path->nodes[0];
1524                 item_size = btrfs_item_size_nr(l, path->slots[0]);
1525         }
1526 #endif
1527         BUG_ON(item_size < sizeof(*item));
1528         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1529         flags |= btrfs_extent_flags(l, item);
1530         btrfs_set_extent_flags(l, item, flags);
1531 out:
1532         btrfs_free_path(path);
1533         finish_current_insert(trans, root->fs_info->extent_root);
1534         del_pending_extents(trans, root->fs_info->extent_root);
1535         return ret;
1536 }
1537
1538 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1539                            struct btrfs_root *root,
1540                            struct extent_buffer *buf,
1541                            int record_parent, int inc)
1542 {
1543         u64 bytenr;
1544         u64 num_bytes;
1545         u64 parent;
1546         u64 ref_root;
1547         u32 nritems;
1548         struct btrfs_key key;
1549         struct btrfs_file_extent_item *fi;
1550         int i;
1551         int level;
1552         int ret = 0;
1553         int (*process_func)(struct btrfs_trans_handle *trans,
1554                             struct btrfs_root *root,
1555                             u64, u64, u64, u64, u64, u64);
1556
1557         ref_root = btrfs_header_owner(buf);
1558         nritems = btrfs_header_nritems(buf);
1559         level = btrfs_header_level(buf);
1560
1561         if (!root->ref_cows && level == 0)
1562                 return 0;
1563
1564         if (inc)
1565                 process_func = btrfs_inc_extent_ref;
1566         else
1567                 process_func = btrfs_free_extent;
1568
1569         if (record_parent)
1570                 parent = buf->start;
1571         else
1572                 parent = 0;
1573
1574         for (i = 0; i < nritems; i++) {
1575                 cond_resched();
1576                 if (level == 0) {
1577                         btrfs_item_key_to_cpu(buf, &key, i);
1578                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1579                                 continue;
1580                         fi = btrfs_item_ptr(buf, i,
1581                                             struct btrfs_file_extent_item);
1582                         if (btrfs_file_extent_type(buf, fi) ==
1583                             BTRFS_FILE_EXTENT_INLINE)
1584                                 continue;
1585                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1586                         if (bytenr == 0)
1587                                 continue;
1588
1589                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1590                         key.offset -= btrfs_file_extent_offset(buf, fi);
1591                         ret = process_func(trans, root, bytenr, num_bytes,
1592                                            parent, ref_root, key.objectid,
1593                                            key.offset);
1594                         if (ret) {
1595                                 WARN_ON(1);
1596                                 goto fail;
1597                         }
1598                 } else {
1599                         bytenr = btrfs_node_blockptr(buf, i);
1600                         num_bytes = btrfs_level_size(root, level - 1);
1601                         ret = process_func(trans, root, bytenr, num_bytes,
1602                                            parent, ref_root, level - 1, 0);
1603                         if (ret) {
1604                                 WARN_ON(1);
1605                                 goto fail;
1606                         }
1607                 }
1608         }
1609         return 0;
1610 fail:
1611         WARN_ON(1);
1612         return ret;
1613 }
1614
1615 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1616                   struct extent_buffer *buf, int record_parent)
1617 {
1618         return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1619 }
1620
1621 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1622                   struct extent_buffer *buf, int record_parent)
1623 {
1624         return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1625 }
1626
1627 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1628                                  struct btrfs_root *root,
1629                                  struct btrfs_path *path,
1630                                  struct btrfs_block_group_cache *cache)
1631 {
1632         int ret;
1633         int pending_ret;
1634         struct btrfs_root *extent_root = root->fs_info->extent_root;
1635         unsigned long bi;
1636         struct extent_buffer *leaf;
1637
1638         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1639         if (ret < 0)
1640                 goto fail;
1641         BUG_ON(ret);
1642
1643         leaf = path->nodes[0];
1644         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1645         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1646         btrfs_mark_buffer_dirty(leaf);
1647         btrfs_release_path(extent_root, path);
1648 fail:
1649         finish_current_insert(trans, extent_root);
1650         pending_ret = del_pending_extents(trans, extent_root);
1651         if (ret)
1652                 return ret;
1653         if (pending_ret)
1654                 return pending_ret;
1655         return 0;
1656
1657 }
1658
1659 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1660                                    struct btrfs_root *root)
1661 {
1662         struct extent_io_tree *block_group_cache;
1663         struct btrfs_block_group_cache *cache;
1664         int ret;
1665         struct btrfs_path *path;
1666         u64 last = 0;
1667         u64 start;
1668         u64 end;
1669         u64 ptr;
1670
1671         block_group_cache = &root->fs_info->block_group_cache;
1672         path = btrfs_alloc_path();
1673         if (!path)
1674                 return -ENOMEM;
1675
1676         while(1) {
1677                 ret = find_first_extent_bit(block_group_cache, last,
1678                                             &start, &end, BLOCK_GROUP_DIRTY);
1679                 if (ret) {
1680                         if (last == 0)
1681                                 break;
1682                         last = 0;
1683                         continue;
1684                 }
1685
1686                 last = end + 1;
1687                 ret = get_state_private(block_group_cache, start, &ptr);
1688                 BUG_ON(ret);
1689
1690                 clear_extent_bits(block_group_cache, start, end,
1691                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1692
1693                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1694                 ret = write_one_cache_group(trans, root, path, cache);
1695                 BUG_ON(ret);
1696         }
1697         btrfs_free_path(path);
1698         return 0;
1699 }
1700
1701 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1702                                                   u64 flags)
1703 {
1704         struct list_head *head = &info->space_info;
1705         struct list_head *cur;
1706         struct btrfs_space_info *found;
1707         list_for_each(cur, head) {
1708                 found = list_entry(cur, struct btrfs_space_info, list);
1709                 if (found->flags & flags)
1710                         return found;
1711         }
1712         return NULL;
1713
1714 }
1715
1716 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1717                              u64 total_bytes, u64 bytes_used,
1718                              struct btrfs_space_info **space_info)
1719 {
1720         struct btrfs_space_info *found;
1721
1722         found = __find_space_info(info, flags);
1723         if (found) {
1724                 found->total_bytes += total_bytes;
1725                 found->bytes_used += bytes_used;
1726                 WARN_ON(found->total_bytes < found->bytes_used);
1727                 *space_info = found;
1728                 return 0;
1729         }
1730         found = kmalloc(sizeof(*found), GFP_NOFS);
1731         if (!found)
1732                 return -ENOMEM;
1733
1734         list_add(&found->list, &info->space_info);
1735         found->flags = flags;
1736         found->total_bytes = total_bytes;
1737         found->bytes_used = bytes_used;
1738         found->bytes_pinned = 0;
1739         found->full = 0;
1740         *space_info = found;
1741         return 0;
1742 }
1743
1744
1745 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1746 {
1747         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1748                                    BTRFS_BLOCK_GROUP_RAID1 |
1749                                    BTRFS_BLOCK_GROUP_RAID10 |
1750                                    BTRFS_BLOCK_GROUP_DUP);
1751         if (extra_flags) {
1752                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1753                         fs_info->avail_data_alloc_bits |= extra_flags;
1754                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1755                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1756                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1757                         fs_info->avail_system_alloc_bits |= extra_flags;
1758         }
1759 }
1760
1761 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1762                           struct btrfs_root *extent_root, u64 alloc_bytes,
1763                           u64 flags)
1764 {
1765         struct btrfs_space_info *space_info;
1766         u64 thresh;
1767         u64 start;
1768         u64 num_bytes;
1769         int ret;
1770
1771         space_info = __find_space_info(extent_root->fs_info, flags);
1772         if (!space_info) {
1773                 ret = update_space_info(extent_root->fs_info, flags,
1774                                         0, 0, &space_info);
1775                 BUG_ON(ret);
1776         }
1777         BUG_ON(!space_info);
1778
1779         if (space_info->full)
1780                 return 0;
1781
1782         thresh = div_factor(space_info->total_bytes, 7);
1783         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1784             thresh)
1785                 return 0;
1786
1787         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1788                                 space_info->flags);
1789         if (ret == -ENOSPC) {
1790                 space_info->full = 1;
1791                 return 0;
1792         }
1793
1794         BUG_ON(ret);
1795
1796         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1797                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1798         BUG_ON(ret);
1799         return 0;
1800 }
1801
1802 static int update_block_group(struct btrfs_trans_handle *trans,
1803                               struct btrfs_root *root,
1804                               u64 bytenr, u64 num_bytes, int alloc,
1805                               int mark_free)
1806 {
1807         struct btrfs_block_group_cache *cache;
1808         struct btrfs_fs_info *info = root->fs_info;
1809         u64 total = num_bytes;
1810         u64 old_val;
1811         u64 byte_in_group;
1812         u64 start;
1813         u64 end;
1814
1815         /* block accounting for super block */
1816         old_val = btrfs_super_bytes_used(&info->super_copy);
1817         if (alloc)
1818                 old_val += num_bytes;
1819         else
1820                 old_val -= num_bytes;
1821         btrfs_set_super_bytes_used(&info->super_copy, old_val);
1822
1823         /* block accounting for root item */
1824         old_val = btrfs_root_used(&root->root_item);
1825         if (alloc)
1826                 old_val += num_bytes;
1827         else
1828                 old_val -= num_bytes;
1829         btrfs_set_root_used(&root->root_item, old_val);
1830
1831         while(total) {
1832                 cache = btrfs_lookup_block_group(info, bytenr);
1833                 if (!cache) {
1834                         return -1;
1835                 }
1836                 byte_in_group = bytenr - cache->key.objectid;
1837                 WARN_ON(byte_in_group > cache->key.offset);
1838                 start = cache->key.objectid;
1839                 end = start + cache->key.offset - 1;
1840                 set_extent_bits(&info->block_group_cache, start, end,
1841                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1842
1843                 old_val = btrfs_block_group_used(&cache->item);
1844                 num_bytes = min(total, cache->key.offset - byte_in_group);
1845                 if (alloc) {
1846                         old_val += num_bytes;
1847                         cache->space_info->bytes_used += num_bytes;
1848                 } else {
1849                         old_val -= num_bytes;
1850                         cache->space_info->bytes_used -= num_bytes;
1851                         if (mark_free) {
1852                                 set_extent_dirty(&info->free_space_cache,
1853                                                  bytenr, bytenr + num_bytes - 1,
1854                                                  GFP_NOFS);
1855                         }
1856                 }
1857                 btrfs_set_block_group_used(&cache->item, old_val);
1858                 total -= num_bytes;
1859                 bytenr += num_bytes;
1860         }
1861         return 0;
1862 }
1863
1864 static int update_pinned_extents(struct btrfs_root *root,
1865                                 u64 bytenr, u64 num, int pin)
1866 {
1867         u64 len;
1868         struct btrfs_block_group_cache *cache;
1869         struct btrfs_fs_info *fs_info = root->fs_info;
1870
1871         if (pin) {
1872                 set_extent_dirty(&fs_info->pinned_extents,
1873                                 bytenr, bytenr + num - 1, GFP_NOFS);
1874         } else {
1875                 clear_extent_dirty(&fs_info->pinned_extents,
1876                                 bytenr, bytenr + num - 1, GFP_NOFS);
1877         }
1878         while (num > 0) {
1879                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1880                 WARN_ON(!cache);
1881                 len = min(num, cache->key.offset -
1882                           (bytenr - cache->key.objectid));
1883                 if (pin) {
1884                         cache->pinned += len;
1885                         cache->space_info->bytes_pinned += len;
1886                         fs_info->total_pinned += len;
1887                 } else {
1888                         cache->pinned -= len;
1889                         cache->space_info->bytes_pinned -= len;
1890                         fs_info->total_pinned -= len;
1891                 }
1892                 bytenr += len;
1893                 num -= len;
1894         }
1895         return 0;
1896 }
1897
1898 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1899 {
1900         u64 last = 0;
1901         u64 start;
1902         u64 end;
1903         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1904         int ret;
1905
1906         while(1) {
1907                 ret = find_first_extent_bit(pinned_extents, last,
1908                                             &start, &end, EXTENT_DIRTY);
1909                 if (ret)
1910                         break;
1911                 set_extent_dirty(copy, start, end, GFP_NOFS);
1912                 last = end + 1;
1913         }
1914         return 0;
1915 }
1916
1917 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1918                                struct btrfs_root *root,
1919                                struct extent_io_tree *unpin)
1920 {
1921         u64 start;
1922         u64 end;
1923         int ret;
1924         struct extent_io_tree *free_space_cache;
1925         free_space_cache = &root->fs_info->free_space_cache;
1926
1927         while(1) {
1928                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1929                                             EXTENT_DIRTY);
1930                 if (ret)
1931                         break;
1932                 update_pinned_extents(root, start, end + 1 - start, 0);
1933                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1934                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1935         }
1936         return 0;
1937 }
1938
1939 static int finish_current_insert(struct btrfs_trans_handle *trans,
1940                                  struct btrfs_root *extent_root)
1941 {
1942         u64 start;
1943         u64 end;
1944         u64 priv;
1945         struct btrfs_fs_info *info = extent_root->fs_info;
1946         struct btrfs_path *path;
1947         struct pending_extent_op *extent_op;
1948         struct btrfs_key key;
1949         int ret;
1950
1951         path = btrfs_alloc_path();
1952
1953         while(1) {
1954                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1955                                             &end, EXTENT_LOCKED);
1956                 if (ret)
1957                         break;
1958
1959                 ret = get_state_private(&info->extent_ins, start, &priv);
1960                 BUG_ON(ret);
1961                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1962
1963                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1964                         key.objectid = start;
1965                         key.offset = end + 1 - start;
1966                         key.type = BTRFS_EXTENT_ITEM_KEY;
1967                         ret = alloc_reserved_tree_block(trans, extent_root,
1968                                                 extent_root->root_key.objectid,
1969                                                 trans->transid,
1970                                                 extent_op->flags,
1971                                                 &extent_op->key,
1972                                                 extent_op->level, &key);
1973                 } else {
1974                         BUG_ON(1);
1975                 }
1976
1977                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1978                                   GFP_NOFS);
1979                 kfree(extent_op);
1980         }
1981         btrfs_free_path(path);
1982         return 0;
1983 }
1984
1985 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1986                           struct btrfs_root *root,
1987                           u64 bytenr, u64 num_bytes, int is_data)
1988 {
1989         int err = 0;
1990         struct extent_buffer *buf;
1991
1992         if (is_data)
1993                 goto pinit;
1994
1995         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1996         if (!buf)
1997                 goto pinit;
1998
1999         /* we can reuse a block if it hasn't been written
2000          * and it is from this transaction.  We can't
2001          * reuse anything from the tree log root because
2002          * it has tiny sub-transactions.
2003          */
2004         if (btrfs_buffer_uptodate(buf, 0)) {
2005                 u64 header_owner = btrfs_header_owner(buf);
2006                 u64 header_transid = btrfs_header_generation(buf);
2007                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2008                     header_transid == trans->transid &&
2009                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2010                         clean_tree_block(NULL, root, buf);
2011                         free_extent_buffer(buf);
2012                         return 1;
2013                 }
2014         }
2015         free_extent_buffer(buf);
2016 pinit:
2017         update_pinned_extents(root, bytenr, num_bytes, 1);
2018
2019         BUG_ON(err < 0);
2020         return 0;
2021 }
2022
2023 /*
2024  * remove an extent from the root, returns 0 on success
2025  */
2026 static int __free_extent(struct btrfs_trans_handle *trans,
2027                          struct btrfs_root *root,
2028                          u64 bytenr, u64 num_bytes, u64 parent,
2029                          u64 root_objectid, u64 owner_objectid,
2030                          u64 owner_offset, int refs_to_drop)
2031 {
2032
2033         struct btrfs_key key;
2034         struct btrfs_path *path;
2035         struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2036         struct btrfs_root *extent_root = root->fs_info->extent_root;
2037         struct extent_buffer *leaf;
2038         struct btrfs_extent_item *ei;
2039         struct btrfs_extent_inline_ref *iref;
2040         int ret;
2041         int is_data;
2042         int extent_slot = 0;
2043         int found_extent = 0;
2044         int num_to_del = 1;
2045         u32 item_size;
2046         u64 refs;
2047
2048         path = btrfs_alloc_path();
2049         if (!path)
2050                 return -ENOMEM;
2051
2052         path->reada = 1;
2053         path->leave_spinning = 1;
2054
2055         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2056         BUG_ON(!is_data && refs_to_drop != 1);
2057
2058         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2059                                     bytenr, num_bytes, parent,
2060                                     root_objectid, owner_objectid,
2061                                     owner_offset);
2062         if (ret == 0) {
2063                 extent_slot = path->slots[0];
2064                 while (extent_slot >= 0) {
2065                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2066                                               extent_slot);
2067                         if (key.objectid != bytenr)
2068                                 break;
2069                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2070                             key.offset == num_bytes) {
2071                                 found_extent = 1;
2072                                 break;
2073                         }
2074                         if (path->slots[0] - extent_slot > 5)
2075                                 break;
2076                         extent_slot--;
2077                 }
2078 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2079                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2080                 if (found_extent && item_size < sizeof(*ei))
2081                         found_extent = 0;
2082 #endif
2083                 if (!found_extent) {
2084                         BUG_ON(iref);
2085                         ret = remove_extent_backref(trans, extent_root, path,
2086                                                     NULL, refs_to_drop,
2087                                                     is_data);
2088                         BUG_ON(ret);
2089                         btrfs_release_path(extent_root, path);
2090                         path->leave_spinning = 1;
2091
2092                         key.objectid = bytenr;
2093                         key.type = BTRFS_EXTENT_ITEM_KEY;
2094                         key.offset = num_bytes;
2095
2096                         ret = btrfs_search_slot(trans, extent_root,
2097                                                 &key, path, -1, 1);
2098                         if (ret) {
2099                                 printk(KERN_ERR "umm, got %d back from search"
2100                                        ", was looking for %llu\n", ret,
2101                                        (unsigned long long)bytenr);
2102                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2103                         }
2104                         BUG_ON(ret);
2105                         extent_slot = path->slots[0];
2106                 }
2107         } else {
2108                 btrfs_print_leaf(extent_root, path->nodes[0]);
2109                 WARN_ON(1);
2110                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2111                        "parent %llu root %llu  owner %llu offset %llu\n",
2112                        (unsigned long long)bytenr,
2113                        (unsigned long long)parent,
2114                        (unsigned long long)root_objectid,
2115                        (unsigned long long)owner_objectid,
2116                        (unsigned long long)owner_offset);
2117         }
2118
2119         leaf = path->nodes[0];
2120         item_size = btrfs_item_size_nr(leaf, extent_slot);
2121 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2122         if (item_size < sizeof(*ei)) {
2123                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2124                 ret = convert_extent_item_v0(trans, extent_root, path,
2125                                              owner_objectid, 0);
2126                 BUG_ON(ret < 0);
2127
2128                 btrfs_release_path(extent_root, path);
2129                 path->leave_spinning = 1;
2130
2131                 key.objectid = bytenr;
2132                 key.type = BTRFS_EXTENT_ITEM_KEY;
2133                 key.offset = num_bytes;
2134
2135                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2136                                         -1, 1);
2137                 if (ret) {
2138                         printk(KERN_ERR "umm, got %d back from search"
2139                                ", was looking for %llu\n", ret,
2140                                (unsigned long long)bytenr);
2141                         btrfs_print_leaf(extent_root, path->nodes[0]);
2142                 }
2143                 BUG_ON(ret);
2144                 extent_slot = path->slots[0];
2145                 leaf = path->nodes[0];
2146                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2147         }
2148 #endif
2149         BUG_ON(item_size < sizeof(*ei));
2150         ei = btrfs_item_ptr(leaf, extent_slot,
2151                             struct btrfs_extent_item);
2152         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2153                 struct btrfs_tree_block_info *bi;
2154                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2155                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2156                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2157         }
2158
2159         refs = btrfs_extent_refs(leaf, ei);
2160         BUG_ON(refs < refs_to_drop);
2161         refs -= refs_to_drop;
2162
2163         if (refs > 0) {
2164                 /*
2165                  * In the case of inline back ref, reference count will
2166                  * be updated by remove_extent_backref
2167                  */
2168                 if (iref) {
2169                         BUG_ON(!found_extent);
2170                 } else {
2171                         btrfs_set_extent_refs(leaf, ei, refs);
2172                         btrfs_mark_buffer_dirty(leaf);
2173                 }
2174                 if (found_extent) {
2175                         ret = remove_extent_backref(trans, extent_root, path,
2176                                                     iref, refs_to_drop,
2177                                                     is_data);
2178                         BUG_ON(ret);
2179                 }
2180         } else {
2181                 int mark_free = 0;
2182                 int pin = 1;
2183
2184                 if (found_extent) {
2185                         BUG_ON(is_data && refs_to_drop !=
2186                                extent_data_ref_count(root, path, iref));
2187                         if (iref) {
2188                                 BUG_ON(path->slots[0] != extent_slot);
2189                         } else {
2190                                 BUG_ON(path->slots[0] != extent_slot + 1);
2191                                 path->slots[0] = extent_slot;
2192                                 num_to_del = 2;
2193                         }
2194                 }
2195
2196                 if (ops && ops->free_extent) {
2197                         ret = ops->free_extent(root, bytenr, num_bytes);
2198                         if (ret > 0) {
2199                                 pin = 0;
2200                                 mark_free = 0;
2201                         }
2202                 }
2203
2204                 if (pin) {
2205                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2206                                              is_data);
2207                         if (ret > 0)
2208                                 mark_free = 1;
2209                         BUG_ON(ret < 0);
2210                 }
2211
2212                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2213                                       num_to_del);
2214                 BUG_ON(ret);
2215                 btrfs_release_path(extent_root, path);
2216
2217                 if (is_data) {
2218                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2219                         BUG_ON(ret);
2220                 }
2221
2222                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2223                                          mark_free);
2224                 BUG_ON(ret);
2225         }
2226         btrfs_free_path(path);
2227         finish_current_insert(trans, extent_root);
2228         return ret;
2229 }
2230
2231 /*
2232  * find all the blocks marked as pending in the radix tree and remove
2233  * them from the extent map
2234  */
2235 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2236                                btrfs_root *extent_root)
2237 {
2238         int ret;
2239         int err = 0;
2240         u64 start;
2241         u64 end;
2242         u64 priv;
2243         struct extent_io_tree *pending_del;
2244         struct extent_io_tree *extent_ins;
2245         struct pending_extent_op *extent_op;
2246
2247         extent_ins = &extent_root->fs_info->extent_ins;
2248         pending_del = &extent_root->fs_info->pending_del;
2249
2250         while(1) {
2251                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2252                                             EXTENT_LOCKED);
2253                 if (ret)
2254                         break;
2255
2256                 ret = get_state_private(pending_del, start, &priv);
2257                 BUG_ON(ret);
2258                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2259
2260                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2261                                   GFP_NOFS);
2262
2263                 if (!test_range_bit(extent_ins, start, end,
2264                                     EXTENT_LOCKED, 0)) {
2265                         ret = __free_extent(trans, extent_root,
2266                                             start, end + 1 - start, 0,
2267                                             extent_root->root_key.objectid,
2268                                             extent_op->level, 0, 1);
2269                         kfree(extent_op);
2270                 } else {
2271                         kfree(extent_op);
2272                         ret = get_state_private(extent_ins, start, &priv);
2273                         BUG_ON(ret);
2274                         extent_op = (struct pending_extent_op *)
2275                                                         (unsigned long)priv;
2276
2277                         clear_extent_bits(extent_ins, start, end,
2278                                           EXTENT_LOCKED, GFP_NOFS);
2279
2280                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2281                                 BUG_ON(1);
2282
2283                         kfree(extent_op);
2284                 }
2285                 if (ret)
2286                         err = ret;
2287         }
2288         return err;
2289 }
2290
2291 /*
2292  * remove an extent from the root, returns 0 on success
2293  */
2294
2295 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2296                       struct btrfs_root *root,
2297                       u64 bytenr, u64 num_bytes, u64 parent,
2298                       u64 root_objectid, u64 owner, u64 offset)
2299 {
2300         struct btrfs_root *extent_root = root->fs_info->extent_root;
2301         int pending_ret;
2302         int ret;
2303
2304         WARN_ON(num_bytes < root->sectorsize);
2305         if (root == extent_root) {
2306                 struct pending_extent_op *extent_op;
2307
2308                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2309                 BUG_ON(!extent_op);
2310
2311                 extent_op->type = PENDING_EXTENT_DELETE;
2312                 extent_op->bytenr = bytenr;
2313                 extent_op->num_bytes = num_bytes;
2314                 extent_op->level = (int)owner;
2315
2316                 set_extent_bits(&root->fs_info->pending_del,
2317                                 bytenr, bytenr + num_bytes - 1,
2318                                 EXTENT_LOCKED, GFP_NOFS);
2319                 set_state_private(&root->fs_info->pending_del,
2320                                   bytenr, (unsigned long)extent_op);
2321                 return 0;
2322         }
2323         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2324                             root_objectid, owner, offset, 1);
2325         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2326         return ret ? ret : pending_ret;
2327 }
2328
2329 static u64 stripe_align(struct btrfs_root *root, u64 val)
2330 {
2331         u64 mask = ((u64)root->stripesize - 1);
2332         u64 ret = (val + mask) & ~mask;
2333         return ret;
2334 }
2335
2336 /*
2337  * walks the btree of allocated extents and find a hole of a given size.
2338  * The key ins is changed to record the hole:
2339  * ins->objectid == block start
2340  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2341  * ins->offset == number of blocks
2342  * Any available blocks before search_start are skipped.
2343  */
2344 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2345                                      struct btrfs_root *orig_root,
2346                                      u64 num_bytes, u64 empty_size,
2347                                      u64 search_start, u64 search_end,
2348                                      u64 hint_byte, struct btrfs_key *ins,
2349                                      u64 exclude_start, u64 exclude_nr,
2350                                      int data)
2351 {
2352         int ret;
2353         u64 orig_search_start = search_start;
2354         struct btrfs_root * root = orig_root->fs_info->extent_root;
2355         struct btrfs_fs_info *info = root->fs_info;
2356         u64 total_needed = num_bytes;
2357         struct btrfs_block_group_cache *block_group;
2358         int full_scan = 0;
2359         int wrapped = 0;
2360
2361         WARN_ON(num_bytes < root->sectorsize);
2362         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2363
2364         if (hint_byte) {
2365                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2366                 if (!block_group)
2367                         hint_byte = search_start;
2368                 block_group = btrfs_find_block_group(root, block_group,
2369                                                      hint_byte, data, 1);
2370         } else {
2371                 block_group = btrfs_find_block_group(root,
2372                                                      trans->block_group,
2373                                                      search_start, data, 1);
2374         }
2375
2376         total_needed += empty_size;
2377
2378 check_failed:
2379         if (!block_group) {
2380                 block_group = btrfs_lookup_first_block_group(info,
2381                                                              search_start);
2382                 if (!block_group)
2383                         block_group = btrfs_lookup_first_block_group(info,
2384                                                        orig_search_start);
2385         }
2386         ret = find_search_start(root, &block_group, &search_start,
2387                                 total_needed, data);
2388         if (ret)
2389                 goto error;
2390
2391         search_start = stripe_align(root, search_start);
2392         ins->objectid = search_start;
2393         ins->offset = num_bytes;
2394
2395         if (ins->objectid + num_bytes >
2396             block_group->key.objectid + block_group->key.offset) {
2397                 search_start = block_group->key.objectid +
2398                         block_group->key.offset;
2399                 goto new_group;
2400         }
2401
2402         if (test_range_bit(&info->extent_ins, ins->objectid,
2403                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2404                 search_start = ins->objectid + num_bytes;
2405                 goto new_group;
2406         }
2407
2408         if (test_range_bit(&info->pinned_extents, ins->objectid,
2409                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2410                 search_start = ins->objectid + num_bytes;
2411                 goto new_group;
2412         }
2413
2414         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2415             ins->objectid < exclude_start + exclude_nr)) {
2416                 search_start = exclude_start + exclude_nr;
2417                 goto new_group;
2418         }
2419
2420         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2421                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2422                 if (block_group)
2423                         trans->block_group = block_group;
2424         }
2425         ins->offset = num_bytes;
2426         return 0;
2427
2428 new_group:
2429         block_group = btrfs_lookup_first_block_group(info, search_start);
2430         if (!block_group) {
2431                 search_start = orig_search_start;
2432                 if (full_scan) {
2433                         ret = -ENOSPC;
2434                         goto error;
2435                 }
2436                 if (wrapped) {
2437                         if (!full_scan)
2438                                 total_needed -= empty_size;
2439                         full_scan = 1;
2440                 } else
2441                         wrapped = 1;
2442         }
2443         cond_resched();
2444         block_group = btrfs_find_block_group(root, block_group,
2445                                              search_start, data, 0);
2446         goto check_failed;
2447
2448 error:
2449         return ret;
2450 }
2451
2452 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2453                                 struct btrfs_root *root,
2454                                 u64 num_bytes, u64 empty_size,
2455                                 u64 hint_byte, u64 search_end,
2456                                 struct btrfs_key *ins, int data)
2457 {
2458         int ret;
2459         u64 search_start = 0;
2460         u64 alloc_profile;
2461         struct btrfs_fs_info *info = root->fs_info;
2462
2463         if (info->extent_ops) {
2464                 struct btrfs_extent_ops *ops = info->extent_ops;
2465                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2466                 BUG_ON(ret);
2467                 goto found;
2468         }
2469
2470         if (data) {
2471                 alloc_profile = info->avail_data_alloc_bits &
2472                                 info->data_alloc_profile;
2473                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2474         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2475                    info->system_allocs >= 0) {
2476                 alloc_profile = info->avail_system_alloc_bits &
2477                                 info->system_alloc_profile;
2478                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2479         } else {
2480                 alloc_profile = info->avail_metadata_alloc_bits &
2481                                 info->metadata_alloc_profile;
2482                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2483         }
2484
2485         if (root->ref_cows) {
2486                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2487                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2488                                              num_bytes,
2489                                              BTRFS_BLOCK_GROUP_METADATA);
2490                         BUG_ON(ret);
2491                 }
2492                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2493                                      num_bytes + 2 * 1024 * 1024, data);
2494                 BUG_ON(ret);
2495         }
2496
2497         WARN_ON(num_bytes < root->sectorsize);
2498         ret = find_free_extent(trans, root, num_bytes, empty_size,
2499                                search_start, search_end, hint_byte, ins,
2500                                trans->alloc_exclude_start,
2501                                trans->alloc_exclude_nr, data);
2502         BUG_ON(ret);
2503 found:
2504         clear_extent_dirty(&root->fs_info->free_space_cache,
2505                            ins->objectid, ins->objectid + ins->offset - 1,
2506                            GFP_NOFS);
2507         return ret;
2508 }
2509
2510 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2511                                      struct btrfs_root *root,
2512                                      u64 root_objectid, u64 generation,
2513                                      u64 flags, struct btrfs_disk_key *key,
2514                                      int level, struct btrfs_key *ins)
2515 {
2516         int ret;
2517         struct btrfs_fs_info *fs_info = root->fs_info;
2518         struct btrfs_extent_item *extent_item;
2519         struct btrfs_tree_block_info *block_info;
2520         struct btrfs_extent_inline_ref *iref;
2521         struct btrfs_path *path;
2522         struct extent_buffer *leaf;
2523         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2524
2525         path = btrfs_alloc_path();
2526         BUG_ON(!path);
2527
2528         path->leave_spinning = 1;
2529         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2530                                       ins, size);
2531         BUG_ON(ret);
2532
2533         leaf = path->nodes[0];
2534         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2535                                      struct btrfs_extent_item);
2536         btrfs_set_extent_refs(leaf, extent_item, 1);
2537         btrfs_set_extent_generation(leaf, extent_item, generation);
2538         btrfs_set_extent_flags(leaf, extent_item,
2539                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2540         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2541
2542         btrfs_set_tree_block_key(leaf, block_info, key);
2543         btrfs_set_tree_block_level(leaf, block_info, level);
2544
2545         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2546         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2547         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2548
2549         btrfs_mark_buffer_dirty(leaf);
2550         btrfs_free_path(path);
2551
2552         ret = update_block_group(trans, root, ins->objectid, ins->offset,
2553                                  1, 0);
2554         if (ret) {
2555                 printk(KERN_ERR "btrfs update block group failed for %llu "
2556                        "%llu\n", (unsigned long long)ins->objectid,
2557                        (unsigned long long)ins->offset);
2558                 BUG();
2559         }
2560         return ret;
2561 }
2562
2563 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2564                             struct btrfs_root *root, u64 num_bytes,
2565                             u64 root_objectid, u64 generation,
2566                             u64 flags, struct btrfs_disk_key *key,
2567                             int level, u64 empty_size, u64 hint_byte,
2568                             u64 search_end, struct btrfs_key *ins)
2569 {
2570         int ret;
2571         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2572                                    hint_byte, search_end, ins, 0);
2573         BUG_ON(ret);
2574
2575         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2576                 struct pending_extent_op *extent_op;
2577
2578                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2579                 BUG_ON(!extent_op);
2580
2581                 extent_op->type = PENDING_EXTENT_INSERT;
2582                 extent_op->bytenr = ins->objectid;
2583                 extent_op->num_bytes = ins->offset;
2584                 extent_op->level = level;
2585                 extent_op->flags = flags;
2586                 memcpy(&extent_op->key, key, sizeof(*key));
2587
2588                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2589                                 ins->objectid + ins->offset - 1,
2590                                 EXTENT_LOCKED, GFP_NOFS);
2591                 set_state_private(&root->fs_info->extent_ins,
2592                                   ins->objectid, (unsigned long)extent_op);
2593         } else {
2594                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2595                                                 generation, flags,
2596                                                 key, level, ins);
2597                 finish_current_insert(trans, root->fs_info->extent_root);
2598                 del_pending_extents(trans, root->fs_info->extent_root);
2599         }
2600         return ret;
2601 }
2602
2603 /*
2604  * helper function to allocate a block for a given tree
2605  * returns the tree buffer or NULL.
2606  */
2607 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2608                                         struct btrfs_root *root,
2609                                         u32 blocksize, u64 root_objectid,
2610                                         struct btrfs_disk_key *key, int level,
2611                                         u64 hint, u64 empty_size)
2612 {
2613         struct btrfs_key ins;
2614         int ret;
2615         struct extent_buffer *buf;
2616
2617         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2618                                trans->transid, 0, key, level,
2619                                empty_size, hint, (u64)-1, &ins);
2620         if (ret) {
2621                 BUG_ON(ret > 0);
2622                 return ERR_PTR(ret);
2623         }
2624
2625         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2626         if (!buf) {
2627                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2628                                   0, root->root_key.objectid, level, 0);
2629                 BUG_ON(1);
2630                 return ERR_PTR(-ENOMEM);
2631         }
2632         btrfs_set_buffer_uptodate(buf);
2633         trans->blocks_used++;
2634
2635         return buf;
2636 }
2637
2638 #if 0
2639
2640 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2641                                   struct btrfs_root *root,
2642                                   struct extent_buffer *leaf)
2643 {
2644         u64 leaf_owner;
2645         u64 leaf_generation;
2646         struct btrfs_key key;
2647         struct btrfs_file_extent_item *fi;
2648         int i;
2649         int nritems;
2650         int ret;
2651
2652         BUG_ON(!btrfs_is_leaf(leaf));
2653         nritems = btrfs_header_nritems(leaf);
2654         leaf_owner = btrfs_header_owner(leaf);
2655         leaf_generation = btrfs_header_generation(leaf);
2656
2657         for (i = 0; i < nritems; i++) {
2658                 u64 disk_bytenr;
2659
2660                 btrfs_item_key_to_cpu(leaf, &key, i);
2661                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2662                         continue;
2663                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2664                 if (btrfs_file_extent_type(leaf, fi) ==
2665                     BTRFS_FILE_EXTENT_INLINE)
2666                         continue;
2667                 /*
2668                  * FIXME make sure to insert a trans record that
2669                  * repeats the snapshot del on crash
2670                  */
2671                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2672                 if (disk_bytenr == 0)
2673                         continue;
2674                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2675                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2676                                 leaf->start, leaf_owner, leaf_generation,
2677                                 key.objectid, 0);
2678                 BUG_ON(ret);
2679         }
2680         return 0;
2681 }
2682
2683 static void noinline reada_walk_down(struct btrfs_root *root,
2684                                      struct extent_buffer *node,
2685                                      int slot)
2686 {
2687         u64 bytenr;
2688         u64 last = 0;
2689         u32 nritems;
2690         u32 refs;
2691         u32 blocksize;
2692         int ret;
2693         int i;
2694         int level;
2695         int skipped = 0;
2696
2697         nritems = btrfs_header_nritems(node);
2698         level = btrfs_header_level(node);
2699         if (level)
2700                 return;
2701
2702         for (i = slot; i < nritems && skipped < 32; i++) {
2703                 bytenr = btrfs_node_blockptr(node, i);
2704                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2705                              (last > bytenr && last - bytenr > 32 * 1024))) {
2706                         skipped++;
2707                         continue;
2708                 }
2709                 blocksize = btrfs_level_size(root, level - 1);
2710                 if (i != slot) {
2711                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2712                                                       blocksize, &refs);
2713                         BUG_ON(ret);
2714                         if (refs != 1) {
2715                                 skipped++;
2716                                 continue;
2717                         }
2718                 }
2719                 mutex_unlock(&root->fs_info->fs_mutex);
2720                 ret = readahead_tree_block(root, bytenr, blocksize,
2721                                            btrfs_node_ptr_generation(node, i));
2722                 last = bytenr + blocksize;
2723                 cond_resched();
2724                 mutex_lock(&root->fs_info->fs_mutex);
2725                 if (ret)
2726                         break;
2727         }
2728 }
2729
2730 /*
2731  * helper function for drop_snapshot, this walks down the tree dropping ref
2732  * counts as it goes.
2733  */
2734 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2735                                    struct btrfs_root *root,
2736                                    struct btrfs_path *path, int *level)
2737 {
2738         u64 root_owner;
2739         u64 root_gen;
2740         u64 bytenr;
2741         u64 ptr_gen;
2742         struct extent_buffer *next;
2743         struct extent_buffer *cur;
2744         struct extent_buffer *parent;
2745         u32 blocksize;
2746         int ret;
2747         u32 refs;
2748
2749         WARN_ON(*level < 0);
2750         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2751         ret = btrfs_lookup_extent_ref(trans, root,
2752                                       path->nodes[*level]->start,
2753                                       path->nodes[*level]->len, &refs);
2754         BUG_ON(ret);
2755         if (refs > 1)
2756                 goto out;
2757
2758         /*
2759          * walk down to the last node level and free all the leaves
2760          */
2761         while(*level >= 0) {
2762                 WARN_ON(*level < 0);
2763                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2764                 cur = path->nodes[*level];
2765
2766                 if (btrfs_header_level(cur) != *level)
2767                         WARN_ON(1);
2768
2769                 if (path->slots[*level] >=
2770                     btrfs_header_nritems(cur))
2771                         break;
2772                 if (*level == 0) {
2773                         ret = drop_leaf_ref(trans, root, cur);
2774                         BUG_ON(ret);
2775                         break;
2776                 }
2777                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2778                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2779                 blocksize = btrfs_level_size(root, *level - 1);
2780                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2781                                               &refs);
2782                 BUG_ON(ret);
2783                 if (refs != 1) {
2784                         parent = path->nodes[*level];
2785                         root_owner = btrfs_header_owner(parent);
2786                         root_gen = btrfs_header_generation(parent);
2787                         path->slots[*level]++;
2788                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2789                                                 parent->start, root_owner,
2790                                                 root_gen, *level - 1, 1);
2791                         BUG_ON(ret);
2792                         continue;
2793                 }
2794                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2795                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2796                         free_extent_buffer(next);
2797                         reada_walk_down(root, cur, path->slots[*level]);
2798                         mutex_unlock(&root->fs_info->fs_mutex);
2799                         next = read_tree_block(root, bytenr, blocksize,
2800                                                ptr_gen);
2801                         mutex_lock(&root->fs_info->fs_mutex);
2802                 }
2803                 WARN_ON(*level <= 0);
2804                 if (path->nodes[*level-1])
2805                         free_extent_buffer(path->nodes[*level-1]);
2806                 path->nodes[*level-1] = next;
2807                 *level = btrfs_header_level(next);
2808                 path->slots[*level] = 0;
2809         }
2810 out:
2811         WARN_ON(*level < 0);
2812         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2813
2814         if (path->nodes[*level] == root->node) {
2815                 root_owner = root->root_key.objectid;
2816                 parent = path->nodes[*level];
2817         } else {
2818                 parent = path->nodes[*level + 1];
2819                 root_owner = btrfs_header_owner(parent);
2820         }
2821
2822         root_gen = btrfs_header_generation(parent);
2823         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2824                                 path->nodes[*level]->len, parent->start,
2825                                 root_owner, root_gen, *level, 1);
2826         free_extent_buffer(path->nodes[*level]);
2827         path->nodes[*level] = NULL;
2828         *level += 1;
2829         BUG_ON(ret);
2830         return 0;
2831 }
2832
2833 /*
2834  * helper for dropping snapshots.  This walks back up the tree in the path
2835  * to find the first node higher up where we haven't yet gone through
2836  * all the slots
2837  */
2838 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2839                                  struct btrfs_root *root,
2840                                  struct btrfs_path *path, int *level)
2841 {
2842         u64 root_owner;
2843         u64 root_gen;
2844         struct btrfs_root_item *root_item = &root->root_item;
2845         int i;
2846         int slot;
2847         int ret;
2848
2849         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2850                 slot = path->slots[i];
2851                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2852                         struct extent_buffer *node;
2853                         struct btrfs_disk_key disk_key;
2854                         node = path->nodes[i];
2855                         path->slots[i]++;
2856                         *level = i;
2857                         WARN_ON(*level == 0);
2858                         btrfs_node_key(node, &disk_key, path->slots[i]);
2859                         memcpy(&root_item->drop_progress,
2860                                &disk_key, sizeof(disk_key));
2861                         root_item->drop_level = i;
2862                         return 0;
2863                 } else {
2864                         struct extent_buffer *parent;
2865                         if (path->nodes[*level] == root->node)
2866                                 parent = path->nodes[*level];
2867                         else
2868                                 parent = path->nodes[*level + 1];
2869
2870                         root_owner = btrfs_header_owner(parent);
2871                         root_gen = btrfs_header_generation(parent);
2872                         ret = btrfs_free_extent(trans, root,
2873                                                 path->nodes[*level]->start,
2874                                                 path->nodes[*level]->len,
2875                                                 parent->start, root_owner,
2876                                                 root_gen, *level, 1);
2877                         BUG_ON(ret);
2878                         free_extent_buffer(path->nodes[*level]);
2879                         path->nodes[*level] = NULL;
2880                         *level = i + 1;
2881                 }
2882         }
2883         return 1;
2884 }
2885
2886 /*
2887  * drop the reference count on the tree rooted at 'snap'.  This traverses
2888  * the tree freeing any blocks that have a ref count of zero after being
2889  * decremented.
2890  */
2891 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2892                         *root)
2893 {
2894         int ret = 0;
2895         int wret;
2896         int level;
2897         struct btrfs_path *path;
2898         int i;
2899         int orig_level;
2900         struct btrfs_root_item *root_item = &root->root_item;
2901
2902         path = btrfs_alloc_path();
2903         BUG_ON(!path);
2904
2905         level = btrfs_header_level(root->node);
2906         orig_level = level;
2907         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2908                 path->nodes[level] = root->node;
2909                 extent_buffer_get(root->node);
2910                 path->slots[level] = 0;
2911         } else {
2912                 struct btrfs_key key;
2913                 struct btrfs_disk_key found_key;
2914                 struct extent_buffer *node;
2915
2916                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2917                 level = root_item->drop_level;
2918                 path->lowest_level = level;
2919                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2920                 if (wret < 0) {
2921                         ret = wret;
2922                         goto out;
2923                 }
2924                 node = path->nodes[level];
2925                 btrfs_node_key(node, &found_key, path->slots[level]);
2926                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2927                                sizeof(found_key)));
2928         }
2929         while(1) {
2930                 wret = walk_down_tree(trans, root, path, &level);
2931                 if (wret < 0)
2932                         ret = wret;
2933                 if (wret != 0)
2934                         break;
2935
2936                 wret = walk_up_tree(trans, root, path, &level);
2937                 if (wret < 0)
2938                         ret = wret;
2939                 if (wret != 0)
2940                         break;
2941                 /*
2942                 ret = -EAGAIN;
2943                 break;
2944                 */
2945         }
2946         for (i = 0; i <= orig_level; i++) {
2947                 if (path->nodes[i]) {
2948                         free_extent_buffer(path->nodes[i]);
2949                         path->nodes[i] = NULL;
2950                 }
2951         }
2952 out:
2953         btrfs_free_path(path);
2954         return ret;
2955 }
2956
2957 #endif
2958
2959 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2960 {
2961         u64 start;
2962         u64 end;
2963         u64 ptr;
2964         int ret;
2965         while(1) {
2966                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2967                                             &start, &end, (unsigned int)-1);
2968                 if (ret)
2969                         break;
2970                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2971                 if (!ret)
2972                         kfree((void *)(unsigned long)ptr);
2973                 clear_extent_bits(&info->block_group_cache, start,
2974                                   end, (unsigned int)-1, GFP_NOFS);
2975         }
2976         while(1) {
2977                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2978                                             &start, &end, EXTENT_DIRTY);
2979                 if (ret)
2980                         break;
2981                 clear_extent_dirty(&info->free_space_cache, start,
2982                                    end, GFP_NOFS);
2983         }
2984         return 0;
2985 }
2986
2987 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2988                            struct btrfs_key *key)
2989 {
2990         int ret;
2991         struct btrfs_key found_key;
2992         struct extent_buffer *leaf;
2993         int slot;
2994
2995         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2996         if (ret < 0)
2997                 return ret;
2998         while(1) {
2999                 slot = path->slots[0];
3000                 leaf = path->nodes[0];
3001                 if (slot >= btrfs_header_nritems(leaf)) {
3002                         ret = btrfs_next_leaf(root, path);
3003                         if (ret == 0)
3004                                 continue;
3005                         if (ret < 0)
3006                                 goto error;
3007                         break;
3008                 }
3009                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3010
3011                 if (found_key.objectid >= key->objectid &&
3012                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3013                         return 0;
3014                 path->slots[0]++;
3015         }
3016         ret = -ENOENT;
3017 error:
3018         return ret;
3019 }
3020
3021 int btrfs_read_block_groups(struct btrfs_root *root)
3022 {
3023         struct btrfs_path *path;
3024         int ret;
3025         int bit;
3026         struct btrfs_block_group_cache *cache;
3027         struct btrfs_fs_info *info = root->fs_info;
3028         struct btrfs_space_info *space_info;
3029         struct extent_io_tree *block_group_cache;
3030         struct btrfs_key key;
3031         struct btrfs_key found_key;
3032         struct extent_buffer *leaf;
3033
3034         block_group_cache = &info->block_group_cache;
3035
3036         root = info->extent_root;
3037         key.objectid = 0;
3038         key.offset = 0;
3039         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3040         path = btrfs_alloc_path();
3041         if (!path)
3042                 return -ENOMEM;
3043
3044         while(1) {
3045                 ret = find_first_block_group(root, path, &key);
3046                 if (ret > 0) {
3047                         ret = 0;
3048                         goto error;
3049                 }
3050                 if (ret != 0) {
3051                         goto error;
3052                 }
3053                 leaf = path->nodes[0];
3054                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3055                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3056                 if (!cache) {
3057                         ret = -ENOMEM;
3058                         break;
3059                 }
3060
3061                 read_extent_buffer(leaf, &cache->item,
3062                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3063                                    sizeof(cache->item));
3064                 memcpy(&cache->key, &found_key, sizeof(found_key));
3065                 cache->cached = 0;
3066                 cache->pinned = 0;
3067                 key.objectid = found_key.objectid + found_key.offset;
3068                 btrfs_release_path(root, path);
3069                 cache->flags = btrfs_block_group_flags(&cache->item);
3070                 bit = 0;
3071                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3072                         bit = BLOCK_GROUP_DATA;
3073                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3074                         bit = BLOCK_GROUP_SYSTEM;
3075                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3076                         bit = BLOCK_GROUP_METADATA;
3077                 }
3078                 set_avail_alloc_bits(info, cache->flags);
3079                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3080                         cache->ro = 1;
3081
3082                 ret = update_space_info(info, cache->flags, found_key.offset,
3083                                         btrfs_block_group_used(&cache->item),
3084                                         &space_info);
3085                 BUG_ON(ret);
3086                 cache->space_info = space_info;
3087
3088                 /* use EXTENT_LOCKED to prevent merging */
3089                 set_extent_bits(block_group_cache, found_key.objectid,
3090                                 found_key.objectid + found_key.offset - 1,
3091                                 bit | EXTENT_LOCKED, GFP_NOFS);
3092                 set_state_private(block_group_cache, found_key.objectid,
3093                                   (unsigned long)cache);
3094         }
3095         ret = 0;
3096 error:
3097         btrfs_free_path(path);
3098         return ret;
3099 }
3100
3101 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3102                            struct btrfs_root *root, u64 bytes_used,
3103                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3104                            u64 size)
3105 {
3106         int ret;
3107         int bit = 0;
3108         struct btrfs_root *extent_root;
3109         struct btrfs_block_group_cache *cache;
3110         struct extent_io_tree *block_group_cache;
3111
3112         extent_root = root->fs_info->extent_root;
3113         block_group_cache = &root->fs_info->block_group_cache;
3114
3115         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3116         BUG_ON(!cache);
3117         cache->key.objectid = chunk_offset;
3118         cache->key.offset = size;
3119
3120         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3121         btrfs_set_block_group_used(&cache->item, bytes_used);
3122         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3123         cache->flags = type;
3124         btrfs_set_block_group_flags(&cache->item, type);
3125
3126         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3127                                 &cache->space_info);
3128         BUG_ON(ret);
3129
3130         bit = block_group_state_bits(type);
3131         set_extent_bits(block_group_cache, chunk_offset,
3132                         chunk_offset + size - 1,
3133                         bit | EXTENT_LOCKED, GFP_NOFS);
3134
3135         set_state_private(block_group_cache, chunk_offset,
3136                           (unsigned long)cache);
3137         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3138                                 sizeof(cache->item));
3139         BUG_ON(ret);
3140
3141         finish_current_insert(trans, extent_root);
3142         ret = del_pending_extents(trans, extent_root);
3143         BUG_ON(ret);
3144         set_avail_alloc_bits(extent_root->fs_info, type);
3145         return 0;
3146 }
3147
3148 /*
3149  * This is for converter use only.
3150  *
3151  * In that case, we don't know where are free blocks located.
3152  * Therefore all block group cache entries must be setup properly
3153  * before doing any block allocation.
3154  */
3155 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3156                             struct btrfs_root *root)
3157 {
3158         u64 total_bytes;
3159         u64 cur_start;
3160         u64 group_type;
3161         u64 group_size;
3162         u64 group_align;
3163         u64 total_data = 0;
3164         u64 total_metadata = 0;
3165         u64 chunk_objectid;
3166         int ret;
3167         int bit;
3168         struct btrfs_root *extent_root;
3169         struct btrfs_block_group_cache *cache;
3170         struct extent_io_tree *block_group_cache;
3171
3172         extent_root = root->fs_info->extent_root;
3173         block_group_cache = &root->fs_info->block_group_cache;
3174         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3175         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3176         group_align = 64 * root->sectorsize;
3177
3178         cur_start = 0;
3179         while (cur_start < total_bytes) {
3180                 group_size = total_bytes / 12;
3181                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3182                 if (cur_start == 0) {
3183                         bit = BLOCK_GROUP_SYSTEM;
3184                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3185                         group_size /= 4;
3186                         group_size &= ~(group_align - 1);
3187                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3188                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3189                 } else {
3190                         group_size &= ~(group_align - 1);
3191                         if (total_data >= total_metadata * 2) {
3192                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3193                                 group_size = min_t(u64, group_size,
3194                                                    1ULL * 1024 * 1024 * 1024);
3195                                 total_metadata += group_size;
3196                         } else {
3197                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3198                                 group_size = min_t(u64, group_size,
3199                                                    5ULL * 1024 * 1024 * 1024);
3200                                 total_data += group_size;
3201                         }
3202                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3203                                 group_size = total_bytes - cur_start;
3204                 }
3205
3206                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3207                 BUG_ON(!cache);
3208
3209                 cache->key.objectid = cur_start;
3210                 cache->key.offset = group_size;
3211                 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3212
3213                 btrfs_set_block_group_used(&cache->item, 0);
3214                 btrfs_set_block_group_chunk_objectid(&cache->item,
3215                                                      chunk_objectid);
3216                 btrfs_set_block_group_flags(&cache->item, group_type);
3217
3218                 cache->flags = group_type;
3219
3220                 ret = update_space_info(root->fs_info, group_type, group_size,
3221                                         0, &cache->space_info);
3222                 BUG_ON(ret);
3223                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3224
3225                 set_extent_bits(block_group_cache, cur_start,
3226                                 cur_start + group_size - 1,
3227                                 bit | EXTENT_LOCKED, GFP_NOFS);
3228                 set_state_private(block_group_cache, cur_start,
3229                                   (unsigned long)cache);
3230                 cur_start += group_size;
3231         }
3232         /* then insert all the items */
3233         cur_start = 0;
3234         while(cur_start < total_bytes) {
3235                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3236                 BUG_ON(!cache);
3237
3238                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3239                                         sizeof(cache->item));
3240                 BUG_ON(ret);
3241
3242                 finish_current_insert(trans, extent_root);
3243                 ret = del_pending_extents(trans, extent_root);
3244                 BUG_ON(ret);
3245
3246                 cur_start = cache->key.objectid + cache->key.offset;
3247         }
3248         return 0;
3249 }
3250
3251 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3252                              struct btrfs_root *root,
3253                              u64 bytenr, u64 num_bytes, int alloc,
3254                              int mark_free)
3255 {
3256         return update_block_group(trans, root, bytenr, num_bytes,
3257                                   alloc, mark_free);
3258 }