extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <stdint.h>
  22 #include "kerncompat.h"
  23 #include "radix-tree.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "crc32c.h"
  29 #include "volumes.h"
  30 #include "free-space-cache.h"
  31 #include "math.h"
  32 #include "utils.h"
  33
  34 #define PENDING_EXTENT_INSERT 0
  35 #define PENDING_EXTENT_DELETE 1
  36 #define PENDING_BACKREF_UPDATE 2
  37
  38 struct pending_extent_op {
  39         int type;
  40         u64 bytenr;
  41         u64 num_bytes;
  42         u64 flags;
  43         struct btrfs_disk_key key;
  44         int level;
  45 };
  46
  47 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  48                                      struct btrfs_root *root,
  49                                      u64 root_objectid, u64 generation,
  50                                      u64 flags, struct btrfs_disk_key *key,
  51                                      int level, struct btrfs_key *ins);
  52 static int __free_extent(struct btrfs_trans_handle *trans,
  53                          struct btrfs_root *root,
  54                          u64 bytenr, u64 num_bytes, u64 parent,
  55                          u64 root_objectid, u64 owner_objectid,
  56                          u64 owner_offset, int refs_to_drop);
  57 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  58                                  btrfs_root *extent_root);
  59 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  60                                btrfs_root *extent_root);
  61 static struct btrfs_block_group_cache *
  62 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
  63                        *hint, u64 search_start, int data, int owner);
  64
  65 static int remove_sb_from_cache(struct btrfs_root *root,
  66                                 struct btrfs_block_group_cache *cache)
  67 {
  68         u64 bytenr;
  69         u64 *logical;
  70         int stripe_len;
  71         int i, nr, ret;
  72         struct extent_io_tree *free_space_cache;
  73
  74         free_space_cache = &root->fs_info->free_space_cache;
  75         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
  76                 bytenr = btrfs_sb_offset(i);
  77                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
  78                                        cache->key.objectid, bytenr, 0,
  79                                        &logical, &nr, &stripe_len);
  80                 BUG_ON(ret);
  81                 while (nr--) {
  82                         clear_extent_dirty(free_space_cache, logical[nr],
  83                                 logical[nr] + stripe_len - 1, GFP_NOFS);
  84                 }
  85                 kfree(logical);
  86         }
  87         return 0;
  88 }
  89
  90 static int cache_block_group(struct btrfs_root *root,
  91                              struct btrfs_block_group_cache *block_group)
  92 {
  93         struct btrfs_path *path;
  94         int ret;
  95         struct btrfs_key key;
  96         struct extent_buffer *leaf;
  97         struct extent_io_tree *free_space_cache;
  98         int slot;
  99         u64 last;
 100         u64 hole_size;
 101
 102         if (!block_group)
 103                 return 0;
 104
 105         root = root->fs_info->extent_root;
 106         free_space_cache = &root->fs_info->free_space_cache;
 107
 108         if (block_group->cached)
 109                 return 0;
 110
 111         path = btrfs_alloc_path();
 112         if (!path)
 113                 return -ENOMEM;
 114
 115         path->reada = 2;
 116         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 117         key.objectid = last;
 118         key.offset = 0;
 119         key.type = 0;
 120
 121         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 122         if (ret < 0)
 123                 goto err;
 124
 125         while(1) {
 126                 leaf = path->nodes[0];
 127                 slot = path->slots[0];
 128                 if (slot >= btrfs_header_nritems(leaf)) {
 129                         ret = btrfs_next_leaf(root, path);
 130                         if (ret < 0)
 131                                 goto err;
 132                         if (ret == 0) {
 133                                 continue;
 134                         } else {
 135                                 break;
 136                         }
 137                 }
 138                 btrfs_item_key_to_cpu(leaf, &key, slot);
 139                 if (key.objectid < block_group->key.objectid) {
 140                         goto next;
 141                 }
 142                 if (key.objectid >= block_group->key.objectid +
 143                     block_group->key.offset) {
 144                         break;
 145                 }
 146
 147                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 148                     key.type == BTRFS_METADATA_ITEM_KEY) {
 149                         if (key.objectid > last) {
 150                                 hole_size = key.objectid - last;
 151                                 set_extent_dirty(free_space_cache, last,
 152                                                  last + hole_size - 1,
 153                                                  GFP_NOFS);
 154                         }
 155                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 156                                 last = key.objectid + root->leafsize;
 157                         else
 158                                 last = key.objectid + key.offset;
 159                 }
 160 next:
 161                 path->slots[0]++;
 162         }
 163
 164         if (block_group->key.objectid +
 165             block_group->key.offset > last) {
 166                 hole_size = block_group->key.objectid +
 167                         block_group->key.offset - last;
 168                 set_extent_dirty(free_space_cache, last,
 169                                  last + hole_size - 1, GFP_NOFS);
 170         }
 171         remove_sb_from_cache(root, block_group);
 172         block_group->cached = 1;
 173 err:
 174         btrfs_free_path(path);
 175         return 0;
 176 }
 177
 178 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 179                                                        btrfs_fs_info *info,
 180                                                        u64 bytenr)
 181 {
 182         struct extent_io_tree *block_group_cache;
 183         struct btrfs_block_group_cache *block_group = NULL;
 184         u64 ptr;
 185         u64 start;
 186         u64 end;
 187         int ret;
 188
 189         bytenr = max_t(u64, bytenr,
 190                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 191         block_group_cache = &info->block_group_cache;
 192         ret = find_first_extent_bit(block_group_cache,
 193                                     bytenr, &start, &end,
 194                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 195                                     BLOCK_GROUP_SYSTEM);
 196         if (ret) {
 197                 return NULL;
 198         }
 199         ret = get_state_private(block_group_cache, start, &ptr);
 200         if (ret)
 201                 return NULL;
 202
 203         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 204         return block_group;
 205 }
 206
 207 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 208                                                          btrfs_fs_info *info,
 209                                                          u64 bytenr)
 210 {
 211         struct extent_io_tree *block_group_cache;
 212         struct btrfs_block_group_cache *block_group = NULL;
 213         u64 ptr;
 214         u64 start;
 215         u64 end;
 216         int ret;
 217
 218         block_group_cache = &info->block_group_cache;
 219         ret = find_first_extent_bit(block_group_cache,
 220                                     bytenr, &start, &end,
 221                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 222                                     BLOCK_GROUP_SYSTEM);
 223         if (ret) {
 224                 return NULL;
 225         }
 226         ret = get_state_private(block_group_cache, start, &ptr);
 227         if (ret)
 228                 return NULL;
 229
 230         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 231         if (block_group->key.objectid <= bytenr && bytenr <
 232             block_group->key.objectid + block_group->key.offset)
 233                 return block_group;
 234         return NULL;
 235 }
 236
 237 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 238 {
 239         return (cache->flags & bits) == bits;
 240 }
 241
 242 static int noinline find_search_start(struct btrfs_root *root,
 243                               struct btrfs_block_group_cache **cache_ret,
 244                               u64 *start_ret, int num, int data)
 245 {
 246         int ret;
 247         struct btrfs_block_group_cache *cache = *cache_ret;
 248         u64 last = *start_ret;
 249         u64 start = 0;
 250         u64 end = 0;
 251         u64 search_start = *start_ret;
 252         int wrapped = 0;
 253
 254         if (!cache)
 255                 goto out;
 256 again:
 257         ret = cache_block_group(root, cache);
 258         if (ret)
 259                 goto out;
 260
 261         last = max(search_start, cache->key.objectid);
 262         if (cache->ro || !block_group_bits(cache, data))
 263                 goto new_group;
 264
 265         while(1) {
 266                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 267                                             last, &start, &end, EXTENT_DIRTY);
 268                 if (ret) {
 269                         goto new_group;
 270                 }
 271
 272                 start = max(last, start);
 273                 last = end + 1;
 274                 if (last - start < num) {
 275                         continue;
 276                 }
 277                 if (start + num > cache->key.objectid + cache->key.offset) {
 278                         goto new_group;
 279                 }
 280                 *start_ret = start;
 281                 return 0;
 282         }
 283 out:
 284         *start_ret = last;
 285         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 286         if (!cache) {
 287                 printk("Unable to find block group for %llu\n",
 288                         (unsigned long long)search_start);
 289                 WARN_ON(1);
 290         }
 291         return -ENOSPC;
 292
 293 new_group:
 294         last = cache->key.objectid + cache->key.offset;
 295 wrapped:
 296         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 297         if (!cache) {
 298                 if (!wrapped) {
 299                         wrapped = 1;
 300                         last = search_start;
 301                         goto wrapped;
 302                 }
 303                 goto out;
 304         }
 305         *cache_ret = cache;
 306         goto again;
 307 }
 308
 309 static int block_group_state_bits(u64 flags)
 310 {
 311         int bits = 0;
 312         if (flags & BTRFS_BLOCK_GROUP_DATA)
 313                 bits |= BLOCK_GROUP_DATA;
 314         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 315                 bits |= BLOCK_GROUP_METADATA;
 316         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 317                 bits |= BLOCK_GROUP_SYSTEM;
 318         return bits;
 319 }
 320
 321 static struct btrfs_block_group_cache *
 322 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
 323                        *hint, u64 search_start, int data, int owner)
 324 {
 325         struct btrfs_block_group_cache *cache;
 326         struct extent_io_tree *block_group_cache;
 327         struct btrfs_block_group_cache *found_group = NULL;
 328         struct btrfs_fs_info *info = root->fs_info;
 329         u64 used;
 330         u64 last = 0;
 331         u64 hint_last;
 332         u64 start;
 333         u64 end;
 334         u64 free_check;
 335         u64 ptr;
 336         int bit;
 337         int ret;
 338         int full_search = 0;
 339         int factor = 10;
 340
 341         block_group_cache = &info->block_group_cache;
 342
 343         if (!owner)
 344                 factor = 10;
 345
 346         bit = block_group_state_bits(data);
 347
 348         if (search_start) {
 349                 struct btrfs_block_group_cache *shint;
 350                 shint = btrfs_lookup_block_group(info, search_start);
 351                 if (shint && !shint->ro && block_group_bits(shint, data)) {
 352                         used = btrfs_block_group_used(&shint->item);
 353                         if (used + shint->pinned <
 354                             div_factor(shint->key.offset, factor)) {
 355                                 return shint;
 356                         }
 357                 }
 358         }
 359         if (hint && !hint->ro && block_group_bits(hint, data)) {
 360                 used = btrfs_block_group_used(&hint->item);
 361                 if (used + hint->pinned <
 362                     div_factor(hint->key.offset, factor)) {
 363                         return hint;
 364                 }
 365                 last = hint->key.objectid + hint->key.offset;
 366                 hint_last = last;
 367         } else {
 368                 if (hint)
 369                         hint_last = max(hint->key.objectid, search_start);
 370                 else
 371                         hint_last = search_start;
 372
 373                 last = hint_last;
 374         }
 375 again:
 376         while(1) {
 377                 ret = find_first_extent_bit(block_group_cache, last,
 378                                             &start, &end, bit);
 379                 if (ret)
 380                         break;
 381
 382                 ret = get_state_private(block_group_cache, start, &ptr);
 383                 if (ret)
 384                         break;
 385
 386                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 387                 last = cache->key.objectid + cache->key.offset;
 388                 used = btrfs_block_group_used(&cache->item);
 389
 390                 if (!cache->ro && block_group_bits(cache, data)) {
 391                         if (full_search)
 392                                 free_check = cache->key.offset;
 393                         else
 394                                 free_check = div_factor(cache->key.offset,
 395                                                         factor);
 396
 397                         if (used + cache->pinned < free_check) {
 398                                 found_group = cache;
 399                                 goto found;
 400                         }
 401                 }
 402                 cond_resched();
 403         }
 404         if (!full_search) {
 405                 last = search_start;
 406                 full_search = 1;
 407                 goto again;
 408         }
 409 found:
 410         return found_group;
 411 }
 412
 413 /*
 414  * Back reference rules.  Back refs have three main goals:
 415  *
 416  * 1) differentiate between all holders of references to an extent so that
 417  *    when a reference is dropped we can make sure it was a valid reference
 418  *    before freeing the extent.
 419  *
 420  * 2) Provide enough information to quickly find the holders of an extent
 421  *    if we notice a given block is corrupted or bad.
 422  *
 423  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 424  *    maintenance.  This is actually the same as #2, but with a slightly
 425  *    different use case.
 426  *
 427  * There are two kinds of back refs. The implicit back refs is optimized
 428  * for pointers in non-shared tree blocks. For a given pointer in a block,
 429  * back refs of this kind provide information about the block's owner tree
 430  * and the pointer's key. These information allow us to find the block by
 431  * b-tree searching. The full back refs is for pointers in tree blocks not
 432  * referenced by their owner trees. The location of tree block is recorded
 433  * in the back refs. Actually the full back refs is generic, and can be
 434  * used in all cases the implicit back refs is used. The major shortcoming
 435  * of the full back refs is its overhead. Every time a tree block gets
 436  * COWed, we have to update back refs entry for all pointers in it.
 437  *
 438  * For a newly allocated tree block, we use implicit back refs for
 439  * pointers in it. This means most tree related operations only involve
 440  * implicit back refs. For a tree block created in old transaction, the
 441  * only way to drop a reference to it is COW it. So we can detect the
 442  * event that tree block loses its owner tree's reference and do the
 443  * back refs conversion.
 444  *
 445  * When a tree block is COW'd through a tree, there are four cases:
 446  *
 447  * The reference count of the block is one and the tree is the block's
 448  * owner tree. Nothing to do in this case.
 449  *
 450  * The reference count of the block is one and the tree is not the
 451  * block's owner tree. In this case, full back refs is used for pointers
 452  * in the block. Remove these full back refs, add implicit back refs for
 453  * every pointers in the new block.
 454  *
 455  * The reference count of the block is greater than one and the tree is
 456  * the block's owner tree. In this case, implicit back refs is used for
 457  * pointers in the block. Add full back refs for every pointers in the
 458  * block, increase lower level extents' reference counts. The original
 459  * implicit back refs are entailed to the new block.
 460  *
 461  * The reference count of the block is greater than one and the tree is
 462  * not the block's owner tree. Add implicit back refs for every pointer in
 463  * the new block, increase lower level extents' reference count.
 464  *
 465  * Back Reference Key composing:
 466  *
 467  * The key objectid corresponds to the first byte in the extent,
 468  * The key type is used to differentiate between types of back refs.
 469  * There are different meanings of the key offset for different types
 470  * of back refs.
 471  *
 472  * File extents can be referenced by:
 473  *
 474  * - multiple snapshots, subvolumes, or different generations in one subvol
 475  * - different files inside a single subvolume
 476  * - different offsets inside a file (bookend extents in file.c)
 477  *
 478  * The extent ref structure for the implicit back refs has fields for:
 479  *
 480  * - Objectid of the subvolume root
 481  * - objectid of the file holding the reference
 482  * - original offset in the file
 483  * - how many bookend extents
 484  *
 485  * The key offset for the implicit back refs is hash of the first
 486  * three fields.
 487  *
 488  * The extent ref structure for the full back refs has field for:
 489  *
 490  * - number of pointers in the tree leaf
 491  *
 492  * The key offset for the implicit back refs is the first byte of
 493  * the tree leaf
 494  *
 495  * When a file extent is allocated, The implicit back refs is used.
 496  * the fields are filled in:
 497  *
 498  *     (root_key.objectid, inode objectid, offset in file, 1)
 499  *
 500  * When a file extent is removed file truncation, we find the
 501  * corresponding implicit back refs and check the following fields:
 502  *
 503  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 504  *
 505  * Btree extents can be referenced by:
 506  *
 507  * - Different subvolumes
 508  *
 509  * Both the implicit back refs and the full back refs for tree blocks
 510  * only consist of key. The key offset for the implicit back refs is
 511  * objectid of block's owner tree. The key offset for the full back refs
 512  * is the first byte of parent block.
 513  *
 514  * When implicit back refs is used, information about the lowest key and
 515  * level of the tree block are required. These information are stored in
 516  * tree block info structure.
 517  */
 518
 519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 520 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 521                                   struct btrfs_root *root,
 522                                   struct btrfs_path *path,
 523                                   u64 owner, u32 extra_size)
 524 {
 525         struct btrfs_extent_item *item;
 526         struct btrfs_extent_item_v0 *ei0;
 527         struct btrfs_extent_ref_v0 *ref0;
 528         struct btrfs_tree_block_info *bi;
 529         struct extent_buffer *leaf;
 530         struct btrfs_key key;
 531         struct btrfs_key found_key;
 532         u32 new_size = sizeof(*item);
 533         u64 refs;
 534         int ret;
 535
 536         leaf = path->nodes[0];
 537         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 538
 539         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 540         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 541                              struct btrfs_extent_item_v0);
 542         refs = btrfs_extent_refs_v0(leaf, ei0);
 543
 544         if (owner == (u64)-1) {
 545                 while (1) {
 546                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 547                                 ret = btrfs_next_leaf(root, path);
 548                                 if (ret < 0)
 549                                         return ret;
 550                                 BUG_ON(ret > 0);
 551                                 leaf = path->nodes[0];
 552                         }
 553                         btrfs_item_key_to_cpu(leaf, &found_key,
 554                                               path->slots[0]);
 555                         BUG_ON(key.objectid != found_key.objectid);
 556                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 557                                 path->slots[0]++;
 558                                 continue;
 559                         }
 560                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 561                                               struct btrfs_extent_ref_v0);
 562                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 563                         break;
 564                 }
 565         }
 566         btrfs_release_path(path);
 567
 568         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 569                 new_size += sizeof(*bi);
 570
 571         new_size -= sizeof(*ei0);
 572         ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
 573         if (ret < 0)
 574                 return ret;
 575         BUG_ON(ret);
 576
 577         ret = btrfs_extend_item(trans, root, path, new_size);
 578         BUG_ON(ret);
 579
 580         leaf = path->nodes[0];
 581         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 582         btrfs_set_extent_refs(leaf, item, refs);
 583         /* FIXME: get real generation */
 584         btrfs_set_extent_generation(leaf, item, 0);
 585         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 586                 btrfs_set_extent_flags(leaf, item,
 587                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 588                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 589                 bi = (struct btrfs_tree_block_info *)(item + 1);
 590                 /* FIXME: get first key of the block */
 591                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 592                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 593         } else {
 594                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 595         }
 596         btrfs_mark_buffer_dirty(leaf);
 597         return 0;
 598 }
 599 #endif
 600
 601 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 602 {
 603         u32 high_crc = ~(u32)0;
 604         u32 low_crc = ~(u32)0;
 605         __le64 lenum;
 606
 607         lenum = cpu_to_le64(root_objectid);
 608         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 609         lenum = cpu_to_le64(owner);
 610         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 611         lenum = cpu_to_le64(offset);
 612         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 613
 614         return ((u64)high_crc << 31) ^ (u64)low_crc;
 615 }
 616
 617 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 618                                      struct btrfs_extent_data_ref *ref)
 619 {
 620         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 621                                     btrfs_extent_data_ref_objectid(leaf, ref),
 622                                     btrfs_extent_data_ref_offset(leaf, ref));
 623 }
 624
 625 static int match_extent_data_ref(struct extent_buffer *leaf,
 626                                  struct btrfs_extent_data_ref *ref,
 627                                  u64 root_objectid, u64 owner, u64 offset)
 628 {
 629         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 630             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 631             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 632                 return 0;
 633         return 1;
 634 }
 635
 636 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 637                                            struct btrfs_root *root,
 638                                            struct btrfs_path *path,
 639                                            u64 bytenr, u64 parent,
 640                                            u64 root_objectid,
 641                                            u64 owner, u64 offset)
 642 {
 643         struct btrfs_key key;
 644         struct btrfs_extent_data_ref *ref;
 645         struct extent_buffer *leaf;
 646         u32 nritems;
 647         int ret;
 648         int recow;
 649         int err = -ENOENT;
 650
 651         key.objectid = bytenr;
 652         if (parent) {
 653                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 654                 key.offset = parent;
 655         } else {
 656                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 657                 key.offset = hash_extent_data_ref(root_objectid,
 658                                                   owner, offset);
 659         }
 660 again:
 661         recow = 0;
 662         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 663         if (ret < 0) {
 664                 err = ret;
 665                 goto fail;
 666         }
 667
 668         if (parent) {
 669                 if (!ret)
 670                         return 0;
 671 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 672                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 673                 btrfs_release_path(path);
 674                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 675                 if (ret < 0) {
 676                         err = ret;
 677                         goto fail;
 678                 }
 679                 if (!ret)
 680                         return 0;
 681 #endif
 682                 goto fail;
 683         }
 684
 685         leaf = path->nodes[0];
 686         nritems = btrfs_header_nritems(leaf);
 687         while (1) {
 688                 if (path->slots[0] >= nritems) {
 689                         ret = btrfs_next_leaf(root, path);
 690                         if (ret < 0)
 691                                 err = ret;
 692                         if (ret)
 693                                 goto fail;
 694
 695                         leaf = path->nodes[0];
 696                         nritems = btrfs_header_nritems(leaf);
 697                         recow = 1;
 698                 }
 699
 700                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 701                 if (key.objectid != bytenr ||
 702                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 703                         goto fail;
 704
 705                 ref = btrfs_item_ptr(leaf, path->slots[0],
 706                                      struct btrfs_extent_data_ref);
 707
 708                 if (match_extent_data_ref(leaf, ref, root_objectid,
 709                                           owner, offset)) {
 710                         if (recow) {
 711                                 btrfs_release_path(path);
 712                                 goto again;
 713                         }
 714                         err = 0;
 715                         break;
 716                 }
 717                 path->slots[0]++;
 718         }
 719 fail:
 720         return err;
 721 }
 722
 723 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 724                                            struct btrfs_root *root,
 725                                            struct btrfs_path *path,
 726                                            u64 bytenr, u64 parent,
 727                                            u64 root_objectid, u64 owner,
 728                                            u64 offset, int refs_to_add)
 729 {
 730         struct btrfs_key key;
 731         struct extent_buffer *leaf;
 732         u32 size;
 733         u32 num_refs;
 734         int ret;
 735
 736         key.objectid = bytenr;
 737         if (parent) {
 738                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 739                 key.offset = parent;
 740                 size = sizeof(struct btrfs_shared_data_ref);
 741         } else {
 742                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 743                 key.offset = hash_extent_data_ref(root_objectid,
 744                                                   owner, offset);
 745                 size = sizeof(struct btrfs_extent_data_ref);
 746         }
 747
 748         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 749         if (ret && ret != -EEXIST)
 750                 goto fail;
 751
 752         leaf = path->nodes[0];
 753         if (parent) {
 754                 struct btrfs_shared_data_ref *ref;
 755                 ref = btrfs_item_ptr(leaf, path->slots[0],
 756                                      struct btrfs_shared_data_ref);
 757                 if (ret == 0) {
 758                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 759                 } else {
 760                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 761                         num_refs += refs_to_add;
 762                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 763                 }
 764         } else {
 765                 struct btrfs_extent_data_ref *ref;
 766                 while (ret == -EEXIST) {
 767                         ref = btrfs_item_ptr(leaf, path->slots[0],
 768                                              struct btrfs_extent_data_ref);
 769                         if (match_extent_data_ref(leaf, ref, root_objectid,
 770                                                   owner, offset))
 771                                 break;
 772                         btrfs_release_path(path);
 773
 774                         key.offset++;
 775                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 776                                                       size);
 777                         if (ret && ret != -EEXIST)
 778                                 goto fail;
 779
 780                         leaf = path->nodes[0];
 781                 }
 782                 ref = btrfs_item_ptr(leaf, path->slots[0],
 783                                      struct btrfs_extent_data_ref);
 784                 if (ret == 0) {
 785                         btrfs_set_extent_data_ref_root(leaf, ref,
 786                                                        root_objectid);
 787                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 788                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 789                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 790                 } else {
 791                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 792                         num_refs += refs_to_add;
 793                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 794                 }
 795         }
 796         btrfs_mark_buffer_dirty(leaf);
 797         ret = 0;
 798 fail:
 799         btrfs_release_path(path);
 800         return ret;
 801 }
 802
 803 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 804                                            struct btrfs_root *root,
 805                                            struct btrfs_path *path,
 806                                            int refs_to_drop)
 807 {
 808         struct btrfs_key key;
 809         struct btrfs_extent_data_ref *ref1 = NULL;
 810         struct btrfs_shared_data_ref *ref2 = NULL;
 811         struct extent_buffer *leaf;
 812         u32 num_refs = 0;
 813         int ret = 0;
 814
 815         leaf = path->nodes[0];
 816         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 817
 818         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 819                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 820                                       struct btrfs_extent_data_ref);
 821                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 822         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 823                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 824                                       struct btrfs_shared_data_ref);
 825                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 826 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 827         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 828                 struct btrfs_extent_ref_v0 *ref0;
 829                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 830                                       struct btrfs_extent_ref_v0);
 831                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 832 #endif
 833         } else {
 834                 BUG();
 835         }
 836
 837         BUG_ON(num_refs < refs_to_drop);
 838         num_refs -= refs_to_drop;
 839
 840         if (num_refs == 0) {
 841                 ret = btrfs_del_item(trans, root, path);
 842         } else {
 843                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 844                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 845                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 846                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 847 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 848                 else {
 849                         struct btrfs_extent_ref_v0 *ref0;
 850                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 851                                         struct btrfs_extent_ref_v0);
 852                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
 853                 }
 854 #endif
 855                 btrfs_mark_buffer_dirty(leaf);
 856         }
 857         return ret;
 858 }
 859
 860 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
 861                                           struct btrfs_path *path,
 862                                           struct btrfs_extent_inline_ref *iref)
 863 {
 864         struct btrfs_key key;
 865         struct extent_buffer *leaf;
 866         struct btrfs_extent_data_ref *ref1;
 867         struct btrfs_shared_data_ref *ref2;
 868         u32 num_refs = 0;
 869
 870         leaf = path->nodes[0];
 871         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 872         if (iref) {
 873                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
 874                     BTRFS_EXTENT_DATA_REF_KEY) {
 875                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 876                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 877                 } else {
 878                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 879                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 880                 }
 881         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 882                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 883                                       struct btrfs_extent_data_ref);
 884                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 885         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 886                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 887                                       struct btrfs_shared_data_ref);
 888                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 889 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 890         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 891                 struct btrfs_extent_ref_v0 *ref0;
 892                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 893                                       struct btrfs_extent_ref_v0);
 894                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 895 #endif
 896         } else {
 897                 BUG();
 898         }
 899         return num_refs;
 900 }
 901
 902 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 903                                           struct btrfs_root *root,
 904                                           struct btrfs_path *path,
 905                                           u64 bytenr, u64 parent,
 906                                           u64 root_objectid)
 907 {
 908         struct btrfs_key key;
 909         int ret;
 910
 911         key.objectid = bytenr;
 912         if (parent) {
 913                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 914                 key.offset = parent;
 915         } else {
 916                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 917                 key.offset = root_objectid;
 918         }
 919
 920         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 921         if (ret > 0)
 922                 ret = -ENOENT;
 923 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 924         if (ret == -ENOENT && parent) {
 925                 btrfs_release_path(path);
 926                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 927                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 928                 if (ret > 0)
 929                         ret = -ENOENT;
 930         }
 931 #endif
 932         return ret;
 933 }
 934
 935 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 936                                           struct btrfs_root *root,
 937                                           struct btrfs_path *path,
 938                                           u64 bytenr, u64 parent,
 939                                           u64 root_objectid)
 940 {
 941         struct btrfs_key key;
 942         int ret;
 943
 944         key.objectid = bytenr;
 945         if (parent) {
 946                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 947                 key.offset = parent;
 948         } else {
 949                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 950                 key.offset = root_objectid;
 951         }
 952
 953         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 954
 955         btrfs_release_path(path);
 956         return ret;
 957 }
 958
 959 static inline int extent_ref_type(u64 parent, u64 owner)
 960 {
 961         int type;
 962         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 963                 if (parent > 0)
 964                         type = BTRFS_SHARED_BLOCK_REF_KEY;
 965                 else
 966                         type = BTRFS_TREE_BLOCK_REF_KEY;
 967         } else {
 968                 if (parent > 0)
 969                         type = BTRFS_SHARED_DATA_REF_KEY;
 970                 else
 971                         type = BTRFS_EXTENT_DATA_REF_KEY;
 972         }
 973         return type;
 974 }
 975
 976 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 977                                  struct btrfs_root *root,
 978                                  struct btrfs_path *path,
 979                                  struct btrfs_extent_inline_ref **ref_ret,
 980                                  u64 bytenr, u64 num_bytes,
 981                                  u64 parent, u64 root_objectid,
 982                                  u64 owner, u64 offset, int insert)
 983 {
 984         struct btrfs_key key;
 985         struct extent_buffer *leaf;
 986         struct btrfs_extent_item *ei;
 987         struct btrfs_extent_inline_ref *iref;
 988         u64 flags;
 989         u32 item_size;
 990         unsigned long ptr;
 991         unsigned long end;
 992         int extra_size;
 993         int type;
 994         int want;
 995         int ret;
 996         int err = 0;
 997         int skinny_metadata =
 998                 btrfs_fs_incompat(root->fs_info,
 999                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1000
1001         key.objectid = bytenr;
1002         key.type = BTRFS_EXTENT_ITEM_KEY;
1003         key.offset = num_bytes;
1004
1005         want = extent_ref_type(parent, owner);
1006         if (insert)
1007                 extra_size = btrfs_extent_inline_ref_size(want);
1008         else
1009                 extra_size = -1;
1010
1011         if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1012                 skinny_metadata = 1;
1013                 key.type = BTRFS_METADATA_ITEM_KEY;
1014                 key.offset = owner;
1015         } else if (skinny_metadata) {
1016                 skinny_metadata = 0;
1017         }
1018
1019 again:
1020         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1021         if (ret < 0) {
1022                 err = ret;
1023                 goto out;
1024         }
1025
1026         /*
1027          * We may be a newly converted file system which still has the old fat
1028          * extent entries for metadata, so try and see if we have one of those.
1029          */
1030         if (ret > 0 && skinny_metadata) {
1031                 skinny_metadata = 0;
1032                 if (path->slots[0]) {
1033                         path->slots[0]--;
1034                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1035                                               path->slots[0]);
1036                         if (key.objectid == bytenr &&
1037                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1038                             key.offset == num_bytes)
1039                                 ret = 0;
1040                 }
1041                 if (ret) {
1042                         key.type = BTRFS_EXTENT_ITEM_KEY;
1043                         key.offset = num_bytes;
1044                         btrfs_release_path(path);
1045                         goto again;
1046                 }
1047         }
1048
1049         if (ret) {
1050                 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1051                 return -ENOENT;
1052         }
1053
1054         BUG_ON(ret);
1055
1056         leaf = path->nodes[0];
1057         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1058 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1059         if (item_size < sizeof(*ei)) {
1060                 if (!insert) {
1061                         err = -ENOENT;
1062                         goto out;
1063                 }
1064                 ret = convert_extent_item_v0(trans, root, path, owner,
1065                                              extra_size);
1066                 if (ret < 0) {
1067                         err = ret;
1068                         goto out;
1069                 }
1070                 leaf = path->nodes[0];
1071                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1072         }
1073 #endif
1074         if (item_size < sizeof(*ei)) {
1075                 printf("Size is %u, needs to be %u, slot %d\n",
1076                        (unsigned)item_size,
1077                        (unsigned)sizeof(*ei), path->slots[0]);
1078                 btrfs_print_leaf(root, leaf);
1079                 return -EINVAL;
1080         }
1081         BUG_ON(item_size < sizeof(*ei));
1082
1083         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1084         flags = btrfs_extent_flags(leaf, ei);
1085
1086         ptr = (unsigned long)(ei + 1);
1087         end = (unsigned long)ei + item_size;
1088
1089         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1090                 ptr += sizeof(struct btrfs_tree_block_info);
1091                 BUG_ON(ptr > end);
1092         } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1093                 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1094                         return -EIO;
1095                 }
1096         }
1097
1098         err = -ENOENT;
1099         while (1) {
1100                 if (ptr >= end) {
1101                         WARN_ON(ptr > end);
1102                         break;
1103                 }
1104                 iref = (struct btrfs_extent_inline_ref *)ptr;
1105                 type = btrfs_extent_inline_ref_type(leaf, iref);
1106                 if (want < type)
1107                         break;
1108                 if (want > type) {
1109                         ptr += btrfs_extent_inline_ref_size(type);
1110                         continue;
1111                 }
1112
1113                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1114                         struct btrfs_extent_data_ref *dref;
1115                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1116                         if (match_extent_data_ref(leaf, dref, root_objectid,
1117                                                   owner, offset)) {
1118                                 err = 0;
1119                                 break;
1120                         }
1121                         if (hash_extent_data_ref_item(leaf, dref) <
1122                             hash_extent_data_ref(root_objectid, owner, offset))
1123                                 break;
1124                 } else {
1125                         u64 ref_offset;
1126                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1127                         if (parent > 0) {
1128                                 if (parent == ref_offset) {
1129                                         err = 0;
1130                                         break;
1131                                 }
1132                                 if (ref_offset < parent)
1133                                         break;
1134                         } else {
1135                                 if (root_objectid == ref_offset) {
1136                                         err = 0;
1137                                         break;
1138                                 }
1139                                 if (ref_offset < root_objectid)
1140                                         break;
1141                         }
1142                 }
1143                 ptr += btrfs_extent_inline_ref_size(type);
1144         }
1145         if (err == -ENOENT && insert) {
1146                 if (item_size + extra_size >=
1147                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1148                         err = -EAGAIN;
1149                         goto out;
1150                 }
1151                 /*
1152                  * To add new inline back ref, we have to make sure
1153                  * there is no corresponding back ref item.
1154                  * For simplicity, we just do not add new inline back
1155                  * ref if there is any back ref item.
1156                  */
1157                 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1158                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1159                         err = -EAGAIN;
1160                         goto out;
1161                 }
1162         }
1163         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1164 out:
1165         return err;
1166 }
1167
1168 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1169                                 struct btrfs_root *root,
1170                                 struct btrfs_path *path,
1171                                 struct btrfs_extent_inline_ref *iref,
1172                                 u64 parent, u64 root_objectid,
1173                                 u64 owner, u64 offset, int refs_to_add)
1174 {
1175         struct extent_buffer *leaf;
1176         struct btrfs_extent_item *ei;
1177         unsigned long ptr;
1178         unsigned long end;
1179         unsigned long item_offset;
1180         u64 refs;
1181         int size;
1182         int type;
1183         int ret;
1184
1185         leaf = path->nodes[0];
1186         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1187         item_offset = (unsigned long)iref - (unsigned long)ei;
1188
1189         type = extent_ref_type(parent, owner);
1190         size = btrfs_extent_inline_ref_size(type);
1191
1192         ret = btrfs_extend_item(trans, root, path, size);
1193         BUG_ON(ret);
1194
1195         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1196         refs = btrfs_extent_refs(leaf, ei);
1197         refs += refs_to_add;
1198         btrfs_set_extent_refs(leaf, ei, refs);
1199
1200         ptr = (unsigned long)ei + item_offset;
1201         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1202         if (ptr < end - size)
1203                 memmove_extent_buffer(leaf, ptr + size, ptr,
1204                                       end - size - ptr);
1205
1206         iref = (struct btrfs_extent_inline_ref *)ptr;
1207         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1208         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1209                 struct btrfs_extent_data_ref *dref;
1210                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1211                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1212                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1213                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1214                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1215         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1216                 struct btrfs_shared_data_ref *sref;
1217                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1218                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1219                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1220         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1221                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1222         } else {
1223                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1224         }
1225         btrfs_mark_buffer_dirty(leaf);
1226         return 0;
1227 }
1228
1229 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1230                                  struct btrfs_root *root,
1231                                  struct btrfs_path *path,
1232                                  struct btrfs_extent_inline_ref **ref_ret,
1233                                  u64 bytenr, u64 num_bytes, u64 parent,
1234                                  u64 root_objectid, u64 owner, u64 offset)
1235 {
1236         int ret;
1237
1238         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1239                                            bytenr, num_bytes, parent,
1240                                            root_objectid, owner, offset, 0);
1241         if (ret != -ENOENT)
1242                 return ret;
1243
1244         btrfs_release_path(path);
1245         *ref_ret = NULL;
1246
1247         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1248                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1249                                             root_objectid);
1250         } else {
1251                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1252                                              root_objectid, owner, offset);
1253         }
1254         return ret;
1255 }
1256
1257 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1258                                  struct btrfs_root *root,
1259                                  struct btrfs_path *path,
1260                                  struct btrfs_extent_inline_ref *iref,
1261                                  int refs_to_mod)
1262 {
1263         struct extent_buffer *leaf;
1264         struct btrfs_extent_item *ei;
1265         struct btrfs_extent_data_ref *dref = NULL;
1266         struct btrfs_shared_data_ref *sref = NULL;
1267         unsigned long ptr;
1268         unsigned long end;
1269         u32 item_size;
1270         int size;
1271         int type;
1272         int ret;
1273         u64 refs;
1274
1275         leaf = path->nodes[0];
1276         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1277         refs = btrfs_extent_refs(leaf, ei);
1278         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1279         refs += refs_to_mod;
1280         btrfs_set_extent_refs(leaf, ei, refs);
1281
1282         type = btrfs_extent_inline_ref_type(leaf, iref);
1283
1284         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1285                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1286                 refs = btrfs_extent_data_ref_count(leaf, dref);
1287         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1288                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1289                 refs = btrfs_shared_data_ref_count(leaf, sref);
1290         } else {
1291                 refs = 1;
1292                 BUG_ON(refs_to_mod != -1);
1293         }
1294
1295         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1296         refs += refs_to_mod;
1297
1298         if (refs > 0) {
1299                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1300                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1301                 else
1302                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1303         } else {
1304                 size =  btrfs_extent_inline_ref_size(type);
1305                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1306                 ptr = (unsigned long)iref;
1307                 end = (unsigned long)ei + item_size;
1308                 if (ptr + size < end)
1309                         memmove_extent_buffer(leaf, ptr, ptr + size,
1310                                               end - ptr - size);
1311                 item_size -= size;
1312                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1313                 BUG_ON(ret);
1314         }
1315         btrfs_mark_buffer_dirty(leaf);
1316         return 0;
1317 }
1318
1319 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1320                                  struct btrfs_root *root,
1321                                  struct btrfs_path *path,
1322                                  u64 bytenr, u64 num_bytes, u64 parent,
1323                                  u64 root_objectid, u64 owner,
1324                                  u64 offset, int refs_to_add)
1325 {
1326         struct btrfs_extent_inline_ref *iref;
1327         int ret;
1328
1329         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1330                                            bytenr, num_bytes, parent,
1331                                            root_objectid, owner, offset, 1);
1332         if (ret == 0) {
1333                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1334                 ret = update_inline_extent_backref(trans, root, path, iref,
1335                                                    refs_to_add);
1336         } else if (ret == -ENOENT) {
1337                 ret = setup_inline_extent_backref(trans, root, path, iref,
1338                                                   parent, root_objectid,
1339                                                   owner, offset, refs_to_add);
1340         }
1341         return ret;
1342 }
1343
1344 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1345                                  struct btrfs_root *root,
1346                                  struct btrfs_path *path,
1347                                  u64 bytenr, u64 parent, u64 root_objectid,
1348                                  u64 owner, u64 offset, int refs_to_add)
1349 {
1350         int ret;
1351
1352         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1353                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1354                                              parent, root_objectid,
1355                                              owner, offset, refs_to_add);
1356         } else {
1357                 BUG_ON(refs_to_add != 1);
1358                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1359                                             parent, root_objectid);
1360         }
1361         return ret;
1362 }
1363
1364 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref *iref,
1368                                  int refs_to_drop, int is_data)
1369 {
1370         int ret;
1371
1372         BUG_ON(!is_data && refs_to_drop != 1);
1373         if (iref) {
1374                 ret = update_inline_extent_backref(trans, root, path, iref,
1375                                                    -refs_to_drop);
1376         } else if (is_data) {
1377                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1378         } else {
1379                 ret = btrfs_del_item(trans, root, path);
1380         }
1381         return ret;
1382 }
1383
1384 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1385                          struct btrfs_root *root,
1386                          u64 bytenr, u64 num_bytes, u64 parent,
1387                          u64 root_objectid, u64 owner, u64 offset)
1388 {
1389         struct btrfs_path *path;
1390         struct extent_buffer *leaf;
1391         struct btrfs_extent_item *item;
1392         u64 refs;
1393         int ret;
1394         int err = 0;
1395
1396         path = btrfs_alloc_path();
1397         if (!path)
1398                 return -ENOMEM;
1399
1400         path->reada = 1;
1401
1402         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1403                                            path, bytenr, num_bytes, parent,
1404                                            root_objectid, owner, offset, 1);
1405         if (ret == 0)
1406                 goto out;
1407
1408         if (ret != -EAGAIN) {
1409                 err = ret;
1410                 goto out;
1411         }
1412
1413         leaf = path->nodes[0];
1414         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1415         refs = btrfs_extent_refs(leaf, item);
1416         btrfs_set_extent_refs(leaf, item, refs + 1);
1417
1418         btrfs_mark_buffer_dirty(leaf);
1419         btrfs_release_path(path);
1420
1421         path->reada = 1;
1422
1423         /* now insert the actual backref */
1424         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1425                                     path, bytenr, parent, root_objectid,
1426                                     owner, offset, 1);
1427         if (ret)
1428                 err = ret;
1429 out:
1430         btrfs_free_path(path);
1431         finish_current_insert(trans, root->fs_info->extent_root);
1432         del_pending_extents(trans, root->fs_info->extent_root);
1433         BUG_ON(err);
1434         return err;
1435 }
1436
1437 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1438                          struct btrfs_root *root)
1439 {
1440         finish_current_insert(trans, root->fs_info->extent_root);
1441         del_pending_extents(trans, root->fs_info->extent_root);
1442         return 0;
1443 }
1444
1445 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1446                              struct btrfs_root *root, u64 bytenr,
1447                              u64 offset, int metadata, u64 *refs, u64 *flags)
1448 {
1449         struct btrfs_path *path;
1450         int ret;
1451         struct btrfs_key key;
1452         struct extent_buffer *l;
1453         struct btrfs_extent_item *item;
1454         u32 item_size;
1455         u64 num_refs;
1456         u64 extent_flags;
1457
1458         if (metadata &&
1459             !btrfs_fs_incompat(root->fs_info,
1460                                BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
1461                 offset = root->leafsize;
1462                 metadata = 0;
1463         }
1464
1465         path = btrfs_alloc_path();
1466         if (!path)
1467                 return -ENOMEM;
1468         path->reada = 1;
1469
1470         key.objectid = bytenr;
1471         key.offset = offset;
1472         if (metadata)
1473                 key.type = BTRFS_METADATA_ITEM_KEY;
1474         else
1475                 key.type = BTRFS_EXTENT_ITEM_KEY;
1476
1477 again:
1478         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1479                                 0, 0);
1480         if (ret < 0)
1481                 goto out;
1482
1483         /*
1484          * Deal with the fact that we may have mixed SKINNY and normal refs.  If
1485          * we didn't find what we wanted check and see if we have a normal ref
1486          * right next to us, or re-search if we are on the edge of the leaf just
1487          * to make sure.
1488          */
1489         if (ret > 0 && metadata) {
1490                 if (path->slots[0]) {
1491                         path->slots[0]--;
1492                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1493                                               path->slots[0]);
1494                         if (key.objectid == bytenr &&
1495                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1496                             key.offset == root->leafsize)
1497                                 ret = 0;
1498                 }
1499
1500                 if (ret) {
1501                         btrfs_release_path(path);
1502                         key.type = BTRFS_EXTENT_ITEM_KEY;
1503                         key.offset = root->leafsize;
1504                         metadata = 0;
1505                         goto again;
1506                 }
1507         }
1508
1509         if (ret != 0) {
1510                 ret = -EIO;
1511                 goto out;
1512         }
1513
1514         l = path->nodes[0];
1515         item_size = btrfs_item_size_nr(l, path->slots[0]);
1516         if (item_size >= sizeof(*item)) {
1517                 item = btrfs_item_ptr(l, path->slots[0],
1518                                       struct btrfs_extent_item);
1519                 num_refs = btrfs_extent_refs(l, item);
1520                 extent_flags = btrfs_extent_flags(l, item);
1521         } else {
1522 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1523                         struct btrfs_extent_item_v0 *ei0;
1524                         BUG_ON(item_size != sizeof(*ei0));
1525                         ei0 = btrfs_item_ptr(l, path->slots[0],
1526                                              struct btrfs_extent_item_v0);
1527                         num_refs = btrfs_extent_refs_v0(l, ei0);
1528                         /* FIXME: this isn't correct for data */
1529                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1530 #else
1531                         BUG();
1532 #endif
1533         }
1534         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1535         if (refs)
1536                 *refs = num_refs;
1537         if (flags)
1538                 *flags = extent_flags;
1539 out:
1540         btrfs_free_path(path);
1541         return ret;
1542 }
1543
1544 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1545                           struct btrfs_root *root,
1546                           u64 bytenr, int level, u64 flags)
1547 {
1548         struct btrfs_path *path;
1549         int ret;
1550         struct btrfs_key key;
1551         struct extent_buffer *l;
1552         struct btrfs_extent_item *item;
1553         u32 item_size;
1554         int skinny_metadata =
1555                 btrfs_fs_incompat(root->fs_info,
1556                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1557
1558         path = btrfs_alloc_path();
1559         if (!path)
1560                 return -ENOMEM;
1561         path->reada = 1;
1562
1563         key.objectid = bytenr;
1564         if (skinny_metadata) {
1565                 key.offset = level;
1566                 key.type = BTRFS_METADATA_ITEM_KEY;
1567         } else {
1568                 key.offset = root->leafsize;
1569                 key.type = BTRFS_EXTENT_ITEM_KEY;
1570         }
1571
1572 again:
1573         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1574                                 0, 0);
1575         if (ret < 0)
1576                 goto out;
1577
1578         if (ret > 0 && skinny_metadata) {
1579                 skinny_metadata = 0;
1580                 if (path->slots[0]) {
1581                         path->slots[0]--;
1582                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1583                                               path->slots[0]);
1584                         if (key.objectid == bytenr &&
1585                             key.offset == root->leafsize &&
1586                             key.type == BTRFS_EXTENT_ITEM_KEY)
1587                                 ret = 0;
1588                 }
1589                 if (ret) {
1590                         btrfs_release_path(path);
1591                         key.offset = root->leafsize;
1592                         key.type = BTRFS_EXTENT_ITEM_KEY;
1593                         goto again;
1594                 }
1595         }
1596
1597         if (ret != 0) {
1598                 btrfs_print_leaf(root, path->nodes[0]);
1599                 printk("failed to find block number %Lu\n",
1600                         (unsigned long long)bytenr);
1601                 BUG();
1602         }
1603         l = path->nodes[0];
1604         item_size = btrfs_item_size_nr(l, path->slots[0]);
1605 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1606         if (item_size < sizeof(*item)) {
1607                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1608                                              path, (u64)-1, 0);
1609                 if (ret < 0)
1610                         goto out;
1611
1612                 l = path->nodes[0];
1613                 item_size = btrfs_item_size_nr(l, path->slots[0]);
1614         }
1615 #endif
1616         BUG_ON(item_size < sizeof(*item));
1617         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1618         flags |= btrfs_extent_flags(l, item);
1619         btrfs_set_extent_flags(l, item, flags);
1620 out:
1621         btrfs_free_path(path);
1622         finish_current_insert(trans, root->fs_info->extent_root);
1623         del_pending_extents(trans, root->fs_info->extent_root);
1624         return ret;
1625 }
1626
1627 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1628                            struct btrfs_root *root,
1629                            struct extent_buffer *buf,
1630                            int record_parent, int inc)
1631 {
1632         u64 bytenr;
1633         u64 num_bytes;
1634         u64 parent;
1635         u64 ref_root;
1636         u32 nritems;
1637         struct btrfs_key key;
1638         struct btrfs_file_extent_item *fi;
1639         int i;
1640         int level;
1641         int ret = 0;
1642         int (*process_func)(struct btrfs_trans_handle *trans,
1643                             struct btrfs_root *root,
1644                             u64, u64, u64, u64, u64, u64);
1645
1646         ref_root = btrfs_header_owner(buf);
1647         nritems = btrfs_header_nritems(buf);
1648         level = btrfs_header_level(buf);
1649
1650         if (!root->ref_cows && level == 0)
1651                 return 0;
1652
1653         if (inc)
1654                 process_func = btrfs_inc_extent_ref;
1655         else
1656                 process_func = btrfs_free_extent;
1657
1658         if (record_parent)
1659                 parent = buf->start;
1660         else
1661                 parent = 0;
1662
1663         for (i = 0; i < nritems; i++) {
1664                 cond_resched();
1665                 if (level == 0) {
1666                         btrfs_item_key_to_cpu(buf, &key, i);
1667                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1668                                 continue;
1669                         fi = btrfs_item_ptr(buf, i,
1670                                             struct btrfs_file_extent_item);
1671                         if (btrfs_file_extent_type(buf, fi) ==
1672                             BTRFS_FILE_EXTENT_INLINE)
1673                                 continue;
1674                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1675                         if (bytenr == 0)
1676                                 continue;
1677
1678                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1679                         key.offset -= btrfs_file_extent_offset(buf, fi);
1680                         ret = process_func(trans, root, bytenr, num_bytes,
1681                                            parent, ref_root, key.objectid,
1682                                            key.offset);
1683                         if (ret) {
1684                                 WARN_ON(1);
1685                                 goto fail;
1686                         }
1687                 } else {
1688                         bytenr = btrfs_node_blockptr(buf, i);
1689                         num_bytes = btrfs_level_size(root, level - 1);
1690                         ret = process_func(trans, root, bytenr, num_bytes,
1691                                            parent, ref_root, level - 1, 0);
1692                         if (ret) {
1693                                 WARN_ON(1);
1694                                 goto fail;
1695                         }
1696                 }
1697         }
1698         return 0;
1699 fail:
1700         WARN_ON(1);
1701         return ret;
1702 }
1703
1704 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1705                   struct extent_buffer *buf, int record_parent)
1706 {
1707         return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1708 }
1709
1710 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1711                   struct extent_buffer *buf, int record_parent)
1712 {
1713         return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1714 }
1715
1716 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1717                                  struct btrfs_root *root,
1718                                  struct btrfs_path *path,
1719                                  struct btrfs_block_group_cache *cache)
1720 {
1721         int ret;
1722         int pending_ret;
1723         struct btrfs_root *extent_root = root->fs_info->extent_root;
1724         unsigned long bi;
1725         struct extent_buffer *leaf;
1726
1727         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1728         if (ret < 0)
1729                 goto fail;
1730         BUG_ON(ret);
1731
1732         leaf = path->nodes[0];
1733         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1734         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1735         btrfs_mark_buffer_dirty(leaf);
1736         btrfs_release_path(path);
1737 fail:
1738         finish_current_insert(trans, extent_root);
1739         pending_ret = del_pending_extents(trans, extent_root);
1740         if (ret)
1741                 return ret;
1742         if (pending_ret)
1743                 return pending_ret;
1744         return 0;
1745
1746 }
1747
1748 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1749                                    struct btrfs_root *root)
1750 {
1751         struct extent_io_tree *block_group_cache;
1752         struct btrfs_block_group_cache *cache;
1753         int ret;
1754         struct btrfs_path *path;
1755         u64 last = 0;
1756         u64 start;
1757         u64 end;
1758         u64 ptr;
1759
1760         block_group_cache = &root->fs_info->block_group_cache;
1761         path = btrfs_alloc_path();
1762         if (!path)
1763                 return -ENOMEM;
1764
1765         while(1) {
1766                 ret = find_first_extent_bit(block_group_cache, last,
1767                                             &start, &end, BLOCK_GROUP_DIRTY);
1768                 if (ret) {
1769                         if (last == 0)
1770                                 break;
1771                         last = 0;
1772                         continue;
1773                 }
1774
1775                 last = end + 1;
1776                 ret = get_state_private(block_group_cache, start, &ptr);
1777                 BUG_ON(ret);
1778
1779                 clear_extent_bits(block_group_cache, start, end,
1780                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1781
1782                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1783                 ret = write_one_cache_group(trans, root, path, cache);
1784         }
1785         btrfs_free_path(path);
1786         return 0;
1787 }
1788
1789 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1790                                                   u64 flags)
1791 {
1792         struct list_head *head = &info->space_info;
1793         struct list_head *cur;
1794         struct btrfs_space_info *found;
1795         list_for_each(cur, head) {
1796                 found = list_entry(cur, struct btrfs_space_info, list);
1797                 if (found->flags & flags)
1798                         return found;
1799         }
1800         return NULL;
1801
1802 }
1803
1804 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1805                              u64 total_bytes, u64 bytes_used,
1806                              struct btrfs_space_info **space_info)
1807 {
1808         struct btrfs_space_info *found;
1809
1810         found = __find_space_info(info, flags);
1811         if (found) {
1812                 found->total_bytes += total_bytes;
1813                 found->bytes_used += bytes_used;
1814                 if (found->total_bytes < found->bytes_used) {
1815                         fprintf(stderr, "warning, bad space info total_bytes "
1816                                 "%llu used %llu\n",
1817                                (unsigned long long)found->total_bytes,
1818                                (unsigned long long)found->bytes_used);
1819                 }
1820                 *space_info = found;
1821                 return 0;
1822         }
1823         found = kmalloc(sizeof(*found), GFP_NOFS);
1824         if (!found)
1825                 return -ENOMEM;
1826
1827         list_add(&found->list, &info->space_info);
1828         found->flags = flags;
1829         found->total_bytes = total_bytes;
1830         found->bytes_used = bytes_used;
1831         found->bytes_pinned = 0;
1832         found->full = 0;
1833         *space_info = found;
1834         return 0;
1835 }
1836
1837
1838 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1839 {
1840         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1841                                    BTRFS_BLOCK_GROUP_RAID1 |
1842                                    BTRFS_BLOCK_GROUP_RAID10 |
1843                                    BTRFS_BLOCK_GROUP_RAID5 |
1844                                    BTRFS_BLOCK_GROUP_RAID6 |
1845                                    BTRFS_BLOCK_GROUP_DUP);
1846         if (extra_flags) {
1847                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1848                         fs_info->avail_data_alloc_bits |= extra_flags;
1849                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1850                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1851                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1852                         fs_info->avail_system_alloc_bits |= extra_flags;
1853         }
1854 }
1855
1856 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1857                           struct btrfs_root *extent_root, u64 alloc_bytes,
1858                           u64 flags)
1859 {
1860         struct btrfs_space_info *space_info;
1861         u64 thresh;
1862         u64 start;
1863         u64 num_bytes;
1864         int ret;
1865
1866         space_info = __find_space_info(extent_root->fs_info, flags);
1867         if (!space_info) {
1868                 ret = update_space_info(extent_root->fs_info, flags,
1869                                         0, 0, &space_info);
1870                 BUG_ON(ret);
1871         }
1872         BUG_ON(!space_info);
1873
1874         if (space_info->full)
1875                 return 0;
1876
1877         thresh = div_factor(space_info->total_bytes, 7);
1878         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1879             thresh)
1880                 return 0;
1881
1882         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1883                                 space_info->flags);
1884         if (ret == -ENOSPC) {
1885                 space_info->full = 1;
1886                 return 0;
1887         }
1888
1889         BUG_ON(ret);
1890
1891         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1892                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1893         BUG_ON(ret);
1894         return 0;
1895 }
1896
1897 static int update_block_group(struct btrfs_trans_handle *trans,
1898                               struct btrfs_root *root,
1899                               u64 bytenr, u64 num_bytes, int alloc,
1900                               int mark_free)
1901 {
1902         struct btrfs_block_group_cache *cache;
1903         struct btrfs_fs_info *info = root->fs_info;
1904         u64 total = num_bytes;
1905         u64 old_val;
1906         u64 byte_in_group;
1907         u64 start;
1908         u64 end;
1909
1910         /* block accounting for super block */
1911         old_val = btrfs_super_bytes_used(info->super_copy);
1912         if (alloc)
1913                 old_val += num_bytes;
1914         else
1915                 old_val -= num_bytes;
1916         btrfs_set_super_bytes_used(info->super_copy, old_val);
1917
1918         /* block accounting for root item */
1919         old_val = btrfs_root_used(&root->root_item);
1920         if (alloc)
1921                 old_val += num_bytes;
1922         else
1923                 old_val -= num_bytes;
1924         btrfs_set_root_used(&root->root_item, old_val);
1925
1926         while(total) {
1927                 cache = btrfs_lookup_block_group(info, bytenr);
1928                 if (!cache) {
1929                         return -1;
1930                 }
1931                 byte_in_group = bytenr - cache->key.objectid;
1932                 WARN_ON(byte_in_group > cache->key.offset);
1933                 start = cache->key.objectid;
1934                 end = start + cache->key.offset - 1;
1935                 set_extent_bits(&info->block_group_cache, start, end,
1936                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1937
1938                 old_val = btrfs_block_group_used(&cache->item);
1939                 num_bytes = min(total, cache->key.offset - byte_in_group);
1940
1941                 if (alloc) {
1942                         old_val += num_bytes;
1943                         cache->space_info->bytes_used += num_bytes;
1944                 } else {
1945                         old_val -= num_bytes;
1946                         cache->space_info->bytes_used -= num_bytes;
1947                         if (mark_free) {
1948                                 set_extent_dirty(&info->free_space_cache,
1949                                                  bytenr, bytenr + num_bytes - 1,
1950                                                  GFP_NOFS);
1951                         }
1952                 }
1953                 btrfs_set_block_group_used(&cache->item, old_val);
1954                 total -= num_bytes;
1955                 bytenr += num_bytes;
1956         }
1957         return 0;
1958 }
1959
1960 static int update_pinned_extents(struct btrfs_root *root,
1961                                 u64 bytenr, u64 num, int pin)
1962 {
1963         u64 len;
1964         struct btrfs_block_group_cache *cache;
1965         struct btrfs_fs_info *fs_info = root->fs_info;
1966
1967         if (pin) {
1968                 set_extent_dirty(&fs_info->pinned_extents,
1969                                 bytenr, bytenr + num - 1, GFP_NOFS);
1970         } else {
1971                 clear_extent_dirty(&fs_info->pinned_extents,
1972                                 bytenr, bytenr + num - 1, GFP_NOFS);
1973         }
1974         while (num > 0) {
1975                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1976                 if (!cache) {
1977                         len = min((u64)root->sectorsize, num);
1978                         goto next;
1979                 }
1980                 WARN_ON(!cache);
1981                 len = min(num, cache->key.offset -
1982                           (bytenr - cache->key.objectid));
1983                 if (pin) {
1984                         cache->pinned += len;
1985                         cache->space_info->bytes_pinned += len;
1986                         fs_info->total_pinned += len;
1987                 } else {
1988                         cache->pinned -= len;
1989                         cache->space_info->bytes_pinned -= len;
1990                         fs_info->total_pinned -= len;
1991                 }
1992 next:
1993                 bytenr += len;
1994                 num -= len;
1995         }
1996         return 0;
1997 }
1998
1999 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2000                                struct btrfs_root *root,
2001                                struct extent_io_tree *unpin)
2002 {
2003         u64 start;
2004         u64 end;
2005         int ret;
2006         struct extent_io_tree *free_space_cache;
2007         free_space_cache = &root->fs_info->free_space_cache;
2008
2009         while(1) {
2010                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2011                                             EXTENT_DIRTY);
2012                 if (ret)
2013                         break;
2014                 update_pinned_extents(root, start, end + 1 - start, 0);
2015                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2016                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2017         }
2018         return 0;
2019 }
2020
2021 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2022 {
2023         u64 start;
2024         u64 end;
2025         int ret;
2026
2027         ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2028                                     &end, EXTENT_LOCKED);
2029         if (!ret) {
2030                 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2031                                             EXTENT_LOCKED);
2032         }
2033         return ret == 0;
2034
2035 }
2036 static int finish_current_insert(struct btrfs_trans_handle *trans,
2037                                  struct btrfs_root *extent_root)
2038 {
2039         u64 start;
2040         u64 end;
2041         u64 priv;
2042         struct btrfs_fs_info *info = extent_root->fs_info;
2043         struct pending_extent_op *extent_op;
2044         struct btrfs_key key;
2045         int ret;
2046         int skinny_metadata =
2047                 btrfs_fs_incompat(extent_root->fs_info,
2048                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2049
2050         while(1) {
2051                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2052                                             &end, EXTENT_LOCKED);
2053                 if (ret)
2054                         break;
2055
2056                 ret = get_state_private(&info->extent_ins, start, &priv);
2057                 BUG_ON(ret);
2058                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2059
2060                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2061                         key.objectid = start;
2062                         if (skinny_metadata) {
2063                                 key.offset = extent_op->level;
2064                                 key.type = BTRFS_METADATA_ITEM_KEY;
2065                         } else {
2066                                 key.offset = extent_op->num_bytes;
2067                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2068                         }
2069                         ret = alloc_reserved_tree_block(trans, extent_root,
2070                                                 extent_root->root_key.objectid,
2071                                                 trans->transid,
2072                                                 extent_op->flags,
2073                                                 &extent_op->key,
2074                                                 extent_op->level, &key);
2075                         BUG_ON(ret);
2076                 } else {
2077                         BUG_ON(1);
2078                 }
2079
2080                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2081                                   GFP_NOFS);
2082                 kfree(extent_op);
2083         }
2084         return 0;
2085 }
2086
2087 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2088                           struct btrfs_root *root,
2089                           u64 bytenr, u64 num_bytes, int is_data)
2090 {
2091         int err = 0;
2092         struct extent_buffer *buf;
2093
2094         if (is_data)
2095                 goto pinit;
2096
2097         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2098         if (!buf)
2099                 goto pinit;
2100
2101         /* we can reuse a block if it hasn't been written
2102          * and it is from this transaction.  We can't
2103          * reuse anything from the tree log root because
2104          * it has tiny sub-transactions.
2105          */
2106         if (btrfs_buffer_uptodate(buf, 0)) {
2107                 u64 header_owner = btrfs_header_owner(buf);
2108                 u64 header_transid = btrfs_header_generation(buf);
2109                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2110                     header_transid == trans->transid &&
2111                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2112                         clean_tree_block(NULL, root, buf);
2113                         free_extent_buffer(buf);
2114                         return 1;
2115                 }
2116         }
2117         free_extent_buffer(buf);
2118 pinit:
2119         update_pinned_extents(root, bytenr, num_bytes, 1);
2120
2121         BUG_ON(err < 0);
2122         return 0;
2123 }
2124
2125 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2126                        u64 bytenr, u64 num_bytes)
2127 {
2128         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2129 }
2130
2131 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2132                         u64 bytenr, u64 num_bytes)
2133 {
2134         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2135 }
2136
2137 /*
2138  * remove an extent from the root, returns 0 on success
2139  */
2140 static int __free_extent(struct btrfs_trans_handle *trans,
2141                          struct btrfs_root *root,
2142                          u64 bytenr, u64 num_bytes, u64 parent,
2143                          u64 root_objectid, u64 owner_objectid,
2144                          u64 owner_offset, int refs_to_drop)
2145 {
2146
2147         struct btrfs_key key;
2148         struct btrfs_path *path;
2149         struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2150         struct btrfs_root *extent_root = root->fs_info->extent_root;
2151         struct extent_buffer *leaf;
2152         struct btrfs_extent_item *ei;
2153         struct btrfs_extent_inline_ref *iref;
2154         int ret;
2155         int is_data;
2156         int extent_slot = 0;
2157         int found_extent = 0;
2158         int num_to_del = 1;
2159         u32 item_size;
2160         u64 refs;
2161         int skinny_metadata =
2162                 btrfs_fs_incompat(extent_root->fs_info,
2163                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2164
2165         if (root->fs_info->free_extent_hook) {
2166                 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2167                                                 parent, root_objectid, owner_objectid,
2168                                                 owner_offset, refs_to_drop);
2169
2170         }
2171         path = btrfs_alloc_path();
2172         if (!path)
2173                 return -ENOMEM;
2174
2175         path->reada = 1;
2176
2177         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2178         if (is_data)
2179                 skinny_metadata = 0;
2180         BUG_ON(!is_data && refs_to_drop != 1);
2181
2182         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2183                                     bytenr, num_bytes, parent,
2184                                     root_objectid, owner_objectid,
2185                                     owner_offset);
2186         if (ret == 0) {
2187                 extent_slot = path->slots[0];
2188                 while (extent_slot >= 0) {
2189                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2190                                               extent_slot);
2191                         if (key.objectid != bytenr)
2192                                 break;
2193                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2194                             key.offset == num_bytes) {
2195                                 found_extent = 1;
2196                                 break;
2197                         }
2198                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
2199                             key.offset == owner_objectid) {
2200                                 found_extent = 1;
2201                                 break;
2202                         }
2203                         if (path->slots[0] - extent_slot > 5)
2204                                 break;
2205                         extent_slot--;
2206                 }
2207 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2208                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2209                 if (found_extent && item_size < sizeof(*ei))
2210                         found_extent = 0;
2211 #endif
2212                 if (!found_extent) {
2213                         BUG_ON(iref);
2214                         ret = remove_extent_backref(trans, extent_root, path,
2215                                                     NULL, refs_to_drop,
2216                                                     is_data);
2217                         BUG_ON(ret);
2218                         btrfs_release_path(path);
2219
2220                         key.objectid = bytenr;
2221
2222                         if (skinny_metadata) {
2223                                 key.type = BTRFS_METADATA_ITEM_KEY;
2224                                 key.offset = owner_objectid;
2225                         } else {
2226                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2227                                 key.offset = num_bytes;
2228                         }
2229
2230                         ret = btrfs_search_slot(trans, extent_root,
2231                                                 &key, path, -1, 1);
2232                         if (ret > 0 && skinny_metadata && path->slots[0]) {
2233                                 path->slots[0]--;
2234                                 btrfs_item_key_to_cpu(path->nodes[0],
2235                                                       &key,
2236                                                       path->slots[0]);
2237                                 if (key.objectid == bytenr &&
2238                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2239                                     key.offset == num_bytes)
2240                                         ret = 0;
2241                         }
2242
2243                         if (ret > 0 && skinny_metadata) {
2244                                 skinny_metadata = 0;
2245                                 btrfs_release_path(path);
2246                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2247                                 key.offset = num_bytes;
2248                                 ret = btrfs_search_slot(trans, extent_root,
2249                                                         &key, path, -1, 1);
2250                         }
2251
2252                         if (ret) {
2253                                 printk(KERN_ERR "umm, got %d back from search"
2254                                        ", was looking for %llu\n", ret,
2255                                        (unsigned long long)bytenr);
2256                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2257                         }
2258                         BUG_ON(ret);
2259                         extent_slot = path->slots[0];
2260                 }
2261         } else {
2262                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2263                        "parent %llu root %llu  owner %llu offset %llu\n",
2264                        (unsigned long long)bytenr,
2265                        (unsigned long long)parent,
2266                        (unsigned long long)root_objectid,
2267                        (unsigned long long)owner_objectid,
2268                        (unsigned long long)owner_offset);
2269                 ret = -EIO;
2270                 goto fail;
2271         }
2272
2273         leaf = path->nodes[0];
2274         item_size = btrfs_item_size_nr(leaf, extent_slot);
2275 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2276         if (item_size < sizeof(*ei)) {
2277                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2278                 ret = convert_extent_item_v0(trans, extent_root, path,
2279                                              owner_objectid, 0);
2280                 BUG_ON(ret < 0);
2281
2282                 btrfs_release_path(path);
2283
2284                 key.objectid = bytenr;
2285                 key.type = BTRFS_EXTENT_ITEM_KEY;
2286                 key.offset = num_bytes;
2287
2288                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2289                                         -1, 1);
2290                 if (ret) {
2291                         printk(KERN_ERR "umm, got %d back from search"
2292                                ", was looking for %llu\n", ret,
2293                                (unsigned long long)bytenr);
2294                         btrfs_print_leaf(extent_root, path->nodes[0]);
2295                 }
2296                 BUG_ON(ret);
2297                 extent_slot = path->slots[0];
2298                 leaf = path->nodes[0];
2299                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2300         }
2301 #endif
2302         BUG_ON(item_size < sizeof(*ei));
2303         ei = btrfs_item_ptr(leaf, extent_slot,
2304                             struct btrfs_extent_item);
2305         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2306             key.type == BTRFS_EXTENT_ITEM_KEY) {
2307                 struct btrfs_tree_block_info *bi;
2308                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2309                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2310                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2311         }
2312
2313         refs = btrfs_extent_refs(leaf, ei);
2314         BUG_ON(refs < refs_to_drop);
2315         refs -= refs_to_drop;
2316
2317         if (refs > 0) {
2318                 /*
2319                  * In the case of inline back ref, reference count will
2320                  * be updated by remove_extent_backref
2321                  */
2322                 if (iref) {
2323                         BUG_ON(!found_extent);
2324                 } else {
2325                         btrfs_set_extent_refs(leaf, ei, refs);
2326                         btrfs_mark_buffer_dirty(leaf);
2327                 }
2328                 if (found_extent) {
2329                         ret = remove_extent_backref(trans, extent_root, path,
2330                                                     iref, refs_to_drop,
2331                                                     is_data);
2332                         BUG_ON(ret);
2333                 }
2334         } else {
2335                 int mark_free = 0;
2336                 int pin = 1;
2337
2338                 if (found_extent) {
2339                         BUG_ON(is_data && refs_to_drop !=
2340                                extent_data_ref_count(root, path, iref));
2341                         if (iref) {
2342                                 BUG_ON(path->slots[0] != extent_slot);
2343                         } else {
2344                                 BUG_ON(path->slots[0] != extent_slot + 1);
2345                                 path->slots[0] = extent_slot;
2346                                 num_to_del = 2;
2347                         }
2348                 }
2349
2350                 if (ops && ops->free_extent) {
2351                         ret = ops->free_extent(root, bytenr, num_bytes);
2352                         if (ret > 0) {
2353                                 pin = 0;
2354                                 mark_free = 0;
2355                         }
2356                 }
2357
2358                 if (pin) {
2359                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2360                                              is_data);
2361                         if (ret > 0)
2362                                 mark_free = 1;
2363                         BUG_ON(ret < 0);
2364                 }
2365
2366                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2367                                       num_to_del);
2368                 BUG_ON(ret);
2369                 btrfs_release_path(path);
2370
2371                 if (is_data) {
2372                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2373                         BUG_ON(ret);
2374                 }
2375
2376                 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2377         }
2378 fail:
2379         btrfs_free_path(path);
2380         finish_current_insert(trans, extent_root);
2381         return ret;
2382 }
2383
2384 /*
2385  * find all the blocks marked as pending in the radix tree and remove
2386  * them from the extent map
2387  */
2388 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2389                                btrfs_root *extent_root)
2390 {
2391         int ret;
2392         int err = 0;
2393         u64 start;
2394         u64 end;
2395         u64 priv;
2396         struct extent_io_tree *pending_del;
2397         struct extent_io_tree *extent_ins;
2398         struct pending_extent_op *extent_op;
2399
2400         extent_ins = &extent_root->fs_info->extent_ins;
2401         pending_del = &extent_root->fs_info->pending_del;
2402
2403         while(1) {
2404                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2405                                             EXTENT_LOCKED);
2406                 if (ret)
2407                         break;
2408
2409                 ret = get_state_private(pending_del, start, &priv);
2410                 BUG_ON(ret);
2411                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2412
2413                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2414                                   GFP_NOFS);
2415
2416                 if (!test_range_bit(extent_ins, start, end,
2417                                     EXTENT_LOCKED, 0)) {
2418                         ret = __free_extent(trans, extent_root,
2419                                             start, end + 1 - start, 0,
2420                                             extent_root->root_key.objectid,
2421                                             extent_op->level, 0, 1);
2422                         kfree(extent_op);
2423                 } else {
2424                         kfree(extent_op);
2425                         ret = get_state_private(extent_ins, start, &priv);
2426                         BUG_ON(ret);
2427                         extent_op = (struct pending_extent_op *)
2428                                                         (unsigned long)priv;
2429
2430                         clear_extent_bits(extent_ins, start, end,
2431                                           EXTENT_LOCKED, GFP_NOFS);
2432
2433                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2434                                 BUG_ON(1);
2435
2436                         kfree(extent_op);
2437                 }
2438                 if (ret)
2439                         err = ret;
2440         }
2441         return err;
2442 }
2443
2444 /*
2445  * remove an extent from the root, returns 0 on success
2446  */
2447
2448 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2449                       struct btrfs_root *root,
2450                       u64 bytenr, u64 num_bytes, u64 parent,
2451                       u64 root_objectid, u64 owner, u64 offset)
2452 {
2453         struct btrfs_root *extent_root = root->fs_info->extent_root;
2454         int pending_ret;
2455         int ret;
2456
2457         WARN_ON(num_bytes < root->sectorsize);
2458         if (root == extent_root) {
2459                 struct pending_extent_op *extent_op;
2460
2461                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2462                 BUG_ON(!extent_op);
2463
2464                 extent_op->type = PENDING_EXTENT_DELETE;
2465                 extent_op->bytenr = bytenr;
2466                 extent_op->num_bytes = num_bytes;
2467                 extent_op->level = (int)owner;
2468
2469                 set_extent_bits(&root->fs_info->pending_del,
2470                                 bytenr, bytenr + num_bytes - 1,
2471                                 EXTENT_LOCKED, GFP_NOFS);
2472                 set_state_private(&root->fs_info->pending_del,
2473                                   bytenr, (unsigned long)extent_op);
2474                 return 0;
2475         }
2476         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2477                             root_objectid, owner, offset, 1);
2478         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2479         return ret ? ret : pending_ret;
2480 }
2481
2482 static u64 stripe_align(struct btrfs_root *root, u64 val)
2483 {
2484         u64 mask = ((u64)root->stripesize - 1);
2485         u64 ret = (val + mask) & ~mask;
2486         return ret;
2487 }
2488
2489 /*
2490  * walks the btree of allocated extents and find a hole of a given size.
2491  * The key ins is changed to record the hole:
2492  * ins->objectid == block start
2493  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2494  * ins->offset == number of blocks
2495  * Any available blocks before search_start are skipped.
2496  */
2497 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2498                                      struct btrfs_root *orig_root,
2499                                      u64 num_bytes, u64 empty_size,
2500                                      u64 search_start, u64 search_end,
2501                                      u64 hint_byte, struct btrfs_key *ins,
2502                                      u64 exclude_start, u64 exclude_nr,
2503                                      int data)
2504 {
2505         int ret;
2506         u64 orig_search_start = search_start;
2507         struct btrfs_root * root = orig_root->fs_info->extent_root;
2508         struct btrfs_fs_info *info = root->fs_info;
2509         u64 total_needed = num_bytes;
2510         struct btrfs_block_group_cache *block_group;
2511         int full_scan = 0;
2512         int wrapped = 0;
2513
2514         WARN_ON(num_bytes < root->sectorsize);
2515         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2516
2517         search_start = stripe_align(root, search_start);
2518
2519         if (hint_byte) {
2520                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2521                 if (!block_group)
2522                         hint_byte = search_start;
2523                 block_group = btrfs_find_block_group(root, block_group,
2524                                                      hint_byte, data, 1);
2525         } else {
2526                 block_group = btrfs_find_block_group(root,
2527                                                      trans->block_group,
2528                                                      search_start, data, 1);
2529         }
2530
2531         total_needed += empty_size;
2532
2533 check_failed:
2534         search_start = stripe_align(root, search_start);
2535         if (!block_group) {
2536                 block_group = btrfs_lookup_first_block_group(info,
2537                                                              search_start);
2538                 if (!block_group)
2539                         block_group = btrfs_lookup_first_block_group(info,
2540                                                        orig_search_start);
2541         }
2542         ret = find_search_start(root, &block_group, &search_start,
2543                                 total_needed, data);
2544         if (ret)
2545                 goto new_group;
2546
2547         ins->objectid = search_start;
2548         ins->offset = num_bytes;
2549
2550         if (ins->objectid + num_bytes >
2551             block_group->key.objectid + block_group->key.offset) {
2552                 search_start = block_group->key.objectid +
2553                         block_group->key.offset;
2554                 goto new_group;
2555         }
2556
2557         if (test_range_bit(&info->extent_ins, ins->objectid,
2558                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2559                 search_start = ins->objectid + num_bytes;
2560                 goto new_group;
2561         }
2562
2563         if (test_range_bit(&info->pinned_extents, ins->objectid,
2564                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2565                 search_start = ins->objectid + num_bytes;
2566                 goto new_group;
2567         }
2568
2569         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2570             ins->objectid < exclude_start + exclude_nr)) {
2571                 search_start = exclude_start + exclude_nr;
2572                 goto new_group;
2573         }
2574
2575         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2576                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2577                 if (block_group)
2578                         trans->block_group = block_group;
2579         }
2580         ins->offset = num_bytes;
2581         return 0;
2582
2583 new_group:
2584         block_group = btrfs_lookup_first_block_group(info, search_start);
2585         if (!block_group) {
2586                 search_start = orig_search_start;
2587                 if (full_scan) {
2588                         ret = -ENOSPC;
2589                         goto error;
2590                 }
2591                 if (wrapped) {
2592                         if (!full_scan)
2593                                 total_needed -= empty_size;
2594                         full_scan = 1;
2595                 } else
2596                         wrapped = 1;
2597         }
2598         cond_resched();
2599         block_group = btrfs_find_block_group(root, block_group,
2600                                              search_start, data, 0);
2601         goto check_failed;
2602
2603 error:
2604         return ret;
2605 }
2606
2607 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2608                          struct btrfs_root *root,
2609                          u64 num_bytes, u64 empty_size,
2610                          u64 hint_byte, u64 search_end,
2611                          struct btrfs_key *ins, int data)
2612 {
2613         int ret;
2614         u64 search_start = 0;
2615         u64 alloc_profile;
2616         struct btrfs_fs_info *info = root->fs_info;
2617
2618         if (info->extent_ops) {
2619                 struct btrfs_extent_ops *ops = info->extent_ops;
2620                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2621                 BUG_ON(ret);
2622                 goto found;
2623         }
2624
2625         if (data) {
2626                 alloc_profile = info->avail_data_alloc_bits &
2627                                 info->data_alloc_profile;
2628                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2629         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2630                    info->system_allocs >= 0) {
2631                 alloc_profile = info->avail_system_alloc_bits &
2632                                 info->system_alloc_profile;
2633                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2634         } else {
2635                 alloc_profile = info->avail_metadata_alloc_bits &
2636                                 info->metadata_alloc_profile;
2637                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2638         }
2639
2640         if (root->ref_cows) {
2641                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2642                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2643                                              num_bytes,
2644                                              BTRFS_BLOCK_GROUP_METADATA);
2645                         BUG_ON(ret);
2646                 }
2647                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2648                                      num_bytes + 2 * 1024 * 1024, data);
2649                 BUG_ON(ret);
2650         }
2651
2652         WARN_ON(num_bytes < root->sectorsize);
2653         ret = find_free_extent(trans, root, num_bytes, empty_size,
2654                                search_start, search_end, hint_byte, ins,
2655                                trans->alloc_exclude_start,
2656                                trans->alloc_exclude_nr, data);
2657         BUG_ON(ret);
2658 found:
2659         clear_extent_dirty(&root->fs_info->free_space_cache,
2660                            ins->objectid, ins->objectid + ins->offset - 1,
2661                            GFP_NOFS);
2662         return ret;
2663 }
2664
2665 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2666                                      struct btrfs_root *root,
2667                                      u64 root_objectid, u64 generation,
2668                                      u64 flags, struct btrfs_disk_key *key,
2669                                      int level, struct btrfs_key *ins)
2670 {
2671         int ret;
2672         struct btrfs_fs_info *fs_info = root->fs_info;
2673         struct btrfs_extent_item *extent_item;
2674         struct btrfs_tree_block_info *block_info;
2675         struct btrfs_extent_inline_ref *iref;
2676         struct btrfs_path *path;
2677         struct extent_buffer *leaf;
2678         u32 size = sizeof(*extent_item) + sizeof(*iref);
2679         int skinny_metadata =
2680                 btrfs_fs_incompat(fs_info,
2681                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2682
2683         if (!skinny_metadata)
2684                 size += sizeof(*block_info);
2685
2686         path = btrfs_alloc_path();
2687         BUG_ON(!path);
2688
2689         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2690                                       ins, size);
2691         BUG_ON(ret);
2692
2693         leaf = path->nodes[0];
2694         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2695                                      struct btrfs_extent_item);
2696         btrfs_set_extent_refs(leaf, extent_item, 1);
2697         btrfs_set_extent_generation(leaf, extent_item, generation);
2698         btrfs_set_extent_flags(leaf, extent_item,
2699                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2700
2701         if (skinny_metadata) {
2702                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2703         } else {
2704                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2705                 btrfs_set_tree_block_key(leaf, block_info, key);
2706                 btrfs_set_tree_block_level(leaf, block_info, level);
2707                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2708         }
2709
2710         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2711         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2712
2713         btrfs_mark_buffer_dirty(leaf);
2714         btrfs_free_path(path);
2715
2716         ret = update_block_group(trans, root, ins->objectid, root->leafsize,
2717                                  1, 0);
2718         return ret;
2719 }
2720
2721 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2722                             struct btrfs_root *root, u64 num_bytes,
2723                             u64 root_objectid, u64 generation,
2724                             u64 flags, struct btrfs_disk_key *key,
2725                             int level, u64 empty_size, u64 hint_byte,
2726                             u64 search_end, struct btrfs_key *ins)
2727 {
2728         int ret;
2729         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2730                                    hint_byte, search_end, ins, 0);
2731         BUG_ON(ret);
2732
2733         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2734                 struct pending_extent_op *extent_op;
2735
2736                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2737                 BUG_ON(!extent_op);
2738
2739                 extent_op->type = PENDING_EXTENT_INSERT;
2740                 extent_op->bytenr = ins->objectid;
2741                 extent_op->num_bytes = ins->offset;
2742                 extent_op->level = level;
2743                 extent_op->flags = flags;
2744                 memcpy(&extent_op->key, key, sizeof(*key));
2745
2746                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2747                                 ins->objectid + ins->offset - 1,
2748                                 EXTENT_LOCKED, GFP_NOFS);
2749                 set_state_private(&root->fs_info->extent_ins,
2750                                   ins->objectid, (unsigned long)extent_op);
2751         } else {
2752                 if (btrfs_fs_incompat(root->fs_info,
2753                                 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
2754                         ins->offset = level;
2755                         ins->type = BTRFS_METADATA_ITEM_KEY;
2756                 }
2757                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2758                                                 generation, flags,
2759                                                 key, level, ins);
2760                 finish_current_insert(trans, root->fs_info->extent_root);
2761                 del_pending_extents(trans, root->fs_info->extent_root);
2762         }
2763         return ret;
2764 }
2765
2766 /*
2767  * helper function to allocate a block for a given tree
2768  * returns the tree buffer or NULL.
2769  */
2770 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2771                                         struct btrfs_root *root,
2772                                         u32 blocksize, u64 root_objectid,
2773                                         struct btrfs_disk_key *key, int level,
2774                                         u64 hint, u64 empty_size)
2775 {
2776         struct btrfs_key ins;
2777         int ret;
2778         struct extent_buffer *buf;
2779
2780         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2781                                trans->transid, 0, key, level,
2782                                empty_size, hint, (u64)-1, &ins);
2783         if (ret) {
2784                 BUG_ON(ret > 0);
2785                 return ERR_PTR(ret);
2786         }
2787
2788         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2789         if (!buf) {
2790                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2791                                   0, root->root_key.objectid, level, 0);
2792                 BUG_ON(1);
2793                 return ERR_PTR(-ENOMEM);
2794         }
2795         btrfs_set_buffer_uptodate(buf);
2796         trans->blocks_used++;
2797
2798         return buf;
2799 }
2800
2801 #if 0
2802
2803 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2804                                   struct btrfs_root *root,
2805                                   struct extent_buffer *leaf)
2806 {
2807         u64 leaf_owner;
2808         u64 leaf_generation;
2809         struct btrfs_key key;
2810         struct btrfs_file_extent_item *fi;
2811         int i;
2812         int nritems;
2813         int ret;
2814
2815         BUG_ON(!btrfs_is_leaf(leaf));
2816         nritems = btrfs_header_nritems(leaf);
2817         leaf_owner = btrfs_header_owner(leaf);
2818         leaf_generation = btrfs_header_generation(leaf);
2819
2820         for (i = 0; i < nritems; i++) {
2821                 u64 disk_bytenr;
2822
2823                 btrfs_item_key_to_cpu(leaf, &key, i);
2824                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2825                         continue;
2826                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2827                 if (btrfs_file_extent_type(leaf, fi) ==
2828                     BTRFS_FILE_EXTENT_INLINE)
2829                         continue;
2830                 /*
2831                  * FIXME make sure to insert a trans record that
2832                  * repeats the snapshot del on crash
2833                  */
2834                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2835                 if (disk_bytenr == 0)
2836                         continue;
2837                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2838                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2839                                 leaf->start, leaf_owner, leaf_generation,
2840                                 key.objectid, 0);
2841                 BUG_ON(ret);
2842         }
2843         return 0;
2844 }
2845
2846 static void noinline reada_walk_down(struct btrfs_root *root,
2847                                      struct extent_buffer *node,
2848                                      int slot)
2849 {
2850         u64 bytenr;
2851         u64 last = 0;
2852         u32 nritems;
2853         u32 refs;
2854         u32 blocksize;
2855         int ret;
2856         int i;
2857         int level;
2858         int skipped = 0;
2859
2860         nritems = btrfs_header_nritems(node);
2861         level = btrfs_header_level(node);
2862         if (level)
2863                 return;
2864
2865         for (i = slot; i < nritems && skipped < 32; i++) {
2866                 bytenr = btrfs_node_blockptr(node, i);
2867                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2868                              (last > bytenr && last - bytenr > 32 * 1024))) {
2869                         skipped++;
2870                         continue;
2871                 }
2872                 blocksize = btrfs_level_size(root, level - 1);
2873                 if (i != slot) {
2874                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2875                                                       blocksize, &refs);
2876                         BUG_ON(ret);
2877                         if (refs != 1) {
2878                                 skipped++;
2879                                 continue;
2880                         }
2881                 }
2882                 mutex_unlock(&root->fs_info->fs_mutex);
2883                 ret = readahead_tree_block(root, bytenr, blocksize,
2884                                            btrfs_node_ptr_generation(node, i));
2885                 last = bytenr + blocksize;
2886                 cond_resched();
2887                 mutex_lock(&root->fs_info->fs_mutex);
2888                 if (ret)
2889                         break;
2890         }
2891 }
2892
2893 /*
2894  * helper function for drop_snapshot, this walks down the tree dropping ref
2895  * counts as it goes.
2896  */
2897 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2898                                    struct btrfs_root *root,
2899                                    struct btrfs_path *path, int *level)
2900 {
2901         u64 root_owner;
2902         u64 root_gen;
2903         u64 bytenr;
2904         u64 ptr_gen;
2905         struct extent_buffer *next;
2906         struct extent_buffer *cur;
2907         struct extent_buffer *parent;
2908         u32 blocksize;
2909         int ret;
2910         u32 refs;
2911
2912         WARN_ON(*level < 0);
2913         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2914         ret = btrfs_lookup_extent_ref(trans, root,
2915                                       path->nodes[*level]->start,
2916                                       path->nodes[*level]->len, &refs);
2917         BUG_ON(ret);
2918         if (refs > 1)
2919                 goto out;
2920
2921         /*
2922          * walk down to the last node level and free all the leaves
2923          */
2924         while(*level >= 0) {
2925                 WARN_ON(*level < 0);
2926                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2927                 cur = path->nodes[*level];
2928
2929                 if (btrfs_header_level(cur) != *level)
2930                         WARN_ON(1);
2931
2932                 if (path->slots[*level] >=
2933                     btrfs_header_nritems(cur))
2934                         break;
2935                 if (*level == 0) {
2936                         ret = drop_leaf_ref(trans, root, cur);
2937                         BUG_ON(ret);
2938                         break;
2939                 }
2940                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2941                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2942                 blocksize = btrfs_level_size(root, *level - 1);
2943                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2944                                               &refs);
2945                 BUG_ON(ret);
2946                 if (refs != 1) {
2947                         parent = path->nodes[*level];
2948                         root_owner = btrfs_header_owner(parent);
2949                         root_gen = btrfs_header_generation(parent);
2950                         path->slots[*level]++;
2951                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2952                                                 parent->start, root_owner,
2953                                                 root_gen, *level - 1, 1);
2954                         BUG_ON(ret);
2955                         continue;
2956                 }
2957                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2958                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2959                         free_extent_buffer(next);
2960                         reada_walk_down(root, cur, path->slots[*level]);
2961                         mutex_unlock(&root->fs_info->fs_mutex);
2962                         next = read_tree_block(root, bytenr, blocksize,
2963                                                ptr_gen);
2964                         mutex_lock(&root->fs_info->fs_mutex);
2965                         if (!extent_buffer_uptodate(next)) {
2966                                 if (IS_ERR(next))
2967                                         ret = PTR_ERR(next);
2968                                 else
2969                                         ret = -EIO;
2970                                 break;
2971                         }
2972                 }
2973                 WARN_ON(*level <= 0);
2974                 if (path->nodes[*level-1])
2975                         free_extent_buffer(path->nodes[*level-1]);
2976                 path->nodes[*level-1] = next;
2977                 *level = btrfs_header_level(next);
2978                 path->slots[*level] = 0;
2979         }
2980 out:
2981         WARN_ON(*level < 0);
2982         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2983
2984         if (path->nodes[*level] == root->node) {
2985                 root_owner = root->root_key.objectid;
2986                 parent = path->nodes[*level];
2987         } else {
2988                 parent = path->nodes[*level + 1];
2989                 root_owner = btrfs_header_owner(parent);
2990         }
2991
2992         root_gen = btrfs_header_generation(parent);
2993         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2994                                 path->nodes[*level]->len, parent->start,
2995                                 root_owner, root_gen, *level, 1);
2996         free_extent_buffer(path->nodes[*level]);
2997         path->nodes[*level] = NULL;
2998         *level += 1;
2999         BUG_ON(ret);
3000         return 0;
3001 }
3002
3003 /*
3004  * helper for dropping snapshots.  This walks back up the tree in the path
3005  * to find the first node higher up where we haven't yet gone through
3006  * all the slots
3007  */
3008 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3009                                  struct btrfs_root *root,
3010                                  struct btrfs_path *path, int *level)
3011 {
3012         u64 root_owner;
3013         u64 root_gen;
3014         struct btrfs_root_item *root_item = &root->root_item;
3015         int i;
3016         int slot;
3017         int ret;
3018
3019         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3020                 slot = path->slots[i];
3021                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3022                         struct extent_buffer *node;
3023                         struct btrfs_disk_key disk_key;
3024                         node = path->nodes[i];
3025                         path->slots[i]++;
3026                         *level = i;
3027                         WARN_ON(*level == 0);
3028                         btrfs_node_key(node, &disk_key, path->slots[i]);
3029                         memcpy(&root_item->drop_progress,
3030                                &disk_key, sizeof(disk_key));
3031                         root_item->drop_level = i;
3032                         return 0;
3033                 } else {
3034                         struct extent_buffer *parent;
3035                         if (path->nodes[*level] == root->node)
3036                                 parent = path->nodes[*level];
3037                         else
3038                                 parent = path->nodes[*level + 1];
3039
3040                         root_owner = btrfs_header_owner(parent);
3041                         root_gen = btrfs_header_generation(parent);
3042                         ret = btrfs_free_extent(trans, root,
3043                                                 path->nodes[*level]->start,
3044                                                 path->nodes[*level]->len,
3045                                                 parent->start, root_owner,
3046                                                 root_gen, *level, 1);
3047                         BUG_ON(ret);
3048                         free_extent_buffer(path->nodes[*level]);
3049                         path->nodes[*level] = NULL;
3050                         *level = i + 1;
3051                 }
3052         }
3053         return 1;
3054 }
3055
3056 #endif
3057
3058 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3059 {
3060         struct btrfs_space_info *sinfo;
3061         struct btrfs_block_group_cache *cache;
3062         u64 start;
3063         u64 end;
3064         u64 ptr;
3065         int ret;
3066
3067         while(1) {
3068                 ret = find_first_extent_bit(&info->block_group_cache, 0,
3069                                             &start, &end, (unsigned int)-1);
3070                 if (ret)
3071                         break;
3072                 ret = get_state_private(&info->block_group_cache, start, &ptr);
3073                 if (!ret) {
3074                         cache = u64_to_ptr(ptr);
3075                         if (cache->free_space_ctl) {
3076                                 btrfs_remove_free_space_cache(cache);
3077                                 kfree(cache->free_space_ctl);
3078                         }
3079                         kfree(cache);
3080                 }
3081                 clear_extent_bits(&info->block_group_cache, start,
3082                                   end, (unsigned int)-1, GFP_NOFS);
3083         }
3084         while(1) {
3085                 ret = find_first_extent_bit(&info->free_space_cache, 0,
3086                                             &start, &end, EXTENT_DIRTY);
3087                 if (ret)
3088                         break;
3089                 clear_extent_dirty(&info->free_space_cache, start,
3090                                    end, GFP_NOFS);
3091         }
3092
3093         while (!list_empty(&info->space_info)) {
3094                 sinfo = list_entry(info->space_info.next,
3095                                    struct btrfs_space_info, list);
3096                 list_del_init(&sinfo->list);
3097                 kfree(sinfo);
3098         }
3099         return 0;
3100 }
3101
3102 static int find_first_block_group(struct btrfs_root *root,
3103                 struct btrfs_path *path, struct btrfs_key *key)
3104 {
3105         int ret;
3106         struct btrfs_key found_key;
3107         struct extent_buffer *leaf;
3108         int slot;
3109
3110         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3111         if (ret < 0)
3112                 return ret;
3113         while(1) {
3114                 slot = path->slots[0];
3115                 leaf = path->nodes[0];
3116                 if (slot >= btrfs_header_nritems(leaf)) {
3117                         ret = btrfs_next_leaf(root, path);
3118                         if (ret == 0)
3119                                 continue;
3120                         if (ret < 0)
3121                                 goto error;
3122                         break;
3123                 }
3124                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3125
3126                 if (found_key.objectid >= key->objectid &&
3127                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3128                         return 0;
3129                 path->slots[0]++;
3130         }
3131         ret = -ENOENT;
3132 error:
3133         return ret;
3134 }
3135
3136 int btrfs_read_block_groups(struct btrfs_root *root)
3137 {
3138         struct btrfs_path *path;
3139         int ret;
3140         int bit;
3141         struct btrfs_block_group_cache *cache;
3142         struct btrfs_fs_info *info = root->fs_info;
3143         struct btrfs_space_info *space_info;
3144         struct extent_io_tree *block_group_cache;
3145         struct btrfs_key key;
3146         struct btrfs_key found_key;
3147         struct extent_buffer *leaf;
3148
3149         block_group_cache = &info->block_group_cache;
3150
3151         root = info->extent_root;
3152         key.objectid = 0;
3153         key.offset = 0;
3154         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3155         path = btrfs_alloc_path();
3156         if (!path)
3157                 return -ENOMEM;
3158
3159         while(1) {
3160                 ret = find_first_block_group(root, path, &key);
3161                 if (ret > 0) {
3162                         ret = 0;
3163                         goto error;
3164                 }
3165                 if (ret != 0) {
3166                         goto error;
3167                 }
3168                 leaf = path->nodes[0];
3169                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3170                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3171                 if (!cache) {
3172                         ret = -ENOMEM;
3173                         break;
3174                 }
3175
3176                 read_extent_buffer(leaf, &cache->item,
3177                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3178                                    sizeof(cache->item));
3179                 memcpy(&cache->key, &found_key, sizeof(found_key));
3180                 cache->cached = 0;
3181                 cache->pinned = 0;
3182                 key.objectid = found_key.objectid + found_key.offset;
3183                 btrfs_release_path(path);
3184                 cache->flags = btrfs_block_group_flags(&cache->item);
3185                 bit = 0;
3186                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3187                         bit = BLOCK_GROUP_DATA;
3188                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3189                         bit = BLOCK_GROUP_SYSTEM;
3190                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3191                         bit = BLOCK_GROUP_METADATA;
3192                 }
3193                 set_avail_alloc_bits(info, cache->flags);
3194                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3195                         cache->ro = 1;
3196
3197                 ret = update_space_info(info, cache->flags, found_key.offset,
3198                                         btrfs_block_group_used(&cache->item),
3199                                         &space_info);
3200                 BUG_ON(ret);
3201                 cache->space_info = space_info;
3202
3203                 /* use EXTENT_LOCKED to prevent merging */
3204                 set_extent_bits(block_group_cache, found_key.objectid,
3205                                 found_key.objectid + found_key.offset - 1,
3206                                 bit | EXTENT_LOCKED, GFP_NOFS);
3207                 set_state_private(block_group_cache, found_key.objectid,
3208                                   (unsigned long)cache);
3209         }
3210         ret = 0;
3211 error:
3212         btrfs_free_path(path);
3213         return ret;
3214 }
3215
3216 struct btrfs_block_group_cache *
3217 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3218                       u64 chunk_objectid, u64 chunk_offset, u64 size)
3219 {
3220         int ret;
3221         int bit = 0;
3222         struct btrfs_block_group_cache *cache;
3223         struct extent_io_tree *block_group_cache;
3224
3225         block_group_cache = &fs_info->block_group_cache;
3226
3227         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3228         BUG_ON(!cache);
3229         cache->key.objectid = chunk_offset;
3230         cache->key.offset = size;
3231
3232         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3233         btrfs_set_block_group_used(&cache->item, bytes_used);
3234         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3235         cache->flags = type;
3236         btrfs_set_block_group_flags(&cache->item, type);
3237
3238         ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3239                                 &cache->space_info);
3240         BUG_ON(ret);
3241
3242         bit = block_group_state_bits(type);
3243         ret = set_extent_bits(block_group_cache, chunk_offset,
3244                               chunk_offset + size - 1,
3245                               bit | EXTENT_LOCKED, GFP_NOFS);
3246         BUG_ON(ret);
3247
3248         ret = set_state_private(block_group_cache, chunk_offset,
3249                                 (unsigned long)cache);
3250         BUG_ON(ret);
3251         set_avail_alloc_bits(fs_info, type);
3252
3253         return cache;
3254 }
3255
3256 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3257                            struct btrfs_root *root, u64 bytes_used,
3258                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3259                            u64 size)
3260 {
3261         int ret;
3262         struct btrfs_root *extent_root;
3263         struct btrfs_block_group_cache *cache;
3264
3265         cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3266                                       chunk_objectid, chunk_offset, size);
3267         extent_root = root->fs_info->extent_root;
3268         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3269                                 sizeof(cache->item));
3270         BUG_ON(ret);
3271
3272         ret = finish_current_insert(trans, extent_root);
3273         BUG_ON(ret);
3274         ret = del_pending_extents(trans, extent_root);
3275         BUG_ON(ret);
3276
3277         return 0;
3278 }
3279
3280 /*
3281  * This is for converter use only.
3282  *
3283  * In that case, we don't know where are free blocks located.
3284  * Therefore all block group cache entries must be setup properly
3285  * before doing any block allocation.
3286  */
3287 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3288                             struct btrfs_root *root)
3289 {
3290         u64 total_bytes;
3291         u64 cur_start;
3292         u64 group_type;
3293         u64 group_size;
3294         u64 group_align;
3295         u64 total_data = 0;
3296         u64 total_metadata = 0;
3297         u64 chunk_objectid;
3298         int ret;
3299         int bit;
3300         struct btrfs_root *extent_root;
3301         struct btrfs_block_group_cache *cache;
3302         struct extent_io_tree *block_group_cache;
3303
3304         extent_root = root->fs_info->extent_root;
3305         block_group_cache = &root->fs_info->block_group_cache;
3306         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3307         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3308         group_align = 64 * root->sectorsize;
3309
3310         cur_start = 0;
3311         while (cur_start < total_bytes) {
3312                 group_size = total_bytes / 12;
3313                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3314                 if (cur_start == 0) {
3315                         bit = BLOCK_GROUP_SYSTEM;
3316                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3317                         group_size /= 4;
3318                         group_size &= ~(group_align - 1);
3319                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3320                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3321                 } else {
3322                         group_size &= ~(group_align - 1);
3323                         if (total_data >= total_metadata * 2) {
3324                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3325                                 group_size = min_t(u64, group_size,
3326                                                    1ULL * 1024 * 1024 * 1024);
3327                                 total_metadata += group_size;
3328                         } else {
3329                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3330                                 group_size = min_t(u64, group_size,
3331                                                    5ULL * 1024 * 1024 * 1024);
3332                                 total_data += group_size;
3333                         }
3334                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3335                                 group_size = total_bytes - cur_start;
3336                 }
3337
3338                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3339                 BUG_ON(!cache);
3340
3341                 cache->key.objectid = cur_start;
3342                 cache->key.offset = group_size;
3343                 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3344
3345                 btrfs_set_block_group_used(&cache->item, 0);
3346                 btrfs_set_block_group_chunk_objectid(&cache->item,
3347                                                      chunk_objectid);
3348                 btrfs_set_block_group_flags(&cache->item, group_type);
3349
3350                 cache->flags = group_type;
3351
3352                 ret = update_space_info(root->fs_info, group_type, group_size,
3353                                         0, &cache->space_info);
3354                 BUG_ON(ret);
3355                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3356
3357                 set_extent_bits(block_group_cache, cur_start,
3358                                 cur_start + group_size - 1,
3359                                 bit | EXTENT_LOCKED, GFP_NOFS);
3360                 set_state_private(block_group_cache, cur_start,
3361                                   (unsigned long)cache);
3362                 cur_start += group_size;
3363         }
3364         /* then insert all the items */
3365         cur_start = 0;
3366         while(cur_start < total_bytes) {
3367                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3368                 BUG_ON(!cache);
3369
3370                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3371                                         sizeof(cache->item));
3372                 BUG_ON(ret);
3373
3374                 finish_current_insert(trans, extent_root);
3375                 ret = del_pending_extents(trans, extent_root);
3376                 BUG_ON(ret);
3377
3378                 cur_start = cache->key.objectid + cache->key.offset;
3379         }
3380         return 0;
3381 }
3382
3383 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3384                              struct btrfs_root *root,
3385                              u64 bytenr, u64 num_bytes, int alloc,
3386                              int mark_free)
3387 {
3388         return update_block_group(trans, root, bytenr, num_bytes,
3389                                   alloc, mark_free);
3390 }
3391
3392 /*
3393  * Fixup block accounting. The initial block accounting created by
3394  * make_block_groups isn't accuracy in this case.
3395  */
3396 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3397                                struct btrfs_root *root)
3398 {
3399         int ret;
3400         int slot;
3401         u64 start = 0;
3402         u64 bytes_used = 0;
3403         struct btrfs_path path;
3404         struct btrfs_key key;
3405         struct extent_buffer *leaf;
3406         struct btrfs_block_group_cache *cache;
3407         struct btrfs_fs_info *fs_info = root->fs_info;
3408
3409         root = root->fs_info->extent_root;
3410
3411         while(extent_root_pending_ops(fs_info)) {
3412                 ret = finish_current_insert(trans, root);
3413                 if (ret)
3414                         return ret;
3415                 ret = del_pending_extents(trans, root);
3416                 if (ret)
3417                         return ret;
3418         }
3419
3420         while(1) {
3421                 cache = btrfs_lookup_first_block_group(fs_info, start);
3422                 if (!cache)
3423                         break;
3424                 start = cache->key.objectid + cache->key.offset;
3425                 btrfs_set_block_group_used(&cache->item, 0);
3426                 cache->space_info->bytes_used = 0;
3427                 set_extent_bits(&root->fs_info->block_group_cache,
3428                                 cache->key.objectid,
3429                                 cache->key.objectid + cache->key.offset -1,
3430                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
3431         }
3432
3433         btrfs_init_path(&path);
3434         key.offset = 0;
3435         key.objectid = 0;
3436         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
3437         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3438                                 &key, &path, 0, 0);
3439         if (ret < 0)
3440                 return ret;
3441         while(1) {
3442                 leaf = path.nodes[0];
3443                 slot = path.slots[0];
3444                 if (slot >= btrfs_header_nritems(leaf)) {
3445                         ret = btrfs_next_leaf(root, &path);
3446                         if (ret < 0)
3447                                 return ret;
3448                         if (ret > 0)
3449                                 break;
3450                         leaf = path.nodes[0];
3451                         slot = path.slots[0];
3452                 }
3453                 btrfs_item_key_to_cpu(leaf, &key, slot);
3454                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3455                         bytes_used += key.offset;
3456                         ret = btrfs_update_block_group(trans, root,
3457                                   key.objectid, key.offset, 1, 0);
3458                         BUG_ON(ret);
3459                 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3460                         bytes_used += root->leafsize;
3461                         ret = btrfs_update_block_group(trans, root,
3462                                   key.objectid, root->leafsize, 1, 0);
3463                         BUG_ON(ret);
3464                 }
3465                 path.slots[0]++;
3466         }
3467         btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3468         btrfs_release_path(&path);
3469         return 0;
3470 }
3471
3472 /*
3473  * Record a file extent. Do all the required works, such as inserting
3474  * file extent item, inserting extent item and backref item into extent
3475  * tree and updating block accounting.
3476  */
3477 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3478                               struct btrfs_root *root, u64 objectid,
3479                               struct btrfs_inode_item *inode,
3480                               u64 file_pos, u64 disk_bytenr,
3481                               u64 num_bytes)
3482 {
3483         int ret;
3484         struct btrfs_fs_info *info = root->fs_info;
3485         struct btrfs_root *extent_root = info->extent_root;
3486         struct extent_buffer *leaf;
3487         struct btrfs_file_extent_item *fi;
3488         struct btrfs_key ins_key;
3489         struct btrfs_path path;
3490         struct btrfs_extent_item *ei;
3491         u64 nbytes;
3492
3493         if (disk_bytenr == 0) {
3494                 ret = btrfs_insert_file_extent(trans, root, objectid,
3495                                                 file_pos, disk_bytenr,
3496                                                 num_bytes, num_bytes);
3497                 return ret;
3498         }
3499
3500         btrfs_init_path(&path);
3501
3502         ins_key.objectid = objectid;
3503         ins_key.offset = file_pos;
3504         btrfs_set_key_type(&ins_key, BTRFS_EXTENT_DATA_KEY);
3505         ret = btrfs_insert_empty_item(trans, root, &path, &ins_key,
3506                                       sizeof(*fi));
3507         if (ret)
3508                 goto fail;
3509         leaf = path.nodes[0];
3510         fi = btrfs_item_ptr(leaf, path.slots[0],
3511                             struct btrfs_file_extent_item);
3512         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
3513         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
3514         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
3515         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
3516         btrfs_set_file_extent_offset(leaf, fi, 0);
3517         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
3518         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
3519         btrfs_set_file_extent_compression(leaf, fi, 0);
3520         btrfs_set_file_extent_encryption(leaf, fi, 0);
3521         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
3522         btrfs_mark_buffer_dirty(leaf);
3523
3524         nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
3525         btrfs_set_stack_inode_nbytes(inode, nbytes);
3526
3527         btrfs_release_path(&path);
3528
3529         ins_key.objectid = disk_bytenr;
3530         ins_key.offset = num_bytes;
3531         ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3532
3533         ret = btrfs_insert_empty_item(trans, extent_root, &path,
3534                                       &ins_key, sizeof(*ei));
3535         if (ret == 0) {
3536                 leaf = path.nodes[0];
3537                 ei = btrfs_item_ptr(leaf, path.slots[0],
3538                                     struct btrfs_extent_item);
3539
3540                 btrfs_set_extent_refs(leaf, ei, 0);
3541                 btrfs_set_extent_generation(leaf, ei, 0);
3542                 btrfs_set_extent_flags(leaf, ei, BTRFS_EXTENT_FLAG_DATA);
3543
3544                 btrfs_mark_buffer_dirty(leaf);
3545
3546                 ret = btrfs_update_block_group(trans, root, disk_bytenr,
3547                                                num_bytes, 1, 0);
3548                 if (ret)
3549                         goto fail;
3550         } else if (ret != -EEXIST) {
3551                 goto fail;
3552         }
3553         btrfs_extent_post_op(trans, extent_root);
3554
3555         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, num_bytes, 0,
3556                                    root->root_key.objectid,
3557                                    objectid, file_pos);
3558         if (ret)
3559                 goto fail;
3560         ret = 0;
3561 fail:
3562         btrfs_release_path(&path);
3563         return ret;
3564 }