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 <math.h>
  23 #include "kerncompat.h"
  24 #include "radix-tree.h"
  25 #include "ctree.h"
  26 #include "disk-io.h"
  27 #include "print-tree.h"
  28 #include "transaction.h"
  29 #include "crc32c.h"
  30 #include "volumes.h"
  31 #include "free-space-cache.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->nodesize;
 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 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->nodesize;
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->nodesize)
1497                                 ret = 0;
1498                 }
1499
1500                 if (ret) {
1501                         btrfs_release_path(path);
1502                         key.type = BTRFS_EXTENT_ITEM_KEY;
1503                         key.offset = root->nodesize;
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->nodesize;
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->nodesize &&
1586                             key.type == BTRFS_EXTENT_ITEM_KEY)
1587                                 ret = 0;
1588                 }
1589                 if (ret) {
1590                         btrfs_release_path(path);
1591                         key.offset = root->nodesize;
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 = root->nodesize;
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 btrfs_space_info *found;
1793
1794         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1795
1796         list_for_each_entry(found, &info->space_info, list) {
1797                 if (found->flags & flags)
1798                         return found;
1799         }
1800         return NULL;
1801
1802 }
1803
1804 static int free_space_info(struct btrfs_fs_info *fs_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         /* only support free block group which is empty */
1811         if (bytes_used)
1812                 return -ENOTEMPTY;
1813
1814         found = __find_space_info(fs_info, flags);
1815         if (!found)
1816                 return -ENOENT;
1817         if (found->total_bytes < total_bytes) {
1818                 fprintf(stderr,
1819                         "WARNING: bad space info to free %llu only have %llu\n",
1820                         total_bytes, found->total_bytes);
1821                 return -EINVAL;
1822         }
1823         found->total_bytes -= total_bytes;
1824         if (space_info)
1825                 *space_info = found;
1826         return 0;
1827 }
1828
1829 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1830                              u64 total_bytes, u64 bytes_used,
1831                              struct btrfs_space_info **space_info)
1832 {
1833         struct btrfs_space_info *found;
1834
1835         found = __find_space_info(info, flags);
1836         if (found) {
1837                 found->total_bytes += total_bytes;
1838                 found->bytes_used += bytes_used;
1839                 if (found->total_bytes < found->bytes_used) {
1840                         fprintf(stderr, "warning, bad space info total_bytes "
1841                                 "%llu used %llu\n",
1842                                (unsigned long long)found->total_bytes,
1843                                (unsigned long long)found->bytes_used);
1844                 }
1845                 *space_info = found;
1846                 return 0;
1847         }
1848         found = kmalloc(sizeof(*found), GFP_NOFS);
1849         if (!found)
1850                 return -ENOMEM;
1851
1852         list_add(&found->list, &info->space_info);
1853         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1854         found->total_bytes = total_bytes;
1855         found->bytes_used = bytes_used;
1856         found->bytes_pinned = 0;
1857         found->full = 0;
1858         *space_info = found;
1859         return 0;
1860 }
1861
1862
1863 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1864 {
1865         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1866                                    BTRFS_BLOCK_GROUP_RAID1 |
1867                                    BTRFS_BLOCK_GROUP_RAID10 |
1868                                    BTRFS_BLOCK_GROUP_RAID5 |
1869                                    BTRFS_BLOCK_GROUP_RAID6 |
1870                                    BTRFS_BLOCK_GROUP_DUP);
1871         if (extra_flags) {
1872                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1873                         fs_info->avail_data_alloc_bits |= extra_flags;
1874                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1875                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1876                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1877                         fs_info->avail_system_alloc_bits |= extra_flags;
1878         }
1879 }
1880
1881 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1882                           struct btrfs_root *extent_root, u64 alloc_bytes,
1883                           u64 flags)
1884 {
1885         struct btrfs_space_info *space_info;
1886         u64 thresh;
1887         u64 start;
1888         u64 num_bytes;
1889         int ret;
1890
1891         space_info = __find_space_info(extent_root->fs_info, flags);
1892         if (!space_info) {
1893                 ret = update_space_info(extent_root->fs_info, flags,
1894                                         0, 0, &space_info);
1895                 BUG_ON(ret);
1896         }
1897         BUG_ON(!space_info);
1898
1899         if (space_info->full)
1900                 return 0;
1901
1902         thresh = div_factor(space_info->total_bytes, 7);
1903         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1904             thresh)
1905                 return 0;
1906
1907         /*
1908          * Avoid allocating given chunk type
1909          */
1910         if (extent_root->fs_info->avoid_meta_chunk_alloc &&
1911             (flags & BTRFS_BLOCK_GROUP_METADATA))
1912                 return 0;
1913         if (extent_root->fs_info->avoid_sys_chunk_alloc &&
1914             (flags & BTRFS_BLOCK_GROUP_SYSTEM))
1915                 return 0;
1916
1917         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1918                                 space_info->flags);
1919         if (ret == -ENOSPC) {
1920                 space_info->full = 1;
1921                 return 0;
1922         }
1923
1924         BUG_ON(ret);
1925
1926         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1927                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1928         BUG_ON(ret);
1929         return 0;
1930 }
1931
1932 static int update_block_group(struct btrfs_trans_handle *trans,
1933                               struct btrfs_root *root,
1934                               u64 bytenr, u64 num_bytes, int alloc,
1935                               int mark_free)
1936 {
1937         struct btrfs_block_group_cache *cache;
1938         struct btrfs_fs_info *info = root->fs_info;
1939         u64 total = num_bytes;
1940         u64 old_val;
1941         u64 byte_in_group;
1942         u64 start;
1943         u64 end;
1944
1945         /* block accounting for super block */
1946         old_val = btrfs_super_bytes_used(info->super_copy);
1947         if (alloc)
1948                 old_val += num_bytes;
1949         else
1950                 old_val -= num_bytes;
1951         btrfs_set_super_bytes_used(info->super_copy, old_val);
1952
1953         /* block accounting for root item */
1954         old_val = btrfs_root_used(&root->root_item);
1955         if (alloc)
1956                 old_val += num_bytes;
1957         else
1958                 old_val -= num_bytes;
1959         btrfs_set_root_used(&root->root_item, old_val);
1960
1961         while(total) {
1962                 cache = btrfs_lookup_block_group(info, bytenr);
1963                 if (!cache) {
1964                         return -1;
1965                 }
1966                 byte_in_group = bytenr - cache->key.objectid;
1967                 WARN_ON(byte_in_group > cache->key.offset);
1968                 start = cache->key.objectid;
1969                 end = start + cache->key.offset - 1;
1970                 set_extent_bits(&info->block_group_cache, start, end,
1971                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1972
1973                 old_val = btrfs_block_group_used(&cache->item);
1974                 num_bytes = min(total, cache->key.offset - byte_in_group);
1975
1976                 if (alloc) {
1977                         old_val += num_bytes;
1978                         cache->space_info->bytes_used += num_bytes;
1979                 } else {
1980                         old_val -= num_bytes;
1981                         cache->space_info->bytes_used -= num_bytes;
1982                         if (mark_free) {
1983                                 set_extent_dirty(&info->free_space_cache,
1984                                                  bytenr, bytenr + num_bytes - 1,
1985                                                  GFP_NOFS);
1986                         }
1987                 }
1988                 btrfs_set_block_group_used(&cache->item, old_val);
1989                 total -= num_bytes;
1990                 bytenr += num_bytes;
1991         }
1992         return 0;
1993 }
1994
1995 static int update_pinned_extents(struct btrfs_root *root,
1996                                 u64 bytenr, u64 num, int pin)
1997 {
1998         u64 len;
1999         struct btrfs_block_group_cache *cache;
2000         struct btrfs_fs_info *fs_info = root->fs_info;
2001
2002         if (pin) {
2003                 set_extent_dirty(&fs_info->pinned_extents,
2004                                 bytenr, bytenr + num - 1, GFP_NOFS);
2005         } else {
2006                 clear_extent_dirty(&fs_info->pinned_extents,
2007                                 bytenr, bytenr + num - 1, GFP_NOFS);
2008         }
2009         while (num > 0) {
2010                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2011                 if (!cache) {
2012                         len = min((u64)root->sectorsize, num);
2013                         goto next;
2014                 }
2015                 WARN_ON(!cache);
2016                 len = min(num, cache->key.offset -
2017                           (bytenr - cache->key.objectid));
2018                 if (pin) {
2019                         cache->pinned += len;
2020                         cache->space_info->bytes_pinned += len;
2021                         fs_info->total_pinned += len;
2022                 } else {
2023                         cache->pinned -= len;
2024                         cache->space_info->bytes_pinned -= len;
2025                         fs_info->total_pinned -= len;
2026                 }
2027 next:
2028                 bytenr += len;
2029                 num -= len;
2030         }
2031         return 0;
2032 }
2033
2034 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2035                                struct btrfs_root *root,
2036                                struct extent_io_tree *unpin)
2037 {
2038         u64 start;
2039         u64 end;
2040         int ret;
2041         struct extent_io_tree *free_space_cache;
2042         free_space_cache = &root->fs_info->free_space_cache;
2043
2044         while(1) {
2045                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2046                                             EXTENT_DIRTY);
2047                 if (ret)
2048                         break;
2049                 update_pinned_extents(root, start, end + 1 - start, 0);
2050                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2051                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2052         }
2053         return 0;
2054 }
2055
2056 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2057 {
2058         u64 start;
2059         u64 end;
2060         int ret;
2061
2062         ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2063                                     &end, EXTENT_LOCKED);
2064         if (!ret) {
2065                 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2066                                             EXTENT_LOCKED);
2067         }
2068         return ret == 0;
2069
2070 }
2071 static int finish_current_insert(struct btrfs_trans_handle *trans,
2072                                  struct btrfs_root *extent_root)
2073 {
2074         u64 start;
2075         u64 end;
2076         u64 priv;
2077         struct btrfs_fs_info *info = extent_root->fs_info;
2078         struct pending_extent_op *extent_op;
2079         struct btrfs_key key;
2080         int ret;
2081         int skinny_metadata =
2082                 btrfs_fs_incompat(extent_root->fs_info,
2083                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2084
2085         while(1) {
2086                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2087                                             &end, EXTENT_LOCKED);
2088                 if (ret)
2089                         break;
2090
2091                 ret = get_state_private(&info->extent_ins, start, &priv);
2092                 BUG_ON(ret);
2093                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2094
2095                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2096                         key.objectid = start;
2097                         if (skinny_metadata) {
2098                                 key.offset = extent_op->level;
2099                                 key.type = BTRFS_METADATA_ITEM_KEY;
2100                         } else {
2101                                 key.offset = extent_op->num_bytes;
2102                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2103                         }
2104                         ret = alloc_reserved_tree_block(trans, extent_root,
2105                                                 extent_root->root_key.objectid,
2106                                                 trans->transid,
2107                                                 extent_op->flags,
2108                                                 &extent_op->key,
2109                                                 extent_op->level, &key);
2110                         BUG_ON(ret);
2111                 } else {
2112                         BUG_ON(1);
2113                 }
2114
2115                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2116                                   GFP_NOFS);
2117                 kfree(extent_op);
2118         }
2119         return 0;
2120 }
2121
2122 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2123                           struct btrfs_root *root,
2124                           u64 bytenr, u64 num_bytes, int is_data)
2125 {
2126         int err = 0;
2127         struct extent_buffer *buf;
2128
2129         if (is_data)
2130                 goto pinit;
2131
2132         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2133         if (!buf)
2134                 goto pinit;
2135
2136         /* we can reuse a block if it hasn't been written
2137          * and it is from this transaction.  We can't
2138          * reuse anything from the tree log root because
2139          * it has tiny sub-transactions.
2140          */
2141         if (btrfs_buffer_uptodate(buf, 0)) {
2142                 u64 header_owner = btrfs_header_owner(buf);
2143                 u64 header_transid = btrfs_header_generation(buf);
2144                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2145                     header_transid == trans->transid &&
2146                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2147                         clean_tree_block(NULL, root, buf);
2148                         free_extent_buffer(buf);
2149                         return 1;
2150                 }
2151         }
2152         free_extent_buffer(buf);
2153 pinit:
2154         update_pinned_extents(root, bytenr, num_bytes, 1);
2155
2156         BUG_ON(err < 0);
2157         return 0;
2158 }
2159
2160 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2161                        u64 bytenr, u64 num_bytes)
2162 {
2163         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2164 }
2165
2166 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2167                         u64 bytenr, u64 num_bytes)
2168 {
2169         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2170 }
2171
2172 /*
2173  * remove an extent from the root, returns 0 on success
2174  */
2175 static int __free_extent(struct btrfs_trans_handle *trans,
2176                          struct btrfs_root *root,
2177                          u64 bytenr, u64 num_bytes, u64 parent,
2178                          u64 root_objectid, u64 owner_objectid,
2179                          u64 owner_offset, int refs_to_drop)
2180 {
2181
2182         struct btrfs_key key;
2183         struct btrfs_path *path;
2184         struct btrfs_root *extent_root = root->fs_info->extent_root;
2185         struct extent_buffer *leaf;
2186         struct btrfs_extent_item *ei;
2187         struct btrfs_extent_inline_ref *iref;
2188         int ret;
2189         int is_data;
2190         int extent_slot = 0;
2191         int found_extent = 0;
2192         int num_to_del = 1;
2193         u32 item_size;
2194         u64 refs;
2195         int skinny_metadata =
2196                 btrfs_fs_incompat(extent_root->fs_info,
2197                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2198
2199         if (root->fs_info->free_extent_hook) {
2200                 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2201                                                 parent, root_objectid, owner_objectid,
2202                                                 owner_offset, refs_to_drop);
2203
2204         }
2205         path = btrfs_alloc_path();
2206         if (!path)
2207                 return -ENOMEM;
2208
2209         path->reada = 1;
2210
2211         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2212         if (is_data)
2213                 skinny_metadata = 0;
2214         BUG_ON(!is_data && refs_to_drop != 1);
2215
2216         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2217                                     bytenr, num_bytes, parent,
2218                                     root_objectid, owner_objectid,
2219                                     owner_offset);
2220         if (ret == 0) {
2221                 extent_slot = path->slots[0];
2222                 while (extent_slot >= 0) {
2223                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2224                                               extent_slot);
2225                         if (key.objectid != bytenr)
2226                                 break;
2227                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2228                             key.offset == num_bytes) {
2229                                 found_extent = 1;
2230                                 break;
2231                         }
2232                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
2233                             key.offset == owner_objectid) {
2234                                 found_extent = 1;
2235                                 break;
2236                         }
2237                         if (path->slots[0] - extent_slot > 5)
2238                                 break;
2239                         extent_slot--;
2240                 }
2241 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2242                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2243                 if (found_extent && item_size < sizeof(*ei))
2244                         found_extent = 0;
2245 #endif
2246                 if (!found_extent) {
2247                         BUG_ON(iref);
2248                         ret = remove_extent_backref(trans, extent_root, path,
2249                                                     NULL, refs_to_drop,
2250                                                     is_data);
2251                         BUG_ON(ret);
2252                         btrfs_release_path(path);
2253
2254                         key.objectid = bytenr;
2255
2256                         if (skinny_metadata) {
2257                                 key.type = BTRFS_METADATA_ITEM_KEY;
2258                                 key.offset = owner_objectid;
2259                         } else {
2260                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2261                                 key.offset = num_bytes;
2262                         }
2263
2264                         ret = btrfs_search_slot(trans, extent_root,
2265                                                 &key, path, -1, 1);
2266                         if (ret > 0 && skinny_metadata && path->slots[0]) {
2267                                 path->slots[0]--;
2268                                 btrfs_item_key_to_cpu(path->nodes[0],
2269                                                       &key,
2270                                                       path->slots[0]);
2271                                 if (key.objectid == bytenr &&
2272                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2273                                     key.offset == num_bytes)
2274                                         ret = 0;
2275                         }
2276
2277                         if (ret > 0 && skinny_metadata) {
2278                                 skinny_metadata = 0;
2279                                 btrfs_release_path(path);
2280                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2281                                 key.offset = num_bytes;
2282                                 ret = btrfs_search_slot(trans, extent_root,
2283                                                         &key, path, -1, 1);
2284                         }
2285
2286                         if (ret) {
2287                                 printk(KERN_ERR "umm, got %d back from search"
2288                                        ", was looking for %llu\n", ret,
2289                                        (unsigned long long)bytenr);
2290                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2291                         }
2292                         BUG_ON(ret);
2293                         extent_slot = path->slots[0];
2294                 }
2295         } else {
2296                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2297                        "parent %llu root %llu  owner %llu offset %llu\n",
2298                        (unsigned long long)bytenr,
2299                        (unsigned long long)parent,
2300                        (unsigned long long)root_objectid,
2301                        (unsigned long long)owner_objectid,
2302                        (unsigned long long)owner_offset);
2303                 ret = -EIO;
2304                 goto fail;
2305         }
2306
2307         leaf = path->nodes[0];
2308         item_size = btrfs_item_size_nr(leaf, extent_slot);
2309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2310         if (item_size < sizeof(*ei)) {
2311                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2312                 ret = convert_extent_item_v0(trans, extent_root, path,
2313                                              owner_objectid, 0);
2314                 BUG_ON(ret < 0);
2315
2316                 btrfs_release_path(path);
2317
2318                 key.objectid = bytenr;
2319                 key.type = BTRFS_EXTENT_ITEM_KEY;
2320                 key.offset = num_bytes;
2321
2322                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2323                                         -1, 1);
2324                 if (ret) {
2325                         printk(KERN_ERR "umm, got %d back from search"
2326                                ", was looking for %llu\n", ret,
2327                                (unsigned long long)bytenr);
2328                         btrfs_print_leaf(extent_root, path->nodes[0]);
2329                 }
2330                 BUG_ON(ret);
2331                 extent_slot = path->slots[0];
2332                 leaf = path->nodes[0];
2333                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2334         }
2335 #endif
2336         BUG_ON(item_size < sizeof(*ei));
2337         ei = btrfs_item_ptr(leaf, extent_slot,
2338                             struct btrfs_extent_item);
2339         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2340             key.type == BTRFS_EXTENT_ITEM_KEY) {
2341                 struct btrfs_tree_block_info *bi;
2342                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2343                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2344                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2345         }
2346
2347         refs = btrfs_extent_refs(leaf, ei);
2348         BUG_ON(refs < refs_to_drop);
2349         refs -= refs_to_drop;
2350
2351         if (refs > 0) {
2352                 /*
2353                  * In the case of inline back ref, reference count will
2354                  * be updated by remove_extent_backref
2355                  */
2356                 if (iref) {
2357                         BUG_ON(!found_extent);
2358                 } else {
2359                         btrfs_set_extent_refs(leaf, ei, refs);
2360                         btrfs_mark_buffer_dirty(leaf);
2361                 }
2362                 if (found_extent) {
2363                         ret = remove_extent_backref(trans, extent_root, path,
2364                                                     iref, refs_to_drop,
2365                                                     is_data);
2366                         BUG_ON(ret);
2367                 }
2368         } else {
2369                 int mark_free = 0;
2370                 int pin = 1;
2371
2372                 if (found_extent) {
2373                         BUG_ON(is_data && refs_to_drop !=
2374                                extent_data_ref_count(root, path, iref));
2375                         if (iref) {
2376                                 BUG_ON(path->slots[0] != extent_slot);
2377                         } else {
2378                                 BUG_ON(path->slots[0] != extent_slot + 1);
2379                                 path->slots[0] = extent_slot;
2380                                 num_to_del = 2;
2381                         }
2382                 }
2383
2384                 if (pin) {
2385                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2386                                              is_data);
2387                         if (ret > 0)
2388                                 mark_free = 1;
2389                         BUG_ON(ret < 0);
2390                 }
2391
2392                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2393                                       num_to_del);
2394                 BUG_ON(ret);
2395                 btrfs_release_path(path);
2396
2397                 if (is_data) {
2398                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2399                         BUG_ON(ret);
2400                 }
2401
2402                 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2403         }
2404 fail:
2405         btrfs_free_path(path);
2406         finish_current_insert(trans, extent_root);
2407         return ret;
2408 }
2409
2410 /*
2411  * find all the blocks marked as pending in the radix tree and remove
2412  * them from the extent map
2413  */
2414 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2415                                btrfs_root *extent_root)
2416 {
2417         int ret;
2418         int err = 0;
2419         u64 start;
2420         u64 end;
2421         u64 priv;
2422         struct extent_io_tree *pending_del;
2423         struct extent_io_tree *extent_ins;
2424         struct pending_extent_op *extent_op;
2425
2426         extent_ins = &extent_root->fs_info->extent_ins;
2427         pending_del = &extent_root->fs_info->pending_del;
2428
2429         while(1) {
2430                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2431                                             EXTENT_LOCKED);
2432                 if (ret)
2433                         break;
2434
2435                 ret = get_state_private(pending_del, start, &priv);
2436                 BUG_ON(ret);
2437                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2438
2439                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2440                                   GFP_NOFS);
2441
2442                 if (!test_range_bit(extent_ins, start, end,
2443                                     EXTENT_LOCKED, 0)) {
2444                         ret = __free_extent(trans, extent_root,
2445                                             start, end + 1 - start, 0,
2446                                             extent_root->root_key.objectid,
2447                                             extent_op->level, 0, 1);
2448                         kfree(extent_op);
2449                 } else {
2450                         kfree(extent_op);
2451                         ret = get_state_private(extent_ins, start, &priv);
2452                         BUG_ON(ret);
2453                         extent_op = (struct pending_extent_op *)
2454                                                         (unsigned long)priv;
2455
2456                         clear_extent_bits(extent_ins, start, end,
2457                                           EXTENT_LOCKED, GFP_NOFS);
2458
2459                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2460                                 BUG_ON(1);
2461
2462                         kfree(extent_op);
2463                 }
2464                 if (ret)
2465                         err = ret;
2466         }
2467         return err;
2468 }
2469
2470 /*
2471  * remove an extent from the root, returns 0 on success
2472  */
2473
2474 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2475                       struct btrfs_root *root,
2476                       u64 bytenr, u64 num_bytes, u64 parent,
2477                       u64 root_objectid, u64 owner, u64 offset)
2478 {
2479         struct btrfs_root *extent_root = root->fs_info->extent_root;
2480         int pending_ret;
2481         int ret;
2482
2483         WARN_ON(num_bytes < root->sectorsize);
2484         if (root == extent_root) {
2485                 struct pending_extent_op *extent_op;
2486
2487                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2488                 BUG_ON(!extent_op);
2489
2490                 extent_op->type = PENDING_EXTENT_DELETE;
2491                 extent_op->bytenr = bytenr;
2492                 extent_op->num_bytes = num_bytes;
2493                 extent_op->level = (int)owner;
2494
2495                 set_extent_bits(&root->fs_info->pending_del,
2496                                 bytenr, bytenr + num_bytes - 1,
2497                                 EXTENT_LOCKED, GFP_NOFS);
2498                 set_state_private(&root->fs_info->pending_del,
2499                                   bytenr, (unsigned long)extent_op);
2500                 return 0;
2501         }
2502         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2503                             root_objectid, owner, offset, 1);
2504         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2505         return ret ? ret : pending_ret;
2506 }
2507
2508 static u64 stripe_align(struct btrfs_root *root, u64 val)
2509 {
2510         u64 mask = ((u64)root->stripesize - 1);
2511         u64 ret = (val + mask) & ~mask;
2512         return ret;
2513 }
2514
2515 /*
2516  * walks the btree of allocated extents and find a hole of a given size.
2517  * The key ins is changed to record the hole:
2518  * ins->objectid == block start
2519  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2520  * ins->offset == number of blocks
2521  * Any available blocks before search_start are skipped.
2522  */
2523 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2524                                      struct btrfs_root *orig_root,
2525                                      u64 num_bytes, u64 empty_size,
2526                                      u64 search_start, u64 search_end,
2527                                      u64 hint_byte, struct btrfs_key *ins,
2528                                      u64 exclude_start, u64 exclude_nr,
2529                                      int data)
2530 {
2531         int ret;
2532         u64 orig_search_start = search_start;
2533         struct btrfs_root * root = orig_root->fs_info->extent_root;
2534         struct btrfs_fs_info *info = root->fs_info;
2535         u64 total_needed = num_bytes;
2536         struct btrfs_block_group_cache *block_group;
2537         int full_scan = 0;
2538         int wrapped = 0;
2539
2540         WARN_ON(num_bytes < root->sectorsize);
2541         ins->type = BTRFS_EXTENT_ITEM_KEY;
2542
2543         search_start = stripe_align(root, search_start);
2544
2545         if (hint_byte) {
2546                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2547                 if (!block_group)
2548                         hint_byte = search_start;
2549                 block_group = btrfs_find_block_group(root, block_group,
2550                                                      hint_byte, data, 1);
2551         } else {
2552                 block_group = btrfs_find_block_group(root,
2553                                                      trans->block_group,
2554                                                      search_start, data, 1);
2555         }
2556
2557         total_needed += empty_size;
2558
2559 check_failed:
2560         search_start = stripe_align(root, search_start);
2561         if (!block_group) {
2562                 block_group = btrfs_lookup_first_block_group(info,
2563                                                              search_start);
2564                 if (!block_group)
2565                         block_group = btrfs_lookup_first_block_group(info,
2566                                                        orig_search_start);
2567         }
2568         ret = find_search_start(root, &block_group, &search_start,
2569                                 total_needed, data);
2570         if (ret)
2571                 goto new_group;
2572
2573         ins->objectid = search_start;
2574         ins->offset = num_bytes;
2575
2576         if (ins->objectid + num_bytes >
2577             block_group->key.objectid + block_group->key.offset) {
2578                 search_start = block_group->key.objectid +
2579                         block_group->key.offset;
2580                 goto new_group;
2581         }
2582
2583         if (test_range_bit(&info->extent_ins, ins->objectid,
2584                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2585                 search_start = ins->objectid + num_bytes;
2586                 goto new_group;
2587         }
2588
2589         if (test_range_bit(&info->pinned_extents, ins->objectid,
2590                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2591                 search_start = ins->objectid + num_bytes;
2592                 goto new_group;
2593         }
2594
2595         if (info->excluded_extents &&
2596             test_range_bit(info->excluded_extents, ins->objectid,
2597                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2598                 search_start = ins->objectid + num_bytes;
2599                 goto new_group;
2600         }
2601
2602         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2603             ins->objectid < exclude_start + exclude_nr)) {
2604                 search_start = exclude_start + exclude_nr;
2605                 goto new_group;
2606         }
2607
2608         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2609                 if (check_crossing_stripes(ins->objectid, num_bytes)) {
2610                         search_start = round_down(ins->objectid + num_bytes,
2611                                                   BTRFS_STRIPE_LEN);
2612                         goto new_group;
2613                 }
2614                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2615                 if (block_group)
2616                         trans->block_group = block_group;
2617         }
2618         ins->offset = num_bytes;
2619         return 0;
2620
2621 new_group:
2622         block_group = btrfs_lookup_first_block_group(info, search_start);
2623         if (!block_group) {
2624                 search_start = orig_search_start;
2625                 if (full_scan) {
2626                         ret = -ENOSPC;
2627                         goto error;
2628                 }
2629                 if (wrapped) {
2630                         if (!full_scan)
2631                                 total_needed -= empty_size;
2632                         full_scan = 1;
2633                 } else
2634                         wrapped = 1;
2635         }
2636         cond_resched();
2637         block_group = btrfs_find_block_group(root, block_group,
2638                                              search_start, data, 0);
2639         goto check_failed;
2640
2641 error:
2642         return ret;
2643 }
2644
2645 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2646                          struct btrfs_root *root,
2647                          u64 num_bytes, u64 empty_size,
2648                          u64 hint_byte, u64 search_end,
2649                          struct btrfs_key *ins, int data)
2650 {
2651         int ret;
2652         u64 search_start = 0;
2653         u64 alloc_profile;
2654         struct btrfs_fs_info *info = root->fs_info;
2655
2656         if (data) {
2657                 alloc_profile = info->avail_data_alloc_bits &
2658                                 info->data_alloc_profile;
2659                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2660         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2661                    info->system_allocs >= 0) {
2662                 alloc_profile = info->avail_system_alloc_bits &
2663                                 info->system_alloc_profile;
2664                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2665         } else {
2666                 alloc_profile = info->avail_metadata_alloc_bits &
2667                                 info->metadata_alloc_profile;
2668                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2669         }
2670
2671         if (root->ref_cows) {
2672                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2673                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2674                                              num_bytes,
2675                                              BTRFS_BLOCK_GROUP_METADATA);
2676                         BUG_ON(ret);
2677                 }
2678                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2679                                      num_bytes + 2 * 1024 * 1024, data);
2680                 BUG_ON(ret);
2681         }
2682
2683         WARN_ON(num_bytes < root->sectorsize);
2684         ret = find_free_extent(trans, root, num_bytes, empty_size,
2685                                search_start, search_end, hint_byte, ins,
2686                                trans->alloc_exclude_start,
2687                                trans->alloc_exclude_nr, data);
2688         BUG_ON(ret);
2689         clear_extent_dirty(&root->fs_info->free_space_cache,
2690                            ins->objectid, ins->objectid + ins->offset - 1,
2691                            GFP_NOFS);
2692         return ret;
2693 }
2694
2695 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2696                                      struct btrfs_root *root,
2697                                      u64 root_objectid, u64 generation,
2698                                      u64 flags, struct btrfs_disk_key *key,
2699                                      int level, struct btrfs_key *ins)
2700 {
2701         int ret;
2702         struct btrfs_fs_info *fs_info = root->fs_info;
2703         struct btrfs_extent_item *extent_item;
2704         struct btrfs_tree_block_info *block_info;
2705         struct btrfs_extent_inline_ref *iref;
2706         struct btrfs_path *path;
2707         struct extent_buffer *leaf;
2708         u32 size = sizeof(*extent_item) + sizeof(*iref);
2709         int skinny_metadata =
2710                 btrfs_fs_incompat(fs_info,
2711                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2712
2713         if (!skinny_metadata)
2714                 size += sizeof(*block_info);
2715
2716         path = btrfs_alloc_path();
2717         if (!path)
2718                 return -ENOMEM;
2719
2720         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2721                                       ins, size);
2722         BUG_ON(ret);
2723
2724         leaf = path->nodes[0];
2725         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2726                                      struct btrfs_extent_item);
2727         btrfs_set_extent_refs(leaf, extent_item, 1);
2728         btrfs_set_extent_generation(leaf, extent_item, generation);
2729         btrfs_set_extent_flags(leaf, extent_item,
2730                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2731
2732         if (skinny_metadata) {
2733                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2734         } else {
2735                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2736                 btrfs_set_tree_block_key(leaf, block_info, key);
2737                 btrfs_set_tree_block_level(leaf, block_info, level);
2738                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2739         }
2740
2741         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2742         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2743
2744         btrfs_mark_buffer_dirty(leaf);
2745         btrfs_free_path(path);
2746
2747         ret = update_block_group(trans, root, ins->objectid, root->nodesize,
2748                                  1, 0);
2749         return ret;
2750 }
2751
2752 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2753                             struct btrfs_root *root, u64 num_bytes,
2754                             u64 root_objectid, u64 generation,
2755                             u64 flags, struct btrfs_disk_key *key,
2756                             int level, u64 empty_size, u64 hint_byte,
2757                             u64 search_end, struct btrfs_key *ins)
2758 {
2759         int ret;
2760         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2761                                    hint_byte, search_end, ins, 0);
2762         BUG_ON(ret);
2763
2764         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2765                 struct pending_extent_op *extent_op;
2766
2767                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2768                 BUG_ON(!extent_op);
2769
2770                 extent_op->type = PENDING_EXTENT_INSERT;
2771                 extent_op->bytenr = ins->objectid;
2772                 extent_op->num_bytes = ins->offset;
2773                 extent_op->level = level;
2774                 extent_op->flags = flags;
2775                 memcpy(&extent_op->key, key, sizeof(*key));
2776
2777                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2778                                 ins->objectid + ins->offset - 1,
2779                                 EXTENT_LOCKED, GFP_NOFS);
2780                 set_state_private(&root->fs_info->extent_ins,
2781                                   ins->objectid, (unsigned long)extent_op);
2782         } else {
2783                 if (btrfs_fs_incompat(root->fs_info,
2784                                 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
2785                         ins->offset = level;
2786                         ins->type = BTRFS_METADATA_ITEM_KEY;
2787                 }
2788                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2789                                                 generation, flags,
2790                                                 key, level, ins);
2791                 finish_current_insert(trans, root->fs_info->extent_root);
2792                 del_pending_extents(trans, root->fs_info->extent_root);
2793         }
2794         return ret;
2795 }
2796
2797 /*
2798  * helper function to allocate a block for a given tree
2799  * returns the tree buffer or NULL.
2800  */
2801 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2802                                         struct btrfs_root *root,
2803                                         u32 blocksize, u64 root_objectid,
2804                                         struct btrfs_disk_key *key, int level,
2805                                         u64 hint, u64 empty_size)
2806 {
2807         struct btrfs_key ins;
2808         int ret;
2809         struct extent_buffer *buf;
2810
2811         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2812                                trans->transid, 0, key, level,
2813                                empty_size, hint, (u64)-1, &ins);
2814         if (ret) {
2815                 BUG_ON(ret > 0);
2816                 return ERR_PTR(ret);
2817         }
2818
2819         buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid,
2820                                            blocksize);
2821         if (!buf) {
2822                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2823                                   0, root->root_key.objectid, level, 0);
2824                 BUG_ON(1);
2825                 return ERR_PTR(-ENOMEM);
2826         }
2827         btrfs_set_buffer_uptodate(buf);
2828         trans->blocks_used++;
2829
2830         return buf;
2831 }
2832
2833 #if 0
2834
2835 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2836                                   struct btrfs_root *root,
2837                                   struct extent_buffer *leaf)
2838 {
2839         u64 leaf_owner;
2840         u64 leaf_generation;
2841         struct btrfs_key key;
2842         struct btrfs_file_extent_item *fi;
2843         int i;
2844         int nritems;
2845         int ret;
2846
2847         BUG_ON(!btrfs_is_leaf(leaf));
2848         nritems = btrfs_header_nritems(leaf);
2849         leaf_owner = btrfs_header_owner(leaf);
2850         leaf_generation = btrfs_header_generation(leaf);
2851
2852         for (i = 0; i < nritems; i++) {
2853                 u64 disk_bytenr;
2854
2855                 btrfs_item_key_to_cpu(leaf, &key, i);
2856                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2857                         continue;
2858                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2859                 if (btrfs_file_extent_type(leaf, fi) ==
2860                     BTRFS_FILE_EXTENT_INLINE)
2861                         continue;
2862                 /*
2863                  * FIXME make sure to insert a trans record that
2864                  * repeats the snapshot del on crash
2865                  */
2866                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2867                 if (disk_bytenr == 0)
2868                         continue;
2869                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2870                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2871                                 leaf->start, leaf_owner, leaf_generation,
2872                                 key.objectid, 0);
2873                 BUG_ON(ret);
2874         }
2875         return 0;
2876 }
2877
2878 static void noinline reada_walk_down(struct btrfs_root *root,
2879                                      struct extent_buffer *node,
2880                                      int slot)
2881 {
2882         u64 bytenr;
2883         u64 last = 0;
2884         u32 nritems;
2885         u32 refs;
2886         u32 blocksize;
2887         int ret;
2888         int i;
2889         int level;
2890         int skipped = 0;
2891
2892         nritems = btrfs_header_nritems(node);
2893         level = btrfs_header_level(node);
2894         if (level)
2895                 return;
2896
2897         for (i = slot; i < nritems && skipped < 32; i++) {
2898                 bytenr = btrfs_node_blockptr(node, i);
2899                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2900                              (last > bytenr && last - bytenr > 32 * 1024))) {
2901                         skipped++;
2902                         continue;
2903                 }
2904                 blocksize = btrfs_level_size(root, level - 1);
2905                 if (i != slot) {
2906                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2907                                                       blocksize, &refs);
2908                         BUG_ON(ret);
2909                         if (refs != 1) {
2910                                 skipped++;
2911                                 continue;
2912                         }
2913                 }
2914                 mutex_unlock(&root->fs_info->fs_mutex);
2915                 ret = readahead_tree_block(root, bytenr, blocksize,
2916                                            btrfs_node_ptr_generation(node, i));
2917                 last = bytenr + blocksize;
2918                 cond_resched();
2919                 mutex_lock(&root->fs_info->fs_mutex);
2920                 if (ret)
2921                         break;
2922         }
2923 }
2924
2925 /*
2926  * helper function for drop_snapshot, this walks down the tree dropping ref
2927  * counts as it goes.
2928  */
2929 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2930                                    struct btrfs_root *root,
2931                                    struct btrfs_path *path, int *level)
2932 {
2933         u64 root_owner;
2934         u64 root_gen;
2935         u64 bytenr;
2936         u64 ptr_gen;
2937         struct extent_buffer *next;
2938         struct extent_buffer *cur;
2939         struct extent_buffer *parent;
2940         u32 blocksize;
2941         int ret;
2942         u32 refs;
2943
2944         WARN_ON(*level < 0);
2945         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2946         ret = btrfs_lookup_extent_ref(trans, root,
2947                                       path->nodes[*level]->start,
2948                                       path->nodes[*level]->len, &refs);
2949         BUG_ON(ret);
2950         if (refs > 1)
2951                 goto out;
2952
2953         /*
2954          * walk down to the last node level and free all the leaves
2955          */
2956         while(*level >= 0) {
2957                 WARN_ON(*level < 0);
2958                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2959                 cur = path->nodes[*level];
2960
2961                 if (btrfs_header_level(cur) != *level)
2962                         WARN_ON(1);
2963
2964                 if (path->slots[*level] >=
2965                     btrfs_header_nritems(cur))
2966                         break;
2967                 if (*level == 0) {
2968                         ret = drop_leaf_ref(trans, root, cur);
2969                         BUG_ON(ret);
2970                         break;
2971                 }
2972                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2973                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2974                 blocksize = btrfs_level_size(root, *level - 1);
2975                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2976                                               &refs);
2977                 BUG_ON(ret);
2978                 if (refs != 1) {
2979                         parent = path->nodes[*level];
2980                         root_owner = btrfs_header_owner(parent);
2981                         root_gen = btrfs_header_generation(parent);
2982                         path->slots[*level]++;
2983                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2984                                                 parent->start, root_owner,
2985                                                 root_gen, *level - 1, 1);
2986                         BUG_ON(ret);
2987                         continue;
2988                 }
2989                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2990                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2991                         free_extent_buffer(next);
2992                         reada_walk_down(root, cur, path->slots[*level]);
2993                         mutex_unlock(&root->fs_info->fs_mutex);
2994                         next = read_tree_block(root, bytenr, blocksize,
2995                                                ptr_gen);
2996                         mutex_lock(&root->fs_info->fs_mutex);
2997                         if (!extent_buffer_uptodate(next)) {
2998                                 if (IS_ERR(next))
2999                                         ret = PTR_ERR(next);
3000                                 else
3001                                         ret = -EIO;
3002                                 break;
3003                         }
3004                 }
3005                 WARN_ON(*level <= 0);
3006                 if (path->nodes[*level-1])
3007                         free_extent_buffer(path->nodes[*level-1]);
3008                 path->nodes[*level-1] = next;
3009                 *level = btrfs_header_level(next);
3010                 path->slots[*level] = 0;
3011         }
3012 out:
3013         WARN_ON(*level < 0);
3014         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3015
3016         if (path->nodes[*level] == root->node) {
3017                 root_owner = root->root_key.objectid;
3018                 parent = path->nodes[*level];
3019         } else {
3020                 parent = path->nodes[*level + 1];
3021                 root_owner = btrfs_header_owner(parent);
3022         }
3023
3024         root_gen = btrfs_header_generation(parent);
3025         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3026                                 path->nodes[*level]->len, parent->start,
3027                                 root_owner, root_gen, *level, 1);
3028         free_extent_buffer(path->nodes[*level]);
3029         path->nodes[*level] = NULL;
3030         *level += 1;
3031         BUG_ON(ret);
3032         return 0;
3033 }
3034
3035 /*
3036  * helper for dropping snapshots.  This walks back up the tree in the path
3037  * to find the first node higher up where we haven't yet gone through
3038  * all the slots
3039  */
3040 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3041                                  struct btrfs_root *root,
3042                                  struct btrfs_path *path, int *level)
3043 {
3044         u64 root_owner;
3045         u64 root_gen;
3046         struct btrfs_root_item *root_item = &root->root_item;
3047         int i;
3048         int slot;
3049         int ret;
3050
3051         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3052                 slot = path->slots[i];
3053                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3054                         struct extent_buffer *node;
3055                         struct btrfs_disk_key disk_key;
3056                         node = path->nodes[i];
3057                         path->slots[i]++;
3058                         *level = i;
3059                         WARN_ON(*level == 0);
3060                         btrfs_node_key(node, &disk_key, path->slots[i]);
3061                         memcpy(&root_item->drop_progress,
3062                                &disk_key, sizeof(disk_key));
3063                         root_item->drop_level = i;
3064                         return 0;
3065                 } else {
3066                         struct extent_buffer *parent;
3067                         if (path->nodes[*level] == root->node)
3068                                 parent = path->nodes[*level];
3069                         else
3070                                 parent = path->nodes[*level + 1];
3071
3072                         root_owner = btrfs_header_owner(parent);
3073                         root_gen = btrfs_header_generation(parent);
3074                         ret = btrfs_free_extent(trans, root,
3075                                                 path->nodes[*level]->start,
3076                                                 path->nodes[*level]->len,
3077                                                 parent->start, root_owner,
3078                                                 root_gen, *level, 1);
3079                         BUG_ON(ret);
3080                         free_extent_buffer(path->nodes[*level]);
3081                         path->nodes[*level] = NULL;
3082                         *level = i + 1;
3083                 }
3084         }
3085         return 1;
3086 }
3087
3088 #endif
3089
3090 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3091 {
3092         struct btrfs_space_info *sinfo;
3093         struct btrfs_block_group_cache *cache;
3094         u64 start;
3095         u64 end;
3096         u64 ptr;
3097         int ret;
3098
3099         while(1) {
3100                 ret = find_first_extent_bit(&info->block_group_cache, 0,
3101                                             &start, &end, (unsigned int)-1);
3102                 if (ret)
3103                         break;
3104                 ret = get_state_private(&info->block_group_cache, start, &ptr);
3105                 if (!ret) {
3106                         cache = u64_to_ptr(ptr);
3107                         if (cache->free_space_ctl) {
3108                                 btrfs_remove_free_space_cache(cache);
3109                                 kfree(cache->free_space_ctl);
3110                         }
3111                         kfree(cache);
3112                 }
3113                 clear_extent_bits(&info->block_group_cache, start,
3114                                   end, (unsigned int)-1, GFP_NOFS);
3115         }
3116         while(1) {
3117                 ret = find_first_extent_bit(&info->free_space_cache, 0,
3118                                             &start, &end, EXTENT_DIRTY);
3119                 if (ret)
3120                         break;
3121                 clear_extent_dirty(&info->free_space_cache, start,
3122                                    end, GFP_NOFS);
3123         }
3124
3125         while (!list_empty(&info->space_info)) {
3126                 sinfo = list_entry(info->space_info.next,
3127                                    struct btrfs_space_info, list);
3128                 list_del_init(&sinfo->list);
3129                 kfree(sinfo);
3130         }
3131         return 0;
3132 }
3133
3134 static int find_first_block_group(struct btrfs_root *root,
3135                 struct btrfs_path *path, struct btrfs_key *key)
3136 {
3137         int ret;
3138         struct btrfs_key found_key;
3139         struct extent_buffer *leaf;
3140         int slot;
3141
3142         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3143         if (ret < 0)
3144                 return ret;
3145         while(1) {
3146                 slot = path->slots[0];
3147                 leaf = path->nodes[0];
3148                 if (slot >= btrfs_header_nritems(leaf)) {
3149                         ret = btrfs_next_leaf(root, path);
3150                         if (ret == 0)
3151                                 continue;
3152                         if (ret < 0)
3153                                 goto error;
3154                         break;
3155                 }
3156                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3157
3158                 if (found_key.objectid >= key->objectid &&
3159                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3160                         return 0;
3161                 path->slots[0]++;
3162         }
3163         ret = -ENOENT;
3164 error:
3165         return ret;
3166 }
3167
3168 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3169                                 struct btrfs_block_group_cache *cache)
3170 {
3171         u64 bytenr;
3172         u64 *logical;
3173         int stripe_len;
3174         int i, nr, ret;
3175
3176         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3177                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3178                 cache->bytes_super += stripe_len;
3179         }
3180
3181         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3182                 bytenr = btrfs_sb_offset(i);
3183                 ret = btrfs_rmap_block(&fs_info->mapping_tree,
3184                                        cache->key.objectid, bytenr,
3185                                        0, &logical, &nr, &stripe_len);
3186                 if (ret)
3187                         return;
3188
3189                 while (nr--) {
3190                         u64 start, len;
3191
3192                         if (logical[nr] > cache->key.objectid +
3193                             cache->key.offset)
3194                                 continue;
3195
3196                         if (logical[nr] + stripe_len <= cache->key.objectid)
3197                                 continue;
3198
3199                         start = logical[nr];
3200                         if (start < cache->key.objectid) {
3201                                 start = cache->key.objectid;
3202                                 len = (logical[nr] + stripe_len) - start;
3203                         } else {
3204                                 len = min_t(u64, stripe_len,
3205                                             cache->key.objectid +
3206                                             cache->key.offset - start);
3207                         }
3208
3209                         cache->bytes_super += len;
3210                 }
3211
3212                 kfree(logical);
3213         }
3214 }
3215
3216 int btrfs_read_block_groups(struct btrfs_root *root)
3217 {
3218         struct btrfs_path *path;
3219         int ret;
3220         int bit;
3221         struct btrfs_block_group_cache *cache;
3222         struct btrfs_fs_info *info = root->fs_info;
3223         struct btrfs_space_info *space_info;
3224         struct extent_io_tree *block_group_cache;
3225         struct btrfs_key key;
3226         struct btrfs_key found_key;
3227         struct extent_buffer *leaf;
3228
3229         block_group_cache = &info->block_group_cache;
3230
3231         root = info->extent_root;
3232         key.objectid = 0;
3233         key.offset = 0;
3234         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3235         path = btrfs_alloc_path();
3236         if (!path)
3237                 return -ENOMEM;
3238
3239         while(1) {
3240                 ret = find_first_block_group(root, path, &key);
3241                 if (ret > 0) {
3242                         ret = 0;
3243                         goto error;
3244                 }
3245                 if (ret != 0) {
3246                         goto error;
3247                 }
3248                 leaf = path->nodes[0];
3249                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3250                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3251                 if (!cache) {
3252                         ret = -ENOMEM;
3253                         goto error;
3254                 }
3255
3256                 read_extent_buffer(leaf, &cache->item,
3257                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3258                                    sizeof(cache->item));
3259                 memcpy(&cache->key, &found_key, sizeof(found_key));
3260                 cache->cached = 0;
3261                 cache->pinned = 0;
3262                 key.objectid = found_key.objectid + found_key.offset;
3263                 if (found_key.offset == 0)
3264                         key.objectid++;
3265                 btrfs_release_path(path);
3266                 cache->flags = btrfs_block_group_flags(&cache->item);
3267                 bit = 0;
3268                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3269                         bit = BLOCK_GROUP_DATA;
3270                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3271                         bit = BLOCK_GROUP_SYSTEM;
3272                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3273                         bit = BLOCK_GROUP_METADATA;
3274                 }
3275                 set_avail_alloc_bits(info, cache->flags);
3276                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3277                         cache->ro = 1;
3278
3279                 account_super_bytes(info, cache);
3280
3281                 ret = update_space_info(info, cache->flags, found_key.offset,
3282                                         btrfs_block_group_used(&cache->item),
3283                                         &space_info);
3284                 BUG_ON(ret);
3285                 cache->space_info = space_info;
3286
3287                 /* use EXTENT_LOCKED to prevent merging */
3288                 set_extent_bits(block_group_cache, found_key.objectid,
3289                                 found_key.objectid + found_key.offset - 1,
3290                                 bit | EXTENT_LOCKED, GFP_NOFS);
3291                 set_state_private(block_group_cache, found_key.objectid,
3292                                   (unsigned long)cache);
3293         }
3294         ret = 0;
3295 error:
3296         btrfs_free_path(path);
3297         return ret;
3298 }
3299
3300 struct btrfs_block_group_cache *
3301 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3302                       u64 chunk_objectid, u64 chunk_offset, u64 size)
3303 {
3304         int ret;
3305         int bit = 0;
3306         struct btrfs_block_group_cache *cache;
3307         struct extent_io_tree *block_group_cache;
3308
3309         block_group_cache = &fs_info->block_group_cache;
3310
3311         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3312         BUG_ON(!cache);
3313         cache->key.objectid = chunk_offset;
3314         cache->key.offset = size;
3315
3316         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3317         btrfs_set_block_group_used(&cache->item, bytes_used);
3318         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3319         cache->flags = type;
3320         btrfs_set_block_group_flags(&cache->item, type);
3321
3322         account_super_bytes(fs_info, cache);
3323         ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3324                                 &cache->space_info);
3325         BUG_ON(ret);
3326
3327         bit = block_group_state_bits(type);
3328         ret = set_extent_bits(block_group_cache, chunk_offset,
3329                               chunk_offset + size - 1,
3330                               bit | EXTENT_LOCKED, GFP_NOFS);
3331         BUG_ON(ret);
3332
3333         ret = set_state_private(block_group_cache, chunk_offset,
3334                                 (unsigned long)cache);
3335         BUG_ON(ret);
3336         set_avail_alloc_bits(fs_info, type);
3337
3338         return cache;
3339 }
3340
3341 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3342                            struct btrfs_root *root, u64 bytes_used,
3343                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3344                            u64 size)
3345 {
3346         int ret;
3347         struct btrfs_root *extent_root;
3348         struct btrfs_block_group_cache *cache;
3349
3350         cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3351                                       chunk_objectid, chunk_offset, size);
3352         extent_root = root->fs_info->extent_root;
3353         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3354                                 sizeof(cache->item));
3355         BUG_ON(ret);
3356
3357         ret = finish_current_insert(trans, extent_root);
3358         BUG_ON(ret);
3359         ret = del_pending_extents(trans, extent_root);
3360         BUG_ON(ret);
3361
3362         return 0;
3363 }
3364
3365 /*
3366  * This is for converter use only.
3367  *
3368  * In that case, we don't know where are free blocks located.
3369  * Therefore all block group cache entries must be setup properly
3370  * before doing any block allocation.
3371  */
3372 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3373                             struct btrfs_root *root)
3374 {
3375         u64 total_bytes;
3376         u64 cur_start;
3377         u64 group_type;
3378         u64 group_size;
3379         u64 group_align;
3380         u64 total_data = 0;
3381         u64 total_metadata = 0;
3382         u64 chunk_objectid;
3383         int ret;
3384         int bit;
3385         struct btrfs_root *extent_root;
3386         struct btrfs_block_group_cache *cache;
3387         struct extent_io_tree *block_group_cache;
3388
3389         extent_root = root->fs_info->extent_root;
3390         block_group_cache = &root->fs_info->block_group_cache;
3391         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3392         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3393         group_align = 64 * root->sectorsize;
3394
3395         cur_start = 0;
3396         while (cur_start < total_bytes) {
3397                 group_size = total_bytes / 12;
3398                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3399                 if (cur_start == 0) {
3400                         bit = BLOCK_GROUP_SYSTEM;
3401                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3402                         group_size /= 4;
3403                         group_size &= ~(group_align - 1);
3404                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3405                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3406                 } else {
3407                         group_size &= ~(group_align - 1);
3408                         if (total_data >= total_metadata * 2) {
3409                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3410                                 group_size = min_t(u64, group_size,
3411                                                    1ULL * 1024 * 1024 * 1024);
3412                                 total_metadata += group_size;
3413                         } else {
3414                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3415                                 group_size = min_t(u64, group_size,
3416                                                    5ULL * 1024 * 1024 * 1024);
3417                                 total_data += group_size;
3418                         }
3419                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3420                                 group_size = total_bytes - cur_start;
3421                 }
3422
3423                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3424                 BUG_ON(!cache);
3425
3426                 cache->key.objectid = cur_start;
3427                 cache->key.offset = group_size;
3428                 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3429
3430                 btrfs_set_block_group_used(&cache->item, 0);
3431                 btrfs_set_block_group_chunk_objectid(&cache->item,
3432                                                      chunk_objectid);
3433                 btrfs_set_block_group_flags(&cache->item, group_type);
3434
3435                 cache->flags = group_type;
3436
3437                 ret = update_space_info(root->fs_info, group_type, group_size,
3438                                         0, &cache->space_info);
3439                 BUG_ON(ret);
3440                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3441
3442                 set_extent_bits(block_group_cache, cur_start,
3443                                 cur_start + group_size - 1,
3444                                 bit | EXTENT_LOCKED, GFP_NOFS);
3445                 set_state_private(block_group_cache, cur_start,
3446                                   (unsigned long)cache);
3447                 cur_start += group_size;
3448         }
3449         /* then insert all the items */
3450         cur_start = 0;
3451         while(cur_start < total_bytes) {
3452                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3453                 BUG_ON(!cache);
3454
3455                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3456                                         sizeof(cache->item));
3457                 BUG_ON(ret);
3458
3459                 finish_current_insert(trans, extent_root);
3460                 ret = del_pending_extents(trans, extent_root);
3461                 BUG_ON(ret);
3462
3463                 cur_start = cache->key.objectid + cache->key.offset;
3464         }
3465         return 0;
3466 }
3467
3468 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3469                              struct btrfs_root *root,
3470                              u64 bytenr, u64 num_bytes, int alloc,
3471                              int mark_free)
3472 {
3473         return update_block_group(trans, root, bytenr, num_bytes,
3474                                   alloc, mark_free);
3475 }
3476
3477 /*
3478  * Just remove a block group item in extent tree
3479  * Caller should ensure the block group is empty and all space is pinned.
3480  * Or new tree block/data may be allocated into it.
3481  */
3482 static int free_block_group_item(struct btrfs_trans_handle *trans,
3483                                  struct btrfs_fs_info *fs_info,
3484                                  u64 bytenr, u64 len)
3485 {
3486         struct btrfs_path *path;
3487         struct btrfs_key key;
3488         struct btrfs_root *root = fs_info->extent_root;
3489         int ret = 0;
3490
3491         key.objectid = bytenr;
3492         key.offset = len;
3493         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3494
3495         path = btrfs_alloc_path();
3496         if (!path)
3497                 return -ENOMEM;
3498
3499         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3500         if (ret > 0) {
3501                 ret = -ENOENT;
3502                 goto out;
3503         }
3504         if (ret < 0)
3505                 goto out;
3506
3507         ret = btrfs_del_item(trans, root, path);
3508 out:
3509         btrfs_free_path(path);
3510         return ret;
3511 }
3512
3513 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3514                                 struct btrfs_fs_info *fs_info,
3515                                 u64 devid, u64 dev_offset)
3516 {
3517         struct btrfs_root *root = fs_info->dev_root;
3518         struct btrfs_path *path;
3519         struct btrfs_key key;
3520         int ret;
3521
3522         path = btrfs_alloc_path();
3523         if (!path)
3524                 return -ENOMEM;
3525
3526         key.objectid = devid;
3527         key.type = BTRFS_DEV_EXTENT_KEY;
3528         key.offset = dev_offset;
3529
3530         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3531         if (ret < 0)
3532                 goto out;
3533         if (ret > 0) {
3534                 ret = -ENOENT;
3535                 goto out;
3536         }
3537
3538         ret = btrfs_del_item(trans, root, path);
3539 out:
3540         btrfs_free_path(path);
3541         return ret;
3542 }
3543
3544 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3545                                        struct btrfs_fs_info *fs_info,
3546                                        u64 chunk_offset)
3547 {
3548         struct btrfs_chunk *chunk = NULL;
3549         struct btrfs_root *root= fs_info->chunk_root;
3550         struct btrfs_path *path;
3551         struct btrfs_key key;
3552         u16 num_stripes;
3553         int i;
3554         int ret;
3555
3556         path = btrfs_alloc_path();
3557         if (!path)
3558                 return -ENOMEM;
3559
3560         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3561         key.type = BTRFS_CHUNK_ITEM_KEY;
3562         key.offset = chunk_offset;
3563
3564         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3565         if (ret < 0)
3566                 goto out;
3567         if (ret > 0) {
3568                 ret = -ENOENT;
3569                 goto out;
3570         }
3571         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3572                                struct btrfs_chunk);
3573         num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3574         for (i = 0; i < num_stripes; i++) {
3575                 ret = free_dev_extent_item(trans, fs_info,
3576                         btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3577                         btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3578                 if (ret < 0)
3579                         goto out;
3580         }
3581 out:
3582         btrfs_free_path(path);
3583         return ret;
3584 }
3585
3586 static int free_system_chunk_item(struct btrfs_super_block *super,
3587                                   struct btrfs_key *key)
3588 {
3589         struct btrfs_disk_key *disk_key;
3590         struct btrfs_key cpu_key;
3591         u32 array_size = btrfs_super_sys_array_size(super);
3592         char *ptr = (char *)super->sys_chunk_array;
3593         int cur = 0;
3594         int ret = -ENOENT;
3595
3596         while (cur < btrfs_super_sys_array_size(super)) {
3597                 struct btrfs_chunk *chunk;
3598                 u32 num_stripes;
3599                 u32 chunk_len;
3600
3601                 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3602                 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3603                 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3604                         /* just in case */
3605                         ret = -EIO;
3606                         goto out;
3607                 }
3608
3609                 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3610                 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3611                 chunk_len = btrfs_chunk_item_size(num_stripes) +
3612                             sizeof(*disk_key);
3613
3614                 if (key->objectid == cpu_key.objectid &&
3615                     key->offset == cpu_key.offset &&
3616                     key->type == cpu_key.type) {
3617                         memmove(ptr + cur, ptr + cur + chunk_len,
3618                                 array_size - cur - chunk_len);
3619                         array_size -= chunk_len;
3620                         btrfs_set_super_sys_array_size(super, array_size);
3621                         ret = 0;
3622                         goto out;
3623                 }
3624
3625                 cur += chunk_len;
3626         }
3627 out:
3628         return ret;
3629 }
3630
3631 static int free_chunk_item(struct btrfs_trans_handle *trans,
3632                            struct btrfs_fs_info *fs_info,
3633                            u64 bytenr, u64 len)
3634 {
3635         struct btrfs_path *path;
3636         struct btrfs_key key;
3637         struct btrfs_root *root = fs_info->chunk_root;
3638         struct btrfs_chunk *chunk;
3639         u64 chunk_type;
3640         int ret;
3641
3642         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3643         key.offset = bytenr;
3644         key.type = BTRFS_CHUNK_ITEM_KEY;
3645
3646         path = btrfs_alloc_path();
3647         if (!path)
3648                 return -ENOMEM;
3649
3650         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3651         if (ret > 0) {
3652                 ret = -ENOENT;
3653                 goto out;
3654         }
3655         if (ret < 0)
3656                 goto out;
3657         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3658                                struct btrfs_chunk);
3659         chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3660
3661         ret = btrfs_del_item(trans, root, path);
3662         if (ret < 0)
3663                 goto out;
3664
3665         if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3666                 ret = free_system_chunk_item(fs_info->super_copy, &key);
3667 out:
3668         btrfs_free_path(path);
3669         return ret;
3670 }
3671
3672 static u64 get_dev_extent_len(struct map_lookup *map)
3673 {
3674         int div;
3675
3676         switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3677         case 0: /* Single */
3678         case BTRFS_BLOCK_GROUP_DUP:
3679         case BTRFS_BLOCK_GROUP_RAID1:
3680                 div = 1;
3681                 break;
3682         case BTRFS_BLOCK_GROUP_RAID5:
3683                 div = (map->num_stripes - 1);
3684                 break;
3685         case BTRFS_BLOCK_GROUP_RAID6:
3686                 div = (map->num_stripes - 2);
3687                 break;
3688         case BTRFS_BLOCK_GROUP_RAID10:
3689                 div = (map->num_stripes / map->sub_stripes);
3690                 break;
3691         default:
3692                 /* normally, read chunk security hook should handled it */
3693                 BUG_ON(1);
3694         }
3695         return map->ce.size / div;
3696 }
3697
3698 /* free block group/chunk related caches */
3699 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3700                                   struct btrfs_fs_info *fs_info,
3701                                   u64 bytenr, u64 len)
3702 {
3703         struct btrfs_block_group_cache *cache;
3704         struct cache_extent *ce;
3705         struct map_lookup *map;
3706         int ret;
3707         int i;
3708         u64 flags;
3709
3710         /* Free block group cache first */
3711         cache = btrfs_lookup_block_group(fs_info, bytenr);
3712         if (!cache)
3713                 return -ENOENT;
3714         flags = cache->flags;
3715         if (cache->free_space_ctl) {
3716                 btrfs_remove_free_space_cache(cache);
3717                 kfree(cache->free_space_ctl);
3718         }
3719         clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len,
3720                           (unsigned int)-1, GFP_NOFS);
3721         ret = free_space_info(fs_info, flags, len, 0, NULL);
3722         if (ret < 0)
3723                 goto out;
3724         kfree(cache);
3725
3726         /* Then free mapping info and dev usage info */
3727         ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3728         if (!ce || ce->start != bytenr) {
3729                 ret = -ENOENT;
3730                 goto out;
3731         }
3732         map = container_of(ce, struct map_lookup, ce);
3733         for (i = 0; i < map->num_stripes; i++) {
3734                 struct btrfs_device *device;
3735
3736                 device = map->stripes[i].dev;
3737                 device->bytes_used -= get_dev_extent_len(map);
3738                 ret = btrfs_update_device(trans, device);
3739                 if (ret < 0)
3740                         goto out;
3741         }
3742         remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3743         free(map);
3744 out:
3745         return ret;
3746 }
3747
3748 int btrfs_free_block_group(struct btrfs_trans_handle *trans,
3749                            struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
3750 {
3751         struct btrfs_root *extent_root = fs_info->extent_root;
3752         struct btrfs_path *path;
3753         struct btrfs_block_group_item *bgi;
3754         struct btrfs_key key;
3755         int ret = 0;
3756
3757         path = btrfs_alloc_path();
3758         if (!path)
3759                 return -ENOMEM;
3760
3761         key.objectid = bytenr;
3762         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3763         key.offset = len;
3764
3765         /* Double check the block group to ensure it's empty */
3766         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3767         if (ret > 0) {
3768                 ret = -ENONET;
3769                 goto out;
3770         }
3771         if (ret < 0)
3772                 goto out;
3773
3774         bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3775                              struct btrfs_block_group_item);
3776         if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
3777                 fprintf(stderr,
3778                         "WARNING: block group [%llu,%llu) is not empty\n",
3779                         bytenr, bytenr + len);
3780                 ret = -EINVAL;
3781                 goto out;
3782         }
3783         btrfs_release_path(path);
3784
3785         /*
3786          * Now pin all space in the block group, to prevent further transaction
3787          * allocate space from it.
3788          * Every operation needs a transaction must be in the range.
3789          */
3790         btrfs_pin_extent(fs_info, bytenr, len);
3791
3792         /* delete block group item and chunk item */
3793         ret = free_block_group_item(trans, fs_info, bytenr, len);
3794         if (ret < 0) {
3795                 fprintf(stderr,
3796                         "failed to free block group item for [%llu,%llu)\n",
3797                         bytenr, bytenr + len);
3798                 btrfs_unpin_extent(fs_info, bytenr, len);
3799                 goto out;
3800         }
3801
3802         ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
3803         if (ret < 0) {
3804                 fprintf(stderr,
3805                         "failed to dev extents belongs to [%llu,%llu)\n",
3806                         bytenr, bytenr + len);
3807                 btrfs_unpin_extent(fs_info, bytenr, len);
3808                 goto out;
3809         }
3810         ret = free_chunk_item(trans, fs_info, bytenr, len);
3811         if (ret < 0) {
3812                 fprintf(stderr,
3813                         "failed to free chunk for [%llu,%llu)\n",
3814                         bytenr, bytenr + len);
3815                 btrfs_unpin_extent(fs_info, bytenr, len);
3816                 goto out;
3817         }
3818
3819         /* Now release the block_group_cache */
3820         ret = free_block_group_cache(trans, fs_info, bytenr, len);
3821         btrfs_unpin_extent(fs_info, bytenr, len);
3822
3823 out:
3824         btrfs_free_path(path);
3825         return ret;
3826 }
3827
3828 /*
3829  * Fixup block accounting. The initial block accounting created by
3830  * make_block_groups isn't accuracy in this case.
3831  */
3832 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3833                                struct btrfs_root *root)
3834 {
3835         int ret;
3836         int slot;
3837         u64 start = 0;
3838         u64 bytes_used = 0;
3839         struct btrfs_path path;
3840         struct btrfs_key key;
3841         struct extent_buffer *leaf;
3842         struct btrfs_block_group_cache *cache;
3843         struct btrfs_fs_info *fs_info = root->fs_info;
3844
3845         root = root->fs_info->extent_root;
3846
3847         while(extent_root_pending_ops(fs_info)) {
3848                 ret = finish_current_insert(trans, root);
3849                 if (ret)
3850                         return ret;
3851                 ret = del_pending_extents(trans, root);
3852                 if (ret)
3853                         return ret;
3854         }
3855
3856         while(1) {
3857                 cache = btrfs_lookup_first_block_group(fs_info, start);
3858                 if (!cache)
3859                         break;
3860                 start = cache->key.objectid + cache->key.offset;
3861                 btrfs_set_block_group_used(&cache->item, 0);
3862                 cache->space_info->bytes_used = 0;
3863                 set_extent_bits(&root->fs_info->block_group_cache,
3864                                 cache->key.objectid,
3865                                 cache->key.objectid + cache->key.offset -1,
3866                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
3867         }
3868
3869         btrfs_init_path(&path);
3870         key.offset = 0;
3871         key.objectid = 0;
3872         key.type = BTRFS_EXTENT_ITEM_KEY;
3873         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3874                                 &key, &path, 0, 0);
3875         if (ret < 0)
3876                 return ret;
3877         while(1) {
3878                 leaf = path.nodes[0];
3879                 slot = path.slots[0];
3880                 if (slot >= btrfs_header_nritems(leaf)) {
3881                         ret = btrfs_next_leaf(root, &path);
3882                         if (ret < 0)
3883                                 return ret;
3884                         if (ret > 0)
3885                                 break;
3886                         leaf = path.nodes[0];
3887                         slot = path.slots[0];
3888                 }
3889                 btrfs_item_key_to_cpu(leaf, &key, slot);
3890                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3891                         bytes_used += key.offset;
3892                         ret = btrfs_update_block_group(trans, root,
3893                                   key.objectid, key.offset, 1, 0);
3894                         BUG_ON(ret);
3895                 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3896                         bytes_used += root->nodesize;
3897                         ret = btrfs_update_block_group(trans, root,
3898                                   key.objectid, root->nodesize, 1, 0);
3899                         BUG_ON(ret);
3900                 }
3901                 path.slots[0]++;
3902         }
3903         btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3904         btrfs_release_path(&path);
3905         return 0;
3906 }
3907
3908 static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
3909                               u64 *start, u64 *len)
3910 {
3911         struct btrfs_key key;
3912
3913         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3914         BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
3915                  key.type == BTRFS_METADATA_ITEM_KEY));
3916         *start = key.objectid;
3917         if (key.type == BTRFS_EXTENT_ITEM_KEY)
3918                 *len = key.offset;
3919         else
3920                 *len = root->nodesize;
3921 }
3922
3923 /*
3924  * Find first overlap extent for range [bytenr, bytenr + len)
3925  * Return 0 for found and point path to it.
3926  * Return >0 for not found.
3927  * Return <0 for err
3928  */
3929 int btrfs_search_overlap_extent(struct btrfs_root *root,
3930                                 struct btrfs_path *path, u64 bytenr, u64 len)
3931 {
3932         struct btrfs_key key;
3933         u64 cur_start;
3934         u64 cur_len;
3935         int ret;
3936
3937         key.objectid = bytenr;
3938         key.type = BTRFS_EXTENT_DATA_KEY;
3939         key.offset = (u64)-1;
3940
3941         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3942         if (ret < 0)
3943                 return ret;
3944         BUG_ON(ret == 0);
3945
3946         ret = btrfs_previous_extent_item(root, path, 0);
3947         if (ret < 0)
3948                 return ret;
3949         /* no previous, check next extent */
3950         if (ret > 0)
3951                 goto next;
3952         __get_extent_size(root, path, &cur_start, &cur_len);
3953         /* Tail overlap */
3954         if (cur_start + cur_len > bytenr)
3955                 return 1;
3956
3957 next:
3958         ret = btrfs_next_extent_item(root, path, bytenr + len);
3959         if (ret < 0)
3960                 return ret;
3961         /* No next, prev already checked, no overlap */
3962         if (ret > 0)
3963                 return 0;
3964         __get_extent_size(root, path, &cur_start, &cur_len);
3965         /* head overlap*/
3966         if (cur_start < bytenr + len)
3967                 return 1;
3968         return 0;
3969 }
3970
3971 static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3972                                       struct btrfs_root *root, u64 objectid,
3973                                       struct btrfs_inode_item *inode,
3974                                       u64 file_pos, u64 disk_bytenr,
3975                                       u64 *ret_num_bytes)
3976 {
3977         int ret;
3978         struct btrfs_fs_info *info = root->fs_info;
3979         struct btrfs_root *extent_root = info->extent_root;
3980         struct extent_buffer *leaf;
3981         struct btrfs_file_extent_item *fi;
3982         struct btrfs_key ins_key;
3983         struct btrfs_path *path;
3984         struct btrfs_extent_item *ei;
3985         u64 nbytes;
3986         u64 extent_num_bytes;
3987         u64 extent_bytenr;
3988         u64 extent_offset;
3989         u64 num_bytes = *ret_num_bytes;
3990
3991         /*
3992          * All supported file system should not use its 0 extent.
3993          * As it's for hole
3994          *
3995          * And hole extent has no size limit, no need to loop.
3996          */
3997         if (disk_bytenr == 0) {
3998                 ret = btrfs_insert_file_extent(trans, root, objectid,
3999                                                 file_pos, disk_bytenr,
4000                                                 num_bytes, num_bytes);
4001                 return ret;
4002         }
4003         num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
4004
4005         path = btrfs_alloc_path();
4006         if (!path)
4007                 return -ENOMEM;
4008
4009         /* First to check extent overlap */
4010         ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
4011                                           num_bytes);
4012         if (ret < 0)
4013                 goto fail;
4014         if (ret > 0) {
4015                 /* Found overlap */
4016                 u64 cur_start;
4017                 u64 cur_len;
4018
4019                 __get_extent_size(extent_root, path, &cur_start, &cur_len);
4020                 /*
4021                  * For convert case, this extent should be a subset of
4022                  * existing one.
4023                  */
4024                 BUG_ON(disk_bytenr < cur_start);
4025
4026                 extent_bytenr = cur_start;
4027                 extent_num_bytes = cur_len;
4028                 extent_offset = disk_bytenr - extent_bytenr;
4029         } else {
4030                 /* No overlap, create new extent */
4031                 btrfs_release_path(path);
4032                 ins_key.objectid = disk_bytenr;
4033                 ins_key.offset = num_bytes;
4034                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4035
4036                 ret = btrfs_insert_empty_item(trans, extent_root, path,
4037                                               &ins_key, sizeof(*ei));
4038                 if (ret == 0) {
4039                         leaf = path->nodes[0];
4040                         ei = btrfs_item_ptr(leaf, path->slots[0],
4041                                             struct btrfs_extent_item);
4042
4043                         btrfs_set_extent_refs(leaf, ei, 0);
4044                         btrfs_set_extent_generation(leaf, ei, 0);
4045                         btrfs_set_extent_flags(leaf, ei,
4046                                                BTRFS_EXTENT_FLAG_DATA);
4047                         btrfs_mark_buffer_dirty(leaf);
4048
4049                         ret = btrfs_update_block_group(trans, root, disk_bytenr,
4050                                                        num_bytes, 1, 0);
4051                         if (ret)
4052                                 goto fail;
4053                 } else if (ret != -EEXIST) {
4054                         goto fail;
4055                 }
4056                 btrfs_extent_post_op(trans, extent_root);
4057                 extent_bytenr = disk_bytenr;
4058                 extent_num_bytes = num_bytes;
4059                 extent_offset = 0;
4060         }
4061         btrfs_release_path(path);
4062         ins_key.objectid = objectid;
4063         ins_key.offset = file_pos;
4064         ins_key.type = BTRFS_EXTENT_DATA_KEY;
4065         ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
4066                                       sizeof(*fi));
4067         if (ret)
4068                 goto fail;
4069         leaf = path->nodes[0];
4070         fi = btrfs_item_ptr(leaf, path->slots[0],
4071                             struct btrfs_file_extent_item);
4072         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4073         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
4074         btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
4075         btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
4076         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
4077         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
4078         btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
4079         btrfs_set_file_extent_compression(leaf, fi, 0);
4080         btrfs_set_file_extent_encryption(leaf, fi, 0);
4081         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
4082         btrfs_mark_buffer_dirty(leaf);
4083
4084         nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
4085         btrfs_set_stack_inode_nbytes(inode, nbytes);
4086         btrfs_release_path(path);
4087
4088         ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
4089                                    0, root->root_key.objectid, objectid,
4090                                    file_pos - extent_offset);
4091         if (ret)
4092                 goto fail;
4093         ret = 0;
4094         *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
4095 fail:
4096         btrfs_free_path(path);
4097         return ret;
4098 }
4099
4100 /*
4101  * Record a file extent. Do all the required works, such as inserting
4102  * file extent item, inserting extent item and backref item into extent
4103  * tree and updating block accounting.
4104  */
4105 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
4106                               struct btrfs_root *root, u64 objectid,
4107                               struct btrfs_inode_item *inode,
4108                               u64 file_pos, u64 disk_bytenr,
4109                               u64 num_bytes)
4110 {
4111         u64 cur_disk_bytenr = disk_bytenr;
4112         u64 cur_file_pos = file_pos;
4113         u64 cur_num_bytes = num_bytes;
4114         int ret = 0;
4115
4116         while (num_bytes > 0) {
4117                 ret = __btrfs_record_file_extent(trans, root, objectid,
4118                                                  inode, cur_file_pos,
4119                                                  cur_disk_bytenr,
4120                                                  &cur_num_bytes);
4121                 if (ret < 0)
4122                         break;
4123                 cur_disk_bytenr += cur_num_bytes;
4124                 cur_file_pos += cur_num_bytes;
4125                 num_bytes -= cur_num_bytes;
4126         }
4127         return ret;
4128 }
4129
4130
4131 static int add_excluded_extent(struct btrfs_root *root,
4132                                u64 start, u64 num_bytes)
4133 {
4134         u64 end = start + num_bytes - 1;
4135         set_extent_bits(&root->fs_info->pinned_extents,
4136                         start, end, EXTENT_UPTODATE, GFP_NOFS);
4137         return 0;
4138 }
4139
4140 void free_excluded_extents(struct btrfs_root *root,
4141                            struct btrfs_block_group_cache *cache)
4142 {
4143         u64 start, end;
4144
4145         start = cache->key.objectid;
4146         end = start + cache->key.offset - 1;
4147
4148         clear_extent_bits(&root->fs_info->pinned_extents,
4149                           start, end, EXTENT_UPTODATE, GFP_NOFS);
4150 }
4151
4152 int exclude_super_stripes(struct btrfs_root *root,
4153                           struct btrfs_block_group_cache *cache)
4154 {
4155         u64 bytenr;
4156         u64 *logical;
4157         int stripe_len;
4158         int i, nr, ret;
4159
4160         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
4161                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
4162                 cache->bytes_super += stripe_len;
4163                 ret = add_excluded_extent(root, cache->key.objectid,
4164                                           stripe_len);
4165                 if (ret)
4166                         return ret;
4167         }
4168
4169         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4170                 bytenr = btrfs_sb_offset(i);
4171                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4172                                        cache->key.objectid, bytenr,
4173                                        0, &logical, &nr, &stripe_len);
4174                 if (ret)
4175                         return ret;
4176
4177                 while (nr--) {
4178                         u64 start, len;
4179
4180                         if (logical[nr] > cache->key.objectid +
4181                             cache->key.offset)
4182                                 continue;
4183
4184                         if (logical[nr] + stripe_len <= cache->key.objectid)
4185                                 continue;
4186
4187                         start = logical[nr];
4188                         if (start < cache->key.objectid) {
4189                                 start = cache->key.objectid;
4190                                 len = (logical[nr] + stripe_len) - start;
4191                         } else {
4192                                 len = min_t(u64, stripe_len,
4193                                             cache->key.objectid +
4194                                             cache->key.offset - start);
4195                         }
4196
4197                         cache->bytes_super += len;
4198                         ret = add_excluded_extent(root, start, len);
4199                         if (ret) {
4200                                 kfree(logical);
4201                                 return ret;
4202                         }
4203                 }
4204
4205                 kfree(logical);
4206         }
4207         return 0;
4208 }
4209
4210 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
4211                        struct btrfs_fs_info *info, u64 start, u64 end)
4212 {
4213         u64 extent_start, extent_end, size, total_added = 0;
4214         int ret;
4215
4216         while (start < end) {
4217                 ret = find_first_extent_bit(&info->pinned_extents, start,
4218                                             &extent_start, &extent_end,
4219                                             EXTENT_DIRTY | EXTENT_UPTODATE);
4220                 if (ret)
4221                         break;
4222
4223                 if (extent_start <= start) {
4224                         start = extent_end + 1;
4225                 } else if (extent_start > start && extent_start < end) {
4226                         size = extent_start - start;
4227                         total_added += size;
4228                         ret = btrfs_add_free_space(block_group->free_space_ctl,
4229                                                    start, size);
4230                         BUG_ON(ret); /* -ENOMEM or logic error */
4231                         start = extent_end + 1;
4232                 } else {
4233                         break;
4234                 }
4235         }
4236
4237         if (start < end) {
4238                 size = end - start;
4239                 total_added += size;
4240                 ret = btrfs_add_free_space(block_group->free_space_ctl, start,
4241                                            size);
4242                 BUG_ON(ret); /* -ENOMEM or logic error */
4243         }
4244
4245         return total_added;
4246 }