extent-tree.c

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