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                 btrfs_print_leaf(root, path->nodes[0]);
1045                 btrfs_free_path(path);
1046                 return -ENOENT;
1047         }
1048
1049         BUG_ON(ret);
1050
1051         leaf = path->nodes[0];
1052         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1054         if (item_size < sizeof(*ei)) {
1055                 if (!insert) {
1056                         err = -ENOENT;
1057                         goto out;
1058                 }
1059                 ret = convert_extent_item_v0(trans, root, path, owner,
1060                                              extra_size);
1061                 if (ret < 0) {
1062                         err = ret;
1063                         goto out;
1064                 }
1065                 leaf = path->nodes[0];
1066                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1067         }
1068 #endif
1069         if (item_size < sizeof(*ei)) {
1070                 printf("Size is %u, needs to be %u, slot %d\n", item_size,
1071                        sizeof(*ei), path->slots[0]);
1072                 btrfs_print_leaf(root, leaf);
1073                 return -EINVAL;
1074         }
1075         BUG_ON(item_size < sizeof(*ei));
1076
1077         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1078         flags = btrfs_extent_flags(leaf, ei);
1079
1080         ptr = (unsigned long)(ei + 1);
1081         end = (unsigned long)ei + item_size;
1082
1083         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1084                 ptr += sizeof(struct btrfs_tree_block_info);
1085                 BUG_ON(ptr > end);
1086         } else {
1087                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1088         }
1089
1090         err = -ENOENT;
1091         while (1) {
1092                 if (ptr >= end) {
1093                         WARN_ON(ptr > end);
1094                         break;
1095                 }
1096                 iref = (struct btrfs_extent_inline_ref *)ptr;
1097                 type = btrfs_extent_inline_ref_type(leaf, iref);
1098                 if (want < type)
1099                         break;
1100                 if (want > type) {
1101                         ptr += btrfs_extent_inline_ref_size(type);
1102                         continue;
1103                 }
1104
1105                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1106                         struct btrfs_extent_data_ref *dref;
1107                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1108                         if (match_extent_data_ref(leaf, dref, root_objectid,
1109                                                   owner, offset)) {
1110                                 err = 0;
1111                                 break;
1112                         }
1113                         if (hash_extent_data_ref_item(leaf, dref) <
1114                             hash_extent_data_ref(root_objectid, owner, offset))
1115                                 break;
1116                 } else {
1117                         u64 ref_offset;
1118                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1119                         if (parent > 0) {
1120                                 if (parent == ref_offset) {
1121                                         err = 0;
1122                                         break;
1123                                 }
1124                                 if (ref_offset < parent)
1125                                         break;
1126                         } else {
1127                                 if (root_objectid == ref_offset) {
1128                                         err = 0;
1129                                         break;
1130                                 }
1131                                 if (ref_offset < root_objectid)
1132                                         break;
1133                         }
1134                 }
1135                 ptr += btrfs_extent_inline_ref_size(type);
1136         }
1137         if (err == -ENOENT && insert) {
1138                 if (item_size + extra_size >=
1139                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1140                         err = -EAGAIN;
1141                         goto out;
1142                 }
1143                 /*
1144                  * To add new inline back ref, we have to make sure
1145                  * there is no corresponding back ref item.
1146                  * For simplicity, we just do not add new inline back
1147                  * ref if there is any back ref item.
1148                  */
1149                 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1150                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1151                         err = -EAGAIN;
1152                         goto out;
1153                 }
1154         }
1155         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1156 out:
1157         return err;
1158 }
1159
1160 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1161                                 struct btrfs_root *root,
1162                                 struct btrfs_path *path,
1163                                 struct btrfs_extent_inline_ref *iref,
1164                                 u64 parent, u64 root_objectid,
1165                                 u64 owner, u64 offset, int refs_to_add)
1166 {
1167         struct extent_buffer *leaf;
1168         struct btrfs_extent_item *ei;
1169         unsigned long ptr;
1170         unsigned long end;
1171         unsigned long item_offset;
1172         u64 refs;
1173         int size;
1174         int type;
1175         int ret;
1176
1177         leaf = path->nodes[0];
1178         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1179         item_offset = (unsigned long)iref - (unsigned long)ei;
1180
1181         type = extent_ref_type(parent, owner);
1182         size = btrfs_extent_inline_ref_size(type);
1183
1184         ret = btrfs_extend_item(trans, root, path, size);
1185         BUG_ON(ret);
1186
1187         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1188         refs = btrfs_extent_refs(leaf, ei);
1189         refs += refs_to_add;
1190         btrfs_set_extent_refs(leaf, ei, refs);
1191
1192         ptr = (unsigned long)ei + item_offset;
1193         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1194         if (ptr < end - size)
1195                 memmove_extent_buffer(leaf, ptr + size, ptr,
1196                                       end - size - ptr);
1197
1198         iref = (struct btrfs_extent_inline_ref *)ptr;
1199         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1200         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1201                 struct btrfs_extent_data_ref *dref;
1202                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1203                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1204                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1205                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1206                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1207         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1208                 struct btrfs_shared_data_ref *sref;
1209                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1210                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1211                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1212         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1213                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1214         } else {
1215                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1216         }
1217         btrfs_mark_buffer_dirty(leaf);
1218         return 0;
1219 }
1220
1221 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1222                                  struct btrfs_root *root,
1223                                  struct btrfs_path *path,
1224                                  struct btrfs_extent_inline_ref **ref_ret,
1225                                  u64 bytenr, u64 num_bytes, u64 parent,
1226                                  u64 root_objectid, u64 owner, u64 offset)
1227 {
1228         int ret;
1229
1230         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1231                                            bytenr, num_bytes, parent,
1232                                            root_objectid, owner, offset, 0);
1233         if (ret != -ENOENT)
1234                 return ret;
1235
1236         btrfs_release_path(root, path);
1237         *ref_ret = NULL;
1238
1239         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1240                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1241                                             root_objectid);
1242         } else {
1243                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1244                                              root_objectid, owner, offset);
1245         }
1246         return ret;
1247 }
1248
1249 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1250                                  struct btrfs_root *root,
1251                                  struct btrfs_path *path,
1252                                  struct btrfs_extent_inline_ref *iref,
1253                                  int refs_to_mod)
1254 {
1255         struct extent_buffer *leaf;
1256         struct btrfs_extent_item *ei;
1257         struct btrfs_extent_data_ref *dref = NULL;
1258         struct btrfs_shared_data_ref *sref = NULL;
1259         unsigned long ptr;
1260         unsigned long end;
1261         u32 item_size;
1262         int size;
1263         int type;
1264         int ret;
1265         u64 refs;
1266
1267         leaf = path->nodes[0];
1268         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1269         refs = btrfs_extent_refs(leaf, ei);
1270         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1271         refs += refs_to_mod;
1272         btrfs_set_extent_refs(leaf, ei, refs);
1273
1274         type = btrfs_extent_inline_ref_type(leaf, iref);
1275
1276         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1277                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1278                 refs = btrfs_extent_data_ref_count(leaf, dref);
1279         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1280                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1281                 refs = btrfs_shared_data_ref_count(leaf, sref);
1282         } else {
1283                 refs = 1;
1284                 BUG_ON(refs_to_mod != -1);
1285         }
1286
1287         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1288         refs += refs_to_mod;
1289
1290         if (refs > 0) {
1291                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1292                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1293                 else
1294                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1295         } else {
1296                 size =  btrfs_extent_inline_ref_size(type);
1297                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1298                 ptr = (unsigned long)iref;
1299                 end = (unsigned long)ei + item_size;
1300                 if (ptr + size < end)
1301                         memmove_extent_buffer(leaf, ptr, ptr + size,
1302                                               end - ptr - size);
1303                 item_size -= size;
1304                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1305                 BUG_ON(ret);
1306         }
1307         btrfs_mark_buffer_dirty(leaf);
1308         return 0;
1309 }
1310
1311 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1312                                  struct btrfs_root *root,
1313                                  struct btrfs_path *path,
1314                                  u64 bytenr, u64 num_bytes, u64 parent,
1315                                  u64 root_objectid, u64 owner,
1316                                  u64 offset, int refs_to_add)
1317 {
1318         struct btrfs_extent_inline_ref *iref;
1319         int ret;
1320
1321         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1322                                            bytenr, num_bytes, parent,
1323                                            root_objectid, owner, offset, 1);
1324         if (ret == 0) {
1325                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1326                 ret = update_inline_extent_backref(trans, root, path, iref,
1327                                                    refs_to_add);
1328         } else if (ret == -ENOENT) {
1329                 ret = setup_inline_extent_backref(trans, root, path, iref,
1330                                                   parent, root_objectid,
1331                                                   owner, offset, refs_to_add);
1332         }
1333         return ret;
1334 }
1335
1336 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1337                                  struct btrfs_root *root,
1338                                  struct btrfs_path *path,
1339                                  u64 bytenr, u64 parent, u64 root_objectid,
1340                                  u64 owner, u64 offset, int refs_to_add)
1341 {
1342         int ret;
1343
1344         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1345                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1346                                              parent, root_objectid,
1347                                              owner, offset, refs_to_add);
1348         } else {
1349                 BUG_ON(refs_to_add != 1);
1350                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1351                                             parent, root_objectid);
1352         }
1353         return ret;
1354 }
1355
1356 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1357                                  struct btrfs_root *root,
1358                                  struct btrfs_path *path,
1359                                  struct btrfs_extent_inline_ref *iref,
1360                                  int refs_to_drop, int is_data)
1361 {
1362         int ret;
1363
1364         BUG_ON(!is_data && refs_to_drop != 1);
1365         if (iref) {
1366                 ret = update_inline_extent_backref(trans, root, path, iref,
1367                                                    -refs_to_drop);
1368         } else if (is_data) {
1369                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1370         } else {
1371                 ret = btrfs_del_item(trans, root, path);
1372         }
1373         return ret;
1374 }
1375
1376 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1377                          struct btrfs_root *root,
1378                          u64 bytenr, u64 num_bytes, u64 parent,
1379                          u64 root_objectid, u64 owner, u64 offset)
1380 {
1381         struct btrfs_path *path;
1382         struct extent_buffer *leaf;
1383         struct btrfs_extent_item *item;
1384         u64 refs;
1385         int ret;
1386         int err = 0;
1387
1388         path = btrfs_alloc_path();
1389         if (!path)
1390                 return -ENOMEM;
1391
1392         path->reada = 1;
1393         path->leave_spinning = 1;
1394
1395         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1396                                            path, bytenr, num_bytes, parent,
1397                                            root_objectid, owner, offset, 1);
1398         if (ret == 0)
1399                 goto out;
1400
1401         if (ret != -EAGAIN) {
1402                 err = ret;
1403                 goto out;
1404         }
1405
1406         leaf = path->nodes[0];
1407         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1408         refs = btrfs_extent_refs(leaf, item);
1409         btrfs_set_extent_refs(leaf, item, refs + 1);
1410
1411         btrfs_mark_buffer_dirty(leaf);
1412         btrfs_release_path(root->fs_info->extent_root, path);
1413
1414         path->reada = 1;
1415         path->leave_spinning = 1;
1416
1417         /* now insert the actual backref */
1418         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1419                                     path, bytenr, parent, root_objectid,
1420                                     owner, offset, 1);
1421         if (ret)
1422                 err = ret;
1423 out:
1424         btrfs_free_path(path);
1425         finish_current_insert(trans, root->fs_info->extent_root);
1426         del_pending_extents(trans, root->fs_info->extent_root);
1427         BUG_ON(err);
1428         return err;
1429 }
1430
1431 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1432                          struct btrfs_root *root)
1433 {
1434         finish_current_insert(trans, root->fs_info->extent_root);
1435         del_pending_extents(trans, root->fs_info->extent_root);
1436         return 0;
1437 }
1438
1439 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1440                              struct btrfs_root *root, u64 bytenr,
1441                              u64 num_bytes, u64 *refs, u64 *flags)
1442 {
1443         struct btrfs_path *path;
1444         int ret;
1445         struct btrfs_key key;
1446         struct extent_buffer *l;
1447         struct btrfs_extent_item *item;
1448         u32 item_size;
1449         u64 num_refs;
1450         u64 extent_flags;
1451
1452         WARN_ON(num_bytes < root->sectorsize);
1453         path = btrfs_alloc_path();
1454         path->reada = 1;
1455         key.objectid = bytenr;
1456         key.offset = num_bytes;
1457         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1458         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1459                                 0, 0);
1460         if (ret < 0)
1461                 goto out;
1462         if (ret != 0) {
1463                 btrfs_print_leaf(root, path->nodes[0]);
1464                 printk("failed to find block number %Lu\n",
1465                        (unsigned long long)bytenr);
1466                 BUG();
1467         }
1468
1469         l = path->nodes[0];
1470         item_size = btrfs_item_size_nr(l, path->slots[0]);
1471         if (item_size >= sizeof(*item)) {
1472                 item = btrfs_item_ptr(l, path->slots[0],
1473                                       struct btrfs_extent_item);
1474                 num_refs = btrfs_extent_refs(l, item);
1475                 extent_flags = btrfs_extent_flags(l, item);
1476         } else {
1477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1478                         struct btrfs_extent_item_v0 *ei0;
1479                         BUG_ON(item_size != sizeof(*ei0));
1480                         ei0 = btrfs_item_ptr(l, path->slots[0],
1481                                              struct btrfs_extent_item_v0);
1482                         num_refs = btrfs_extent_refs_v0(l, ei0);
1483                         /* FIXME: this isn't correct for data */
1484                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1485 #else
1486                         BUG();
1487 #endif
1488                 }
1489                 BUG_ON(num_refs == 0);
1490         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1491         if (refs)
1492                 *refs = num_refs;
1493         if (flags)
1494                 *flags = extent_flags;
1495 out:
1496         btrfs_free_path(path);
1497         return 0;
1498 }
1499
1500 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1501                           struct btrfs_root *root,
1502                           u64 bytenr, u64 num_bytes, u64 flags)
1503 {
1504         struct btrfs_path *path;
1505         int ret;
1506         struct btrfs_key key;
1507         struct extent_buffer *l;
1508         struct btrfs_extent_item *item;
1509         u32 item_size;
1510
1511         WARN_ON(num_bytes < root->sectorsize);
1512         path = btrfs_alloc_path();
1513         path->reada = 1;
1514         key.objectid = bytenr;
1515         key.offset = num_bytes;
1516         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1517         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1518                                 0, 0);
1519         if (ret < 0)
1520                 goto out;
1521         if (ret != 0) {
1522                 btrfs_print_leaf(root, path->nodes[0]);
1523                 printk("failed to find block number %Lu\n",
1524                         (unsigned long long)bytenr);
1525                 BUG();
1526         }
1527         l = path->nodes[0];
1528         item_size = btrfs_item_size_nr(l, path->slots[0]);
1529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1530         if (item_size < sizeof(*item)) {
1531                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1532                                              path, (u64)-1, 0);
1533                 if (ret < 0)
1534                         goto out;
1535
1536                 l = path->nodes[0];
1537                 item_size = btrfs_item_size_nr(l, path->slots[0]);
1538         }
1539 #endif
1540         BUG_ON(item_size < sizeof(*item));
1541         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1542         flags |= btrfs_extent_flags(l, item);
1543         btrfs_set_extent_flags(l, item, flags);
1544 out:
1545         btrfs_free_path(path);
1546         finish_current_insert(trans, root->fs_info->extent_root);
1547         del_pending_extents(trans, root->fs_info->extent_root);
1548         return ret;
1549 }
1550
1551 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1552                            struct btrfs_root *root,
1553                            struct extent_buffer *buf,
1554                            int record_parent, int inc)
1555 {
1556         u64 bytenr;
1557         u64 num_bytes;
1558         u64 parent;
1559         u64 ref_root;
1560         u32 nritems;
1561         struct btrfs_key key;
1562         struct btrfs_file_extent_item *fi;
1563         int i;
1564         int level;
1565         int ret = 0;
1566         int (*process_func)(struct btrfs_trans_handle *trans,
1567                             struct btrfs_root *root,
1568                             u64, u64, u64, u64, u64, u64);
1569
1570         ref_root = btrfs_header_owner(buf);
1571         nritems = btrfs_header_nritems(buf);
1572         level = btrfs_header_level(buf);
1573
1574         if (!root->ref_cows && level == 0)
1575                 return 0;
1576
1577         if (inc)
1578                 process_func = btrfs_inc_extent_ref;
1579         else
1580                 process_func = btrfs_free_extent;
1581
1582         if (record_parent)
1583                 parent = buf->start;
1584         else
1585                 parent = 0;
1586
1587         for (i = 0; i < nritems; i++) {
1588                 cond_resched();
1589                 if (level == 0) {
1590                         btrfs_item_key_to_cpu(buf, &key, i);
1591                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1592                                 continue;
1593                         fi = btrfs_item_ptr(buf, i,
1594                                             struct btrfs_file_extent_item);
1595                         if (btrfs_file_extent_type(buf, fi) ==
1596                             BTRFS_FILE_EXTENT_INLINE)
1597                                 continue;
1598                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1599                         if (bytenr == 0)
1600                                 continue;
1601
1602                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1603                         key.offset -= btrfs_file_extent_offset(buf, fi);
1604                         ret = process_func(trans, root, bytenr, num_bytes,
1605                                            parent, ref_root, key.objectid,
1606                                            key.offset);
1607                         if (ret) {
1608                                 WARN_ON(1);
1609                                 goto fail;
1610                         }
1611                 } else {
1612                         bytenr = btrfs_node_blockptr(buf, i);
1613                         num_bytes = btrfs_level_size(root, level - 1);
1614                         ret = process_func(trans, root, bytenr, num_bytes,
1615                                            parent, ref_root, level - 1, 0);
1616                         if (ret) {
1617                                 WARN_ON(1);
1618                                 goto fail;
1619                         }
1620                 }
1621         }
1622         return 0;
1623 fail:
1624         WARN_ON(1);
1625         return ret;
1626 }
1627
1628 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1629                   struct extent_buffer *buf, int record_parent)
1630 {
1631         return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1632 }
1633
1634 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1635                   struct extent_buffer *buf, int record_parent)
1636 {
1637         return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1638 }
1639
1640 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1641                                  struct btrfs_root *root,
1642                                  struct btrfs_path *path,
1643                                  struct btrfs_block_group_cache *cache)
1644 {
1645         int ret;
1646         int pending_ret;
1647         struct btrfs_root *extent_root = root->fs_info->extent_root;
1648         unsigned long bi;
1649         struct extent_buffer *leaf;
1650
1651         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1652         if (ret < 0)
1653                 goto fail;
1654         BUG_ON(ret);
1655
1656         leaf = path->nodes[0];
1657         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1658         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1659         btrfs_mark_buffer_dirty(leaf);
1660         btrfs_release_path(extent_root, path);
1661 fail:
1662         finish_current_insert(trans, extent_root);
1663         pending_ret = del_pending_extents(trans, extent_root);
1664         if (ret)
1665                 return ret;
1666         if (pending_ret)
1667                 return pending_ret;
1668         return 0;
1669
1670 }
1671
1672 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1673                                    struct btrfs_root *root)
1674 {
1675         struct extent_io_tree *block_group_cache;
1676         struct btrfs_block_group_cache *cache;
1677         int ret;
1678         struct btrfs_path *path;
1679         u64 last = 0;
1680         u64 start;
1681         u64 end;
1682         u64 ptr;
1683
1684         block_group_cache = &root->fs_info->block_group_cache;
1685         path = btrfs_alloc_path();
1686         if (!path)
1687                 return -ENOMEM;
1688
1689         while(1) {
1690                 ret = find_first_extent_bit(block_group_cache, last,
1691                                             &start, &end, BLOCK_GROUP_DIRTY);
1692                 if (ret) {
1693                         if (last == 0)
1694                                 break;
1695                         last = 0;
1696                         continue;
1697                 }
1698
1699                 last = end + 1;
1700                 ret = get_state_private(block_group_cache, start, &ptr);
1701                 BUG_ON(ret);
1702
1703                 clear_extent_bits(block_group_cache, start, end,
1704                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1705
1706                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1707                 ret = write_one_cache_group(trans, root, path, cache);
1708                 BUG_ON(ret);
1709         }
1710         btrfs_free_path(path);
1711         return 0;
1712 }
1713
1714 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1715                                                   u64 flags)
1716 {
1717         struct list_head *head = &info->space_info;
1718         struct list_head *cur;
1719         struct btrfs_space_info *found;
1720         list_for_each(cur, head) {
1721                 found = list_entry(cur, struct btrfs_space_info, list);
1722                 if (found->flags & flags)
1723                         return found;
1724         }
1725         return NULL;
1726
1727 }
1728
1729 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1730                              u64 total_bytes, u64 bytes_used,
1731                              struct btrfs_space_info **space_info)
1732 {
1733         struct btrfs_space_info *found;
1734
1735         found = __find_space_info(info, flags);
1736         if (found) {
1737                 found->total_bytes += total_bytes;
1738                 found->bytes_used += bytes_used;
1739                 WARN_ON(found->total_bytes < found->bytes_used);
1740                 *space_info = found;
1741                 return 0;
1742         }
1743         found = kmalloc(sizeof(*found), GFP_NOFS);
1744         if (!found)
1745                 return -ENOMEM;
1746
1747         list_add(&found->list, &info->space_info);
1748         found->flags = flags;
1749         found->total_bytes = total_bytes;
1750         found->bytes_used = bytes_used;
1751         found->bytes_pinned = 0;
1752         found->full = 0;
1753         *space_info = found;
1754         return 0;
1755 }
1756
1757
1758 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1759 {
1760         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1761                                    BTRFS_BLOCK_GROUP_RAID1 |
1762                                    BTRFS_BLOCK_GROUP_RAID10 |
1763                                    BTRFS_BLOCK_GROUP_DUP);
1764         if (extra_flags) {
1765                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1766                         fs_info->avail_data_alloc_bits |= extra_flags;
1767                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1768                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1769                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1770                         fs_info->avail_system_alloc_bits |= extra_flags;
1771         }
1772 }
1773
1774 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1775                           struct btrfs_root *extent_root, u64 alloc_bytes,
1776                           u64 flags)
1777 {
1778         struct btrfs_space_info *space_info;
1779         u64 thresh;
1780         u64 start;
1781         u64 num_bytes;
1782         int ret;
1783
1784         space_info = __find_space_info(extent_root->fs_info, flags);
1785         if (!space_info) {
1786                 ret = update_space_info(extent_root->fs_info, flags,
1787                                         0, 0, &space_info);
1788                 BUG_ON(ret);
1789         }
1790         BUG_ON(!space_info);
1791
1792         if (space_info->full)
1793                 return 0;
1794
1795         thresh = div_factor(space_info->total_bytes, 7);
1796         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1797             thresh)
1798                 return 0;
1799
1800         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1801                                 space_info->flags);
1802         if (ret == -ENOSPC) {
1803                 space_info->full = 1;
1804                 return 0;
1805         }
1806
1807         BUG_ON(ret);
1808
1809         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1810                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1811         BUG_ON(ret);
1812         return 0;
1813 }
1814
1815 static int update_block_group(struct btrfs_trans_handle *trans,
1816                               struct btrfs_root *root,
1817                               u64 bytenr, u64 num_bytes, int alloc,
1818                               int mark_free)
1819 {
1820         struct btrfs_block_group_cache *cache;
1821         struct btrfs_fs_info *info = root->fs_info;
1822         u64 total = num_bytes;
1823         u64 old_val;
1824         u64 byte_in_group;
1825         u64 start;
1826         u64 end;
1827
1828         /* block accounting for super block */
1829         old_val = btrfs_super_bytes_used(&info->super_copy);
1830         if (alloc)
1831                 old_val += num_bytes;
1832         else
1833                 old_val -= num_bytes;
1834         btrfs_set_super_bytes_used(&info->super_copy, old_val);
1835
1836         /* block accounting for root item */
1837         old_val = btrfs_root_used(&root->root_item);
1838         if (alloc)
1839                 old_val += num_bytes;
1840         else
1841                 old_val -= num_bytes;
1842         btrfs_set_root_used(&root->root_item, old_val);
1843
1844         while(total) {
1845                 cache = btrfs_lookup_block_group(info, bytenr);
1846                 if (!cache) {
1847                         return -1;
1848                 }
1849                 byte_in_group = bytenr - cache->key.objectid;
1850                 WARN_ON(byte_in_group > cache->key.offset);
1851                 start = cache->key.objectid;
1852                 end = start + cache->key.offset - 1;
1853                 set_extent_bits(&info->block_group_cache, start, end,
1854                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1855
1856                 old_val = btrfs_block_group_used(&cache->item);
1857                 num_bytes = min(total, cache->key.offset - byte_in_group);
1858                 if (alloc) {
1859                         old_val += num_bytes;
1860                         cache->space_info->bytes_used += num_bytes;
1861                 } else {
1862                         old_val -= num_bytes;
1863                         cache->space_info->bytes_used -= num_bytes;
1864                         if (mark_free) {
1865                                 set_extent_dirty(&info->free_space_cache,
1866                                                  bytenr, bytenr + num_bytes - 1,
1867                                                  GFP_NOFS);
1868                         }
1869                 }
1870                 btrfs_set_block_group_used(&cache->item, old_val);
1871                 total -= num_bytes;
1872                 bytenr += num_bytes;
1873         }
1874         return 0;
1875 }
1876
1877 static int update_pinned_extents(struct btrfs_root *root,
1878                                 u64 bytenr, u64 num, int pin)
1879 {
1880         u64 len;
1881         struct btrfs_block_group_cache *cache;
1882         struct btrfs_fs_info *fs_info = root->fs_info;
1883
1884         if (pin) {
1885                 set_extent_dirty(&fs_info->pinned_extents,
1886                                 bytenr, bytenr + num - 1, GFP_NOFS);
1887         } else {
1888                 clear_extent_dirty(&fs_info->pinned_extents,
1889                                 bytenr, bytenr + num - 1, GFP_NOFS);
1890         }
1891         while (num > 0) {
1892                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1893                 WARN_ON(!cache);
1894                 len = min(num, cache->key.offset -
1895                           (bytenr - cache->key.objectid));
1896                 if (pin) {
1897                         cache->pinned += len;
1898                         cache->space_info->bytes_pinned += len;
1899                         fs_info->total_pinned += len;
1900                 } else {
1901                         cache->pinned -= len;
1902                         cache->space_info->bytes_pinned -= len;
1903                         fs_info->total_pinned -= len;
1904                 }
1905                 bytenr += len;
1906                 num -= len;
1907         }
1908         return 0;
1909 }
1910
1911 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1912 {
1913         u64 last = 0;
1914         u64 start;
1915         u64 end;
1916         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1917         int ret;
1918
1919         while(1) {
1920                 ret = find_first_extent_bit(pinned_extents, last,
1921                                             &start, &end, EXTENT_DIRTY);
1922                 if (ret)
1923                         break;
1924                 set_extent_dirty(copy, start, end, GFP_NOFS);
1925                 last = end + 1;
1926         }
1927         return 0;
1928 }
1929
1930 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1931                                struct btrfs_root *root,
1932                                struct extent_io_tree *unpin)
1933 {
1934         u64 start;
1935         u64 end;
1936         int ret;
1937         struct extent_io_tree *free_space_cache;
1938         free_space_cache = &root->fs_info->free_space_cache;
1939
1940         while(1) {
1941                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1942                                             EXTENT_DIRTY);
1943                 if (ret)
1944                         break;
1945                 update_pinned_extents(root, start, end + 1 - start, 0);
1946                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1947                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1948         }
1949         return 0;
1950 }
1951
1952 static int finish_current_insert(struct btrfs_trans_handle *trans,
1953                                  struct btrfs_root *extent_root)
1954 {
1955         u64 start;
1956         u64 end;
1957         u64 priv;
1958         struct btrfs_fs_info *info = extent_root->fs_info;
1959         struct btrfs_path *path;
1960         struct pending_extent_op *extent_op;
1961         struct btrfs_key key;
1962         int ret;
1963
1964         path = btrfs_alloc_path();
1965
1966         while(1) {
1967                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1968                                             &end, EXTENT_LOCKED);
1969                 if (ret)
1970                         break;
1971
1972                 ret = get_state_private(&info->extent_ins, start, &priv);
1973                 BUG_ON(ret);
1974                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1975
1976                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1977                         key.objectid = start;
1978                         key.offset = end + 1 - start;
1979                         key.type = BTRFS_EXTENT_ITEM_KEY;
1980                         ret = alloc_reserved_tree_block(trans, extent_root,
1981                                                 extent_root->root_key.objectid,
1982                                                 trans->transid,
1983                                                 extent_op->flags,
1984                                                 &extent_op->key,
1985                                                 extent_op->level, &key);
1986                 } else {
1987                         BUG_ON(1);
1988                 }
1989
1990                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1991                                   GFP_NOFS);
1992                 kfree(extent_op);
1993         }
1994         btrfs_free_path(path);
1995         return 0;
1996 }
1997
1998 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1999                           struct btrfs_root *root,
2000                           u64 bytenr, u64 num_bytes, int is_data)
2001 {
2002         int err = 0;
2003         struct extent_buffer *buf;
2004
2005         if (is_data)
2006                 goto pinit;
2007
2008         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2009         if (!buf)
2010                 goto pinit;
2011
2012         /* we can reuse a block if it hasn't been written
2013          * and it is from this transaction.  We can't
2014          * reuse anything from the tree log root because
2015          * it has tiny sub-transactions.
2016          */
2017         if (btrfs_buffer_uptodate(buf, 0)) {
2018                 u64 header_owner = btrfs_header_owner(buf);
2019                 u64 header_transid = btrfs_header_generation(buf);
2020                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2021                     header_transid == trans->transid &&
2022                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2023                         clean_tree_block(NULL, root, buf);
2024                         free_extent_buffer(buf);
2025                         return 1;
2026                 }
2027         }
2028         free_extent_buffer(buf);
2029 pinit:
2030         update_pinned_extents(root, bytenr, num_bytes, 1);
2031
2032         BUG_ON(err < 0);
2033         return 0;
2034 }
2035
2036 /*
2037  * remove an extent from the root, returns 0 on success
2038  */
2039 static int __free_extent(struct btrfs_trans_handle *trans,
2040                          struct btrfs_root *root,
2041                          u64 bytenr, u64 num_bytes, u64 parent,
2042                          u64 root_objectid, u64 owner_objectid,
2043                          u64 owner_offset, int refs_to_drop)
2044 {
2045
2046         struct btrfs_key key;
2047         struct btrfs_path *path;
2048         struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2049         struct btrfs_root *extent_root = root->fs_info->extent_root;
2050         struct extent_buffer *leaf;
2051         struct btrfs_extent_item *ei;
2052         struct btrfs_extent_inline_ref *iref;
2053         int ret;
2054         int is_data;
2055         int extent_slot = 0;
2056         int found_extent = 0;
2057         int num_to_del = 1;
2058         u32 item_size;
2059         u64 refs;
2060
2061         path = btrfs_alloc_path();
2062         if (!path)
2063                 return -ENOMEM;
2064
2065         path->reada = 1;
2066         path->leave_spinning = 1;
2067
2068         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2069         BUG_ON(!is_data && refs_to_drop != 1);
2070
2071         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2072                                     bytenr, num_bytes, parent,
2073                                     root_objectid, owner_objectid,
2074                                     owner_offset);
2075         if (ret == 0) {
2076                 extent_slot = path->slots[0];
2077                 while (extent_slot >= 0) {
2078                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2079                                               extent_slot);
2080                         if (key.objectid != bytenr)
2081                                 break;
2082                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2083                             key.offset == num_bytes) {
2084                                 found_extent = 1;
2085                                 break;
2086                         }
2087                         if (path->slots[0] - extent_slot > 5)
2088                                 break;
2089                         extent_slot--;
2090                 }
2091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2092                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2093                 if (found_extent && item_size < sizeof(*ei))
2094                         found_extent = 0;
2095 #endif
2096                 if (!found_extent) {
2097                         BUG_ON(iref);
2098                         ret = remove_extent_backref(trans, extent_root, path,
2099                                                     NULL, refs_to_drop,
2100                                                     is_data);
2101                         BUG_ON(ret);
2102                         btrfs_release_path(extent_root, path);
2103                         path->leave_spinning = 1;
2104
2105                         key.objectid = bytenr;
2106                         key.type = BTRFS_EXTENT_ITEM_KEY;
2107                         key.offset = num_bytes;
2108
2109                         ret = btrfs_search_slot(trans, extent_root,
2110                                                 &key, path, -1, 1);
2111                         if (ret) {
2112                                 printk(KERN_ERR "umm, got %d back from search"
2113                                        ", was looking for %llu\n", ret,
2114                                        (unsigned long long)bytenr);
2115                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2116                         }
2117                         BUG_ON(ret);
2118                         extent_slot = path->slots[0];
2119                 }
2120         } else {
2121                 btrfs_print_leaf(extent_root, path->nodes[0]);
2122                 WARN_ON(1);
2123                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2124                        "parent %llu root %llu  owner %llu offset %llu\n",
2125                        (unsigned long long)bytenr,
2126                        (unsigned long long)parent,
2127                        (unsigned long long)root_objectid,
2128                        (unsigned long long)owner_objectid,
2129                        (unsigned long long)owner_offset);
2130         }
2131
2132         leaf = path->nodes[0];
2133         item_size = btrfs_item_size_nr(leaf, extent_slot);
2134 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2135         if (item_size < sizeof(*ei)) {
2136                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2137                 ret = convert_extent_item_v0(trans, extent_root, path,
2138                                              owner_objectid, 0);
2139                 BUG_ON(ret < 0);
2140
2141                 btrfs_release_path(extent_root, path);
2142                 path->leave_spinning = 1;
2143
2144                 key.objectid = bytenr;
2145                 key.type = BTRFS_EXTENT_ITEM_KEY;
2146                 key.offset = num_bytes;
2147
2148                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2149                                         -1, 1);
2150                 if (ret) {
2151                         printk(KERN_ERR "umm, got %d back from search"
2152                                ", was looking for %llu\n", ret,
2153                                (unsigned long long)bytenr);
2154                         btrfs_print_leaf(extent_root, path->nodes[0]);
2155                 }
2156                 BUG_ON(ret);
2157                 extent_slot = path->slots[0];
2158                 leaf = path->nodes[0];
2159                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2160         }
2161 #endif
2162         BUG_ON(item_size < sizeof(*ei));
2163         ei = btrfs_item_ptr(leaf, extent_slot,
2164                             struct btrfs_extent_item);
2165         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2166                 struct btrfs_tree_block_info *bi;
2167                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2168                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2169                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2170         }
2171
2172         refs = btrfs_extent_refs(leaf, ei);
2173         BUG_ON(refs < refs_to_drop);
2174         refs -= refs_to_drop;
2175
2176         if (refs > 0) {
2177                 /*
2178                  * In the case of inline back ref, reference count will
2179                  * be updated by remove_extent_backref
2180                  */
2181                 if (iref) {
2182                         BUG_ON(!found_extent);
2183                 } else {
2184                         btrfs_set_extent_refs(leaf, ei, refs);
2185                         btrfs_mark_buffer_dirty(leaf);
2186                 }
2187                 if (found_extent) {
2188                         ret = remove_extent_backref(trans, extent_root, path,
2189                                                     iref, refs_to_drop,
2190                                                     is_data);
2191                         BUG_ON(ret);
2192                 }
2193         } else {
2194                 int mark_free = 0;
2195                 int pin = 1;
2196
2197                 if (found_extent) {
2198                         BUG_ON(is_data && refs_to_drop !=
2199                                extent_data_ref_count(root, path, iref));
2200                         if (iref) {
2201                                 BUG_ON(path->slots[0] != extent_slot);
2202                         } else {
2203                                 BUG_ON(path->slots[0] != extent_slot + 1);
2204                                 path->slots[0] = extent_slot;
2205                                 num_to_del = 2;
2206                         }
2207                 }
2208
2209                 if (ops && ops->free_extent) {
2210                         ret = ops->free_extent(root, bytenr, num_bytes);
2211                         if (ret > 0) {
2212                                 pin = 0;
2213                                 mark_free = 0;
2214                         }
2215                 }
2216
2217                 if (pin) {
2218                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2219                                              is_data);
2220                         if (ret > 0)
2221                                 mark_free = 1;
2222                         BUG_ON(ret < 0);
2223                 }
2224
2225                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2226                                       num_to_del);
2227                 BUG_ON(ret);
2228                 btrfs_release_path(extent_root, path);
2229
2230                 if (is_data) {
2231                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2232                         BUG_ON(ret);
2233                 }
2234
2235                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2236                                          mark_free);
2237                 BUG_ON(ret);
2238         }
2239         btrfs_free_path(path);
2240         finish_current_insert(trans, extent_root);
2241         return ret;
2242 }
2243
2244 /*
2245  * find all the blocks marked as pending in the radix tree and remove
2246  * them from the extent map
2247  */
2248 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2249                                btrfs_root *extent_root)
2250 {
2251         int ret;
2252         int err = 0;
2253         u64 start;
2254         u64 end;
2255         u64 priv;
2256         struct extent_io_tree *pending_del;
2257         struct extent_io_tree *extent_ins;
2258         struct pending_extent_op *extent_op;
2259
2260         extent_ins = &extent_root->fs_info->extent_ins;
2261         pending_del = &extent_root->fs_info->pending_del;
2262
2263         while(1) {
2264                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2265                                             EXTENT_LOCKED);
2266                 if (ret)
2267                         break;
2268
2269                 ret = get_state_private(pending_del, start, &priv);
2270                 BUG_ON(ret);
2271                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2272
2273                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2274                                   GFP_NOFS);
2275
2276                 if (!test_range_bit(extent_ins, start, end,
2277                                     EXTENT_LOCKED, 0)) {
2278                         ret = __free_extent(trans, extent_root,
2279                                             start, end + 1 - start, 0,
2280                                             extent_root->root_key.objectid,
2281                                             extent_op->level, 0, 1);
2282                         kfree(extent_op);
2283                 } else {
2284                         kfree(extent_op);
2285                         ret = get_state_private(extent_ins, start, &priv);
2286                         BUG_ON(ret);
2287                         extent_op = (struct pending_extent_op *)
2288                                                         (unsigned long)priv;
2289
2290                         clear_extent_bits(extent_ins, start, end,
2291                                           EXTENT_LOCKED, GFP_NOFS);
2292
2293                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2294                                 BUG_ON(1);
2295
2296                         kfree(extent_op);
2297                 }
2298                 if (ret)
2299                         err = ret;
2300         }
2301         return err;
2302 }
2303
2304 /*
2305  * remove an extent from the root, returns 0 on success
2306  */
2307
2308 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2309                       struct btrfs_root *root,
2310                       u64 bytenr, u64 num_bytes, u64 parent,
2311                       u64 root_objectid, u64 owner, u64 offset)
2312 {
2313         struct btrfs_root *extent_root = root->fs_info->extent_root;
2314         int pending_ret;
2315         int ret;
2316
2317         WARN_ON(num_bytes < root->sectorsize);
2318         if (root == extent_root) {
2319                 struct pending_extent_op *extent_op;
2320
2321                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2322                 BUG_ON(!extent_op);
2323
2324                 extent_op->type = PENDING_EXTENT_DELETE;
2325                 extent_op->bytenr = bytenr;
2326                 extent_op->num_bytes = num_bytes;
2327                 extent_op->level = (int)owner;
2328
2329                 set_extent_bits(&root->fs_info->pending_del,
2330                                 bytenr, bytenr + num_bytes - 1,
2331                                 EXTENT_LOCKED, GFP_NOFS);
2332                 set_state_private(&root->fs_info->pending_del,
2333                                   bytenr, (unsigned long)extent_op);
2334                 return 0;
2335         }
2336         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2337                             root_objectid, owner, offset, 1);
2338         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2339         return ret ? ret : pending_ret;
2340 }
2341
2342 static u64 stripe_align(struct btrfs_root *root, u64 val)
2343 {
2344         u64 mask = ((u64)root->stripesize - 1);
2345         u64 ret = (val + mask) & ~mask;
2346         return ret;
2347 }
2348
2349 /*
2350  * walks the btree of allocated extents and find a hole of a given size.
2351  * The key ins is changed to record the hole:
2352  * ins->objectid == block start
2353  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2354  * ins->offset == number of blocks
2355  * Any available blocks before search_start are skipped.
2356  */
2357 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2358                                      struct btrfs_root *orig_root,
2359                                      u64 num_bytes, u64 empty_size,
2360                                      u64 search_start, u64 search_end,
2361                                      u64 hint_byte, struct btrfs_key *ins,
2362                                      u64 exclude_start, u64 exclude_nr,
2363                                      int data)
2364 {
2365         int ret;
2366         u64 orig_search_start = search_start;
2367         struct btrfs_root * root = orig_root->fs_info->extent_root;
2368         struct btrfs_fs_info *info = root->fs_info;
2369         u64 total_needed = num_bytes;
2370         struct btrfs_block_group_cache *block_group;
2371         int full_scan = 0;
2372         int wrapped = 0;
2373
2374         WARN_ON(num_bytes < root->sectorsize);
2375         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2376
2377         if (hint_byte) {
2378                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2379                 if (!block_group)
2380                         hint_byte = search_start;
2381                 block_group = btrfs_find_block_group(root, block_group,
2382                                                      hint_byte, data, 1);
2383         } else {
2384                 block_group = btrfs_find_block_group(root,
2385                                                      trans->block_group,
2386                                                      search_start, data, 1);
2387         }
2388
2389         total_needed += empty_size;
2390
2391 check_failed:
2392         if (!block_group) {
2393                 block_group = btrfs_lookup_first_block_group(info,
2394                                                              search_start);
2395                 if (!block_group)
2396                         block_group = btrfs_lookup_first_block_group(info,
2397                                                        orig_search_start);
2398         }
2399         ret = find_search_start(root, &block_group, &search_start,
2400                                 total_needed, data);
2401         if (ret)
2402                 goto error;
2403
2404         search_start = stripe_align(root, search_start);
2405         ins->objectid = search_start;
2406         ins->offset = num_bytes;
2407
2408         if (ins->objectid + num_bytes >
2409             block_group->key.objectid + block_group->key.offset) {
2410                 search_start = block_group->key.objectid +
2411                         block_group->key.offset;
2412                 goto new_group;
2413         }
2414
2415         if (test_range_bit(&info->extent_ins, ins->objectid,
2416                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2417                 search_start = ins->objectid + num_bytes;
2418                 goto new_group;
2419         }
2420
2421         if (test_range_bit(&info->pinned_extents, ins->objectid,
2422                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2423                 search_start = ins->objectid + num_bytes;
2424                 goto new_group;
2425         }
2426
2427         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2428             ins->objectid < exclude_start + exclude_nr)) {
2429                 search_start = exclude_start + exclude_nr;
2430                 goto new_group;
2431         }
2432
2433         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2434                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2435                 if (block_group)
2436                         trans->block_group = block_group;
2437         }
2438         ins->offset = num_bytes;
2439         return 0;
2440
2441 new_group:
2442         block_group = btrfs_lookup_first_block_group(info, search_start);
2443         if (!block_group) {
2444                 search_start = orig_search_start;
2445                 if (full_scan) {
2446                         ret = -ENOSPC;
2447                         goto error;
2448                 }
2449                 if (wrapped) {
2450                         if (!full_scan)
2451                                 total_needed -= empty_size;
2452                         full_scan = 1;
2453                 } else
2454                         wrapped = 1;
2455         }
2456         cond_resched();
2457         block_group = btrfs_find_block_group(root, block_group,
2458                                              search_start, data, 0);
2459         goto check_failed;
2460
2461 error:
2462         return ret;
2463 }
2464
2465 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2466                                 struct btrfs_root *root,
2467                                 u64 num_bytes, u64 empty_size,
2468                                 u64 hint_byte, u64 search_end,
2469                                 struct btrfs_key *ins, int data)
2470 {
2471         int ret;
2472         u64 search_start = 0;
2473         u64 alloc_profile;
2474         struct btrfs_fs_info *info = root->fs_info;
2475
2476         if (info->extent_ops) {
2477                 struct btrfs_extent_ops *ops = info->extent_ops;
2478                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2479                 BUG_ON(ret);
2480                 goto found;
2481         }
2482
2483         if (data) {
2484                 alloc_profile = info->avail_data_alloc_bits &
2485                                 info->data_alloc_profile;
2486                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2487         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2488                    info->system_allocs >= 0) {
2489                 alloc_profile = info->avail_system_alloc_bits &
2490                                 info->system_alloc_profile;
2491                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2492         } else {
2493                 alloc_profile = info->avail_metadata_alloc_bits &
2494                                 info->metadata_alloc_profile;
2495                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2496         }
2497
2498         if (root->ref_cows) {
2499                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2500                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2501                                              num_bytes,
2502                                              BTRFS_BLOCK_GROUP_METADATA);
2503                         BUG_ON(ret);
2504                 }
2505                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2506                                      num_bytes + 2 * 1024 * 1024, data);
2507                 BUG_ON(ret);
2508         }
2509
2510         WARN_ON(num_bytes < root->sectorsize);
2511         ret = find_free_extent(trans, root, num_bytes, empty_size,
2512                                search_start, search_end, hint_byte, ins,
2513                                trans->alloc_exclude_start,
2514                                trans->alloc_exclude_nr, data);
2515         BUG_ON(ret);
2516 found:
2517         clear_extent_dirty(&root->fs_info->free_space_cache,
2518                            ins->objectid, ins->objectid + ins->offset - 1,
2519                            GFP_NOFS);
2520         return ret;
2521 }
2522
2523 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2524                                      struct btrfs_root *root,
2525                                      u64 root_objectid, u64 generation,
2526                                      u64 flags, struct btrfs_disk_key *key,
2527                                      int level, struct btrfs_key *ins)
2528 {
2529         int ret;
2530         struct btrfs_fs_info *fs_info = root->fs_info;
2531         struct btrfs_extent_item *extent_item;
2532         struct btrfs_tree_block_info *block_info;
2533         struct btrfs_extent_inline_ref *iref;
2534         struct btrfs_path *path;
2535         struct extent_buffer *leaf;
2536         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2537
2538         path = btrfs_alloc_path();
2539         BUG_ON(!path);
2540
2541         path->leave_spinning = 1;
2542         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2543                                       ins, size);
2544         BUG_ON(ret);
2545
2546         leaf = path->nodes[0];
2547         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2548                                      struct btrfs_extent_item);
2549         btrfs_set_extent_refs(leaf, extent_item, 1);
2550         btrfs_set_extent_generation(leaf, extent_item, generation);
2551         btrfs_set_extent_flags(leaf, extent_item,
2552                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2553         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2554
2555         btrfs_set_tree_block_key(leaf, block_info, key);
2556         btrfs_set_tree_block_level(leaf, block_info, level);
2557
2558         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2559         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2560         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2561
2562         btrfs_mark_buffer_dirty(leaf);
2563         btrfs_free_path(path);
2564
2565         ret = update_block_group(trans, root, ins->objectid, ins->offset,
2566                                  1, 0);
2567         if (ret) {
2568                 printk(KERN_ERR "btrfs update block group failed for %llu "
2569                        "%llu\n", (unsigned long long)ins->objectid,
2570                        (unsigned long long)ins->offset);
2571                 BUG();
2572         }
2573         return ret;
2574 }
2575
2576 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2577                             struct btrfs_root *root, u64 num_bytes,
2578                             u64 root_objectid, u64 generation,
2579                             u64 flags, struct btrfs_disk_key *key,
2580                             int level, u64 empty_size, u64 hint_byte,
2581                             u64 search_end, struct btrfs_key *ins)
2582 {
2583         int ret;
2584         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2585                                    hint_byte, search_end, ins, 0);
2586         BUG_ON(ret);
2587
2588         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2589                 struct pending_extent_op *extent_op;
2590
2591                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2592                 BUG_ON(!extent_op);
2593
2594                 extent_op->type = PENDING_EXTENT_INSERT;
2595                 extent_op->bytenr = ins->objectid;
2596                 extent_op->num_bytes = ins->offset;
2597                 extent_op->level = level;
2598                 extent_op->flags = flags;
2599                 memcpy(&extent_op->key, key, sizeof(*key));
2600
2601                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2602                                 ins->objectid + ins->offset - 1,
2603                                 EXTENT_LOCKED, GFP_NOFS);
2604                 set_state_private(&root->fs_info->extent_ins,
2605                                   ins->objectid, (unsigned long)extent_op);
2606         } else {
2607                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2608                                                 generation, flags,
2609                                                 key, level, ins);
2610                 finish_current_insert(trans, root->fs_info->extent_root);
2611                 del_pending_extents(trans, root->fs_info->extent_root);
2612         }
2613         return ret;
2614 }
2615
2616 /*
2617  * helper function to allocate a block for a given tree
2618  * returns the tree buffer or NULL.
2619  */
2620 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2621                                         struct btrfs_root *root,
2622                                         u32 blocksize, u64 root_objectid,
2623                                         struct btrfs_disk_key *key, int level,
2624                                         u64 hint, u64 empty_size)
2625 {
2626         struct btrfs_key ins;
2627         int ret;
2628         struct extent_buffer *buf;
2629
2630         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2631                                trans->transid, 0, key, level,
2632                                empty_size, hint, (u64)-1, &ins);
2633         if (ret) {
2634                 BUG_ON(ret > 0);
2635                 return ERR_PTR(ret);
2636         }
2637
2638         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2639         if (!buf) {
2640                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2641                                   0, root->root_key.objectid, level, 0);
2642                 BUG_ON(1);
2643                 return ERR_PTR(-ENOMEM);
2644         }
2645         btrfs_set_buffer_uptodate(buf);
2646         trans->blocks_used++;
2647
2648         return buf;
2649 }
2650
2651 #if 0
2652
2653 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2654                                   struct btrfs_root *root,
2655                                   struct extent_buffer *leaf)
2656 {
2657         u64 leaf_owner;
2658         u64 leaf_generation;
2659         struct btrfs_key key;
2660         struct btrfs_file_extent_item *fi;
2661         int i;
2662         int nritems;
2663         int ret;
2664
2665         BUG_ON(!btrfs_is_leaf(leaf));
2666         nritems = btrfs_header_nritems(leaf);
2667         leaf_owner = btrfs_header_owner(leaf);
2668         leaf_generation = btrfs_header_generation(leaf);
2669
2670         for (i = 0; i < nritems; i++) {
2671                 u64 disk_bytenr;
2672
2673                 btrfs_item_key_to_cpu(leaf, &key, i);
2674                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2675                         continue;
2676                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2677                 if (btrfs_file_extent_type(leaf, fi) ==
2678                     BTRFS_FILE_EXTENT_INLINE)
2679                         continue;
2680                 /*
2681                  * FIXME make sure to insert a trans record that
2682                  * repeats the snapshot del on crash
2683                  */
2684                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2685                 if (disk_bytenr == 0)
2686                         continue;
2687                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2688                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2689                                 leaf->start, leaf_owner, leaf_generation,
2690                                 key.objectid, 0);
2691                 BUG_ON(ret);
2692         }
2693         return 0;
2694 }
2695
2696 static void noinline reada_walk_down(struct btrfs_root *root,
2697                                      struct extent_buffer *node,
2698                                      int slot)
2699 {
2700         u64 bytenr;
2701         u64 last = 0;
2702         u32 nritems;
2703         u32 refs;
2704         u32 blocksize;
2705         int ret;
2706         int i;
2707         int level;
2708         int skipped = 0;
2709
2710         nritems = btrfs_header_nritems(node);
2711         level = btrfs_header_level(node);
2712         if (level)
2713                 return;
2714
2715         for (i = slot; i < nritems && skipped < 32; i++) {
2716                 bytenr = btrfs_node_blockptr(node, i);
2717                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2718                              (last > bytenr && last - bytenr > 32 * 1024))) {
2719                         skipped++;
2720                         continue;
2721                 }
2722                 blocksize = btrfs_level_size(root, level - 1);
2723                 if (i != slot) {
2724                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2725                                                       blocksize, &refs);
2726                         BUG_ON(ret);
2727                         if (refs != 1) {
2728                                 skipped++;
2729                                 continue;
2730                         }
2731                 }
2732                 mutex_unlock(&root->fs_info->fs_mutex);
2733                 ret = readahead_tree_block(root, bytenr, blocksize,
2734                                            btrfs_node_ptr_generation(node, i));
2735                 last = bytenr + blocksize;
2736                 cond_resched();
2737                 mutex_lock(&root->fs_info->fs_mutex);
2738                 if (ret)
2739                         break;
2740         }
2741 }
2742
2743 /*
2744  * helper function for drop_snapshot, this walks down the tree dropping ref
2745  * counts as it goes.
2746  */
2747 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2748                                    struct btrfs_root *root,
2749                                    struct btrfs_path *path, int *level)
2750 {
2751         u64 root_owner;
2752         u64 root_gen;
2753         u64 bytenr;
2754         u64 ptr_gen;
2755         struct extent_buffer *next;
2756         struct extent_buffer *cur;
2757         struct extent_buffer *parent;
2758         u32 blocksize;
2759         int ret;
2760         u32 refs;
2761
2762         WARN_ON(*level < 0);
2763         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2764         ret = btrfs_lookup_extent_ref(trans, root,
2765                                       path->nodes[*level]->start,
2766                                       path->nodes[*level]->len, &refs);
2767         BUG_ON(ret);
2768         if (refs > 1)
2769                 goto out;
2770
2771         /*
2772          * walk down to the last node level and free all the leaves
2773          */
2774         while(*level >= 0) {
2775                 WARN_ON(*level < 0);
2776                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2777                 cur = path->nodes[*level];
2778
2779                 if (btrfs_header_level(cur) != *level)
2780                         WARN_ON(1);
2781
2782                 if (path->slots[*level] >=
2783                     btrfs_header_nritems(cur))
2784                         break;
2785                 if (*level == 0) {
2786                         ret = drop_leaf_ref(trans, root, cur);
2787                         BUG_ON(ret);
2788                         break;
2789                 }
2790                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2791                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2792                 blocksize = btrfs_level_size(root, *level - 1);
2793                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2794                                               &refs);
2795                 BUG_ON(ret);
2796                 if (refs != 1) {
2797                         parent = path->nodes[*level];
2798                         root_owner = btrfs_header_owner(parent);
2799                         root_gen = btrfs_header_generation(parent);
2800                         path->slots[*level]++;
2801                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2802                                                 parent->start, root_owner,
2803                                                 root_gen, *level - 1, 1);
2804                         BUG_ON(ret);
2805                         continue;
2806                 }
2807                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2808                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2809                         free_extent_buffer(next);
2810                         reada_walk_down(root, cur, path->slots[*level]);
2811                         mutex_unlock(&root->fs_info->fs_mutex);
2812                         next = read_tree_block(root, bytenr, blocksize,
2813                                                ptr_gen);
2814                         mutex_lock(&root->fs_info->fs_mutex);
2815                 }
2816                 WARN_ON(*level <= 0);
2817                 if (path->nodes[*level-1])
2818                         free_extent_buffer(path->nodes[*level-1]);
2819                 path->nodes[*level-1] = next;
2820                 *level = btrfs_header_level(next);
2821                 path->slots[*level] = 0;
2822         }
2823 out:
2824         WARN_ON(*level < 0);
2825         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2826
2827         if (path->nodes[*level] == root->node) {
2828                 root_owner = root->root_key.objectid;
2829                 parent = path->nodes[*level];
2830         } else {
2831                 parent = path->nodes[*level + 1];
2832                 root_owner = btrfs_header_owner(parent);
2833         }
2834
2835         root_gen = btrfs_header_generation(parent);
2836         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2837                                 path->nodes[*level]->len, parent->start,
2838                                 root_owner, root_gen, *level, 1);
2839         free_extent_buffer(path->nodes[*level]);
2840         path->nodes[*level] = NULL;
2841         *level += 1;
2842         BUG_ON(ret);
2843         return 0;
2844 }
2845
2846 /*
2847  * helper for dropping snapshots.  This walks back up the tree in the path
2848  * to find the first node higher up where we haven't yet gone through
2849  * all the slots
2850  */
2851 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2852                                  struct btrfs_root *root,
2853                                  struct btrfs_path *path, int *level)
2854 {
2855         u64 root_owner;
2856         u64 root_gen;
2857         struct btrfs_root_item *root_item = &root->root_item;
2858         int i;
2859         int slot;
2860         int ret;
2861
2862         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2863                 slot = path->slots[i];
2864                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2865                         struct extent_buffer *node;
2866                         struct btrfs_disk_key disk_key;
2867                         node = path->nodes[i];
2868                         path->slots[i]++;
2869                         *level = i;
2870                         WARN_ON(*level == 0);
2871                         btrfs_node_key(node, &disk_key, path->slots[i]);
2872                         memcpy(&root_item->drop_progress,
2873                                &disk_key, sizeof(disk_key));
2874                         root_item->drop_level = i;
2875                         return 0;
2876                 } else {
2877                         struct extent_buffer *parent;
2878                         if (path->nodes[*level] == root->node)
2879                                 parent = path->nodes[*level];
2880                         else
2881                                 parent = path->nodes[*level + 1];
2882
2883                         root_owner = btrfs_header_owner(parent);
2884                         root_gen = btrfs_header_generation(parent);
2885                         ret = btrfs_free_extent(trans, root,
2886                                                 path->nodes[*level]->start,
2887                                                 path->nodes[*level]->len,
2888                                                 parent->start, root_owner,
2889                                                 root_gen, *level, 1);
2890                         BUG_ON(ret);
2891                         free_extent_buffer(path->nodes[*level]);
2892                         path->nodes[*level] = NULL;
2893                         *level = i + 1;
2894                 }
2895         }
2896         return 1;
2897 }
2898
2899 /*
2900  * drop the reference count on the tree rooted at 'snap'.  This traverses
2901  * the tree freeing any blocks that have a ref count of zero after being
2902  * decremented.
2903  */
2904 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2905                         *root)
2906 {
2907         int ret = 0;
2908         int wret;
2909         int level;
2910         struct btrfs_path *path;
2911         int i;
2912         int orig_level;
2913         struct btrfs_root_item *root_item = &root->root_item;
2914
2915         path = btrfs_alloc_path();
2916         BUG_ON(!path);
2917
2918         level = btrfs_header_level(root->node);
2919         orig_level = level;
2920         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2921                 path->nodes[level] = root->node;
2922                 extent_buffer_get(root->node);
2923                 path->slots[level] = 0;
2924         } else {
2925                 struct btrfs_key key;
2926                 struct btrfs_disk_key found_key;
2927                 struct extent_buffer *node;
2928
2929                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2930                 level = root_item->drop_level;
2931                 path->lowest_level = level;
2932                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2933                 if (wret < 0) {
2934                         ret = wret;
2935                         goto out;
2936                 }
2937                 node = path->nodes[level];
2938                 btrfs_node_key(node, &found_key, path->slots[level]);
2939                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2940                                sizeof(found_key)));
2941         }
2942         while(1) {
2943                 wret = walk_down_tree(trans, root, path, &level);
2944                 if (wret < 0)
2945                         ret = wret;
2946                 if (wret != 0)
2947                         break;
2948
2949                 wret = walk_up_tree(trans, root, path, &level);
2950                 if (wret < 0)
2951                         ret = wret;
2952                 if (wret != 0)
2953                         break;
2954                 /*
2955                 ret = -EAGAIN;
2956                 break;
2957                 */
2958         }
2959         for (i = 0; i <= orig_level; i++) {
2960                 if (path->nodes[i]) {
2961                         free_extent_buffer(path->nodes[i]);
2962                         path->nodes[i] = NULL;
2963                 }
2964         }
2965 out:
2966         btrfs_free_path(path);
2967         return ret;
2968 }
2969
2970 #endif
2971
2972 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2973 {
2974         u64 start;
2975         u64 end;
2976         u64 ptr;
2977         int ret;
2978         while(1) {
2979                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2980                                             &start, &end, (unsigned int)-1);
2981                 if (ret)
2982                         break;
2983                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2984                 if (!ret)
2985                         kfree((void *)(unsigned long)ptr);
2986                 clear_extent_bits(&info->block_group_cache, start,
2987                                   end, (unsigned int)-1, GFP_NOFS);
2988         }
2989         while(1) {
2990                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2991                                             &start, &end, EXTENT_DIRTY);
2992                 if (ret)
2993                         break;
2994                 clear_extent_dirty(&info->free_space_cache, start,
2995                                    end, GFP_NOFS);
2996         }
2997         return 0;
2998 }
2999
3000 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3001                            struct btrfs_key *key)
3002 {
3003         int ret;
3004         struct btrfs_key found_key;
3005         struct extent_buffer *leaf;
3006         int slot;
3007
3008         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3009         if (ret < 0)
3010                 return ret;
3011         while(1) {
3012                 slot = path->slots[0];
3013                 leaf = path->nodes[0];
3014                 if (slot >= btrfs_header_nritems(leaf)) {
3015                         ret = btrfs_next_leaf(root, path);
3016                         if (ret == 0)
3017                                 continue;
3018                         if (ret < 0)
3019                                 goto error;
3020                         break;
3021                 }
3022                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3023
3024                 if (found_key.objectid >= key->objectid &&
3025                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3026                         return 0;
3027                 path->slots[0]++;
3028         }
3029         ret = -ENOENT;
3030 error:
3031         return ret;
3032 }
3033
3034 int btrfs_read_block_groups(struct btrfs_root *root)
3035 {
3036         struct btrfs_path *path;
3037         int ret;
3038         int bit;
3039         struct btrfs_block_group_cache *cache;
3040         struct btrfs_fs_info *info = root->fs_info;
3041         struct btrfs_space_info *space_info;
3042         struct extent_io_tree *block_group_cache;
3043         struct btrfs_key key;
3044         struct btrfs_key found_key;
3045         struct extent_buffer *leaf;
3046
3047         block_group_cache = &info->block_group_cache;
3048
3049         root = info->extent_root;
3050         key.objectid = 0;
3051         key.offset = 0;
3052         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3053         path = btrfs_alloc_path();
3054         if (!path)
3055                 return -ENOMEM;
3056
3057         while(1) {
3058                 ret = find_first_block_group(root, path, &key);
3059                 if (ret > 0) {
3060                         ret = 0;
3061                         goto error;
3062                 }
3063                 if (ret != 0) {
3064                         goto error;
3065                 }
3066                 leaf = path->nodes[0];
3067                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3068                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3069                 if (!cache) {
3070                         ret = -ENOMEM;
3071                         break;
3072                 }
3073
3074                 read_extent_buffer(leaf, &cache->item,
3075                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3076                                    sizeof(cache->item));
3077                 memcpy(&cache->key, &found_key, sizeof(found_key));
3078                 cache->cached = 0;
3079                 cache->pinned = 0;
3080                 key.objectid = found_key.objectid + found_key.offset;
3081                 btrfs_release_path(root, path);
3082                 cache->flags = btrfs_block_group_flags(&cache->item);
3083                 bit = 0;
3084                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3085                         bit = BLOCK_GROUP_DATA;
3086                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3087                         bit = BLOCK_GROUP_SYSTEM;
3088                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3089                         bit = BLOCK_GROUP_METADATA;
3090                 }
3091                 set_avail_alloc_bits(info, cache->flags);
3092                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3093                         cache->ro = 1;
3094
3095                 ret = update_space_info(info, cache->flags, found_key.offset,
3096                                         btrfs_block_group_used(&cache->item),
3097                                         &space_info);
3098                 BUG_ON(ret);
3099                 cache->space_info = space_info;
3100
3101                 /* use EXTENT_LOCKED to prevent merging */
3102                 set_extent_bits(block_group_cache, found_key.objectid,
3103                                 found_key.objectid + found_key.offset - 1,
3104                                 bit | EXTENT_LOCKED, GFP_NOFS);
3105                 set_state_private(block_group_cache, found_key.objectid,
3106                                   (unsigned long)cache);
3107         }
3108         ret = 0;
3109 error:
3110         btrfs_free_path(path);
3111         return ret;
3112 }
3113
3114 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3115                            struct btrfs_root *root, u64 bytes_used,
3116                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3117                            u64 size)
3118 {
3119         int ret;
3120         int bit = 0;
3121         struct btrfs_root *extent_root;
3122         struct btrfs_block_group_cache *cache;
3123         struct extent_io_tree *block_group_cache;
3124
3125         extent_root = root->fs_info->extent_root;
3126         block_group_cache = &root->fs_info->block_group_cache;
3127
3128         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3129         BUG_ON(!cache);
3130         cache->key.objectid = chunk_offset;
3131         cache->key.offset = size;
3132
3133         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3134         btrfs_set_block_group_used(&cache->item, bytes_used);
3135         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3136         cache->flags = type;
3137         btrfs_set_block_group_flags(&cache->item, type);
3138
3139         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3140                                 &cache->space_info);
3141         BUG_ON(ret);
3142
3143         bit = block_group_state_bits(type);
3144         set_extent_bits(block_group_cache, chunk_offset,
3145                         chunk_offset + size - 1,
3146                         bit | EXTENT_LOCKED, GFP_NOFS);
3147
3148         set_state_private(block_group_cache, chunk_offset,
3149                           (unsigned long)cache);
3150         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3151                                 sizeof(cache->item));
3152         BUG_ON(ret);
3153
3154         finish_current_insert(trans, extent_root);
3155         ret = del_pending_extents(trans, extent_root);
3156         BUG_ON(ret);
3157         set_avail_alloc_bits(extent_root->fs_info, type);
3158         return 0;
3159 }
3160
3161 /*
3162  * This is for converter use only.
3163  *
3164  * In that case, we don't know where are free blocks located.
3165  * Therefore all block group cache entries must be setup properly
3166  * before doing any block allocation.
3167  */
3168 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3169                             struct btrfs_root *root)
3170 {
3171         u64 total_bytes;
3172         u64 cur_start;
3173         u64 group_type;
3174         u64 group_size;
3175         u64 group_align;
3176         u64 total_data = 0;
3177         u64 total_metadata = 0;
3178         u64 chunk_objectid;
3179         int ret;
3180         int bit;
3181         struct btrfs_root *extent_root;
3182         struct btrfs_block_group_cache *cache;
3183         struct extent_io_tree *block_group_cache;
3184
3185         extent_root = root->fs_info->extent_root;
3186         block_group_cache = &root->fs_info->block_group_cache;
3187         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3188         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3189         group_align = 64 * root->sectorsize;
3190
3191         cur_start = 0;
3192         while (cur_start < total_bytes) {
3193                 group_size = total_bytes / 12;
3194                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3195                 if (cur_start == 0) {
3196                         bit = BLOCK_GROUP_SYSTEM;
3197                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3198                         group_size /= 4;
3199                         group_size &= ~(group_align - 1);
3200                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3201                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3202                 } else {
3203                         group_size &= ~(group_align - 1);
3204                         if (total_data >= total_metadata * 2) {
3205                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3206                                 group_size = min_t(u64, group_size,
3207                                                    1ULL * 1024 * 1024 * 1024);
3208                                 total_metadata += group_size;
3209                         } else {
3210                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3211                                 group_size = min_t(u64, group_size,
3212                                                    5ULL * 1024 * 1024 * 1024);
3213                                 total_data += group_size;
3214                         }
3215                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3216                                 group_size = total_bytes - cur_start;
3217                 }
3218
3219                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3220                 BUG_ON(!cache);
3221
3222                 cache->key.objectid = cur_start;
3223                 cache->key.offset = group_size;
3224                 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3225
3226                 btrfs_set_block_group_used(&cache->item, 0);
3227                 btrfs_set_block_group_chunk_objectid(&cache->item,
3228                                                      chunk_objectid);
3229                 btrfs_set_block_group_flags(&cache->item, group_type);
3230
3231                 cache->flags = group_type;
3232
3233                 ret = update_space_info(root->fs_info, group_type, group_size,
3234                                         0, &cache->space_info);
3235                 BUG_ON(ret);
3236                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3237
3238                 set_extent_bits(block_group_cache, cur_start,
3239                                 cur_start + group_size - 1,
3240                                 bit | EXTENT_LOCKED, GFP_NOFS);
3241                 set_state_private(block_group_cache, cur_start,
3242                                   (unsigned long)cache);
3243                 cur_start += group_size;
3244         }
3245         /* then insert all the items */
3246         cur_start = 0;
3247         while(cur_start < total_bytes) {
3248                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3249                 BUG_ON(!cache);
3250
3251                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3252                                         sizeof(cache->item));
3253                 BUG_ON(ret);
3254
3255                 finish_current_insert(trans, extent_root);
3256                 ret = del_pending_extents(trans, extent_root);
3257                 BUG_ON(ret);
3258
3259                 cur_start = cache->key.objectid + cache->key.offset;
3260         }
3261         return 0;
3262 }
3263
3264 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3265                              struct btrfs_root *root,
3266                              u64 bytenr, u64 num_bytes, int alloc,
3267                              int mark_free)
3268 {
3269         return update_block_group(trans, root, bytenr, num_bytes,
3270                                   alloc, mark_free);
3271 }