git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/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 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include "compat.h"
  25 #include "hash.h"
  26 #include "ctree.h"
  27 #include "disk-io.h"
  28 #include "print-tree.h"
  29 #include "transaction.h"
  30 #include "volumes.h"
  31 #include "locking.h"
  32 #include "free-space-cache.h"
  33
  34 static int update_reserved_extents(struct btrfs_root *root,
  35                                    u64 bytenr, u64 num, int reserve);
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc,
  39                               int mark_free);
  40 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  41                                 struct btrfs_root *root,
  42                                 u64 bytenr, u64 num_bytes, u64 parent,
  43                                 u64 root_objectid, u64 owner_objectid,
  44                                 u64 owner_offset, int refs_to_drop,
  45                                 struct btrfs_delayed_extent_op *extra_op);
  46 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  47                                     struct extent_buffer *leaf,
  48                                     struct btrfs_extent_item *ei);
  49 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  50                                       struct btrfs_root *root,
  51                                       u64 parent, u64 root_objectid,
  52                                       u64 flags, u64 owner, u64 offset,
  53                                       struct btrfs_key *ins, int ref_mod);
  54 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  55                                      struct btrfs_root *root,
  56                                      u64 parent, u64 root_objectid,
  57                                      u64 flags, struct btrfs_disk_key *key,
  58                                      int level, struct btrfs_key *ins);
  59
  60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  61                           struct btrfs_root *extent_root, u64 alloc_bytes,
  62                           u64 flags, int force);
  63
  64 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  65 {
  66         return (cache->flags & bits) == bits;
  67 }
  68
  69 /*
  70  * this adds the block group to the fs_info rb tree for the block group
  71  * cache
  72  */
  73 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  74                                 struct btrfs_block_group_cache *block_group)
  75 {
  76         struct rb_node **p;
  77         struct rb_node *parent = NULL;
  78         struct btrfs_block_group_cache *cache;
  79
  80         spin_lock(&info->block_group_cache_lock);
  81         p = &info->block_group_cache_tree.rb_node;
  82
  83         while (*p) {
  84                 parent = *p;
  85                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  86                                  cache_node);
  87                 if (block_group->key.objectid < cache->key.objectid) {
  88                         p = &(*p)->rb_left;
  89                 } else if (block_group->key.objectid > cache->key.objectid) {
  90                         p = &(*p)->rb_right;
  91                 } else {
  92                         spin_unlock(&info->block_group_cache_lock);
  93                         return -EEXIST;
  94                 }
  95         }
  96
  97         rb_link_node(&block_group->cache_node, parent, p);
  98         rb_insert_color(&block_group->cache_node,
  99                         &info->block_group_cache_tree);
 100         spin_unlock(&info->block_group_cache_lock);
 101
 102         return 0;
 103 }
 104
 105 /*
 106  * This will return the block group at or after bytenr if contains is 0, else
 107  * it will return the block group that contains the bytenr
 108  */
 109 static struct btrfs_block_group_cache *
 110 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 111                               int contains)
 112 {
 113         struct btrfs_block_group_cache *cache, *ret = NULL;
 114         struct rb_node *n;
 115         u64 end, start;
 116
 117         spin_lock(&info->block_group_cache_lock);
 118         n = info->block_group_cache_tree.rb_node;
 119
 120         while (n) {
 121                 cache = rb_entry(n, struct btrfs_block_group_cache,
 122                                  cache_node);
 123                 end = cache->key.objectid + cache->key.offset - 1;
 124                 start = cache->key.objectid;
 125
 126                 if (bytenr < start) {
 127                         if (!contains && (!ret || start < ret->key.objectid))
 128                                 ret = cache;
 129                         n = n->rb_left;
 130                 } else if (bytenr > start) {
 131                         if (contains && bytenr <= end) {
 132                                 ret = cache;
 133                                 break;
 134                         }
 135                         n = n->rb_right;
 136                 } else {
 137                         ret = cache;
 138                         break;
 139                 }
 140         }
 141         if (ret)
 142                 atomic_inc(&ret->count);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return ret;
 146 }
 147
 148 /*
 149  * this is only called by cache_block_group, since we could have freed extents
 150  * we need to check the pinned_extents for any extents that can't be used yet
 151  * since their free space will be released as soon as the transaction commits.
 152  */
 153 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 154                               struct btrfs_fs_info *info, u64 start, u64 end)
 155 {
 156         u64 extent_start, extent_end, size;
 157         int ret;
 158
 159         while (start < end) {
 160                 ret = find_first_extent_bit(&info->pinned_extents, start,
 161                                             &extent_start, &extent_end,
 162                                             EXTENT_DIRTY);
 163                 if (ret)
 164                         break;
 165
 166                 if (extent_start == start) {
 167                         start = extent_end + 1;
 168                 } else if (extent_start > start && extent_start < end) {
 169                         size = extent_start - start;
 170                         ret = btrfs_add_free_space(block_group, start,
 171                                                    size);
 172                         BUG_ON(ret);
 173                         start = extent_end + 1;
 174                 } else {
 175                         break;
 176                 }
 177         }
 178
 179         if (start < end) {
 180                 size = end - start;
 181                 ret = btrfs_add_free_space(block_group, start, size);
 182                 BUG_ON(ret);
 183         }
 184
 185         return 0;
 186 }
 187
 188 static int remove_sb_from_cache(struct btrfs_root *root,
 189                                 struct btrfs_block_group_cache *cache)
 190 {
 191         u64 bytenr;
 192         u64 *logical;
 193         int stripe_len;
 194         int i, nr, ret;
 195
 196         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 197                 bytenr = btrfs_sb_offset(i);
 198                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 199                                        cache->key.objectid, bytenr, 0,
 200                                        &logical, &nr, &stripe_len);
 201                 BUG_ON(ret);
 202                 while (nr--) {
 203                         btrfs_remove_free_space(cache, logical[nr],
 204                                                 stripe_len);
 205                 }
 206                 kfree(logical);
 207         }
 208         return 0;
 209 }
 210
 211 static int cache_block_group(struct btrfs_root *root,
 212                              struct btrfs_block_group_cache *block_group)
 213 {
 214         struct btrfs_path *path;
 215         int ret = 0;
 216         struct btrfs_key key;
 217         struct extent_buffer *leaf;
 218         int slot;
 219         u64 last;
 220
 221         if (!block_group)
 222                 return 0;
 223
 224         root = root->fs_info->extent_root;
 225
 226         if (block_group->cached)
 227                 return 0;
 228
 229         path = btrfs_alloc_path();
 230         if (!path)
 231                 return -ENOMEM;
 232
 233         path->reada = 2;
 234         /*
 235          * we get into deadlocks with paths held by callers of this function.
 236          * since the alloc_mutex is protecting things right now, just
 237          * skip the locking here
 238          */
 239         path->skip_locking = 1;
 240         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 241         key.objectid = last;
 242         key.offset = 0;
 243         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 244         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 245         if (ret < 0)
 246                 goto err;
 247
 248         while (1) {
 249                 leaf = path->nodes[0];
 250                 slot = path->slots[0];
 251                 if (slot >= btrfs_header_nritems(leaf)) {
 252                         ret = btrfs_next_leaf(root, path);
 253                         if (ret < 0)
 254                                 goto err;
 255                         if (ret == 0)
 256                                 continue;
 257                         else
 258                                 break;
 259                 }
 260                 btrfs_item_key_to_cpu(leaf, &key, slot);
 261                 if (key.objectid < block_group->key.objectid)
 262                         goto next;
 263
 264                 if (key.objectid >= block_group->key.objectid +
 265                     block_group->key.offset)
 266                         break;
 267
 268                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 269                         add_new_free_space(block_group, root->fs_info, last,
 270                                            key.objectid);
 271
 272                         last = key.objectid + key.offset;
 273                 }
 274 next:
 275                 path->slots[0]++;
 276         }
 277
 278         add_new_free_space(block_group, root->fs_info, last,
 279                            block_group->key.objectid +
 280                            block_group->key.offset);
 281
 282         block_group->cached = 1;
 283         remove_sb_from_cache(root, block_group);
 284         ret = 0;
 285 err:
 286         btrfs_free_path(path);
 287         return ret;
 288 }
 289
 290 /*
 291  * return the block group that starts at or after bytenr
 292  */
 293 static struct btrfs_block_group_cache *
 294 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 295 {
 296         struct btrfs_block_group_cache *cache;
 297
 298         cache = block_group_cache_tree_search(info, bytenr, 0);
 299
 300         return cache;
 301 }
 302
 303 /*
 304  * return the block group that contains the given bytenr
 305  */
 306 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 307                                                  struct btrfs_fs_info *info,
 308                                                  u64 bytenr)
 309 {
 310         struct btrfs_block_group_cache *cache;
 311
 312         cache = block_group_cache_tree_search(info, bytenr, 1);
 313
 314         return cache;
 315 }
 316
 317 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 318 {
 319         if (atomic_dec_and_test(&cache->count))
 320                 kfree(cache);
 321 }
 322
 323 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 324                                                   u64 flags)
 325 {
 326         struct list_head *head = &info->space_info;
 327         struct btrfs_space_info *found;
 328
 329         rcu_read_lock();
 330         list_for_each_entry_rcu(found, head, list) {
 331                 if (found->flags == flags) {
 332                         rcu_read_unlock();
 333                         return found;
 334                 }
 335         }
 336         rcu_read_unlock();
 337         return NULL;
 338 }
 339
 340 /*
 341  * after adding space to the filesystem, we need to clear the full flags
 342  * on all the space infos.
 343  */
 344 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 345 {
 346         struct list_head *head = &info->space_info;
 347         struct btrfs_space_info *found;
 348
 349         rcu_read_lock();
 350         list_for_each_entry_rcu(found, head, list)
 351                 found->full = 0;
 352         rcu_read_unlock();
 353 }
 354
 355 static u64 div_factor(u64 num, int factor)
 356 {
 357         if (factor == 10)
 358                 return num;
 359         num *= factor;
 360         do_div(num, 10);
 361         return num;
 362 }
 363
 364 u64 btrfs_find_block_group(struct btrfs_root *root,
 365                            u64 search_start, u64 search_hint, int owner)
 366 {
 367         struct btrfs_block_group_cache *cache;
 368         u64 used;
 369         u64 last = max(search_hint, search_start);
 370         u64 group_start = 0;
 371         int full_search = 0;
 372         int factor = 9;
 373         int wrapped = 0;
 374 again:
 375         while (1) {
 376                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 377                 if (!cache)
 378                         break;
 379
 380                 spin_lock(&cache->lock);
 381                 last = cache->key.objectid + cache->key.offset;
 382                 used = btrfs_block_group_used(&cache->item);
 383
 384                 if ((full_search || !cache->ro) &&
 385                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 386                         if (used + cache->pinned + cache->reserved <
 387                             div_factor(cache->key.offset, factor)) {
 388                                 group_start = cache->key.objectid;
 389                                 spin_unlock(&cache->lock);
 390                                 btrfs_put_block_group(cache);
 391                                 goto found;
 392                         }
 393                 }
 394                 spin_unlock(&cache->lock);
 395                 btrfs_put_block_group(cache);
 396                 cond_resched();
 397         }
 398         if (!wrapped) {
 399                 last = search_start;
 400                 wrapped = 1;
 401                 goto again;
 402         }
 403         if (!full_search && factor < 10) {
 404                 last = search_start;
 405                 full_search = 1;
 406                 factor = 10;
 407                 goto again;
 408         }
 409 found:
 410         return group_start;
 411 }
 412
 413 /* simple helper to search for an existing extent at a given offset */
 414 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 415 {
 416         int ret;
 417         struct btrfs_key key;
 418         struct btrfs_path *path;
 419
 420         path = btrfs_alloc_path();
 421         BUG_ON(!path);
 422         key.objectid = start;
 423         key.offset = len;
 424         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 425         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 426                                 0, 0);
 427         btrfs_free_path(path);
 428         return ret;
 429 }
 430
 431 /*
 432  * Back reference rules.  Back refs have three main goals:
 433  *
 434  * 1) differentiate between all holders of references to an extent so that
 435  *    when a reference is dropped we can make sure it was a valid reference
 436  *    before freeing the extent.
 437  *
 438  * 2) Provide enough information to quickly find the holders of an extent
 439  *    if we notice a given block is corrupted or bad.
 440  *
 441  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 442  *    maintenance.  This is actually the same as #2, but with a slightly
 443  *    different use case.
 444  *
 445  * There are two kinds of back refs. The implicit back refs is optimized
 446  * for pointers in non-shared tree blocks. For a given pointer in a block,
 447  * back refs of this kind provide information about the block's owner tree
 448  * and the pointer's key. These information allow us to find the block by
 449  * b-tree searching. The full back refs is for pointers in tree blocks not
 450  * referenced by their owner trees. The location of tree block is recorded
 451  * in the back refs. Actually the full back refs is generic, and can be
 452  * used in all cases the implicit back refs is used. The major shortcoming
 453  * of the full back refs is its overhead. Every time a tree block gets
 454  * COWed, we have to update back refs entry for all pointers in it.
 455  *
 456  * For a newly allocated tree block, we use implicit back refs for
 457  * pointers in it. This means most tree related operations only involve
 458  * implicit back refs. For a tree block created in old transaction, the
 459  * only way to drop a reference to it is COW it. So we can detect the
 460  * event that tree block loses its owner tree's reference and do the
 461  * back refs conversion.
 462  *
 463  * When a tree block is COW'd through a tree, there are four cases:
 464  *
 465  * The reference count of the block is one and the tree is the block's
 466  * owner tree. Nothing to do in this case.
 467  *
 468  * The reference count of the block is one and the tree is not the
 469  * block's owner tree. In this case, full back refs is used for pointers
 470  * in the block. Remove these full back refs, add implicit back refs for
 471  * every pointers in the new block.
 472  *
 473  * The reference count of the block is greater than one and the tree is
 474  * the block's owner tree. In this case, implicit back refs is used for
 475  * pointers in the block. Add full back refs for every pointers in the
 476  * block, increase lower level extents' reference counts. The original
 477  * implicit back refs are entailed to the new block.
 478  *
 479  * The reference count of the block is greater than one and the tree is
 480  * not the block's owner tree. Add implicit back refs for every pointer in
 481  * the new block, increase lower level extents' reference count.
 482  *
 483  * Back Reference Key composing:
 484  *
 485  * The key objectid corresponds to the first byte in the extent,
 486  * The key type is used to differentiate between types of back refs.
 487  * There are different meanings of the key offset for different types
 488  * of back refs.
 489  *
 490  * File extents can be referenced by:
 491  *
 492  * - multiple snapshots, subvolumes, or different generations in one subvol
 493  * - different files inside a single subvolume
 494  * - different offsets inside a file (bookend extents in file.c)
 495  *
 496  * The extent ref structure for the implicit back refs has fields for:
 497  *
 498  * - Objectid of the subvolume root
 499  * - objectid of the file holding the reference
 500  * - original offset in the file
 501  * - how many bookend extents
 502  *
 503  * The key offset for the implicit back refs is hash of the first
 504  * three fields.
 505  *
 506  * The extent ref structure for the full back refs has field for:
 507  *
 508  * - number of pointers in the tree leaf
 509  *
 510  * The key offset for the implicit back refs is the first byte of
 511  * the tree leaf
 512  *
 513  * When a file extent is allocated, The implicit back refs is used.
 514  * the fields are filled in:
 515  *
 516  *     (root_key.objectid, inode objectid, offset in file, 1)
 517  *
 518  * When a file extent is removed file truncation, we find the
 519  * corresponding implicit back refs and check the following fields:
 520  *
 521  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 522  *
 523  * Btree extents can be referenced by:
 524  *
 525  * - Different subvolumes
 526  *
 527  * Both the implicit back refs and the full back refs for tree blocks
 528  * only consist of key. The key offset for the implicit back refs is
 529  * objectid of block's owner tree. The key offset for the full back refs
 530  * is the first byte of parent block.
 531  *
 532  * When implicit back refs is used, information about the lowest key and
 533  * level of the tree block are required. These information are stored in
 534  * tree block info structure.
 535  */
 536
 537 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 538 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 539                                   struct btrfs_root *root,
 540                                   struct btrfs_path *path,
 541                                   u64 owner, u32 extra_size)
 542 {
 543         struct btrfs_extent_item *item;
 544         struct btrfs_extent_item_v0 *ei0;
 545         struct btrfs_extent_ref_v0 *ref0;
 546         struct btrfs_tree_block_info *bi;
 547         struct extent_buffer *leaf;
 548         struct btrfs_key key;
 549         struct btrfs_key found_key;
 550         u32 new_size = sizeof(*item);
 551         u64 refs;
 552         int ret;
 553
 554         leaf = path->nodes[0];
 555         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 556
 557         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 558         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 559                              struct btrfs_extent_item_v0);
 560         refs = btrfs_extent_refs_v0(leaf, ei0);
 561
 562         if (owner == (u64)-1) {
 563                 while (1) {
 564                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 565                                 ret = btrfs_next_leaf(root, path);
 566                                 if (ret < 0)
 567                                         return ret;
 568                                 BUG_ON(ret > 0);
 569                                 leaf = path->nodes[0];
 570                         }
 571                         btrfs_item_key_to_cpu(leaf, &found_key,
 572                                               path->slots[0]);
 573                         BUG_ON(key.objectid != found_key.objectid);
 574                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 575                                 path->slots[0]++;
 576                                 continue;
 577                         }
 578                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 579                                               struct btrfs_extent_ref_v0);
 580                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 581                         break;
 582                 }
 583         }
 584         btrfs_release_path(root, path);
 585
 586         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 587                 new_size += sizeof(*bi);
 588
 589         new_size -= sizeof(*ei0);
 590         ret = btrfs_search_slot(trans, root, &key, path,
 591                                 new_size + extra_size, 1);
 592         if (ret < 0)
 593                 return ret;
 594         BUG_ON(ret);
 595
 596         ret = btrfs_extend_item(trans, root, path, new_size);
 597         BUG_ON(ret);
 598
 599         leaf = path->nodes[0];
 600         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 601         btrfs_set_extent_refs(leaf, item, refs);
 602         /* FIXME: get real generation */
 603         btrfs_set_extent_generation(leaf, item, 0);
 604         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 605                 btrfs_set_extent_flags(leaf, item,
 606                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 607                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 608                 bi = (struct btrfs_tree_block_info *)(item + 1);
 609                 /* FIXME: get first key of the block */
 610                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 611                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 612         } else {
 613                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 614         }
 615         btrfs_mark_buffer_dirty(leaf);
 616         return 0;
 617 }
 618 #endif
 619
 620 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 621 {
 622         u32 high_crc = ~(u32)0;
 623         u32 low_crc = ~(u32)0;
 624         __le64 lenum;
 625
 626         lenum = cpu_to_le64(root_objectid);
 627         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 628         lenum = cpu_to_le64(owner);
 629         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 630         lenum = cpu_to_le64(offset);
 631         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 632
 633         return ((u64)high_crc << 31) ^ (u64)low_crc;
 634 }
 635
 636 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 637                                      struct btrfs_extent_data_ref *ref)
 638 {
 639         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 640                                     btrfs_extent_data_ref_objectid(leaf, ref),
 641                                     btrfs_extent_data_ref_offset(leaf, ref));
 642 }
 643
 644 static int match_extent_data_ref(struct extent_buffer *leaf,
 645                                  struct btrfs_extent_data_ref *ref,
 646                                  u64 root_objectid, u64 owner, u64 offset)
 647 {
 648         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 649             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 650             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 651                 return 0;
 652         return 1;
 653 }
 654
 655 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 656                                            struct btrfs_root *root,
 657                                            struct btrfs_path *path,
 658                                            u64 bytenr, u64 parent,
 659                                            u64 root_objectid,
 660                                            u64 owner, u64 offset)
 661 {
 662         struct btrfs_key key;
 663         struct btrfs_extent_data_ref *ref;
 664         struct extent_buffer *leaf;
 665         u32 nritems;
 666         int ret;
 667         int recow;
 668         int err = -ENOENT;
 669
 670         key.objectid = bytenr;
 671         if (parent) {
 672                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 673                 key.offset = parent;
 674         } else {
 675                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 676                 key.offset = hash_extent_data_ref(root_objectid,
 677                                                   owner, offset);
 678         }
 679 again:
 680         recow = 0;
 681         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 682         if (ret < 0) {
 683                 err = ret;
 684                 goto fail;
 685         }
 686
 687         if (parent) {
 688                 if (!ret)
 689                         return 0;
 690 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 691                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 692                 btrfs_release_path(root, path);
 693                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 694                 if (ret < 0) {
 695                         err = ret;
 696                         goto fail;
 697                 }
 698                 if (!ret)
 699                         return 0;
 700 #endif
 701                 goto fail;
 702         }
 703
 704         leaf = path->nodes[0];
 705         nritems = btrfs_header_nritems(leaf);
 706         while (1) {
 707                 if (path->slots[0] >= nritems) {
 708                         ret = btrfs_next_leaf(root, path);
 709                         if (ret < 0)
 710                                 err = ret;
 711                         if (ret)
 712                                 goto fail;
 713
 714                         leaf = path->nodes[0];
 715                         nritems = btrfs_header_nritems(leaf);
 716                         recow = 1;
 717                 }
 718
 719                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 720                 if (key.objectid != bytenr ||
 721                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 722                         goto fail;
 723
 724                 ref = btrfs_item_ptr(leaf, path->slots[0],
 725                                      struct btrfs_extent_data_ref);
 726
 727                 if (match_extent_data_ref(leaf, ref, root_objectid,
 728                                           owner, offset)) {
 729                         if (recow) {
 730                                 btrfs_release_path(root, path);
 731                                 goto again;
 732                         }
 733                         err = 0;
 734                         break;
 735                 }
 736                 path->slots[0]++;
 737         }
 738 fail:
 739         return err;
 740 }
 741
 742 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 743                                            struct btrfs_root *root,
 744                                            struct btrfs_path *path,
 745                                            u64 bytenr, u64 parent,
 746                                            u64 root_objectid, u64 owner,
 747                                            u64 offset, int refs_to_add)
 748 {
 749         struct btrfs_key key;
 750         struct extent_buffer *leaf;
 751         u32 size;
 752         u32 num_refs;
 753         int ret;
 754
 755         key.objectid = bytenr;
 756         if (parent) {
 757                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 758                 key.offset = parent;
 759                 size = sizeof(struct btrfs_shared_data_ref);
 760         } else {
 761                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 762                 key.offset = hash_extent_data_ref(root_objectid,
 763                                                   owner, offset);
 764                 size = sizeof(struct btrfs_extent_data_ref);
 765         }
 766
 767         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 768         if (ret && ret != -EEXIST)
 769                 goto fail;
 770
 771         leaf = path->nodes[0];
 772         if (parent) {
 773                 struct btrfs_shared_data_ref *ref;
 774                 ref = btrfs_item_ptr(leaf, path->slots[0],
 775                                      struct btrfs_shared_data_ref);
 776                 if (ret == 0) {
 777                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 778                 } else {
 779                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 780                         num_refs += refs_to_add;
 781                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 782                 }
 783         } else {
 784                 struct btrfs_extent_data_ref *ref;
 785                 while (ret == -EEXIST) {
 786                         ref = btrfs_item_ptr(leaf, path->slots[0],
 787                                              struct btrfs_extent_data_ref);
 788                         if (match_extent_data_ref(leaf, ref, root_objectid,
 789                                                   owner, offset))
 790                                 break;
 791                         btrfs_release_path(root, path);
 792                         key.offset++;
 793                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 794                                                       size);
 795                         if (ret && ret != -EEXIST)
 796                                 goto fail;
 797
 798                         leaf = path->nodes[0];
 799                 }
 800                 ref = btrfs_item_ptr(leaf, path->slots[0],
 801                                      struct btrfs_extent_data_ref);
 802                 if (ret == 0) {
 803                         btrfs_set_extent_data_ref_root(leaf, ref,
 804                                                        root_objectid);
 805                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 806                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 807                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 808                 } else {
 809                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 810                         num_refs += refs_to_add;
 811                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 812                 }
 813         }
 814         btrfs_mark_buffer_dirty(leaf);
 815         ret = 0;
 816 fail:
 817         btrfs_release_path(root, path);
 818         return ret;
 819 }
 820
 821 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 822                                            struct btrfs_root *root,
 823                                            struct btrfs_path *path,
 824                                            int refs_to_drop)
 825 {
 826         struct btrfs_key key;
 827         struct btrfs_extent_data_ref *ref1 = NULL;
 828         struct btrfs_shared_data_ref *ref2 = NULL;
 829         struct extent_buffer *leaf;
 830         u32 num_refs = 0;
 831         int ret = 0;
 832
 833         leaf = path->nodes[0];
 834         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 835
 836         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 837                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 838                                       struct btrfs_extent_data_ref);
 839                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 840         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 841                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 842                                       struct btrfs_shared_data_ref);
 843                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 844 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 845         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 846                 struct btrfs_extent_ref_v0 *ref0;
 847                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 848                                       struct btrfs_extent_ref_v0);
 849                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 850 #endif
 851         } else {
 852                 BUG();
 853         }
 854
 855         BUG_ON(num_refs < refs_to_drop);
 856         num_refs -= refs_to_drop;
 857
 858         if (num_refs == 0) {
 859                 ret = btrfs_del_item(trans, root, path);
 860         } else {
 861                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 862                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 863                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 864                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 865 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 866                 else {
 867                         struct btrfs_extent_ref_v0 *ref0;
 868                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 869                                         struct btrfs_extent_ref_v0);
 870                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
 871                 }
 872 #endif
 873                 btrfs_mark_buffer_dirty(leaf);
 874         }
 875         return ret;
 876 }
 877
 878 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
 879                                           struct btrfs_path *path,
 880                                           struct btrfs_extent_inline_ref *iref)
 881 {
 882         struct btrfs_key key;
 883         struct extent_buffer *leaf;
 884         struct btrfs_extent_data_ref *ref1;
 885         struct btrfs_shared_data_ref *ref2;
 886         u32 num_refs = 0;
 887
 888         leaf = path->nodes[0];
 889         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 890         if (iref) {
 891                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
 892                     BTRFS_EXTENT_DATA_REF_KEY) {
 893                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 894                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 895                 } else {
 896                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 897                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 898                 }
 899         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 900                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 901                                       struct btrfs_extent_data_ref);
 902                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 903         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 904                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 905                                       struct btrfs_shared_data_ref);
 906                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 907 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 908         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 909                 struct btrfs_extent_ref_v0 *ref0;
 910                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 911                                       struct btrfs_extent_ref_v0);
 912                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 913 #endif
 914         } else {
 915                 WARN_ON(1);
 916         }
 917         return num_refs;
 918 }
 919
 920 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 921                                           struct btrfs_root *root,
 922                                           struct btrfs_path *path,
 923                                           u64 bytenr, u64 parent,
 924                                           u64 root_objectid)
 925 {
 926         struct btrfs_key key;
 927         int ret;
 928
 929         key.objectid = bytenr;
 930         if (parent) {
 931                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 932                 key.offset = parent;
 933         } else {
 934                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 935                 key.offset = root_objectid;
 936         }
 937
 938         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 939         if (ret > 0)
 940                 ret = -ENOENT;
 941 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 942         if (ret == -ENOENT && parent) {
 943                 btrfs_release_path(root, path);
 944                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 945                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 946                 if (ret > 0)
 947                         ret = -ENOENT;
 948         }
 949 #endif
 950         return ret;
 951 }
 952
 953 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 954                                           struct btrfs_root *root,
 955                                           struct btrfs_path *path,
 956                                           u64 bytenr, u64 parent,
 957                                           u64 root_objectid)
 958 {
 959         struct btrfs_key key;
 960         int ret;
 961
 962         key.objectid = bytenr;
 963         if (parent) {
 964                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 965                 key.offset = parent;
 966         } else {
 967                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 968                 key.offset = root_objectid;
 969         }
 970
 971         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 972         btrfs_release_path(root, path);
 973         return ret;
 974 }
 975
 976 static inline int extent_ref_type(u64 parent, u64 owner)
 977 {
 978         int type;
 979         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 980                 if (parent > 0)
 981                         type = BTRFS_SHARED_BLOCK_REF_KEY;
 982                 else
 983                         type = BTRFS_TREE_BLOCK_REF_KEY;
 984         } else {
 985                 if (parent > 0)
 986                         type = BTRFS_SHARED_DATA_REF_KEY;
 987                 else
 988                         type = BTRFS_EXTENT_DATA_REF_KEY;
 989         }
 990         return type;
 991 }
 992
 993 static int find_next_key(struct btrfs_path *path, int level,
 994                          struct btrfs_key *key)
 995
 996 {
 997         for (; level < BTRFS_MAX_LEVEL; level++) {
 998                 if (!path->nodes[level])
 999                         break;
1000                 if (path->slots[level] + 1 >=
1001                     btrfs_header_nritems(path->nodes[level]))
1002                         continue;
1003                 if (level == 0)
1004                         btrfs_item_key_to_cpu(path->nodes[level], key,
1005                                               path->slots[level] + 1);
1006                 else
1007                         btrfs_node_key_to_cpu(path->nodes[level], key,
1008                                               path->slots[level] + 1);
1009                 return 0;
1010         }
1011         return 1;
1012 }
1013
1014 /*
1015  * look for inline back ref. if back ref is found, *ref_ret is set
1016  * to the address of inline back ref, and 0 is returned.
1017  *
1018  * if back ref isn't found, *ref_ret is set to the address where it
1019  * should be inserted, and -ENOENT is returned.
1020  *
1021  * if insert is true and there are too many inline back refs, the path
1022  * points to the extent item, and -EAGAIN is returned.
1023  *
1024  * NOTE: inline back refs are ordered in the same way that back ref
1025  *       items in the tree are ordered.
1026  */
1027 static noinline_for_stack
1028 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1029                                  struct btrfs_root *root,
1030                                  struct btrfs_path *path,
1031                                  struct btrfs_extent_inline_ref **ref_ret,
1032                                  u64 bytenr, u64 num_bytes,
1033                                  u64 parent, u64 root_objectid,
1034                                  u64 owner, u64 offset, int insert)
1035 {
1036         struct btrfs_key key;
1037         struct extent_buffer *leaf;
1038         struct btrfs_extent_item *ei;
1039         struct btrfs_extent_inline_ref *iref;
1040         u64 flags;
1041         u64 item_size;
1042         unsigned long ptr;
1043         unsigned long end;
1044         int extra_size;
1045         int type;
1046         int want;
1047         int ret;
1048         int err = 0;
1049
1050         key.objectid = bytenr;
1051         key.type = BTRFS_EXTENT_ITEM_KEY;
1052         key.offset = num_bytes;
1053
1054         want = extent_ref_type(parent, owner);
1055         if (insert) {
1056                 extra_size = btrfs_extent_inline_ref_size(want);
1057                 path->keep_locks = 1;
1058         } else
1059                 extra_size = -1;
1060         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1061         if (ret < 0) {
1062                 err = ret;
1063                 goto out;
1064         }
1065         BUG_ON(ret);
1066
1067         leaf = path->nodes[0];
1068         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1069 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1070         if (item_size < sizeof(*ei)) {
1071                 if (!insert) {
1072                         err = -ENOENT;
1073                         goto out;
1074                 }
1075                 ret = convert_extent_item_v0(trans, root, path, owner,
1076                                              extra_size);
1077                 if (ret < 0) {
1078                         err = ret;
1079                         goto out;
1080                 }
1081                 leaf = path->nodes[0];
1082                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1083         }
1084 #endif
1085         BUG_ON(item_size < sizeof(*ei));
1086
1087         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1088         flags = btrfs_extent_flags(leaf, ei);
1089
1090         ptr = (unsigned long)(ei + 1);
1091         end = (unsigned long)ei + item_size;
1092
1093         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1094                 ptr += sizeof(struct btrfs_tree_block_info);
1095                 BUG_ON(ptr > end);
1096         } else {
1097                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1098         }
1099
1100         err = -ENOENT;
1101         while (1) {
1102                 if (ptr >= end) {
1103                         WARN_ON(ptr > end);
1104                         break;
1105                 }
1106                 iref = (struct btrfs_extent_inline_ref *)ptr;
1107                 type = btrfs_extent_inline_ref_type(leaf, iref);
1108                 if (want < type)
1109                         break;
1110                 if (want > type) {
1111                         ptr += btrfs_extent_inline_ref_size(type);
1112                         continue;
1113                 }
1114
1115                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1116                         struct btrfs_extent_data_ref *dref;
1117                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1118                         if (match_extent_data_ref(leaf, dref, root_objectid,
1119                                                   owner, offset)) {
1120                                 err = 0;
1121                                 break;
1122                         }
1123                         if (hash_extent_data_ref_item(leaf, dref) <
1124                             hash_extent_data_ref(root_objectid, owner, offset))
1125                                 break;
1126                 } else {
1127                         u64 ref_offset;
1128                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1129                         if (parent > 0) {
1130                                 if (parent == ref_offset) {
1131                                         err = 0;
1132                                         break;
1133                                 }
1134                                 if (ref_offset < parent)
1135                                         break;
1136                         } else {
1137                                 if (root_objectid == ref_offset) {
1138                                         err = 0;
1139                                         break;
1140                                 }
1141                                 if (ref_offset < root_objectid)
1142                                         break;
1143                         }
1144                 }
1145                 ptr += btrfs_extent_inline_ref_size(type);
1146         }
1147         if (err == -ENOENT && insert) {
1148                 if (item_size + extra_size >=
1149                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1150                         err = -EAGAIN;
1151                         goto out;
1152                 }
1153                 /*
1154                  * To add new inline back ref, we have to make sure
1155                  * there is no corresponding back ref item.
1156                  * For simplicity, we just do not add new inline back
1157                  * ref if there is any kind of item for this block
1158                  */
1159                 if (find_next_key(path, 0, &key) == 0 &&
1160                     key.objectid == bytenr &&
1161                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1162                         err = -EAGAIN;
1163                         goto out;
1164                 }
1165         }
1166         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1167 out:
1168         if (insert) {
1169                 path->keep_locks = 0;
1170                 btrfs_unlock_up_safe(path, 1);
1171         }
1172         return err;
1173 }
1174
1175 /*
1176  * helper to add new inline back ref
1177  */
1178 static noinline_for_stack
1179 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1180                                 struct btrfs_root *root,
1181                                 struct btrfs_path *path,
1182                                 struct btrfs_extent_inline_ref *iref,
1183                                 u64 parent, u64 root_objectid,
1184                                 u64 owner, u64 offset, int refs_to_add,
1185                                 struct btrfs_delayed_extent_op *extent_op)
1186 {
1187         struct extent_buffer *leaf;
1188         struct btrfs_extent_item *ei;
1189         unsigned long ptr;
1190         unsigned long end;
1191         unsigned long item_offset;
1192         u64 refs;
1193         int size;
1194         int type;
1195         int ret;
1196
1197         leaf = path->nodes[0];
1198         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1199         item_offset = (unsigned long)iref - (unsigned long)ei;
1200
1201         type = extent_ref_type(parent, owner);
1202         size = btrfs_extent_inline_ref_size(type);
1203
1204         ret = btrfs_extend_item(trans, root, path, size);
1205         BUG_ON(ret);
1206
1207         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1208         refs = btrfs_extent_refs(leaf, ei);
1209         refs += refs_to_add;
1210         btrfs_set_extent_refs(leaf, ei, refs);
1211         if (extent_op)
1212                 __run_delayed_extent_op(extent_op, leaf, ei);
1213
1214         ptr = (unsigned long)ei + item_offset;
1215         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1216         if (ptr < end - size)
1217                 memmove_extent_buffer(leaf, ptr + size, ptr,
1218                                       end - size - ptr);
1219
1220         iref = (struct btrfs_extent_inline_ref *)ptr;
1221         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1222         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1223                 struct btrfs_extent_data_ref *dref;
1224                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1225                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1226                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1227                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1228                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1229         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1230                 struct btrfs_shared_data_ref *sref;
1231                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1232                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1233                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1234         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1235                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1236         } else {
1237                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1238         }
1239         btrfs_mark_buffer_dirty(leaf);
1240         return 0;
1241 }
1242
1243 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1244                                  struct btrfs_root *root,
1245                                  struct btrfs_path *path,
1246                                  struct btrfs_extent_inline_ref **ref_ret,
1247                                  u64 bytenr, u64 num_bytes, u64 parent,
1248                                  u64 root_objectid, u64 owner, u64 offset)
1249 {
1250         int ret;
1251
1252         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1253                                            bytenr, num_bytes, parent,
1254                                            root_objectid, owner, offset, 0);
1255         if (ret != -ENOENT)
1256                 return ret;
1257
1258         btrfs_release_path(root, path);
1259         *ref_ret = NULL;
1260
1261         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1262                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1263                                             root_objectid);
1264         } else {
1265                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1266                                              root_objectid, owner, offset);
1267         }
1268         return ret;
1269 }
1270
1271 /*
1272  * helper to update/remove inline back ref
1273  */
1274 static noinline_for_stack
1275 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1276                                  struct btrfs_root *root,
1277                                  struct btrfs_path *path,
1278                                  struct btrfs_extent_inline_ref *iref,
1279                                  int refs_to_mod,
1280                                  struct btrfs_delayed_extent_op *extent_op)
1281 {
1282         struct extent_buffer *leaf;
1283         struct btrfs_extent_item *ei;
1284         struct btrfs_extent_data_ref *dref = NULL;
1285         struct btrfs_shared_data_ref *sref = NULL;
1286         unsigned long ptr;
1287         unsigned long end;
1288         u32 item_size;
1289         int size;
1290         int type;
1291         int ret;
1292         u64 refs;
1293
1294         leaf = path->nodes[0];
1295         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1296         refs = btrfs_extent_refs(leaf, ei);
1297         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1298         refs += refs_to_mod;
1299         btrfs_set_extent_refs(leaf, ei, refs);
1300         if (extent_op)
1301                 __run_delayed_extent_op(extent_op, leaf, ei);
1302
1303         type = btrfs_extent_inline_ref_type(leaf, iref);
1304
1305         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1306                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1307                 refs = btrfs_extent_data_ref_count(leaf, dref);
1308         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1309                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1310                 refs = btrfs_shared_data_ref_count(leaf, sref);
1311         } else {
1312                 refs = 1;
1313                 BUG_ON(refs_to_mod != -1);
1314         }
1315
1316         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1317         refs += refs_to_mod;
1318
1319         if (refs > 0) {
1320                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1321                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1322                 else
1323                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1324         } else {
1325                 size =  btrfs_extent_inline_ref_size(type);
1326                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1327                 ptr = (unsigned long)iref;
1328                 end = (unsigned long)ei + item_size;
1329                 if (ptr + size < end)
1330                         memmove_extent_buffer(leaf, ptr, ptr + size,
1331                                               end - ptr - size);
1332                 item_size -= size;
1333                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1334                 BUG_ON(ret);
1335         }
1336         btrfs_mark_buffer_dirty(leaf);
1337         return 0;
1338 }
1339
1340 static noinline_for_stack
1341 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1342                                  struct btrfs_root *root,
1343                                  struct btrfs_path *path,
1344                                  u64 bytenr, u64 num_bytes, u64 parent,
1345                                  u64 root_objectid, u64 owner,
1346                                  u64 offset, int refs_to_add,
1347                                  struct btrfs_delayed_extent_op *extent_op)
1348 {
1349         struct btrfs_extent_inline_ref *iref;
1350         int ret;
1351
1352         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1353                                            bytenr, num_bytes, parent,
1354                                            root_objectid, owner, offset, 1);
1355         if (ret == 0) {
1356                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1357                 ret = update_inline_extent_backref(trans, root, path, iref,
1358                                                    refs_to_add, extent_op);
1359         } else if (ret == -ENOENT) {
1360                 ret = setup_inline_extent_backref(trans, root, path, iref,
1361                                                   parent, root_objectid,
1362                                                   owner, offset, refs_to_add,
1363                                                   extent_op);
1364         }
1365         return ret;
1366 }
1367
1368 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1369                                  struct btrfs_root *root,
1370                                  struct btrfs_path *path,
1371                                  u64 bytenr, u64 parent, u64 root_objectid,
1372                                  u64 owner, u64 offset, int refs_to_add)
1373 {
1374         int ret;
1375         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1376                 BUG_ON(refs_to_add != 1);
1377                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1378                                             parent, root_objectid);
1379         } else {
1380                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1381                                              parent, root_objectid,
1382                                              owner, offset, refs_to_add);
1383         }
1384         return ret;
1385 }
1386
1387 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1388                                  struct btrfs_root *root,
1389                                  struct btrfs_path *path,
1390                                  struct btrfs_extent_inline_ref *iref,
1391                                  int refs_to_drop, int is_data)
1392 {
1393         int ret;
1394
1395         BUG_ON(!is_data && refs_to_drop != 1);
1396         if (iref) {
1397                 ret = update_inline_extent_backref(trans, root, path, iref,
1398                                                    -refs_to_drop, NULL);
1399         } else if (is_data) {
1400                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1401         } else {
1402                 ret = btrfs_del_item(trans, root, path);
1403         }
1404         return ret;
1405 }
1406
1407 #ifdef BIO_RW_DISCARD
1408 static void btrfs_issue_discard(struct block_device *bdev,
1409                                 u64 start, u64 len)
1410 {
1411         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1412 }
1413 #endif
1414
1415 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1416                                 u64 num_bytes)
1417 {
1418 #ifdef BIO_RW_DISCARD
1419         int ret;
1420         u64 map_length = num_bytes;
1421         struct btrfs_multi_bio *multi = NULL;
1422
1423         /* Tell the block device(s) that the sectors can be discarded */
1424         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1425                               bytenr, &map_length, &multi, 0);
1426         if (!ret) {
1427                 struct btrfs_bio_stripe *stripe = multi->stripes;
1428                 int i;
1429
1430                 if (map_length > num_bytes)
1431                         map_length = num_bytes;
1432
1433                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1434                         btrfs_issue_discard(stripe->dev->bdev,
1435                                             stripe->physical,
1436                                             map_length);
1437                 }
1438                 kfree(multi);
1439         }
1440
1441         return ret;
1442 #else
1443         return 0;
1444 #endif
1445 }
1446
1447 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1448                          struct btrfs_root *root,
1449                          u64 bytenr, u64 num_bytes, u64 parent,
1450                          u64 root_objectid, u64 owner, u64 offset)
1451 {
1452         int ret;
1453         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1454                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1455
1456         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1457                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1458                                         parent, root_objectid, (int)owner,
1459                                         BTRFS_ADD_DELAYED_REF, NULL);
1460         } else {
1461                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1462                                         parent, root_objectid, owner, offset,
1463                                         BTRFS_ADD_DELAYED_REF, NULL);
1464         }
1465         return ret;
1466 }
1467
1468 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1469                                   struct btrfs_root *root,
1470                                   u64 bytenr, u64 num_bytes,
1471                                   u64 parent, u64 root_objectid,
1472                                   u64 owner, u64 offset, int refs_to_add,
1473                                   struct btrfs_delayed_extent_op *extent_op)
1474 {
1475         struct btrfs_path *path;
1476         struct extent_buffer *leaf;
1477         struct btrfs_extent_item *item;
1478         u64 refs;
1479         int ret;
1480         int err = 0;
1481
1482         path = btrfs_alloc_path();
1483         if (!path)
1484                 return -ENOMEM;
1485
1486         path->reada = 1;
1487         path->leave_spinning = 1;
1488         /* this will setup the path even if it fails to insert the back ref */
1489         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1490                                            path, bytenr, num_bytes, parent,
1491                                            root_objectid, owner, offset,
1492                                            refs_to_add, extent_op);
1493         if (ret == 0)
1494                 goto out;
1495
1496         if (ret != -EAGAIN) {
1497                 err = ret;
1498                 goto out;
1499         }
1500
1501         leaf = path->nodes[0];
1502         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1503         refs = btrfs_extent_refs(leaf, item);
1504         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1505         if (extent_op)
1506                 __run_delayed_extent_op(extent_op, leaf, item);
1507
1508         btrfs_mark_buffer_dirty(leaf);
1509         btrfs_release_path(root->fs_info->extent_root, path);
1510
1511         path->reada = 1;
1512         path->leave_spinning = 1;
1513
1514         /* now insert the actual backref */
1515         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1516                                     path, bytenr, parent, root_objectid,
1517                                     owner, offset, refs_to_add);
1518         BUG_ON(ret);
1519 out:
1520         btrfs_free_path(path);
1521         return err;
1522 }
1523
1524 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1525                                 struct btrfs_root *root,
1526                                 struct btrfs_delayed_ref_node *node,
1527                                 struct btrfs_delayed_extent_op *extent_op,
1528                                 int insert_reserved)
1529 {
1530         int ret = 0;
1531         struct btrfs_delayed_data_ref *ref;
1532         struct btrfs_key ins;
1533         u64 parent = 0;
1534         u64 ref_root = 0;
1535         u64 flags = 0;
1536
1537         ins.objectid = node->bytenr;
1538         ins.offset = node->num_bytes;
1539         ins.type = BTRFS_EXTENT_ITEM_KEY;
1540
1541         ref = btrfs_delayed_node_to_data_ref(node);
1542         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1543                 parent = ref->parent;
1544         else
1545                 ref_root = ref->root;
1546
1547         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1548                 if (extent_op) {
1549                         BUG_ON(extent_op->update_key);
1550                         flags |= extent_op->flags_to_set;
1551                 }
1552                 ret = alloc_reserved_file_extent(trans, root,
1553                                                  parent, ref_root, flags,
1554                                                  ref->objectid, ref->offset,
1555                                                  &ins, node->ref_mod);
1556                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1557         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1558                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1559                                              node->num_bytes, parent,
1560                                              ref_root, ref->objectid,
1561                                              ref->offset, node->ref_mod,
1562                                              extent_op);
1563         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1564                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1565                                           node->num_bytes, parent,
1566                                           ref_root, ref->objectid,
1567                                           ref->offset, node->ref_mod,
1568                                           extent_op);
1569         } else {
1570                 BUG();
1571         }
1572         return ret;
1573 }
1574
1575 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1576                                     struct extent_buffer *leaf,
1577                                     struct btrfs_extent_item *ei)
1578 {
1579         u64 flags = btrfs_extent_flags(leaf, ei);
1580         if (extent_op->update_flags) {
1581                 flags |= extent_op->flags_to_set;
1582                 btrfs_set_extent_flags(leaf, ei, flags);
1583         }
1584
1585         if (extent_op->update_key) {
1586                 struct btrfs_tree_block_info *bi;
1587                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1588                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1589                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1590         }
1591 }
1592
1593 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1594                                  struct btrfs_root *root,
1595                                  struct btrfs_delayed_ref_node *node,
1596                                  struct btrfs_delayed_extent_op *extent_op)
1597 {
1598         struct btrfs_key key;
1599         struct btrfs_path *path;
1600         struct btrfs_extent_item *ei;
1601         struct extent_buffer *leaf;
1602         u32 item_size;
1603         int ret;
1604         int err = 0;
1605
1606         path = btrfs_alloc_path();
1607         if (!path)
1608                 return -ENOMEM;
1609
1610         key.objectid = node->bytenr;
1611         key.type = BTRFS_EXTENT_ITEM_KEY;
1612         key.offset = node->num_bytes;
1613
1614         path->reada = 1;
1615         path->leave_spinning = 1;
1616         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1617                                 path, 0, 1);
1618         if (ret < 0) {
1619                 err = ret;
1620                 goto out;
1621         }
1622         if (ret > 0) {
1623                 err = -EIO;
1624                 goto out;
1625         }
1626
1627         leaf = path->nodes[0];
1628         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1629 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1630         if (item_size < sizeof(*ei)) {
1631                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1632                                              path, (u64)-1, 0);
1633                 if (ret < 0) {
1634                         err = ret;
1635                         goto out;
1636                 }
1637                 leaf = path->nodes[0];
1638                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1639         }
1640 #endif
1641         BUG_ON(item_size < sizeof(*ei));
1642         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1643         __run_delayed_extent_op(extent_op, leaf, ei);
1644
1645         btrfs_mark_buffer_dirty(leaf);
1646 out:
1647         btrfs_free_path(path);
1648         return err;
1649 }
1650
1651 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1652                                 struct btrfs_root *root,
1653                                 struct btrfs_delayed_ref_node *node,
1654                                 struct btrfs_delayed_extent_op *extent_op,
1655                                 int insert_reserved)
1656 {
1657         int ret = 0;
1658         struct btrfs_delayed_tree_ref *ref;
1659         struct btrfs_key ins;
1660         u64 parent = 0;
1661         u64 ref_root = 0;
1662
1663         ins.objectid = node->bytenr;
1664         ins.offset = node->num_bytes;
1665         ins.type = BTRFS_EXTENT_ITEM_KEY;
1666
1667         ref = btrfs_delayed_node_to_tree_ref(node);
1668         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1669                 parent = ref->parent;
1670         else
1671                 ref_root = ref->root;
1672
1673         BUG_ON(node->ref_mod != 1);
1674         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1675                 BUG_ON(!extent_op || !extent_op->update_flags ||
1676                        !extent_op->update_key);
1677                 ret = alloc_reserved_tree_block(trans, root,
1678                                                 parent, ref_root,
1679                                                 extent_op->flags_to_set,
1680                                                 &extent_op->key,
1681                                                 ref->level, &ins);
1682                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1683         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1684                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1685                                              node->num_bytes, parent, ref_root,
1686                                              ref->level, 0, 1, extent_op);
1687         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1688                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1689                                           node->num_bytes, parent, ref_root,
1690                                           ref->level, 0, 1, extent_op);
1691         } else {
1692                 BUG();
1693         }
1694         return ret;
1695 }
1696
1697
1698 /* helper function to actually process a single delayed ref entry */
1699 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1700                                struct btrfs_root *root,
1701                                struct btrfs_delayed_ref_node *node,
1702                                struct btrfs_delayed_extent_op *extent_op,
1703                                int insert_reserved)
1704 {
1705         int ret;
1706         if (btrfs_delayed_ref_is_head(node)) {
1707                 struct btrfs_delayed_ref_head *head;
1708                 /*
1709                  * we've hit the end of the chain and we were supposed
1710                  * to insert this extent into the tree.  But, it got
1711                  * deleted before we ever needed to insert it, so all
1712                  * we have to do is clean up the accounting
1713                  */
1714                 BUG_ON(extent_op);
1715                 head = btrfs_delayed_node_to_head(node);
1716                 if (insert_reserved) {
1717                         if (head->is_data) {
1718                                 ret = btrfs_del_csums(trans, root,
1719                                                       node->bytenr,
1720                                                       node->num_bytes);
1721                                 BUG_ON(ret);
1722                         }
1723                         btrfs_update_pinned_extents(root, node->bytenr,
1724                                                     node->num_bytes, 1);
1725                         update_reserved_extents(root, node->bytenr,
1726                                                 node->num_bytes, 0);
1727                 }
1728                 mutex_unlock(&head->mutex);
1729                 return 0;
1730         }
1731
1732         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1733             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1734                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1735                                            insert_reserved);
1736         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1737                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1738                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1739                                            insert_reserved);
1740         else
1741                 BUG();
1742         return ret;
1743 }
1744
1745 static noinline struct btrfs_delayed_ref_node *
1746 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1747 {
1748         struct rb_node *node;
1749         struct btrfs_delayed_ref_node *ref;
1750         int action = BTRFS_ADD_DELAYED_REF;
1751 again:
1752         /*
1753          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1754          * this prevents ref count from going down to zero when
1755          * there still are pending delayed ref.
1756          */
1757         node = rb_prev(&head->node.rb_node);
1758         while (1) {
1759                 if (!node)
1760                         break;
1761                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1762                                 rb_node);
1763                 if (ref->bytenr != head->node.bytenr)
1764                         break;
1765                 if (ref->action == action)
1766                         return ref;
1767                 node = rb_prev(node);
1768         }
1769         if (action == BTRFS_ADD_DELAYED_REF) {
1770                 action = BTRFS_DROP_DELAYED_REF;
1771                 goto again;
1772         }
1773         return NULL;
1774 }
1775
1776 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1777                                        struct btrfs_root *root,
1778                                        struct list_head *cluster)
1779 {
1780         struct btrfs_delayed_ref_root *delayed_refs;
1781         struct btrfs_delayed_ref_node *ref;
1782         struct btrfs_delayed_ref_head *locked_ref = NULL;
1783         struct btrfs_delayed_extent_op *extent_op;
1784         int ret;
1785         int count = 0;
1786         int must_insert_reserved = 0;
1787
1788         delayed_refs = &trans->transaction->delayed_refs;
1789         while (1) {
1790                 if (!locked_ref) {
1791                         /* pick a new head ref from the cluster list */
1792                         if (list_empty(cluster))
1793                                 break;
1794
1795                         locked_ref = list_entry(cluster->next,
1796                                      struct btrfs_delayed_ref_head, cluster);
1797
1798                         /* grab the lock that says we are going to process
1799                          * all the refs for this head */
1800                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1801
1802                         /*
1803                          * we may have dropped the spin lock to get the head
1804                          * mutex lock, and that might have given someone else
1805                          * time to free the head.  If that's true, it has been
1806                          * removed from our list and we can move on.
1807                          */
1808                         if (ret == -EAGAIN) {
1809                                 locked_ref = NULL;
1810                                 count++;
1811                                 continue;
1812                         }
1813                 }
1814
1815                 /*
1816                  * record the must insert reserved flag before we
1817                  * drop the spin lock.
1818                  */
1819                 must_insert_reserved = locked_ref->must_insert_reserved;
1820                 locked_ref->must_insert_reserved = 0;
1821
1822                 extent_op = locked_ref->extent_op;
1823                 locked_ref->extent_op = NULL;
1824
1825                 /*
1826                  * locked_ref is the head node, so we have to go one
1827                  * node back for any delayed ref updates
1828                  */
1829                 ref = select_delayed_ref(locked_ref);
1830                 if (!ref) {
1831                         /* All delayed refs have been processed, Go ahead
1832                          * and send the head node to run_one_delayed_ref,
1833                          * so that any accounting fixes can happen
1834                          */
1835                         ref = &locked_ref->node;
1836
1837                         if (extent_op && must_insert_reserved) {
1838                                 kfree(extent_op);
1839                                 extent_op = NULL;
1840                         }
1841
1842                         if (extent_op) {
1843                                 spin_unlock(&delayed_refs->lock);
1844
1845                                 ret = run_delayed_extent_op(trans, root,
1846                                                             ref, extent_op);
1847                                 BUG_ON(ret);
1848                                 kfree(extent_op);
1849
1850                                 cond_resched();
1851                                 spin_lock(&delayed_refs->lock);
1852                                 continue;
1853                         }
1854
1855                         list_del_init(&locked_ref->cluster);
1856                         locked_ref = NULL;
1857                 }
1858
1859                 ref->in_tree = 0;
1860                 rb_erase(&ref->rb_node, &delayed_refs->root);
1861                 delayed_refs->num_entries--;
1862
1863                 spin_unlock(&delayed_refs->lock);
1864
1865                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
1866                                           must_insert_reserved);
1867                 BUG_ON(ret);
1868
1869                 btrfs_put_delayed_ref(ref);
1870                 kfree(extent_op);
1871                 count++;
1872
1873                 cond_resched();
1874                 spin_lock(&delayed_refs->lock);
1875         }
1876         return count;
1877 }
1878
1879 /*
1880  * this starts processing the delayed reference count updates and
1881  * extent insertions we have queued up so far.  count can be
1882  * 0, which means to process everything in the tree at the start
1883  * of the run (but not newly added entries), or it can be some target
1884  * number you'd like to process.
1885  */
1886 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1887                            struct btrfs_root *root, unsigned long count)
1888 {
1889         struct rb_node *node;
1890         struct btrfs_delayed_ref_root *delayed_refs;
1891         struct btrfs_delayed_ref_node *ref;
1892         struct list_head cluster;
1893         int ret;
1894         int run_all = count == (unsigned long)-1;
1895         int run_most = 0;
1896
1897         if (root == root->fs_info->extent_root)
1898                 root = root->fs_info->tree_root;
1899
1900         delayed_refs = &trans->transaction->delayed_refs;
1901         INIT_LIST_HEAD(&cluster);
1902 again:
1903         spin_lock(&delayed_refs->lock);
1904         if (count == 0) {
1905                 count = delayed_refs->num_entries * 2;
1906                 run_most = 1;
1907         }
1908         while (1) {
1909                 if (!(run_all || run_most) &&
1910                     delayed_refs->num_heads_ready < 64)
1911                         break;
1912
1913                 /*
1914                  * go find something we can process in the rbtree.  We start at
1915                  * the beginning of the tree, and then build a cluster
1916                  * of refs to process starting at the first one we are able to
1917                  * lock
1918                  */
1919                 ret = btrfs_find_ref_cluster(trans, &cluster,
1920                                              delayed_refs->run_delayed_start);
1921                 if (ret)
1922                         break;
1923
1924                 ret = run_clustered_refs(trans, root, &cluster);
1925                 BUG_ON(ret < 0);
1926
1927                 count -= min_t(unsigned long, ret, count);
1928
1929                 if (count == 0)
1930                         break;
1931         }
1932
1933         if (run_all) {
1934                 node = rb_first(&delayed_refs->root);
1935                 if (!node)
1936                         goto out;
1937                 count = (unsigned long)-1;
1938
1939                 while (node) {
1940                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
1941                                        rb_node);
1942                         if (btrfs_delayed_ref_is_head(ref)) {
1943                                 struct btrfs_delayed_ref_head *head;
1944
1945                                 head = btrfs_delayed_node_to_head(ref);
1946                                 atomic_inc(&ref->refs);
1947
1948                                 spin_unlock(&delayed_refs->lock);
1949                                 mutex_lock(&head->mutex);
1950                                 mutex_unlock(&head->mutex);
1951
1952                                 btrfs_put_delayed_ref(ref);
1953                                 cond_resched();
1954                                 goto again;
1955                         }
1956                         node = rb_next(node);
1957                 }
1958                 spin_unlock(&delayed_refs->lock);
1959                 schedule_timeout(1);
1960                 goto again;
1961         }
1962 out:
1963         spin_unlock(&delayed_refs->lock);
1964         return 0;
1965 }
1966
1967 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
1968                                 struct btrfs_root *root,
1969                                 u64 bytenr, u64 num_bytes, u64 flags,
1970                                 int is_data)
1971 {
1972         struct btrfs_delayed_extent_op *extent_op;
1973         int ret;
1974
1975         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1976         if (!extent_op)
1977                 return -ENOMEM;
1978
1979         extent_op->flags_to_set = flags;
1980         extent_op->update_flags = 1;
1981         extent_op->update_key = 0;
1982         extent_op->is_data = is_data ? 1 : 0;
1983
1984         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
1985         if (ret)
1986                 kfree(extent_op);
1987         return ret;
1988 }
1989
1990 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
1991                                       struct btrfs_root *root,
1992                                       struct btrfs_path *path,
1993                                       u64 objectid, u64 offset, u64 bytenr)
1994 {
1995         struct btrfs_delayed_ref_head *head;
1996         struct btrfs_delayed_ref_node *ref;
1997         struct btrfs_delayed_data_ref *data_ref;
1998         struct btrfs_delayed_ref_root *delayed_refs;
1999         struct rb_node *node;
2000         int ret = 0;
2001
2002         ret = -ENOENT;
2003         delayed_refs = &trans->transaction->delayed_refs;
2004         spin_lock(&delayed_refs->lock);
2005         head = btrfs_find_delayed_ref_head(trans, bytenr);
2006         if (!head)
2007                 goto out;
2008
2009         if (!mutex_trylock(&head->mutex)) {
2010                 atomic_inc(&head->node.refs);
2011                 spin_unlock(&delayed_refs->lock);
2012
2013                 btrfs_release_path(root->fs_info->extent_root, path);
2014
2015                 mutex_lock(&head->mutex);
2016                 mutex_unlock(&head->mutex);
2017                 btrfs_put_delayed_ref(&head->node);
2018                 return -EAGAIN;
2019         }
2020
2021         node = rb_prev(&head->node.rb_node);
2022         if (!node)
2023                 goto out_unlock;
2024
2025         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2026
2027         if (ref->bytenr != bytenr)
2028                 goto out_unlock;
2029
2030         ret = 1;
2031         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2032                 goto out_unlock;
2033
2034         data_ref = btrfs_delayed_node_to_data_ref(ref);
2035
2036         node = rb_prev(node);
2037         if (node) {
2038                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2039                 if (ref->bytenr == bytenr)
2040                         goto out_unlock;
2041         }
2042
2043         if (data_ref->root != root->root_key.objectid ||
2044             data_ref->objectid != objectid || data_ref->offset != offset)
2045                 goto out_unlock;
2046
2047         ret = 0;
2048 out_unlock:
2049         mutex_unlock(&head->mutex);
2050 out:
2051         spin_unlock(&delayed_refs->lock);
2052         return ret;
2053 }
2054
2055 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2056                                         struct btrfs_root *root,
2057                                         struct btrfs_path *path,
2058                                         u64 objectid, u64 offset, u64 bytenr)
2059 {
2060         struct btrfs_root *extent_root = root->fs_info->extent_root;
2061         struct extent_buffer *leaf;
2062         struct btrfs_extent_data_ref *ref;
2063         struct btrfs_extent_inline_ref *iref;
2064         struct btrfs_extent_item *ei;
2065         struct btrfs_key key;
2066         u32 item_size;
2067         int ret;
2068
2069         key.objectid = bytenr;
2070         key.offset = (u64)-1;
2071         key.type = BTRFS_EXTENT_ITEM_KEY;
2072
2073         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2074         if (ret < 0)
2075                 goto out;
2076         BUG_ON(ret == 0);
2077
2078         ret = -ENOENT;
2079         if (path->slots[0] == 0)
2080                 goto out;
2081
2082         path->slots[0]--;
2083         leaf = path->nodes[0];
2084         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2085
2086         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2087                 goto out;
2088
2089         ret = 1;
2090         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2092         if (item_size < sizeof(*ei)) {
2093                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2094                 goto out;
2095         }
2096 #endif
2097         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2098
2099         if (item_size != sizeof(*ei) +
2100             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2101                 goto out;
2102
2103         if (btrfs_extent_generation(leaf, ei) <=
2104             btrfs_root_last_snapshot(&root->root_item))
2105                 goto out;
2106
2107         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2108         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2109             BTRFS_EXTENT_DATA_REF_KEY)
2110                 goto out;
2111
2112         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2113         if (btrfs_extent_refs(leaf, ei) !=
2114             btrfs_extent_data_ref_count(leaf, ref) ||
2115             btrfs_extent_data_ref_root(leaf, ref) !=
2116             root->root_key.objectid ||
2117             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2118             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2119                 goto out;
2120
2121         ret = 0;
2122 out:
2123         return ret;
2124 }
2125
2126 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2127                           struct btrfs_root *root,
2128                           u64 objectid, u64 offset, u64 bytenr)
2129 {
2130         struct btrfs_path *path;
2131         int ret;
2132         int ret2;
2133
2134         path = btrfs_alloc_path();
2135         if (!path)
2136                 return -ENOENT;
2137
2138         do {
2139                 ret = check_committed_ref(trans, root, path, objectid,
2140                                           offset, bytenr);
2141                 if (ret && ret != -ENOENT)
2142                         goto out;
2143
2144                 ret2 = check_delayed_ref(trans, root, path, objectid,
2145                                          offset, bytenr);
2146         } while (ret2 == -EAGAIN);
2147
2148         if (ret2 && ret2 != -ENOENT) {
2149                 ret = ret2;
2150                 goto out;
2151         }
2152
2153         if (ret != -ENOENT || ret2 != -ENOENT)
2154                 ret = 0;
2155 out:
2156         btrfs_free_path(path);
2157         return ret;
2158 }
2159
2160 #if 0
2161 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2162                     struct extent_buffer *buf, u32 nr_extents)
2163 {
2164         struct btrfs_key key;
2165         struct btrfs_file_extent_item *fi;
2166         u64 root_gen;
2167         u32 nritems;
2168         int i;
2169         int level;
2170         int ret = 0;
2171         int shared = 0;
2172
2173         if (!root->ref_cows)
2174                 return 0;
2175
2176         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2177                 shared = 0;
2178                 root_gen = root->root_key.offset;
2179         } else {
2180                 shared = 1;
2181                 root_gen = trans->transid - 1;
2182         }
2183
2184         level = btrfs_header_level(buf);
2185         nritems = btrfs_header_nritems(buf);
2186
2187         if (level == 0) {
2188                 struct btrfs_leaf_ref *ref;
2189                 struct btrfs_extent_info *info;
2190
2191                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2192                 if (!ref) {
2193                         ret = -ENOMEM;
2194                         goto out;
2195                 }
2196
2197                 ref->root_gen = root_gen;
2198                 ref->bytenr = buf->start;
2199                 ref->owner = btrfs_header_owner(buf);
2200                 ref->generation = btrfs_header_generation(buf);
2201                 ref->nritems = nr_extents;
2202                 info = ref->extents;
2203
2204                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2205                         u64 disk_bytenr;
2206                         btrfs_item_key_to_cpu(buf, &key, i);
2207                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2208                                 continue;
2209                         fi = btrfs_item_ptr(buf, i,
2210                                             struct btrfs_file_extent_item);
2211                         if (btrfs_file_extent_type(buf, fi) ==
2212                             BTRFS_FILE_EXTENT_INLINE)
2213                                 continue;
2214                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2215                         if (disk_bytenr == 0)
2216                                 continue;
2217
2218                         info->bytenr = disk_bytenr;
2219                         info->num_bytes =
2220                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2221                         info->objectid = key.objectid;
2222                         info->offset = key.offset;
2223                         info++;
2224                 }
2225
2226                 ret = btrfs_add_leaf_ref(root, ref, shared);
2227                 if (ret == -EEXIST && shared) {
2228                         struct btrfs_leaf_ref *old;
2229                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2230                         BUG_ON(!old);
2231                         btrfs_remove_leaf_ref(root, old);
2232                         btrfs_free_leaf_ref(root, old);
2233                         ret = btrfs_add_leaf_ref(root, ref, shared);
2234                 }
2235                 WARN_ON(ret);
2236                 btrfs_free_leaf_ref(root, ref);
2237         }
2238 out:
2239         return ret;
2240 }
2241
2242 /* when a block goes through cow, we update the reference counts of
2243  * everything that block points to.  The internal pointers of the block
2244  * can be in just about any order, and it is likely to have clusters of
2245  * things that are close together and clusters of things that are not.
2246  *
2247  * To help reduce the seeks that come with updating all of these reference
2248  * counts, sort them by byte number before actual updates are done.
2249  *
2250  * struct refsort is used to match byte number to slot in the btree block.
2251  * we sort based on the byte number and then use the slot to actually
2252  * find the item.
2253  *
2254  * struct refsort is smaller than strcut btrfs_item and smaller than
2255  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2256  * for a btree block, there's no way for a kmalloc of refsorts for a
2257  * single node to be bigger than a page.
2258  */
2259 struct refsort {
2260         u64 bytenr;
2261         u32 slot;
2262 };
2263
2264 /*
2265  * for passing into sort()
2266  */
2267 static int refsort_cmp(const void *a_void, const void *b_void)
2268 {
2269         const struct refsort *a = a_void;
2270         const struct refsort *b = b_void;
2271
2272         if (a->bytenr < b->bytenr)
2273                 return -1;
2274         if (a->bytenr > b->bytenr)
2275                 return 1;
2276         return 0;
2277 }
2278 #endif
2279
2280 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2281                            struct btrfs_root *root,
2282                            struct extent_buffer *buf,
2283                            int full_backref, int inc)
2284 {
2285         u64 bytenr;
2286         u64 num_bytes;
2287         u64 parent;
2288         u64 ref_root;
2289         u32 nritems;
2290         struct btrfs_key key;
2291         struct btrfs_file_extent_item *fi;
2292         int i;
2293         int level;
2294         int ret = 0;
2295         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2296                             u64, u64, u64, u64, u64, u64);
2297
2298         ref_root = btrfs_header_owner(buf);
2299         nritems = btrfs_header_nritems(buf);
2300         level = btrfs_header_level(buf);
2301
2302         if (!root->ref_cows && level == 0)
2303                 return 0;
2304
2305         if (inc)
2306                 process_func = btrfs_inc_extent_ref;
2307         else
2308                 process_func = btrfs_free_extent;
2309
2310         if (full_backref)
2311                 parent = buf->start;
2312         else
2313                 parent = 0;
2314
2315         for (i = 0; i < nritems; i++) {
2316                 if (level == 0) {
2317                         btrfs_item_key_to_cpu(buf, &key, i);
2318                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2319                                 continue;
2320                         fi = btrfs_item_ptr(buf, i,
2321                                             struct btrfs_file_extent_item);
2322                         if (btrfs_file_extent_type(buf, fi) ==
2323                             BTRFS_FILE_EXTENT_INLINE)
2324                                 continue;
2325                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2326                         if (bytenr == 0)
2327                                 continue;
2328
2329                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2330                         key.offset -= btrfs_file_extent_offset(buf, fi);
2331                         ret = process_func(trans, root, bytenr, num_bytes,
2332                                            parent, ref_root, key.objectid,
2333                                            key.offset);
2334                         if (ret)
2335                                 goto fail;
2336                 } else {
2337                         bytenr = btrfs_node_blockptr(buf, i);
2338                         num_bytes = btrfs_level_size(root, level - 1);
2339                         ret = process_func(trans, root, bytenr, num_bytes,
2340                                            parent, ref_root, level - 1, 0);
2341                         if (ret)
2342                                 goto fail;
2343                 }
2344         }
2345         return 0;
2346 fail:
2347         BUG();
2348         return ret;
2349 }
2350
2351 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2352                   struct extent_buffer *buf, int full_backref)
2353 {
2354         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2355 }
2356
2357 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2358                   struct extent_buffer *buf, int full_backref)
2359 {
2360         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2361 }
2362
2363 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2364                                  struct btrfs_root *root,
2365                                  struct btrfs_path *path,
2366                                  struct btrfs_block_group_cache *cache)
2367 {
2368         int ret;
2369         struct btrfs_root *extent_root = root->fs_info->extent_root;
2370         unsigned long bi;
2371         struct extent_buffer *leaf;
2372
2373         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2374         if (ret < 0)
2375                 goto fail;
2376         BUG_ON(ret);
2377
2378         leaf = path->nodes[0];
2379         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2380         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2381         btrfs_mark_buffer_dirty(leaf);
2382         btrfs_release_path(extent_root, path);
2383 fail:
2384         if (ret)
2385                 return ret;
2386         return 0;
2387
2388 }
2389
2390 static struct btrfs_block_group_cache *
2391 next_block_group(struct btrfs_root *root,
2392                  struct btrfs_block_group_cache *cache)
2393 {
2394         struct rb_node *node;
2395         spin_lock(&root->fs_info->block_group_cache_lock);
2396         node = rb_next(&cache->cache_node);
2397         btrfs_put_block_group(cache);
2398         if (node) {
2399                 cache = rb_entry(node, struct btrfs_block_group_cache,
2400                                  cache_node);
2401                 atomic_inc(&cache->count);
2402         } else
2403                 cache = NULL;
2404         spin_unlock(&root->fs_info->block_group_cache_lock);
2405         return cache;
2406 }
2407
2408 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2409                                    struct btrfs_root *root)
2410 {
2411         struct btrfs_block_group_cache *cache;
2412         int err = 0;
2413         struct btrfs_path *path;
2414         u64 last = 0;
2415
2416         path = btrfs_alloc_path();
2417         if (!path)
2418                 return -ENOMEM;
2419
2420         while (1) {
2421                 if (last == 0) {
2422                         err = btrfs_run_delayed_refs(trans, root,
2423                                                      (unsigned long)-1);
2424                         BUG_ON(err);
2425                 }
2426
2427                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2428                 while (cache) {
2429                         if (cache->dirty)
2430                                 break;
2431                         cache = next_block_group(root, cache);
2432                 }
2433                 if (!cache) {
2434                         if (last == 0)
2435                                 break;
2436                         last = 0;
2437                         continue;
2438                 }
2439
2440                 cache->dirty = 0;
2441                 last = cache->key.objectid + cache->key.offset;
2442
2443                 err = write_one_cache_group(trans, root, path, cache);
2444                 BUG_ON(err);
2445                 btrfs_put_block_group(cache);
2446         }
2447
2448         btrfs_free_path(path);
2449         return 0;
2450 }
2451
2452 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2453 {
2454         struct btrfs_block_group_cache *block_group;
2455         int readonly = 0;
2456
2457         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2458         if (!block_group || block_group->ro)
2459                 readonly = 1;
2460         if (block_group)
2461                 btrfs_put_block_group(block_group);
2462         return readonly;
2463 }
2464
2465 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2466                              u64 total_bytes, u64 bytes_used,
2467                              struct btrfs_space_info **space_info)
2468 {
2469         struct btrfs_space_info *found;
2470
2471         found = __find_space_info(info, flags);
2472         if (found) {
2473                 spin_lock(&found->lock);
2474                 found->total_bytes += total_bytes;
2475                 found->bytes_used += bytes_used;
2476                 found->full = 0;
2477                 spin_unlock(&found->lock);
2478                 *space_info = found;
2479                 return 0;
2480         }
2481         found = kzalloc(sizeof(*found), GFP_NOFS);
2482         if (!found)
2483                 return -ENOMEM;
2484
2485         INIT_LIST_HEAD(&found->block_groups);
2486         init_rwsem(&found->groups_sem);
2487         spin_lock_init(&found->lock);
2488         found->flags = flags;
2489         found->total_bytes = total_bytes;
2490         found->bytes_used = bytes_used;
2491         found->bytes_pinned = 0;
2492         found->bytes_reserved = 0;
2493         found->bytes_readonly = 0;
2494         found->bytes_delalloc = 0;
2495         found->full = 0;
2496         found->force_alloc = 0;
2497         *space_info = found;
2498         list_add_rcu(&found->list, &info->space_info);
2499         return 0;
2500 }
2501
2502 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2503 {
2504         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2505                                    BTRFS_BLOCK_GROUP_RAID1 |
2506                                    BTRFS_BLOCK_GROUP_RAID10 |
2507                                    BTRFS_BLOCK_GROUP_DUP);
2508         if (extra_flags) {
2509                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2510                         fs_info->avail_data_alloc_bits |= extra_flags;
2511                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2512                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2513                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2514                         fs_info->avail_system_alloc_bits |= extra_flags;
2515         }
2516 }
2517
2518 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2519 {
2520         spin_lock(&cache->space_info->lock);
2521         spin_lock(&cache->lock);
2522         if (!cache->ro) {
2523                 cache->space_info->bytes_readonly += cache->key.offset -
2524                                         btrfs_block_group_used(&cache->item);
2525                 cache->ro = 1;
2526         }
2527         spin_unlock(&cache->lock);
2528         spin_unlock(&cache->space_info->lock);
2529 }
2530
2531 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2532 {
2533         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2534
2535         if (num_devices == 1)
2536                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2537         if (num_devices < 4)
2538                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2539
2540         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2541             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2542                       BTRFS_BLOCK_GROUP_RAID10))) {
2543                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2544         }
2545
2546         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2547             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2548                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2549         }
2550
2551         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2552             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2553              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2554              (flags & BTRFS_BLOCK_GROUP_DUP)))
2555                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2556         return flags;
2557 }
2558
2559 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2560 {
2561         struct btrfs_fs_info *info = root->fs_info;
2562         u64 alloc_profile;
2563
2564         if (data) {
2565                 alloc_profile = info->avail_data_alloc_bits &
2566                         info->data_alloc_profile;
2567                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2568         } else if (root == root->fs_info->chunk_root) {
2569                 alloc_profile = info->avail_system_alloc_bits &
2570                         info->system_alloc_profile;
2571                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2572         } else {
2573                 alloc_profile = info->avail_metadata_alloc_bits &
2574                         info->metadata_alloc_profile;
2575                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2576         }
2577
2578         return btrfs_reduce_alloc_profile(root, data);
2579 }
2580
2581 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2582 {
2583         u64 alloc_target;
2584
2585         alloc_target = btrfs_get_alloc_profile(root, 1);
2586         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2587                                                        alloc_target);
2588 }
2589
2590 /*
2591  * for now this just makes sure we have at least 5% of our metadata space free
2592  * for use.
2593  */
2594 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2595 {
2596         struct btrfs_fs_info *info = root->fs_info;
2597         struct btrfs_space_info *meta_sinfo;
2598         u64 alloc_target, thresh;
2599         int committed = 0, ret;
2600
2601         /* get the space info for where the metadata will live */
2602         alloc_target = btrfs_get_alloc_profile(root, 0);
2603         meta_sinfo = __find_space_info(info, alloc_target);
2604
2605 again:
2606         spin_lock(&meta_sinfo->lock);
2607         if (!meta_sinfo->full)
2608                 thresh = meta_sinfo->total_bytes * 80;
2609         else
2610                 thresh = meta_sinfo->total_bytes * 95;
2611
2612         do_div(thresh, 100);
2613
2614         if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2615             meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2616                 struct btrfs_trans_handle *trans;
2617                 if (!meta_sinfo->full) {
2618                         meta_sinfo->force_alloc = 1;
2619                         spin_unlock(&meta_sinfo->lock);
2620
2621                         trans = btrfs_start_transaction(root, 1);
2622                         if (!trans)
2623                                 return -ENOMEM;
2624
2625                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2626                                              2 * 1024 * 1024, alloc_target, 0);
2627                         btrfs_end_transaction(trans, root);
2628                         goto again;
2629                 }
2630                 spin_unlock(&meta_sinfo->lock);
2631
2632                 if (!committed) {
2633                         committed = 1;
2634                         trans = btrfs_join_transaction(root, 1);
2635                         if (!trans)
2636                                 return -ENOMEM;
2637                         ret = btrfs_commit_transaction(trans, root);
2638                         if (ret)
2639                                 return ret;
2640                         goto again;
2641                 }
2642                 return -ENOSPC;
2643         }
2644         spin_unlock(&meta_sinfo->lock);
2645
2646         return 0;
2647 }
2648
2649 /*
2650  * This will check the space that the inode allocates from to make sure we have
2651  * enough space for bytes.
2652  */
2653 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2654                                 u64 bytes)
2655 {
2656         struct btrfs_space_info *data_sinfo;
2657         int ret = 0, committed = 0;
2658
2659         /* make sure bytes are sectorsize aligned */
2660         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2661
2662         data_sinfo = BTRFS_I(inode)->space_info;
2663 again:
2664         /* make sure we have enough space to handle the data first */
2665         spin_lock(&data_sinfo->lock);
2666         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2667             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2668             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2669             data_sinfo->bytes_may_use < bytes) {
2670                 struct btrfs_trans_handle *trans;
2671
2672                 /*
2673                  * if we don't have enough free bytes in this space then we need
2674                  * to alloc a new chunk.
2675                  */
2676                 if (!data_sinfo->full) {
2677                         u64 alloc_target;
2678
2679                         data_sinfo->force_alloc = 1;
2680                         spin_unlock(&data_sinfo->lock);
2681
2682                         alloc_target = btrfs_get_alloc_profile(root, 1);
2683                         trans = btrfs_start_transaction(root, 1);
2684                         if (!trans)
2685                                 return -ENOMEM;
2686
2687                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2688                                              bytes + 2 * 1024 * 1024,
2689                                              alloc_target, 0);
2690                         btrfs_end_transaction(trans, root);
2691                         if (ret)
2692                                 return ret;
2693                         goto again;
2694                 }
2695                 spin_unlock(&data_sinfo->lock);
2696
2697                 /* commit the current transaction and try again */
2698                 if (!committed) {
2699                         committed = 1;
2700                         trans = btrfs_join_transaction(root, 1);
2701                         if (!trans)
2702                                 return -ENOMEM;
2703                         ret = btrfs_commit_transaction(trans, root);
2704                         if (ret)
2705                                 return ret;
2706                         goto again;
2707                 }
2708
2709                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2710                        ", %llu bytes_used, %llu bytes_reserved, "
2711                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2712                        "%llu total\n", (unsigned long long)bytes,
2713                        (unsigned long long)data_sinfo->bytes_delalloc,
2714                        (unsigned long long)data_sinfo->bytes_used,
2715                        (unsigned long long)data_sinfo->bytes_reserved,
2716                        (unsigned long long)data_sinfo->bytes_pinned,
2717                        (unsigned long long)data_sinfo->bytes_readonly,
2718                        (unsigned long long)data_sinfo->bytes_may_use,
2719                        (unsigned long long)data_sinfo->total_bytes);
2720                 return -ENOSPC;
2721         }
2722         data_sinfo->bytes_may_use += bytes;
2723         BTRFS_I(inode)->reserved_bytes += bytes;
2724         spin_unlock(&data_sinfo->lock);
2725
2726         return btrfs_check_metadata_free_space(root);
2727 }
2728
2729 /*
2730  * if there was an error for whatever reason after calling
2731  * btrfs_check_data_free_space, call this so we can cleanup the counters.
2732  */
2733 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2734                                     struct inode *inode, u64 bytes)
2735 {
2736         struct btrfs_space_info *data_sinfo;
2737
2738         /* make sure bytes are sectorsize aligned */
2739         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2740
2741         data_sinfo = BTRFS_I(inode)->space_info;
2742         spin_lock(&data_sinfo->lock);
2743         data_sinfo->bytes_may_use -= bytes;
2744         BTRFS_I(inode)->reserved_bytes -= bytes;
2745         spin_unlock(&data_sinfo->lock);
2746 }
2747
2748 /* called when we are adding a delalloc extent to the inode's io_tree */
2749 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2750                                   u64 bytes)
2751 {
2752         struct btrfs_space_info *data_sinfo;
2753
2754         /* get the space info for where this inode will be storing its data */
2755         data_sinfo = BTRFS_I(inode)->space_info;
2756
2757         /* make sure we have enough space to handle the data first */
2758         spin_lock(&data_sinfo->lock);
2759         data_sinfo->bytes_delalloc += bytes;
2760
2761         /*
2762          * we are adding a delalloc extent without calling
2763          * btrfs_check_data_free_space first.  This happens on a weird
2764          * writepage condition, but shouldn't hurt our accounting
2765          */
2766         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2767                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2768                 BTRFS_I(inode)->reserved_bytes = 0;
2769         } else {
2770                 data_sinfo->bytes_may_use -= bytes;
2771                 BTRFS_I(inode)->reserved_bytes -= bytes;
2772         }
2773
2774         spin_unlock(&data_sinfo->lock);
2775 }
2776
2777 /* called when we are clearing an delalloc extent from the inode's io_tree */
2778 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2779                               u64 bytes)
2780 {
2781         struct btrfs_space_info *info;
2782
2783         info = BTRFS_I(inode)->space_info;
2784
2785         spin_lock(&info->lock);
2786         info->bytes_delalloc -= bytes;
2787         spin_unlock(&info->lock);
2788 }
2789
2790 static void force_metadata_allocation(struct btrfs_fs_info *info)
2791 {
2792         struct list_head *head = &info->space_info;
2793         struct btrfs_space_info *found;
2794
2795         rcu_read_lock();
2796         list_for_each_entry_rcu(found, head, list) {
2797                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2798                         found->force_alloc = 1;
2799         }
2800         rcu_read_unlock();
2801 }
2802
2803 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2804                           struct btrfs_root *extent_root, u64 alloc_bytes,
2805                           u64 flags, int force)
2806 {
2807         struct btrfs_space_info *space_info;
2808         struct btrfs_fs_info *fs_info = extent_root->fs_info;
2809         u64 thresh;
2810         int ret = 0;
2811
2812         mutex_lock(&fs_info->chunk_mutex);
2813
2814         flags = btrfs_reduce_alloc_profile(extent_root, flags);
2815
2816         space_info = __find_space_info(extent_root->fs_info, flags);
2817         if (!space_info) {
2818                 ret = update_space_info(extent_root->fs_info, flags,
2819                                         0, 0, &space_info);
2820                 BUG_ON(ret);
2821         }
2822         BUG_ON(!space_info);
2823
2824         spin_lock(&space_info->lock);
2825         if (space_info->force_alloc) {
2826                 force = 1;
2827                 space_info->force_alloc = 0;
2828         }
2829         if (space_info->full) {
2830                 spin_unlock(&space_info->lock);
2831                 goto out;
2832         }
2833
2834         thresh = space_info->total_bytes - space_info->bytes_readonly;
2835         thresh = div_factor(thresh, 6);
2836         if (!force &&
2837            (space_info->bytes_used + space_info->bytes_pinned +
2838             space_info->bytes_reserved + alloc_bytes) < thresh) {
2839                 spin_unlock(&space_info->lock);
2840                 goto out;
2841         }
2842         spin_unlock(&space_info->lock);
2843
2844         /*
2845          * if we're doing a data chunk, go ahead and make sure that
2846          * we keep a reasonable number of metadata chunks allocated in the
2847          * FS as well.
2848          */
2849         if (flags & BTRFS_BLOCK_GROUP_DATA) {
2850                 fs_info->data_chunk_allocations++;
2851                 if (!(fs_info->data_chunk_allocations %
2852                       fs_info->metadata_ratio))
2853                         force_metadata_allocation(fs_info);
2854         }
2855
2856         ret = btrfs_alloc_chunk(trans, extent_root, flags);
2857         if (ret)
2858                 space_info->full = 1;
2859 out:
2860         mutex_unlock(&extent_root->fs_info->chunk_mutex);
2861         return ret;
2862 }
2863
2864 static int update_block_group(struct btrfs_trans_handle *trans,
2865                               struct btrfs_root *root,
2866                               u64 bytenr, u64 num_bytes, int alloc,
2867                               int mark_free)
2868 {
2869         struct btrfs_block_group_cache *cache;
2870         struct btrfs_fs_info *info = root->fs_info;
2871         u64 total = num_bytes;
2872         u64 old_val;
2873         u64 byte_in_group;
2874
2875         /* block accounting for super block */
2876         spin_lock(&info->delalloc_lock);
2877         old_val = btrfs_super_bytes_used(&info->super_copy);
2878         if (alloc)
2879                 old_val += num_bytes;
2880         else
2881                 old_val -= num_bytes;
2882         btrfs_set_super_bytes_used(&info->super_copy, old_val);
2883
2884         /* block accounting for root item */
2885         old_val = btrfs_root_used(&root->root_item);
2886         if (alloc)
2887                 old_val += num_bytes;
2888         else
2889                 old_val -= num_bytes;
2890         btrfs_set_root_used(&root->root_item, old_val);
2891         spin_unlock(&info->delalloc_lock);
2892
2893         while (total) {
2894                 cache = btrfs_lookup_block_group(info, bytenr);
2895                 if (!cache)
2896                         return -1;
2897                 byte_in_group = bytenr - cache->key.objectid;
2898                 WARN_ON(byte_in_group > cache->key.offset);
2899
2900                 spin_lock(&cache->space_info->lock);
2901                 spin_lock(&cache->lock);
2902                 cache->dirty = 1;
2903                 old_val = btrfs_block_group_used(&cache->item);
2904                 num_bytes = min(total, cache->key.offset - byte_in_group);
2905                 if (alloc) {
2906                         old_val += num_bytes;
2907                         cache->space_info->bytes_used += num_bytes;
2908                         if (cache->ro)
2909                                 cache->space_info->bytes_readonly -= num_bytes;
2910                         btrfs_set_block_group_used(&cache->item, old_val);
2911                         spin_unlock(&cache->lock);
2912                         spin_unlock(&cache->space_info->lock);
2913                 } else {
2914                         old_val -= num_bytes;
2915                         cache->space_info->bytes_used -= num_bytes;
2916                         if (cache->ro)
2917                                 cache->space_info->bytes_readonly += num_bytes;
2918                         btrfs_set_block_group_used(&cache->item, old_val);
2919                         spin_unlock(&cache->lock);
2920                         spin_unlock(&cache->space_info->lock);
2921                         if (mark_free) {
2922                                 int ret;
2923
2924                                 ret = btrfs_discard_extent(root, bytenr,
2925                                                            num_bytes);
2926                                 WARN_ON(ret);
2927
2928                                 ret = btrfs_add_free_space(cache, bytenr,
2929                                                            num_bytes);
2930                                 WARN_ON(ret);
2931                         }
2932                 }
2933                 btrfs_put_block_group(cache);
2934                 total -= num_bytes;
2935                 bytenr += num_bytes;
2936         }
2937         return 0;
2938 }
2939
2940 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2941 {
2942         struct btrfs_block_group_cache *cache;
2943         u64 bytenr;
2944
2945         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2946         if (!cache)
2947                 return 0;
2948
2949         bytenr = cache->key.objectid;
2950         btrfs_put_block_group(cache);
2951
2952         return bytenr;
2953 }
2954
2955 int btrfs_update_pinned_extents(struct btrfs_root *root,
2956                                 u64 bytenr, u64 num, int pin)
2957 {
2958         u64 len;
2959         struct btrfs_block_group_cache *cache;
2960         struct btrfs_fs_info *fs_info = root->fs_info;
2961
2962         if (pin) {
2963                 set_extent_dirty(&fs_info->pinned_extents,
2964                                 bytenr, bytenr + num - 1, GFP_NOFS);
2965         } else {
2966                 clear_extent_dirty(&fs_info->pinned_extents,
2967                                 bytenr, bytenr + num - 1, GFP_NOFS);
2968         }
2969
2970         while (num > 0) {
2971                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2972                 BUG_ON(!cache);
2973                 len = min(num, cache->key.offset -
2974                           (bytenr - cache->key.objectid));
2975                 if (pin) {
2976                         spin_lock(&cache->space_info->lock);
2977                         spin_lock(&cache->lock);
2978                         cache->pinned += len;
2979                         cache->space_info->bytes_pinned += len;
2980                         spin_unlock(&cache->lock);
2981                         spin_unlock(&cache->space_info->lock);
2982                         fs_info->total_pinned += len;
2983                 } else {
2984                         spin_lock(&cache->space_info->lock);
2985                         spin_lock(&cache->lock);
2986                         cache->pinned -= len;
2987                         cache->space_info->bytes_pinned -= len;
2988                         spin_unlock(&cache->lock);
2989                         spin_unlock(&cache->space_info->lock);
2990                         fs_info->total_pinned -= len;
2991                         if (cache->cached)
2992                                 btrfs_add_free_space(cache, bytenr, len);
2993                 }
2994                 btrfs_put_block_group(cache);
2995                 bytenr += len;
2996                 num -= len;
2997         }
2998         return 0;
2999 }
3000
3001 static int update_reserved_extents(struct btrfs_root *root,
3002                                    u64 bytenr, u64 num, int reserve)
3003 {
3004         u64 len;
3005         struct btrfs_block_group_cache *cache;
3006         struct btrfs_fs_info *fs_info = root->fs_info;
3007
3008         while (num > 0) {
3009                 cache = btrfs_lookup_block_group(fs_info, bytenr);
3010                 BUG_ON(!cache);
3011                 len = min(num, cache->key.offset -
3012                           (bytenr - cache->key.objectid));
3013
3014                 spin_lock(&cache->space_info->lock);
3015                 spin_lock(&cache->lock);
3016                 if (reserve) {
3017                         cache->reserved += len;
3018                         cache->space_info->bytes_reserved += len;
3019                 } else {
3020                         cache->reserved -= len;
3021                         cache->space_info->bytes_reserved -= len;
3022                 }
3023                 spin_unlock(&cache->lock);
3024                 spin_unlock(&cache->space_info->lock);
3025                 btrfs_put_block_group(cache);
3026                 bytenr += len;
3027                 num -= len;
3028         }
3029         return 0;
3030 }
3031
3032 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
3033 {
3034         u64 last = 0;
3035         u64 start;
3036         u64 end;
3037         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3038         int ret;
3039
3040         while (1) {
3041                 ret = find_first_extent_bit(pinned_extents, last,
3042                                             &start, &end, EXTENT_DIRTY);
3043                 if (ret)
3044                         break;
3045                 set_extent_dirty(copy, start, end, GFP_NOFS);
3046                 last = end + 1;
3047         }
3048         return 0;
3049 }
3050
3051 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3052                                struct btrfs_root *root,
3053                                struct extent_io_tree *unpin)
3054 {
3055         u64 start;
3056         u64 end;
3057         int ret;
3058
3059         while (1) {
3060                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3061                                             EXTENT_DIRTY);
3062                 if (ret)
3063                         break;
3064
3065                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3066
3067                 /* unlocks the pinned mutex */
3068                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
3069                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3070
3071                 cond_resched();
3072         }
3073         return ret;
3074 }
3075
3076 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3077                           struct btrfs_root *root,
3078                           struct btrfs_path *path,
3079                           u64 bytenr, u64 num_bytes, int is_data,
3080                           struct extent_buffer **must_clean)
3081 {
3082         int err = 0;
3083         struct extent_buffer *buf;
3084
3085         if (is_data)
3086                 goto pinit;
3087
3088         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3089         if (!buf)
3090                 goto pinit;
3091
3092         /* we can reuse a block if it hasn't been written
3093          * and it is from this transaction.  We can't
3094          * reuse anything from the tree log root because
3095          * it has tiny sub-transactions.
3096          */
3097         if (btrfs_buffer_uptodate(buf, 0) &&
3098             btrfs_try_tree_lock(buf)) {
3099                 u64 header_owner = btrfs_header_owner(buf);
3100                 u64 header_transid = btrfs_header_generation(buf);
3101                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3102                     header_transid == trans->transid &&
3103                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3104                         *must_clean = buf;
3105                         return 1;
3106                 }
3107                 btrfs_tree_unlock(buf);
3108         }
3109         free_extent_buffer(buf);
3110 pinit:
3111         btrfs_set_path_blocking(path);
3112         /* unlocks the pinned mutex */
3113         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3114
3115         BUG_ON(err < 0);
3116         return 0;
3117 }
3118
3119
3120 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3121                                 struct btrfs_root *root,
3122                                 u64 bytenr, u64 num_bytes, u64 parent,
3123                                 u64 root_objectid, u64 owner_objectid,
3124                                 u64 owner_offset, int refs_to_drop,
3125                                 struct btrfs_delayed_extent_op *extent_op)
3126 {
3127         struct btrfs_key key;
3128         struct btrfs_path *path;
3129         struct btrfs_fs_info *info = root->fs_info;
3130         struct btrfs_root *extent_root = info->extent_root;
3131         struct extent_buffer *leaf;
3132         struct btrfs_extent_item *ei;
3133         struct btrfs_extent_inline_ref *iref;
3134         int ret;
3135         int is_data;
3136         int extent_slot = 0;
3137         int found_extent = 0;
3138         int num_to_del = 1;
3139         u32 item_size;
3140         u64 refs;
3141
3142         path = btrfs_alloc_path();
3143         if (!path)
3144                 return -ENOMEM;
3145
3146         path->reada = 1;
3147         path->leave_spinning = 1;
3148
3149         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3150         BUG_ON(!is_data && refs_to_drop != 1);
3151
3152         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3153                                     bytenr, num_bytes, parent,
3154                                     root_objectid, owner_objectid,
3155                                     owner_offset);
3156         if (ret == 0) {
3157                 extent_slot = path->slots[0];
3158                 while (extent_slot >= 0) {
3159                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3160                                               extent_slot);
3161                         if (key.objectid != bytenr)
3162                                 break;
3163                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3164                             key.offset == num_bytes) {
3165                                 found_extent = 1;
3166                                 break;
3167                         }
3168                         if (path->slots[0] - extent_slot > 5)
3169                                 break;
3170                         extent_slot--;
3171                 }
3172 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3173                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3174                 if (found_extent && item_size < sizeof(*ei))
3175                         found_extent = 0;
3176 #endif
3177                 if (!found_extent) {
3178                         BUG_ON(iref);
3179                         ret = remove_extent_backref(trans, extent_root, path,
3180                                                     NULL, refs_to_drop,
3181                                                     is_data);
3182                         BUG_ON(ret);
3183                         btrfs_release_path(extent_root, path);
3184                         path->leave_spinning = 1;
3185
3186                         key.objectid = bytenr;
3187                         key.type = BTRFS_EXTENT_ITEM_KEY;
3188                         key.offset = num_bytes;
3189
3190                         ret = btrfs_search_slot(trans, extent_root,
3191                                                 &key, path, -1, 1);
3192                         if (ret) {
3193                                 printk(KERN_ERR "umm, got %d back from search"
3194                                        ", was looking for %llu\n", ret,
3195                                        (unsigned long long)bytenr);
3196                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3197                         }
3198                         BUG_ON(ret);
3199                         extent_slot = path->slots[0];
3200                 }
3201         } else {
3202                 btrfs_print_leaf(extent_root, path->nodes[0]);
3203                 WARN_ON(1);
3204                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3205                        "parent %llu root %llu  owner %llu offset %llu\n",
3206                        (unsigned long long)bytenr,
3207                        (unsigned long long)parent,
3208                        (unsigned long long)root_objectid,
3209                        (unsigned long long)owner_objectid,
3210                        (unsigned long long)owner_offset);
3211         }
3212
3213         leaf = path->nodes[0];
3214         item_size = btrfs_item_size_nr(leaf, extent_slot);
3215 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3216         if (item_size < sizeof(*ei)) {
3217                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3218                 ret = convert_extent_item_v0(trans, extent_root, path,
3219                                              owner_objectid, 0);
3220                 BUG_ON(ret < 0);
3221
3222                 btrfs_release_path(extent_root, path);
3223                 path->leave_spinning = 1;
3224
3225                 key.objectid = bytenr;
3226                 key.type = BTRFS_EXTENT_ITEM_KEY;
3227                 key.offset = num_bytes;
3228
3229                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3230                                         -1, 1);
3231                 if (ret) {
3232                         printk(KERN_ERR "umm, got %d back from search"
3233                                ", was looking for %llu\n", ret,
3234                                (unsigned long long)bytenr);
3235                         btrfs_print_leaf(extent_root, path->nodes[0]);
3236                 }
3237                 BUG_ON(ret);
3238                 extent_slot = path->slots[0];
3239                 leaf = path->nodes[0];
3240                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3241         }
3242 #endif
3243         BUG_ON(item_size < sizeof(*ei));
3244         ei = btrfs_item_ptr(leaf, extent_slot,
3245                             struct btrfs_extent_item);
3246         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3247                 struct btrfs_tree_block_info *bi;
3248                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3249                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3250                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3251         }
3252
3253         refs = btrfs_extent_refs(leaf, ei);
3254         BUG_ON(refs < refs_to_drop);
3255         refs -= refs_to_drop;
3256
3257         if (refs > 0) {
3258                 if (extent_op)
3259                         __run_delayed_extent_op(extent_op, leaf, ei);
3260                 /*
3261                  * In the case of inline back ref, reference count will
3262                  * be updated by remove_extent_backref
3263                  */
3264                 if (iref) {
3265                         BUG_ON(!found_extent);
3266                 } else {
3267                         btrfs_set_extent_refs(leaf, ei, refs);
3268                         btrfs_mark_buffer_dirty(leaf);
3269                 }
3270                 if (found_extent) {
3271                         ret = remove_extent_backref(trans, extent_root, path,
3272                                                     iref, refs_to_drop,
3273                                                     is_data);
3274                         BUG_ON(ret);
3275                 }
3276         } else {
3277                 int mark_free = 0;
3278                 struct extent_buffer *must_clean = NULL;
3279
3280                 if (found_extent) {
3281                         BUG_ON(is_data && refs_to_drop !=
3282                                extent_data_ref_count(root, path, iref));
3283                         if (iref) {
3284                                 BUG_ON(path->slots[0] != extent_slot);
3285                         } else {
3286                                 BUG_ON(path->slots[0] != extent_slot + 1);
3287                                 path->slots[0] = extent_slot;
3288                                 num_to_del = 2;
3289                         }
3290                 }
3291
3292                 ret = pin_down_bytes(trans, root, path, bytenr,
3293                                      num_bytes, is_data, &must_clean);
3294                 if (ret > 0)
3295                         mark_free = 1;
3296                 BUG_ON(ret < 0);
3297                 /*
3298                  * it is going to be very rare for someone to be waiting
3299                  * on the block we're freeing.  del_items might need to
3300                  * schedule, so rather than get fancy, just force it
3301                  * to blocking here
3302                  */
3303                 if (must_clean)
3304                         btrfs_set_lock_blocking(must_clean);
3305
3306                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3307                                       num_to_del);
3308                 BUG_ON(ret);
3309                 btrfs_release_path(extent_root, path);
3310
3311                 if (must_clean) {
3312                         clean_tree_block(NULL, root, must_clean);
3313                         btrfs_tree_unlock(must_clean);
3314                         free_extent_buffer(must_clean);
3315                 }
3316
3317                 if (is_data) {
3318                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3319                         BUG_ON(ret);
3320                 } else {
3321                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3322                              bytenr >> PAGE_CACHE_SHIFT,
3323                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3324                 }
3325
3326                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3327                                          mark_free);
3328                 BUG_ON(ret);
3329         }
3330         btrfs_free_path(path);
3331         return ret;
3332 }
3333
3334 /*
3335  * when we free an extent, it is possible (and likely) that we free the last
3336  * delayed ref for that extent as well.  This searches the delayed ref tree for
3337  * a given extent, and if there are no other delayed refs to be processed, it
3338  * removes it from the tree.
3339  */
3340 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3341                                       struct btrfs_root *root, u64 bytenr)
3342 {
3343         struct btrfs_delayed_ref_head *head;
3344         struct btrfs_delayed_ref_root *delayed_refs;
3345         struct btrfs_delayed_ref_node *ref;
3346         struct rb_node *node;
3347         int ret;
3348
3349         delayed_refs = &trans->transaction->delayed_refs;
3350         spin_lock(&delayed_refs->lock);
3351         head = btrfs_find_delayed_ref_head(trans, bytenr);
3352         if (!head)
3353                 goto out;
3354
3355         node = rb_prev(&head->node.rb_node);
3356         if (!node)
3357                 goto out;
3358
3359         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3360
3361         /* there are still entries for this ref, we can't drop it */
3362         if (ref->bytenr == bytenr)
3363                 goto out;
3364
3365         if (head->extent_op) {
3366                 if (!head->must_insert_reserved)
3367                         goto out;
3368                 kfree(head->extent_op);
3369                 head->extent_op = NULL;
3370         }
3371
3372         /*
3373          * waiting for the lock here would deadlock.  If someone else has it
3374          * locked they are already in the process of dropping it anyway
3375          */
3376         if (!mutex_trylock(&head->mutex))
3377                 goto out;
3378
3379         /*
3380          * at this point we have a head with no other entries.  Go
3381          * ahead and process it.
3382          */
3383         head->node.in_tree = 0;
3384         rb_erase(&head->node.rb_node, &delayed_refs->root);
3385
3386         delayed_refs->num_entries--;
3387
3388         /*
3389          * we don't take a ref on the node because we're removing it from the
3390          * tree, so we just steal the ref the tree was holding.
3391          */
3392         delayed_refs->num_heads--;
3393         if (list_empty(&head->cluster))
3394                 delayed_refs->num_heads_ready--;
3395
3396         list_del_init(&head->cluster);
3397         spin_unlock(&delayed_refs->lock);
3398
3399         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3400                                   &head->node, head->extent_op,
3401                                   head->must_insert_reserved);
3402         BUG_ON(ret);
3403         btrfs_put_delayed_ref(&head->node);
3404         return 0;
3405 out:
3406         spin_unlock(&delayed_refs->lock);
3407         return 0;
3408 }
3409
3410 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3411                       struct btrfs_root *root,
3412                       u64 bytenr, u64 num_bytes, u64 parent,
3413                       u64 root_objectid, u64 owner, u64 offset)
3414 {
3415         int ret;
3416
3417         /*
3418          * tree log blocks never actually go into the extent allocation
3419          * tree, just update pinning info and exit early.
3420          */
3421         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3422                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3423                 /* unlocks the pinned mutex */
3424                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3425                 update_reserved_extents(root, bytenr, num_bytes, 0);
3426                 ret = 0;
3427         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3428                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3429                                         parent, root_objectid, (int)owner,
3430                                         BTRFS_DROP_DELAYED_REF, NULL);
3431                 BUG_ON(ret);
3432                 ret = check_ref_cleanup(trans, root, bytenr);
3433                 BUG_ON(ret);
3434         } else {
3435                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3436                                         parent, root_objectid, owner,
3437                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
3438                 BUG_ON(ret);
3439         }
3440         return ret;
3441 }
3442
3443 static u64 stripe_align(struct btrfs_root *root, u64 val)
3444 {
3445         u64 mask = ((u64)root->stripesize - 1);
3446         u64 ret = (val + mask) & ~mask;
3447         return ret;
3448 }
3449
3450 /*
3451  * walks the btree of allocated extents and find a hole of a given size.
3452  * The key ins is changed to record the hole:
3453  * ins->objectid == block start
3454  * ins->flags = BTRFS_EXTENT_ITEM_KEY
3455  * ins->offset == number of blocks
3456  * Any available blocks before search_start are skipped.
3457  */
3458 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3459                                      struct btrfs_root *orig_root,
3460                                      u64 num_bytes, u64 empty_size,
3461                                      u64 search_start, u64 search_end,
3462                                      u64 hint_byte, struct btrfs_key *ins,
3463                                      u64 exclude_start, u64 exclude_nr,
3464                                      int data)
3465 {
3466         int ret = 0;
3467         struct btrfs_root *root = orig_root->fs_info->extent_root;
3468         struct btrfs_free_cluster *last_ptr = NULL;
3469         struct btrfs_block_group_cache *block_group = NULL;
3470         int empty_cluster = 2 * 1024 * 1024;
3471         int allowed_chunk_alloc = 0;
3472         struct btrfs_space_info *space_info;
3473         int last_ptr_loop = 0;
3474         int loop = 0;
3475
3476         WARN_ON(num_bytes < root->sectorsize);
3477         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3478         ins->objectid = 0;
3479         ins->offset = 0;
3480
3481         space_info = __find_space_info(root->fs_info, data);
3482
3483         if (orig_root->ref_cows || empty_size)
3484                 allowed_chunk_alloc = 1;
3485
3486         if (data & BTRFS_BLOCK_GROUP_METADATA) {
3487                 last_ptr = &root->fs_info->meta_alloc_cluster;
3488                 if (!btrfs_test_opt(root, SSD))
3489                         empty_cluster = 64 * 1024;
3490         }
3491
3492         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3493                 last_ptr = &root->fs_info->data_alloc_cluster;
3494         }
3495
3496         if (last_ptr) {
3497                 spin_lock(&last_ptr->lock);
3498                 if (last_ptr->block_group)
3499                         hint_byte = last_ptr->window_start;
3500                 spin_unlock(&last_ptr->lock);
3501         }
3502
3503         search_start = max(search_start, first_logical_byte(root, 0));
3504         search_start = max(search_start, hint_byte);
3505
3506         if (!last_ptr) {
3507                 empty_cluster = 0;
3508                 loop = 1;
3509         }
3510
3511         if (search_start == hint_byte) {
3512                 block_group = btrfs_lookup_block_group(root->fs_info,
3513                                                        search_start);
3514                 if (block_group && block_group_bits(block_group, data)) {
3515                         down_read(&space_info->groups_sem);
3516                         if (list_empty(&block_group->list) ||
3517                             block_group->ro) {
3518                                 /*
3519                                  * someone is removing this block group,
3520                                  * we can't jump into the have_block_group
3521                                  * target because our list pointers are not
3522                                  * valid
3523                                  */
3524                                 btrfs_put_block_group(block_group);
3525                                 up_read(&space_info->groups_sem);
3526                         } else
3527                                 goto have_block_group;
3528                 } else if (block_group) {
3529                         btrfs_put_block_group(block_group);
3530                 }
3531         }
3532
3533 search:
3534         down_read(&space_info->groups_sem);
3535         list_for_each_entry(block_group, &space_info->block_groups, list) {
3536                 u64 offset;
3537
3538                 atomic_inc(&block_group->count);
3539                 search_start = block_group->key.objectid;
3540
3541 have_block_group:
3542                 if (unlikely(!block_group->cached)) {
3543                         mutex_lock(&block_group->cache_mutex);
3544                         ret = cache_block_group(root, block_group);
3545                         mutex_unlock(&block_group->cache_mutex);
3546                         if (ret) {
3547                                 btrfs_put_block_group(block_group);
3548                                 break;
3549                         }
3550                 }
3551
3552                 if (unlikely(block_group->ro))
3553                         goto loop;
3554
3555                 if (last_ptr) {
3556                         /*
3557                          * the refill lock keeps out other
3558                          * people trying to start a new cluster
3559                          */
3560                         spin_lock(&last_ptr->refill_lock);
3561                         if (last_ptr->block_group &&
3562                             (last_ptr->block_group->ro ||
3563                             !block_group_bits(last_ptr->block_group, data))) {
3564                                 offset = 0;
3565                                 goto refill_cluster;
3566                         }
3567
3568                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3569                                                  num_bytes, search_start);
3570                         if (offset) {
3571                                 /* we have a block, we're done */
3572                                 spin_unlock(&last_ptr->refill_lock);
3573                                 goto checks;
3574                         }
3575
3576                         spin_lock(&last_ptr->lock);
3577                         /*
3578                          * whoops, this cluster doesn't actually point to
3579                          * this block group.  Get a ref on the block
3580                          * group is does point to and try again
3581                          */
3582                         if (!last_ptr_loop && last_ptr->block_group &&
3583                             last_ptr->block_group != block_group) {
3584
3585                                 btrfs_put_block_group(block_group);
3586                                 block_group = last_ptr->block_group;
3587                                 atomic_inc(&block_group->count);
3588                                 spin_unlock(&last_ptr->lock);
3589                                 spin_unlock(&last_ptr->refill_lock);
3590
3591                                 last_ptr_loop = 1;
3592                                 search_start = block_group->key.objectid;
3593                                 /*
3594                                  * we know this block group is properly
3595                                  * in the list because
3596                                  * btrfs_remove_block_group, drops the
3597                                  * cluster before it removes the block
3598                                  * group from the list
3599                                  */
3600                                 goto have_block_group;
3601                         }
3602                         spin_unlock(&last_ptr->lock);
3603 refill_cluster:
3604                         /*
3605                          * this cluster didn't work out, free it and
3606                          * start over
3607                          */
3608                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3609
3610                         last_ptr_loop = 0;
3611
3612                         /* allocate a cluster in this block group */
3613                         ret = btrfs_find_space_cluster(trans, root,
3614                                                block_group, last_ptr,
3615                                                offset, num_bytes,
3616                                                empty_cluster + empty_size);
3617                         if (ret == 0) {
3618                                 /*
3619                                  * now pull our allocation out of this
3620                                  * cluster
3621                                  */
3622                                 offset = btrfs_alloc_from_cluster(block_group,
3623                                                   last_ptr, num_bytes,
3624                                                   search_start);
3625                                 if (offset) {
3626                                         /* we found one, proceed */
3627                                         spin_unlock(&last_ptr->refill_lock);
3628                                         goto checks;
3629                                 }
3630                         }
3631                         /*
3632                          * at this point we either didn't find a cluster
3633                          * or we weren't able to allocate a block from our
3634                          * cluster.  Free the cluster we've been trying
3635                          * to use, and go to the next block group
3636                          */
3637                         if (loop < 2) {
3638                                 btrfs_return_cluster_to_free_space(NULL,
3639                                                                    last_ptr);
3640                                 spin_unlock(&last_ptr->refill_lock);
3641                                 goto loop;
3642                         }
3643                         spin_unlock(&last_ptr->refill_lock);
3644                 }
3645
3646                 offset = btrfs_find_space_for_alloc(block_group, search_start,
3647                                                     num_bytes, empty_size);
3648                 if (!offset)
3649                         goto loop;
3650 checks:
3651                 search_start = stripe_align(root, offset);
3652
3653                 /* move on to the next group */
3654                 if (search_start + num_bytes >= search_end) {
3655                         btrfs_add_free_space(block_group, offset, num_bytes);
3656                         goto loop;
3657                 }
3658
3659                 /* move on to the next group */
3660                 if (search_start + num_bytes >
3661                     block_group->key.objectid + block_group->key.offset) {
3662                         btrfs_add_free_space(block_group, offset, num_bytes);
3663                         goto loop;
3664                 }
3665
3666                 if (exclude_nr > 0 &&
3667                     (search_start + num_bytes > exclude_start &&
3668                      search_start < exclude_start + exclude_nr)) {
3669                         search_start = exclude_start + exclude_nr;
3670
3671                         btrfs_add_free_space(block_group, offset, num_bytes);
3672                         /*
3673                          * if search_start is still in this block group
3674                          * then we just re-search this block group
3675                          */
3676                         if (search_start >= block_group->key.objectid &&
3677                             search_start < (block_group->key.objectid +
3678                                             block_group->key.offset))
3679                                 goto have_block_group;
3680                         goto loop;
3681                 }
3682
3683                 ins->objectid = search_start;
3684                 ins->offset = num_bytes;
3685
3686                 if (offset < search_start)
3687                         btrfs_add_free_space(block_group, offset,
3688                                              search_start - offset);
3689                 BUG_ON(offset > search_start);
3690
3691                 /* we are all good, lets return */
3692                 break;
3693 loop:
3694                 btrfs_put_block_group(block_group);
3695         }
3696         up_read(&space_info->groups_sem);
3697
3698         /* loop == 0, try to find a clustered alloc in every block group
3699          * loop == 1, try again after forcing a chunk allocation
3700          * loop == 2, set empty_size and empty_cluster to 0 and try again
3701          */
3702         if (!ins->objectid && loop < 3 &&
3703             (empty_size || empty_cluster || allowed_chunk_alloc)) {
3704                 if (loop >= 2) {
3705                         empty_size = 0;
3706                         empty_cluster = 0;
3707                 }
3708
3709                 if (allowed_chunk_alloc) {
3710                         ret = do_chunk_alloc(trans, root, num_bytes +
3711                                              2 * 1024 * 1024, data, 1);
3712                         allowed_chunk_alloc = 0;
3713                 } else {
3714                         space_info->force_alloc = 1;
3715                 }
3716
3717                 if (loop < 3) {
3718                         loop++;
3719                         goto search;
3720                 }
3721                 ret = -ENOSPC;
3722         } else if (!ins->objectid) {
3723                 ret = -ENOSPC;
3724         }
3725
3726         /* we found what we needed */
3727         if (ins->objectid) {
3728                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3729                         trans->block_group = block_group->key.objectid;
3730
3731                 btrfs_put_block_group(block_group);
3732                 ret = 0;
3733         }
3734
3735         return ret;
3736 }
3737
3738 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3739 {
3740         struct btrfs_block_group_cache *cache;
3741
3742         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3743                (unsigned long long)(info->total_bytes - info->bytes_used -
3744                                     info->bytes_pinned - info->bytes_reserved),
3745                (info->full) ? "" : "not ");
3746         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3747                " may_use=%llu, used=%llu\n",
3748                (unsigned long long)info->total_bytes,
3749                (unsigned long long)info->bytes_pinned,
3750                (unsigned long long)info->bytes_delalloc,
3751                (unsigned long long)info->bytes_may_use,
3752                (unsigned long long)info->bytes_used);
3753
3754         down_read(&info->groups_sem);
3755         list_for_each_entry(cache, &info->block_groups, list) {
3756                 spin_lock(&cache->lock);
3757                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3758                        "%llu pinned %llu reserved\n",
3759                        (unsigned long long)cache->key.objectid,
3760                        (unsigned long long)cache->key.offset,
3761                        (unsigned long long)btrfs_block_group_used(&cache->item),
3762                        (unsigned long long)cache->pinned,
3763                        (unsigned long long)cache->reserved);
3764                 btrfs_dump_free_space(cache, bytes);
3765                 spin_unlock(&cache->lock);
3766         }
3767         up_read(&info->groups_sem);
3768 }
3769
3770 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3771                                   struct btrfs_root *root,
3772                                   u64 num_bytes, u64 min_alloc_size,
3773                                   u64 empty_size, u64 hint_byte,
3774                                   u64 search_end, struct btrfs_key *ins,
3775                                   u64 data)
3776 {
3777         int ret;
3778         u64 search_start = 0;
3779         struct btrfs_fs_info *info = root->fs_info;
3780
3781         data = btrfs_get_alloc_profile(root, data);
3782 again:
3783         /*
3784          * the only place that sets empty_size is btrfs_realloc_node, which
3785          * is not called recursively on allocations
3786          */
3787         if (empty_size || root->ref_cows) {
3788                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3789                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3790                                      2 * 1024 * 1024,
3791                                      BTRFS_BLOCK_GROUP_METADATA |
3792                                      (info->metadata_alloc_profile &
3793                                       info->avail_metadata_alloc_bits), 0);
3794                 }
3795                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3796                                      num_bytes + 2 * 1024 * 1024, data, 0);
3797         }
3798
3799         WARN_ON(num_bytes < root->sectorsize);
3800         ret = find_free_extent(trans, root, num_bytes, empty_size,
3801                                search_start, search_end, hint_byte, ins,
3802                                trans->alloc_exclude_start,
3803                                trans->alloc_exclude_nr, data);
3804
3805         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3806                 num_bytes = num_bytes >> 1;
3807                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3808                 num_bytes = max(num_bytes, min_alloc_size);
3809                 do_chunk_alloc(trans, root->fs_info->extent_root,
3810                                num_bytes, data, 1);
3811                 goto again;
3812         }
3813         if (ret) {
3814                 struct btrfs_space_info *sinfo;
3815
3816                 sinfo = __find_space_info(root->fs_info, data);
3817                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3818                        "wanted %llu\n", (unsigned long long)data,
3819                        (unsigned long long)num_bytes);
3820                 dump_space_info(sinfo, num_bytes);
3821                 BUG();
3822         }
3823
3824         return ret;
3825 }
3826
3827 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3828 {
3829         struct btrfs_block_group_cache *cache;
3830         int ret = 0;
3831
3832         cache = btrfs_lookup_block_group(root->fs_info, start);
3833         if (!cache) {
3834                 printk(KERN_ERR "Unable to find block group for %llu\n",
3835                        (unsigned long long)start);
3836                 return -ENOSPC;
3837         }
3838
3839         ret = btrfs_discard_extent(root, start, len);
3840
3841         btrfs_add_free_space(cache, start, len);
3842         btrfs_put_block_group(cache);
3843         update_reserved_extents(root, start, len, 0);
3844
3845         return ret;
3846 }
3847
3848 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3849                                   struct btrfs_root *root,
3850                                   u64 num_bytes, u64 min_alloc_size,
3851                                   u64 empty_size, u64 hint_byte,
3852                                   u64 search_end, struct btrfs_key *ins,
3853                                   u64 data)
3854 {
3855         int ret;
3856         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3857                                      empty_size, hint_byte, search_end, ins,
3858                                      data);
3859         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3860         return ret;
3861 }
3862
3863 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
3864                                       struct btrfs_root *root,
3865                                       u64 parent, u64 root_objectid,
3866                                       u64 flags, u64 owner, u64 offset,
3867                                       struct btrfs_key *ins, int ref_mod)
3868 {
3869         int ret;
3870         struct btrfs_fs_info *fs_info = root->fs_info;
3871         struct btrfs_extent_item *extent_item;
3872         struct btrfs_extent_inline_ref *iref;
3873         struct btrfs_path *path;
3874         struct extent_buffer *leaf;
3875         int type;
3876         u32 size;
3877
3878         if (parent > 0)
3879                 type = BTRFS_SHARED_DATA_REF_KEY;
3880         else
3881                 type = BTRFS_EXTENT_DATA_REF_KEY;
3882
3883         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
3884
3885         path = btrfs_alloc_path();
3886         BUG_ON(!path);
3887
3888         path->leave_spinning = 1;
3889         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
3890                                       ins, size);
3891         BUG_ON(ret);
3892
3893         leaf = path->nodes[0];
3894         extent_item = btrfs_item_ptr(leaf, path->slots[0],
3895                                      struct btrfs_extent_item);
3896         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
3897         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
3898         btrfs_set_extent_flags(leaf, extent_item,
3899                                flags | BTRFS_EXTENT_FLAG_DATA);
3900
3901         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
3902         btrfs_set_extent_inline_ref_type(leaf, iref, type);
3903         if (parent > 0) {
3904                 struct btrfs_shared_data_ref *ref;
3905                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
3906                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
3907                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
3908         } else {
3909                 struct btrfs_extent_data_ref *ref;
3910                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3911                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
3912                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
3913                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
3914                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
3915         }
3916
3917         btrfs_mark_buffer_dirty(path->nodes[0]);
3918         btrfs_free_path(path);
3919
3920         ret = update_block_group(trans, root, ins->objectid, ins->offset,
3921                                  1, 0);
3922         if (ret) {
3923                 printk(KERN_ERR "btrfs update block group failed for %llu "
3924                        "%llu\n", (unsigned long long)ins->objectid,
3925                        (unsigned long long)ins->offset);
3926                 BUG();
3927         }
3928         return ret;
3929 }
3930
3931 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
3932                                      struct btrfs_root *root,
3933                                      u64 parent, u64 root_objectid,
3934                                      u64 flags, struct btrfs_disk_key *key,
3935                                      int level, struct btrfs_key *ins)
3936 {
3937         int ret;
3938         struct btrfs_fs_info *fs_info = root->fs_info;
3939         struct btrfs_extent_item *extent_item;
3940         struct btrfs_tree_block_info *block_info;
3941         struct btrfs_extent_inline_ref *iref;
3942         struct btrfs_path *path;
3943         struct extent_buffer *leaf;
3944         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
3945
3946         path = btrfs_alloc_path();
3947         BUG_ON(!path);
3948
3949         path->leave_spinning = 1;
3950         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
3951                                       ins, size);
3952         BUG_ON(ret);
3953
3954         leaf = path->nodes[0];
3955         extent_item = btrfs_item_ptr(leaf, path->slots[0],
3956                                      struct btrfs_extent_item);
3957         btrfs_set_extent_refs(leaf, extent_item, 1);
3958         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
3959         btrfs_set_extent_flags(leaf, extent_item,
3960                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
3961         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
3962
3963         btrfs_set_tree_block_key(leaf, block_info, key);
3964         btrfs_set_tree_block_level(leaf, block_info, level);
3965
3966         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
3967         if (parent > 0) {
3968                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
3969                 btrfs_set_extent_inline_ref_type(leaf, iref,
3970                                                  BTRFS_SHARED_BLOCK_REF_KEY);
3971                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
3972         } else {
3973                 btrfs_set_extent_inline_ref_type(leaf, iref,
3974                                                  BTRFS_TREE_BLOCK_REF_KEY);
3975                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
3976         }
3977
3978         btrfs_mark_buffer_dirty(leaf);
3979         btrfs_free_path(path);
3980
3981         ret = update_block_group(trans, root, ins->objectid, ins->offset,
3982                                  1, 0);
3983         if (ret) {
3984                 printk(KERN_ERR "btrfs update block group failed for %llu "
3985                        "%llu\n", (unsigned long long)ins->objectid,
3986                        (unsigned long long)ins->offset);
3987                 BUG();
3988         }
3989         return ret;
3990 }
3991
3992 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
3993                                      struct btrfs_root *root,
3994                                      u64 root_objectid, u64 owner,
3995                                      u64 offset, struct btrfs_key *ins)
3996 {
3997         int ret;
3998
3999         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4000
4001         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4002                                          0, root_objectid, owner, offset,
4003                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4004         return ret;
4005 }
4006
4007 /*
4008  * this is used by the tree logging recovery code.  It records that
4009  * an extent has been allocated and makes sure to clear the free
4010  * space cache bits as well
4011  */
4012 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4013                                    struct btrfs_root *root,
4014                                    u64 root_objectid, u64 owner, u64 offset,
4015                                    struct btrfs_key *ins)
4016 {
4017         int ret;
4018         struct btrfs_block_group_cache *block_group;
4019
4020         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4021         mutex_lock(&block_group->cache_mutex);
4022         cache_block_group(root, block_group);
4023         mutex_unlock(&block_group->cache_mutex);
4024
4025         ret = btrfs_remove_free_space(block_group, ins->objectid,
4026                                       ins->offset);
4027         BUG_ON(ret);
4028         btrfs_put_block_group(block_group);
4029         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4030                                          0, owner, offset, ins, 1);
4031         return ret;
4032 }
4033
4034 /*
4035  * finds a free extent and does all the dirty work required for allocation
4036  * returns the key for the extent through ins, and a tree buffer for
4037  * the first block of the extent through buf.
4038  *
4039  * returns 0 if everything worked, non-zero otherwise.
4040  */
4041 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4042                             struct btrfs_root *root,
4043                             u64 num_bytes, u64 parent, u64 root_objectid,
4044                             struct btrfs_disk_key *key, int level,
4045                             u64 empty_size, u64 hint_byte, u64 search_end,
4046                             struct btrfs_key *ins)
4047 {
4048         int ret;
4049         u64 flags = 0;
4050
4051         ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4052                                      empty_size, hint_byte, search_end,
4053                                      ins, 0);
4054         BUG_ON(ret);
4055
4056         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4057                 if (parent == 0)
4058                         parent = ins->objectid;
4059                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4060         } else
4061                 BUG_ON(parent > 0);
4062
4063         update_reserved_extents(root, ins->objectid, ins->offset, 1);
4064         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4065                 struct btrfs_delayed_extent_op *extent_op;
4066                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4067                 BUG_ON(!extent_op);
4068                 if (key)
4069                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4070                 else
4071                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4072                 extent_op->flags_to_set = flags;
4073                 extent_op->update_key = 1;
4074                 extent_op->update_flags = 1;
4075                 extent_op->is_data = 0;
4076
4077                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4078                                         ins->offset, parent, root_objectid,
4079                                         level, BTRFS_ADD_DELAYED_EXTENT,
4080                                         extent_op);
4081                 BUG_ON(ret);
4082         }
4083         return ret;
4084 }
4085
4086 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4087                                             struct btrfs_root *root,
4088                                             u64 bytenr, u32 blocksize,
4089                                             int level)
4090 {
4091         struct extent_buffer *buf;
4092
4093         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4094         if (!buf)
4095                 return ERR_PTR(-ENOMEM);
4096         btrfs_set_header_generation(buf, trans->transid);
4097         btrfs_set_buffer_lockdep_class(buf, level);
4098         btrfs_tree_lock(buf);
4099         clean_tree_block(trans, root, buf);
4100
4101         btrfs_set_lock_blocking(buf);
4102         btrfs_set_buffer_uptodate(buf);
4103
4104         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4105                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4106                          buf->start + buf->len - 1, GFP_NOFS);
4107         } else {
4108                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4109                          buf->start + buf->len - 1, GFP_NOFS);
4110         }
4111         trans->blocks_used++;
4112         /* this returns a buffer locked for blocking */
4113         return buf;
4114 }
4115
4116 /*
4117  * helper function to allocate a block for a given tree
4118  * returns the tree buffer or NULL.
4119  */
4120 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4121                                         struct btrfs_root *root, u32 blocksize,
4122                                         u64 parent, u64 root_objectid,
4123                                         struct btrfs_disk_key *key, int level,
4124                                         u64 hint, u64 empty_size)
4125 {
4126         struct btrfs_key ins;
4127         int ret;
4128         struct extent_buffer *buf;
4129
4130         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4131                                key, level, empty_size, hint, (u64)-1, &ins);
4132         if (ret) {
4133                 BUG_ON(ret > 0);
4134                 return ERR_PTR(ret);
4135         }
4136
4137         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4138                                     blocksize, level);
4139         return buf;
4140 }
4141
4142 #if 0
4143 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4144                         struct btrfs_root *root, struct extent_buffer *leaf)
4145 {
4146         u64 disk_bytenr;
4147         u64 num_bytes;
4148         struct btrfs_key key;
4149         struct btrfs_file_extent_item *fi;
4150         u32 nritems;
4151         int i;
4152         int ret;
4153
4154         BUG_ON(!btrfs_is_leaf(leaf));
4155         nritems = btrfs_header_nritems(leaf);
4156
4157         for (i = 0; i < nritems; i++) {
4158                 cond_resched();
4159                 btrfs_item_key_to_cpu(leaf, &key, i);
4160
4161                 /* only extents have references, skip everything else */
4162                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4163                         continue;
4164
4165                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4166
4167                 /* inline extents live in the btree, they don't have refs */
4168                 if (btrfs_file_extent_type(leaf, fi) ==
4169                     BTRFS_FILE_EXTENT_INLINE)
4170                         continue;
4171
4172                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4173
4174                 /* holes don't have refs */
4175                 if (disk_bytenr == 0)
4176                         continue;
4177
4178                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4179                 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4180                                         leaf->start, 0, key.objectid, 0);
4181                 BUG_ON(ret);
4182         }
4183         return 0;
4184 }
4185
4186 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4187                                         struct btrfs_root *root,
4188                                         struct btrfs_leaf_ref *ref)
4189 {
4190         int i;
4191         int ret;
4192         struct btrfs_extent_info *info;
4193         struct refsort *sorted;
4194
4195         if (ref->nritems == 0)
4196                 return 0;
4197
4198         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4199         for (i = 0; i < ref->nritems; i++) {
4200                 sorted[i].bytenr = ref->extents[i].bytenr;
4201                 sorted[i].slot = i;
4202         }
4203         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4204
4205         /*
4206          * the items in the ref were sorted when the ref was inserted
4207          * into the ref cache, so this is already in order
4208          */
4209         for (i = 0; i < ref->nritems; i++) {
4210                 info = ref->extents + sorted[i].slot;
4211                 ret = btrfs_free_extent(trans, root, info->bytenr,
4212                                           info->num_bytes, ref->bytenr,
4213                                           ref->owner, ref->generation,
4214                                           info->objectid, 0);
4215
4216                 atomic_inc(&root->fs_info->throttle_gen);
4217                 wake_up(&root->fs_info->transaction_throttle);
4218                 cond_resched();
4219
4220                 BUG_ON(ret);
4221                 info++;
4222         }
4223
4224         kfree(sorted);
4225         return 0;
4226 }
4227
4228
4229 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4230                                      struct btrfs_root *root, u64 start,
4231                                      u64 len, u32 *refs)
4232 {
4233         int ret;
4234
4235         ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4236         BUG_ON(ret);
4237
4238 #if 0 /* some debugging code in case we see problems here */
4239         /* if the refs count is one, it won't get increased again.  But
4240          * if the ref count is > 1, someone may be decreasing it at
4241          * the same time we are.
4242          */
4243         if (*refs != 1) {
4244                 struct extent_buffer *eb = NULL;
4245                 eb = btrfs_find_create_tree_block(root, start, len);
4246                 if (eb)
4247                         btrfs_tree_lock(eb);
4248
4249                 mutex_lock(&root->fs_info->alloc_mutex);
4250                 ret = lookup_extent_ref(NULL, root, start, len, refs);
4251                 BUG_ON(ret);
4252                 mutex_unlock(&root->fs_info->alloc_mutex);
4253
4254                 if (eb) {
4255                         btrfs_tree_unlock(eb);
4256                         free_extent_buffer(eb);
4257                 }
4258                 if (*refs == 1) {
4259                         printk(KERN_ERR "btrfs block %llu went down to one "
4260                                "during drop_snap\n", (unsigned long long)start);
4261                 }
4262
4263         }
4264 #endif
4265
4266         cond_resched();
4267         return ret;
4268 }
4269
4270
4271 /*
4272  * this is used while deleting old snapshots, and it drops the refs
4273  * on a whole subtree starting from a level 1 node.
4274  *
4275  * The idea is to sort all the leaf pointers, and then drop the
4276  * ref on all the leaves in order.  Most of the time the leaves
4277  * will have ref cache entries, so no leaf IOs will be required to
4278  * find the extents they have references on.
4279  *
4280  * For each leaf, any references it has are also dropped in order
4281  *
4282  * This ends up dropping the references in something close to optimal
4283  * order for reading and modifying the extent allocation tree.
4284  */
4285 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4286                                         struct btrfs_root *root,
4287                                         struct btrfs_path *path)
4288 {
4289         u64 bytenr;
4290         u64 root_owner;
4291         u64 root_gen;
4292         struct extent_buffer *eb = path->nodes[1];
4293         struct extent_buffer *leaf;
4294         struct btrfs_leaf_ref *ref;
4295         struct refsort *sorted = NULL;
4296         int nritems = btrfs_header_nritems(eb);
4297         int ret;
4298         int i;
4299         int refi = 0;
4300         int slot = path->slots[1];
4301         u32 blocksize = btrfs_level_size(root, 0);
4302         u32 refs;
4303
4304         if (nritems == 0)
4305                 goto out;
4306
4307         root_owner = btrfs_header_owner(eb);
4308         root_gen = btrfs_header_generation(eb);
4309         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4310
4311         /*
4312          * step one, sort all the leaf pointers so we don't scribble
4313          * randomly into the extent allocation tree
4314          */
4315         for (i = slot; i < nritems; i++) {
4316                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4317                 sorted[refi].slot = i;
4318                 refi++;
4319         }
4320
4321         /*
4322          * nritems won't be zero, but if we're picking up drop_snapshot
4323          * after a crash, slot might be > 0, so double check things
4324          * just in case.
4325          */
4326         if (refi == 0)
4327                 goto out;
4328
4329         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4330
4331         /*
4332          * the first loop frees everything the leaves point to
4333          */
4334         for (i = 0; i < refi; i++) {
4335                 u64 ptr_gen;
4336
4337                 bytenr = sorted[i].bytenr;
4338
4339                 /*
4340                  * check the reference count on this leaf.  If it is > 1
4341                  * we just decrement it below and don't update any
4342                  * of the refs the leaf points to.
4343                  */
4344                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4345                                                 blocksize, &refs);
4346                 BUG_ON(ret);
4347                 if (refs != 1)
4348                         continue;
4349
4350                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4351
4352                 /*
4353                  * the leaf only had one reference, which means the
4354                  * only thing pointing to this leaf is the snapshot
4355                  * we're deleting.  It isn't possible for the reference
4356                  * count to increase again later
4357                  *
4358                  * The reference cache is checked for the leaf,
4359                  * and if found we'll be able to drop any refs held by
4360                  * the leaf without needing to read it in.
4361                  */
4362                 ref = btrfs_lookup_leaf_ref(root, bytenr);
4363                 if (ref && ref->generation != ptr_gen) {
4364                         btrfs_free_leaf_ref(root, ref);
4365                         ref = NULL;
4366                 }
4367                 if (ref) {
4368                         ret = cache_drop_leaf_ref(trans, root, ref);
4369                         BUG_ON(ret);
4370                         btrfs_remove_leaf_ref(root, ref);
4371                         btrfs_free_leaf_ref(root, ref);
4372                 } else {
4373                         /*
4374                          * the leaf wasn't in the reference cache, so
4375                          * we have to read it.
4376                          */
4377                         leaf = read_tree_block(root, bytenr, blocksize,
4378                                                ptr_gen);
4379                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
4380                         BUG_ON(ret);
4381                         free_extent_buffer(leaf);
4382                 }
4383                 atomic_inc(&root->fs_info->throttle_gen);
4384                 wake_up(&root->fs_info->transaction_throttle);
4385                 cond_resched();
4386         }
4387
4388         /*
4389          * run through the loop again to free the refs on the leaves.
4390          * This is faster than doing it in the loop above because
4391          * the leaves are likely to be clustered together.  We end up
4392          * working in nice chunks on the extent allocation tree.
4393          */
4394         for (i = 0; i < refi; i++) {
4395                 bytenr = sorted[i].bytenr;
4396                 ret = btrfs_free_extent(trans, root, bytenr,
4397                                         blocksize, eb->start,
4398                                         root_owner, root_gen, 0, 1);
4399                 BUG_ON(ret);
4400
4401                 atomic_inc(&root->fs_info->throttle_gen);
4402                 wake_up(&root->fs_info->transaction_throttle);
4403                 cond_resched();
4404         }
4405 out:
4406         kfree(sorted);
4407
4408         /*
4409          * update the path to show we've processed the entire level 1
4410          * node.  This will get saved into the root's drop_snapshot_progress
4411          * field so these drops are not repeated again if this transaction
4412          * commits.
4413          */
4414         path->slots[1] = nritems;
4415         return 0;
4416 }
4417
4418 /*
4419  * helper function for drop_snapshot, this walks down the tree dropping ref
4420  * counts as it goes.
4421  */
4422 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4423                                    struct btrfs_root *root,
4424                                    struct btrfs_path *path, int *level)
4425 {
4426         u64 root_owner;
4427         u64 root_gen;
4428         u64 bytenr;
4429         u64 ptr_gen;
4430         struct extent_buffer *next;
4431         struct extent_buffer *cur;
4432         struct extent_buffer *parent;
4433         u32 blocksize;
4434         int ret;
4435         u32 refs;
4436
4437         WARN_ON(*level < 0);
4438         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4439         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4440                                 path->nodes[*level]->len, &refs);
4441         BUG_ON(ret);
4442         if (refs > 1)
4443                 goto out;
4444
4445         /*
4446          * walk down to the last node level and free all the leaves
4447          */
4448         while (*level >= 0) {
4449                 WARN_ON(*level < 0);
4450                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4451                 cur = path->nodes[*level];
4452
4453                 if (btrfs_header_level(cur) != *level)
4454                         WARN_ON(1);
4455
4456                 if (path->slots[*level] >=
4457                     btrfs_header_nritems(cur))
4458                         break;
4459
4460                 /* the new code goes down to level 1 and does all the
4461                  * leaves pointed to that node in bulk.  So, this check
4462                  * for level 0 will always be false.
4463                  *
4464                  * But, the disk format allows the drop_snapshot_progress
4465                  * field in the root to leave things in a state where
4466                  * a leaf will need cleaning up here.  If someone crashes
4467                  * with the old code and then boots with the new code,
4468                  * we might find a leaf here.
4469                  */
4470                 if (*level == 0) {
4471                         ret = btrfs_drop_leaf_ref(trans, root, cur);
4472                         BUG_ON(ret);
4473                         break;
4474                 }
4475
4476                 /*
4477                  * once we get to level one, process the whole node
4478                  * at once, including everything below it.
4479                  */
4480                 if (*level == 1) {
4481                         ret = drop_level_one_refs(trans, root, path);
4482                         BUG_ON(ret);
4483                         break;
4484                 }
4485
4486                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4487                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4488                 blocksize = btrfs_level_size(root, *level - 1);
4489
4490                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4491                                                 blocksize, &refs);
4492                 BUG_ON(ret);
4493
4494                 /*
4495                  * if there is more than one reference, we don't need
4496                  * to read that node to drop any references it has.  We
4497                  * just drop the ref we hold on that node and move on to the
4498                  * next slot in this level.
4499                  */
4500                 if (refs != 1) {
4501                         parent = path->nodes[*level];
4502                         root_owner = btrfs_header_owner(parent);
4503                         root_gen = btrfs_header_generation(parent);
4504                         path->slots[*level]++;
4505
4506                         ret = btrfs_free_extent(trans, root, bytenr,
4507                                                 blocksize, parent->start,
4508                                                 root_owner, root_gen,
4509                                                 *level - 1, 1);
4510                         BUG_ON(ret);
4511
4512                         atomic_inc(&root->fs_info->throttle_gen);
4513                         wake_up(&root->fs_info->transaction_throttle);
4514                         cond_resched();
4515
4516                         continue;
4517                 }
4518
4519                 /*
4520                  * we need to keep freeing things in the next level down.
4521                  * read the block and loop around to process it
4522                  */
4523                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4524                 WARN_ON(*level <= 0);
4525                 if (path->nodes[*level-1])
4526                         free_extent_buffer(path->nodes[*level-1]);
4527                 path->nodes[*level-1] = next;
4528                 *level = btrfs_header_level(next);
4529                 path->slots[*level] = 0;
4530                 cond_resched();
4531         }
4532 out:
4533         WARN_ON(*level < 0);
4534         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4535
4536         if (path->nodes[*level] == root->node) {
4537                 parent = path->nodes[*level];
4538                 bytenr = path->nodes[*level]->start;
4539         } else {
4540                 parent = path->nodes[*level + 1];
4541                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4542         }
4543
4544         blocksize = btrfs_level_size(root, *level);
4545         root_owner = btrfs_header_owner(parent);
4546         root_gen = btrfs_header_generation(parent);
4547
4548         /*
4549          * cleanup and free the reference on the last node
4550          * we processed
4551          */
4552         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4553                                   parent->start, root_owner, root_gen,
4554                                   *level, 1);
4555         free_extent_buffer(path->nodes[*level]);
4556         path->nodes[*level] = NULL;
4557
4558         *level += 1;
4559         BUG_ON(ret);
4560
4561         cond_resched();
4562         return 0;
4563 }
4564 #endif
4565
4566 struct walk_control {
4567         u64 refs[BTRFS_MAX_LEVEL];
4568         u64 flags[BTRFS_MAX_LEVEL];
4569         struct btrfs_key update_progress;
4570         int stage;
4571         int level;
4572         int shared_level;
4573         int update_ref;
4574         int keep_locks;
4575 };
4576
4577 #define DROP_REFERENCE  1
4578 #define UPDATE_BACKREF  2
4579
4580 /*
4581  * hepler to process tree block while walking down the tree.
4582  *
4583  * when wc->stage == DROP_REFERENCE, this function checks
4584  * reference count of the block. if the block is shared and
4585  * we need update back refs for the subtree rooted at the
4586  * block, this function changes wc->stage to UPDATE_BACKREF
4587  *
4588  * when wc->stage == UPDATE_BACKREF, this function updates
4589  * back refs for pointers in the block.
4590  *
4591  * NOTE: return value 1 means we should stop walking down.
4592  */
4593 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4594                                    struct btrfs_root *root,
4595                                    struct btrfs_path *path,
4596                                    struct walk_control *wc)
4597 {
4598         int level = wc->level;
4599         struct extent_buffer *eb = path->nodes[level];
4600         struct btrfs_key key;
4601         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4602         int ret;
4603
4604         if (wc->stage == UPDATE_BACKREF &&
4605             btrfs_header_owner(eb) != root->root_key.objectid)
4606                 return 1;
4607
4608         /*
4609          * when reference count of tree block is 1, it won't increase
4610          * again. once full backref flag is set, we never clear it.
4611          */
4612         if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4613             (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4614                 BUG_ON(!path->locks[level]);
4615                 ret = btrfs_lookup_extent_info(trans, root,
4616                                                eb->start, eb->len,
4617                                                &wc->refs[level],
4618                                                &wc->flags[level]);
4619                 BUG_ON(ret);
4620                 BUG_ON(wc->refs[level] == 0);
4621         }
4622
4623         if (wc->stage == DROP_REFERENCE &&
4624             wc->update_ref && wc->refs[level] > 1) {
4625                 BUG_ON(eb == root->node);
4626                 BUG_ON(path->slots[level] > 0);
4627                 if (level == 0)
4628                         btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4629                 else
4630                         btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4631                 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4632                     btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4633                         wc->stage = UPDATE_BACKREF;
4634                         wc->shared_level = level;
4635                 }
4636         }
4637
4638         if (wc->stage == DROP_REFERENCE) {
4639                 if (wc->refs[level] > 1)
4640                         return 1;
4641
4642                 if (path->locks[level] && !wc->keep_locks) {
4643                         btrfs_tree_unlock(eb);
4644                         path->locks[level] = 0;
4645                 }
4646                 return 0;
4647         }
4648
4649         /* wc->stage == UPDATE_BACKREF */
4650         if (!(wc->flags[level] & flag)) {
4651                 BUG_ON(!path->locks[level]);
4652                 ret = btrfs_inc_ref(trans, root, eb, 1);
4653                 BUG_ON(ret);
4654                 ret = btrfs_dec_ref(trans, root, eb, 0);
4655                 BUG_ON(ret);
4656                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4657                                                   eb->len, flag, 0);
4658                 BUG_ON(ret);
4659                 wc->flags[level] |= flag;
4660         }
4661
4662         /*
4663          * the block is shared by multiple trees, so it's not good to
4664          * keep the tree lock
4665          */
4666         if (path->locks[level] && level > 0) {
4667                 btrfs_tree_unlock(eb);
4668                 path->locks[level] = 0;
4669         }
4670         return 0;
4671 }
4672
4673 /*
4674  * hepler to process tree block while walking up the tree.
4675  *
4676  * when wc->stage == DROP_REFERENCE, this function drops
4677  * reference count on the block.
4678  *
4679  * when wc->stage == UPDATE_BACKREF, this function changes
4680  * wc->stage back to DROP_REFERENCE if we changed wc->stage
4681  * to UPDATE_BACKREF previously while processing the block.
4682  *
4683  * NOTE: return value 1 means we should stop walking up.
4684  */
4685 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4686                                  struct btrfs_root *root,
4687                                  struct btrfs_path *path,
4688                                  struct walk_control *wc)
4689 {
4690         int ret = 0;
4691         int level = wc->level;
4692         struct extent_buffer *eb = path->nodes[level];
4693         u64 parent = 0;
4694
4695         if (wc->stage == UPDATE_BACKREF) {
4696                 BUG_ON(wc->shared_level < level);
4697                 if (level < wc->shared_level)
4698                         goto out;
4699
4700                 BUG_ON(wc->refs[level] <= 1);
4701                 ret = find_next_key(path, level + 1, &wc->update_progress);
4702                 if (ret > 0)
4703                         wc->update_ref = 0;
4704
4705                 wc->stage = DROP_REFERENCE;
4706                 wc->shared_level = -1;
4707                 path->slots[level] = 0;
4708
4709                 /*
4710                  * check reference count again if the block isn't locked.
4711                  * we should start walking down the tree again if reference
4712                  * count is one.
4713                  */
4714                 if (!path->locks[level]) {
4715                         BUG_ON(level == 0);
4716                         btrfs_tree_lock(eb);
4717                         btrfs_set_lock_blocking(eb);
4718                         path->locks[level] = 1;
4719
4720                         ret = btrfs_lookup_extent_info(trans, root,
4721                                                        eb->start, eb->len,
4722                                                        &wc->refs[level],
4723                                                        &wc->flags[level]);
4724                         BUG_ON(ret);
4725                         BUG_ON(wc->refs[level] == 0);
4726                         if (wc->refs[level] == 1) {
4727                                 btrfs_tree_unlock(eb);
4728                                 path->locks[level] = 0;
4729                                 return 1;
4730                         }
4731                 } else {
4732                         BUG_ON(level != 0);
4733                 }
4734         }
4735
4736         /* wc->stage == DROP_REFERENCE */
4737         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4738
4739         if (wc->refs[level] == 1) {
4740                 if (level == 0) {
4741                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4742                                 ret = btrfs_dec_ref(trans, root, eb, 1);
4743                         else
4744                                 ret = btrfs_dec_ref(trans, root, eb, 0);
4745                         BUG_ON(ret);
4746                 }
4747                 /* make block locked assertion in clean_tree_block happy */
4748                 if (!path->locks[level] &&
4749                     btrfs_header_generation(eb) == trans->transid) {
4750                         btrfs_tree_lock(eb);
4751                         btrfs_set_lock_blocking(eb);
4752                         path->locks[level] = 1;
4753                 }
4754                 clean_tree_block(trans, root, eb);
4755         }
4756
4757         if (eb == root->node) {
4758                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4759                         parent = eb->start;
4760                 else
4761                         BUG_ON(root->root_key.objectid !=
4762                                btrfs_header_owner(eb));
4763         } else {
4764                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4765                         parent = path->nodes[level + 1]->start;
4766                 else
4767                         BUG_ON(root->root_key.objectid !=
4768                                btrfs_header_owner(path->nodes[level + 1]));
4769         }
4770
4771         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
4772                                 root->root_key.objectid, level, 0);
4773         BUG_ON(ret);
4774 out:
4775         wc->refs[level] = 0;
4776         wc->flags[level] = 0;
4777         return ret;
4778 }
4779
4780 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4781                                    struct btrfs_root *root,
4782                                    struct btrfs_path *path,
4783                                    struct walk_control *wc)
4784 {
4785         struct extent_buffer *next;
4786         struct extent_buffer *cur;
4787         u64 bytenr;
4788         u64 ptr_gen;
4789         u32 blocksize;
4790         int level = wc->level;
4791         int ret;
4792
4793         while (level >= 0) {
4794                 cur = path->nodes[level];
4795                 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
4796
4797                 ret = walk_down_proc(trans, root, path, wc);
4798                 if (ret > 0)
4799                         break;
4800
4801                 if (level == 0)
4802                         break;
4803
4804                 bytenr = btrfs_node_blockptr(cur, path->slots[level]);
4805                 blocksize = btrfs_level_size(root, level - 1);
4806                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
4807
4808                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4809                 btrfs_tree_lock(next);
4810                 btrfs_set_lock_blocking(next);
4811
4812                 level--;
4813                 BUG_ON(level != btrfs_header_level(next));
4814                 path->nodes[level] = next;
4815                 path->slots[level] = 0;
4816                 path->locks[level] = 1;
4817                 wc->level = level;
4818         }
4819         return 0;
4820 }
4821
4822 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4823                                  struct btrfs_root *root,
4824                                  struct btrfs_path *path,
4825                                  struct walk_control *wc, int max_level)
4826 {
4827         int level = wc->level;
4828         int ret;
4829
4830         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
4831         while (level < max_level && path->nodes[level]) {
4832                 wc->level = level;
4833                 if (path->slots[level] + 1 <
4834                     btrfs_header_nritems(path->nodes[level])) {
4835                         path->slots[level]++;
4836                         return 0;
4837                 } else {
4838                         ret = walk_up_proc(trans, root, path, wc);
4839                         if (ret > 0)
4840                                 return 0;
4841
4842                         if (path->locks[level]) {
4843                                 btrfs_tree_unlock(path->nodes[level]);
4844                                 path->locks[level] = 0;
4845                         }
4846                         free_extent_buffer(path->nodes[level]);
4847                         path->nodes[level] = NULL;
4848                         level++;
4849                 }
4850         }
4851         return 1;
4852 }
4853
4854 /*
4855  * drop a subvolume tree.
4856  *
4857  * this function traverses the tree freeing any blocks that only
4858  * referenced by the tree.
4859  *
4860  * when a shared tree block is found. this function decreases its
4861  * reference count by one. if update_ref is true, this function
4862  * also make sure backrefs for the shared block and all lower level
4863  * blocks are properly updated.
4864  */
4865 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
4866 {
4867         struct btrfs_path *path;
4868         struct btrfs_trans_handle *trans;
4869         struct btrfs_root *tree_root = root->fs_info->tree_root;
4870         struct btrfs_root_item *root_item = &root->root_item;
4871         struct walk_control *wc;
4872         struct btrfs_key key;
4873         int err = 0;
4874         int ret;
4875         int level;
4876
4877         path = btrfs_alloc_path();
4878         BUG_ON(!path);
4879
4880         wc = kzalloc(sizeof(*wc), GFP_NOFS);
4881         BUG_ON(!wc);
4882
4883         trans = btrfs_start_transaction(tree_root, 1);
4884
4885         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4886                 level = btrfs_header_level(root->node);
4887                 path->nodes[level] = btrfs_lock_root_node(root);
4888                 btrfs_set_lock_blocking(path->nodes[level]);
4889                 path->slots[level] = 0;
4890                 path->locks[level] = 1;
4891                 memset(&wc->update_progress, 0,
4892                        sizeof(wc->update_progress));
4893         } else {
4894                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4895                 memcpy(&wc->update_progress, &key,
4896                        sizeof(wc->update_progress));
4897
4898                 level = root_item->drop_level;
4899                 BUG_ON(level == 0);
4900                 path->lowest_level = level;
4901                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4902                 path->lowest_level = 0;
4903                 if (ret < 0) {
4904                         err = ret;
4905                         goto out;
4906                 }
4907                 btrfs_node_key_to_cpu(path->nodes[level], &key,
4908                                       path->slots[level]);
4909                 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
4910
4911                 /*
4912                  * unlock our path, this is safe because only this
4913                  * function is allowed to delete this snapshot
4914                  */
4915                 btrfs_unlock_up_safe(path, 0);
4916
4917                 level = btrfs_header_level(root->node);
4918                 while (1) {
4919                         btrfs_tree_lock(path->nodes[level]);
4920                         btrfs_set_lock_blocking(path->nodes[level]);
4921
4922                         ret = btrfs_lookup_extent_info(trans, root,
4923                                                 path->nodes[level]->start,
4924                                                 path->nodes[level]->len,
4925                                                 &wc->refs[level],
4926                                                 &wc->flags[level]);
4927                         BUG_ON(ret);
4928                         BUG_ON(wc->refs[level] == 0);
4929
4930                         if (level == root_item->drop_level)
4931                                 break;
4932
4933                         btrfs_tree_unlock(path->nodes[level]);
4934                         WARN_ON(wc->refs[level] != 1);
4935                         level--;
4936                 }
4937         }
4938
4939         wc->level = level;
4940         wc->shared_level = -1;
4941         wc->stage = DROP_REFERENCE;
4942         wc->update_ref = update_ref;
4943         wc->keep_locks = 0;
4944
4945         while (1) {
4946                 ret = walk_down_tree(trans, root, path, wc);
4947                 if (ret < 0) {
4948                         err = ret;
4949                         break;
4950                 }
4951
4952                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
4953                 if (ret < 0) {
4954                         err = ret;
4955                         break;
4956                 }
4957
4958                 if (ret > 0) {
4959                         BUG_ON(wc->stage != DROP_REFERENCE);
4960                         break;
4961                 }
4962
4963                 if (wc->stage == DROP_REFERENCE) {
4964                         level = wc->level;
4965                         btrfs_node_key(path->nodes[level],
4966                                        &root_item->drop_progress,
4967                                        path->slots[level]);
4968                         root_item->drop_level = level;
4969                 }
4970
4971                 BUG_ON(wc->level == 0);
4972                 if (trans->transaction->in_commit ||
4973                     trans->transaction->delayed_refs.flushing) {
4974                         ret = btrfs_update_root(trans, tree_root,
4975                                                 &root->root_key,
4976                                                 root_item);
4977                         BUG_ON(ret);
4978
4979                         btrfs_end_transaction(trans, tree_root);
4980                         trans = btrfs_start_transaction(tree_root, 1);
4981                 } else {
4982                         unsigned long update;
4983                         update = trans->delayed_ref_updates;
4984                         trans->delayed_ref_updates = 0;
4985                         if (update)
4986                                 btrfs_run_delayed_refs(trans, tree_root,
4987                                                        update);
4988                 }
4989         }
4990         btrfs_release_path(root, path);
4991         BUG_ON(err);
4992
4993         ret = btrfs_del_root(trans, tree_root, &root->root_key);
4994         BUG_ON(ret);
4995
4996         free_extent_buffer(root->node);
4997         free_extent_buffer(root->commit_root);
4998         kfree(root);
4999 out:
5000         btrfs_end_transaction(trans, tree_root);
5001         kfree(wc);
5002         btrfs_free_path(path);
5003         return err;
5004 }
5005
5006 /*
5007  * drop subtree rooted at tree block 'node'.
5008  *
5009  * NOTE: this function will unlock and release tree block 'node'
5010  */
5011 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5012                         struct btrfs_root *root,
5013                         struct extent_buffer *node,
5014                         struct extent_buffer *parent)
5015 {
5016         struct btrfs_path *path;
5017         struct walk_control *wc;
5018         int level;
5019         int parent_level;
5020         int ret = 0;
5021         int wret;
5022
5023         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5024
5025         path = btrfs_alloc_path();
5026         BUG_ON(!path);
5027
5028         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5029         BUG_ON(!wc);
5030
5031         btrfs_assert_tree_locked(parent);
5032         parent_level = btrfs_header_level(parent);
5033         extent_buffer_get(parent);
5034         path->nodes[parent_level] = parent;
5035         path->slots[parent_level] = btrfs_header_nritems(parent);
5036
5037         btrfs_assert_tree_locked(node);
5038         level = btrfs_header_level(node);
5039         path->nodes[level] = node;
5040         path->slots[level] = 0;
5041         path->locks[level] = 1;
5042
5043         wc->refs[parent_level] = 1;
5044         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5045         wc->level = level;
5046         wc->shared_level = -1;
5047         wc->stage = DROP_REFERENCE;
5048         wc->update_ref = 0;
5049         wc->keep_locks = 1;
5050
5051         while (1) {
5052                 wret = walk_down_tree(trans, root, path, wc);
5053                 if (wret < 0) {
5054                         ret = wret;
5055                         break;
5056                 }
5057
5058                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5059                 if (wret < 0)
5060                         ret = wret;
5061                 if (wret != 0)
5062                         break;
5063         }
5064
5065         kfree(wc);
5066         btrfs_free_path(path);
5067         return ret;
5068 }
5069
5070 #if 0
5071 static unsigned long calc_ra(unsigned long start, unsigned long last,
5072                              unsigned long nr)
5073 {
5074         return min(last, start + nr - 1);
5075 }
5076
5077 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5078                                          u64 len)
5079 {
5080         u64 page_start;
5081         u64 page_end;
5082         unsigned long first_index;
5083         unsigned long last_index;
5084         unsigned long i;
5085         struct page *page;
5086         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5087         struct file_ra_state *ra;
5088         struct btrfs_ordered_extent *ordered;
5089         unsigned int total_read = 0;
5090         unsigned int total_dirty = 0;
5091         int ret = 0;
5092
5093         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5094
5095         mutex_lock(&inode->i_mutex);
5096         first_index = start >> PAGE_CACHE_SHIFT;
5097         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5098
5099         /* make sure the dirty trick played by the caller work */
5100         ret = invalidate_inode_pages2_range(inode->i_mapping,
5101                                             first_index, last_index);
5102         if (ret)
5103                 goto out_unlock;
5104
5105         file_ra_state_init(ra, inode->i_mapping);
5106
5107         for (i = first_index ; i <= last_index; i++) {
5108                 if (total_read % ra->ra_pages == 0) {
5109                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5110                                        calc_ra(i, last_index, ra->ra_pages));
5111                 }
5112                 total_read++;
5113 again:
5114                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5115                         BUG_ON(1);
5116                 page = grab_cache_page(inode->i_mapping, i);
5117                 if (!page) {
5118                         ret = -ENOMEM;
5119                         goto out_unlock;
5120                 }
5121                 if (!PageUptodate(page)) {
5122                         btrfs_readpage(NULL, page);
5123                         lock_page(page);
5124                         if (!PageUptodate(page)) {
5125                                 unlock_page(page);
5126                                 page_cache_release(page);
5127                                 ret = -EIO;
5128                                 goto out_unlock;
5129                         }
5130                 }
5131                 wait_on_page_writeback(page);
5132
5133                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5134                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5135                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5136
5137                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5138                 if (ordered) {
5139                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5140                         unlock_page(page);
5141                         page_cache_release(page);
5142                         btrfs_start_ordered_extent(inode, ordered, 1);
5143                         btrfs_put_ordered_extent(ordered);
5144                         goto again;
5145                 }
5146                 set_page_extent_mapped(page);
5147
5148                 if (i == first_index)
5149                         set_extent_bits(io_tree, page_start, page_end,
5150                                         EXTENT_BOUNDARY, GFP_NOFS);
5151                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5152
5153                 set_page_dirty(page);
5154                 total_dirty++;
5155
5156                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5157                 unlock_page(page);
5158                 page_cache_release(page);
5159         }
5160
5161 out_unlock:
5162         kfree(ra);
5163         mutex_unlock(&inode->i_mutex);
5164         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5165         return ret;
5166 }
5167
5168 static noinline int relocate_data_extent(struct inode *reloc_inode,
5169                                          struct btrfs_key *extent_key,
5170                                          u64 offset)
5171 {
5172         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5173         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5174         struct extent_map *em;
5175         u64 start = extent_key->objectid - offset;
5176         u64 end = start + extent_key->offset - 1;
5177
5178         em = alloc_extent_map(GFP_NOFS);
5179         BUG_ON(!em || IS_ERR(em));
5180
5181         em->start = start;
5182         em->len = extent_key->offset;
5183         em->block_len = extent_key->offset;
5184         em->block_start = extent_key->objectid;
5185         em->bdev = root->fs_info->fs_devices->latest_bdev;
5186         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5187
5188         /* setup extent map to cheat btrfs_readpage */
5189         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5190         while (1) {
5191                 int ret;
5192                 spin_lock(&em_tree->lock);
5193                 ret = add_extent_mapping(em_tree, em);
5194                 spin_unlock(&em_tree->lock);
5195                 if (ret != -EEXIST) {
5196                         free_extent_map(em);
5197                         break;
5198                 }
5199                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5200         }
5201         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5202
5203         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5204 }
5205
5206 struct btrfs_ref_path {
5207         u64 extent_start;
5208         u64 nodes[BTRFS_MAX_LEVEL];
5209         u64 root_objectid;
5210         u64 root_generation;
5211         u64 owner_objectid;
5212         u32 num_refs;
5213         int lowest_level;
5214         int current_level;
5215         int shared_level;
5216
5217         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5218         u64 new_nodes[BTRFS_MAX_LEVEL];
5219 };
5220
5221 struct disk_extent {
5222         u64 ram_bytes;
5223         u64 disk_bytenr;
5224         u64 disk_num_bytes;
5225         u64 offset;
5226         u64 num_bytes;
5227         u8 compression;
5228         u8 encryption;
5229         u16 other_encoding;
5230 };
5231
5232 static int is_cowonly_root(u64 root_objectid)
5233 {
5234         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5235             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5236             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5237             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5238             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5239             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5240                 return 1;
5241         return 0;
5242 }
5243
5244 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5245                                     struct btrfs_root *extent_root,
5246                                     struct btrfs_ref_path *ref_path,
5247                                     int first_time)
5248 {
5249         struct extent_buffer *leaf;
5250         struct btrfs_path *path;
5251         struct btrfs_extent_ref *ref;
5252         struct btrfs_key key;
5253         struct btrfs_key found_key;
5254         u64 bytenr;
5255         u32 nritems;
5256         int level;
5257         int ret = 1;
5258
5259         path = btrfs_alloc_path();
5260         if (!path)
5261                 return -ENOMEM;
5262
5263         if (first_time) {
5264                 ref_path->lowest_level = -1;
5265                 ref_path->current_level = -1;
5266                 ref_path->shared_level = -1;
5267                 goto walk_up;
5268         }
5269 walk_down:
5270         level = ref_path->current_level - 1;
5271         while (level >= -1) {
5272                 u64 parent;
5273                 if (level < ref_path->lowest_level)
5274                         break;
5275
5276                 if (level >= 0)
5277                         bytenr = ref_path->nodes[level];
5278                 else
5279                         bytenr = ref_path->extent_start;
5280                 BUG_ON(bytenr == 0);
5281
5282                 parent = ref_path->nodes[level + 1];
5283                 ref_path->nodes[level + 1] = 0;
5284                 ref_path->current_level = level;
5285                 BUG_ON(parent == 0);
5286
5287                 key.objectid = bytenr;
5288                 key.offset = parent + 1;
5289                 key.type = BTRFS_EXTENT_REF_KEY;
5290
5291                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5292                 if (ret < 0)
5293                         goto out;
5294                 BUG_ON(ret == 0);
5295
5296                 leaf = path->nodes[0];
5297                 nritems = btrfs_header_nritems(leaf);
5298                 if (path->slots[0] >= nritems) {
5299                         ret = btrfs_next_leaf(extent_root, path);
5300                         if (ret < 0)
5301                                 goto out;
5302                         if (ret > 0)
5303                                 goto next;
5304                         leaf = path->nodes[0];
5305                 }
5306
5307                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5308                 if (found_key.objectid == bytenr &&
5309                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5310                         if (level < ref_path->shared_level)
5311                                 ref_path->shared_level = level;
5312                         goto found;
5313                 }
5314 next:
5315                 level--;
5316                 btrfs_release_path(extent_root, path);
5317                 cond_resched();
5318         }
5319         /* reached lowest level */
5320         ret = 1;
5321         goto out;
5322 walk_up:
5323         level = ref_path->current_level;
5324         while (level < BTRFS_MAX_LEVEL - 1) {
5325                 u64 ref_objectid;
5326
5327                 if (level >= 0)
5328                         bytenr = ref_path->nodes[level];
5329                 else
5330                         bytenr = ref_path->extent_start;
5331
5332                 BUG_ON(bytenr == 0);
5333
5334                 key.objectid = bytenr;
5335                 key.offset = 0;
5336                 key.type = BTRFS_EXTENT_REF_KEY;
5337
5338                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5339                 if (ret < 0)
5340                         goto out;
5341
5342                 leaf = path->nodes[0];
5343                 nritems = btrfs_header_nritems(leaf);
5344                 if (path->slots[0] >= nritems) {
5345                         ret = btrfs_next_leaf(extent_root, path);
5346                         if (ret < 0)
5347                                 goto out;
5348                         if (ret > 0) {
5349                                 /* the extent was freed by someone */
5350                                 if (ref_path->lowest_level == level)
5351                                         goto out;
5352                                 btrfs_release_path(extent_root, path);
5353                                 goto walk_down;
5354                         }
5355                         leaf = path->nodes[0];
5356                 }
5357
5358                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5359                 if (found_key.objectid != bytenr ||
5360                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5361                         /* the extent was freed by someone */
5362                         if (ref_path->lowest_level == level) {
5363                                 ret = 1;
5364                                 goto out;
5365                         }
5366                         btrfs_release_path(extent_root, path);
5367                         goto walk_down;
5368                 }
5369 found:
5370                 ref = btrfs_item_ptr(leaf, path->slots[0],
5371                                 struct btrfs_extent_ref);
5372                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5373                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5374                         if (first_time) {
5375                                 level = (int)ref_objectid;
5376                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5377                                 ref_path->lowest_level = level;
5378                                 ref_path->current_level = level;
5379                                 ref_path->nodes[level] = bytenr;
5380                         } else {
5381                                 WARN_ON(ref_objectid != level);
5382                         }
5383                 } else {
5384                         WARN_ON(level != -1);
5385                 }
5386                 first_time = 0;
5387
5388                 if (ref_path->lowest_level == level) {
5389                         ref_path->owner_objectid = ref_objectid;
5390                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5391                 }
5392
5393                 /*
5394                  * the block is tree root or the block isn't in reference
5395                  * counted tree.
5396                  */
5397                 if (found_key.objectid == found_key.offset ||
5398                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5399                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5400                         ref_path->root_generation =
5401                                 btrfs_ref_generation(leaf, ref);
5402                         if (level < 0) {
5403                                 /* special reference from the tree log */
5404                                 ref_path->nodes[0] = found_key.offset;
5405                                 ref_path->current_level = 0;
5406                         }
5407                         ret = 0;
5408                         goto out;
5409                 }
5410
5411                 level++;
5412                 BUG_ON(ref_path->nodes[level] != 0);
5413                 ref_path->nodes[level] = found_key.offset;
5414                 ref_path->current_level = level;
5415
5416                 /*
5417                  * the reference was created in the running transaction,
5418                  * no need to continue walking up.
5419                  */
5420                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5421                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5422                         ref_path->root_generation =
5423                                 btrfs_ref_generation(leaf, ref);
5424                         ret = 0;
5425                         goto out;
5426                 }
5427
5428                 btrfs_release_path(extent_root, path);
5429                 cond_resched();
5430         }
5431         /* reached max tree level, but no tree root found. */
5432         BUG();
5433 out:
5434         btrfs_free_path(path);
5435         return ret;
5436 }
5437
5438 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5439                                 struct btrfs_root *extent_root,
5440                                 struct btrfs_ref_path *ref_path,
5441                                 u64 extent_start)
5442 {
5443         memset(ref_path, 0, sizeof(*ref_path));
5444         ref_path->extent_start = extent_start;
5445
5446         return __next_ref_path(trans, extent_root, ref_path, 1);
5447 }
5448
5449 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5450                                struct btrfs_root *extent_root,
5451                                struct btrfs_ref_path *ref_path)
5452 {
5453         return __next_ref_path(trans, extent_root, ref_path, 0);
5454 }
5455
5456 static noinline int get_new_locations(struct inode *reloc_inode,
5457                                       struct btrfs_key *extent_key,
5458                                       u64 offset, int no_fragment,
5459                                       struct disk_extent **extents,
5460                                       int *nr_extents)
5461 {
5462         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5463         struct btrfs_path *path;
5464         struct btrfs_file_extent_item *fi;
5465         struct extent_buffer *leaf;
5466         struct disk_extent *exts = *extents;
5467         struct btrfs_key found_key;
5468         u64 cur_pos;
5469         u64 last_byte;
5470         u32 nritems;
5471         int nr = 0;
5472         int max = *nr_extents;
5473         int ret;
5474
5475         WARN_ON(!no_fragment && *extents);
5476         if (!exts) {
5477                 max = 1;
5478                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5479                 if (!exts)
5480                         return -ENOMEM;
5481         }
5482
5483         path = btrfs_alloc_path();
5484         BUG_ON(!path);
5485
5486         cur_pos = extent_key->objectid - offset;
5487         last_byte = extent_key->objectid + extent_key->offset;
5488         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5489                                        cur_pos, 0);
5490         if (ret < 0)
5491                 goto out;
5492         if (ret > 0) {
5493                 ret = -ENOENT;
5494                 goto out;
5495         }
5496
5497         while (1) {
5498                 leaf = path->nodes[0];
5499                 nritems = btrfs_header_nritems(leaf);
5500                 if (path->slots[0] >= nritems) {
5501                         ret = btrfs_next_leaf(root, path);
5502                         if (ret < 0)
5503                                 goto out;
5504                         if (ret > 0)
5505                                 break;
5506                         leaf = path->nodes[0];
5507                 }
5508
5509                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5510                 if (found_key.offset != cur_pos ||
5511                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
5512                     found_key.objectid != reloc_inode->i_ino)
5513                         break;
5514
5515                 fi = btrfs_item_ptr(leaf, path->slots[0],
5516                                     struct btrfs_file_extent_item);
5517                 if (btrfs_file_extent_type(leaf, fi) !=
5518                     BTRFS_FILE_EXTENT_REG ||
5519                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5520                         break;
5521
5522                 if (nr == max) {
5523                         struct disk_extent *old = exts;
5524                         max *= 2;
5525                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5526                         memcpy(exts, old, sizeof(*exts) * nr);
5527                         if (old != *extents)
5528                                 kfree(old);
5529                 }
5530
5531                 exts[nr].disk_bytenr =
5532                         btrfs_file_extent_disk_bytenr(leaf, fi);
5533                 exts[nr].disk_num_bytes =
5534                         btrfs_file_extent_disk_num_bytes(leaf, fi);
5535                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5536                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5537                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5538                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5539                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5540                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5541                                                                            fi);
5542                 BUG_ON(exts[nr].offset > 0);
5543                 BUG_ON(exts[nr].compression || exts[nr].encryption);
5544                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5545
5546                 cur_pos += exts[nr].num_bytes;
5547                 nr++;
5548
5549                 if (cur_pos + offset >= last_byte)
5550                         break;
5551
5552                 if (no_fragment) {
5553                         ret = 1;
5554                         goto out;
5555                 }
5556                 path->slots[0]++;
5557         }
5558
5559         BUG_ON(cur_pos + offset > last_byte);
5560         if (cur_pos + offset < last_byte) {
5561                 ret = -ENOENT;
5562                 goto out;
5563         }
5564         ret = 0;
5565 out:
5566         btrfs_free_path(path);
5567         if (ret) {
5568                 if (exts != *extents)
5569                         kfree(exts);
5570         } else {
5571                 *extents = exts;
5572                 *nr_extents = nr;
5573         }
5574         return ret;
5575 }
5576
5577 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5578                                         struct btrfs_root *root,
5579                                         struct btrfs_path *path,
5580                                         struct btrfs_key *extent_key,
5581                                         struct btrfs_key *leaf_key,
5582                                         struct btrfs_ref_path *ref_path,
5583                                         struct disk_extent *new_extents,
5584                                         int nr_extents)
5585 {
5586         struct extent_buffer *leaf;
5587         struct btrfs_file_extent_item *fi;
5588         struct inode *inode = NULL;
5589         struct btrfs_key key;
5590         u64 lock_start = 0;
5591         u64 lock_end = 0;
5592         u64 num_bytes;
5593         u64 ext_offset;
5594         u64 search_end = (u64)-1;
5595         u32 nritems;
5596         int nr_scaned = 0;
5597         int extent_locked = 0;
5598         int extent_type;
5599         int ret;
5600
5601         memcpy(&key, leaf_key, sizeof(key));
5602         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5603                 if (key.objectid < ref_path->owner_objectid ||
5604                     (key.objectid == ref_path->owner_objectid &&
5605                      key.type < BTRFS_EXTENT_DATA_KEY)) {
5606                         key.objectid = ref_path->owner_objectid;
5607                         key.type = BTRFS_EXTENT_DATA_KEY;
5608                         key.offset = 0;
5609                 }
5610         }
5611
5612         while (1) {
5613                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5614                 if (ret < 0)
5615                         goto out;
5616
5617                 leaf = path->nodes[0];
5618                 nritems = btrfs_header_nritems(leaf);
5619 next:
5620                 if (extent_locked && ret > 0) {
5621                         /*
5622                          * the file extent item was modified by someone
5623                          * before the extent got locked.
5624                          */
5625                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5626                                       lock_end, GFP_NOFS);
5627                         extent_locked = 0;
5628                 }
5629
5630                 if (path->slots[0] >= nritems) {
5631                         if (++nr_scaned > 2)
5632                                 break;
5633
5634                         BUG_ON(extent_locked);
5635                         ret = btrfs_next_leaf(root, path);
5636                         if (ret < 0)
5637                                 goto out;
5638                         if (ret > 0)
5639                                 break;
5640                         leaf = path->nodes[0];
5641                         nritems = btrfs_header_nritems(leaf);
5642                 }
5643
5644                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5645
5646                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5647                         if ((key.objectid > ref_path->owner_objectid) ||
5648                             (key.objectid == ref_path->owner_objectid &&
5649                              key.type > BTRFS_EXTENT_DATA_KEY) ||
5650                             key.offset >= search_end)
5651                                 break;
5652                 }
5653
5654                 if (inode && key.objectid != inode->i_ino) {
5655                         BUG_ON(extent_locked);
5656                         btrfs_release_path(root, path);
5657                         mutex_unlock(&inode->i_mutex);
5658                         iput(inode);
5659                         inode = NULL;
5660                         continue;
5661                 }
5662
5663                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5664                         path->slots[0]++;
5665                         ret = 1;
5666                         goto next;
5667                 }
5668                 fi = btrfs_item_ptr(leaf, path->slots[0],
5669                                     struct btrfs_file_extent_item);
5670                 extent_type = btrfs_file_extent_type(leaf, fi);
5671                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5672                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5673                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5674                      extent_key->objectid)) {
5675                         path->slots[0]++;
5676                         ret = 1;
5677                         goto next;
5678                 }
5679
5680                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5681                 ext_offset = btrfs_file_extent_offset(leaf, fi);
5682
5683                 if (search_end == (u64)-1) {
5684                         search_end = key.offset - ext_offset +
5685                                 btrfs_file_extent_ram_bytes(leaf, fi);
5686                 }
5687
5688                 if (!extent_locked) {
5689                         lock_start = key.offset;
5690                         lock_end = lock_start + num_bytes - 1;
5691                 } else {
5692                         if (lock_start > key.offset ||
5693                             lock_end + 1 < key.offset + num_bytes) {
5694                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5695                                               lock_start, lock_end, GFP_NOFS);
5696                                 extent_locked = 0;
5697                         }
5698                 }
5699
5700                 if (!inode) {
5701                         btrfs_release_path(root, path);
5702
5703                         inode = btrfs_iget_locked(root->fs_info->sb,
5704                                                   key.objectid, root);
5705                         if (inode->i_state & I_NEW) {
5706                                 BTRFS_I(inode)->root = root;
5707                                 BTRFS_I(inode)->location.objectid =
5708                                         key.objectid;
5709                                 BTRFS_I(inode)->location.type =
5710                                         BTRFS_INODE_ITEM_KEY;
5711                                 BTRFS_I(inode)->location.offset = 0;
5712                                 btrfs_read_locked_inode(inode);
5713                                 unlock_new_inode(inode);
5714                         }
5715                         /*
5716                          * some code call btrfs_commit_transaction while
5717                          * holding the i_mutex, so we can't use mutex_lock
5718                          * here.
5719                          */
5720                         if (is_bad_inode(inode) ||
5721                             !mutex_trylock(&inode->i_mutex)) {
5722                                 iput(inode);
5723                                 inode = NULL;
5724                                 key.offset = (u64)-1;
5725                                 goto skip;
5726                         }
5727                 }
5728
5729                 if (!extent_locked) {
5730                         struct btrfs_ordered_extent *ordered;
5731
5732                         btrfs_release_path(root, path);
5733
5734                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5735                                     lock_end, GFP_NOFS);
5736                         ordered = btrfs_lookup_first_ordered_extent(inode,
5737                                                                     lock_end);
5738                         if (ordered &&
5739                             ordered->file_offset <= lock_end &&
5740                             ordered->file_offset + ordered->len > lock_start) {
5741                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5742                                               lock_start, lock_end, GFP_NOFS);
5743                                 btrfs_start_ordered_extent(inode, ordered, 1);
5744                                 btrfs_put_ordered_extent(ordered);
5745                                 key.offset += num_bytes;
5746                                 goto skip;
5747                         }
5748                         if (ordered)
5749                                 btrfs_put_ordered_extent(ordered);
5750
5751                         extent_locked = 1;
5752                         continue;
5753                 }
5754
5755                 if (nr_extents == 1) {
5756                         /* update extent pointer in place */
5757                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
5758                                                 new_extents[0].disk_bytenr);
5759                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5760                                                 new_extents[0].disk_num_bytes);
5761                         btrfs_mark_buffer_dirty(leaf);
5762
5763                         btrfs_drop_extent_cache(inode, key.offset,
5764                                                 key.offset + num_bytes - 1, 0);
5765
5766                         ret = btrfs_inc_extent_ref(trans, root,
5767                                                 new_extents[0].disk_bytenr,
5768                                                 new_extents[0].disk_num_bytes,
5769                                                 leaf->start,
5770                                                 root->root_key.objectid,
5771                                                 trans->transid,
5772                                                 key.objectid);
5773                         BUG_ON(ret);
5774
5775                         ret = btrfs_free_extent(trans, root,
5776                                                 extent_key->objectid,
5777                                                 extent_key->offset,
5778                                                 leaf->start,
5779                                                 btrfs_header_owner(leaf),
5780                                                 btrfs_header_generation(leaf),
5781                                                 key.objectid, 0);
5782                         BUG_ON(ret);
5783
5784                         btrfs_release_path(root, path);
5785                         key.offset += num_bytes;
5786                 } else {
5787                         BUG_ON(1);
5788 #if 0
5789                         u64 alloc_hint;
5790                         u64 extent_len;
5791                         int i;
5792                         /*
5793                          * drop old extent pointer at first, then insert the
5794                          * new pointers one bye one
5795                          */
5796                         btrfs_release_path(root, path);
5797                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
5798                                                  key.offset + num_bytes,
5799                                                  key.offset, &alloc_hint);
5800                         BUG_ON(ret);
5801
5802                         for (i = 0; i < nr_extents; i++) {
5803                                 if (ext_offset >= new_extents[i].num_bytes) {
5804                                         ext_offset -= new_extents[i].num_bytes;
5805                                         continue;
5806                                 }
5807                                 extent_len = min(new_extents[i].num_bytes -
5808                                                  ext_offset, num_bytes);
5809
5810                                 ret = btrfs_insert_empty_item(trans, root,
5811                                                               path, &key,
5812                                                               sizeof(*fi));
5813                                 BUG_ON(ret);
5814
5815                                 leaf = path->nodes[0];
5816                                 fi = btrfs_item_ptr(leaf, path->slots[0],
5817                                                 struct btrfs_file_extent_item);
5818                                 btrfs_set_file_extent_generation(leaf, fi,
5819                                                         trans->transid);
5820                                 btrfs_set_file_extent_type(leaf, fi,
5821                                                         BTRFS_FILE_EXTENT_REG);
5822                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5823                                                 new_extents[i].disk_bytenr);
5824                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5825                                                 new_extents[i].disk_num_bytes);
5826                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
5827                                                 new_extents[i].ram_bytes);
5828
5829                                 btrfs_set_file_extent_compression(leaf, fi,
5830                                                 new_extents[i].compression);
5831                                 btrfs_set_file_extent_encryption(leaf, fi,
5832                                                 new_extents[i].encryption);
5833                                 btrfs_set_file_extent_other_encoding(leaf, fi,
5834                                                 new_extents[i].other_encoding);
5835
5836                                 btrfs_set_file_extent_num_bytes(leaf, fi,
5837                                                         extent_len);
5838                                 ext_offset += new_extents[i].offset;
5839                                 btrfs_set_file_extent_offset(leaf, fi,
5840                                                         ext_offset);
5841                                 btrfs_mark_buffer_dirty(leaf);
5842
5843                                 btrfs_drop_extent_cache(inode, key.offset,
5844                                                 key.offset + extent_len - 1, 0);
5845
5846                                 ret = btrfs_inc_extent_ref(trans, root,
5847                                                 new_extents[i].disk_bytenr,
5848                                                 new_extents[i].disk_num_bytes,
5849                                                 leaf->start,
5850                                                 root->root_key.objectid,
5851                                                 trans->transid, key.objectid);
5852                                 BUG_ON(ret);
5853                                 btrfs_release_path(root, path);
5854
5855                                 inode_add_bytes(inode, extent_len);
5856
5857                                 ext_offset = 0;
5858                                 num_bytes -= extent_len;
5859                                 key.offset += extent_len;
5860
5861                                 if (num_bytes == 0)
5862                                         break;
5863                         }
5864                         BUG_ON(i >= nr_extents);
5865 #endif
5866                 }
5867
5868                 if (extent_locked) {
5869                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5870                                       lock_end, GFP_NOFS);
5871                         extent_locked = 0;
5872                 }
5873 skip:
5874                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5875                     key.offset >= search_end)
5876                         break;
5877
5878                 cond_resched();
5879         }
5880         ret = 0;
5881 out:
5882         btrfs_release_path(root, path);
5883         if (inode) {
5884                 mutex_unlock(&inode->i_mutex);
5885                 if (extent_locked) {
5886                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5887                                       lock_end, GFP_NOFS);
5888                 }
5889                 iput(inode);
5890         }
5891         return ret;
5892 }
5893
5894 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5895                                struct btrfs_root *root,
5896                                struct extent_buffer *buf, u64 orig_start)
5897 {
5898         int level;
5899         int ret;
5900
5901         BUG_ON(btrfs_header_generation(buf) != trans->transid);
5902         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5903
5904         level = btrfs_header_level(buf);
5905         if (level == 0) {
5906                 struct btrfs_leaf_ref *ref;
5907                 struct btrfs_leaf_ref *orig_ref;
5908
5909                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5910                 if (!orig_ref)
5911                         return -ENOENT;
5912
5913                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5914                 if (!ref) {
5915                         btrfs_free_leaf_ref(root, orig_ref);
5916                         return -ENOMEM;
5917                 }
5918
5919                 ref->nritems = orig_ref->nritems;
5920                 memcpy(ref->extents, orig_ref->extents,
5921                         sizeof(ref->extents[0]) * ref->nritems);
5922
5923                 btrfs_free_leaf_ref(root, orig_ref);
5924
5925                 ref->root_gen = trans->transid;
5926                 ref->bytenr = buf->start;
5927                 ref->owner = btrfs_header_owner(buf);
5928                 ref->generation = btrfs_header_generation(buf);
5929
5930                 ret = btrfs_add_leaf_ref(root, ref, 0);
5931                 WARN_ON(ret);
5932                 btrfs_free_leaf_ref(root, ref);
5933         }
5934         return 0;
5935 }
5936
5937 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5938                                         struct extent_buffer *leaf,
5939                                         struct btrfs_block_group_cache *group,
5940                                         struct btrfs_root *target_root)
5941 {
5942         struct btrfs_key key;
5943         struct inode *inode = NULL;
5944         struct btrfs_file_extent_item *fi;
5945         u64 num_bytes;
5946         u64 skip_objectid = 0;
5947         u32 nritems;
5948         u32 i;
5949
5950         nritems = btrfs_header_nritems(leaf);
5951         for (i = 0; i < nritems; i++) {
5952                 btrfs_item_key_to_cpu(leaf, &key, i);
5953                 if (key.objectid == skip_objectid ||
5954                     key.type != BTRFS_EXTENT_DATA_KEY)
5955                         continue;
5956                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5957                 if (btrfs_file_extent_type(leaf, fi) ==
5958                     BTRFS_FILE_EXTENT_INLINE)
5959                         continue;
5960                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5961                         continue;
5962                 if (!inode || inode->i_ino != key.objectid) {
5963                         iput(inode);
5964                         inode = btrfs_ilookup(target_root->fs_info->sb,
5965                                               key.objectid, target_root, 1);
5966                 }
5967                 if (!inode) {
5968                         skip_objectid = key.objectid;
5969                         continue;
5970                 }
5971                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5972
5973                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5974                             key.offset + num_bytes - 1, GFP_NOFS);
5975                 btrfs_drop_extent_cache(inode, key.offset,
5976                                         key.offset + num_bytes - 1, 1);
5977                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5978                               key.offset + num_bytes - 1, GFP_NOFS);
5979                 cond_resched();
5980         }
5981         iput(inode);
5982         return 0;
5983 }
5984
5985 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5986                                         struct btrfs_root *root,
5987                                         struct extent_buffer *leaf,
5988                                         struct btrfs_block_group_cache *group,
5989                                         struct inode *reloc_inode)
5990 {
5991         struct btrfs_key key;
5992         struct btrfs_key extent_key;
5993         struct btrfs_file_extent_item *fi;
5994         struct btrfs_leaf_ref *ref;
5995         struct disk_extent *new_extent;
5996         u64 bytenr;
5997         u64 num_bytes;
5998         u32 nritems;
5999         u32 i;
6000         int ext_index;
6001         int nr_extent;
6002         int ret;
6003
6004         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6005         BUG_ON(!new_extent);
6006
6007         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6008         BUG_ON(!ref);
6009
6010         ext_index = -1;
6011         nritems = btrfs_header_nritems(leaf);
6012         for (i = 0; i < nritems; i++) {
6013                 btrfs_item_key_to_cpu(leaf, &key, i);
6014                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6015                         continue;
6016                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6017                 if (btrfs_file_extent_type(leaf, fi) ==
6018                     BTRFS_FILE_EXTENT_INLINE)
6019                         continue;
6020                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6021                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6022                 if (bytenr == 0)
6023                         continue;
6024
6025                 ext_index++;
6026                 if (bytenr >= group->key.objectid + group->key.offset ||
6027                     bytenr + num_bytes <= group->key.objectid)
6028                         continue;
6029
6030                 extent_key.objectid = bytenr;
6031                 extent_key.offset = num_bytes;
6032                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6033                 nr_extent = 1;
6034                 ret = get_new_locations(reloc_inode, &extent_key,
6035                                         group->key.objectid, 1,
6036                                         &new_extent, &nr_extent);
6037                 if (ret > 0)
6038                         continue;
6039                 BUG_ON(ret < 0);
6040
6041                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6042                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6043                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6044                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6045
6046                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6047                                                 new_extent->disk_bytenr);
6048                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6049                                                 new_extent->disk_num_bytes);
6050                 btrfs_mark_buffer_dirty(leaf);
6051
6052                 ret = btrfs_inc_extent_ref(trans, root,
6053                                         new_extent->disk_bytenr,
6054                                         new_extent->disk_num_bytes,
6055                                         leaf->start,
6056                                         root->root_key.objectid,
6057                                         trans->transid, key.objectid);
6058                 BUG_ON(ret);
6059
6060                 ret = btrfs_free_extent(trans, root,
6061                                         bytenr, num_bytes, leaf->start,
6062                                         btrfs_header_owner(leaf),
6063                                         btrfs_header_generation(leaf),
6064                                         key.objectid, 0);
6065                 BUG_ON(ret);
6066                 cond_resched();
6067         }
6068         kfree(new_extent);
6069         BUG_ON(ext_index + 1 != ref->nritems);
6070         btrfs_free_leaf_ref(root, ref);
6071         return 0;
6072 }
6073
6074 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6075                           struct btrfs_root *root)
6076 {
6077         struct btrfs_root *reloc_root;
6078         int ret;
6079
6080         if (root->reloc_root) {
6081                 reloc_root = root->reloc_root;
6082                 root->reloc_root = NULL;
6083                 list_add(&reloc_root->dead_list,
6084                          &root->fs_info->dead_reloc_roots);
6085
6086                 btrfs_set_root_bytenr(&reloc_root->root_item,
6087                                       reloc_root->node->start);
6088                 btrfs_set_root_level(&root->root_item,
6089                                      btrfs_header_level(reloc_root->node));
6090                 memset(&reloc_root->root_item.drop_progress, 0,
6091                         sizeof(struct btrfs_disk_key));
6092                 reloc_root->root_item.drop_level = 0;
6093
6094                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6095                                         &reloc_root->root_key,
6096                                         &reloc_root->root_item);
6097                 BUG_ON(ret);
6098         }
6099         return 0;
6100 }
6101
6102 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6103 {
6104         struct btrfs_trans_handle *trans;
6105         struct btrfs_root *reloc_root;
6106         struct btrfs_root *prev_root = NULL;
6107         struct list_head dead_roots;
6108         int ret;
6109         unsigned long nr;
6110
6111         INIT_LIST_HEAD(&dead_roots);
6112         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6113
6114         while (!list_empty(&dead_roots)) {
6115                 reloc_root = list_entry(dead_roots.prev,
6116                                         struct btrfs_root, dead_list);
6117                 list_del_init(&reloc_root->dead_list);
6118
6119                 BUG_ON(reloc_root->commit_root != NULL);
6120                 while (1) {
6121                         trans = btrfs_join_transaction(root, 1);
6122                         BUG_ON(!trans);
6123
6124                         mutex_lock(&root->fs_info->drop_mutex);
6125                         ret = btrfs_drop_snapshot(trans, reloc_root);
6126                         if (ret != -EAGAIN)
6127                                 break;
6128                         mutex_unlock(&root->fs_info->drop_mutex);
6129
6130                         nr = trans->blocks_used;
6131                         ret = btrfs_end_transaction(trans, root);
6132                         BUG_ON(ret);
6133                         btrfs_btree_balance_dirty(root, nr);
6134                 }
6135
6136                 free_extent_buffer(reloc_root->node);
6137
6138                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6139                                      &reloc_root->root_key);
6140                 BUG_ON(ret);
6141                 mutex_unlock(&root->fs_info->drop_mutex);
6142
6143                 nr = trans->blocks_used;
6144                 ret = btrfs_end_transaction(trans, root);
6145                 BUG_ON(ret);
6146                 btrfs_btree_balance_dirty(root, nr);
6147
6148                 kfree(prev_root);
6149                 prev_root = reloc_root;
6150         }
6151         if (prev_root) {
6152                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6153                 kfree(prev_root);
6154         }
6155         return 0;
6156 }
6157
6158 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6159 {
6160         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6161         return 0;
6162 }
6163
6164 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6165 {
6166         struct btrfs_root *reloc_root;
6167         struct btrfs_trans_handle *trans;
6168         struct btrfs_key location;
6169         int found;
6170         int ret;
6171
6172         mutex_lock(&root->fs_info->tree_reloc_mutex);
6173         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6174         BUG_ON(ret);
6175         found = !list_empty(&root->fs_info->dead_reloc_roots);
6176         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6177
6178         if (found) {
6179                 trans = btrfs_start_transaction(root, 1);
6180                 BUG_ON(!trans);
6181                 ret = btrfs_commit_transaction(trans, root);
6182                 BUG_ON(ret);
6183         }
6184
6185         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6186         location.offset = (u64)-1;
6187         location.type = BTRFS_ROOT_ITEM_KEY;
6188
6189         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6190         BUG_ON(!reloc_root);
6191         btrfs_orphan_cleanup(reloc_root);
6192         return 0;
6193 }
6194
6195 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6196                                     struct btrfs_root *root)
6197 {
6198         struct btrfs_root *reloc_root;
6199         struct extent_buffer *eb;
6200         struct btrfs_root_item *root_item;
6201         struct btrfs_key root_key;
6202         int ret;
6203
6204         BUG_ON(!root->ref_cows);
6205         if (root->reloc_root)
6206                 return 0;
6207
6208         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6209         BUG_ON(!root_item);
6210
6211         ret = btrfs_copy_root(trans, root, root->commit_root,
6212                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6213         BUG_ON(ret);
6214
6215         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6216         root_key.offset = root->root_key.objectid;
6217         root_key.type = BTRFS_ROOT_ITEM_KEY;
6218
6219         memcpy(root_item, &root->root_item, sizeof(root_item));
6220         btrfs_set_root_refs(root_item, 0);
6221         btrfs_set_root_bytenr(root_item, eb->start);
6222         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6223         btrfs_set_root_generation(root_item, trans->transid);
6224
6225         btrfs_tree_unlock(eb);
6226         free_extent_buffer(eb);
6227
6228         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6229                                 &root_key, root_item);
6230         BUG_ON(ret);
6231         kfree(root_item);
6232
6233         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6234                                                  &root_key);
6235         BUG_ON(!reloc_root);
6236         reloc_root->last_trans = trans->transid;
6237         reloc_root->commit_root = NULL;
6238         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6239
6240         root->reloc_root = reloc_root;
6241         return 0;
6242 }
6243
6244 /*
6245  * Core function of space balance.
6246  *
6247  * The idea is using reloc trees to relocate tree blocks in reference
6248  * counted roots. There is one reloc tree for each subvol, and all
6249  * reloc trees share same root key objectid. Reloc trees are snapshots
6250  * of the latest committed roots of subvols (root->commit_root).
6251  *
6252  * To relocate a tree block referenced by a subvol, there are two steps.
6253  * COW the block through subvol's reloc tree, then update block pointer
6254  * in the subvol to point to the new block. Since all reloc trees share
6255  * same root key objectid, doing special handing for tree blocks owned
6256  * by them is easy. Once a tree block has been COWed in one reloc tree,
6257  * we can use the resulting new block directly when the same block is
6258  * required to COW again through other reloc trees. By this way, relocated
6259  * tree blocks are shared between reloc trees, so they are also shared
6260  * between subvols.
6261  */
6262 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6263                                       struct btrfs_root *root,
6264                                       struct btrfs_path *path,
6265                                       struct btrfs_key *first_key,
6266                                       struct btrfs_ref_path *ref_path,
6267                                       struct btrfs_block_group_cache *group,
6268                                       struct inode *reloc_inode)
6269 {
6270         struct btrfs_root *reloc_root;
6271         struct extent_buffer *eb = NULL;
6272         struct btrfs_key *keys;
6273         u64 *nodes;
6274         int level;
6275         int shared_level;
6276         int lowest_level = 0;
6277         int ret;
6278
6279         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6280                 lowest_level = ref_path->owner_objectid;
6281
6282         if (!root->ref_cows) {
6283                 path->lowest_level = lowest_level;
6284                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6285                 BUG_ON(ret < 0);
6286                 path->lowest_level = 0;
6287                 btrfs_release_path(root, path);
6288                 return 0;
6289         }
6290
6291         mutex_lock(&root->fs_info->tree_reloc_mutex);
6292         ret = init_reloc_tree(trans, root);
6293         BUG_ON(ret);
6294         reloc_root = root->reloc_root;
6295
6296         shared_level = ref_path->shared_level;
6297         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6298
6299         keys = ref_path->node_keys;
6300         nodes = ref_path->new_nodes;
6301         memset(&keys[shared_level + 1], 0,
6302                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6303         memset(&nodes[shared_level + 1], 0,
6304                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6305
6306         if (nodes[lowest_level] == 0) {
6307                 path->lowest_level = lowest_level;
6308                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6309                                         0, 1);
6310                 BUG_ON(ret);
6311                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6312                         eb = path->nodes[level];
6313                         if (!eb || eb == reloc_root->node)
6314                                 break;
6315                         nodes[level] = eb->start;
6316                         if (level == 0)
6317                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6318                         else
6319                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6320                 }
6321                 if (nodes[0] &&
6322                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6323                         eb = path->nodes[0];
6324                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6325                                                       group, reloc_inode);
6326                         BUG_ON(ret);
6327                 }
6328                 btrfs_release_path(reloc_root, path);
6329         } else {
6330                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6331                                        lowest_level);
6332                 BUG_ON(ret);
6333         }
6334
6335         /*
6336          * replace tree blocks in the fs tree with tree blocks in
6337          * the reloc tree.
6338          */
6339         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6340         BUG_ON(ret < 0);
6341
6342         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6343                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6344                                         0, 0);
6345                 BUG_ON(ret);
6346                 extent_buffer_get(path->nodes[0]);
6347                 eb = path->nodes[0];
6348                 btrfs_release_path(reloc_root, path);
6349                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6350                 BUG_ON(ret);
6351                 free_extent_buffer(eb);
6352         }
6353
6354         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6355         path->lowest_level = 0;
6356         return 0;
6357 }
6358
6359 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6360                                         struct btrfs_root *root,
6361                                         struct btrfs_path *path,
6362                                         struct btrfs_key *first_key,
6363                                         struct btrfs_ref_path *ref_path)
6364 {
6365         int ret;
6366
6367         ret = relocate_one_path(trans, root, path, first_key,
6368                                 ref_path, NULL, NULL);
6369         BUG_ON(ret);
6370
6371         return 0;
6372 }
6373
6374 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6375                                     struct btrfs_root *extent_root,
6376                                     struct btrfs_path *path,
6377                                     struct btrfs_key *extent_key)
6378 {
6379         int ret;
6380
6381         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6382         if (ret)
6383                 goto out;
6384         ret = btrfs_del_item(trans, extent_root, path);
6385 out:
6386         btrfs_release_path(extent_root, path);
6387         return ret;
6388 }
6389
6390 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6391                                                 struct btrfs_ref_path *ref_path)
6392 {
6393         struct btrfs_key root_key;
6394
6395         root_key.objectid = ref_path->root_objectid;
6396         root_key.type = BTRFS_ROOT_ITEM_KEY;
6397         if (is_cowonly_root(ref_path->root_objectid))
6398                 root_key.offset = 0;
6399         else
6400                 root_key.offset = (u64)-1;
6401
6402         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6403 }
6404
6405 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6406                                         struct btrfs_path *path,
6407                                         struct btrfs_key *extent_key,
6408                                         struct btrfs_block_group_cache *group,
6409                                         struct inode *reloc_inode, int pass)
6410 {
6411         struct btrfs_trans_handle *trans;
6412         struct btrfs_root *found_root;
6413         struct btrfs_ref_path *ref_path = NULL;
6414         struct disk_extent *new_extents = NULL;
6415         int nr_extents = 0;
6416         int loops;
6417         int ret;
6418         int level;
6419         struct btrfs_key first_key;
6420         u64 prev_block = 0;
6421
6422
6423         trans = btrfs_start_transaction(extent_root, 1);
6424         BUG_ON(!trans);
6425
6426         if (extent_key->objectid == 0) {
6427                 ret = del_extent_zero(trans, extent_root, path, extent_key);
6428                 goto out;
6429         }
6430
6431         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6432         if (!ref_path) {
6433                 ret = -ENOMEM;
6434                 goto out;
6435         }
6436
6437         for (loops = 0; ; loops++) {
6438                 if (loops == 0) {
6439                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6440                                                    extent_key->objectid);
6441                 } else {
6442                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6443                 }
6444                 if (ret < 0)
6445                         goto out;
6446                 if (ret > 0)
6447                         break;
6448
6449                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6450                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6451                         continue;
6452
6453                 found_root = read_ref_root(extent_root->fs_info, ref_path);
6454                 BUG_ON(!found_root);
6455                 /*
6456                  * for reference counted tree, only process reference paths
6457                  * rooted at the latest committed root.
6458                  */
6459                 if (found_root->ref_cows &&
6460                     ref_path->root_generation != found_root->root_key.offset)
6461                         continue;
6462
6463                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6464                         if (pass == 0) {
6465                                 /*
6466                                  * copy data extents to new locations
6467                                  */
6468                                 u64 group_start = group->key.objectid;
6469                                 ret = relocate_data_extent(reloc_inode,
6470                                                            extent_key,
6471                                                            group_start);
6472                                 if (ret < 0)
6473                                         goto out;
6474                                 break;
6475                         }
6476                         level = 0;
6477                 } else {
6478                         level = ref_path->owner_objectid;
6479                 }
6480
6481                 if (prev_block != ref_path->nodes[level]) {
6482                         struct extent_buffer *eb;
6483                         u64 block_start = ref_path->nodes[level];
6484                         u64 block_size = btrfs_level_size(found_root, level);
6485
6486                         eb = read_tree_block(found_root, block_start,
6487                                              block_size, 0);
6488                         btrfs_tree_lock(eb);
6489                         BUG_ON(level != btrfs_header_level(eb));
6490
6491                         if (level == 0)
6492                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
6493                         else
6494                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
6495
6496                         btrfs_tree_unlock(eb);
6497                         free_extent_buffer(eb);
6498                         prev_block = block_start;
6499                 }
6500
6501                 mutex_lock(&extent_root->fs_info->trans_mutex);
6502                 btrfs_record_root_in_trans(found_root);
6503                 mutex_unlock(&extent_root->fs_info->trans_mutex);
6504                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6505                         /*
6506                          * try to update data extent references while
6507                          * keeping metadata shared between snapshots.
6508                          */
6509                         if (pass == 1) {
6510                                 ret = relocate_one_path(trans, found_root,
6511                                                 path, &first_key, ref_path,
6512                                                 group, reloc_inode);
6513                                 if (ret < 0)
6514                                         goto out;
6515                                 continue;
6516                         }
6517                         /*
6518                          * use fallback method to process the remaining
6519                          * references.
6520                          */
6521                         if (!new_extents) {
6522                                 u64 group_start = group->key.objectid;
6523                                 new_extents = kmalloc(sizeof(*new_extents),
6524                                                       GFP_NOFS);
6525                                 nr_extents = 1;
6526                                 ret = get_new_locations(reloc_inode,
6527                                                         extent_key,
6528                                                         group_start, 1,
6529                                                         &new_extents,
6530                                                         &nr_extents);
6531                                 if (ret)
6532                                         goto out;
6533                         }
6534                         ret = replace_one_extent(trans, found_root,
6535                                                 path, extent_key,
6536                                                 &first_key, ref_path,
6537                                                 new_extents, nr_extents);
6538                 } else {
6539                         ret = relocate_tree_block(trans, found_root, path,
6540                                                   &first_key, ref_path);
6541                 }
6542                 if (ret < 0)
6543                         goto out;
6544         }
6545         ret = 0;
6546 out:
6547         btrfs_end_transaction(trans, extent_root);
6548         kfree(new_extents);
6549         kfree(ref_path);
6550         return ret;
6551 }
6552 #endif
6553
6554 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6555 {
6556         u64 num_devices;
6557         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6558                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6559
6560         num_devices = root->fs_info->fs_devices->rw_devices;
6561         if (num_devices == 1) {
6562                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6563                 stripped = flags & ~stripped;
6564
6565                 /* turn raid0 into single device chunks */
6566                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6567                         return stripped;
6568
6569                 /* turn mirroring into duplication */
6570                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6571                              BTRFS_BLOCK_GROUP_RAID10))
6572                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6573                 return flags;
6574         } else {
6575                 /* they already had raid on here, just return */
6576                 if (flags & stripped)
6577                         return flags;
6578
6579                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6580                 stripped = flags & ~stripped;
6581
6582                 /* switch duplicated blocks with raid1 */
6583                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6584                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6585
6586                 /* turn single device chunks into raid0 */
6587                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6588         }
6589         return flags;
6590 }
6591
6592 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6593                      struct btrfs_block_group_cache *shrink_block_group,
6594                      int force)
6595 {
6596         struct btrfs_trans_handle *trans;
6597         u64 new_alloc_flags;
6598         u64 calc;
6599
6600         spin_lock(&shrink_block_group->lock);
6601         if (btrfs_block_group_used(&shrink_block_group->item) +
6602             shrink_block_group->reserved > 0) {
6603                 spin_unlock(&shrink_block_group->lock);
6604
6605                 trans = btrfs_start_transaction(root, 1);
6606                 spin_lock(&shrink_block_group->lock);
6607
6608                 new_alloc_flags = update_block_group_flags(root,
6609                                                    shrink_block_group->flags);
6610                 if (new_alloc_flags != shrink_block_group->flags) {
6611                         calc =
6612                              btrfs_block_group_used(&shrink_block_group->item);
6613                 } else {
6614                         calc = shrink_block_group->key.offset;
6615                 }
6616                 spin_unlock(&shrink_block_group->lock);
6617
6618                 do_chunk_alloc(trans, root->fs_info->extent_root,
6619                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
6620
6621                 btrfs_end_transaction(trans, root);
6622         } else
6623                 spin_unlock(&shrink_block_group->lock);
6624         return 0;
6625 }
6626
6627
6628 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6629                                          struct btrfs_block_group_cache *group)
6630
6631 {
6632         __alloc_chunk_for_shrink(root, group, 1);
6633         set_block_group_readonly(group);
6634         return 0;
6635 }
6636
6637 #if 0
6638 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6639                                  struct btrfs_root *root,
6640                                  u64 objectid, u64 size)
6641 {
6642         struct btrfs_path *path;
6643         struct btrfs_inode_item *item;
6644         struct extent_buffer *leaf;
6645         int ret;
6646
6647         path = btrfs_alloc_path();
6648         if (!path)
6649                 return -ENOMEM;
6650
6651         path->leave_spinning = 1;
6652         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6653         if (ret)
6654                 goto out;
6655
6656         leaf = path->nodes[0];
6657         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6658         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6659         btrfs_set_inode_generation(leaf, item, 1);
6660         btrfs_set_inode_size(leaf, item, size);
6661         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6662         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6663         btrfs_mark_buffer_dirty(leaf);
6664         btrfs_release_path(root, path);
6665 out:
6666         btrfs_free_path(path);
6667         return ret;
6668 }
6669
6670 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6671                                         struct btrfs_block_group_cache *group)
6672 {
6673         struct inode *inode = NULL;
6674         struct btrfs_trans_handle *trans;
6675         struct btrfs_root *root;
6676         struct btrfs_key root_key;
6677         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6678         int err = 0;
6679
6680         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6681         root_key.type = BTRFS_ROOT_ITEM_KEY;
6682         root_key.offset = (u64)-1;
6683         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6684         if (IS_ERR(root))
6685                 return ERR_CAST(root);
6686
6687         trans = btrfs_start_transaction(root, 1);
6688         BUG_ON(!trans);
6689
6690         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6691         if (err)
6692                 goto out;
6693
6694         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6695         BUG_ON(err);
6696
6697         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6698                                        group->key.offset, 0, group->key.offset,
6699                                        0, 0, 0);
6700         BUG_ON(err);
6701
6702         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6703         if (inode->i_state & I_NEW) {
6704                 BTRFS_I(inode)->root = root;
6705                 BTRFS_I(inode)->location.objectid = objectid;
6706                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6707                 BTRFS_I(inode)->location.offset = 0;
6708                 btrfs_read_locked_inode(inode);
6709                 unlock_new_inode(inode);
6710                 BUG_ON(is_bad_inode(inode));
6711         } else {
6712                 BUG_ON(1);
6713         }
6714         BTRFS_I(inode)->index_cnt = group->key.objectid;
6715
6716         err = btrfs_orphan_add(trans, inode);
6717 out:
6718         btrfs_end_transaction(trans, root);
6719         if (err) {
6720                 if (inode)
6721                         iput(inode);
6722                 inode = ERR_PTR(err);
6723         }
6724         return inode;
6725 }
6726
6727 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6728 {
6729
6730         struct btrfs_ordered_sum *sums;
6731         struct btrfs_sector_sum *sector_sum;
6732         struct btrfs_ordered_extent *ordered;
6733         struct btrfs_root *root = BTRFS_I(inode)->root;
6734         struct list_head list;
6735         size_t offset;
6736         int ret;
6737         u64 disk_bytenr;
6738
6739         INIT_LIST_HEAD(&list);
6740
6741         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6742         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6743
6744         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6745         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6746                                        disk_bytenr + len - 1, &list);
6747
6748         while (!list_empty(&list)) {
6749                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6750                 list_del_init(&sums->list);
6751
6752                 sector_sum = sums->sums;
6753                 sums->bytenr = ordered->start;
6754
6755                 offset = 0;
6756                 while (offset < sums->len) {
6757                         sector_sum->bytenr += ordered->start - disk_bytenr;
6758                         sector_sum++;
6759                         offset += root->sectorsize;
6760                 }
6761
6762                 btrfs_add_ordered_sum(inode, ordered, sums);
6763         }
6764         btrfs_put_ordered_extent(ordered);
6765         return 0;
6766 }
6767
6768 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6769 {
6770         struct btrfs_trans_handle *trans;
6771         struct btrfs_path *path;
6772         struct btrfs_fs_info *info = root->fs_info;
6773         struct extent_buffer *leaf;
6774         struct inode *reloc_inode;
6775         struct btrfs_block_group_cache *block_group;
6776         struct btrfs_key key;
6777         u64 skipped;
6778         u64 cur_byte;
6779         u64 total_found;
6780         u32 nritems;
6781         int ret;
6782         int progress;
6783         int pass = 0;
6784
6785         root = root->fs_info->extent_root;
6786
6787         block_group = btrfs_lookup_block_group(info, group_start);
6788         BUG_ON(!block_group);
6789
6790         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6791                (unsigned long long)block_group->key.objectid,
6792                (unsigned long long)block_group->flags);
6793
6794         path = btrfs_alloc_path();
6795         BUG_ON(!path);
6796
6797         reloc_inode = create_reloc_inode(info, block_group);
6798         BUG_ON(IS_ERR(reloc_inode));
6799
6800         __alloc_chunk_for_shrink(root, block_group, 1);
6801         set_block_group_readonly(block_group);
6802
6803         btrfs_start_delalloc_inodes(info->tree_root);
6804         btrfs_wait_ordered_extents(info->tree_root, 0);
6805 again:
6806         skipped = 0;
6807         total_found = 0;
6808         progress = 0;
6809         key.objectid = block_group->key.objectid;
6810         key.offset = 0;
6811         key.type = 0;
6812         cur_byte = key.objectid;
6813
6814         trans = btrfs_start_transaction(info->tree_root, 1);
6815         btrfs_commit_transaction(trans, info->tree_root);
6816
6817         mutex_lock(&root->fs_info->cleaner_mutex);
6818         btrfs_clean_old_snapshots(info->tree_root);
6819         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6820         mutex_unlock(&root->fs_info->cleaner_mutex);
6821
6822         trans = btrfs_start_transaction(info->tree_root, 1);
6823         btrfs_commit_transaction(trans, info->tree_root);
6824
6825         while (1) {
6826                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6827                 if (ret < 0)
6828                         goto out;
6829 next:
6830                 leaf = path->nodes[0];
6831                 nritems = btrfs_header_nritems(leaf);
6832                 if (path->slots[0] >= nritems) {
6833                         ret = btrfs_next_leaf(root, path);
6834                         if (ret < 0)
6835                                 goto out;
6836                         if (ret == 1) {
6837                                 ret = 0;
6838                                 break;
6839                         }
6840                         leaf = path->nodes[0];
6841                         nritems = btrfs_header_nritems(leaf);
6842                 }
6843
6844                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6845
6846                 if (key.objectid >= block_group->key.objectid +
6847                     block_group->key.offset)
6848                         break;
6849
6850                 if (progress && need_resched()) {
6851                         btrfs_release_path(root, path);
6852                         cond_resched();
6853                         progress = 0;
6854                         continue;
6855                 }
6856                 progress = 1;
6857
6858                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
6859                     key.objectid + key.offset <= cur_byte) {
6860                         path->slots[0]++;
6861                         goto next;
6862                 }
6863
6864                 total_found++;
6865                 cur_byte = key.objectid + key.offset;
6866                 btrfs_release_path(root, path);
6867
6868                 __alloc_chunk_for_shrink(root, block_group, 0);
6869                 ret = relocate_one_extent(root, path, &key, block_group,
6870                                           reloc_inode, pass);
6871                 BUG_ON(ret < 0);
6872                 if (ret > 0)
6873                         skipped++;
6874
6875                 key.objectid = cur_byte;
6876                 key.type = 0;
6877                 key.offset = 0;
6878         }
6879
6880         btrfs_release_path(root, path);
6881
6882         if (pass == 0) {
6883                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6884                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6885         }
6886
6887         if (total_found > 0) {
6888                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6889                        (unsigned long long)total_found, pass);
6890                 pass++;
6891                 if (total_found == skipped && pass > 2) {
6892                         iput(reloc_inode);
6893                         reloc_inode = create_reloc_inode(info, block_group);
6894                         pass = 0;
6895                 }
6896                 goto again;
6897         }
6898
6899         /* delete reloc_inode */
6900         iput(reloc_inode);
6901
6902         /* unpin extents in this range */
6903         trans = btrfs_start_transaction(info->tree_root, 1);
6904         btrfs_commit_transaction(trans, info->tree_root);
6905
6906         spin_lock(&block_group->lock);
6907         WARN_ON(block_group->pinned > 0);
6908         WARN_ON(block_group->reserved > 0);
6909         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6910         spin_unlock(&block_group->lock);
6911         btrfs_put_block_group(block_group);
6912         ret = 0;
6913 out:
6914         btrfs_free_path(path);
6915         return ret;
6916 }
6917 #endif
6918
6919 static int find_first_block_group(struct btrfs_root *root,
6920                 struct btrfs_path *path, struct btrfs_key *key)
6921 {
6922         int ret = 0;
6923         struct btrfs_key found_key;
6924         struct extent_buffer *leaf;
6925         int slot;
6926
6927         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6928         if (ret < 0)
6929                 goto out;
6930
6931         while (1) {
6932                 slot = path->slots[0];
6933                 leaf = path->nodes[0];
6934                 if (slot >= btrfs_header_nritems(leaf)) {
6935                         ret = btrfs_next_leaf(root, path);
6936                         if (ret == 0)
6937                                 continue;
6938                         if (ret < 0)
6939                                 goto out;
6940                         break;
6941                 }
6942                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6943
6944                 if (found_key.objectid >= key->objectid &&
6945                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6946                         ret = 0;
6947                         goto out;
6948                 }
6949                 path->slots[0]++;
6950         }
6951         ret = -ENOENT;
6952 out:
6953         return ret;
6954 }
6955
6956 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6957 {
6958         struct btrfs_block_group_cache *block_group;
6959         struct btrfs_space_info *space_info;
6960         struct rb_node *n;
6961
6962         spin_lock(&info->block_group_cache_lock);
6963         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6964                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6965                                        cache_node);
6966                 rb_erase(&block_group->cache_node,
6967                          &info->block_group_cache_tree);
6968                 spin_unlock(&info->block_group_cache_lock);
6969
6970                 btrfs_remove_free_space_cache(block_group);
6971                 down_write(&block_group->space_info->groups_sem);
6972                 list_del(&block_group->list);
6973                 up_write(&block_group->space_info->groups_sem);
6974
6975                 WARN_ON(atomic_read(&block_group->count) != 1);
6976                 kfree(block_group);
6977
6978                 spin_lock(&info->block_group_cache_lock);
6979         }
6980         spin_unlock(&info->block_group_cache_lock);
6981
6982         /* now that all the block groups are freed, go through and
6983          * free all the space_info structs.  This is only called during
6984          * the final stages of unmount, and so we know nobody is
6985          * using them.  We call synchronize_rcu() once before we start,
6986          * just to be on the safe side.
6987          */
6988         synchronize_rcu();
6989
6990         while(!list_empty(&info->space_info)) {
6991                 space_info = list_entry(info->space_info.next,
6992                                         struct btrfs_space_info,
6993                                         list);
6994
6995                 list_del(&space_info->list);
6996                 kfree(space_info);
6997         }
6998         return 0;
6999 }
7000
7001 int btrfs_read_block_groups(struct btrfs_root *root)
7002 {
7003         struct btrfs_path *path;
7004         int ret;
7005         struct btrfs_block_group_cache *cache;
7006         struct btrfs_fs_info *info = root->fs_info;
7007         struct btrfs_space_info *space_info;
7008         struct btrfs_key key;
7009         struct btrfs_key found_key;
7010         struct extent_buffer *leaf;
7011
7012         root = info->extent_root;
7013         key.objectid = 0;
7014         key.offset = 0;
7015         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7016         path = btrfs_alloc_path();
7017         if (!path)
7018                 return -ENOMEM;
7019
7020         while (1) {
7021                 ret = find_first_block_group(root, path, &key);
7022                 if (ret > 0) {
7023                         ret = 0;
7024                         goto error;
7025                 }
7026                 if (ret != 0)
7027                         goto error;
7028
7029                 leaf = path->nodes[0];
7030                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7031                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7032                 if (!cache) {
7033                         ret = -ENOMEM;
7034                         break;
7035                 }
7036
7037                 atomic_set(&cache->count, 1);
7038                 spin_lock_init(&cache->lock);
7039                 spin_lock_init(&cache->tree_lock);
7040                 mutex_init(&cache->cache_mutex);
7041                 INIT_LIST_HEAD(&cache->list);
7042                 INIT_LIST_HEAD(&cache->cluster_list);
7043                 read_extent_buffer(leaf, &cache->item,
7044                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7045                                    sizeof(cache->item));
7046                 memcpy(&cache->key, &found_key, sizeof(found_key));
7047
7048                 key.objectid = found_key.objectid + found_key.offset;
7049                 btrfs_release_path(root, path);
7050                 cache->flags = btrfs_block_group_flags(&cache->item);
7051
7052                 ret = update_space_info(info, cache->flags, found_key.offset,
7053                                         btrfs_block_group_used(&cache->item),
7054                                         &space_info);
7055                 BUG_ON(ret);
7056                 cache->space_info = space_info;
7057                 down_write(&space_info->groups_sem);
7058                 list_add_tail(&cache->list, &space_info->block_groups);
7059                 up_write(&space_info->groups_sem);
7060
7061                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7062                 BUG_ON(ret);
7063
7064                 set_avail_alloc_bits(root->fs_info, cache->flags);
7065                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7066                         set_block_group_readonly(cache);
7067         }
7068         ret = 0;
7069 error:
7070         btrfs_free_path(path);
7071         return ret;
7072 }
7073
7074 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7075                            struct btrfs_root *root, u64 bytes_used,
7076                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7077                            u64 size)
7078 {
7079         int ret;
7080         struct btrfs_root *extent_root;
7081         struct btrfs_block_group_cache *cache;
7082
7083         extent_root = root->fs_info->extent_root;
7084
7085         root->fs_info->last_trans_log_full_commit = trans->transid;
7086
7087         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7088         if (!cache)
7089                 return -ENOMEM;
7090
7091         cache->key.objectid = chunk_offset;
7092         cache->key.offset = size;
7093         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7094         atomic_set(&cache->count, 1);
7095         spin_lock_init(&cache->lock);
7096         spin_lock_init(&cache->tree_lock);
7097         mutex_init(&cache->cache_mutex);
7098         INIT_LIST_HEAD(&cache->list);
7099         INIT_LIST_HEAD(&cache->cluster_list);
7100
7101         btrfs_set_block_group_used(&cache->item, bytes_used);
7102         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7103         cache->flags = type;
7104         btrfs_set_block_group_flags(&cache->item, type);
7105
7106         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7107                                 &cache->space_info);
7108         BUG_ON(ret);
7109         down_write(&cache->space_info->groups_sem);
7110         list_add_tail(&cache->list, &cache->space_info->block_groups);
7111         up_write(&cache->space_info->groups_sem);
7112
7113         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7114         BUG_ON(ret);
7115
7116         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7117                                 sizeof(cache->item));
7118         BUG_ON(ret);
7119
7120         set_avail_alloc_bits(extent_root->fs_info, type);
7121
7122         return 0;
7123 }
7124
7125 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7126                              struct btrfs_root *root, u64 group_start)
7127 {
7128         struct btrfs_path *path;
7129         struct btrfs_block_group_cache *block_group;
7130         struct btrfs_free_cluster *cluster;
7131         struct btrfs_key key;
7132         int ret;
7133
7134         root = root->fs_info->extent_root;
7135
7136         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7137         BUG_ON(!block_group);
7138         BUG_ON(!block_group->ro);
7139
7140         memcpy(&key, &block_group->key, sizeof(key));
7141
7142         /* make sure this block group isn't part of an allocation cluster */
7143         cluster = &root->fs_info->data_alloc_cluster;
7144         spin_lock(&cluster->refill_lock);
7145         btrfs_return_cluster_to_free_space(block_group, cluster);
7146         spin_unlock(&cluster->refill_lock);
7147
7148         /*
7149          * make sure this block group isn't part of a metadata
7150          * allocation cluster
7151          */
7152         cluster = &root->fs_info->meta_alloc_cluster;
7153         spin_lock(&cluster->refill_lock);
7154         btrfs_return_cluster_to_free_space(block_group, cluster);
7155         spin_unlock(&cluster->refill_lock);
7156
7157         path = btrfs_alloc_path();
7158         BUG_ON(!path);
7159
7160         spin_lock(&root->fs_info->block_group_cache_lock);
7161         rb_erase(&block_group->cache_node,
7162                  &root->fs_info->block_group_cache_tree);
7163         spin_unlock(&root->fs_info->block_group_cache_lock);
7164         btrfs_remove_free_space_cache(block_group);
7165         down_write(&block_group->space_info->groups_sem);
7166         /*
7167          * we must use list_del_init so people can check to see if they
7168          * are still on the list after taking the semaphore
7169          */
7170         list_del_init(&block_group->list);
7171         up_write(&block_group->space_info->groups_sem);
7172
7173         spin_lock(&block_group->space_info->lock);
7174         block_group->space_info->total_bytes -= block_group->key.offset;
7175         block_group->space_info->bytes_readonly -= block_group->key.offset;
7176         spin_unlock(&block_group->space_info->lock);
7177         block_group->space_info->full = 0;
7178
7179         btrfs_put_block_group(block_group);
7180         btrfs_put_block_group(block_group);
7181
7182         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7183         if (ret > 0)
7184                 ret = -EIO;
7185         if (ret < 0)
7186                 goto out;
7187
7188         ret = btrfs_del_item(trans, root, path);
7189 out:
7190         btrfs_free_path(path);
7191         return ret;
7192 }