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                 /* move on to the next group */
3653                 if (search_start + num_bytes >= search_end) {
3654                         btrfs_add_free_space(block_group, offset, num_bytes);
3655                         goto loop;
3656                 }
3657
3658                 /* move on to the next group */
3659                 if (search_start + num_bytes >
3660                     block_group->key.objectid + block_group->key.offset) {
3661                         btrfs_add_free_space(block_group, offset, num_bytes);
3662                         goto loop;
3663                 }
3664
3665                 if (exclude_nr > 0 &&
3666                     (search_start + num_bytes > exclude_start &&
3667                      search_start < exclude_start + exclude_nr)) {
3668                         search_start = exclude_start + exclude_nr;
3669
3670                         btrfs_add_free_space(block_group, offset, num_bytes);
3671                         /*
3672                          * if search_start is still in this block group
3673                          * then we just re-search this block group
3674                          */
3675                         if (search_start >= block_group->key.objectid &&
3676                             search_start < (block_group->key.objectid +
3677                                             block_group->key.offset))
3678                                 goto have_block_group;
3679                         goto loop;
3680                 }
3681
3682                 ins->objectid = search_start;
3683                 ins->offset = num_bytes;
3684
3685                 if (offset < search_start)
3686                         btrfs_add_free_space(block_group, offset,
3687                                              search_start - offset);
3688                 BUG_ON(offset > search_start);
3689
3690                 /* we are all good, lets return */
3691                 break;
3692 loop:
3693                 btrfs_put_block_group(block_group);
3694         }
3695         up_read(&space_info->groups_sem);
3696
3697         /* loop == 0, try to find a clustered alloc in every block group
3698          * loop == 1, try again after forcing a chunk allocation
3699          * loop == 2, set empty_size and empty_cluster to 0 and try again
3700          */
3701         if (!ins->objectid && loop < 3 &&
3702             (empty_size || empty_cluster || allowed_chunk_alloc)) {
3703                 if (loop >= 2) {
3704                         empty_size = 0;
3705                         empty_cluster = 0;
3706                 }
3707
3708                 if (allowed_chunk_alloc) {
3709                         ret = do_chunk_alloc(trans, root, num_bytes +
3710                                              2 * 1024 * 1024, data, 1);
3711                         allowed_chunk_alloc = 0;
3712                 } else {
3713                         space_info->force_alloc = 1;
3714                 }
3715
3716                 if (loop < 3) {
3717                         loop++;
3718                         goto search;
3719                 }
3720                 ret = -ENOSPC;
3721         } else if (!ins->objectid) {
3722                 ret = -ENOSPC;
3723         }
3724
3725         /* we found what we needed */
3726         if (ins->objectid) {
3727                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3728                         trans->block_group = block_group->key.objectid;
3729
3730                 btrfs_put_block_group(block_group);
3731                 ret = 0;
3732         }
3733
3734         return ret;
3735 }
3736
3737 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3738 {
3739         struct btrfs_block_group_cache *cache;
3740
3741         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3742                (unsigned long long)(info->total_bytes - info->bytes_used -
3743                                     info->bytes_pinned - info->bytes_reserved),
3744                (info->full) ? "" : "not ");
3745         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3746                " may_use=%llu, used=%llu\n",
3747                (unsigned long long)info->total_bytes,
3748                (unsigned long long)info->bytes_pinned,
3749                (unsigned long long)info->bytes_delalloc,
3750                (unsigned long long)info->bytes_may_use,
3751                (unsigned long long)info->bytes_used);
3752
3753         down_read(&info->groups_sem);
3754         list_for_each_entry(cache, &info->block_groups, list) {
3755                 spin_lock(&cache->lock);
3756                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3757                        "%llu pinned %llu reserved\n",
3758                        (unsigned long long)cache->key.objectid,
3759                        (unsigned long long)cache->key.offset,
3760                        (unsigned long long)btrfs_block_group_used(&cache->item),
3761                        (unsigned long long)cache->pinned,
3762                        (unsigned long long)cache->reserved);
3763                 btrfs_dump_free_space(cache, bytes);
3764                 spin_unlock(&cache->lock);
3765         }
3766         up_read(&info->groups_sem);
3767 }
3768
3769 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3770                                   struct btrfs_root *root,
3771                                   u64 num_bytes, u64 min_alloc_size,
3772                                   u64 empty_size, u64 hint_byte,
3773                                   u64 search_end, struct btrfs_key *ins,
3774                                   u64 data)
3775 {
3776         int ret;
3777         u64 search_start = 0;
3778         struct btrfs_fs_info *info = root->fs_info;
3779
3780         data = btrfs_get_alloc_profile(root, data);
3781 again:
3782         /*
3783          * the only place that sets empty_size is btrfs_realloc_node, which
3784          * is not called recursively on allocations
3785          */
3786         if (empty_size || root->ref_cows) {
3787                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3788                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3789                                      2 * 1024 * 1024,
3790                                      BTRFS_BLOCK_GROUP_METADATA |
3791                                      (info->metadata_alloc_profile &
3792                                       info->avail_metadata_alloc_bits), 0);
3793                 }
3794                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3795                                      num_bytes + 2 * 1024 * 1024, data, 0);
3796         }
3797
3798         WARN_ON(num_bytes < root->sectorsize);
3799         ret = find_free_extent(trans, root, num_bytes, empty_size,
3800                                search_start, search_end, hint_byte, ins,
3801                                trans->alloc_exclude_start,
3802                                trans->alloc_exclude_nr, data);
3803
3804         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3805                 num_bytes = num_bytes >> 1;
3806                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3807                 num_bytes = max(num_bytes, min_alloc_size);
3808                 do_chunk_alloc(trans, root->fs_info->extent_root,
3809                                num_bytes, data, 1);
3810                 goto again;
3811         }
3812         if (ret) {
3813                 struct btrfs_space_info *sinfo;
3814
3815                 sinfo = __find_space_info(root->fs_info, data);
3816                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3817                        "wanted %llu\n", (unsigned long long)data,
3818                        (unsigned long long)num_bytes);
3819                 dump_space_info(sinfo, num_bytes);
3820                 BUG();
3821         }
3822
3823         return ret;
3824 }
3825
3826 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3827 {
3828         struct btrfs_block_group_cache *cache;
3829         int ret = 0;
3830
3831         cache = btrfs_lookup_block_group(root->fs_info, start);
3832         if (!cache) {
3833                 printk(KERN_ERR "Unable to find block group for %llu\n",
3834                        (unsigned long long)start);
3835                 return -ENOSPC;
3836         }
3837
3838         ret = btrfs_discard_extent(root, start, len);
3839
3840         btrfs_add_free_space(cache, start, len);
3841         btrfs_put_block_group(cache);
3842         update_reserved_extents(root, start, len, 0);
3843
3844         return ret;
3845 }
3846
3847 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3848                                   struct btrfs_root *root,
3849                                   u64 num_bytes, u64 min_alloc_size,
3850                                   u64 empty_size, u64 hint_byte,
3851                                   u64 search_end, struct btrfs_key *ins,
3852                                   u64 data)
3853 {
3854         int ret;
3855         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3856                                      empty_size, hint_byte, search_end, ins,
3857                                      data);
3858         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3859         return ret;
3860 }
3861
3862 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
3863                                       struct btrfs_root *root,
3864                                       u64 parent, u64 root_objectid,
3865                                       u64 flags, u64 owner, u64 offset,
3866                                       struct btrfs_key *ins, int ref_mod)
3867 {
3868         int ret;
3869         struct btrfs_fs_info *fs_info = root->fs_info;
3870         struct btrfs_extent_item *extent_item;
3871         struct btrfs_extent_inline_ref *iref;
3872         struct btrfs_path *path;
3873         struct extent_buffer *leaf;
3874         int type;
3875         u32 size;
3876
3877         if (parent > 0)
3878                 type = BTRFS_SHARED_DATA_REF_KEY;
3879         else
3880                 type = BTRFS_EXTENT_DATA_REF_KEY;
3881
3882         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
3883
3884         path = btrfs_alloc_path();
3885         BUG_ON(!path);
3886
3887         path->leave_spinning = 1;
3888         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
3889                                       ins, size);
3890         BUG_ON(ret);
3891
3892         leaf = path->nodes[0];
3893         extent_item = btrfs_item_ptr(leaf, path->slots[0],
3894                                      struct btrfs_extent_item);
3895         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
3896         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
3897         btrfs_set_extent_flags(leaf, extent_item,
3898                                flags | BTRFS_EXTENT_FLAG_DATA);
3899
3900         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
3901         btrfs_set_extent_inline_ref_type(leaf, iref, type);
3902         if (parent > 0) {
3903                 struct btrfs_shared_data_ref *ref;
3904                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
3905                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
3906                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
3907         } else {
3908                 struct btrfs_extent_data_ref *ref;
3909                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3910                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
3911                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
3912                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
3913                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
3914         }
3915
3916         btrfs_mark_buffer_dirty(path->nodes[0]);
3917         btrfs_free_path(path);
3918
3919         ret = update_block_group(trans, root, ins->objectid, ins->offset,
3920                                  1, 0);
3921         if (ret) {
3922                 printk(KERN_ERR "btrfs update block group failed for %llu "
3923                        "%llu\n", (unsigned long long)ins->objectid,
3924                        (unsigned long long)ins->offset);
3925                 BUG();
3926         }
3927         return ret;
3928 }
3929
3930 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
3931                                      struct btrfs_root *root,
3932                                      u64 parent, u64 root_objectid,
3933                                      u64 flags, struct btrfs_disk_key *key,
3934                                      int level, struct btrfs_key *ins)
3935 {
3936         int ret;
3937         struct btrfs_fs_info *fs_info = root->fs_info;
3938         struct btrfs_extent_item *extent_item;
3939         struct btrfs_tree_block_info *block_info;
3940         struct btrfs_extent_inline_ref *iref;
3941         struct btrfs_path *path;
3942         struct extent_buffer *leaf;
3943         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
3944
3945         path = btrfs_alloc_path();
3946         BUG_ON(!path);
3947
3948         path->leave_spinning = 1;
3949         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
3950                                       ins, size);
3951         BUG_ON(ret);
3952
3953         leaf = path->nodes[0];
3954         extent_item = btrfs_item_ptr(leaf, path->slots[0],
3955                                      struct btrfs_extent_item);
3956         btrfs_set_extent_refs(leaf, extent_item, 1);
3957         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
3958         btrfs_set_extent_flags(leaf, extent_item,
3959                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
3960         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
3961
3962         btrfs_set_tree_block_key(leaf, block_info, key);
3963         btrfs_set_tree_block_level(leaf, block_info, level);
3964
3965         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
3966         if (parent > 0) {
3967                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
3968                 btrfs_set_extent_inline_ref_type(leaf, iref,
3969                                                  BTRFS_SHARED_BLOCK_REF_KEY);
3970                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
3971         } else {
3972                 btrfs_set_extent_inline_ref_type(leaf, iref,
3973                                                  BTRFS_TREE_BLOCK_REF_KEY);
3974                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
3975         }
3976
3977         btrfs_mark_buffer_dirty(leaf);
3978         btrfs_free_path(path);
3979
3980         ret = update_block_group(trans, root, ins->objectid, ins->offset,
3981                                  1, 0);
3982         if (ret) {
3983                 printk(KERN_ERR "btrfs update block group failed for %llu "
3984                        "%llu\n", (unsigned long long)ins->objectid,
3985                        (unsigned long long)ins->offset);
3986                 BUG();
3987         }
3988         return ret;
3989 }
3990
3991 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
3992                                      struct btrfs_root *root,
3993                                      u64 root_objectid, u64 owner,
3994                                      u64 offset, struct btrfs_key *ins)
3995 {
3996         int ret;
3997
3998         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
3999
4000         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4001                                          0, root_objectid, owner, offset,
4002                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4003         return ret;
4004 }
4005
4006 /*
4007  * this is used by the tree logging recovery code.  It records that
4008  * an extent has been allocated and makes sure to clear the free
4009  * space cache bits as well
4010  */
4011 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4012                                    struct btrfs_root *root,
4013                                    u64 root_objectid, u64 owner, u64 offset,
4014                                    struct btrfs_key *ins)
4015 {
4016         int ret;
4017         struct btrfs_block_group_cache *block_group;
4018
4019         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4020         mutex_lock(&block_group->cache_mutex);
4021         cache_block_group(root, block_group);
4022         mutex_unlock(&block_group->cache_mutex);
4023
4024         ret = btrfs_remove_free_space(block_group, ins->objectid,
4025                                       ins->offset);
4026         BUG_ON(ret);
4027         btrfs_put_block_group(block_group);
4028         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4029                                          0, owner, offset, ins, 1);
4030         return ret;
4031 }
4032
4033 /*
4034  * finds a free extent and does all the dirty work required for allocation
4035  * returns the key for the extent through ins, and a tree buffer for
4036  * the first block of the extent through buf.
4037  *
4038  * returns 0 if everything worked, non-zero otherwise.
4039  */
4040 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4041                             struct btrfs_root *root,
4042                             u64 num_bytes, u64 parent, u64 root_objectid,
4043                             struct btrfs_disk_key *key, int level,
4044                             u64 empty_size, u64 hint_byte, u64 search_end,
4045                             struct btrfs_key *ins)
4046 {
4047         int ret;
4048         u64 flags = 0;
4049
4050         ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4051                                      empty_size, hint_byte, search_end,
4052                                      ins, 0);
4053         BUG_ON(ret);
4054
4055         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4056                 if (parent == 0)
4057                         parent = ins->objectid;
4058                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4059         } else
4060                 BUG_ON(parent > 0);
4061
4062         update_reserved_extents(root, ins->objectid, ins->offset, 1);
4063         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4064                 struct btrfs_delayed_extent_op *extent_op;
4065                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4066                 BUG_ON(!extent_op);
4067                 if (key)
4068                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4069                 else
4070                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4071                 extent_op->flags_to_set = flags;
4072                 extent_op->update_key = 1;
4073                 extent_op->update_flags = 1;
4074                 extent_op->is_data = 0;
4075
4076                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4077                                         ins->offset, parent, root_objectid,
4078                                         level, BTRFS_ADD_DELAYED_EXTENT,
4079                                         extent_op);
4080                 BUG_ON(ret);
4081         }
4082         return ret;
4083 }
4084
4085 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4086                                             struct btrfs_root *root,
4087                                             u64 bytenr, u32 blocksize,
4088                                             int level)
4089 {
4090         struct extent_buffer *buf;
4091
4092         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4093         if (!buf)
4094                 return ERR_PTR(-ENOMEM);
4095         btrfs_set_header_generation(buf, trans->transid);
4096         btrfs_set_buffer_lockdep_class(buf, level);
4097         btrfs_tree_lock(buf);
4098         clean_tree_block(trans, root, buf);
4099
4100         btrfs_set_lock_blocking(buf);
4101         btrfs_set_buffer_uptodate(buf);
4102
4103         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4104                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4105                          buf->start + buf->len - 1, GFP_NOFS);
4106         } else {
4107                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4108                          buf->start + buf->len - 1, GFP_NOFS);
4109         }
4110         trans->blocks_used++;
4111         /* this returns a buffer locked for blocking */
4112         return buf;
4113 }
4114
4115 /*
4116  * helper function to allocate a block for a given tree
4117  * returns the tree buffer or NULL.
4118  */
4119 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4120                                         struct btrfs_root *root, u32 blocksize,
4121                                         u64 parent, u64 root_objectid,
4122                                         struct btrfs_disk_key *key, int level,
4123                                         u64 hint, u64 empty_size)
4124 {
4125         struct btrfs_key ins;
4126         int ret;
4127         struct extent_buffer *buf;
4128
4129         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4130                                key, level, empty_size, hint, (u64)-1, &ins);
4131         if (ret) {
4132                 BUG_ON(ret > 0);
4133                 return ERR_PTR(ret);
4134         }
4135
4136         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4137                                     blocksize, level);
4138         return buf;
4139 }
4140
4141 #if 0
4142 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4143                         struct btrfs_root *root, struct extent_buffer *leaf)
4144 {
4145         u64 disk_bytenr;
4146         u64 num_bytes;
4147         struct btrfs_key key;
4148         struct btrfs_file_extent_item *fi;
4149         u32 nritems;
4150         int i;
4151         int ret;
4152
4153         BUG_ON(!btrfs_is_leaf(leaf));
4154         nritems = btrfs_header_nritems(leaf);
4155
4156         for (i = 0; i < nritems; i++) {
4157                 cond_resched();
4158                 btrfs_item_key_to_cpu(leaf, &key, i);
4159
4160                 /* only extents have references, skip everything else */
4161                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4162                         continue;
4163
4164                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4165
4166                 /* inline extents live in the btree, they don't have refs */
4167                 if (btrfs_file_extent_type(leaf, fi) ==
4168                     BTRFS_FILE_EXTENT_INLINE)
4169                         continue;
4170
4171                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4172
4173                 /* holes don't have refs */
4174                 if (disk_bytenr == 0)
4175                         continue;
4176
4177                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4178                 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4179                                         leaf->start, 0, key.objectid, 0);
4180                 BUG_ON(ret);
4181         }
4182         return 0;
4183 }
4184
4185 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4186                                         struct btrfs_root *root,
4187                                         struct btrfs_leaf_ref *ref)
4188 {
4189         int i;
4190         int ret;
4191         struct btrfs_extent_info *info;
4192         struct refsort *sorted;
4193
4194         if (ref->nritems == 0)
4195                 return 0;
4196
4197         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4198         for (i = 0; i < ref->nritems; i++) {
4199                 sorted[i].bytenr = ref->extents[i].bytenr;
4200                 sorted[i].slot = i;
4201         }
4202         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4203
4204         /*
4205          * the items in the ref were sorted when the ref was inserted
4206          * into the ref cache, so this is already in order
4207          */
4208         for (i = 0; i < ref->nritems; i++) {
4209                 info = ref->extents + sorted[i].slot;
4210                 ret = btrfs_free_extent(trans, root, info->bytenr,
4211                                           info->num_bytes, ref->bytenr,
4212                                           ref->owner, ref->generation,
4213                                           info->objectid, 0);
4214
4215                 atomic_inc(&root->fs_info->throttle_gen);
4216                 wake_up(&root->fs_info->transaction_throttle);
4217                 cond_resched();
4218
4219                 BUG_ON(ret);
4220                 info++;
4221         }
4222
4223         kfree(sorted);
4224         return 0;
4225 }
4226
4227
4228 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4229                                      struct btrfs_root *root, u64 start,
4230                                      u64 len, u32 *refs)
4231 {
4232         int ret;
4233
4234         ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4235         BUG_ON(ret);
4236
4237 #if 0 /* some debugging code in case we see problems here */
4238         /* if the refs count is one, it won't get increased again.  But
4239          * if the ref count is > 1, someone may be decreasing it at
4240          * the same time we are.
4241          */
4242         if (*refs != 1) {
4243                 struct extent_buffer *eb = NULL;
4244                 eb = btrfs_find_create_tree_block(root, start, len);
4245                 if (eb)
4246                         btrfs_tree_lock(eb);
4247
4248                 mutex_lock(&root->fs_info->alloc_mutex);
4249                 ret = lookup_extent_ref(NULL, root, start, len, refs);
4250                 BUG_ON(ret);
4251                 mutex_unlock(&root->fs_info->alloc_mutex);
4252
4253                 if (eb) {
4254                         btrfs_tree_unlock(eb);
4255                         free_extent_buffer(eb);
4256                 }
4257                 if (*refs == 1) {
4258                         printk(KERN_ERR "btrfs block %llu went down to one "
4259                                "during drop_snap\n", (unsigned long long)start);
4260                 }
4261
4262         }
4263 #endif
4264
4265         cond_resched();
4266         return ret;
4267 }
4268
4269
4270 /*
4271  * this is used while deleting old snapshots, and it drops the refs
4272  * on a whole subtree starting from a level 1 node.
4273  *
4274  * The idea is to sort all the leaf pointers, and then drop the
4275  * ref on all the leaves in order.  Most of the time the leaves
4276  * will have ref cache entries, so no leaf IOs will be required to
4277  * find the extents they have references on.
4278  *
4279  * For each leaf, any references it has are also dropped in order
4280  *
4281  * This ends up dropping the references in something close to optimal
4282  * order for reading and modifying the extent allocation tree.
4283  */
4284 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4285                                         struct btrfs_root *root,
4286                                         struct btrfs_path *path)
4287 {
4288         u64 bytenr;
4289         u64 root_owner;
4290         u64 root_gen;
4291         struct extent_buffer *eb = path->nodes[1];
4292         struct extent_buffer *leaf;
4293         struct btrfs_leaf_ref *ref;
4294         struct refsort *sorted = NULL;
4295         int nritems = btrfs_header_nritems(eb);
4296         int ret;
4297         int i;
4298         int refi = 0;
4299         int slot = path->slots[1];
4300         u32 blocksize = btrfs_level_size(root, 0);
4301         u32 refs;
4302
4303         if (nritems == 0)
4304                 goto out;
4305
4306         root_owner = btrfs_header_owner(eb);
4307         root_gen = btrfs_header_generation(eb);
4308         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4309
4310         /*
4311          * step one, sort all the leaf pointers so we don't scribble
4312          * randomly into the extent allocation tree
4313          */
4314         for (i = slot; i < nritems; i++) {
4315                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4316                 sorted[refi].slot = i;
4317                 refi++;
4318         }
4319
4320         /*
4321          * nritems won't be zero, but if we're picking up drop_snapshot
4322          * after a crash, slot might be > 0, so double check things
4323          * just in case.
4324          */
4325         if (refi == 0)
4326                 goto out;
4327
4328         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4329
4330         /*
4331          * the first loop frees everything the leaves point to
4332          */
4333         for (i = 0; i < refi; i++) {
4334                 u64 ptr_gen;
4335
4336                 bytenr = sorted[i].bytenr;
4337
4338                 /*
4339                  * check the reference count on this leaf.  If it is > 1
4340                  * we just decrement it below and don't update any
4341                  * of the refs the leaf points to.
4342                  */
4343                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4344                                                 blocksize, &refs);
4345                 BUG_ON(ret);
4346                 if (refs != 1)
4347                         continue;
4348
4349                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4350
4351                 /*
4352                  * the leaf only had one reference, which means the
4353                  * only thing pointing to this leaf is the snapshot
4354                  * we're deleting.  It isn't possible for the reference
4355                  * count to increase again later
4356                  *
4357                  * The reference cache is checked for the leaf,
4358                  * and if found we'll be able to drop any refs held by
4359                  * the leaf without needing to read it in.
4360                  */
4361                 ref = btrfs_lookup_leaf_ref(root, bytenr);
4362                 if (ref && ref->generation != ptr_gen) {
4363                         btrfs_free_leaf_ref(root, ref);
4364                         ref = NULL;
4365                 }
4366                 if (ref) {
4367                         ret = cache_drop_leaf_ref(trans, root, ref);
4368                         BUG_ON(ret);
4369                         btrfs_remove_leaf_ref(root, ref);
4370                         btrfs_free_leaf_ref(root, ref);
4371                 } else {
4372                         /*
4373                          * the leaf wasn't in the reference cache, so
4374                          * we have to read it.
4375                          */
4376                         leaf = read_tree_block(root, bytenr, blocksize,
4377                                                ptr_gen);
4378                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
4379                         BUG_ON(ret);
4380                         free_extent_buffer(leaf);
4381                 }
4382                 atomic_inc(&root->fs_info->throttle_gen);
4383                 wake_up(&root->fs_info->transaction_throttle);
4384                 cond_resched();
4385         }
4386
4387         /*
4388          * run through the loop again to free the refs on the leaves.
4389          * This is faster than doing it in the loop above because
4390          * the leaves are likely to be clustered together.  We end up
4391          * working in nice chunks on the extent allocation tree.
4392          */
4393         for (i = 0; i < refi; i++) {
4394                 bytenr = sorted[i].bytenr;
4395                 ret = btrfs_free_extent(trans, root, bytenr,
4396                                         blocksize, eb->start,
4397                                         root_owner, root_gen, 0, 1);
4398                 BUG_ON(ret);
4399
4400                 atomic_inc(&root->fs_info->throttle_gen);
4401                 wake_up(&root->fs_info->transaction_throttle);
4402                 cond_resched();
4403         }
4404 out:
4405         kfree(sorted);
4406
4407         /*
4408          * update the path to show we've processed the entire level 1
4409          * node.  This will get saved into the root's drop_snapshot_progress
4410          * field so these drops are not repeated again if this transaction
4411          * commits.
4412          */
4413         path->slots[1] = nritems;
4414         return 0;
4415 }
4416
4417 /*
4418  * helper function for drop_snapshot, this walks down the tree dropping ref
4419  * counts as it goes.
4420  */
4421 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4422                                    struct btrfs_root *root,
4423                                    struct btrfs_path *path, int *level)
4424 {
4425         u64 root_owner;
4426         u64 root_gen;
4427         u64 bytenr;
4428         u64 ptr_gen;
4429         struct extent_buffer *next;
4430         struct extent_buffer *cur;
4431         struct extent_buffer *parent;
4432         u32 blocksize;
4433         int ret;
4434         u32 refs;
4435
4436         WARN_ON(*level < 0);
4437         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4438         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4439                                 path->nodes[*level]->len, &refs);
4440         BUG_ON(ret);
4441         if (refs > 1)
4442                 goto out;
4443
4444         /*
4445          * walk down to the last node level and free all the leaves
4446          */
4447         while (*level >= 0) {
4448                 WARN_ON(*level < 0);
4449                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4450                 cur = path->nodes[*level];
4451
4452                 if (btrfs_header_level(cur) != *level)
4453                         WARN_ON(1);
4454
4455                 if (path->slots[*level] >=
4456                     btrfs_header_nritems(cur))
4457                         break;
4458
4459                 /* the new code goes down to level 1 and does all the
4460                  * leaves pointed to that node in bulk.  So, this check
4461                  * for level 0 will always be false.
4462                  *
4463                  * But, the disk format allows the drop_snapshot_progress
4464                  * field in the root to leave things in a state where
4465                  * a leaf will need cleaning up here.  If someone crashes
4466                  * with the old code and then boots with the new code,
4467                  * we might find a leaf here.
4468                  */
4469                 if (*level == 0) {
4470                         ret = btrfs_drop_leaf_ref(trans, root, cur);
4471                         BUG_ON(ret);
4472                         break;
4473                 }
4474
4475                 /*
4476                  * once we get to level one, process the whole node
4477                  * at once, including everything below it.
4478                  */
4479                 if (*level == 1) {
4480                         ret = drop_level_one_refs(trans, root, path);
4481                         BUG_ON(ret);
4482                         break;
4483                 }
4484
4485                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4486                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4487                 blocksize = btrfs_level_size(root, *level - 1);
4488
4489                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4490                                                 blocksize, &refs);
4491                 BUG_ON(ret);
4492
4493                 /*
4494                  * if there is more than one reference, we don't need
4495                  * to read that node to drop any references it has.  We
4496                  * just drop the ref we hold on that node and move on to the
4497                  * next slot in this level.
4498                  */
4499                 if (refs != 1) {
4500                         parent = path->nodes[*level];
4501                         root_owner = btrfs_header_owner(parent);
4502                         root_gen = btrfs_header_generation(parent);
4503                         path->slots[*level]++;
4504
4505                         ret = btrfs_free_extent(trans, root, bytenr,
4506                                                 blocksize, parent->start,
4507                                                 root_owner, root_gen,
4508                                                 *level - 1, 1);
4509                         BUG_ON(ret);
4510
4511                         atomic_inc(&root->fs_info->throttle_gen);
4512                         wake_up(&root->fs_info->transaction_throttle);
4513                         cond_resched();
4514
4515                         continue;
4516                 }
4517
4518                 /*
4519                  * we need to keep freeing things in the next level down.
4520                  * read the block and loop around to process it
4521                  */
4522                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4523                 WARN_ON(*level <= 0);
4524                 if (path->nodes[*level-1])
4525                         free_extent_buffer(path->nodes[*level-1]);
4526                 path->nodes[*level-1] = next;
4527                 *level = btrfs_header_level(next);
4528                 path->slots[*level] = 0;
4529                 cond_resched();
4530         }
4531 out:
4532         WARN_ON(*level < 0);
4533         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4534
4535         if (path->nodes[*level] == root->node) {
4536                 parent = path->nodes[*level];
4537                 bytenr = path->nodes[*level]->start;
4538         } else {
4539                 parent = path->nodes[*level + 1];
4540                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4541         }
4542
4543         blocksize = btrfs_level_size(root, *level);
4544         root_owner = btrfs_header_owner(parent);
4545         root_gen = btrfs_header_generation(parent);
4546
4547         /*
4548          * cleanup and free the reference on the last node
4549          * we processed
4550          */
4551         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4552                                   parent->start, root_owner, root_gen,
4553                                   *level, 1);
4554         free_extent_buffer(path->nodes[*level]);
4555         path->nodes[*level] = NULL;
4556
4557         *level += 1;
4558         BUG_ON(ret);
4559
4560         cond_resched();
4561         return 0;
4562 }
4563 #endif
4564
4565 struct walk_control {
4566         u64 refs[BTRFS_MAX_LEVEL];
4567         u64 flags[BTRFS_MAX_LEVEL];
4568         struct btrfs_key update_progress;
4569         int stage;
4570         int level;
4571         int shared_level;
4572         int update_ref;
4573         int keep_locks;
4574 };
4575
4576 #define DROP_REFERENCE  1
4577 #define UPDATE_BACKREF  2
4578
4579 /*
4580  * hepler to process tree block while walking down the tree.
4581  *
4582  * when wc->stage == DROP_REFERENCE, this function checks
4583  * reference count of the block. if the block is shared and
4584  * we need update back refs for the subtree rooted at the
4585  * block, this function changes wc->stage to UPDATE_BACKREF
4586  *
4587  * when wc->stage == UPDATE_BACKREF, this function updates
4588  * back refs for pointers in the block.
4589  *
4590  * NOTE: return value 1 means we should stop walking down.
4591  */
4592 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4593                                    struct btrfs_root *root,
4594                                    struct btrfs_path *path,
4595                                    struct walk_control *wc)
4596 {
4597         int level = wc->level;
4598         struct extent_buffer *eb = path->nodes[level];
4599         struct btrfs_key key;
4600         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4601         int ret;
4602
4603         if (wc->stage == UPDATE_BACKREF &&
4604             btrfs_header_owner(eb) != root->root_key.objectid)
4605                 return 1;
4606
4607         /*
4608          * when reference count of tree block is 1, it won't increase
4609          * again. once full backref flag is set, we never clear it.
4610          */
4611         if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4612             (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4613                 BUG_ON(!path->locks[level]);
4614                 ret = btrfs_lookup_extent_info(trans, root,
4615                                                eb->start, eb->len,
4616                                                &wc->refs[level],
4617                                                &wc->flags[level]);
4618                 BUG_ON(ret);
4619                 BUG_ON(wc->refs[level] == 0);
4620         }
4621
4622         if (wc->stage == DROP_REFERENCE &&
4623             wc->update_ref && wc->refs[level] > 1) {
4624                 BUG_ON(eb == root->node);
4625                 BUG_ON(path->slots[level] > 0);
4626                 if (level == 0)
4627                         btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4628                 else
4629                         btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4630                 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4631                     btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4632                         wc->stage = UPDATE_BACKREF;
4633                         wc->shared_level = level;
4634                 }
4635         }
4636
4637         if (wc->stage == DROP_REFERENCE) {
4638                 if (wc->refs[level] > 1)
4639                         return 1;
4640
4641                 if (path->locks[level] && !wc->keep_locks) {
4642                         btrfs_tree_unlock(eb);
4643                         path->locks[level] = 0;
4644                 }
4645                 return 0;
4646         }
4647
4648         /* wc->stage == UPDATE_BACKREF */
4649         if (!(wc->flags[level] & flag)) {
4650                 BUG_ON(!path->locks[level]);
4651                 ret = btrfs_inc_ref(trans, root, eb, 1);
4652                 BUG_ON(ret);
4653                 ret = btrfs_dec_ref(trans, root, eb, 0);
4654                 BUG_ON(ret);
4655                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4656                                                   eb->len, flag, 0);
4657                 BUG_ON(ret);
4658                 wc->flags[level] |= flag;
4659         }
4660
4661         /*
4662          * the block is shared by multiple trees, so it's not good to
4663          * keep the tree lock
4664          */
4665         if (path->locks[level] && level > 0) {
4666                 btrfs_tree_unlock(eb);
4667                 path->locks[level] = 0;
4668         }
4669         return 0;
4670 }
4671
4672 /*
4673  * hepler to process tree block while walking up the tree.
4674  *
4675  * when wc->stage == DROP_REFERENCE, this function drops
4676  * reference count on the block.
4677  *
4678  * when wc->stage == UPDATE_BACKREF, this function changes
4679  * wc->stage back to DROP_REFERENCE if we changed wc->stage
4680  * to UPDATE_BACKREF previously while processing the block.
4681  *
4682  * NOTE: return value 1 means we should stop walking up.
4683  */
4684 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4685                                  struct btrfs_root *root,
4686                                  struct btrfs_path *path,
4687                                  struct walk_control *wc)
4688 {
4689         int ret = 0;
4690         int level = wc->level;
4691         struct extent_buffer *eb = path->nodes[level];
4692         u64 parent = 0;
4693
4694         if (wc->stage == UPDATE_BACKREF) {
4695                 BUG_ON(wc->shared_level < level);
4696                 if (level < wc->shared_level)
4697                         goto out;
4698
4699                 BUG_ON(wc->refs[level] <= 1);
4700                 ret = find_next_key(path, level + 1, &wc->update_progress);
4701                 if (ret > 0)
4702                         wc->update_ref = 0;
4703
4704                 wc->stage = DROP_REFERENCE;
4705                 wc->shared_level = -1;
4706                 path->slots[level] = 0;
4707
4708                 /*
4709                  * check reference count again if the block isn't locked.
4710                  * we should start walking down the tree again if reference
4711                  * count is one.
4712                  */
4713                 if (!path->locks[level]) {
4714                         BUG_ON(level == 0);
4715                         btrfs_tree_lock(eb);
4716                         btrfs_set_lock_blocking(eb);
4717                         path->locks[level] = 1;
4718
4719                         ret = btrfs_lookup_extent_info(trans, root,
4720                                                        eb->start, eb->len,
4721                                                        &wc->refs[level],
4722                                                        &wc->flags[level]);
4723                         BUG_ON(ret);
4724                         BUG_ON(wc->refs[level] == 0);
4725                         if (wc->refs[level] == 1) {
4726                                 btrfs_tree_unlock(eb);
4727                                 path->locks[level] = 0;
4728                                 return 1;
4729                         }
4730                 } else {
4731                         BUG_ON(level != 0);
4732                 }
4733         }
4734
4735         /* wc->stage == DROP_REFERENCE */
4736         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4737
4738         if (wc->refs[level] == 1) {
4739                 if (level == 0) {
4740                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4741                                 ret = btrfs_dec_ref(trans, root, eb, 1);
4742                         else
4743                                 ret = btrfs_dec_ref(trans, root, eb, 0);
4744                         BUG_ON(ret);
4745                 }
4746                 /* make block locked assertion in clean_tree_block happy */
4747                 if (!path->locks[level] &&
4748                     btrfs_header_generation(eb) == trans->transid) {
4749                         btrfs_tree_lock(eb);
4750                         btrfs_set_lock_blocking(eb);
4751                         path->locks[level] = 1;
4752                 }
4753                 clean_tree_block(trans, root, eb);
4754         }
4755
4756         if (eb == root->node) {
4757                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4758                         parent = eb->start;
4759                 else
4760                         BUG_ON(root->root_key.objectid !=
4761                                btrfs_header_owner(eb));
4762         } else {
4763                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4764                         parent = path->nodes[level + 1]->start;
4765                 else
4766                         BUG_ON(root->root_key.objectid !=
4767                                btrfs_header_owner(path->nodes[level + 1]));
4768         }
4769
4770         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
4771                                 root->root_key.objectid, level, 0);
4772         BUG_ON(ret);
4773 out:
4774         wc->refs[level] = 0;
4775         wc->flags[level] = 0;
4776         return ret;
4777 }
4778
4779 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4780                                    struct btrfs_root *root,
4781                                    struct btrfs_path *path,
4782                                    struct walk_control *wc)
4783 {
4784         struct extent_buffer *next;
4785         struct extent_buffer *cur;
4786         u64 bytenr;
4787         u64 ptr_gen;
4788         u32 blocksize;
4789         int level = wc->level;
4790         int ret;
4791
4792         while (level >= 0) {
4793                 cur = path->nodes[level];
4794                 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
4795
4796                 ret = walk_down_proc(trans, root, path, wc);
4797                 if (ret > 0)
4798                         break;
4799
4800                 if (level == 0)
4801                         break;
4802
4803                 bytenr = btrfs_node_blockptr(cur, path->slots[level]);
4804                 blocksize = btrfs_level_size(root, level - 1);
4805                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
4806
4807                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4808                 btrfs_tree_lock(next);
4809                 btrfs_set_lock_blocking(next);
4810
4811                 level--;
4812                 BUG_ON(level != btrfs_header_level(next));
4813                 path->nodes[level] = next;
4814                 path->slots[level] = 0;
4815                 path->locks[level] = 1;
4816                 wc->level = level;
4817         }
4818         return 0;
4819 }
4820
4821 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4822                                  struct btrfs_root *root,
4823                                  struct btrfs_path *path,
4824                                  struct walk_control *wc, int max_level)
4825 {
4826         int level = wc->level;
4827         int ret;
4828
4829         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
4830         while (level < max_level && path->nodes[level]) {
4831                 wc->level = level;
4832                 if (path->slots[level] + 1 <
4833                     btrfs_header_nritems(path->nodes[level])) {
4834                         path->slots[level]++;
4835                         return 0;
4836                 } else {
4837                         ret = walk_up_proc(trans, root, path, wc);
4838                         if (ret > 0)
4839                                 return 0;
4840
4841                         if (path->locks[level]) {
4842                                 btrfs_tree_unlock(path->nodes[level]);
4843                                 path->locks[level] = 0;
4844                         }
4845                         free_extent_buffer(path->nodes[level]);
4846                         path->nodes[level] = NULL;
4847                         level++;
4848                 }
4849         }
4850         return 1;
4851 }
4852
4853 /*
4854  * drop a subvolume tree.
4855  *
4856  * this function traverses the tree freeing any blocks that only
4857  * referenced by the tree.
4858  *
4859  * when a shared tree block is found. this function decreases its
4860  * reference count by one. if update_ref is true, this function
4861  * also make sure backrefs for the shared block and all lower level
4862  * blocks are properly updated.
4863  */
4864 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
4865 {
4866         struct btrfs_path *path;
4867         struct btrfs_trans_handle *trans;
4868         struct btrfs_root *tree_root = root->fs_info->tree_root;
4869         struct btrfs_root_item *root_item = &root->root_item;
4870         struct walk_control *wc;
4871         struct btrfs_key key;
4872         int err = 0;
4873         int ret;
4874         int level;
4875
4876         path = btrfs_alloc_path();
4877         BUG_ON(!path);
4878
4879         wc = kzalloc(sizeof(*wc), GFP_NOFS);
4880         BUG_ON(!wc);
4881
4882         trans = btrfs_start_transaction(tree_root, 1);
4883
4884         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4885                 level = btrfs_header_level(root->node);
4886                 path->nodes[level] = btrfs_lock_root_node(root);
4887                 btrfs_set_lock_blocking(path->nodes[level]);
4888                 path->slots[level] = 0;
4889                 path->locks[level] = 1;
4890                 memset(&wc->update_progress, 0,
4891                        sizeof(wc->update_progress));
4892         } else {
4893                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4894                 memcpy(&wc->update_progress, &key,
4895                        sizeof(wc->update_progress));
4896
4897                 level = root_item->drop_level;
4898                 BUG_ON(level == 0);
4899                 path->lowest_level = level;
4900                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4901                 path->lowest_level = 0;
4902                 if (ret < 0) {
4903                         err = ret;
4904                         goto out;
4905                 }
4906                 btrfs_node_key_to_cpu(path->nodes[level], &key,
4907                                       path->slots[level]);
4908                 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
4909
4910                 /*
4911                  * unlock our path, this is safe because only this
4912                  * function is allowed to delete this snapshot
4913                  */
4914                 btrfs_unlock_up_safe(path, 0);
4915
4916                 level = btrfs_header_level(root->node);
4917                 while (1) {
4918                         btrfs_tree_lock(path->nodes[level]);
4919                         btrfs_set_lock_blocking(path->nodes[level]);
4920
4921                         ret = btrfs_lookup_extent_info(trans, root,
4922                                                 path->nodes[level]->start,
4923                                                 path->nodes[level]->len,
4924                                                 &wc->refs[level],
4925                                                 &wc->flags[level]);
4926                         BUG_ON(ret);
4927                         BUG_ON(wc->refs[level] == 0);
4928
4929                         if (level == root_item->drop_level)
4930                                 break;
4931
4932                         btrfs_tree_unlock(path->nodes[level]);
4933                         WARN_ON(wc->refs[level] != 1);
4934                         level--;
4935                 }
4936         }
4937
4938         wc->level = level;
4939         wc->shared_level = -1;
4940         wc->stage = DROP_REFERENCE;
4941         wc->update_ref = update_ref;
4942         wc->keep_locks = 0;
4943
4944         while (1) {
4945                 ret = walk_down_tree(trans, root, path, wc);
4946                 if (ret < 0) {
4947                         err = ret;
4948                         break;
4949                 }
4950
4951                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
4952                 if (ret < 0) {
4953                         err = ret;
4954                         break;
4955                 }
4956
4957                 if (ret > 0) {
4958                         BUG_ON(wc->stage != DROP_REFERENCE);
4959                         break;
4960                 }
4961
4962                 if (wc->stage == DROP_REFERENCE) {
4963                         level = wc->level;
4964                         btrfs_node_key(path->nodes[level],
4965                                        &root_item->drop_progress,
4966                                        path->slots[level]);
4967                         root_item->drop_level = level;
4968                 }
4969
4970                 BUG_ON(wc->level == 0);
4971                 if (trans->transaction->in_commit ||
4972                     trans->transaction->delayed_refs.flushing) {
4973                         ret = btrfs_update_root(trans, tree_root,
4974                                                 &root->root_key,
4975                                                 root_item);
4976                         BUG_ON(ret);
4977
4978                         btrfs_end_transaction(trans, tree_root);
4979                         trans = btrfs_start_transaction(tree_root, 1);
4980                 } else {
4981                         unsigned long update;
4982                         update = trans->delayed_ref_updates;
4983                         trans->delayed_ref_updates = 0;
4984                         if (update)
4985                                 btrfs_run_delayed_refs(trans, tree_root,
4986                                                        update);
4987                 }
4988         }
4989         btrfs_release_path(root, path);
4990         BUG_ON(err);
4991
4992         ret = btrfs_del_root(trans, tree_root, &root->root_key);
4993         BUG_ON(ret);
4994
4995         free_extent_buffer(root->node);
4996         free_extent_buffer(root->commit_root);
4997         kfree(root);
4998 out:
4999         btrfs_end_transaction(trans, tree_root);
5000         kfree(wc);
5001         btrfs_free_path(path);
5002         return err;
5003 }
5004
5005 /*
5006  * drop subtree rooted at tree block 'node'.
5007  *
5008  * NOTE: this function will unlock and release tree block 'node'
5009  */
5010 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5011                         struct btrfs_root *root,
5012                         struct extent_buffer *node,
5013                         struct extent_buffer *parent)
5014 {
5015         struct btrfs_path *path;
5016         struct walk_control *wc;
5017         int level;
5018         int parent_level;
5019         int ret = 0;
5020         int wret;
5021
5022         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5023
5024         path = btrfs_alloc_path();
5025         BUG_ON(!path);
5026
5027         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5028         BUG_ON(!wc);
5029
5030         btrfs_assert_tree_locked(parent);
5031         parent_level = btrfs_header_level(parent);
5032         extent_buffer_get(parent);
5033         path->nodes[parent_level] = parent;
5034         path->slots[parent_level] = btrfs_header_nritems(parent);
5035
5036         btrfs_assert_tree_locked(node);
5037         level = btrfs_header_level(node);
5038         path->nodes[level] = node;
5039         path->slots[level] = 0;
5040         path->locks[level] = 1;
5041
5042         wc->refs[parent_level] = 1;
5043         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5044         wc->level = level;
5045         wc->shared_level = -1;
5046         wc->stage = DROP_REFERENCE;
5047         wc->update_ref = 0;
5048         wc->keep_locks = 1;
5049
5050         while (1) {
5051                 wret = walk_down_tree(trans, root, path, wc);
5052                 if (wret < 0) {
5053                         ret = wret;
5054                         break;
5055                 }
5056
5057                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5058                 if (wret < 0)
5059                         ret = wret;
5060                 if (wret != 0)
5061                         break;
5062         }
5063
5064         kfree(wc);
5065         btrfs_free_path(path);
5066         return ret;
5067 }
5068
5069 #if 0
5070 static unsigned long calc_ra(unsigned long start, unsigned long last,
5071                              unsigned long nr)
5072 {
5073         return min(last, start + nr - 1);
5074 }
5075
5076 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5077                                          u64 len)
5078 {
5079         u64 page_start;
5080         u64 page_end;
5081         unsigned long first_index;
5082         unsigned long last_index;
5083         unsigned long i;
5084         struct page *page;
5085         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5086         struct file_ra_state *ra;
5087         struct btrfs_ordered_extent *ordered;
5088         unsigned int total_read = 0;
5089         unsigned int total_dirty = 0;
5090         int ret = 0;
5091
5092         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5093
5094         mutex_lock(&inode->i_mutex);
5095         first_index = start >> PAGE_CACHE_SHIFT;
5096         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5097
5098         /* make sure the dirty trick played by the caller work */
5099         ret = invalidate_inode_pages2_range(inode->i_mapping,
5100                                             first_index, last_index);
5101         if (ret)
5102                 goto out_unlock;
5103
5104         file_ra_state_init(ra, inode->i_mapping);
5105
5106         for (i = first_index ; i <= last_index; i++) {
5107                 if (total_read % ra->ra_pages == 0) {
5108                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5109                                        calc_ra(i, last_index, ra->ra_pages));
5110                 }
5111                 total_read++;
5112 again:
5113                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5114                         BUG_ON(1);
5115                 page = grab_cache_page(inode->i_mapping, i);
5116                 if (!page) {
5117                         ret = -ENOMEM;
5118                         goto out_unlock;
5119                 }
5120                 if (!PageUptodate(page)) {
5121                         btrfs_readpage(NULL, page);
5122                         lock_page(page);
5123                         if (!PageUptodate(page)) {
5124                                 unlock_page(page);
5125                                 page_cache_release(page);
5126                                 ret = -EIO;
5127                                 goto out_unlock;
5128                         }
5129                 }
5130                 wait_on_page_writeback(page);
5131
5132                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5133                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5134                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5135
5136                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5137                 if (ordered) {
5138                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5139                         unlock_page(page);
5140                         page_cache_release(page);
5141                         btrfs_start_ordered_extent(inode, ordered, 1);
5142                         btrfs_put_ordered_extent(ordered);
5143                         goto again;
5144                 }
5145                 set_page_extent_mapped(page);
5146
5147                 if (i == first_index)
5148                         set_extent_bits(io_tree, page_start, page_end,
5149                                         EXTENT_BOUNDARY, GFP_NOFS);
5150                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5151
5152                 set_page_dirty(page);
5153                 total_dirty++;
5154
5155                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5156                 unlock_page(page);
5157                 page_cache_release(page);
5158         }
5159
5160 out_unlock:
5161         kfree(ra);
5162         mutex_unlock(&inode->i_mutex);
5163         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5164         return ret;
5165 }
5166
5167 static noinline int relocate_data_extent(struct inode *reloc_inode,
5168                                          struct btrfs_key *extent_key,
5169                                          u64 offset)
5170 {
5171         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5172         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5173         struct extent_map *em;
5174         u64 start = extent_key->objectid - offset;
5175         u64 end = start + extent_key->offset - 1;
5176
5177         em = alloc_extent_map(GFP_NOFS);
5178         BUG_ON(!em || IS_ERR(em));
5179
5180         em->start = start;
5181         em->len = extent_key->offset;
5182         em->block_len = extent_key->offset;
5183         em->block_start = extent_key->objectid;
5184         em->bdev = root->fs_info->fs_devices->latest_bdev;
5185         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5186
5187         /* setup extent map to cheat btrfs_readpage */
5188         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5189         while (1) {
5190                 int ret;
5191                 spin_lock(&em_tree->lock);
5192                 ret = add_extent_mapping(em_tree, em);
5193                 spin_unlock(&em_tree->lock);
5194                 if (ret != -EEXIST) {
5195                         free_extent_map(em);
5196                         break;
5197                 }
5198                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5199         }
5200         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5201
5202         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5203 }
5204
5205 struct btrfs_ref_path {
5206         u64 extent_start;
5207         u64 nodes[BTRFS_MAX_LEVEL];
5208         u64 root_objectid;
5209         u64 root_generation;
5210         u64 owner_objectid;
5211         u32 num_refs;
5212         int lowest_level;
5213         int current_level;
5214         int shared_level;
5215
5216         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5217         u64 new_nodes[BTRFS_MAX_LEVEL];
5218 };
5219
5220 struct disk_extent {
5221         u64 ram_bytes;
5222         u64 disk_bytenr;
5223         u64 disk_num_bytes;
5224         u64 offset;
5225         u64 num_bytes;
5226         u8 compression;
5227         u8 encryption;
5228         u16 other_encoding;
5229 };
5230
5231 static int is_cowonly_root(u64 root_objectid)
5232 {
5233         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5234             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5235             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5236             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5237             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5238             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5239                 return 1;
5240         return 0;
5241 }
5242
5243 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5244                                     struct btrfs_root *extent_root,
5245                                     struct btrfs_ref_path *ref_path,
5246                                     int first_time)
5247 {
5248         struct extent_buffer *leaf;
5249         struct btrfs_path *path;
5250         struct btrfs_extent_ref *ref;
5251         struct btrfs_key key;
5252         struct btrfs_key found_key;
5253         u64 bytenr;
5254         u32 nritems;
5255         int level;
5256         int ret = 1;
5257
5258         path = btrfs_alloc_path();
5259         if (!path)
5260                 return -ENOMEM;
5261
5262         if (first_time) {
5263                 ref_path->lowest_level = -1;
5264                 ref_path->current_level = -1;
5265                 ref_path->shared_level = -1;
5266                 goto walk_up;
5267         }
5268 walk_down:
5269         level = ref_path->current_level - 1;
5270         while (level >= -1) {
5271                 u64 parent;
5272                 if (level < ref_path->lowest_level)
5273                         break;
5274
5275                 if (level >= 0)
5276                         bytenr = ref_path->nodes[level];
5277                 else
5278                         bytenr = ref_path->extent_start;
5279                 BUG_ON(bytenr == 0);
5280
5281                 parent = ref_path->nodes[level + 1];
5282                 ref_path->nodes[level + 1] = 0;
5283                 ref_path->current_level = level;
5284                 BUG_ON(parent == 0);
5285
5286                 key.objectid = bytenr;
5287                 key.offset = parent + 1;
5288                 key.type = BTRFS_EXTENT_REF_KEY;
5289
5290                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5291                 if (ret < 0)
5292                         goto out;
5293                 BUG_ON(ret == 0);
5294
5295                 leaf = path->nodes[0];
5296                 nritems = btrfs_header_nritems(leaf);
5297                 if (path->slots[0] >= nritems) {
5298                         ret = btrfs_next_leaf(extent_root, path);
5299                         if (ret < 0)
5300                                 goto out;
5301                         if (ret > 0)
5302                                 goto next;
5303                         leaf = path->nodes[0];
5304                 }
5305
5306                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5307                 if (found_key.objectid == bytenr &&
5308                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5309                         if (level < ref_path->shared_level)
5310                                 ref_path->shared_level = level;
5311                         goto found;
5312                 }
5313 next:
5314                 level--;
5315                 btrfs_release_path(extent_root, path);
5316                 cond_resched();
5317         }
5318         /* reached lowest level */
5319         ret = 1;
5320         goto out;
5321 walk_up:
5322         level = ref_path->current_level;
5323         while (level < BTRFS_MAX_LEVEL - 1) {
5324                 u64 ref_objectid;
5325
5326                 if (level >= 0)
5327                         bytenr = ref_path->nodes[level];
5328                 else
5329                         bytenr = ref_path->extent_start;
5330
5331                 BUG_ON(bytenr == 0);
5332
5333                 key.objectid = bytenr;
5334                 key.offset = 0;
5335                 key.type = BTRFS_EXTENT_REF_KEY;
5336
5337                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5338                 if (ret < 0)
5339                         goto out;
5340
5341                 leaf = path->nodes[0];
5342                 nritems = btrfs_header_nritems(leaf);
5343                 if (path->slots[0] >= nritems) {
5344                         ret = btrfs_next_leaf(extent_root, path);
5345                         if (ret < 0)
5346                                 goto out;
5347                         if (ret > 0) {
5348                                 /* the extent was freed by someone */
5349                                 if (ref_path->lowest_level == level)
5350                                         goto out;
5351                                 btrfs_release_path(extent_root, path);
5352                                 goto walk_down;
5353                         }
5354                         leaf = path->nodes[0];
5355                 }
5356
5357                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5358                 if (found_key.objectid != bytenr ||
5359                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5360                         /* the extent was freed by someone */
5361                         if (ref_path->lowest_level == level) {
5362                                 ret = 1;
5363                                 goto out;
5364                         }
5365                         btrfs_release_path(extent_root, path);
5366                         goto walk_down;
5367                 }
5368 found:
5369                 ref = btrfs_item_ptr(leaf, path->slots[0],
5370                                 struct btrfs_extent_ref);
5371                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5372                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5373                         if (first_time) {
5374                                 level = (int)ref_objectid;
5375                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5376                                 ref_path->lowest_level = level;
5377                                 ref_path->current_level = level;
5378                                 ref_path->nodes[level] = bytenr;
5379                         } else {
5380                                 WARN_ON(ref_objectid != level);
5381                         }
5382                 } else {
5383                         WARN_ON(level != -1);
5384                 }
5385                 first_time = 0;
5386
5387                 if (ref_path->lowest_level == level) {
5388                         ref_path->owner_objectid = ref_objectid;
5389                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5390                 }
5391
5392                 /*
5393                  * the block is tree root or the block isn't in reference
5394                  * counted tree.
5395                  */
5396                 if (found_key.objectid == found_key.offset ||
5397                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5398                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5399                         ref_path->root_generation =
5400                                 btrfs_ref_generation(leaf, ref);
5401                         if (level < 0) {
5402                                 /* special reference from the tree log */
5403                                 ref_path->nodes[0] = found_key.offset;
5404                                 ref_path->current_level = 0;
5405                         }
5406                         ret = 0;
5407                         goto out;
5408                 }
5409
5410                 level++;
5411                 BUG_ON(ref_path->nodes[level] != 0);
5412                 ref_path->nodes[level] = found_key.offset;
5413                 ref_path->current_level = level;
5414
5415                 /*
5416                  * the reference was created in the running transaction,
5417                  * no need to continue walking up.
5418                  */
5419                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5420                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5421                         ref_path->root_generation =
5422                                 btrfs_ref_generation(leaf, ref);
5423                         ret = 0;
5424                         goto out;
5425                 }
5426
5427                 btrfs_release_path(extent_root, path);
5428                 cond_resched();
5429         }
5430         /* reached max tree level, but no tree root found. */
5431         BUG();
5432 out:
5433         btrfs_free_path(path);
5434         return ret;
5435 }
5436
5437 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5438                                 struct btrfs_root *extent_root,
5439                                 struct btrfs_ref_path *ref_path,
5440                                 u64 extent_start)
5441 {
5442         memset(ref_path, 0, sizeof(*ref_path));
5443         ref_path->extent_start = extent_start;
5444
5445         return __next_ref_path(trans, extent_root, ref_path, 1);
5446 }
5447
5448 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5449                                struct btrfs_root *extent_root,
5450                                struct btrfs_ref_path *ref_path)
5451 {
5452         return __next_ref_path(trans, extent_root, ref_path, 0);
5453 }
5454
5455 static noinline int get_new_locations(struct inode *reloc_inode,
5456                                       struct btrfs_key *extent_key,
5457                                       u64 offset, int no_fragment,
5458                                       struct disk_extent **extents,
5459                                       int *nr_extents)
5460 {
5461         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5462         struct btrfs_path *path;
5463         struct btrfs_file_extent_item *fi;
5464         struct extent_buffer *leaf;
5465         struct disk_extent *exts = *extents;
5466         struct btrfs_key found_key;
5467         u64 cur_pos;
5468         u64 last_byte;
5469         u32 nritems;
5470         int nr = 0;
5471         int max = *nr_extents;
5472         int ret;
5473
5474         WARN_ON(!no_fragment && *extents);
5475         if (!exts) {
5476                 max = 1;
5477                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5478                 if (!exts)
5479                         return -ENOMEM;
5480         }
5481
5482         path = btrfs_alloc_path();
5483         BUG_ON(!path);
5484
5485         cur_pos = extent_key->objectid - offset;
5486         last_byte = extent_key->objectid + extent_key->offset;
5487         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5488                                        cur_pos, 0);
5489         if (ret < 0)
5490                 goto out;
5491         if (ret > 0) {
5492                 ret = -ENOENT;
5493                 goto out;
5494         }
5495
5496         while (1) {
5497                 leaf = path->nodes[0];
5498                 nritems = btrfs_header_nritems(leaf);
5499                 if (path->slots[0] >= nritems) {
5500                         ret = btrfs_next_leaf(root, path);
5501                         if (ret < 0)
5502                                 goto out;
5503                         if (ret > 0)
5504                                 break;
5505                         leaf = path->nodes[0];
5506                 }
5507
5508                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5509                 if (found_key.offset != cur_pos ||
5510                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
5511                     found_key.objectid != reloc_inode->i_ino)
5512                         break;
5513
5514                 fi = btrfs_item_ptr(leaf, path->slots[0],
5515                                     struct btrfs_file_extent_item);
5516                 if (btrfs_file_extent_type(leaf, fi) !=
5517                     BTRFS_FILE_EXTENT_REG ||
5518                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5519                         break;
5520
5521                 if (nr == max) {
5522                         struct disk_extent *old = exts;
5523                         max *= 2;
5524                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5525                         memcpy(exts, old, sizeof(*exts) * nr);
5526                         if (old != *extents)
5527                                 kfree(old);
5528                 }
5529
5530                 exts[nr].disk_bytenr =
5531                         btrfs_file_extent_disk_bytenr(leaf, fi);
5532                 exts[nr].disk_num_bytes =
5533                         btrfs_file_extent_disk_num_bytes(leaf, fi);
5534                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5535                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5536                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5537                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5538                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5539                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5540                                                                            fi);
5541                 BUG_ON(exts[nr].offset > 0);
5542                 BUG_ON(exts[nr].compression || exts[nr].encryption);
5543                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5544
5545                 cur_pos += exts[nr].num_bytes;
5546                 nr++;
5547
5548                 if (cur_pos + offset >= last_byte)
5549                         break;
5550
5551                 if (no_fragment) {
5552                         ret = 1;
5553                         goto out;
5554                 }
5555                 path->slots[0]++;
5556         }
5557
5558         BUG_ON(cur_pos + offset > last_byte);
5559         if (cur_pos + offset < last_byte) {
5560                 ret = -ENOENT;
5561                 goto out;
5562         }
5563         ret = 0;
5564 out:
5565         btrfs_free_path(path);
5566         if (ret) {
5567                 if (exts != *extents)
5568                         kfree(exts);
5569         } else {
5570                 *extents = exts;
5571                 *nr_extents = nr;
5572         }
5573         return ret;
5574 }
5575
5576 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5577                                         struct btrfs_root *root,
5578                                         struct btrfs_path *path,
5579                                         struct btrfs_key *extent_key,
5580                                         struct btrfs_key *leaf_key,
5581                                         struct btrfs_ref_path *ref_path,
5582                                         struct disk_extent *new_extents,
5583                                         int nr_extents)
5584 {
5585         struct extent_buffer *leaf;
5586         struct btrfs_file_extent_item *fi;
5587         struct inode *inode = NULL;
5588         struct btrfs_key key;
5589         u64 lock_start = 0;
5590         u64 lock_end = 0;
5591         u64 num_bytes;
5592         u64 ext_offset;
5593         u64 search_end = (u64)-1;
5594         u32 nritems;
5595         int nr_scaned = 0;
5596         int extent_locked = 0;
5597         int extent_type;
5598         int ret;
5599
5600         memcpy(&key, leaf_key, sizeof(key));
5601         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5602                 if (key.objectid < ref_path->owner_objectid ||
5603                     (key.objectid == ref_path->owner_objectid &&
5604                      key.type < BTRFS_EXTENT_DATA_KEY)) {
5605                         key.objectid = ref_path->owner_objectid;
5606                         key.type = BTRFS_EXTENT_DATA_KEY;
5607                         key.offset = 0;
5608                 }
5609         }
5610
5611         while (1) {
5612                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5613                 if (ret < 0)
5614                         goto out;
5615
5616                 leaf = path->nodes[0];
5617                 nritems = btrfs_header_nritems(leaf);
5618 next:
5619                 if (extent_locked && ret > 0) {
5620                         /*
5621                          * the file extent item was modified by someone
5622                          * before the extent got locked.
5623                          */
5624                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5625                                       lock_end, GFP_NOFS);
5626                         extent_locked = 0;
5627                 }
5628
5629                 if (path->slots[0] >= nritems) {
5630                         if (++nr_scaned > 2)
5631                                 break;
5632
5633                         BUG_ON(extent_locked);
5634                         ret = btrfs_next_leaf(root, path);
5635                         if (ret < 0)
5636                                 goto out;
5637                         if (ret > 0)
5638                                 break;
5639                         leaf = path->nodes[0];
5640                         nritems = btrfs_header_nritems(leaf);
5641                 }
5642
5643                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5644
5645                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5646                         if ((key.objectid > ref_path->owner_objectid) ||
5647                             (key.objectid == ref_path->owner_objectid &&
5648                              key.type > BTRFS_EXTENT_DATA_KEY) ||
5649                             key.offset >= search_end)
5650                                 break;
5651                 }
5652
5653                 if (inode && key.objectid != inode->i_ino) {
5654                         BUG_ON(extent_locked);
5655                         btrfs_release_path(root, path);
5656                         mutex_unlock(&inode->i_mutex);
5657                         iput(inode);
5658                         inode = NULL;
5659                         continue;
5660                 }
5661
5662                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5663                         path->slots[0]++;
5664                         ret = 1;
5665                         goto next;
5666                 }
5667                 fi = btrfs_item_ptr(leaf, path->slots[0],
5668                                     struct btrfs_file_extent_item);
5669                 extent_type = btrfs_file_extent_type(leaf, fi);
5670                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5671                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5672                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5673                      extent_key->objectid)) {
5674                         path->slots[0]++;
5675                         ret = 1;
5676                         goto next;
5677                 }
5678
5679                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5680                 ext_offset = btrfs_file_extent_offset(leaf, fi);
5681
5682                 if (search_end == (u64)-1) {
5683                         search_end = key.offset - ext_offset +
5684                                 btrfs_file_extent_ram_bytes(leaf, fi);
5685                 }
5686
5687                 if (!extent_locked) {
5688                         lock_start = key.offset;
5689                         lock_end = lock_start + num_bytes - 1;
5690                 } else {
5691                         if (lock_start > key.offset ||
5692                             lock_end + 1 < key.offset + num_bytes) {
5693                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5694                                               lock_start, lock_end, GFP_NOFS);
5695                                 extent_locked = 0;
5696                         }
5697                 }
5698
5699                 if (!inode) {
5700                         btrfs_release_path(root, path);
5701
5702                         inode = btrfs_iget_locked(root->fs_info->sb,
5703                                                   key.objectid, root);
5704                         if (inode->i_state & I_NEW) {
5705                                 BTRFS_I(inode)->root = root;
5706                                 BTRFS_I(inode)->location.objectid =
5707                                         key.objectid;
5708                                 BTRFS_I(inode)->location.type =
5709                                         BTRFS_INODE_ITEM_KEY;
5710                                 BTRFS_I(inode)->location.offset = 0;
5711                                 btrfs_read_locked_inode(inode);
5712                                 unlock_new_inode(inode);
5713                         }
5714                         /*
5715                          * some code call btrfs_commit_transaction while
5716                          * holding the i_mutex, so we can't use mutex_lock
5717                          * here.
5718                          */
5719                         if (is_bad_inode(inode) ||
5720                             !mutex_trylock(&inode->i_mutex)) {
5721                                 iput(inode);
5722                                 inode = NULL;
5723                                 key.offset = (u64)-1;
5724                                 goto skip;
5725                         }
5726                 }
5727
5728                 if (!extent_locked) {
5729                         struct btrfs_ordered_extent *ordered;
5730
5731                         btrfs_release_path(root, path);
5732
5733                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5734                                     lock_end, GFP_NOFS);
5735                         ordered = btrfs_lookup_first_ordered_extent(inode,
5736                                                                     lock_end);
5737                         if (ordered &&
5738                             ordered->file_offset <= lock_end &&
5739                             ordered->file_offset + ordered->len > lock_start) {
5740                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5741                                               lock_start, lock_end, GFP_NOFS);
5742                                 btrfs_start_ordered_extent(inode, ordered, 1);
5743                                 btrfs_put_ordered_extent(ordered);
5744                                 key.offset += num_bytes;
5745                                 goto skip;
5746                         }
5747                         if (ordered)
5748                                 btrfs_put_ordered_extent(ordered);
5749
5750                         extent_locked = 1;
5751                         continue;
5752                 }
5753
5754                 if (nr_extents == 1) {
5755                         /* update extent pointer in place */
5756                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
5757                                                 new_extents[0].disk_bytenr);
5758                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5759                                                 new_extents[0].disk_num_bytes);
5760                         btrfs_mark_buffer_dirty(leaf);
5761
5762                         btrfs_drop_extent_cache(inode, key.offset,
5763                                                 key.offset + num_bytes - 1, 0);
5764
5765                         ret = btrfs_inc_extent_ref(trans, root,
5766                                                 new_extents[0].disk_bytenr,
5767                                                 new_extents[0].disk_num_bytes,
5768                                                 leaf->start,
5769                                                 root->root_key.objectid,
5770                                                 trans->transid,
5771                                                 key.objectid);
5772                         BUG_ON(ret);
5773
5774                         ret = btrfs_free_extent(trans, root,
5775                                                 extent_key->objectid,
5776                                                 extent_key->offset,
5777                                                 leaf->start,
5778                                                 btrfs_header_owner(leaf),
5779                                                 btrfs_header_generation(leaf),
5780                                                 key.objectid, 0);
5781                         BUG_ON(ret);
5782
5783                         btrfs_release_path(root, path);
5784                         key.offset += num_bytes;
5785                 } else {
5786                         BUG_ON(1);
5787 #if 0
5788                         u64 alloc_hint;
5789                         u64 extent_len;
5790                         int i;
5791                         /*
5792                          * drop old extent pointer at first, then insert the
5793                          * new pointers one bye one
5794                          */
5795                         btrfs_release_path(root, path);
5796                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
5797                                                  key.offset + num_bytes,
5798                                                  key.offset, &alloc_hint);
5799                         BUG_ON(ret);
5800
5801                         for (i = 0; i < nr_extents; i++) {
5802                                 if (ext_offset >= new_extents[i].num_bytes) {
5803                                         ext_offset -= new_extents[i].num_bytes;
5804                                         continue;
5805                                 }
5806                                 extent_len = min(new_extents[i].num_bytes -
5807                                                  ext_offset, num_bytes);
5808
5809                                 ret = btrfs_insert_empty_item(trans, root,
5810                                                               path, &key,
5811                                                               sizeof(*fi));
5812                                 BUG_ON(ret);
5813
5814                                 leaf = path->nodes[0];
5815                                 fi = btrfs_item_ptr(leaf, path->slots[0],
5816                                                 struct btrfs_file_extent_item);
5817                                 btrfs_set_file_extent_generation(leaf, fi,
5818                                                         trans->transid);
5819                                 btrfs_set_file_extent_type(leaf, fi,
5820                                                         BTRFS_FILE_EXTENT_REG);
5821                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5822                                                 new_extents[i].disk_bytenr);
5823                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5824                                                 new_extents[i].disk_num_bytes);
5825                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
5826                                                 new_extents[i].ram_bytes);
5827
5828                                 btrfs_set_file_extent_compression(leaf, fi,
5829                                                 new_extents[i].compression);
5830                                 btrfs_set_file_extent_encryption(leaf, fi,
5831                                                 new_extents[i].encryption);
5832                                 btrfs_set_file_extent_other_encoding(leaf, fi,
5833                                                 new_extents[i].other_encoding);
5834
5835                                 btrfs_set_file_extent_num_bytes(leaf, fi,
5836                                                         extent_len);
5837                                 ext_offset += new_extents[i].offset;
5838                                 btrfs_set_file_extent_offset(leaf, fi,
5839                                                         ext_offset);
5840                                 btrfs_mark_buffer_dirty(leaf);
5841
5842                                 btrfs_drop_extent_cache(inode, key.offset,
5843                                                 key.offset + extent_len - 1, 0);
5844
5845                                 ret = btrfs_inc_extent_ref(trans, root,
5846                                                 new_extents[i].disk_bytenr,
5847                                                 new_extents[i].disk_num_bytes,
5848                                                 leaf->start,
5849                                                 root->root_key.objectid,
5850                                                 trans->transid, key.objectid);
5851                                 BUG_ON(ret);
5852                                 btrfs_release_path(root, path);
5853
5854                                 inode_add_bytes(inode, extent_len);
5855
5856                                 ext_offset = 0;
5857                                 num_bytes -= extent_len;
5858                                 key.offset += extent_len;
5859
5860                                 if (num_bytes == 0)
5861                                         break;
5862                         }
5863                         BUG_ON(i >= nr_extents);
5864 #endif
5865                 }
5866
5867                 if (extent_locked) {
5868                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5869                                       lock_end, GFP_NOFS);
5870                         extent_locked = 0;
5871                 }
5872 skip:
5873                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5874                     key.offset >= search_end)
5875                         break;
5876
5877                 cond_resched();
5878         }
5879         ret = 0;
5880 out:
5881         btrfs_release_path(root, path);
5882         if (inode) {
5883                 mutex_unlock(&inode->i_mutex);
5884                 if (extent_locked) {
5885                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5886                                       lock_end, GFP_NOFS);
5887                 }
5888                 iput(inode);
5889         }
5890         return ret;
5891 }
5892
5893 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5894                                struct btrfs_root *root,
5895                                struct extent_buffer *buf, u64 orig_start)
5896 {
5897         int level;
5898         int ret;
5899
5900         BUG_ON(btrfs_header_generation(buf) != trans->transid);
5901         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5902
5903         level = btrfs_header_level(buf);
5904         if (level == 0) {
5905                 struct btrfs_leaf_ref *ref;
5906                 struct btrfs_leaf_ref *orig_ref;
5907
5908                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5909                 if (!orig_ref)
5910                         return -ENOENT;
5911
5912                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5913                 if (!ref) {
5914                         btrfs_free_leaf_ref(root, orig_ref);
5915                         return -ENOMEM;
5916                 }
5917
5918                 ref->nritems = orig_ref->nritems;
5919                 memcpy(ref->extents, orig_ref->extents,
5920                         sizeof(ref->extents[0]) * ref->nritems);
5921
5922                 btrfs_free_leaf_ref(root, orig_ref);
5923
5924                 ref->root_gen = trans->transid;
5925                 ref->bytenr = buf->start;
5926                 ref->owner = btrfs_header_owner(buf);
5927                 ref->generation = btrfs_header_generation(buf);
5928
5929                 ret = btrfs_add_leaf_ref(root, ref, 0);
5930                 WARN_ON(ret);
5931                 btrfs_free_leaf_ref(root, ref);
5932         }
5933         return 0;
5934 }
5935
5936 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5937                                         struct extent_buffer *leaf,
5938                                         struct btrfs_block_group_cache *group,
5939                                         struct btrfs_root *target_root)
5940 {
5941         struct btrfs_key key;
5942         struct inode *inode = NULL;
5943         struct btrfs_file_extent_item *fi;
5944         u64 num_bytes;
5945         u64 skip_objectid = 0;
5946         u32 nritems;
5947         u32 i;
5948
5949         nritems = btrfs_header_nritems(leaf);
5950         for (i = 0; i < nritems; i++) {
5951                 btrfs_item_key_to_cpu(leaf, &key, i);
5952                 if (key.objectid == skip_objectid ||
5953                     key.type != BTRFS_EXTENT_DATA_KEY)
5954                         continue;
5955                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5956                 if (btrfs_file_extent_type(leaf, fi) ==
5957                     BTRFS_FILE_EXTENT_INLINE)
5958                         continue;
5959                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5960                         continue;
5961                 if (!inode || inode->i_ino != key.objectid) {
5962                         iput(inode);
5963                         inode = btrfs_ilookup(target_root->fs_info->sb,
5964                                               key.objectid, target_root, 1);
5965                 }
5966                 if (!inode) {
5967                         skip_objectid = key.objectid;
5968                         continue;
5969                 }
5970                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5971
5972                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5973                             key.offset + num_bytes - 1, GFP_NOFS);
5974                 btrfs_drop_extent_cache(inode, key.offset,
5975                                         key.offset + num_bytes - 1, 1);
5976                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5977                               key.offset + num_bytes - 1, GFP_NOFS);
5978                 cond_resched();
5979         }
5980         iput(inode);
5981         return 0;
5982 }
5983
5984 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5985                                         struct btrfs_root *root,
5986                                         struct extent_buffer *leaf,
5987                                         struct btrfs_block_group_cache *group,
5988                                         struct inode *reloc_inode)
5989 {
5990         struct btrfs_key key;
5991         struct btrfs_key extent_key;
5992         struct btrfs_file_extent_item *fi;
5993         struct btrfs_leaf_ref *ref;
5994         struct disk_extent *new_extent;
5995         u64 bytenr;
5996         u64 num_bytes;
5997         u32 nritems;
5998         u32 i;
5999         int ext_index;
6000         int nr_extent;
6001         int ret;
6002
6003         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6004         BUG_ON(!new_extent);
6005
6006         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6007         BUG_ON(!ref);
6008
6009         ext_index = -1;
6010         nritems = btrfs_header_nritems(leaf);
6011         for (i = 0; i < nritems; i++) {
6012                 btrfs_item_key_to_cpu(leaf, &key, i);
6013                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6014                         continue;
6015                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6016                 if (btrfs_file_extent_type(leaf, fi) ==
6017                     BTRFS_FILE_EXTENT_INLINE)
6018                         continue;
6019                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6020                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6021                 if (bytenr == 0)
6022                         continue;
6023
6024                 ext_index++;
6025                 if (bytenr >= group->key.objectid + group->key.offset ||
6026                     bytenr + num_bytes <= group->key.objectid)
6027                         continue;
6028
6029                 extent_key.objectid = bytenr;
6030                 extent_key.offset = num_bytes;
6031                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6032                 nr_extent = 1;
6033                 ret = get_new_locations(reloc_inode, &extent_key,
6034                                         group->key.objectid, 1,
6035                                         &new_extent, &nr_extent);
6036                 if (ret > 0)
6037                         continue;
6038                 BUG_ON(ret < 0);
6039
6040                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6041                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6042                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6043                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6044
6045                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6046                                                 new_extent->disk_bytenr);
6047                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6048                                                 new_extent->disk_num_bytes);
6049                 btrfs_mark_buffer_dirty(leaf);
6050
6051                 ret = btrfs_inc_extent_ref(trans, root,
6052                                         new_extent->disk_bytenr,
6053                                         new_extent->disk_num_bytes,
6054                                         leaf->start,
6055                                         root->root_key.objectid,
6056                                         trans->transid, key.objectid);
6057                 BUG_ON(ret);
6058
6059                 ret = btrfs_free_extent(trans, root,
6060                                         bytenr, num_bytes, leaf->start,
6061                                         btrfs_header_owner(leaf),
6062                                         btrfs_header_generation(leaf),
6063                                         key.objectid, 0);
6064                 BUG_ON(ret);
6065                 cond_resched();
6066         }
6067         kfree(new_extent);
6068         BUG_ON(ext_index + 1 != ref->nritems);
6069         btrfs_free_leaf_ref(root, ref);
6070         return 0;
6071 }
6072
6073 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6074                           struct btrfs_root *root)
6075 {
6076         struct btrfs_root *reloc_root;
6077         int ret;
6078
6079         if (root->reloc_root) {
6080                 reloc_root = root->reloc_root;
6081                 root->reloc_root = NULL;
6082                 list_add(&reloc_root->dead_list,
6083                          &root->fs_info->dead_reloc_roots);
6084
6085                 btrfs_set_root_bytenr(&reloc_root->root_item,
6086                                       reloc_root->node->start);
6087                 btrfs_set_root_level(&root->root_item,
6088                                      btrfs_header_level(reloc_root->node));
6089                 memset(&reloc_root->root_item.drop_progress, 0,
6090                         sizeof(struct btrfs_disk_key));
6091                 reloc_root->root_item.drop_level = 0;
6092
6093                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6094                                         &reloc_root->root_key,
6095                                         &reloc_root->root_item);
6096                 BUG_ON(ret);
6097         }
6098         return 0;
6099 }
6100
6101 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6102 {
6103         struct btrfs_trans_handle *trans;
6104         struct btrfs_root *reloc_root;
6105         struct btrfs_root *prev_root = NULL;
6106         struct list_head dead_roots;
6107         int ret;
6108         unsigned long nr;
6109
6110         INIT_LIST_HEAD(&dead_roots);
6111         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6112
6113         while (!list_empty(&dead_roots)) {
6114                 reloc_root = list_entry(dead_roots.prev,
6115                                         struct btrfs_root, dead_list);
6116                 list_del_init(&reloc_root->dead_list);
6117
6118                 BUG_ON(reloc_root->commit_root != NULL);
6119                 while (1) {
6120                         trans = btrfs_join_transaction(root, 1);
6121                         BUG_ON(!trans);
6122
6123                         mutex_lock(&root->fs_info->drop_mutex);
6124                         ret = btrfs_drop_snapshot(trans, reloc_root);
6125                         if (ret != -EAGAIN)
6126                                 break;
6127                         mutex_unlock(&root->fs_info->drop_mutex);
6128
6129                         nr = trans->blocks_used;
6130                         ret = btrfs_end_transaction(trans, root);
6131                         BUG_ON(ret);
6132                         btrfs_btree_balance_dirty(root, nr);
6133                 }
6134
6135                 free_extent_buffer(reloc_root->node);
6136
6137                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6138                                      &reloc_root->root_key);
6139                 BUG_ON(ret);
6140                 mutex_unlock(&root->fs_info->drop_mutex);
6141
6142                 nr = trans->blocks_used;
6143                 ret = btrfs_end_transaction(trans, root);
6144                 BUG_ON(ret);
6145                 btrfs_btree_balance_dirty(root, nr);
6146
6147                 kfree(prev_root);
6148                 prev_root = reloc_root;
6149         }
6150         if (prev_root) {
6151                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6152                 kfree(prev_root);
6153         }
6154         return 0;
6155 }
6156
6157 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6158 {
6159         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6160         return 0;
6161 }
6162
6163 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6164 {
6165         struct btrfs_root *reloc_root;
6166         struct btrfs_trans_handle *trans;
6167         struct btrfs_key location;
6168         int found;
6169         int ret;
6170
6171         mutex_lock(&root->fs_info->tree_reloc_mutex);
6172         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6173         BUG_ON(ret);
6174         found = !list_empty(&root->fs_info->dead_reloc_roots);
6175         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6176
6177         if (found) {
6178                 trans = btrfs_start_transaction(root, 1);
6179                 BUG_ON(!trans);
6180                 ret = btrfs_commit_transaction(trans, root);
6181                 BUG_ON(ret);
6182         }
6183
6184         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6185         location.offset = (u64)-1;
6186         location.type = BTRFS_ROOT_ITEM_KEY;
6187
6188         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6189         BUG_ON(!reloc_root);
6190         btrfs_orphan_cleanup(reloc_root);
6191         return 0;
6192 }
6193
6194 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6195                                     struct btrfs_root *root)
6196 {
6197         struct btrfs_root *reloc_root;
6198         struct extent_buffer *eb;
6199         struct btrfs_root_item *root_item;
6200         struct btrfs_key root_key;
6201         int ret;
6202
6203         BUG_ON(!root->ref_cows);
6204         if (root->reloc_root)
6205                 return 0;
6206
6207         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6208         BUG_ON(!root_item);
6209
6210         ret = btrfs_copy_root(trans, root, root->commit_root,
6211                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6212         BUG_ON(ret);
6213
6214         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6215         root_key.offset = root->root_key.objectid;
6216         root_key.type = BTRFS_ROOT_ITEM_KEY;
6217
6218         memcpy(root_item, &root->root_item, sizeof(root_item));
6219         btrfs_set_root_refs(root_item, 0);
6220         btrfs_set_root_bytenr(root_item, eb->start);
6221         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6222         btrfs_set_root_generation(root_item, trans->transid);
6223
6224         btrfs_tree_unlock(eb);
6225         free_extent_buffer(eb);
6226
6227         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6228                                 &root_key, root_item);
6229         BUG_ON(ret);
6230         kfree(root_item);
6231
6232         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6233                                                  &root_key);
6234         BUG_ON(!reloc_root);
6235         reloc_root->last_trans = trans->transid;
6236         reloc_root->commit_root = NULL;
6237         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6238
6239         root->reloc_root = reloc_root;
6240         return 0;
6241 }
6242
6243 /*
6244  * Core function of space balance.
6245  *
6246  * The idea is using reloc trees to relocate tree blocks in reference
6247  * counted roots. There is one reloc tree for each subvol, and all
6248  * reloc trees share same root key objectid. Reloc trees are snapshots
6249  * of the latest committed roots of subvols (root->commit_root).
6250  *
6251  * To relocate a tree block referenced by a subvol, there are two steps.
6252  * COW the block through subvol's reloc tree, then update block pointer
6253  * in the subvol to point to the new block. Since all reloc trees share
6254  * same root key objectid, doing special handing for tree blocks owned
6255  * by them is easy. Once a tree block has been COWed in one reloc tree,
6256  * we can use the resulting new block directly when the same block is
6257  * required to COW again through other reloc trees. By this way, relocated
6258  * tree blocks are shared between reloc trees, so they are also shared
6259  * between subvols.
6260  */
6261 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6262                                       struct btrfs_root *root,
6263                                       struct btrfs_path *path,
6264                                       struct btrfs_key *first_key,
6265                                       struct btrfs_ref_path *ref_path,
6266                                       struct btrfs_block_group_cache *group,
6267                                       struct inode *reloc_inode)
6268 {
6269         struct btrfs_root *reloc_root;
6270         struct extent_buffer *eb = NULL;
6271         struct btrfs_key *keys;
6272         u64 *nodes;
6273         int level;
6274         int shared_level;
6275         int lowest_level = 0;
6276         int ret;
6277
6278         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6279                 lowest_level = ref_path->owner_objectid;
6280
6281         if (!root->ref_cows) {
6282                 path->lowest_level = lowest_level;
6283                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6284                 BUG_ON(ret < 0);
6285                 path->lowest_level = 0;
6286                 btrfs_release_path(root, path);
6287                 return 0;
6288         }
6289
6290         mutex_lock(&root->fs_info->tree_reloc_mutex);
6291         ret = init_reloc_tree(trans, root);
6292         BUG_ON(ret);
6293         reloc_root = root->reloc_root;
6294
6295         shared_level = ref_path->shared_level;
6296         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6297
6298         keys = ref_path->node_keys;
6299         nodes = ref_path->new_nodes;
6300         memset(&keys[shared_level + 1], 0,
6301                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6302         memset(&nodes[shared_level + 1], 0,
6303                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6304
6305         if (nodes[lowest_level] == 0) {
6306                 path->lowest_level = lowest_level;
6307                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6308                                         0, 1);
6309                 BUG_ON(ret);
6310                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6311                         eb = path->nodes[level];
6312                         if (!eb || eb == reloc_root->node)
6313                                 break;
6314                         nodes[level] = eb->start;
6315                         if (level == 0)
6316                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6317                         else
6318                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6319                 }
6320                 if (nodes[0] &&
6321                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6322                         eb = path->nodes[0];
6323                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6324                                                       group, reloc_inode);
6325                         BUG_ON(ret);
6326                 }
6327                 btrfs_release_path(reloc_root, path);
6328         } else {
6329                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6330                                        lowest_level);
6331                 BUG_ON(ret);
6332         }
6333
6334         /*
6335          * replace tree blocks in the fs tree with tree blocks in
6336          * the reloc tree.
6337          */
6338         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6339         BUG_ON(ret < 0);
6340
6341         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6342                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6343                                         0, 0);
6344                 BUG_ON(ret);
6345                 extent_buffer_get(path->nodes[0]);
6346                 eb = path->nodes[0];
6347                 btrfs_release_path(reloc_root, path);
6348                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6349                 BUG_ON(ret);
6350                 free_extent_buffer(eb);
6351         }
6352
6353         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6354         path->lowest_level = 0;
6355         return 0;
6356 }
6357
6358 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6359                                         struct btrfs_root *root,
6360                                         struct btrfs_path *path,
6361                                         struct btrfs_key *first_key,
6362                                         struct btrfs_ref_path *ref_path)
6363 {
6364         int ret;
6365
6366         ret = relocate_one_path(trans, root, path, first_key,
6367                                 ref_path, NULL, NULL);
6368         BUG_ON(ret);
6369
6370         return 0;
6371 }
6372
6373 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6374                                     struct btrfs_root *extent_root,
6375                                     struct btrfs_path *path,
6376                                     struct btrfs_key *extent_key)
6377 {
6378         int ret;
6379
6380         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6381         if (ret)
6382                 goto out;
6383         ret = btrfs_del_item(trans, extent_root, path);
6384 out:
6385         btrfs_release_path(extent_root, path);
6386         return ret;
6387 }
6388
6389 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6390                                                 struct btrfs_ref_path *ref_path)
6391 {
6392         struct btrfs_key root_key;
6393
6394         root_key.objectid = ref_path->root_objectid;
6395         root_key.type = BTRFS_ROOT_ITEM_KEY;
6396         if (is_cowonly_root(ref_path->root_objectid))
6397                 root_key.offset = 0;
6398         else
6399                 root_key.offset = (u64)-1;
6400
6401         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6402 }
6403
6404 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6405                                         struct btrfs_path *path,
6406                                         struct btrfs_key *extent_key,
6407                                         struct btrfs_block_group_cache *group,
6408                                         struct inode *reloc_inode, int pass)
6409 {
6410         struct btrfs_trans_handle *trans;
6411         struct btrfs_root *found_root;
6412         struct btrfs_ref_path *ref_path = NULL;
6413         struct disk_extent *new_extents = NULL;
6414         int nr_extents = 0;
6415         int loops;
6416         int ret;
6417         int level;
6418         struct btrfs_key first_key;
6419         u64 prev_block = 0;
6420
6421
6422         trans = btrfs_start_transaction(extent_root, 1);
6423         BUG_ON(!trans);
6424
6425         if (extent_key->objectid == 0) {
6426                 ret = del_extent_zero(trans, extent_root, path, extent_key);
6427                 goto out;
6428         }
6429
6430         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6431         if (!ref_path) {
6432                 ret = -ENOMEM;
6433                 goto out;
6434         }
6435
6436         for (loops = 0; ; loops++) {
6437                 if (loops == 0) {
6438                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6439                                                    extent_key->objectid);
6440                 } else {
6441                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6442                 }
6443                 if (ret < 0)
6444                         goto out;
6445                 if (ret > 0)
6446                         break;
6447
6448                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6449                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6450                         continue;
6451
6452                 found_root = read_ref_root(extent_root->fs_info, ref_path);
6453                 BUG_ON(!found_root);
6454                 /*
6455                  * for reference counted tree, only process reference paths
6456                  * rooted at the latest committed root.
6457                  */
6458                 if (found_root->ref_cows &&
6459                     ref_path->root_generation != found_root->root_key.offset)
6460                         continue;
6461
6462                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6463                         if (pass == 0) {
6464                                 /*
6465                                  * copy data extents to new locations
6466                                  */
6467                                 u64 group_start = group->key.objectid;
6468                                 ret = relocate_data_extent(reloc_inode,
6469                                                            extent_key,
6470                                                            group_start);
6471                                 if (ret < 0)
6472                                         goto out;
6473                                 break;
6474                         }
6475                         level = 0;
6476                 } else {
6477                         level = ref_path->owner_objectid;
6478                 }
6479
6480                 if (prev_block != ref_path->nodes[level]) {
6481                         struct extent_buffer *eb;
6482                         u64 block_start = ref_path->nodes[level];
6483                         u64 block_size = btrfs_level_size(found_root, level);
6484
6485                         eb = read_tree_block(found_root, block_start,
6486                                              block_size, 0);
6487                         btrfs_tree_lock(eb);
6488                         BUG_ON(level != btrfs_header_level(eb));
6489
6490                         if (level == 0)
6491                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
6492                         else
6493                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
6494
6495                         btrfs_tree_unlock(eb);
6496                         free_extent_buffer(eb);
6497                         prev_block = block_start;
6498                 }
6499
6500                 mutex_lock(&extent_root->fs_info->trans_mutex);
6501                 btrfs_record_root_in_trans(found_root);
6502                 mutex_unlock(&extent_root->fs_info->trans_mutex);
6503                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6504                         /*
6505                          * try to update data extent references while
6506                          * keeping metadata shared between snapshots.
6507                          */
6508                         if (pass == 1) {
6509                                 ret = relocate_one_path(trans, found_root,
6510                                                 path, &first_key, ref_path,
6511                                                 group, reloc_inode);
6512                                 if (ret < 0)
6513                                         goto out;
6514                                 continue;
6515                         }
6516                         /*
6517                          * use fallback method to process the remaining
6518                          * references.
6519                          */
6520                         if (!new_extents) {
6521                                 u64 group_start = group->key.objectid;
6522                                 new_extents = kmalloc(sizeof(*new_extents),
6523                                                       GFP_NOFS);
6524                                 nr_extents = 1;
6525                                 ret = get_new_locations(reloc_inode,
6526                                                         extent_key,
6527                                                         group_start, 1,
6528                                                         &new_extents,
6529                                                         &nr_extents);
6530                                 if (ret)
6531                                         goto out;
6532                         }
6533                         ret = replace_one_extent(trans, found_root,
6534                                                 path, extent_key,
6535                                                 &first_key, ref_path,
6536                                                 new_extents, nr_extents);
6537                 } else {
6538                         ret = relocate_tree_block(trans, found_root, path,
6539                                                   &first_key, ref_path);
6540                 }
6541                 if (ret < 0)
6542                         goto out;
6543         }
6544         ret = 0;
6545 out:
6546         btrfs_end_transaction(trans, extent_root);
6547         kfree(new_extents);
6548         kfree(ref_path);
6549         return ret;
6550 }
6551 #endif
6552
6553 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6554 {
6555         u64 num_devices;
6556         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6557                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6558
6559         num_devices = root->fs_info->fs_devices->rw_devices;
6560         if (num_devices == 1) {
6561                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6562                 stripped = flags & ~stripped;
6563
6564                 /* turn raid0 into single device chunks */
6565                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6566                         return stripped;
6567
6568                 /* turn mirroring into duplication */
6569                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6570                              BTRFS_BLOCK_GROUP_RAID10))
6571                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6572                 return flags;
6573         } else {
6574                 /* they already had raid on here, just return */
6575                 if (flags & stripped)
6576                         return flags;
6577
6578                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6579                 stripped = flags & ~stripped;
6580
6581                 /* switch duplicated blocks with raid1 */
6582                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6583                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6584
6585                 /* turn single device chunks into raid0 */
6586                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6587         }
6588         return flags;
6589 }
6590
6591 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6592                      struct btrfs_block_group_cache *shrink_block_group,
6593                      int force)
6594 {
6595         struct btrfs_trans_handle *trans;
6596         u64 new_alloc_flags;
6597         u64 calc;
6598
6599         spin_lock(&shrink_block_group->lock);
6600         if (btrfs_block_group_used(&shrink_block_group->item) +
6601             shrink_block_group->reserved > 0) {
6602                 spin_unlock(&shrink_block_group->lock);
6603
6604                 trans = btrfs_start_transaction(root, 1);
6605                 spin_lock(&shrink_block_group->lock);
6606
6607                 new_alloc_flags = update_block_group_flags(root,
6608                                                    shrink_block_group->flags);
6609                 if (new_alloc_flags != shrink_block_group->flags) {
6610                         calc =
6611                              btrfs_block_group_used(&shrink_block_group->item);
6612                 } else {
6613                         calc = shrink_block_group->key.offset;
6614                 }
6615                 spin_unlock(&shrink_block_group->lock);
6616
6617                 do_chunk_alloc(trans, root->fs_info->extent_root,
6618                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
6619
6620                 btrfs_end_transaction(trans, root);
6621         } else
6622                 spin_unlock(&shrink_block_group->lock);
6623         return 0;
6624 }
6625
6626
6627 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6628                                          struct btrfs_block_group_cache *group)
6629
6630 {
6631         __alloc_chunk_for_shrink(root, group, 1);
6632         set_block_group_readonly(group);
6633         return 0;
6634 }
6635
6636 #if 0
6637 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6638                                  struct btrfs_root *root,
6639                                  u64 objectid, u64 size)
6640 {
6641         struct btrfs_path *path;
6642         struct btrfs_inode_item *item;
6643         struct extent_buffer *leaf;
6644         int ret;
6645
6646         path = btrfs_alloc_path();
6647         if (!path)
6648                 return -ENOMEM;
6649
6650         path->leave_spinning = 1;
6651         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6652         if (ret)
6653                 goto out;
6654
6655         leaf = path->nodes[0];
6656         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6657         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6658         btrfs_set_inode_generation(leaf, item, 1);
6659         btrfs_set_inode_size(leaf, item, size);
6660         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6661         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6662         btrfs_mark_buffer_dirty(leaf);
6663         btrfs_release_path(root, path);
6664 out:
6665         btrfs_free_path(path);
6666         return ret;
6667 }
6668
6669 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6670                                         struct btrfs_block_group_cache *group)
6671 {
6672         struct inode *inode = NULL;
6673         struct btrfs_trans_handle *trans;
6674         struct btrfs_root *root;
6675         struct btrfs_key root_key;
6676         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6677         int err = 0;
6678
6679         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6680         root_key.type = BTRFS_ROOT_ITEM_KEY;
6681         root_key.offset = (u64)-1;
6682         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6683         if (IS_ERR(root))
6684                 return ERR_CAST(root);
6685
6686         trans = btrfs_start_transaction(root, 1);
6687         BUG_ON(!trans);
6688
6689         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6690         if (err)
6691                 goto out;
6692
6693         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6694         BUG_ON(err);
6695
6696         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6697                                        group->key.offset, 0, group->key.offset,
6698                                        0, 0, 0);
6699         BUG_ON(err);
6700
6701         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6702         if (inode->i_state & I_NEW) {
6703                 BTRFS_I(inode)->root = root;
6704                 BTRFS_I(inode)->location.objectid = objectid;
6705                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6706                 BTRFS_I(inode)->location.offset = 0;
6707                 btrfs_read_locked_inode(inode);
6708                 unlock_new_inode(inode);
6709                 BUG_ON(is_bad_inode(inode));
6710         } else {
6711                 BUG_ON(1);
6712         }
6713         BTRFS_I(inode)->index_cnt = group->key.objectid;
6714
6715         err = btrfs_orphan_add(trans, inode);
6716 out:
6717         btrfs_end_transaction(trans, root);
6718         if (err) {
6719                 if (inode)
6720                         iput(inode);
6721                 inode = ERR_PTR(err);
6722         }
6723         return inode;
6724 }
6725
6726 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6727 {
6728
6729         struct btrfs_ordered_sum *sums;
6730         struct btrfs_sector_sum *sector_sum;
6731         struct btrfs_ordered_extent *ordered;
6732         struct btrfs_root *root = BTRFS_I(inode)->root;
6733         struct list_head list;
6734         size_t offset;
6735         int ret;
6736         u64 disk_bytenr;
6737
6738         INIT_LIST_HEAD(&list);
6739
6740         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6741         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6742
6743         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6744         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6745                                        disk_bytenr + len - 1, &list);
6746
6747         while (!list_empty(&list)) {
6748                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6749                 list_del_init(&sums->list);
6750
6751                 sector_sum = sums->sums;
6752                 sums->bytenr = ordered->start;
6753
6754                 offset = 0;
6755                 while (offset < sums->len) {
6756                         sector_sum->bytenr += ordered->start - disk_bytenr;
6757                         sector_sum++;
6758                         offset += root->sectorsize;
6759                 }
6760
6761                 btrfs_add_ordered_sum(inode, ordered, sums);
6762         }
6763         btrfs_put_ordered_extent(ordered);
6764         return 0;
6765 }
6766
6767 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6768 {
6769         struct btrfs_trans_handle *trans;
6770         struct btrfs_path *path;
6771         struct btrfs_fs_info *info = root->fs_info;
6772         struct extent_buffer *leaf;
6773         struct inode *reloc_inode;
6774         struct btrfs_block_group_cache *block_group;
6775         struct btrfs_key key;
6776         u64 skipped;
6777         u64 cur_byte;
6778         u64 total_found;
6779         u32 nritems;
6780         int ret;
6781         int progress;
6782         int pass = 0;
6783
6784         root = root->fs_info->extent_root;
6785
6786         block_group = btrfs_lookup_block_group(info, group_start);
6787         BUG_ON(!block_group);
6788
6789         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6790                (unsigned long long)block_group->key.objectid,
6791                (unsigned long long)block_group->flags);
6792
6793         path = btrfs_alloc_path();
6794         BUG_ON(!path);
6795
6796         reloc_inode = create_reloc_inode(info, block_group);
6797         BUG_ON(IS_ERR(reloc_inode));
6798
6799         __alloc_chunk_for_shrink(root, block_group, 1);
6800         set_block_group_readonly(block_group);
6801
6802         btrfs_start_delalloc_inodes(info->tree_root);
6803         btrfs_wait_ordered_extents(info->tree_root, 0);
6804 again:
6805         skipped = 0;
6806         total_found = 0;
6807         progress = 0;
6808         key.objectid = block_group->key.objectid;
6809         key.offset = 0;
6810         key.type = 0;
6811         cur_byte = key.objectid;
6812
6813         trans = btrfs_start_transaction(info->tree_root, 1);
6814         btrfs_commit_transaction(trans, info->tree_root);
6815
6816         mutex_lock(&root->fs_info->cleaner_mutex);
6817         btrfs_clean_old_snapshots(info->tree_root);
6818         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6819         mutex_unlock(&root->fs_info->cleaner_mutex);
6820
6821         trans = btrfs_start_transaction(info->tree_root, 1);
6822         btrfs_commit_transaction(trans, info->tree_root);
6823
6824         while (1) {
6825                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6826                 if (ret < 0)
6827                         goto out;
6828 next:
6829                 leaf = path->nodes[0];
6830                 nritems = btrfs_header_nritems(leaf);
6831                 if (path->slots[0] >= nritems) {
6832                         ret = btrfs_next_leaf(root, path);
6833                         if (ret < 0)
6834                                 goto out;
6835                         if (ret == 1) {
6836                                 ret = 0;
6837                                 break;
6838                         }
6839                         leaf = path->nodes[0];
6840                         nritems = btrfs_header_nritems(leaf);
6841                 }
6842
6843                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6844
6845                 if (key.objectid >= block_group->key.objectid +
6846                     block_group->key.offset)
6847                         break;
6848
6849                 if (progress && need_resched()) {
6850                         btrfs_release_path(root, path);
6851                         cond_resched();
6852                         progress = 0;
6853                         continue;
6854                 }
6855                 progress = 1;
6856
6857                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
6858                     key.objectid + key.offset <= cur_byte) {
6859                         path->slots[0]++;
6860                         goto next;
6861                 }
6862
6863                 total_found++;
6864                 cur_byte = key.objectid + key.offset;
6865                 btrfs_release_path(root, path);
6866
6867                 __alloc_chunk_for_shrink(root, block_group, 0);
6868                 ret = relocate_one_extent(root, path, &key, block_group,
6869                                           reloc_inode, pass);
6870                 BUG_ON(ret < 0);
6871                 if (ret > 0)
6872                         skipped++;
6873
6874                 key.objectid = cur_byte;
6875                 key.type = 0;
6876                 key.offset = 0;
6877         }
6878
6879         btrfs_release_path(root, path);
6880
6881         if (pass == 0) {
6882                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6883                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6884         }
6885
6886         if (total_found > 0) {
6887                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6888                        (unsigned long long)total_found, pass);
6889                 pass++;
6890                 if (total_found == skipped && pass > 2) {
6891                         iput(reloc_inode);
6892                         reloc_inode = create_reloc_inode(info, block_group);
6893                         pass = 0;
6894                 }
6895                 goto again;
6896         }
6897
6898         /* delete reloc_inode */
6899         iput(reloc_inode);
6900
6901         /* unpin extents in this range */
6902         trans = btrfs_start_transaction(info->tree_root, 1);
6903         btrfs_commit_transaction(trans, info->tree_root);
6904
6905         spin_lock(&block_group->lock);
6906         WARN_ON(block_group->pinned > 0);
6907         WARN_ON(block_group->reserved > 0);
6908         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6909         spin_unlock(&block_group->lock);
6910         btrfs_put_block_group(block_group);
6911         ret = 0;
6912 out:
6913         btrfs_free_path(path);
6914         return ret;
6915 }
6916 #endif
6917
6918 static int find_first_block_group(struct btrfs_root *root,
6919                 struct btrfs_path *path, struct btrfs_key *key)
6920 {
6921         int ret = 0;
6922         struct btrfs_key found_key;
6923         struct extent_buffer *leaf;
6924         int slot;
6925
6926         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6927         if (ret < 0)
6928                 goto out;
6929
6930         while (1) {
6931                 slot = path->slots[0];
6932                 leaf = path->nodes[0];
6933                 if (slot >= btrfs_header_nritems(leaf)) {
6934                         ret = btrfs_next_leaf(root, path);
6935                         if (ret == 0)
6936                                 continue;
6937                         if (ret < 0)
6938                                 goto out;
6939                         break;
6940                 }
6941                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6942
6943                 if (found_key.objectid >= key->objectid &&
6944                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6945                         ret = 0;
6946                         goto out;
6947                 }
6948                 path->slots[0]++;
6949         }
6950         ret = -ENOENT;
6951 out:
6952         return ret;
6953 }
6954
6955 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6956 {
6957         struct btrfs_block_group_cache *block_group;
6958         struct btrfs_space_info *space_info;
6959         struct rb_node *n;
6960
6961         spin_lock(&info->block_group_cache_lock);
6962         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6963                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6964                                        cache_node);
6965                 rb_erase(&block_group->cache_node,
6966                          &info->block_group_cache_tree);
6967                 spin_unlock(&info->block_group_cache_lock);
6968
6969                 btrfs_remove_free_space_cache(block_group);
6970                 down_write(&block_group->space_info->groups_sem);
6971                 list_del(&block_group->list);
6972                 up_write(&block_group->space_info->groups_sem);
6973
6974                 WARN_ON(atomic_read(&block_group->count) != 1);
6975                 kfree(block_group);
6976
6977                 spin_lock(&info->block_group_cache_lock);
6978         }
6979         spin_unlock(&info->block_group_cache_lock);
6980
6981         /* now that all the block groups are freed, go through and
6982          * free all the space_info structs.  This is only called during
6983          * the final stages of unmount, and so we know nobody is
6984          * using them.  We call synchronize_rcu() once before we start,
6985          * just to be on the safe side.
6986          */
6987         synchronize_rcu();
6988
6989         while(!list_empty(&info->space_info)) {
6990                 space_info = list_entry(info->space_info.next,
6991                                         struct btrfs_space_info,
6992                                         list);
6993
6994                 list_del(&space_info->list);
6995                 kfree(space_info);
6996         }
6997         return 0;
6998 }
6999
7000 int btrfs_read_block_groups(struct btrfs_root *root)
7001 {
7002         struct btrfs_path *path;
7003         int ret;
7004         struct btrfs_block_group_cache *cache;
7005         struct btrfs_fs_info *info = root->fs_info;
7006         struct btrfs_space_info *space_info;
7007         struct btrfs_key key;
7008         struct btrfs_key found_key;
7009         struct extent_buffer *leaf;
7010
7011         root = info->extent_root;
7012         key.objectid = 0;
7013         key.offset = 0;
7014         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7015         path = btrfs_alloc_path();
7016         if (!path)
7017                 return -ENOMEM;
7018
7019         while (1) {
7020                 ret = find_first_block_group(root, path, &key);
7021                 if (ret > 0) {
7022                         ret = 0;
7023                         goto error;
7024                 }
7025                 if (ret != 0)
7026                         goto error;
7027
7028                 leaf = path->nodes[0];
7029                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7030                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7031                 if (!cache) {
7032                         ret = -ENOMEM;
7033                         break;
7034                 }
7035
7036                 atomic_set(&cache->count, 1);
7037                 spin_lock_init(&cache->lock);
7038                 spin_lock_init(&cache->tree_lock);
7039                 mutex_init(&cache->cache_mutex);
7040                 INIT_LIST_HEAD(&cache->list);
7041                 INIT_LIST_HEAD(&cache->cluster_list);
7042                 cache->sectorsize = root->sectorsize;
7043
7044                 /*
7045                  * we only want to have 32k of ram per block group for keeping
7046                  * track of free space, and if we pass 1/2 of that we want to
7047                  * start converting things over to using bitmaps
7048                  */
7049                 cache->extents_thresh = ((1024 * 32) / 2) /
7050                         sizeof(struct btrfs_free_space);
7051
7052                 read_extent_buffer(leaf, &cache->item,
7053                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7054                                    sizeof(cache->item));
7055                 memcpy(&cache->key, &found_key, sizeof(found_key));
7056
7057                 key.objectid = found_key.objectid + found_key.offset;
7058                 btrfs_release_path(root, path);
7059                 cache->flags = btrfs_block_group_flags(&cache->item);
7060
7061                 ret = update_space_info(info, cache->flags, found_key.offset,
7062                                         btrfs_block_group_used(&cache->item),
7063                                         &space_info);
7064                 BUG_ON(ret);
7065                 cache->space_info = space_info;
7066                 down_write(&space_info->groups_sem);
7067                 list_add_tail(&cache->list, &space_info->block_groups);
7068                 up_write(&space_info->groups_sem);
7069
7070                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7071                 BUG_ON(ret);
7072
7073                 set_avail_alloc_bits(root->fs_info, cache->flags);
7074                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7075                         set_block_group_readonly(cache);
7076         }
7077         ret = 0;
7078 error:
7079         btrfs_free_path(path);
7080         return ret;
7081 }
7082
7083 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7084                            struct btrfs_root *root, u64 bytes_used,
7085                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7086                            u64 size)
7087 {
7088         int ret;
7089         struct btrfs_root *extent_root;
7090         struct btrfs_block_group_cache *cache;
7091
7092         extent_root = root->fs_info->extent_root;
7093
7094         root->fs_info->last_trans_log_full_commit = trans->transid;
7095
7096         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7097         if (!cache)
7098                 return -ENOMEM;
7099
7100         cache->key.objectid = chunk_offset;
7101         cache->key.offset = size;
7102         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7103         cache->sectorsize = root->sectorsize;
7104
7105         /*
7106          * we only want to have 32k of ram per block group for keeping track
7107          * of free space, and if we pass 1/2 of that we want to start
7108          * converting things over to using bitmaps
7109          */
7110         cache->extents_thresh = ((1024 * 32) / 2) /
7111                 sizeof(struct btrfs_free_space);
7112         atomic_set(&cache->count, 1);
7113         spin_lock_init(&cache->lock);
7114         spin_lock_init(&cache->tree_lock);
7115         mutex_init(&cache->cache_mutex);
7116         INIT_LIST_HEAD(&cache->list);
7117         INIT_LIST_HEAD(&cache->cluster_list);
7118
7119         btrfs_set_block_group_used(&cache->item, bytes_used);
7120         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7121         cache->flags = type;
7122         btrfs_set_block_group_flags(&cache->item, type);
7123
7124         cache->cached = 1;
7125         ret = btrfs_add_free_space(cache, chunk_offset, size);
7126         BUG_ON(ret);
7127         remove_sb_from_cache(root, cache);
7128
7129         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7130                                 &cache->space_info);
7131         BUG_ON(ret);
7132         down_write(&cache->space_info->groups_sem);
7133         list_add_tail(&cache->list, &cache->space_info->block_groups);
7134         up_write(&cache->space_info->groups_sem);
7135
7136         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7137         BUG_ON(ret);
7138
7139         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7140                                 sizeof(cache->item));
7141         BUG_ON(ret);
7142
7143         set_avail_alloc_bits(extent_root->fs_info, type);
7144
7145         return 0;
7146 }
7147
7148 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7149                              struct btrfs_root *root, u64 group_start)
7150 {
7151         struct btrfs_path *path;
7152         struct btrfs_block_group_cache *block_group;
7153         struct btrfs_free_cluster *cluster;
7154         struct btrfs_key key;
7155         int ret;
7156
7157         root = root->fs_info->extent_root;
7158
7159         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7160         BUG_ON(!block_group);
7161         BUG_ON(!block_group->ro);
7162
7163         memcpy(&key, &block_group->key, sizeof(key));
7164
7165         /* make sure this block group isn't part of an allocation cluster */
7166         cluster = &root->fs_info->data_alloc_cluster;
7167         spin_lock(&cluster->refill_lock);
7168         btrfs_return_cluster_to_free_space(block_group, cluster);
7169         spin_unlock(&cluster->refill_lock);
7170
7171         /*
7172          * make sure this block group isn't part of a metadata
7173          * allocation cluster
7174          */
7175         cluster = &root->fs_info->meta_alloc_cluster;
7176         spin_lock(&cluster->refill_lock);
7177         btrfs_return_cluster_to_free_space(block_group, cluster);
7178         spin_unlock(&cluster->refill_lock);
7179
7180         path = btrfs_alloc_path();
7181         BUG_ON(!path);
7182
7183         spin_lock(&root->fs_info->block_group_cache_lock);
7184         rb_erase(&block_group->cache_node,
7185                  &root->fs_info->block_group_cache_tree);
7186         spin_unlock(&root->fs_info->block_group_cache_lock);
7187         btrfs_remove_free_space_cache(block_group);
7188         down_write(&block_group->space_info->groups_sem);
7189         /*
7190          * we must use list_del_init so people can check to see if they
7191          * are still on the list after taking the semaphore
7192          */
7193         list_del_init(&block_group->list);
7194         up_write(&block_group->space_info->groups_sem);
7195
7196         spin_lock(&block_group->space_info->lock);
7197         block_group->space_info->total_bytes -= block_group->key.offset;
7198         block_group->space_info->bytes_readonly -= block_group->key.offset;
7199         spin_unlock(&block_group->space_info->lock);
7200         block_group->space_info->full = 0;
7201
7202         btrfs_put_block_group(block_group);
7203         btrfs_put_block_group(block_group);
7204
7205         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7206         if (ret > 0)
7207                 ret = -EIO;
7208         if (ret < 0)
7209                 goto out;
7210
7211         ret = btrfs_del_item(trans, root, path);
7212 out:
7213         btrfs_free_path(path);
7214         return ret;
7215 }