git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include "compat.h"
  26 #include "hash.h"
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "print-tree.h"
  30 #include "transaction.h"
  31 #include "volumes.h"
  32 #include "locking.h"
  33 #include "free-space-cache.h"
  34
  35 static int update_reserved_extents(struct btrfs_root *root,
  36                                    u64 bytenr, u64 num, int reserve);
  37 static int update_block_group(struct btrfs_trans_handle *trans,
  38                               struct btrfs_root *root,
  39                               u64 bytenr, u64 num_bytes, int alloc,
  40                               int mark_free);
  41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 u64 bytenr, u64 num_bytes, u64 parent,
  44                                 u64 root_objectid, u64 owner_objectid,
  45                                 u64 owner_offset, int refs_to_drop,
  46                                 struct btrfs_delayed_extent_op *extra_op);
  47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  48                                     struct extent_buffer *leaf,
  49                                     struct btrfs_extent_item *ei);
  50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  51                                       struct btrfs_root *root,
  52                                       u64 parent, u64 root_objectid,
  53                                       u64 flags, u64 owner, u64 offset,
  54                                       struct btrfs_key *ins, int ref_mod);
  55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  56                                      struct btrfs_root *root,
  57                                      u64 parent, u64 root_objectid,
  58                                      u64 flags, struct btrfs_disk_key *key,
  59                                      int level, struct btrfs_key *ins);
  60
  61 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  62                           struct btrfs_root *extent_root, u64 alloc_bytes,
  63                           u64 flags, int force);
  64
  65 static noinline int
  66 block_group_cache_done(struct btrfs_block_group_cache *cache)
  67 {
  68         smp_mb();
  69         return cache->cached == BTRFS_CACHE_FINISHED;
  70 }
  71
  72 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  73 {
  74         return (cache->flags & bits) == bits;
  75 }
  76
  77 /*
  78  * this adds the block group to the fs_info rb tree for the block group
  79  * cache
  80  */
  81 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  82                                 struct btrfs_block_group_cache *block_group)
  83 {
  84         struct rb_node **p;
  85         struct rb_node *parent = NULL;
  86         struct btrfs_block_group_cache *cache;
  87
  88         spin_lock(&info->block_group_cache_lock);
  89         p = &info->block_group_cache_tree.rb_node;
  90
  91         while (*p) {
  92                 parent = *p;
  93                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  94                                  cache_node);
  95                 if (block_group->key.objectid < cache->key.objectid) {
  96                         p = &(*p)->rb_left;
  97                 } else if (block_group->key.objectid > cache->key.objectid) {
  98                         p = &(*p)->rb_right;
  99                 } else {
 100                         spin_unlock(&info->block_group_cache_lock);
 101                         return -EEXIST;
 102                 }
 103         }
 104
 105         rb_link_node(&block_group->cache_node, parent, p);
 106         rb_insert_color(&block_group->cache_node,
 107                         &info->block_group_cache_tree);
 108         spin_unlock(&info->block_group_cache_lock);
 109
 110         return 0;
 111 }
 112
 113 /*
 114  * This will return the block group at or after bytenr if contains is 0, else
 115  * it will return the block group that contains the bytenr
 116  */
 117 static struct btrfs_block_group_cache *
 118 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 119                               int contains)
 120 {
 121         struct btrfs_block_group_cache *cache, *ret = NULL;
 122         struct rb_node *n;
 123         u64 end, start;
 124
 125         spin_lock(&info->block_group_cache_lock);
 126         n = info->block_group_cache_tree.rb_node;
 127
 128         while (n) {
 129                 cache = rb_entry(n, struct btrfs_block_group_cache,
 130                                  cache_node);
 131                 end = cache->key.objectid + cache->key.offset - 1;
 132                 start = cache->key.objectid;
 133
 134                 if (bytenr < start) {
 135                         if (!contains && (!ret || start < ret->key.objectid))
 136                                 ret = cache;
 137                         n = n->rb_left;
 138                 } else if (bytenr > start) {
 139                         if (contains && bytenr <= end) {
 140                                 ret = cache;
 141                                 break;
 142                         }
 143                         n = n->rb_right;
 144                 } else {
 145                         ret = cache;
 146                         break;
 147                 }
 148         }
 149         if (ret)
 150                 atomic_inc(&ret->count);
 151         spin_unlock(&info->block_group_cache_lock);
 152
 153         return ret;
 154 }
 155
 156 /*
 157  * We always set EXTENT_LOCKED for the super mirror extents so we don't
 158  * overwrite them, so those bits need to be unset.  Also, if we are unmounting
 159  * with pinned extents still sitting there because we had a block group caching,
 160  * we need to clear those now, since we are done.
 161  */
 162 void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
 163 {
 164         u64 start, end, last = 0;
 165         int ret;
 166
 167         while (1) {
 168                 ret = find_first_extent_bit(&info->pinned_extents, last,
 169                                             &start, &end,
 170                                             EXTENT_LOCKED|EXTENT_DIRTY);
 171                 if (ret)
 172                         break;
 173
 174                 clear_extent_bits(&info->pinned_extents, start, end,
 175                                   EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS);
 176                 last = end+1;
 177         }
 178 }
 179
 180 static int remove_sb_from_cache(struct btrfs_root *root,
 181                                 struct btrfs_block_group_cache *cache)
 182 {
 183         struct btrfs_fs_info *fs_info = root->fs_info;
 184         u64 bytenr;
 185         u64 *logical;
 186         int stripe_len;
 187         int i, nr, ret;
 188
 189         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 190                 bytenr = btrfs_sb_offset(i);
 191                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 192                                        cache->key.objectid, bytenr,
 193                                        0, &logical, &nr, &stripe_len);
 194                 BUG_ON(ret);
 195                 while (nr--) {
 196                         try_lock_extent(&fs_info->pinned_extents,
 197                                         logical[nr],
 198                                         logical[nr] + stripe_len - 1, GFP_NOFS);
 199                 }
 200                 kfree(logical);
 201         }
 202
 203         return 0;
 204 }
 205
 206 /*
 207  * this is only called by cache_block_group, since we could have freed extents
 208  * we need to check the pinned_extents for any extents that can't be used yet
 209  * since their free space will be released as soon as the transaction commits.
 210  */
 211 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 212                               struct btrfs_fs_info *info, u64 start, u64 end)
 213 {
 214         u64 extent_start, extent_end, size, total_added = 0;
 215         int ret;
 216
 217         while (start < end) {
 218                 ret = find_first_extent_bit(&info->pinned_extents, start,
 219                                             &extent_start, &extent_end,
 220                                             EXTENT_DIRTY|EXTENT_LOCKED);
 221                 if (ret)
 222                         break;
 223
 224                 if (extent_start == start) {
 225                         start = extent_end + 1;
 226                 } else if (extent_start > start && extent_start < end) {
 227                         size = extent_start - start;
 228                         total_added += size;
 229                         ret = btrfs_add_free_space(block_group, start,
 230                                                    size);
 231                         BUG_ON(ret);
 232                         start = extent_end + 1;
 233                 } else {
 234                         break;
 235                 }
 236         }
 237
 238         if (start < end) {
 239                 size = end - start;
 240                 total_added += size;
 241                 ret = btrfs_add_free_space(block_group, start, size);
 242                 BUG_ON(ret);
 243         }
 244
 245         return total_added;
 246 }
 247
 248 static int caching_kthread(void *data)
 249 {
 250         struct btrfs_block_group_cache *block_group = data;
 251         struct btrfs_fs_info *fs_info = block_group->fs_info;
 252         u64 last = 0;
 253         struct btrfs_path *path;
 254         int ret = 0;
 255         struct btrfs_key key;
 256         struct extent_buffer *leaf;
 257         int slot;
 258         u64 total_found = 0;
 259
 260         BUG_ON(!fs_info);
 261
 262         path = btrfs_alloc_path();
 263         if (!path)
 264                 return -ENOMEM;
 265
 266         atomic_inc(&block_group->space_info->caching_threads);
 267         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 268 again:
 269         /* need to make sure the commit_root doesn't disappear */
 270         down_read(&fs_info->extent_root->commit_root_sem);
 271
 272         /*
 273          * We don't want to deadlock with somebody trying to allocate a new
 274          * extent for the extent root while also trying to search the extent
 275          * root to add free space.  So we skip locking and search the commit
 276          * root, since its read-only
 277          */
 278         path->skip_locking = 1;
 279         path->search_commit_root = 1;
 280         path->reada = 2;
 281
 282         key.objectid = last;
 283         key.offset = 0;
 284         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 285         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
 286         if (ret < 0)
 287                 goto err;
 288
 289         while (1) {
 290                 smp_mb();
 291                 if (block_group->fs_info->closing > 1) {
 292                         last = (u64)-1;
 293                         break;
 294                 }
 295
 296                 leaf = path->nodes[0];
 297                 slot = path->slots[0];
 298                 if (slot >= btrfs_header_nritems(leaf)) {
 299                         ret = btrfs_next_leaf(fs_info->extent_root, path);
 300                         if (ret < 0)
 301                                 goto err;
 302                         else if (ret)
 303                                 break;
 304
 305                         if (need_resched()) {
 306                                 btrfs_release_path(fs_info->extent_root, path);
 307                                 up_read(&fs_info->extent_root->commit_root_sem);
 308                                 cond_resched();
 309                                 goto again;
 310                         }
 311
 312                         continue;
 313                 }
 314                 btrfs_item_key_to_cpu(leaf, &key, slot);
 315                 if (key.objectid < block_group->key.objectid)
 316                         goto next;
 317
 318                 if (key.objectid >= block_group->key.objectid +
 319                     block_group->key.offset)
 320                         break;
 321
 322                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 323                         total_found += add_new_free_space(block_group,
 324                                                           fs_info, last,
 325                                                           key.objectid);
 326                         last = key.objectid + key.offset;
 327                 }
 328
 329                 if (total_found > (1024 * 1024 * 2)) {
 330                         total_found = 0;
 331                         wake_up(&block_group->caching_q);
 332                 }
 333 next:
 334                 path->slots[0]++;
 335         }
 336         ret = 0;
 337
 338         total_found += add_new_free_space(block_group, fs_info, last,
 339                                           block_group->key.objectid +
 340                                           block_group->key.offset);
 341
 342         spin_lock(&block_group->lock);
 343         block_group->cached = BTRFS_CACHE_FINISHED;
 344         spin_unlock(&block_group->lock);
 345
 346 err:
 347         btrfs_free_path(path);
 348         up_read(&fs_info->extent_root->commit_root_sem);
 349         atomic_dec(&block_group->space_info->caching_threads);
 350         wake_up(&block_group->caching_q);
 351
 352         return 0;
 353 }
 354
 355 static int cache_block_group(struct btrfs_block_group_cache *cache)
 356 {
 357         struct task_struct *tsk;
 358         int ret = 0;
 359
 360         spin_lock(&cache->lock);
 361         if (cache->cached != BTRFS_CACHE_NO) {
 362                 spin_unlock(&cache->lock);
 363                 return ret;
 364         }
 365         cache->cached = BTRFS_CACHE_STARTED;
 366         spin_unlock(&cache->lock);
 367
 368         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 369                           cache->key.objectid);
 370         if (IS_ERR(tsk)) {
 371                 ret = PTR_ERR(tsk);
 372                 printk(KERN_ERR "error running thread %d\n", ret);
 373                 BUG();
 374         }
 375
 376         return ret;
 377 }
 378
 379 /*
 380  * return the block group that starts at or after bytenr
 381  */
 382 static struct btrfs_block_group_cache *
 383 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 384 {
 385         struct btrfs_block_group_cache *cache;
 386
 387         cache = block_group_cache_tree_search(info, bytenr, 0);
 388
 389         return cache;
 390 }
 391
 392 /*
 393  * return the block group that contains the given bytenr
 394  */
 395 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 396                                                  struct btrfs_fs_info *info,
 397                                                  u64 bytenr)
 398 {
 399         struct btrfs_block_group_cache *cache;
 400
 401         cache = block_group_cache_tree_search(info, bytenr, 1);
 402
 403         return cache;
 404 }
 405
 406 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 407 {
 408         if (atomic_dec_and_test(&cache->count))
 409                 kfree(cache);
 410 }
 411
 412 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 413                                                   u64 flags)
 414 {
 415         struct list_head *head = &info->space_info;
 416         struct btrfs_space_info *found;
 417
 418         rcu_read_lock();
 419         list_for_each_entry_rcu(found, head, list) {
 420                 if (found->flags == flags) {
 421                         rcu_read_unlock();
 422                         return found;
 423                 }
 424         }
 425         rcu_read_unlock();
 426         return NULL;
 427 }
 428
 429 /*
 430  * after adding space to the filesystem, we need to clear the full flags
 431  * on all the space infos.
 432  */
 433 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 434 {
 435         struct list_head *head = &info->space_info;
 436         struct btrfs_space_info *found;
 437
 438         rcu_read_lock();
 439         list_for_each_entry_rcu(found, head, list)
 440                 found->full = 0;
 441         rcu_read_unlock();
 442 }
 443
 444 static u64 div_factor(u64 num, int factor)
 445 {
 446         if (factor == 10)
 447                 return num;
 448         num *= factor;
 449         do_div(num, 10);
 450         return num;
 451 }
 452
 453 u64 btrfs_find_block_group(struct btrfs_root *root,
 454                            u64 search_start, u64 search_hint, int owner)
 455 {
 456         struct btrfs_block_group_cache *cache;
 457         u64 used;
 458         u64 last = max(search_hint, search_start);
 459         u64 group_start = 0;
 460         int full_search = 0;
 461         int factor = 9;
 462         int wrapped = 0;
 463 again:
 464         while (1) {
 465                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 466                 if (!cache)
 467                         break;
 468
 469                 spin_lock(&cache->lock);
 470                 last = cache->key.objectid + cache->key.offset;
 471                 used = btrfs_block_group_used(&cache->item);
 472
 473                 if ((full_search || !cache->ro) &&
 474                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 475                         if (used + cache->pinned + cache->reserved <
 476                             div_factor(cache->key.offset, factor)) {
 477                                 group_start = cache->key.objectid;
 478                                 spin_unlock(&cache->lock);
 479                                 btrfs_put_block_group(cache);
 480                                 goto found;
 481                         }
 482                 }
 483                 spin_unlock(&cache->lock);
 484                 btrfs_put_block_group(cache);
 485                 cond_resched();
 486         }
 487         if (!wrapped) {
 488                 last = search_start;
 489                 wrapped = 1;
 490                 goto again;
 491         }
 492         if (!full_search && factor < 10) {
 493                 last = search_start;
 494                 full_search = 1;
 495                 factor = 10;
 496                 goto again;
 497         }
 498 found:
 499         return group_start;
 500 }
 501
 502 /* simple helper to search for an existing extent at a given offset */
 503 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 504 {
 505         int ret;
 506         struct btrfs_key key;
 507         struct btrfs_path *path;
 508
 509         path = btrfs_alloc_path();
 510         BUG_ON(!path);
 511         key.objectid = start;
 512         key.offset = len;
 513         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 514         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 515                                 0, 0);
 516         btrfs_free_path(path);
 517         return ret;
 518 }
 519
 520 /*
 521  * Back reference rules.  Back refs have three main goals:
 522  *
 523  * 1) differentiate between all holders of references to an extent so that
 524  *    when a reference is dropped we can make sure it was a valid reference
 525  *    before freeing the extent.
 526  *
 527  * 2) Provide enough information to quickly find the holders of an extent
 528  *    if we notice a given block is corrupted or bad.
 529  *
 530  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 531  *    maintenance.  This is actually the same as #2, but with a slightly
 532  *    different use case.
 533  *
 534  * There are two kinds of back refs. The implicit back refs is optimized
 535  * for pointers in non-shared tree blocks. For a given pointer in a block,
 536  * back refs of this kind provide information about the block's owner tree
 537  * and the pointer's key. These information allow us to find the block by
 538  * b-tree searching. The full back refs is for pointers in tree blocks not
 539  * referenced by their owner trees. The location of tree block is recorded
 540  * in the back refs. Actually the full back refs is generic, and can be
 541  * used in all cases the implicit back refs is used. The major shortcoming
 542  * of the full back refs is its overhead. Every time a tree block gets
 543  * COWed, we have to update back refs entry for all pointers in it.
 544  *
 545  * For a newly allocated tree block, we use implicit back refs for
 546  * pointers in it. This means most tree related operations only involve
 547  * implicit back refs. For a tree block created in old transaction, the
 548  * only way to drop a reference to it is COW it. So we can detect the
 549  * event that tree block loses its owner tree's reference and do the
 550  * back refs conversion.
 551  *
 552  * When a tree block is COW'd through a tree, there are four cases:
 553  *
 554  * The reference count of the block is one and the tree is the block's
 555  * owner tree. Nothing to do in this case.
 556  *
 557  * The reference count of the block is one and the tree is not the
 558  * block's owner tree. In this case, full back refs is used for pointers
 559  * in the block. Remove these full back refs, add implicit back refs for
 560  * every pointers in the new block.
 561  *
 562  * The reference count of the block is greater than one and the tree is
 563  * the block's owner tree. In this case, implicit back refs is used for
 564  * pointers in the block. Add full back refs for every pointers in the
 565  * block, increase lower level extents' reference counts. The original
 566  * implicit back refs are entailed to the new block.
 567  *
 568  * The reference count of the block is greater than one and the tree is
 569  * not the block's owner tree. Add implicit back refs for every pointer in
 570  * the new block, increase lower level extents' reference count.
 571  *
 572  * Back Reference Key composing:
 573  *
 574  * The key objectid corresponds to the first byte in the extent,
 575  * The key type is used to differentiate between types of back refs.
 576  * There are different meanings of the key offset for different types
 577  * of back refs.
 578  *
 579  * File extents can be referenced by:
 580  *
 581  * - multiple snapshots, subvolumes, or different generations in one subvol
 582  * - different files inside a single subvolume
 583  * - different offsets inside a file (bookend extents in file.c)
 584  *
 585  * The extent ref structure for the implicit back refs has fields for:
 586  *
 587  * - Objectid of the subvolume root
 588  * - objectid of the file holding the reference
 589  * - original offset in the file
 590  * - how many bookend extents
 591  *
 592  * The key offset for the implicit back refs is hash of the first
 593  * three fields.
 594  *
 595  * The extent ref structure for the full back refs has field for:
 596  *
 597  * - number of pointers in the tree leaf
 598  *
 599  * The key offset for the implicit back refs is the first byte of
 600  * the tree leaf
 601  *
 602  * When a file extent is allocated, The implicit back refs is used.
 603  * the fields are filled in:
 604  *
 605  *     (root_key.objectid, inode objectid, offset in file, 1)
 606  *
 607  * When a file extent is removed file truncation, we find the
 608  * corresponding implicit back refs and check the following fields:
 609  *
 610  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 611  *
 612  * Btree extents can be referenced by:
 613  *
 614  * - Different subvolumes
 615  *
 616  * Both the implicit back refs and the full back refs for tree blocks
 617  * only consist of key. The key offset for the implicit back refs is
 618  * objectid of block's owner tree. The key offset for the full back refs
 619  * is the first byte of parent block.
 620  *
 621  * When implicit back refs is used, information about the lowest key and
 622  * level of the tree block are required. These information are stored in
 623  * tree block info structure.
 624  */
 625
 626 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 627 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 628                                   struct btrfs_root *root,
 629                                   struct btrfs_path *path,
 630                                   u64 owner, u32 extra_size)
 631 {
 632         struct btrfs_extent_item *item;
 633         struct btrfs_extent_item_v0 *ei0;
 634         struct btrfs_extent_ref_v0 *ref0;
 635         struct btrfs_tree_block_info *bi;
 636         struct extent_buffer *leaf;
 637         struct btrfs_key key;
 638         struct btrfs_key found_key;
 639         u32 new_size = sizeof(*item);
 640         u64 refs;
 641         int ret;
 642
 643         leaf = path->nodes[0];
 644         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 645
 646         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 647         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 648                              struct btrfs_extent_item_v0);
 649         refs = btrfs_extent_refs_v0(leaf, ei0);
 650
 651         if (owner == (u64)-1) {
 652                 while (1) {
 653                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 654                                 ret = btrfs_next_leaf(root, path);
 655                                 if (ret < 0)
 656                                         return ret;
 657                                 BUG_ON(ret > 0);
 658                                 leaf = path->nodes[0];
 659                         }
 660                         btrfs_item_key_to_cpu(leaf, &found_key,
 661                                               path->slots[0]);
 662                         BUG_ON(key.objectid != found_key.objectid);
 663                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 664                                 path->slots[0]++;
 665                                 continue;
 666                         }
 667                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 668                                               struct btrfs_extent_ref_v0);
 669                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 670                         break;
 671                 }
 672         }
 673         btrfs_release_path(root, path);
 674
 675         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 676                 new_size += sizeof(*bi);
 677
 678         new_size -= sizeof(*ei0);
 679         ret = btrfs_search_slot(trans, root, &key, path,
 680                                 new_size + extra_size, 1);
 681         if (ret < 0)
 682                 return ret;
 683         BUG_ON(ret);
 684
 685         ret = btrfs_extend_item(trans, root, path, new_size);
 686         BUG_ON(ret);
 687
 688         leaf = path->nodes[0];
 689         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 690         btrfs_set_extent_refs(leaf, item, refs);
 691         /* FIXME: get real generation */
 692         btrfs_set_extent_generation(leaf, item, 0);
 693         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 694                 btrfs_set_extent_flags(leaf, item,
 695                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 696                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 697                 bi = (struct btrfs_tree_block_info *)(item + 1);
 698                 /* FIXME: get first key of the block */
 699                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 700                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 701         } else {
 702                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 703         }
 704         btrfs_mark_buffer_dirty(leaf);
 705         return 0;
 706 }
 707 #endif
 708
 709 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 710 {
 711         u32 high_crc = ~(u32)0;
 712         u32 low_crc = ~(u32)0;
 713         __le64 lenum;
 714
 715         lenum = cpu_to_le64(root_objectid);
 716         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 717         lenum = cpu_to_le64(owner);
 718         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 719         lenum = cpu_to_le64(offset);
 720         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 721
 722         return ((u64)high_crc << 31) ^ (u64)low_crc;
 723 }
 724
 725 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 726                                      struct btrfs_extent_data_ref *ref)
 727 {
 728         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 729                                     btrfs_extent_data_ref_objectid(leaf, ref),
 730                                     btrfs_extent_data_ref_offset(leaf, ref));
 731 }
 732
 733 static int match_extent_data_ref(struct extent_buffer *leaf,
 734                                  struct btrfs_extent_data_ref *ref,
 735                                  u64 root_objectid, u64 owner, u64 offset)
 736 {
 737         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 738             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 739             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 740                 return 0;
 741         return 1;
 742 }
 743
 744 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 745                                            struct btrfs_root *root,
 746                                            struct btrfs_path *path,
 747                                            u64 bytenr, u64 parent,
 748                                            u64 root_objectid,
 749                                            u64 owner, u64 offset)
 750 {
 751         struct btrfs_key key;
 752         struct btrfs_extent_data_ref *ref;
 753         struct extent_buffer *leaf;
 754         u32 nritems;
 755         int ret;
 756         int recow;
 757         int err = -ENOENT;
 758
 759         key.objectid = bytenr;
 760         if (parent) {
 761                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 762                 key.offset = parent;
 763         } else {
 764                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 765                 key.offset = hash_extent_data_ref(root_objectid,
 766                                                   owner, offset);
 767         }
 768 again:
 769         recow = 0;
 770         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 771         if (ret < 0) {
 772                 err = ret;
 773                 goto fail;
 774         }
 775
 776         if (parent) {
 777                 if (!ret)
 778                         return 0;
 779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 780                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 781                 btrfs_release_path(root, path);
 782                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 783                 if (ret < 0) {
 784                         err = ret;
 785                         goto fail;
 786                 }
 787                 if (!ret)
 788                         return 0;
 789 #endif
 790                 goto fail;
 791         }
 792
 793         leaf = path->nodes[0];
 794         nritems = btrfs_header_nritems(leaf);
 795         while (1) {
 796                 if (path->slots[0] >= nritems) {
 797                         ret = btrfs_next_leaf(root, path);
 798                         if (ret < 0)
 799                                 err = ret;
 800                         if (ret)
 801                                 goto fail;
 802
 803                         leaf = path->nodes[0];
 804                         nritems = btrfs_header_nritems(leaf);
 805                         recow = 1;
 806                 }
 807
 808                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 809                 if (key.objectid != bytenr ||
 810                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 811                         goto fail;
 812
 813                 ref = btrfs_item_ptr(leaf, path->slots[0],
 814                                      struct btrfs_extent_data_ref);
 815
 816                 if (match_extent_data_ref(leaf, ref, root_objectid,
 817                                           owner, offset)) {
 818                         if (recow) {
 819                                 btrfs_release_path(root, path);
 820                                 goto again;
 821                         }
 822                         err = 0;
 823                         break;
 824                 }
 825                 path->slots[0]++;
 826         }
 827 fail:
 828         return err;
 829 }
 830
 831 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 832                                            struct btrfs_root *root,
 833                                            struct btrfs_path *path,
 834                                            u64 bytenr, u64 parent,
 835                                            u64 root_objectid, u64 owner,
 836                                            u64 offset, int refs_to_add)
 837 {
 838         struct btrfs_key key;
 839         struct extent_buffer *leaf;
 840         u32 size;
 841         u32 num_refs;
 842         int ret;
 843
 844         key.objectid = bytenr;
 845         if (parent) {
 846                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 847                 key.offset = parent;
 848                 size = sizeof(struct btrfs_shared_data_ref);
 849         } else {
 850                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 851                 key.offset = hash_extent_data_ref(root_objectid,
 852                                                   owner, offset);
 853                 size = sizeof(struct btrfs_extent_data_ref);
 854         }
 855
 856         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 857         if (ret && ret != -EEXIST)
 858                 goto fail;
 859
 860         leaf = path->nodes[0];
 861         if (parent) {
 862                 struct btrfs_shared_data_ref *ref;
 863                 ref = btrfs_item_ptr(leaf, path->slots[0],
 864                                      struct btrfs_shared_data_ref);
 865                 if (ret == 0) {
 866                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 867                 } else {
 868                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 869                         num_refs += refs_to_add;
 870                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 871                 }
 872         } else {
 873                 struct btrfs_extent_data_ref *ref;
 874                 while (ret == -EEXIST) {
 875                         ref = btrfs_item_ptr(leaf, path->slots[0],
 876                                              struct btrfs_extent_data_ref);
 877                         if (match_extent_data_ref(leaf, ref, root_objectid,
 878                                                   owner, offset))
 879                                 break;
 880                         btrfs_release_path(root, path);
 881                         key.offset++;
 882                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 883                                                       size);
 884                         if (ret && ret != -EEXIST)
 885                                 goto fail;
 886
 887                         leaf = path->nodes[0];
 888                 }
 889                 ref = btrfs_item_ptr(leaf, path->slots[0],
 890                                      struct btrfs_extent_data_ref);
 891                 if (ret == 0) {
 892                         btrfs_set_extent_data_ref_root(leaf, ref,
 893                                                        root_objectid);
 894                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 895                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 896                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 897                 } else {
 898                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 899                         num_refs += refs_to_add;
 900                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 901                 }
 902         }
 903         btrfs_mark_buffer_dirty(leaf);
 904         ret = 0;
 905 fail:
 906         btrfs_release_path(root, path);
 907         return ret;
 908 }
 909
 910 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 911                                            struct btrfs_root *root,
 912                                            struct btrfs_path *path,
 913                                            int refs_to_drop)
 914 {
 915         struct btrfs_key key;
 916         struct btrfs_extent_data_ref *ref1 = NULL;
 917         struct btrfs_shared_data_ref *ref2 = NULL;
 918         struct extent_buffer *leaf;
 919         u32 num_refs = 0;
 920         int ret = 0;
 921
 922         leaf = path->nodes[0];
 923         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 924
 925         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 926                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 927                                       struct btrfs_extent_data_ref);
 928                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 929         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 930                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 931                                       struct btrfs_shared_data_ref);
 932                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 934         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 935                 struct btrfs_extent_ref_v0 *ref0;
 936                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
 937                                       struct btrfs_extent_ref_v0);
 938                 num_refs = btrfs_ref_count_v0(leaf, ref0);
 939 #endif
 940         } else {
 941                 BUG();
 942         }
 943
 944         BUG_ON(num_refs < refs_to_drop);
 945         num_refs -= refs_to_drop;
 946
 947         if (num_refs == 0) {
 948                 ret = btrfs_del_item(trans, root, path);
 949         } else {
 950                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 951                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 952                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 953                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 955                 else {
 956                         struct btrfs_extent_ref_v0 *ref0;
 957                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 958                                         struct btrfs_extent_ref_v0);
 959                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
 960                 }
 961 #endif
 962                 btrfs_mark_buffer_dirty(leaf);
 963         }
 964         return ret;
 965 }
 966
 967 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
 968                                           struct btrfs_path *path,
 969                                           struct btrfs_extent_inline_ref *iref)
 970 {
 971         struct btrfs_key key;
 972         struct extent_buffer *leaf;
 973         struct btrfs_extent_data_ref *ref1;
 974         struct btrfs_shared_data_ref *ref2;
 975         u32 num_refs = 0;
 976
 977         leaf = path->nodes[0];
 978         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 979         if (iref) {
 980                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
 981                     BTRFS_EXTENT_DATA_REF_KEY) {
 982                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 983                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 984                 } else {
 985                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 986                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 987                 }
 988         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 989                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 990                                       struct btrfs_extent_data_ref);
 991                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 992         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 993                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 994                                       struct btrfs_shared_data_ref);
 995                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 996 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 997         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 998                 struct btrfs_extent_ref_v0 *ref0;
 999                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1000                                       struct btrfs_extent_ref_v0);
1001                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1002 #endif
1003         } else {
1004                 WARN_ON(1);
1005         }
1006         return num_refs;
1007 }
1008
1009 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1010                                           struct btrfs_root *root,
1011                                           struct btrfs_path *path,
1012                                           u64 bytenr, u64 parent,
1013                                           u64 root_objectid)
1014 {
1015         struct btrfs_key key;
1016         int ret;
1017
1018         key.objectid = bytenr;
1019         if (parent) {
1020                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1021                 key.offset = parent;
1022         } else {
1023                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1024                 key.offset = root_objectid;
1025         }
1026
1027         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1028         if (ret > 0)
1029                 ret = -ENOENT;
1030 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031         if (ret == -ENOENT && parent) {
1032                 btrfs_release_path(root, path);
1033                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1034                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1035                 if (ret > 0)
1036                         ret = -ENOENT;
1037         }
1038 #endif
1039         return ret;
1040 }
1041
1042 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1043                                           struct btrfs_root *root,
1044                                           struct btrfs_path *path,
1045                                           u64 bytenr, u64 parent,
1046                                           u64 root_objectid)
1047 {
1048         struct btrfs_key key;
1049         int ret;
1050
1051         key.objectid = bytenr;
1052         if (parent) {
1053                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1054                 key.offset = parent;
1055         } else {
1056                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1057                 key.offset = root_objectid;
1058         }
1059
1060         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1061         btrfs_release_path(root, path);
1062         return ret;
1063 }
1064
1065 static inline int extent_ref_type(u64 parent, u64 owner)
1066 {
1067         int type;
1068         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1069                 if (parent > 0)
1070                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1071                 else
1072                         type = BTRFS_TREE_BLOCK_REF_KEY;
1073         } else {
1074                 if (parent > 0)
1075                         type = BTRFS_SHARED_DATA_REF_KEY;
1076                 else
1077                         type = BTRFS_EXTENT_DATA_REF_KEY;
1078         }
1079         return type;
1080 }
1081
1082 static int find_next_key(struct btrfs_path *path, int level,
1083                          struct btrfs_key *key)
1084
1085 {
1086         for (; level < BTRFS_MAX_LEVEL; level++) {
1087                 if (!path->nodes[level])
1088                         break;
1089                 if (path->slots[level] + 1 >=
1090                     btrfs_header_nritems(path->nodes[level]))
1091                         continue;
1092                 if (level == 0)
1093                         btrfs_item_key_to_cpu(path->nodes[level], key,
1094                                               path->slots[level] + 1);
1095                 else
1096                         btrfs_node_key_to_cpu(path->nodes[level], key,
1097                                               path->slots[level] + 1);
1098                 return 0;
1099         }
1100         return 1;
1101 }
1102
1103 /*
1104  * look for inline back ref. if back ref is found, *ref_ret is set
1105  * to the address of inline back ref, and 0 is returned.
1106  *
1107  * if back ref isn't found, *ref_ret is set to the address where it
1108  * should be inserted, and -ENOENT is returned.
1109  *
1110  * if insert is true and there are too many inline back refs, the path
1111  * points to the extent item, and -EAGAIN is returned.
1112  *
1113  * NOTE: inline back refs are ordered in the same way that back ref
1114  *       items in the tree are ordered.
1115  */
1116 static noinline_for_stack
1117 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1118                                  struct btrfs_root *root,
1119                                  struct btrfs_path *path,
1120                                  struct btrfs_extent_inline_ref **ref_ret,
1121                                  u64 bytenr, u64 num_bytes,
1122                                  u64 parent, u64 root_objectid,
1123                                  u64 owner, u64 offset, int insert)
1124 {
1125         struct btrfs_key key;
1126         struct extent_buffer *leaf;
1127         struct btrfs_extent_item *ei;
1128         struct btrfs_extent_inline_ref *iref;
1129         u64 flags;
1130         u64 item_size;
1131         unsigned long ptr;
1132         unsigned long end;
1133         int extra_size;
1134         int type;
1135         int want;
1136         int ret;
1137         int err = 0;
1138
1139         key.objectid = bytenr;
1140         key.type = BTRFS_EXTENT_ITEM_KEY;
1141         key.offset = num_bytes;
1142
1143         want = extent_ref_type(parent, owner);
1144         if (insert) {
1145                 extra_size = btrfs_extent_inline_ref_size(want);
1146                 path->keep_locks = 1;
1147         } else
1148                 extra_size = -1;
1149         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1150         if (ret < 0) {
1151                 err = ret;
1152                 goto out;
1153         }
1154         BUG_ON(ret);
1155
1156         leaf = path->nodes[0];
1157         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1158 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1159         if (item_size < sizeof(*ei)) {
1160                 if (!insert) {
1161                         err = -ENOENT;
1162                         goto out;
1163                 }
1164                 ret = convert_extent_item_v0(trans, root, path, owner,
1165                                              extra_size);
1166                 if (ret < 0) {
1167                         err = ret;
1168                         goto out;
1169                 }
1170                 leaf = path->nodes[0];
1171                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1172         }
1173 #endif
1174         BUG_ON(item_size < sizeof(*ei));
1175
1176         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1177         flags = btrfs_extent_flags(leaf, ei);
1178
1179         ptr = (unsigned long)(ei + 1);
1180         end = (unsigned long)ei + item_size;
1181
1182         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1183                 ptr += sizeof(struct btrfs_tree_block_info);
1184                 BUG_ON(ptr > end);
1185         } else {
1186                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1187         }
1188
1189         err = -ENOENT;
1190         while (1) {
1191                 if (ptr >= end) {
1192                         WARN_ON(ptr > end);
1193                         break;
1194                 }
1195                 iref = (struct btrfs_extent_inline_ref *)ptr;
1196                 type = btrfs_extent_inline_ref_type(leaf, iref);
1197                 if (want < type)
1198                         break;
1199                 if (want > type) {
1200                         ptr += btrfs_extent_inline_ref_size(type);
1201                         continue;
1202                 }
1203
1204                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1205                         struct btrfs_extent_data_ref *dref;
1206                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1207                         if (match_extent_data_ref(leaf, dref, root_objectid,
1208                                                   owner, offset)) {
1209                                 err = 0;
1210                                 break;
1211                         }
1212                         if (hash_extent_data_ref_item(leaf, dref) <
1213                             hash_extent_data_ref(root_objectid, owner, offset))
1214                                 break;
1215                 } else {
1216                         u64 ref_offset;
1217                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1218                         if (parent > 0) {
1219                                 if (parent == ref_offset) {
1220                                         err = 0;
1221                                         break;
1222                                 }
1223                                 if (ref_offset < parent)
1224                                         break;
1225                         } else {
1226                                 if (root_objectid == ref_offset) {
1227                                         err = 0;
1228                                         break;
1229                                 }
1230                                 if (ref_offset < root_objectid)
1231                                         break;
1232                         }
1233                 }
1234                 ptr += btrfs_extent_inline_ref_size(type);
1235         }
1236         if (err == -ENOENT && insert) {
1237                 if (item_size + extra_size >=
1238                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1239                         err = -EAGAIN;
1240                         goto out;
1241                 }
1242                 /*
1243                  * To add new inline back ref, we have to make sure
1244                  * there is no corresponding back ref item.
1245                  * For simplicity, we just do not add new inline back
1246                  * ref if there is any kind of item for this block
1247                  */
1248                 if (find_next_key(path, 0, &key) == 0 &&
1249                     key.objectid == bytenr &&
1250                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1251                         err = -EAGAIN;
1252                         goto out;
1253                 }
1254         }
1255         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1256 out:
1257         if (insert) {
1258                 path->keep_locks = 0;
1259                 btrfs_unlock_up_safe(path, 1);
1260         }
1261         return err;
1262 }
1263
1264 /*
1265  * helper to add new inline back ref
1266  */
1267 static noinline_for_stack
1268 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1269                                 struct btrfs_root *root,
1270                                 struct btrfs_path *path,
1271                                 struct btrfs_extent_inline_ref *iref,
1272                                 u64 parent, u64 root_objectid,
1273                                 u64 owner, u64 offset, int refs_to_add,
1274                                 struct btrfs_delayed_extent_op *extent_op)
1275 {
1276         struct extent_buffer *leaf;
1277         struct btrfs_extent_item *ei;
1278         unsigned long ptr;
1279         unsigned long end;
1280         unsigned long item_offset;
1281         u64 refs;
1282         int size;
1283         int type;
1284         int ret;
1285
1286         leaf = path->nodes[0];
1287         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1288         item_offset = (unsigned long)iref - (unsigned long)ei;
1289
1290         type = extent_ref_type(parent, owner);
1291         size = btrfs_extent_inline_ref_size(type);
1292
1293         ret = btrfs_extend_item(trans, root, path, size);
1294         BUG_ON(ret);
1295
1296         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1297         refs = btrfs_extent_refs(leaf, ei);
1298         refs += refs_to_add;
1299         btrfs_set_extent_refs(leaf, ei, refs);
1300         if (extent_op)
1301                 __run_delayed_extent_op(extent_op, leaf, ei);
1302
1303         ptr = (unsigned long)ei + item_offset;
1304         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1305         if (ptr < end - size)
1306                 memmove_extent_buffer(leaf, ptr + size, ptr,
1307                                       end - size - ptr);
1308
1309         iref = (struct btrfs_extent_inline_ref *)ptr;
1310         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1311         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1312                 struct btrfs_extent_data_ref *dref;
1313                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1314                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1315                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1316                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1317                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1318         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1319                 struct btrfs_shared_data_ref *sref;
1320                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1321                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1322                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1323         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1324                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1325         } else {
1326                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1327         }
1328         btrfs_mark_buffer_dirty(leaf);
1329         return 0;
1330 }
1331
1332 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1333                                  struct btrfs_root *root,
1334                                  struct btrfs_path *path,
1335                                  struct btrfs_extent_inline_ref **ref_ret,
1336                                  u64 bytenr, u64 num_bytes, u64 parent,
1337                                  u64 root_objectid, u64 owner, u64 offset)
1338 {
1339         int ret;
1340
1341         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1342                                            bytenr, num_bytes, parent,
1343                                            root_objectid, owner, offset, 0);
1344         if (ret != -ENOENT)
1345                 return ret;
1346
1347         btrfs_release_path(root, path);
1348         *ref_ret = NULL;
1349
1350         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1351                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1352                                             root_objectid);
1353         } else {
1354                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1355                                              root_objectid, owner, offset);
1356         }
1357         return ret;
1358 }
1359
1360 /*
1361  * helper to update/remove inline back ref
1362  */
1363 static noinline_for_stack
1364 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref *iref,
1368                                  int refs_to_mod,
1369                                  struct btrfs_delayed_extent_op *extent_op)
1370 {
1371         struct extent_buffer *leaf;
1372         struct btrfs_extent_item *ei;
1373         struct btrfs_extent_data_ref *dref = NULL;
1374         struct btrfs_shared_data_ref *sref = NULL;
1375         unsigned long ptr;
1376         unsigned long end;
1377         u32 item_size;
1378         int size;
1379         int type;
1380         int ret;
1381         u64 refs;
1382
1383         leaf = path->nodes[0];
1384         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1385         refs = btrfs_extent_refs(leaf, ei);
1386         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1387         refs += refs_to_mod;
1388         btrfs_set_extent_refs(leaf, ei, refs);
1389         if (extent_op)
1390                 __run_delayed_extent_op(extent_op, leaf, ei);
1391
1392         type = btrfs_extent_inline_ref_type(leaf, iref);
1393
1394         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1395                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1396                 refs = btrfs_extent_data_ref_count(leaf, dref);
1397         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1398                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1399                 refs = btrfs_shared_data_ref_count(leaf, sref);
1400         } else {
1401                 refs = 1;
1402                 BUG_ON(refs_to_mod != -1);
1403         }
1404
1405         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1406         refs += refs_to_mod;
1407
1408         if (refs > 0) {
1409                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1410                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1411                 else
1412                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1413         } else {
1414                 size =  btrfs_extent_inline_ref_size(type);
1415                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1416                 ptr = (unsigned long)iref;
1417                 end = (unsigned long)ei + item_size;
1418                 if (ptr + size < end)
1419                         memmove_extent_buffer(leaf, ptr, ptr + size,
1420                                               end - ptr - size);
1421                 item_size -= size;
1422                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1423                 BUG_ON(ret);
1424         }
1425         btrfs_mark_buffer_dirty(leaf);
1426         return 0;
1427 }
1428
1429 static noinline_for_stack
1430 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1431                                  struct btrfs_root *root,
1432                                  struct btrfs_path *path,
1433                                  u64 bytenr, u64 num_bytes, u64 parent,
1434                                  u64 root_objectid, u64 owner,
1435                                  u64 offset, int refs_to_add,
1436                                  struct btrfs_delayed_extent_op *extent_op)
1437 {
1438         struct btrfs_extent_inline_ref *iref;
1439         int ret;
1440
1441         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1442                                            bytenr, num_bytes, parent,
1443                                            root_objectid, owner, offset, 1);
1444         if (ret == 0) {
1445                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1446                 ret = update_inline_extent_backref(trans, root, path, iref,
1447                                                    refs_to_add, extent_op);
1448         } else if (ret == -ENOENT) {
1449                 ret = setup_inline_extent_backref(trans, root, path, iref,
1450                                                   parent, root_objectid,
1451                                                   owner, offset, refs_to_add,
1452                                                   extent_op);
1453         }
1454         return ret;
1455 }
1456
1457 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1458                                  struct btrfs_root *root,
1459                                  struct btrfs_path *path,
1460                                  u64 bytenr, u64 parent, u64 root_objectid,
1461                                  u64 owner, u64 offset, int refs_to_add)
1462 {
1463         int ret;
1464         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1465                 BUG_ON(refs_to_add != 1);
1466                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1467                                             parent, root_objectid);
1468         } else {
1469                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1470                                              parent, root_objectid,
1471                                              owner, offset, refs_to_add);
1472         }
1473         return ret;
1474 }
1475
1476 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1477                                  struct btrfs_root *root,
1478                                  struct btrfs_path *path,
1479                                  struct btrfs_extent_inline_ref *iref,
1480                                  int refs_to_drop, int is_data)
1481 {
1482         int ret;
1483
1484         BUG_ON(!is_data && refs_to_drop != 1);
1485         if (iref) {
1486                 ret = update_inline_extent_backref(trans, root, path, iref,
1487                                                    -refs_to_drop, NULL);
1488         } else if (is_data) {
1489                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1490         } else {
1491                 ret = btrfs_del_item(trans, root, path);
1492         }
1493         return ret;
1494 }
1495
1496 #ifdef BIO_RW_DISCARD
1497 static void btrfs_issue_discard(struct block_device *bdev,
1498                                 u64 start, u64 len)
1499 {
1500         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1501 }
1502 #endif
1503
1504 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1505                                 u64 num_bytes)
1506 {
1507 #ifdef BIO_RW_DISCARD
1508         int ret;
1509         u64 map_length = num_bytes;
1510         struct btrfs_multi_bio *multi = NULL;
1511
1512         /* Tell the block device(s) that the sectors can be discarded */
1513         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1514                               bytenr, &map_length, &multi, 0);
1515         if (!ret) {
1516                 struct btrfs_bio_stripe *stripe = multi->stripes;
1517                 int i;
1518
1519                 if (map_length > num_bytes)
1520                         map_length = num_bytes;
1521
1522                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1523                         btrfs_issue_discard(stripe->dev->bdev,
1524                                             stripe->physical,
1525                                             map_length);
1526                 }
1527                 kfree(multi);
1528         }
1529
1530         return ret;
1531 #else
1532         return 0;
1533 #endif
1534 }
1535
1536 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1537                          struct btrfs_root *root,
1538                          u64 bytenr, u64 num_bytes, u64 parent,
1539                          u64 root_objectid, u64 owner, u64 offset)
1540 {
1541         int ret;
1542         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1543                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1544
1545         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1546                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1547                                         parent, root_objectid, (int)owner,
1548                                         BTRFS_ADD_DELAYED_REF, NULL);
1549         } else {
1550                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1551                                         parent, root_objectid, owner, offset,
1552                                         BTRFS_ADD_DELAYED_REF, NULL);
1553         }
1554         return ret;
1555 }
1556
1557 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1558                                   struct btrfs_root *root,
1559                                   u64 bytenr, u64 num_bytes,
1560                                   u64 parent, u64 root_objectid,
1561                                   u64 owner, u64 offset, int refs_to_add,
1562                                   struct btrfs_delayed_extent_op *extent_op)
1563 {
1564         struct btrfs_path *path;
1565         struct extent_buffer *leaf;
1566         struct btrfs_extent_item *item;
1567         u64 refs;
1568         int ret;
1569         int err = 0;
1570
1571         path = btrfs_alloc_path();
1572         if (!path)
1573                 return -ENOMEM;
1574
1575         path->reada = 1;
1576         path->leave_spinning = 1;
1577         /* this will setup the path even if it fails to insert the back ref */
1578         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1579                                            path, bytenr, num_bytes, parent,
1580                                            root_objectid, owner, offset,
1581                                            refs_to_add, extent_op);
1582         if (ret == 0)
1583                 goto out;
1584
1585         if (ret != -EAGAIN) {
1586                 err = ret;
1587                 goto out;
1588         }
1589
1590         leaf = path->nodes[0];
1591         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1592         refs = btrfs_extent_refs(leaf, item);
1593         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1594         if (extent_op)
1595                 __run_delayed_extent_op(extent_op, leaf, item);
1596
1597         btrfs_mark_buffer_dirty(leaf);
1598         btrfs_release_path(root->fs_info->extent_root, path);
1599
1600         path->reada = 1;
1601         path->leave_spinning = 1;
1602
1603         /* now insert the actual backref */
1604         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1605                                     path, bytenr, parent, root_objectid,
1606                                     owner, offset, refs_to_add);
1607         BUG_ON(ret);
1608 out:
1609         btrfs_free_path(path);
1610         return err;
1611 }
1612
1613 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1614                                 struct btrfs_root *root,
1615                                 struct btrfs_delayed_ref_node *node,
1616                                 struct btrfs_delayed_extent_op *extent_op,
1617                                 int insert_reserved)
1618 {
1619         int ret = 0;
1620         struct btrfs_delayed_data_ref *ref;
1621         struct btrfs_key ins;
1622         u64 parent = 0;
1623         u64 ref_root = 0;
1624         u64 flags = 0;
1625
1626         ins.objectid = node->bytenr;
1627         ins.offset = node->num_bytes;
1628         ins.type = BTRFS_EXTENT_ITEM_KEY;
1629
1630         ref = btrfs_delayed_node_to_data_ref(node);
1631         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1632                 parent = ref->parent;
1633         else
1634                 ref_root = ref->root;
1635
1636         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1637                 if (extent_op) {
1638                         BUG_ON(extent_op->update_key);
1639                         flags |= extent_op->flags_to_set;
1640                 }
1641                 ret = alloc_reserved_file_extent(trans, root,
1642                                                  parent, ref_root, flags,
1643                                                  ref->objectid, ref->offset,
1644                                                  &ins, node->ref_mod);
1645                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1646         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1647                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1648                                              node->num_bytes, parent,
1649                                              ref_root, ref->objectid,
1650                                              ref->offset, node->ref_mod,
1651                                              extent_op);
1652         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1653                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1654                                           node->num_bytes, parent,
1655                                           ref_root, ref->objectid,
1656                                           ref->offset, node->ref_mod,
1657                                           extent_op);
1658         } else {
1659                 BUG();
1660         }
1661         return ret;
1662 }
1663
1664 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1665                                     struct extent_buffer *leaf,
1666                                     struct btrfs_extent_item *ei)
1667 {
1668         u64 flags = btrfs_extent_flags(leaf, ei);
1669         if (extent_op->update_flags) {
1670                 flags |= extent_op->flags_to_set;
1671                 btrfs_set_extent_flags(leaf, ei, flags);
1672         }
1673
1674         if (extent_op->update_key) {
1675                 struct btrfs_tree_block_info *bi;
1676                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1677                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1678                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1679         }
1680 }
1681
1682 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1683                                  struct btrfs_root *root,
1684                                  struct btrfs_delayed_ref_node *node,
1685                                  struct btrfs_delayed_extent_op *extent_op)
1686 {
1687         struct btrfs_key key;
1688         struct btrfs_path *path;
1689         struct btrfs_extent_item *ei;
1690         struct extent_buffer *leaf;
1691         u32 item_size;
1692         int ret;
1693         int err = 0;
1694
1695         path = btrfs_alloc_path();
1696         if (!path)
1697                 return -ENOMEM;
1698
1699         key.objectid = node->bytenr;
1700         key.type = BTRFS_EXTENT_ITEM_KEY;
1701         key.offset = node->num_bytes;
1702
1703         path->reada = 1;
1704         path->leave_spinning = 1;
1705         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1706                                 path, 0, 1);
1707         if (ret < 0) {
1708                 err = ret;
1709                 goto out;
1710         }
1711         if (ret > 0) {
1712                 err = -EIO;
1713                 goto out;
1714         }
1715
1716         leaf = path->nodes[0];
1717         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1719         if (item_size < sizeof(*ei)) {
1720                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1721                                              path, (u64)-1, 0);
1722                 if (ret < 0) {
1723                         err = ret;
1724                         goto out;
1725                 }
1726                 leaf = path->nodes[0];
1727                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1728         }
1729 #endif
1730         BUG_ON(item_size < sizeof(*ei));
1731         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1732         __run_delayed_extent_op(extent_op, leaf, ei);
1733
1734         btrfs_mark_buffer_dirty(leaf);
1735 out:
1736         btrfs_free_path(path);
1737         return err;
1738 }
1739
1740 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1741                                 struct btrfs_root *root,
1742                                 struct btrfs_delayed_ref_node *node,
1743                                 struct btrfs_delayed_extent_op *extent_op,
1744                                 int insert_reserved)
1745 {
1746         int ret = 0;
1747         struct btrfs_delayed_tree_ref *ref;
1748         struct btrfs_key ins;
1749         u64 parent = 0;
1750         u64 ref_root = 0;
1751
1752         ins.objectid = node->bytenr;
1753         ins.offset = node->num_bytes;
1754         ins.type = BTRFS_EXTENT_ITEM_KEY;
1755
1756         ref = btrfs_delayed_node_to_tree_ref(node);
1757         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1758                 parent = ref->parent;
1759         else
1760                 ref_root = ref->root;
1761
1762         BUG_ON(node->ref_mod != 1);
1763         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1764                 BUG_ON(!extent_op || !extent_op->update_flags ||
1765                        !extent_op->update_key);
1766                 ret = alloc_reserved_tree_block(trans, root,
1767                                                 parent, ref_root,
1768                                                 extent_op->flags_to_set,
1769                                                 &extent_op->key,
1770                                                 ref->level, &ins);
1771                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1772         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1773                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1774                                              node->num_bytes, parent, ref_root,
1775                                              ref->level, 0, 1, extent_op);
1776         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1777                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1778                                           node->num_bytes, parent, ref_root,
1779                                           ref->level, 0, 1, extent_op);
1780         } else {
1781                 BUG();
1782         }
1783         return ret;
1784 }
1785
1786
1787 /* helper function to actually process a single delayed ref entry */
1788 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1789                                struct btrfs_root *root,
1790                                struct btrfs_delayed_ref_node *node,
1791                                struct btrfs_delayed_extent_op *extent_op,
1792                                int insert_reserved)
1793 {
1794         int ret;
1795         if (btrfs_delayed_ref_is_head(node)) {
1796                 struct btrfs_delayed_ref_head *head;
1797                 /*
1798                  * we've hit the end of the chain and we were supposed
1799                  * to insert this extent into the tree.  But, it got
1800                  * deleted before we ever needed to insert it, so all
1801                  * we have to do is clean up the accounting
1802                  */
1803                 BUG_ON(extent_op);
1804                 head = btrfs_delayed_node_to_head(node);
1805                 if (insert_reserved) {
1806                         if (head->is_data) {
1807                                 ret = btrfs_del_csums(trans, root,
1808                                                       node->bytenr,
1809                                                       node->num_bytes);
1810                                 BUG_ON(ret);
1811                         }
1812                         btrfs_update_pinned_extents(root, node->bytenr,
1813                                                     node->num_bytes, 1);
1814                         update_reserved_extents(root, node->bytenr,
1815                                                 node->num_bytes, 0);
1816                 }
1817                 mutex_unlock(&head->mutex);
1818                 return 0;
1819         }
1820
1821         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1822             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1823                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1824                                            insert_reserved);
1825         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1826                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1827                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1828                                            insert_reserved);
1829         else
1830                 BUG();
1831         return ret;
1832 }
1833
1834 static noinline struct btrfs_delayed_ref_node *
1835 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1836 {
1837         struct rb_node *node;
1838         struct btrfs_delayed_ref_node *ref;
1839         int action = BTRFS_ADD_DELAYED_REF;
1840 again:
1841         /*
1842          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1843          * this prevents ref count from going down to zero when
1844          * there still are pending delayed ref.
1845          */
1846         node = rb_prev(&head->node.rb_node);
1847         while (1) {
1848                 if (!node)
1849                         break;
1850                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1851                                 rb_node);
1852                 if (ref->bytenr != head->node.bytenr)
1853                         break;
1854                 if (ref->action == action)
1855                         return ref;
1856                 node = rb_prev(node);
1857         }
1858         if (action == BTRFS_ADD_DELAYED_REF) {
1859                 action = BTRFS_DROP_DELAYED_REF;
1860                 goto again;
1861         }
1862         return NULL;
1863 }
1864
1865 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1866                                        struct btrfs_root *root,
1867                                        struct list_head *cluster)
1868 {
1869         struct btrfs_delayed_ref_root *delayed_refs;
1870         struct btrfs_delayed_ref_node *ref;
1871         struct btrfs_delayed_ref_head *locked_ref = NULL;
1872         struct btrfs_delayed_extent_op *extent_op;
1873         int ret;
1874         int count = 0;
1875         int must_insert_reserved = 0;
1876
1877         delayed_refs = &trans->transaction->delayed_refs;
1878         while (1) {
1879                 if (!locked_ref) {
1880                         /* pick a new head ref from the cluster list */
1881                         if (list_empty(cluster))
1882                                 break;
1883
1884                         locked_ref = list_entry(cluster->next,
1885                                      struct btrfs_delayed_ref_head, cluster);
1886
1887                         /* grab the lock that says we are going to process
1888                          * all the refs for this head */
1889                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1890
1891                         /*
1892                          * we may have dropped the spin lock to get the head
1893                          * mutex lock, and that might have given someone else
1894                          * time to free the head.  If that's true, it has been
1895                          * removed from our list and we can move on.
1896                          */
1897                         if (ret == -EAGAIN) {
1898                                 locked_ref = NULL;
1899                                 count++;
1900                                 continue;
1901                         }
1902                 }
1903
1904                 /*
1905                  * record the must insert reserved flag before we
1906                  * drop the spin lock.
1907                  */
1908                 must_insert_reserved = locked_ref->must_insert_reserved;
1909                 locked_ref->must_insert_reserved = 0;
1910
1911                 extent_op = locked_ref->extent_op;
1912                 locked_ref->extent_op = NULL;
1913
1914                 /*
1915                  * locked_ref is the head node, so we have to go one
1916                  * node back for any delayed ref updates
1917                  */
1918                 ref = select_delayed_ref(locked_ref);
1919                 if (!ref) {
1920                         /* All delayed refs have been processed, Go ahead
1921                          * and send the head node to run_one_delayed_ref,
1922                          * so that any accounting fixes can happen
1923                          */
1924                         ref = &locked_ref->node;
1925
1926                         if (extent_op && must_insert_reserved) {
1927                                 kfree(extent_op);
1928                                 extent_op = NULL;
1929                         }
1930
1931                         if (extent_op) {
1932                                 spin_unlock(&delayed_refs->lock);
1933
1934                                 ret = run_delayed_extent_op(trans, root,
1935                                                             ref, extent_op);
1936                                 BUG_ON(ret);
1937                                 kfree(extent_op);
1938
1939                                 cond_resched();
1940                                 spin_lock(&delayed_refs->lock);
1941                                 continue;
1942                         }
1943
1944                         list_del_init(&locked_ref->cluster);
1945                         locked_ref = NULL;
1946                 }
1947
1948                 ref->in_tree = 0;
1949                 rb_erase(&ref->rb_node, &delayed_refs->root);
1950                 delayed_refs->num_entries--;
1951
1952                 spin_unlock(&delayed_refs->lock);
1953
1954                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
1955                                           must_insert_reserved);
1956                 BUG_ON(ret);
1957
1958                 btrfs_put_delayed_ref(ref);
1959                 kfree(extent_op);
1960                 count++;
1961
1962                 cond_resched();
1963                 spin_lock(&delayed_refs->lock);
1964         }
1965         return count;
1966 }
1967
1968 /*
1969  * this starts processing the delayed reference count updates and
1970  * extent insertions we have queued up so far.  count can be
1971  * 0, which means to process everything in the tree at the start
1972  * of the run (but not newly added entries), or it can be some target
1973  * number you'd like to process.
1974  */
1975 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1976                            struct btrfs_root *root, unsigned long count)
1977 {
1978         struct rb_node *node;
1979         struct btrfs_delayed_ref_root *delayed_refs;
1980         struct btrfs_delayed_ref_node *ref;
1981         struct list_head cluster;
1982         int ret;
1983         int run_all = count == (unsigned long)-1;
1984         int run_most = 0;
1985
1986         if (root == root->fs_info->extent_root)
1987                 root = root->fs_info->tree_root;
1988
1989         delayed_refs = &trans->transaction->delayed_refs;
1990         INIT_LIST_HEAD(&cluster);
1991 again:
1992         spin_lock(&delayed_refs->lock);
1993         if (count == 0) {
1994                 count = delayed_refs->num_entries * 2;
1995                 run_most = 1;
1996         }
1997         while (1) {
1998                 if (!(run_all || run_most) &&
1999                     delayed_refs->num_heads_ready < 64)
2000                         break;
2001
2002                 /*
2003                  * go find something we can process in the rbtree.  We start at
2004                  * the beginning of the tree, and then build a cluster
2005                  * of refs to process starting at the first one we are able to
2006                  * lock
2007                  */
2008                 ret = btrfs_find_ref_cluster(trans, &cluster,
2009                                              delayed_refs->run_delayed_start);
2010                 if (ret)
2011                         break;
2012
2013                 ret = run_clustered_refs(trans, root, &cluster);
2014                 BUG_ON(ret < 0);
2015
2016                 count -= min_t(unsigned long, ret, count);
2017
2018                 if (count == 0)
2019                         break;
2020         }
2021
2022         if (run_all) {
2023                 node = rb_first(&delayed_refs->root);
2024                 if (!node)
2025                         goto out;
2026                 count = (unsigned long)-1;
2027
2028                 while (node) {
2029                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2030                                        rb_node);
2031                         if (btrfs_delayed_ref_is_head(ref)) {
2032                                 struct btrfs_delayed_ref_head *head;
2033
2034                                 head = btrfs_delayed_node_to_head(ref);
2035                                 atomic_inc(&ref->refs);
2036
2037                                 spin_unlock(&delayed_refs->lock);
2038                                 mutex_lock(&head->mutex);
2039                                 mutex_unlock(&head->mutex);
2040
2041                                 btrfs_put_delayed_ref(ref);
2042                                 cond_resched();
2043                                 goto again;
2044                         }
2045                         node = rb_next(node);
2046                 }
2047                 spin_unlock(&delayed_refs->lock);
2048                 schedule_timeout(1);
2049                 goto again;
2050         }
2051 out:
2052         spin_unlock(&delayed_refs->lock);
2053         return 0;
2054 }
2055
2056 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2057                                 struct btrfs_root *root,
2058                                 u64 bytenr, u64 num_bytes, u64 flags,
2059                                 int is_data)
2060 {
2061         struct btrfs_delayed_extent_op *extent_op;
2062         int ret;
2063
2064         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2065         if (!extent_op)
2066                 return -ENOMEM;
2067
2068         extent_op->flags_to_set = flags;
2069         extent_op->update_flags = 1;
2070         extent_op->update_key = 0;
2071         extent_op->is_data = is_data ? 1 : 0;
2072
2073         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2074         if (ret)
2075                 kfree(extent_op);
2076         return ret;
2077 }
2078
2079 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2080                                       struct btrfs_root *root,
2081                                       struct btrfs_path *path,
2082                                       u64 objectid, u64 offset, u64 bytenr)
2083 {
2084         struct btrfs_delayed_ref_head *head;
2085         struct btrfs_delayed_ref_node *ref;
2086         struct btrfs_delayed_data_ref *data_ref;
2087         struct btrfs_delayed_ref_root *delayed_refs;
2088         struct rb_node *node;
2089         int ret = 0;
2090
2091         ret = -ENOENT;
2092         delayed_refs = &trans->transaction->delayed_refs;
2093         spin_lock(&delayed_refs->lock);
2094         head = btrfs_find_delayed_ref_head(trans, bytenr);
2095         if (!head)
2096                 goto out;
2097
2098         if (!mutex_trylock(&head->mutex)) {
2099                 atomic_inc(&head->node.refs);
2100                 spin_unlock(&delayed_refs->lock);
2101
2102                 btrfs_release_path(root->fs_info->extent_root, path);
2103
2104                 mutex_lock(&head->mutex);
2105                 mutex_unlock(&head->mutex);
2106                 btrfs_put_delayed_ref(&head->node);
2107                 return -EAGAIN;
2108         }
2109
2110         node = rb_prev(&head->node.rb_node);
2111         if (!node)
2112                 goto out_unlock;
2113
2114         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2115
2116         if (ref->bytenr != bytenr)
2117                 goto out_unlock;
2118
2119         ret = 1;
2120         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2121                 goto out_unlock;
2122
2123         data_ref = btrfs_delayed_node_to_data_ref(ref);
2124
2125         node = rb_prev(node);
2126         if (node) {
2127                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2128                 if (ref->bytenr == bytenr)
2129                         goto out_unlock;
2130         }
2131
2132         if (data_ref->root != root->root_key.objectid ||
2133             data_ref->objectid != objectid || data_ref->offset != offset)
2134                 goto out_unlock;
2135
2136         ret = 0;
2137 out_unlock:
2138         mutex_unlock(&head->mutex);
2139 out:
2140         spin_unlock(&delayed_refs->lock);
2141         return ret;
2142 }
2143
2144 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2145                                         struct btrfs_root *root,
2146                                         struct btrfs_path *path,
2147                                         u64 objectid, u64 offset, u64 bytenr)
2148 {
2149         struct btrfs_root *extent_root = root->fs_info->extent_root;
2150         struct extent_buffer *leaf;
2151         struct btrfs_extent_data_ref *ref;
2152         struct btrfs_extent_inline_ref *iref;
2153         struct btrfs_extent_item *ei;
2154         struct btrfs_key key;
2155         u32 item_size;
2156         int ret;
2157
2158         key.objectid = bytenr;
2159         key.offset = (u64)-1;
2160         key.type = BTRFS_EXTENT_ITEM_KEY;
2161
2162         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2163         if (ret < 0)
2164                 goto out;
2165         BUG_ON(ret == 0);
2166
2167         ret = -ENOENT;
2168         if (path->slots[0] == 0)
2169                 goto out;
2170
2171         path->slots[0]--;
2172         leaf = path->nodes[0];
2173         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2174
2175         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2176                 goto out;
2177
2178         ret = 1;
2179         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2181         if (item_size < sizeof(*ei)) {
2182                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2183                 goto out;
2184         }
2185 #endif
2186         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2187
2188         if (item_size != sizeof(*ei) +
2189             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2190                 goto out;
2191
2192         if (btrfs_extent_generation(leaf, ei) <=
2193             btrfs_root_last_snapshot(&root->root_item))
2194                 goto out;
2195
2196         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2197         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2198             BTRFS_EXTENT_DATA_REF_KEY)
2199                 goto out;
2200
2201         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2202         if (btrfs_extent_refs(leaf, ei) !=
2203             btrfs_extent_data_ref_count(leaf, ref) ||
2204             btrfs_extent_data_ref_root(leaf, ref) !=
2205             root->root_key.objectid ||
2206             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2207             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2208                 goto out;
2209
2210         ret = 0;
2211 out:
2212         return ret;
2213 }
2214
2215 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2216                           struct btrfs_root *root,
2217                           u64 objectid, u64 offset, u64 bytenr)
2218 {
2219         struct btrfs_path *path;
2220         int ret;
2221         int ret2;
2222
2223         path = btrfs_alloc_path();
2224         if (!path)
2225                 return -ENOENT;
2226
2227         do {
2228                 ret = check_committed_ref(trans, root, path, objectid,
2229                                           offset, bytenr);
2230                 if (ret && ret != -ENOENT)
2231                         goto out;
2232
2233                 ret2 = check_delayed_ref(trans, root, path, objectid,
2234                                          offset, bytenr);
2235         } while (ret2 == -EAGAIN);
2236
2237         if (ret2 && ret2 != -ENOENT) {
2238                 ret = ret2;
2239                 goto out;
2240         }
2241
2242         if (ret != -ENOENT || ret2 != -ENOENT)
2243                 ret = 0;
2244 out:
2245         btrfs_free_path(path);
2246         return ret;
2247 }
2248
2249 #if 0
2250 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2251                     struct extent_buffer *buf, u32 nr_extents)
2252 {
2253         struct btrfs_key key;
2254         struct btrfs_file_extent_item *fi;
2255         u64 root_gen;
2256         u32 nritems;
2257         int i;
2258         int level;
2259         int ret = 0;
2260         int shared = 0;
2261
2262         if (!root->ref_cows)
2263                 return 0;
2264
2265         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2266                 shared = 0;
2267                 root_gen = root->root_key.offset;
2268         } else {
2269                 shared = 1;
2270                 root_gen = trans->transid - 1;
2271         }
2272
2273         level = btrfs_header_level(buf);
2274         nritems = btrfs_header_nritems(buf);
2275
2276         if (level == 0) {
2277                 struct btrfs_leaf_ref *ref;
2278                 struct btrfs_extent_info *info;
2279
2280                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2281                 if (!ref) {
2282                         ret = -ENOMEM;
2283                         goto out;
2284                 }
2285
2286                 ref->root_gen = root_gen;
2287                 ref->bytenr = buf->start;
2288                 ref->owner = btrfs_header_owner(buf);
2289                 ref->generation = btrfs_header_generation(buf);
2290                 ref->nritems = nr_extents;
2291                 info = ref->extents;
2292
2293                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2294                         u64 disk_bytenr;
2295                         btrfs_item_key_to_cpu(buf, &key, i);
2296                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2297                                 continue;
2298                         fi = btrfs_item_ptr(buf, i,
2299                                             struct btrfs_file_extent_item);
2300                         if (btrfs_file_extent_type(buf, fi) ==
2301                             BTRFS_FILE_EXTENT_INLINE)
2302                                 continue;
2303                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2304                         if (disk_bytenr == 0)
2305                                 continue;
2306
2307                         info->bytenr = disk_bytenr;
2308                         info->num_bytes =
2309                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2310                         info->objectid = key.objectid;
2311                         info->offset = key.offset;
2312                         info++;
2313                 }
2314
2315                 ret = btrfs_add_leaf_ref(root, ref, shared);
2316                 if (ret == -EEXIST && shared) {
2317                         struct btrfs_leaf_ref *old;
2318                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2319                         BUG_ON(!old);
2320                         btrfs_remove_leaf_ref(root, old);
2321                         btrfs_free_leaf_ref(root, old);
2322                         ret = btrfs_add_leaf_ref(root, ref, shared);
2323                 }
2324                 WARN_ON(ret);
2325                 btrfs_free_leaf_ref(root, ref);
2326         }
2327 out:
2328         return ret;
2329 }
2330
2331 /* when a block goes through cow, we update the reference counts of
2332  * everything that block points to.  The internal pointers of the block
2333  * can be in just about any order, and it is likely to have clusters of
2334  * things that are close together and clusters of things that are not.
2335  *
2336  * To help reduce the seeks that come with updating all of these reference
2337  * counts, sort them by byte number before actual updates are done.
2338  *
2339  * struct refsort is used to match byte number to slot in the btree block.
2340  * we sort based on the byte number and then use the slot to actually
2341  * find the item.
2342  *
2343  * struct refsort is smaller than strcut btrfs_item and smaller than
2344  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2345  * for a btree block, there's no way for a kmalloc of refsorts for a
2346  * single node to be bigger than a page.
2347  */
2348 struct refsort {
2349         u64 bytenr;
2350         u32 slot;
2351 };
2352
2353 /*
2354  * for passing into sort()
2355  */
2356 static int refsort_cmp(const void *a_void, const void *b_void)
2357 {
2358         const struct refsort *a = a_void;
2359         const struct refsort *b = b_void;
2360
2361         if (a->bytenr < b->bytenr)
2362                 return -1;
2363         if (a->bytenr > b->bytenr)
2364                 return 1;
2365         return 0;
2366 }
2367 #endif
2368
2369 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2370                            struct btrfs_root *root,
2371                            struct extent_buffer *buf,
2372                            int full_backref, int inc)
2373 {
2374         u64 bytenr;
2375         u64 num_bytes;
2376         u64 parent;
2377         u64 ref_root;
2378         u32 nritems;
2379         struct btrfs_key key;
2380         struct btrfs_file_extent_item *fi;
2381         int i;
2382         int level;
2383         int ret = 0;
2384         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2385                             u64, u64, u64, u64, u64, u64);
2386
2387         ref_root = btrfs_header_owner(buf);
2388         nritems = btrfs_header_nritems(buf);
2389         level = btrfs_header_level(buf);
2390
2391         if (!root->ref_cows && level == 0)
2392                 return 0;
2393
2394         if (inc)
2395                 process_func = btrfs_inc_extent_ref;
2396         else
2397                 process_func = btrfs_free_extent;
2398
2399         if (full_backref)
2400                 parent = buf->start;
2401         else
2402                 parent = 0;
2403
2404         for (i = 0; i < nritems; i++) {
2405                 if (level == 0) {
2406                         btrfs_item_key_to_cpu(buf, &key, i);
2407                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2408                                 continue;
2409                         fi = btrfs_item_ptr(buf, i,
2410                                             struct btrfs_file_extent_item);
2411                         if (btrfs_file_extent_type(buf, fi) ==
2412                             BTRFS_FILE_EXTENT_INLINE)
2413                                 continue;
2414                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2415                         if (bytenr == 0)
2416                                 continue;
2417
2418                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2419                         key.offset -= btrfs_file_extent_offset(buf, fi);
2420                         ret = process_func(trans, root, bytenr, num_bytes,
2421                                            parent, ref_root, key.objectid,
2422                                            key.offset);
2423                         if (ret)
2424                                 goto fail;
2425                 } else {
2426                         bytenr = btrfs_node_blockptr(buf, i);
2427                         num_bytes = btrfs_level_size(root, level - 1);
2428                         ret = process_func(trans, root, bytenr, num_bytes,
2429                                            parent, ref_root, level - 1, 0);
2430                         if (ret)
2431                                 goto fail;
2432                 }
2433         }
2434         return 0;
2435 fail:
2436         BUG();
2437         return ret;
2438 }
2439
2440 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2441                   struct extent_buffer *buf, int full_backref)
2442 {
2443         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2444 }
2445
2446 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2447                   struct extent_buffer *buf, int full_backref)
2448 {
2449         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2450 }
2451
2452 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2453                                  struct btrfs_root *root,
2454                                  struct btrfs_path *path,
2455                                  struct btrfs_block_group_cache *cache)
2456 {
2457         int ret;
2458         struct btrfs_root *extent_root = root->fs_info->extent_root;
2459         unsigned long bi;
2460         struct extent_buffer *leaf;
2461
2462         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2463         if (ret < 0)
2464                 goto fail;
2465         BUG_ON(ret);
2466
2467         leaf = path->nodes[0];
2468         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2469         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2470         btrfs_mark_buffer_dirty(leaf);
2471         btrfs_release_path(extent_root, path);
2472 fail:
2473         if (ret)
2474                 return ret;
2475         return 0;
2476
2477 }
2478
2479 static struct btrfs_block_group_cache *
2480 next_block_group(struct btrfs_root *root,
2481                  struct btrfs_block_group_cache *cache)
2482 {
2483         struct rb_node *node;
2484         spin_lock(&root->fs_info->block_group_cache_lock);
2485         node = rb_next(&cache->cache_node);
2486         btrfs_put_block_group(cache);
2487         if (node) {
2488                 cache = rb_entry(node, struct btrfs_block_group_cache,
2489                                  cache_node);
2490                 atomic_inc(&cache->count);
2491         } else
2492                 cache = NULL;
2493         spin_unlock(&root->fs_info->block_group_cache_lock);
2494         return cache;
2495 }
2496
2497 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2498                                    struct btrfs_root *root)
2499 {
2500         struct btrfs_block_group_cache *cache;
2501         int err = 0;
2502         struct btrfs_path *path;
2503         u64 last = 0;
2504
2505         path = btrfs_alloc_path();
2506         if (!path)
2507                 return -ENOMEM;
2508
2509         while (1) {
2510                 if (last == 0) {
2511                         err = btrfs_run_delayed_refs(trans, root,
2512                                                      (unsigned long)-1);
2513                         BUG_ON(err);
2514                 }
2515
2516                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2517                 while (cache) {
2518                         if (cache->dirty)
2519                                 break;
2520                         cache = next_block_group(root, cache);
2521                 }
2522                 if (!cache) {
2523                         if (last == 0)
2524                                 break;
2525                         last = 0;
2526                         continue;
2527                 }
2528
2529                 cache->dirty = 0;
2530                 last = cache->key.objectid + cache->key.offset;
2531
2532                 err = write_one_cache_group(trans, root, path, cache);
2533                 BUG_ON(err);
2534                 btrfs_put_block_group(cache);
2535         }
2536
2537         btrfs_free_path(path);
2538         return 0;
2539 }
2540
2541 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2542 {
2543         struct btrfs_block_group_cache *block_group;
2544         int readonly = 0;
2545
2546         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2547         if (!block_group || block_group->ro)
2548                 readonly = 1;
2549         if (block_group)
2550                 btrfs_put_block_group(block_group);
2551         return readonly;
2552 }
2553
2554 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2555                              u64 total_bytes, u64 bytes_used,
2556                              struct btrfs_space_info **space_info)
2557 {
2558         struct btrfs_space_info *found;
2559
2560         found = __find_space_info(info, flags);
2561         if (found) {
2562                 spin_lock(&found->lock);
2563                 found->total_bytes += total_bytes;
2564                 found->bytes_used += bytes_used;
2565                 found->full = 0;
2566                 spin_unlock(&found->lock);
2567                 *space_info = found;
2568                 return 0;
2569         }
2570         found = kzalloc(sizeof(*found), GFP_NOFS);
2571         if (!found)
2572                 return -ENOMEM;
2573
2574         INIT_LIST_HEAD(&found->block_groups);
2575         init_rwsem(&found->groups_sem);
2576         spin_lock_init(&found->lock);
2577         found->flags = flags;
2578         found->total_bytes = total_bytes;
2579         found->bytes_used = bytes_used;
2580         found->bytes_pinned = 0;
2581         found->bytes_reserved = 0;
2582         found->bytes_readonly = 0;
2583         found->bytes_delalloc = 0;
2584         found->full = 0;
2585         found->force_alloc = 0;
2586         *space_info = found;
2587         list_add_rcu(&found->list, &info->space_info);
2588         atomic_set(&found->caching_threads, 0);
2589         return 0;
2590 }
2591
2592 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2593 {
2594         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2595                                    BTRFS_BLOCK_GROUP_RAID1 |
2596                                    BTRFS_BLOCK_GROUP_RAID10 |
2597                                    BTRFS_BLOCK_GROUP_DUP);
2598         if (extra_flags) {
2599                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2600                         fs_info->avail_data_alloc_bits |= extra_flags;
2601                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2602                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2603                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2604                         fs_info->avail_system_alloc_bits |= extra_flags;
2605         }
2606 }
2607
2608 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2609 {
2610         spin_lock(&cache->space_info->lock);
2611         spin_lock(&cache->lock);
2612         if (!cache->ro) {
2613                 cache->space_info->bytes_readonly += cache->key.offset -
2614                                         btrfs_block_group_used(&cache->item);
2615                 cache->ro = 1;
2616         }
2617         spin_unlock(&cache->lock);
2618         spin_unlock(&cache->space_info->lock);
2619 }
2620
2621 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2622 {
2623         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2624
2625         if (num_devices == 1)
2626                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2627         if (num_devices < 4)
2628                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2629
2630         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2631             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2632                       BTRFS_BLOCK_GROUP_RAID10))) {
2633                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2634         }
2635
2636         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2637             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2638                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2639         }
2640
2641         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2642             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2643              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2644              (flags & BTRFS_BLOCK_GROUP_DUP)))
2645                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2646         return flags;
2647 }
2648
2649 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2650 {
2651         struct btrfs_fs_info *info = root->fs_info;
2652         u64 alloc_profile;
2653
2654         if (data) {
2655                 alloc_profile = info->avail_data_alloc_bits &
2656                         info->data_alloc_profile;
2657                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2658         } else if (root == root->fs_info->chunk_root) {
2659                 alloc_profile = info->avail_system_alloc_bits &
2660                         info->system_alloc_profile;
2661                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2662         } else {
2663                 alloc_profile = info->avail_metadata_alloc_bits &
2664                         info->metadata_alloc_profile;
2665                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2666         }
2667
2668         return btrfs_reduce_alloc_profile(root, data);
2669 }
2670
2671 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2672 {
2673         u64 alloc_target;
2674
2675         alloc_target = btrfs_get_alloc_profile(root, 1);
2676         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2677                                                        alloc_target);
2678 }
2679
2680 /*
2681  * for now this just makes sure we have at least 5% of our metadata space free
2682  * for use.
2683  */
2684 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2685 {
2686         struct btrfs_fs_info *info = root->fs_info;
2687         struct btrfs_space_info *meta_sinfo;
2688         u64 alloc_target, thresh;
2689         int committed = 0, ret;
2690
2691         /* get the space info for where the metadata will live */
2692         alloc_target = btrfs_get_alloc_profile(root, 0);
2693         meta_sinfo = __find_space_info(info, alloc_target);
2694
2695 again:
2696         spin_lock(&meta_sinfo->lock);
2697         if (!meta_sinfo->full)
2698                 thresh = meta_sinfo->total_bytes * 80;
2699         else
2700                 thresh = meta_sinfo->total_bytes * 95;
2701
2702         do_div(thresh, 100);
2703
2704         if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2705             meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2706                 struct btrfs_trans_handle *trans;
2707                 if (!meta_sinfo->full) {
2708                         meta_sinfo->force_alloc = 1;
2709                         spin_unlock(&meta_sinfo->lock);
2710
2711                         trans = btrfs_start_transaction(root, 1);
2712                         if (!trans)
2713                                 return -ENOMEM;
2714
2715                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2716                                              2 * 1024 * 1024, alloc_target, 0);
2717                         btrfs_end_transaction(trans, root);
2718                         goto again;
2719                 }
2720                 spin_unlock(&meta_sinfo->lock);
2721
2722                 if (!committed) {
2723                         committed = 1;
2724                         trans = btrfs_join_transaction(root, 1);
2725                         if (!trans)
2726                                 return -ENOMEM;
2727                         ret = btrfs_commit_transaction(trans, root);
2728                         if (ret)
2729                                 return ret;
2730                         goto again;
2731                 }
2732                 return -ENOSPC;
2733         }
2734         spin_unlock(&meta_sinfo->lock);
2735
2736         return 0;
2737 }
2738
2739 /*
2740  * This will check the space that the inode allocates from to make sure we have
2741  * enough space for bytes.
2742  */
2743 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2744                                 u64 bytes)
2745 {
2746         struct btrfs_space_info *data_sinfo;
2747         int ret = 0, committed = 0;
2748
2749         /* make sure bytes are sectorsize aligned */
2750         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2751
2752         data_sinfo = BTRFS_I(inode)->space_info;
2753 again:
2754         /* make sure we have enough space to handle the data first */
2755         spin_lock(&data_sinfo->lock);
2756         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2757             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2758             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2759             data_sinfo->bytes_may_use < bytes) {
2760                 struct btrfs_trans_handle *trans;
2761
2762                 /*
2763                  * if we don't have enough free bytes in this space then we need
2764                  * to alloc a new chunk.
2765                  */
2766                 if (!data_sinfo->full) {
2767                         u64 alloc_target;
2768
2769                         data_sinfo->force_alloc = 1;
2770                         spin_unlock(&data_sinfo->lock);
2771
2772                         alloc_target = btrfs_get_alloc_profile(root, 1);
2773                         trans = btrfs_start_transaction(root, 1);
2774                         if (!trans)
2775                                 return -ENOMEM;
2776
2777                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2778                                              bytes + 2 * 1024 * 1024,
2779                                              alloc_target, 0);
2780                         btrfs_end_transaction(trans, root);
2781                         if (ret)
2782                                 return ret;
2783                         goto again;
2784                 }
2785                 spin_unlock(&data_sinfo->lock);
2786
2787                 /* commit the current transaction and try again */
2788                 if (!committed) {
2789                         committed = 1;
2790                         trans = btrfs_join_transaction(root, 1);
2791                         if (!trans)
2792                                 return -ENOMEM;
2793                         ret = btrfs_commit_transaction(trans, root);
2794                         if (ret)
2795                                 return ret;
2796                         goto again;
2797                 }
2798
2799                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2800                        ", %llu bytes_used, %llu bytes_reserved, "
2801                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2802                        "%llu total\n", (unsigned long long)bytes,
2803                        (unsigned long long)data_sinfo->bytes_delalloc,
2804                        (unsigned long long)data_sinfo->bytes_used,
2805                        (unsigned long long)data_sinfo->bytes_reserved,
2806                        (unsigned long long)data_sinfo->bytes_pinned,
2807                        (unsigned long long)data_sinfo->bytes_readonly,
2808                        (unsigned long long)data_sinfo->bytes_may_use,
2809                        (unsigned long long)data_sinfo->total_bytes);
2810                 return -ENOSPC;
2811         }
2812         data_sinfo->bytes_may_use += bytes;
2813         BTRFS_I(inode)->reserved_bytes += bytes;
2814         spin_unlock(&data_sinfo->lock);
2815
2816         return btrfs_check_metadata_free_space(root);
2817 }
2818
2819 /*
2820  * if there was an error for whatever reason after calling
2821  * btrfs_check_data_free_space, call this so we can cleanup the counters.
2822  */
2823 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2824                                     struct inode *inode, u64 bytes)
2825 {
2826         struct btrfs_space_info *data_sinfo;
2827
2828         /* make sure bytes are sectorsize aligned */
2829         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2830
2831         data_sinfo = BTRFS_I(inode)->space_info;
2832         spin_lock(&data_sinfo->lock);
2833         data_sinfo->bytes_may_use -= bytes;
2834         BTRFS_I(inode)->reserved_bytes -= bytes;
2835         spin_unlock(&data_sinfo->lock);
2836 }
2837
2838 /* called when we are adding a delalloc extent to the inode's io_tree */
2839 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2840                                   u64 bytes)
2841 {
2842         struct btrfs_space_info *data_sinfo;
2843
2844         /* get the space info for where this inode will be storing its data */
2845         data_sinfo = BTRFS_I(inode)->space_info;
2846
2847         /* make sure we have enough space to handle the data first */
2848         spin_lock(&data_sinfo->lock);
2849         data_sinfo->bytes_delalloc += bytes;
2850
2851         /*
2852          * we are adding a delalloc extent without calling
2853          * btrfs_check_data_free_space first.  This happens on a weird
2854          * writepage condition, but shouldn't hurt our accounting
2855          */
2856         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2857                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2858                 BTRFS_I(inode)->reserved_bytes = 0;
2859         } else {
2860                 data_sinfo->bytes_may_use -= bytes;
2861                 BTRFS_I(inode)->reserved_bytes -= bytes;
2862         }
2863
2864         spin_unlock(&data_sinfo->lock);
2865 }
2866
2867 /* called when we are clearing an delalloc extent from the inode's io_tree */
2868 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2869                               u64 bytes)
2870 {
2871         struct btrfs_space_info *info;
2872
2873         info = BTRFS_I(inode)->space_info;
2874
2875         spin_lock(&info->lock);
2876         info->bytes_delalloc -= bytes;
2877         spin_unlock(&info->lock);
2878 }
2879
2880 static void force_metadata_allocation(struct btrfs_fs_info *info)
2881 {
2882         struct list_head *head = &info->space_info;
2883         struct btrfs_space_info *found;
2884
2885         rcu_read_lock();
2886         list_for_each_entry_rcu(found, head, list) {
2887                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2888                         found->force_alloc = 1;
2889         }
2890         rcu_read_unlock();
2891 }
2892
2893 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2894                           struct btrfs_root *extent_root, u64 alloc_bytes,
2895                           u64 flags, int force)
2896 {
2897         struct btrfs_space_info *space_info;
2898         struct btrfs_fs_info *fs_info = extent_root->fs_info;
2899         u64 thresh;
2900         int ret = 0;
2901
2902         mutex_lock(&fs_info->chunk_mutex);
2903
2904         flags = btrfs_reduce_alloc_profile(extent_root, flags);
2905
2906         space_info = __find_space_info(extent_root->fs_info, flags);
2907         if (!space_info) {
2908                 ret = update_space_info(extent_root->fs_info, flags,
2909                                         0, 0, &space_info);
2910                 BUG_ON(ret);
2911         }
2912         BUG_ON(!space_info);
2913
2914         spin_lock(&space_info->lock);
2915         if (space_info->force_alloc) {
2916                 force = 1;
2917                 space_info->force_alloc = 0;
2918         }
2919         if (space_info->full) {
2920                 spin_unlock(&space_info->lock);
2921                 goto out;
2922         }
2923
2924         thresh = space_info->total_bytes - space_info->bytes_readonly;
2925         thresh = div_factor(thresh, 6);
2926         if (!force &&
2927            (space_info->bytes_used + space_info->bytes_pinned +
2928             space_info->bytes_reserved + alloc_bytes) < thresh) {
2929                 spin_unlock(&space_info->lock);
2930                 goto out;
2931         }
2932         spin_unlock(&space_info->lock);
2933
2934         /*
2935          * if we're doing a data chunk, go ahead and make sure that
2936          * we keep a reasonable number of metadata chunks allocated in the
2937          * FS as well.
2938          */
2939         if (flags & BTRFS_BLOCK_GROUP_DATA) {
2940                 fs_info->data_chunk_allocations++;
2941                 if (!(fs_info->data_chunk_allocations %
2942                       fs_info->metadata_ratio))
2943                         force_metadata_allocation(fs_info);
2944         }
2945
2946         ret = btrfs_alloc_chunk(trans, extent_root, flags);
2947         if (ret)
2948                 space_info->full = 1;
2949 out:
2950         mutex_unlock(&extent_root->fs_info->chunk_mutex);
2951         return ret;
2952 }
2953
2954 static int update_block_group(struct btrfs_trans_handle *trans,
2955                               struct btrfs_root *root,
2956                               u64 bytenr, u64 num_bytes, int alloc,
2957                               int mark_free)
2958 {
2959         struct btrfs_block_group_cache *cache;
2960         struct btrfs_fs_info *info = root->fs_info;
2961         u64 total = num_bytes;
2962         u64 old_val;
2963         u64 byte_in_group;
2964
2965         /* block accounting for super block */
2966         spin_lock(&info->delalloc_lock);
2967         old_val = btrfs_super_bytes_used(&info->super_copy);
2968         if (alloc)
2969                 old_val += num_bytes;
2970         else
2971                 old_val -= num_bytes;
2972         btrfs_set_super_bytes_used(&info->super_copy, old_val);
2973
2974         /* block accounting for root item */
2975         old_val = btrfs_root_used(&root->root_item);
2976         if (alloc)
2977                 old_val += num_bytes;
2978         else
2979                 old_val -= num_bytes;
2980         btrfs_set_root_used(&root->root_item, old_val);
2981         spin_unlock(&info->delalloc_lock);
2982
2983         while (total) {
2984                 cache = btrfs_lookup_block_group(info, bytenr);
2985                 if (!cache)
2986                         return -1;
2987                 byte_in_group = bytenr - cache->key.objectid;
2988                 WARN_ON(byte_in_group > cache->key.offset);
2989
2990                 spin_lock(&cache->space_info->lock);
2991                 spin_lock(&cache->lock);
2992                 cache->dirty = 1;
2993                 old_val = btrfs_block_group_used(&cache->item);
2994                 num_bytes = min(total, cache->key.offset - byte_in_group);
2995                 if (alloc) {
2996                         old_val += num_bytes;
2997                         cache->space_info->bytes_used += num_bytes;
2998                         if (cache->ro)
2999                                 cache->space_info->bytes_readonly -= num_bytes;
3000                         btrfs_set_block_group_used(&cache->item, old_val);
3001                         spin_unlock(&cache->lock);
3002                         spin_unlock(&cache->space_info->lock);
3003                 } else {
3004                         old_val -= num_bytes;
3005                         cache->space_info->bytes_used -= num_bytes;
3006                         if (cache->ro)
3007                                 cache->space_info->bytes_readonly += num_bytes;
3008                         btrfs_set_block_group_used(&cache->item, old_val);
3009                         spin_unlock(&cache->lock);
3010                         spin_unlock(&cache->space_info->lock);
3011                         if (mark_free) {
3012                                 int ret;
3013
3014                                 ret = btrfs_discard_extent(root, bytenr,
3015                                                            num_bytes);
3016                                 WARN_ON(ret);
3017
3018                                 ret = btrfs_add_free_space(cache, bytenr,
3019                                                            num_bytes);
3020                                 WARN_ON(ret);
3021                         }
3022                 }
3023                 btrfs_put_block_group(cache);
3024                 total -= num_bytes;
3025                 bytenr += num_bytes;
3026         }
3027         return 0;
3028 }
3029
3030 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3031 {
3032         struct btrfs_block_group_cache *cache;
3033         u64 bytenr;
3034
3035         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3036         if (!cache)
3037                 return 0;
3038
3039         bytenr = cache->key.objectid;
3040         btrfs_put_block_group(cache);
3041
3042         return bytenr;
3043 }
3044
3045 int btrfs_update_pinned_extents(struct btrfs_root *root,
3046                                 u64 bytenr, u64 num, int pin)
3047 {
3048         u64 len;
3049         struct btrfs_block_group_cache *cache;
3050         struct btrfs_fs_info *fs_info = root->fs_info;
3051
3052         if (pin)
3053                 set_extent_dirty(&fs_info->pinned_extents,
3054                                 bytenr, bytenr + num - 1, GFP_NOFS);
3055
3056         while (num > 0) {
3057                 cache = btrfs_lookup_block_group(fs_info, bytenr);
3058                 BUG_ON(!cache);
3059                 len = min(num, cache->key.offset -
3060                           (bytenr - cache->key.objectid));
3061                 if (pin) {
3062                         spin_lock(&cache->space_info->lock);
3063                         spin_lock(&cache->lock);
3064                         cache->pinned += len;
3065                         cache->space_info->bytes_pinned += len;
3066                         spin_unlock(&cache->lock);
3067                         spin_unlock(&cache->space_info->lock);
3068                         fs_info->total_pinned += len;
3069                 } else {
3070                         int unpin = 0;
3071
3072                         /*
3073                          * in order to not race with the block group caching, we
3074                          * only want to unpin the extent if we are cached.  If
3075                          * we aren't cached, we want to start async caching this
3076                          * block group so we can free the extent the next time
3077                          * around.
3078                          */
3079                         spin_lock(&cache->space_info->lock);
3080                         spin_lock(&cache->lock);
3081                         unpin = (cache->cached == BTRFS_CACHE_FINISHED);
3082                         if (likely(unpin)) {
3083                                 cache->pinned -= len;
3084                                 cache->space_info->bytes_pinned -= len;
3085                                 fs_info->total_pinned -= len;
3086                         }
3087                         spin_unlock(&cache->lock);
3088                         spin_unlock(&cache->space_info->lock);
3089
3090                         if (likely(unpin))
3091                                 clear_extent_dirty(&fs_info->pinned_extents,
3092                                                    bytenr, bytenr + len -1,
3093                                                    GFP_NOFS);
3094                         else
3095                                 cache_block_group(cache);
3096
3097                         if (unpin)
3098                                 btrfs_add_free_space(cache, bytenr, len);
3099                 }
3100                 btrfs_put_block_group(cache);
3101                 bytenr += len;
3102                 num -= len;
3103         }
3104         return 0;
3105 }
3106
3107 static int update_reserved_extents(struct btrfs_root *root,
3108                                    u64 bytenr, u64 num, int reserve)
3109 {
3110         u64 len;
3111         struct btrfs_block_group_cache *cache;
3112         struct btrfs_fs_info *fs_info = root->fs_info;
3113
3114         while (num > 0) {
3115                 cache = btrfs_lookup_block_group(fs_info, bytenr);
3116                 BUG_ON(!cache);
3117                 len = min(num, cache->key.offset -
3118                           (bytenr - cache->key.objectid));
3119
3120                 spin_lock(&cache->space_info->lock);
3121                 spin_lock(&cache->lock);
3122                 if (reserve) {
3123                         cache->reserved += len;
3124                         cache->space_info->bytes_reserved += len;
3125                 } else {
3126                         cache->reserved -= len;
3127                         cache->space_info->bytes_reserved -= len;
3128                 }
3129                 spin_unlock(&cache->lock);
3130                 spin_unlock(&cache->space_info->lock);
3131                 btrfs_put_block_group(cache);
3132                 bytenr += len;
3133                 num -= len;
3134         }
3135         return 0;
3136 }
3137
3138 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
3139 {
3140         u64 last = 0;
3141         u64 start;
3142         u64 end;
3143         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3144         int ret;
3145
3146         while (1) {
3147                 ret = find_first_extent_bit(pinned_extents, last,
3148                                             &start, &end, EXTENT_DIRTY);
3149                 if (ret)
3150                         break;
3151
3152                 set_extent_dirty(copy, start, end, GFP_NOFS);
3153                 last = end + 1;
3154         }
3155         return 0;
3156 }
3157
3158 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3159                                struct btrfs_root *root,
3160                                struct extent_io_tree *unpin)
3161 {
3162         u64 start;
3163         u64 end;
3164         int ret;
3165
3166         while (1) {
3167                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3168                                             EXTENT_DIRTY);
3169                 if (ret)
3170                         break;
3171
3172                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3173
3174                 /* unlocks the pinned mutex */
3175                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
3176                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3177
3178                 cond_resched();
3179         }
3180
3181         return ret;
3182 }
3183
3184 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3185                           struct btrfs_root *root,
3186                           struct btrfs_path *path,
3187                           u64 bytenr, u64 num_bytes, int is_data,
3188                           struct extent_buffer **must_clean)
3189 {
3190         int err = 0;
3191         struct extent_buffer *buf;
3192
3193         if (is_data)
3194                 goto pinit;
3195
3196         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3197         if (!buf)
3198                 goto pinit;
3199
3200         /* we can reuse a block if it hasn't been written
3201          * and it is from this transaction.  We can't
3202          * reuse anything from the tree log root because
3203          * it has tiny sub-transactions.
3204          */
3205         if (btrfs_buffer_uptodate(buf, 0) &&
3206             btrfs_try_tree_lock(buf)) {
3207                 u64 header_owner = btrfs_header_owner(buf);
3208                 u64 header_transid = btrfs_header_generation(buf);
3209                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3210                     header_transid == trans->transid &&
3211                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3212                         *must_clean = buf;
3213                         return 1;
3214                 }
3215                 btrfs_tree_unlock(buf);
3216         }
3217         free_extent_buffer(buf);
3218 pinit:
3219         btrfs_set_path_blocking(path);
3220         /* unlocks the pinned mutex */
3221         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3222
3223         BUG_ON(err < 0);
3224         return 0;
3225 }
3226
3227
3228 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3229                                 struct btrfs_root *root,
3230                                 u64 bytenr, u64 num_bytes, u64 parent,
3231                                 u64 root_objectid, u64 owner_objectid,
3232                                 u64 owner_offset, int refs_to_drop,
3233                                 struct btrfs_delayed_extent_op *extent_op)
3234 {
3235         struct btrfs_key key;
3236         struct btrfs_path *path;
3237         struct btrfs_fs_info *info = root->fs_info;
3238         struct btrfs_root *extent_root = info->extent_root;
3239         struct extent_buffer *leaf;
3240         struct btrfs_extent_item *ei;
3241         struct btrfs_extent_inline_ref *iref;
3242         int ret;
3243         int is_data;
3244         int extent_slot = 0;
3245         int found_extent = 0;
3246         int num_to_del = 1;
3247         u32 item_size;
3248         u64 refs;
3249
3250         path = btrfs_alloc_path();
3251         if (!path)
3252                 return -ENOMEM;
3253
3254         path->reada = 1;
3255         path->leave_spinning = 1;
3256
3257         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3258         BUG_ON(!is_data && refs_to_drop != 1);
3259
3260         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3261                                     bytenr, num_bytes, parent,
3262                                     root_objectid, owner_objectid,
3263                                     owner_offset);
3264         if (ret == 0) {
3265                 extent_slot = path->slots[0];
3266                 while (extent_slot >= 0) {
3267                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3268                                               extent_slot);
3269                         if (key.objectid != bytenr)
3270                                 break;
3271                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3272                             key.offset == num_bytes) {
3273                                 found_extent = 1;
3274                                 break;
3275                         }
3276                         if (path->slots[0] - extent_slot > 5)
3277                                 break;
3278                         extent_slot--;
3279                 }
3280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3281                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3282                 if (found_extent && item_size < sizeof(*ei))
3283                         found_extent = 0;
3284 #endif
3285                 if (!found_extent) {
3286                         BUG_ON(iref);
3287                         ret = remove_extent_backref(trans, extent_root, path,
3288                                                     NULL, refs_to_drop,
3289                                                     is_data);
3290                         BUG_ON(ret);
3291                         btrfs_release_path(extent_root, path);
3292                         path->leave_spinning = 1;
3293
3294                         key.objectid = bytenr;
3295                         key.type = BTRFS_EXTENT_ITEM_KEY;
3296                         key.offset = num_bytes;
3297
3298                         ret = btrfs_search_slot(trans, extent_root,
3299                                                 &key, path, -1, 1);
3300                         if (ret) {
3301                                 printk(KERN_ERR "umm, got %d back from search"
3302                                        ", was looking for %llu\n", ret,
3303                                        (unsigned long long)bytenr);
3304                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3305                         }
3306                         BUG_ON(ret);
3307                         extent_slot = path->slots[0];
3308                 }
3309         } else {
3310                 btrfs_print_leaf(extent_root, path->nodes[0]);
3311                 WARN_ON(1);
3312                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3313                        "parent %llu root %llu  owner %llu offset %llu\n",
3314                        (unsigned long long)bytenr,
3315                        (unsigned long long)parent,
3316                        (unsigned long long)root_objectid,
3317                        (unsigned long long)owner_objectid,
3318                        (unsigned long long)owner_offset);
3319         }
3320
3321         leaf = path->nodes[0];
3322         item_size = btrfs_item_size_nr(leaf, extent_slot);
3323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3324         if (item_size < sizeof(*ei)) {
3325                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3326                 ret = convert_extent_item_v0(trans, extent_root, path,
3327                                              owner_objectid, 0);
3328                 BUG_ON(ret < 0);
3329
3330                 btrfs_release_path(extent_root, path);
3331                 path->leave_spinning = 1;
3332
3333                 key.objectid = bytenr;
3334                 key.type = BTRFS_EXTENT_ITEM_KEY;
3335                 key.offset = num_bytes;
3336
3337                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3338                                         -1, 1);
3339                 if (ret) {
3340                         printk(KERN_ERR "umm, got %d back from search"
3341                                ", was looking for %llu\n", ret,
3342                                (unsigned long long)bytenr);
3343                         btrfs_print_leaf(extent_root, path->nodes[0]);
3344                 }
3345                 BUG_ON(ret);
3346                 extent_slot = path->slots[0];
3347                 leaf = path->nodes[0];
3348                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3349         }
3350 #endif
3351         BUG_ON(item_size < sizeof(*ei));
3352         ei = btrfs_item_ptr(leaf, extent_slot,
3353                             struct btrfs_extent_item);
3354         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3355                 struct btrfs_tree_block_info *bi;
3356                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3357                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3358                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3359         }
3360
3361         refs = btrfs_extent_refs(leaf, ei);
3362         BUG_ON(refs < refs_to_drop);
3363         refs -= refs_to_drop;
3364
3365         if (refs > 0) {
3366                 if (extent_op)
3367                         __run_delayed_extent_op(extent_op, leaf, ei);
3368                 /*
3369                  * In the case of inline back ref, reference count will
3370                  * be updated by remove_extent_backref
3371                  */
3372                 if (iref) {
3373                         BUG_ON(!found_extent);
3374                 } else {
3375                         btrfs_set_extent_refs(leaf, ei, refs);
3376                         btrfs_mark_buffer_dirty(leaf);
3377                 }
3378                 if (found_extent) {
3379                         ret = remove_extent_backref(trans, extent_root, path,
3380                                                     iref, refs_to_drop,
3381                                                     is_data);
3382                         BUG_ON(ret);
3383                 }
3384         } else {
3385                 int mark_free = 0;
3386                 struct extent_buffer *must_clean = NULL;
3387
3388                 if (found_extent) {
3389                         BUG_ON(is_data && refs_to_drop !=
3390                                extent_data_ref_count(root, path, iref));
3391                         if (iref) {
3392                                 BUG_ON(path->slots[0] != extent_slot);
3393                         } else {
3394                                 BUG_ON(path->slots[0] != extent_slot + 1);
3395                                 path->slots[0] = extent_slot;
3396                                 num_to_del = 2;
3397                         }
3398                 }
3399
3400                 ret = pin_down_bytes(trans, root, path, bytenr,
3401                                      num_bytes, is_data, &must_clean);
3402                 if (ret > 0)
3403                         mark_free = 1;
3404                 BUG_ON(ret < 0);
3405                 /*
3406                  * it is going to be very rare for someone to be waiting
3407                  * on the block we're freeing.  del_items might need to
3408                  * schedule, so rather than get fancy, just force it
3409                  * to blocking here
3410                  */
3411                 if (must_clean)
3412                         btrfs_set_lock_blocking(must_clean);
3413
3414                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3415                                       num_to_del);
3416                 BUG_ON(ret);
3417                 btrfs_release_path(extent_root, path);
3418
3419                 if (must_clean) {
3420                         clean_tree_block(NULL, root, must_clean);
3421                         btrfs_tree_unlock(must_clean);
3422                         free_extent_buffer(must_clean);
3423                 }
3424
3425                 if (is_data) {
3426                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3427                         BUG_ON(ret);
3428                 } else {
3429                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3430                              bytenr >> PAGE_CACHE_SHIFT,
3431                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3432                 }
3433
3434                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3435                                          mark_free);
3436                 BUG_ON(ret);
3437         }
3438         btrfs_free_path(path);
3439         return ret;
3440 }
3441
3442 /*
3443  * when we free an extent, it is possible (and likely) that we free the last
3444  * delayed ref for that extent as well.  This searches the delayed ref tree for
3445  * a given extent, and if there are no other delayed refs to be processed, it
3446  * removes it from the tree.
3447  */
3448 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3449                                       struct btrfs_root *root, u64 bytenr)
3450 {
3451         struct btrfs_delayed_ref_head *head;
3452         struct btrfs_delayed_ref_root *delayed_refs;
3453         struct btrfs_delayed_ref_node *ref;
3454         struct rb_node *node;
3455         int ret;
3456
3457         delayed_refs = &trans->transaction->delayed_refs;
3458         spin_lock(&delayed_refs->lock);
3459         head = btrfs_find_delayed_ref_head(trans, bytenr);
3460         if (!head)
3461                 goto out;
3462
3463         node = rb_prev(&head->node.rb_node);
3464         if (!node)
3465                 goto out;
3466
3467         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3468
3469         /* there are still entries for this ref, we can't drop it */
3470         if (ref->bytenr == bytenr)
3471                 goto out;
3472
3473         if (head->extent_op) {
3474                 if (!head->must_insert_reserved)
3475                         goto out;
3476                 kfree(head->extent_op);
3477                 head->extent_op = NULL;
3478         }
3479
3480         /*
3481          * waiting for the lock here would deadlock.  If someone else has it
3482          * locked they are already in the process of dropping it anyway
3483          */
3484         if (!mutex_trylock(&head->mutex))
3485                 goto out;
3486
3487         /*
3488          * at this point we have a head with no other entries.  Go
3489          * ahead and process it.
3490          */
3491         head->node.in_tree = 0;
3492         rb_erase(&head->node.rb_node, &delayed_refs->root);
3493
3494         delayed_refs->num_entries--;
3495
3496         /*
3497          * we don't take a ref on the node because we're removing it from the
3498          * tree, so we just steal the ref the tree was holding.
3499          */
3500         delayed_refs->num_heads--;
3501         if (list_empty(&head->cluster))
3502                 delayed_refs->num_heads_ready--;
3503
3504         list_del_init(&head->cluster);
3505         spin_unlock(&delayed_refs->lock);
3506
3507         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3508                                   &head->node, head->extent_op,
3509                                   head->must_insert_reserved);
3510         BUG_ON(ret);
3511         btrfs_put_delayed_ref(&head->node);
3512         return 0;
3513 out:
3514         spin_unlock(&delayed_refs->lock);
3515         return 0;
3516 }
3517
3518 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3519                       struct btrfs_root *root,
3520                       u64 bytenr, u64 num_bytes, u64 parent,
3521                       u64 root_objectid, u64 owner, u64 offset)
3522 {
3523         int ret;
3524
3525         /*
3526          * tree log blocks never actually go into the extent allocation
3527          * tree, just update pinning info and exit early.
3528          */
3529         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3530                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3531                 /* unlocks the pinned mutex */
3532                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3533                 update_reserved_extents(root, bytenr, num_bytes, 0);
3534                 ret = 0;
3535         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3536                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3537                                         parent, root_objectid, (int)owner,
3538                                         BTRFS_DROP_DELAYED_REF, NULL);
3539                 BUG_ON(ret);
3540                 ret = check_ref_cleanup(trans, root, bytenr);
3541                 BUG_ON(ret);
3542         } else {
3543                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3544                                         parent, root_objectid, owner,
3545                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
3546                 BUG_ON(ret);
3547         }
3548         return ret;
3549 }
3550
3551 static u64 stripe_align(struct btrfs_root *root, u64 val)
3552 {
3553         u64 mask = ((u64)root->stripesize - 1);
3554         u64 ret = (val + mask) & ~mask;
3555         return ret;
3556 }
3557
3558 /*
3559  * when we wait for progress in the block group caching, its because
3560  * our allocation attempt failed at least once.  So, we must sleep
3561  * and let some progress happen before we try again.
3562  *
3563  * This function will sleep at least once waiting for new free space to
3564  * show up, and then it will check the block group free space numbers
3565  * for our min num_bytes.  Another option is to have it go ahead
3566  * and look in the rbtree for a free extent of a given size, but this
3567  * is a good start.
3568  */
3569 static noinline int
3570 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3571                                 u64 num_bytes)
3572 {
3573         DEFINE_WAIT(wait);
3574
3575         prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE);
3576
3577         if (block_group_cache_done(cache)) {
3578                 finish_wait(&cache->caching_q, &wait);
3579                 return 0;
3580         }
3581         schedule();
3582         finish_wait(&cache->caching_q, &wait);
3583
3584         wait_event(cache->caching_q, block_group_cache_done(cache) ||
3585                    (cache->free_space >= num_bytes));
3586         return 0;
3587 }
3588
3589 enum btrfs_loop_type {
3590         LOOP_CACHED_ONLY = 0,
3591         LOOP_CACHING_NOWAIT = 1,
3592         LOOP_CACHING_WAIT = 2,
3593         LOOP_ALLOC_CHUNK = 3,
3594         LOOP_NO_EMPTY_SIZE = 4,
3595 };
3596
3597 /*
3598  * walks the btree of allocated extents and find a hole of a given size.
3599  * The key ins is changed to record the hole:
3600  * ins->objectid == block start
3601  * ins->flags = BTRFS_EXTENT_ITEM_KEY
3602  * ins->offset == number of blocks
3603  * Any available blocks before search_start are skipped.
3604  */
3605 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3606                                      struct btrfs_root *orig_root,
3607                                      u64 num_bytes, u64 empty_size,
3608                                      u64 search_start, u64 search_end,
3609                                      u64 hint_byte, struct btrfs_key *ins,
3610                                      u64 exclude_start, u64 exclude_nr,
3611                                      int data)
3612 {
3613         int ret = 0;
3614         struct btrfs_root *root = orig_root->fs_info->extent_root;
3615         struct btrfs_free_cluster *last_ptr = NULL;
3616         struct btrfs_block_group_cache *block_group = NULL;
3617         int empty_cluster = 2 * 1024 * 1024;
3618         int allowed_chunk_alloc = 0;
3619         struct btrfs_space_info *space_info;
3620         int last_ptr_loop = 0;
3621         int loop = 0;
3622         bool found_uncached_bg = false;
3623
3624         WARN_ON(num_bytes < root->sectorsize);
3625         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3626         ins->objectid = 0;
3627         ins->offset = 0;
3628
3629         space_info = __find_space_info(root->fs_info, data);
3630
3631         if (orig_root->ref_cows || empty_size)
3632                 allowed_chunk_alloc = 1;
3633
3634         if (data & BTRFS_BLOCK_GROUP_METADATA) {
3635                 last_ptr = &root->fs_info->meta_alloc_cluster;
3636                 if (!btrfs_test_opt(root, SSD))
3637                         empty_cluster = 64 * 1024;
3638         }
3639
3640         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3641                 last_ptr = &root->fs_info->data_alloc_cluster;
3642         }
3643
3644         if (last_ptr) {
3645                 spin_lock(&last_ptr->lock);
3646                 if (last_ptr->block_group)
3647                         hint_byte = last_ptr->window_start;
3648                 spin_unlock(&last_ptr->lock);
3649         }
3650
3651         search_start = max(search_start, first_logical_byte(root, 0));
3652         search_start = max(search_start, hint_byte);
3653
3654         if (!last_ptr)
3655                 empty_cluster = 0;
3656
3657         if (search_start == hint_byte) {
3658                 block_group = btrfs_lookup_block_group(root->fs_info,
3659                                                        search_start);
3660                 /*
3661                  * we don't want to use the block group if it doesn't match our
3662                  * allocation bits, or if its not cached.
3663                  */
3664                 if (block_group && block_group_bits(block_group, data) &&
3665                     block_group_cache_done(block_group)) {
3666                         down_read(&space_info->groups_sem);
3667                         if (list_empty(&block_group->list) ||
3668                             block_group->ro) {
3669                                 /*
3670                                  * someone is removing this block group,
3671                                  * we can't jump into the have_block_group
3672                                  * target because our list pointers are not
3673                                  * valid
3674                                  */
3675                                 btrfs_put_block_group(block_group);
3676                                 up_read(&space_info->groups_sem);
3677                         } else
3678                                 goto have_block_group;
3679                 } else if (block_group) {
3680                         btrfs_put_block_group(block_group);
3681                 }
3682         }
3683
3684 search:
3685         down_read(&space_info->groups_sem);
3686         list_for_each_entry(block_group, &space_info->block_groups, list) {
3687                 u64 offset;
3688                 int cached;
3689
3690                 atomic_inc(&block_group->count);
3691                 search_start = block_group->key.objectid;
3692
3693 have_block_group:
3694                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
3695                         /*
3696                          * we want to start caching kthreads, but not too many
3697                          * right off the bat so we don't overwhelm the system,
3698                          * so only start them if there are less than 2 and we're
3699                          * in the initial allocation phase.
3700                          */
3701                         if (loop > LOOP_CACHING_NOWAIT ||
3702                             atomic_read(&space_info->caching_threads) < 2) {
3703                                 ret = cache_block_group(block_group);
3704                                 BUG_ON(ret);
3705                         }
3706                 }
3707
3708                 cached = block_group_cache_done(block_group);
3709                 if (unlikely(!cached)) {
3710                         found_uncached_bg = true;
3711
3712                         /* if we only want cached bgs, loop */
3713                         if (loop == LOOP_CACHED_ONLY)
3714                                 goto loop;
3715                 }
3716
3717                 if (unlikely(block_group->ro))
3718                         goto loop;
3719
3720                 if (last_ptr) {
3721                         /*
3722                          * the refill lock keeps out other
3723                          * people trying to start a new cluster
3724                          */
3725                         spin_lock(&last_ptr->refill_lock);
3726                         if (last_ptr->block_group &&
3727                             (last_ptr->block_group->ro ||
3728                             !block_group_bits(last_ptr->block_group, data))) {
3729                                 offset = 0;
3730                                 goto refill_cluster;
3731                         }
3732
3733                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3734                                                  num_bytes, search_start);
3735                         if (offset) {
3736                                 /* we have a block, we're done */
3737                                 spin_unlock(&last_ptr->refill_lock);
3738                                 goto checks;
3739                         }
3740
3741                         spin_lock(&last_ptr->lock);
3742                         /*
3743                          * whoops, this cluster doesn't actually point to
3744                          * this block group.  Get a ref on the block
3745                          * group is does point to and try again
3746                          */
3747                         if (!last_ptr_loop && last_ptr->block_group &&
3748                             last_ptr->block_group != block_group) {
3749
3750                                 btrfs_put_block_group(block_group);
3751                                 block_group = last_ptr->block_group;
3752                                 atomic_inc(&block_group->count);
3753                                 spin_unlock(&last_ptr->lock);
3754                                 spin_unlock(&last_ptr->refill_lock);
3755
3756                                 last_ptr_loop = 1;
3757                                 search_start = block_group->key.objectid;
3758                                 /*
3759                                  * we know this block group is properly
3760                                  * in the list because
3761                                  * btrfs_remove_block_group, drops the
3762                                  * cluster before it removes the block
3763                                  * group from the list
3764                                  */
3765                                 goto have_block_group;
3766                         }
3767                         spin_unlock(&last_ptr->lock);
3768 refill_cluster:
3769                         /*
3770                          * this cluster didn't work out, free it and
3771                          * start over
3772                          */
3773                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3774
3775                         last_ptr_loop = 0;
3776
3777                         /* allocate a cluster in this block group */
3778                         ret = btrfs_find_space_cluster(trans, root,
3779                                                block_group, last_ptr,
3780                                                offset, num_bytes,
3781                                                empty_cluster + empty_size);
3782                         if (ret == 0) {
3783                                 /*
3784                                  * now pull our allocation out of this
3785                                  * cluster
3786                                  */
3787                                 offset = btrfs_alloc_from_cluster(block_group,
3788                                                   last_ptr, num_bytes,
3789                                                   search_start);
3790                                 if (offset) {
3791                                         /* we found one, proceed */
3792                                         spin_unlock(&last_ptr->refill_lock);
3793                                         goto checks;
3794                                 }
3795                         } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
3796                                 spin_unlock(&last_ptr->refill_lock);
3797
3798                                 wait_block_group_cache_progress(block_group,
3799                                        num_bytes + empty_cluster + empty_size);
3800                                 goto have_block_group;
3801                         }
3802
3803                         /*
3804                          * at this point we either didn't find a cluster
3805                          * or we weren't able to allocate a block from our
3806                          * cluster.  Free the cluster we've been trying
3807                          * to use, and go to the next block group
3808                          */
3809                         if (loop < LOOP_NO_EMPTY_SIZE) {
3810                                 btrfs_return_cluster_to_free_space(NULL,
3811                                                                    last_ptr);
3812                                 spin_unlock(&last_ptr->refill_lock);
3813                                 goto loop;
3814                         }
3815                         spin_unlock(&last_ptr->refill_lock);
3816                 }
3817
3818                 offset = btrfs_find_space_for_alloc(block_group, search_start,
3819                                                     num_bytes, empty_size);
3820                 if (!offset && (cached || (!cached &&
3821                                            loop == LOOP_CACHING_NOWAIT))) {
3822                         goto loop;
3823                 } else if (!offset && (!cached &&
3824                                        loop > LOOP_CACHING_NOWAIT)) {
3825                         wait_block_group_cache_progress(block_group,
3826                                         num_bytes + empty_size);
3827                         goto have_block_group;
3828                 }
3829 checks:
3830                 search_start = stripe_align(root, offset);
3831                 /* move on to the next group */
3832                 if (search_start + num_bytes >= search_end) {
3833                         btrfs_add_free_space(block_group, offset, num_bytes);
3834                         goto loop;
3835                 }
3836
3837                 /* move on to the next group */
3838                 if (search_start + num_bytes >
3839                     block_group->key.objectid + block_group->key.offset) {
3840                         btrfs_add_free_space(block_group, offset, num_bytes);
3841                         goto loop;
3842                 }
3843
3844                 if (exclude_nr > 0 &&
3845                     (search_start + num_bytes > exclude_start &&
3846                      search_start < exclude_start + exclude_nr)) {
3847                         search_start = exclude_start + exclude_nr;
3848
3849                         btrfs_add_free_space(block_group, offset, num_bytes);
3850                         /*
3851                          * if search_start is still in this block group
3852                          * then we just re-search this block group
3853                          */
3854                         if (search_start >= block_group->key.objectid &&
3855                             search_start < (block_group->key.objectid +
3856                                             block_group->key.offset))
3857                                 goto have_block_group;
3858                         goto loop;
3859                 }
3860
3861                 ins->objectid = search_start;
3862                 ins->offset = num_bytes;
3863
3864                 if (offset < search_start)
3865                         btrfs_add_free_space(block_group, offset,
3866                                              search_start - offset);
3867                 BUG_ON(offset > search_start);
3868
3869                 /* we are all good, lets return */
3870                 break;
3871 loop:
3872                 btrfs_put_block_group(block_group);
3873         }
3874         up_read(&space_info->groups_sem);
3875
3876         /* LOOP_CACHED_ONLY, only search fully cached block groups
3877          * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3878          *                      dont wait foR them to finish caching
3879          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3880          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3881          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3882          *                      again
3883          */
3884         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
3885             (found_uncached_bg || empty_size || empty_cluster ||
3886              allowed_chunk_alloc)) {
3887                 if (found_uncached_bg) {
3888                         found_uncached_bg = false;
3889                         if (loop < LOOP_CACHING_WAIT) {
3890                                 loop++;
3891                                 goto search;
3892                         }
3893                 }
3894
3895                 if (loop == LOOP_ALLOC_CHUNK) {
3896                         empty_size = 0;
3897                         empty_cluster = 0;
3898                 }
3899
3900                 if (allowed_chunk_alloc) {
3901                         ret = do_chunk_alloc(trans, root, num_bytes +
3902                                              2 * 1024 * 1024, data, 1);
3903                         allowed_chunk_alloc = 0;
3904                 } else {
3905                         space_info->force_alloc = 1;
3906                 }
3907
3908                 if (loop < LOOP_NO_EMPTY_SIZE) {
3909                         loop++;
3910                         goto search;
3911                 }
3912                 ret = -ENOSPC;
3913         } else if (!ins->objectid) {
3914                 ret = -ENOSPC;
3915         }
3916
3917         /* we found what we needed */
3918         if (ins->objectid) {
3919                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3920                         trans->block_group = block_group->key.objectid;
3921
3922                 btrfs_put_block_group(block_group);
3923                 ret = 0;
3924         }
3925
3926         return ret;
3927 }
3928
3929 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3930 {
3931         struct btrfs_block_group_cache *cache;
3932
3933         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3934                (unsigned long long)(info->total_bytes - info->bytes_used -
3935                                     info->bytes_pinned - info->bytes_reserved),
3936                (info->full) ? "" : "not ");
3937         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3938                " may_use=%llu, used=%llu\n",
3939                (unsigned long long)info->total_bytes,
3940                (unsigned long long)info->bytes_pinned,
3941                (unsigned long long)info->bytes_delalloc,
3942                (unsigned long long)info->bytes_may_use,
3943                (unsigned long long)info->bytes_used);
3944
3945         down_read(&info->groups_sem);
3946         list_for_each_entry(cache, &info->block_groups, list) {
3947                 spin_lock(&cache->lock);
3948                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3949                        "%llu pinned %llu reserved\n",
3950                        (unsigned long long)cache->key.objectid,
3951                        (unsigned long long)cache->key.offset,
3952                        (unsigned long long)btrfs_block_group_used(&cache->item),
3953                        (unsigned long long)cache->pinned,
3954                        (unsigned long long)cache->reserved);
3955                 btrfs_dump_free_space(cache, bytes);
3956                 spin_unlock(&cache->lock);
3957         }
3958         up_read(&info->groups_sem);
3959 }
3960
3961 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3962                                   struct btrfs_root *root,
3963                                   u64 num_bytes, u64 min_alloc_size,
3964                                   u64 empty_size, u64 hint_byte,
3965                                   u64 search_end, struct btrfs_key *ins,
3966                                   u64 data)
3967 {
3968         int ret;
3969         u64 search_start = 0;
3970         struct btrfs_fs_info *info = root->fs_info;
3971
3972         data = btrfs_get_alloc_profile(root, data);
3973 again:
3974         /*
3975          * the only place that sets empty_size is btrfs_realloc_node, which
3976          * is not called recursively on allocations
3977          */
3978         if (empty_size || root->ref_cows) {
3979                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3980                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3981                                      2 * 1024 * 1024,
3982                                      BTRFS_BLOCK_GROUP_METADATA |
3983                                      (info->metadata_alloc_profile &
3984                                       info->avail_metadata_alloc_bits), 0);
3985                 }
3986                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3987                                      num_bytes + 2 * 1024 * 1024, data, 0);
3988         }
3989
3990         WARN_ON(num_bytes < root->sectorsize);
3991         ret = find_free_extent(trans, root, num_bytes, empty_size,
3992                                search_start, search_end, hint_byte, ins,
3993                                trans->alloc_exclude_start,
3994                                trans->alloc_exclude_nr, data);
3995
3996         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3997                 num_bytes = num_bytes >> 1;
3998                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3999                 num_bytes = max(num_bytes, min_alloc_size);
4000                 do_chunk_alloc(trans, root->fs_info->extent_root,
4001                                num_bytes, data, 1);
4002                 goto again;
4003         }
4004         if (ret == -ENOSPC) {
4005                 struct btrfs_space_info *sinfo;
4006
4007                 sinfo = __find_space_info(root->fs_info, data);
4008                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4009                        "wanted %llu\n", (unsigned long long)data,
4010                        (unsigned long long)num_bytes);
4011                 dump_space_info(sinfo, num_bytes);
4012         }
4013
4014         return ret;
4015 }
4016
4017 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4018 {
4019         struct btrfs_block_group_cache *cache;
4020         int ret = 0;
4021
4022         cache = btrfs_lookup_block_group(root->fs_info, start);
4023         if (!cache) {
4024                 printk(KERN_ERR "Unable to find block group for %llu\n",
4025                        (unsigned long long)start);
4026                 return -ENOSPC;
4027         }
4028
4029         ret = btrfs_discard_extent(root, start, len);
4030
4031         btrfs_add_free_space(cache, start, len);
4032         btrfs_put_block_group(cache);
4033         update_reserved_extents(root, start, len, 0);
4034
4035         return ret;
4036 }
4037
4038 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4039                                   struct btrfs_root *root,
4040                                   u64 num_bytes, u64 min_alloc_size,
4041                                   u64 empty_size, u64 hint_byte,
4042                                   u64 search_end, struct btrfs_key *ins,
4043                                   u64 data)
4044 {
4045         int ret;
4046         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
4047                                      empty_size, hint_byte, search_end, ins,
4048                                      data);
4049         if (!ret)
4050                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4051
4052         return ret;
4053 }
4054
4055 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4056                                       struct btrfs_root *root,
4057                                       u64 parent, u64 root_objectid,
4058                                       u64 flags, u64 owner, u64 offset,
4059                                       struct btrfs_key *ins, int ref_mod)
4060 {
4061         int ret;
4062         struct btrfs_fs_info *fs_info = root->fs_info;
4063         struct btrfs_extent_item *extent_item;
4064         struct btrfs_extent_inline_ref *iref;
4065         struct btrfs_path *path;
4066         struct extent_buffer *leaf;
4067         int type;
4068         u32 size;
4069
4070         if (parent > 0)
4071                 type = BTRFS_SHARED_DATA_REF_KEY;
4072         else
4073                 type = BTRFS_EXTENT_DATA_REF_KEY;
4074
4075         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4076
4077         path = btrfs_alloc_path();
4078         BUG_ON(!path);
4079
4080         path->leave_spinning = 1;
4081         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4082                                       ins, size);
4083         BUG_ON(ret);
4084
4085         leaf = path->nodes[0];
4086         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4087                                      struct btrfs_extent_item);
4088         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4089         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4090         btrfs_set_extent_flags(leaf, extent_item,
4091                                flags | BTRFS_EXTENT_FLAG_DATA);
4092
4093         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4094         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4095         if (parent > 0) {
4096                 struct btrfs_shared_data_ref *ref;
4097                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4098                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4099                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4100         } else {
4101                 struct btrfs_extent_data_ref *ref;
4102                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4103                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4104                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4105                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4106                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4107         }
4108
4109         btrfs_mark_buffer_dirty(path->nodes[0]);
4110         btrfs_free_path(path);
4111
4112         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4113                                  1, 0);
4114         if (ret) {
4115                 printk(KERN_ERR "btrfs update block group failed for %llu "
4116                        "%llu\n", (unsigned long long)ins->objectid,
4117                        (unsigned long long)ins->offset);
4118                 BUG();
4119         }
4120         return ret;
4121 }
4122
4123 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4124                                      struct btrfs_root *root,
4125                                      u64 parent, u64 root_objectid,
4126                                      u64 flags, struct btrfs_disk_key *key,
4127                                      int level, struct btrfs_key *ins)
4128 {
4129         int ret;
4130         struct btrfs_fs_info *fs_info = root->fs_info;
4131         struct btrfs_extent_item *extent_item;
4132         struct btrfs_tree_block_info *block_info;
4133         struct btrfs_extent_inline_ref *iref;
4134         struct btrfs_path *path;
4135         struct extent_buffer *leaf;
4136         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4137
4138         path = btrfs_alloc_path();
4139         BUG_ON(!path);
4140
4141         path->leave_spinning = 1;
4142         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4143                                       ins, size);
4144         BUG_ON(ret);
4145
4146         leaf = path->nodes[0];
4147         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4148                                      struct btrfs_extent_item);
4149         btrfs_set_extent_refs(leaf, extent_item, 1);
4150         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4151         btrfs_set_extent_flags(leaf, extent_item,
4152                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4153         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4154
4155         btrfs_set_tree_block_key(leaf, block_info, key);
4156         btrfs_set_tree_block_level(leaf, block_info, level);
4157
4158         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4159         if (parent > 0) {
4160                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4161                 btrfs_set_extent_inline_ref_type(leaf, iref,
4162                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4163                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4164         } else {
4165                 btrfs_set_extent_inline_ref_type(leaf, iref,
4166                                                  BTRFS_TREE_BLOCK_REF_KEY);
4167                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4168         }
4169
4170         btrfs_mark_buffer_dirty(leaf);
4171         btrfs_free_path(path);
4172
4173         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4174                                  1, 0);
4175         if (ret) {
4176                 printk(KERN_ERR "btrfs update block group failed for %llu "
4177                        "%llu\n", (unsigned long long)ins->objectid,
4178                        (unsigned long long)ins->offset);
4179                 BUG();
4180         }
4181         return ret;
4182 }
4183
4184 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4185                                      struct btrfs_root *root,
4186                                      u64 root_objectid, u64 owner,
4187                                      u64 offset, struct btrfs_key *ins)
4188 {
4189         int ret;
4190
4191         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4192
4193         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4194                                          0, root_objectid, owner, offset,
4195                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4196         return ret;
4197 }
4198
4199 /*
4200  * this is used by the tree logging recovery code.  It records that
4201  * an extent has been allocated and makes sure to clear the free
4202  * space cache bits as well
4203  */
4204 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4205                                    struct btrfs_root *root,
4206                                    u64 root_objectid, u64 owner, u64 offset,
4207                                    struct btrfs_key *ins)
4208 {
4209         int ret;
4210         struct btrfs_block_group_cache *block_group;
4211
4212         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4213         cache_block_group(block_group);
4214         wait_event(block_group->caching_q,
4215                    block_group_cache_done(block_group));
4216
4217         ret = btrfs_remove_free_space(block_group, ins->objectid,
4218                                       ins->offset);
4219         BUG_ON(ret);
4220         btrfs_put_block_group(block_group);
4221         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4222                                          0, owner, offset, ins, 1);
4223         return ret;
4224 }
4225
4226 /*
4227  * finds a free extent and does all the dirty work required for allocation
4228  * returns the key for the extent through ins, and a tree buffer for
4229  * the first block of the extent through buf.
4230  *
4231  * returns 0 if everything worked, non-zero otherwise.
4232  */
4233 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4234                             struct btrfs_root *root,
4235                             u64 num_bytes, u64 parent, u64 root_objectid,
4236                             struct btrfs_disk_key *key, int level,
4237                             u64 empty_size, u64 hint_byte, u64 search_end,
4238                             struct btrfs_key *ins)
4239 {
4240         int ret;
4241         u64 flags = 0;
4242
4243         ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4244                                      empty_size, hint_byte, search_end,
4245                                      ins, 0);
4246         if (ret)
4247                 return ret;
4248
4249         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4250                 if (parent == 0)
4251                         parent = ins->objectid;
4252                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4253         } else
4254                 BUG_ON(parent > 0);
4255
4256         update_reserved_extents(root, ins->objectid, ins->offset, 1);
4257         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4258                 struct btrfs_delayed_extent_op *extent_op;
4259                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4260                 BUG_ON(!extent_op);
4261                 if (key)
4262                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4263                 else
4264                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4265                 extent_op->flags_to_set = flags;
4266                 extent_op->update_key = 1;
4267                 extent_op->update_flags = 1;
4268                 extent_op->is_data = 0;
4269
4270                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4271                                         ins->offset, parent, root_objectid,
4272                                         level, BTRFS_ADD_DELAYED_EXTENT,
4273                                         extent_op);
4274                 BUG_ON(ret);
4275         }
4276         return ret;
4277 }
4278
4279 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4280                                             struct btrfs_root *root,
4281                                             u64 bytenr, u32 blocksize,
4282                                             int level)
4283 {
4284         struct extent_buffer *buf;
4285
4286         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4287         if (!buf)
4288                 return ERR_PTR(-ENOMEM);
4289         btrfs_set_header_generation(buf, trans->transid);
4290         btrfs_set_buffer_lockdep_class(buf, level);
4291         btrfs_tree_lock(buf);
4292         clean_tree_block(trans, root, buf);
4293
4294         btrfs_set_lock_blocking(buf);
4295         btrfs_set_buffer_uptodate(buf);
4296
4297         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4298                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4299                          buf->start + buf->len - 1, GFP_NOFS);
4300         } else {
4301                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4302                          buf->start + buf->len - 1, GFP_NOFS);
4303         }
4304         trans->blocks_used++;
4305         /* this returns a buffer locked for blocking */
4306         return buf;
4307 }
4308
4309 /*
4310  * helper function to allocate a block for a given tree
4311  * returns the tree buffer or NULL.
4312  */
4313 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4314                                         struct btrfs_root *root, u32 blocksize,
4315                                         u64 parent, u64 root_objectid,
4316                                         struct btrfs_disk_key *key, int level,
4317                                         u64 hint, u64 empty_size)
4318 {
4319         struct btrfs_key ins;
4320         int ret;
4321         struct extent_buffer *buf;
4322
4323         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4324                                key, level, empty_size, hint, (u64)-1, &ins);
4325         if (ret) {
4326                 BUG_ON(ret > 0);
4327                 return ERR_PTR(ret);
4328         }
4329
4330         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4331                                     blocksize, level);
4332         return buf;
4333 }
4334
4335 #if 0
4336 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4337                         struct btrfs_root *root, struct extent_buffer *leaf)
4338 {
4339         u64 disk_bytenr;
4340         u64 num_bytes;
4341         struct btrfs_key key;
4342         struct btrfs_file_extent_item *fi;
4343         u32 nritems;
4344         int i;
4345         int ret;
4346
4347         BUG_ON(!btrfs_is_leaf(leaf));
4348         nritems = btrfs_header_nritems(leaf);
4349
4350         for (i = 0; i < nritems; i++) {
4351                 cond_resched();
4352                 btrfs_item_key_to_cpu(leaf, &key, i);
4353
4354                 /* only extents have references, skip everything else */
4355                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4356                         continue;
4357
4358                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4359
4360                 /* inline extents live in the btree, they don't have refs */
4361                 if (btrfs_file_extent_type(leaf, fi) ==
4362                     BTRFS_FILE_EXTENT_INLINE)
4363                         continue;
4364
4365                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4366
4367                 /* holes don't have refs */
4368                 if (disk_bytenr == 0)
4369                         continue;
4370
4371                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4372                 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4373                                         leaf->start, 0, key.objectid, 0);
4374                 BUG_ON(ret);
4375         }
4376         return 0;
4377 }
4378
4379 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4380                                         struct btrfs_root *root,
4381                                         struct btrfs_leaf_ref *ref)
4382 {
4383         int i;
4384         int ret;
4385         struct btrfs_extent_info *info;
4386         struct refsort *sorted;
4387
4388         if (ref->nritems == 0)
4389                 return 0;
4390
4391         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4392         for (i = 0; i < ref->nritems; i++) {
4393                 sorted[i].bytenr = ref->extents[i].bytenr;
4394                 sorted[i].slot = i;
4395         }
4396         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4397
4398         /*
4399          * the items in the ref were sorted when the ref was inserted
4400          * into the ref cache, so this is already in order
4401          */
4402         for (i = 0; i < ref->nritems; i++) {
4403                 info = ref->extents + sorted[i].slot;
4404                 ret = btrfs_free_extent(trans, root, info->bytenr,
4405                                           info->num_bytes, ref->bytenr,
4406                                           ref->owner, ref->generation,
4407                                           info->objectid, 0);
4408
4409                 atomic_inc(&root->fs_info->throttle_gen);
4410                 wake_up(&root->fs_info->transaction_throttle);
4411                 cond_resched();
4412
4413                 BUG_ON(ret);
4414                 info++;
4415         }
4416
4417         kfree(sorted);
4418         return 0;
4419 }
4420
4421
4422 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4423                                      struct btrfs_root *root, u64 start,
4424                                      u64 len, u32 *refs)
4425 {
4426         int ret;
4427
4428         ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4429         BUG_ON(ret);
4430
4431 #if 0 /* some debugging code in case we see problems here */
4432         /* if the refs count is one, it won't get increased again.  But
4433          * if the ref count is > 1, someone may be decreasing it at
4434          * the same time we are.
4435          */
4436         if (*refs != 1) {
4437                 struct extent_buffer *eb = NULL;
4438                 eb = btrfs_find_create_tree_block(root, start, len);
4439                 if (eb)
4440                         btrfs_tree_lock(eb);
4441
4442                 mutex_lock(&root->fs_info->alloc_mutex);
4443                 ret = lookup_extent_ref(NULL, root, start, len, refs);
4444                 BUG_ON(ret);
4445                 mutex_unlock(&root->fs_info->alloc_mutex);
4446
4447                 if (eb) {
4448                         btrfs_tree_unlock(eb);
4449                         free_extent_buffer(eb);
4450                 }
4451                 if (*refs == 1) {
4452                         printk(KERN_ERR "btrfs block %llu went down to one "
4453                                "during drop_snap\n", (unsigned long long)start);
4454                 }
4455
4456         }
4457 #endif
4458
4459         cond_resched();
4460         return ret;
4461 }
4462
4463
4464 /*
4465  * this is used while deleting old snapshots, and it drops the refs
4466  * on a whole subtree starting from a level 1 node.
4467  *
4468  * The idea is to sort all the leaf pointers, and then drop the
4469  * ref on all the leaves in order.  Most of the time the leaves
4470  * will have ref cache entries, so no leaf IOs will be required to
4471  * find the extents they have references on.
4472  *
4473  * For each leaf, any references it has are also dropped in order
4474  *
4475  * This ends up dropping the references in something close to optimal
4476  * order for reading and modifying the extent allocation tree.
4477  */
4478 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4479                                         struct btrfs_root *root,
4480                                         struct btrfs_path *path)
4481 {
4482         u64 bytenr;
4483         u64 root_owner;
4484         u64 root_gen;
4485         struct extent_buffer *eb = path->nodes[1];
4486         struct extent_buffer *leaf;
4487         struct btrfs_leaf_ref *ref;
4488         struct refsort *sorted = NULL;
4489         int nritems = btrfs_header_nritems(eb);
4490         int ret;
4491         int i;
4492         int refi = 0;
4493         int slot = path->slots[1];
4494         u32 blocksize = btrfs_level_size(root, 0);
4495         u32 refs;
4496
4497         if (nritems == 0)
4498                 goto out;
4499
4500         root_owner = btrfs_header_owner(eb);
4501         root_gen = btrfs_header_generation(eb);
4502         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4503
4504         /*
4505          * step one, sort all the leaf pointers so we don't scribble
4506          * randomly into the extent allocation tree
4507          */
4508         for (i = slot; i < nritems; i++) {
4509                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4510                 sorted[refi].slot = i;
4511                 refi++;
4512         }
4513
4514         /*
4515          * nritems won't be zero, but if we're picking up drop_snapshot
4516          * after a crash, slot might be > 0, so double check things
4517          * just in case.
4518          */
4519         if (refi == 0)
4520                 goto out;
4521
4522         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4523
4524         /*
4525          * the first loop frees everything the leaves point to
4526          */
4527         for (i = 0; i < refi; i++) {
4528                 u64 ptr_gen;
4529
4530                 bytenr = sorted[i].bytenr;
4531
4532                 /*
4533                  * check the reference count on this leaf.  If it is > 1
4534                  * we just decrement it below and don't update any
4535                  * of the refs the leaf points to.
4536                  */
4537                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4538                                                 blocksize, &refs);
4539                 BUG_ON(ret);
4540                 if (refs != 1)
4541                         continue;
4542
4543                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4544
4545                 /*
4546                  * the leaf only had one reference, which means the
4547                  * only thing pointing to this leaf is the snapshot
4548                  * we're deleting.  It isn't possible for the reference
4549                  * count to increase again later
4550                  *
4551                  * The reference cache is checked for the leaf,
4552                  * and if found we'll be able to drop any refs held by
4553                  * the leaf without needing to read it in.
4554                  */
4555                 ref = btrfs_lookup_leaf_ref(root, bytenr);
4556                 if (ref && ref->generation != ptr_gen) {
4557                         btrfs_free_leaf_ref(root, ref);
4558                         ref = NULL;
4559                 }
4560                 if (ref) {
4561                         ret = cache_drop_leaf_ref(trans, root, ref);
4562                         BUG_ON(ret);
4563                         btrfs_remove_leaf_ref(root, ref);
4564                         btrfs_free_leaf_ref(root, ref);
4565                 } else {
4566                         /*
4567                          * the leaf wasn't in the reference cache, so
4568                          * we have to read it.
4569                          */
4570                         leaf = read_tree_block(root, bytenr, blocksize,
4571                                                ptr_gen);
4572                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
4573                         BUG_ON(ret);
4574                         free_extent_buffer(leaf);
4575                 }
4576                 atomic_inc(&root->fs_info->throttle_gen);
4577                 wake_up(&root->fs_info->transaction_throttle);
4578                 cond_resched();
4579         }
4580
4581         /*
4582          * run through the loop again to free the refs on the leaves.
4583          * This is faster than doing it in the loop above because
4584          * the leaves are likely to be clustered together.  We end up
4585          * working in nice chunks on the extent allocation tree.
4586          */
4587         for (i = 0; i < refi; i++) {
4588                 bytenr = sorted[i].bytenr;
4589                 ret = btrfs_free_extent(trans, root, bytenr,
4590                                         blocksize, eb->start,
4591                                         root_owner, root_gen, 0, 1);
4592                 BUG_ON(ret);
4593
4594                 atomic_inc(&root->fs_info->throttle_gen);
4595                 wake_up(&root->fs_info->transaction_throttle);
4596                 cond_resched();
4597         }
4598 out:
4599         kfree(sorted);
4600
4601         /*
4602          * update the path to show we've processed the entire level 1
4603          * node.  This will get saved into the root's drop_snapshot_progress
4604          * field so these drops are not repeated again if this transaction
4605          * commits.
4606          */
4607         path->slots[1] = nritems;
4608         return 0;
4609 }
4610
4611 /*
4612  * helper function for drop_snapshot, this walks down the tree dropping ref
4613  * counts as it goes.
4614  */
4615 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4616                                    struct btrfs_root *root,
4617                                    struct btrfs_path *path, int *level)
4618 {
4619         u64 root_owner;
4620         u64 root_gen;
4621         u64 bytenr;
4622         u64 ptr_gen;
4623         struct extent_buffer *next;
4624         struct extent_buffer *cur;
4625         struct extent_buffer *parent;
4626         u32 blocksize;
4627         int ret;
4628         u32 refs;
4629
4630         WARN_ON(*level < 0);
4631         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4632         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4633                                 path->nodes[*level]->len, &refs);
4634         BUG_ON(ret);
4635         if (refs > 1)
4636                 goto out;
4637
4638         /*
4639          * walk down to the last node level and free all the leaves
4640          */
4641         while (*level >= 0) {
4642                 WARN_ON(*level < 0);
4643                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4644                 cur = path->nodes[*level];
4645
4646                 if (btrfs_header_level(cur) != *level)
4647                         WARN_ON(1);
4648
4649                 if (path->slots[*level] >=
4650                     btrfs_header_nritems(cur))
4651                         break;
4652
4653                 /* the new code goes down to level 1 and does all the
4654                  * leaves pointed to that node in bulk.  So, this check
4655                  * for level 0 will always be false.
4656                  *
4657                  * But, the disk format allows the drop_snapshot_progress
4658                  * field in the root to leave things in a state where
4659                  * a leaf will need cleaning up here.  If someone crashes
4660                  * with the old code and then boots with the new code,
4661                  * we might find a leaf here.
4662                  */
4663                 if (*level == 0) {
4664                         ret = btrfs_drop_leaf_ref(trans, root, cur);
4665                         BUG_ON(ret);
4666                         break;
4667                 }
4668
4669                 /*
4670                  * once we get to level one, process the whole node
4671                  * at once, including everything below it.
4672                  */
4673                 if (*level == 1) {
4674                         ret = drop_level_one_refs(trans, root, path);
4675                         BUG_ON(ret);
4676                         break;
4677                 }
4678
4679                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4680                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4681                 blocksize = btrfs_level_size(root, *level - 1);
4682
4683                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4684                                                 blocksize, &refs);
4685                 BUG_ON(ret);
4686
4687                 /*
4688                  * if there is more than one reference, we don't need
4689                  * to read that node to drop any references it has.  We
4690                  * just drop the ref we hold on that node and move on to the
4691                  * next slot in this level.
4692                  */
4693                 if (refs != 1) {
4694                         parent = path->nodes[*level];
4695                         root_owner = btrfs_header_owner(parent);
4696                         root_gen = btrfs_header_generation(parent);
4697                         path->slots[*level]++;
4698
4699                         ret = btrfs_free_extent(trans, root, bytenr,
4700                                                 blocksize, parent->start,
4701                                                 root_owner, root_gen,
4702                                                 *level - 1, 1);
4703                         BUG_ON(ret);
4704
4705                         atomic_inc(&root->fs_info->throttle_gen);
4706                         wake_up(&root->fs_info->transaction_throttle);
4707                         cond_resched();
4708
4709                         continue;
4710                 }
4711
4712                 /*
4713                  * we need to keep freeing things in the next level down.
4714                  * read the block and loop around to process it
4715                  */
4716                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4717                 WARN_ON(*level <= 0);
4718                 if (path->nodes[*level-1])
4719                         free_extent_buffer(path->nodes[*level-1]);
4720                 path->nodes[*level-1] = next;
4721                 *level = btrfs_header_level(next);
4722                 path->slots[*level] = 0;
4723                 cond_resched();
4724         }
4725 out:
4726         WARN_ON(*level < 0);
4727         WARN_ON(*level >= BTRFS_MAX_LEVEL);
4728
4729         if (path->nodes[*level] == root->node) {
4730                 parent = path->nodes[*level];
4731                 bytenr = path->nodes[*level]->start;
4732         } else {
4733                 parent = path->nodes[*level + 1];
4734                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4735         }
4736
4737         blocksize = btrfs_level_size(root, *level);
4738         root_owner = btrfs_header_owner(parent);
4739         root_gen = btrfs_header_generation(parent);
4740
4741         /*
4742          * cleanup and free the reference on the last node
4743          * we processed
4744          */
4745         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4746                                   parent->start, root_owner, root_gen,
4747                                   *level, 1);
4748         free_extent_buffer(path->nodes[*level]);
4749         path->nodes[*level] = NULL;
4750
4751         *level += 1;
4752         BUG_ON(ret);
4753
4754         cond_resched();
4755         return 0;
4756 }
4757 #endif
4758
4759 struct walk_control {
4760         u64 refs[BTRFS_MAX_LEVEL];
4761         u64 flags[BTRFS_MAX_LEVEL];
4762         struct btrfs_key update_progress;
4763         int stage;
4764         int level;
4765         int shared_level;
4766         int update_ref;
4767         int keep_locks;
4768 };
4769
4770 #define DROP_REFERENCE  1
4771 #define UPDATE_BACKREF  2
4772
4773 /*
4774  * hepler to process tree block while walking down the tree.
4775  *
4776  * when wc->stage == DROP_REFERENCE, this function checks
4777  * reference count of the block. if the block is shared and
4778  * we need update back refs for the subtree rooted at the
4779  * block, this function changes wc->stage to UPDATE_BACKREF
4780  *
4781  * when wc->stage == UPDATE_BACKREF, this function updates
4782  * back refs for pointers in the block.
4783  *
4784  * NOTE: return value 1 means we should stop walking down.
4785  */
4786 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4787                                    struct btrfs_root *root,
4788                                    struct btrfs_path *path,
4789                                    struct walk_control *wc)
4790 {
4791         int level = wc->level;
4792         struct extent_buffer *eb = path->nodes[level];
4793         struct btrfs_key key;
4794         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4795         int ret;
4796
4797         if (wc->stage == UPDATE_BACKREF &&
4798             btrfs_header_owner(eb) != root->root_key.objectid)
4799                 return 1;
4800
4801         /*
4802          * when reference count of tree block is 1, it won't increase
4803          * again. once full backref flag is set, we never clear it.
4804          */
4805         if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4806             (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4807                 BUG_ON(!path->locks[level]);
4808                 ret = btrfs_lookup_extent_info(trans, root,
4809                                                eb->start, eb->len,
4810                                                &wc->refs[level],
4811                                                &wc->flags[level]);
4812                 BUG_ON(ret);
4813                 BUG_ON(wc->refs[level] == 0);
4814         }
4815
4816         if (wc->stage == DROP_REFERENCE &&
4817             wc->update_ref && wc->refs[level] > 1) {
4818                 BUG_ON(eb == root->node);
4819                 BUG_ON(path->slots[level] > 0);
4820                 if (level == 0)
4821                         btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4822                 else
4823                         btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4824                 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4825                     btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4826                         wc->stage = UPDATE_BACKREF;
4827                         wc->shared_level = level;
4828                 }
4829         }
4830
4831         if (wc->stage == DROP_REFERENCE) {
4832                 if (wc->refs[level] > 1)
4833                         return 1;
4834
4835                 if (path->locks[level] && !wc->keep_locks) {
4836                         btrfs_tree_unlock(eb);
4837                         path->locks[level] = 0;
4838                 }
4839                 return 0;
4840         }
4841
4842         /* wc->stage == UPDATE_BACKREF */
4843         if (!(wc->flags[level] & flag)) {
4844                 BUG_ON(!path->locks[level]);
4845                 ret = btrfs_inc_ref(trans, root, eb, 1);
4846                 BUG_ON(ret);
4847                 ret = btrfs_dec_ref(trans, root, eb, 0);
4848                 BUG_ON(ret);
4849                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4850                                                   eb->len, flag, 0);
4851                 BUG_ON(ret);
4852                 wc->flags[level] |= flag;
4853         }
4854
4855         /*
4856          * the block is shared by multiple trees, so it's not good to
4857          * keep the tree lock
4858          */
4859         if (path->locks[level] && level > 0) {
4860                 btrfs_tree_unlock(eb);
4861                 path->locks[level] = 0;
4862         }
4863         return 0;
4864 }
4865
4866 /*
4867  * hepler to process tree block while walking up the tree.
4868  *
4869  * when wc->stage == DROP_REFERENCE, this function drops
4870  * reference count on the block.
4871  *
4872  * when wc->stage == UPDATE_BACKREF, this function changes
4873  * wc->stage back to DROP_REFERENCE if we changed wc->stage
4874  * to UPDATE_BACKREF previously while processing the block.
4875  *
4876  * NOTE: return value 1 means we should stop walking up.
4877  */
4878 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4879                                  struct btrfs_root *root,
4880                                  struct btrfs_path *path,
4881                                  struct walk_control *wc)
4882 {
4883         int ret = 0;
4884         int level = wc->level;
4885         struct extent_buffer *eb = path->nodes[level];
4886         u64 parent = 0;
4887
4888         if (wc->stage == UPDATE_BACKREF) {
4889                 BUG_ON(wc->shared_level < level);
4890                 if (level < wc->shared_level)
4891                         goto out;
4892
4893                 BUG_ON(wc->refs[level] <= 1);
4894                 ret = find_next_key(path, level + 1, &wc->update_progress);
4895                 if (ret > 0)
4896                         wc->update_ref = 0;
4897
4898                 wc->stage = DROP_REFERENCE;
4899                 wc->shared_level = -1;
4900                 path->slots[level] = 0;
4901
4902                 /*
4903                  * check reference count again if the block isn't locked.
4904                  * we should start walking down the tree again if reference
4905                  * count is one.
4906                  */
4907                 if (!path->locks[level]) {
4908                         BUG_ON(level == 0);
4909                         btrfs_tree_lock(eb);
4910                         btrfs_set_lock_blocking(eb);
4911                         path->locks[level] = 1;
4912
4913                         ret = btrfs_lookup_extent_info(trans, root,
4914                                                        eb->start, eb->len,
4915                                                        &wc->refs[level],
4916                                                        &wc->flags[level]);
4917                         BUG_ON(ret);
4918                         BUG_ON(wc->refs[level] == 0);
4919                         if (wc->refs[level] == 1) {
4920                                 btrfs_tree_unlock(eb);
4921                                 path->locks[level] = 0;
4922                                 return 1;
4923                         }
4924                 } else {
4925                         BUG_ON(level != 0);
4926                 }
4927         }
4928
4929         /* wc->stage == DROP_REFERENCE */
4930         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4931
4932         if (wc->refs[level] == 1) {
4933                 if (level == 0) {
4934                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4935                                 ret = btrfs_dec_ref(trans, root, eb, 1);
4936                         else
4937                                 ret = btrfs_dec_ref(trans, root, eb, 0);
4938                         BUG_ON(ret);
4939                 }
4940                 /* make block locked assertion in clean_tree_block happy */
4941                 if (!path->locks[level] &&
4942                     btrfs_header_generation(eb) == trans->transid) {
4943                         btrfs_tree_lock(eb);
4944                         btrfs_set_lock_blocking(eb);
4945                         path->locks[level] = 1;
4946                 }
4947                 clean_tree_block(trans, root, eb);
4948         }
4949
4950         if (eb == root->node) {
4951                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4952                         parent = eb->start;
4953                 else
4954                         BUG_ON(root->root_key.objectid !=
4955                                btrfs_header_owner(eb));
4956         } else {
4957                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4958                         parent = path->nodes[level + 1]->start;
4959                 else
4960                         BUG_ON(root->root_key.objectid !=
4961                                btrfs_header_owner(path->nodes[level + 1]));
4962         }
4963
4964         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
4965                                 root->root_key.objectid, level, 0);
4966         BUG_ON(ret);
4967 out:
4968         wc->refs[level] = 0;
4969         wc->flags[level] = 0;
4970         return ret;
4971 }
4972
4973 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4974                                    struct btrfs_root *root,
4975                                    struct btrfs_path *path,
4976                                    struct walk_control *wc)
4977 {
4978         struct extent_buffer *next;
4979         struct extent_buffer *cur;
4980         u64 bytenr;
4981         u64 ptr_gen;
4982         u32 blocksize;
4983         int level = wc->level;
4984         int ret;
4985
4986         while (level >= 0) {
4987                 cur = path->nodes[level];
4988                 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
4989
4990                 ret = walk_down_proc(trans, root, path, wc);
4991                 if (ret > 0)
4992                         break;
4993
4994                 if (level == 0)
4995                         break;
4996
4997                 bytenr = btrfs_node_blockptr(cur, path->slots[level]);
4998                 blocksize = btrfs_level_size(root, level - 1);
4999                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
5000
5001                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
5002                 btrfs_tree_lock(next);
5003                 btrfs_set_lock_blocking(next);
5004
5005                 level--;
5006                 BUG_ON(level != btrfs_header_level(next));
5007                 path->nodes[level] = next;
5008                 path->slots[level] = 0;
5009                 path->locks[level] = 1;
5010                 wc->level = level;
5011         }
5012         return 0;
5013 }
5014
5015 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5016                                  struct btrfs_root *root,
5017                                  struct btrfs_path *path,
5018                                  struct walk_control *wc, int max_level)
5019 {
5020         int level = wc->level;
5021         int ret;
5022
5023         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5024         while (level < max_level && path->nodes[level]) {
5025                 wc->level = level;
5026                 if (path->slots[level] + 1 <
5027                     btrfs_header_nritems(path->nodes[level])) {
5028                         path->slots[level]++;
5029                         return 0;
5030                 } else {
5031                         ret = walk_up_proc(trans, root, path, wc);
5032                         if (ret > 0)
5033                                 return 0;
5034
5035                         if (path->locks[level]) {
5036                                 btrfs_tree_unlock(path->nodes[level]);
5037                                 path->locks[level] = 0;
5038                         }
5039                         free_extent_buffer(path->nodes[level]);
5040                         path->nodes[level] = NULL;
5041                         level++;
5042                 }
5043         }
5044         return 1;
5045 }
5046
5047 /*
5048  * drop a subvolume tree.
5049  *
5050  * this function traverses the tree freeing any blocks that only
5051  * referenced by the tree.
5052  *
5053  * when a shared tree block is found. this function decreases its
5054  * reference count by one. if update_ref is true, this function
5055  * also make sure backrefs for the shared block and all lower level
5056  * blocks are properly updated.
5057  */
5058 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5059 {
5060         struct btrfs_path *path;
5061         struct btrfs_trans_handle *trans;
5062         struct btrfs_root *tree_root = root->fs_info->tree_root;
5063         struct btrfs_root_item *root_item = &root->root_item;
5064         struct walk_control *wc;
5065         struct btrfs_key key;
5066         int err = 0;
5067         int ret;
5068         int level;
5069
5070         path = btrfs_alloc_path();
5071         BUG_ON(!path);
5072
5073         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5074         BUG_ON(!wc);
5075
5076         trans = btrfs_start_transaction(tree_root, 1);
5077
5078         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5079                 level = btrfs_header_level(root->node);
5080                 path->nodes[level] = btrfs_lock_root_node(root);
5081                 btrfs_set_lock_blocking(path->nodes[level]);
5082                 path->slots[level] = 0;
5083                 path->locks[level] = 1;
5084                 memset(&wc->update_progress, 0,
5085                        sizeof(wc->update_progress));
5086         } else {
5087                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5088                 memcpy(&wc->update_progress, &key,
5089                        sizeof(wc->update_progress));
5090
5091                 level = root_item->drop_level;
5092                 BUG_ON(level == 0);
5093                 path->lowest_level = level;
5094                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5095                 path->lowest_level = 0;
5096                 if (ret < 0) {
5097                         err = ret;
5098                         goto out;
5099                 }
5100                 btrfs_node_key_to_cpu(path->nodes[level], &key,
5101                                       path->slots[level]);
5102                 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
5103
5104                 /*
5105                  * unlock our path, this is safe because only this
5106                  * function is allowed to delete this snapshot
5107                  */
5108                 btrfs_unlock_up_safe(path, 0);
5109
5110                 level = btrfs_header_level(root->node);
5111                 while (1) {
5112                         btrfs_tree_lock(path->nodes[level]);
5113                         btrfs_set_lock_blocking(path->nodes[level]);
5114
5115                         ret = btrfs_lookup_extent_info(trans, root,
5116                                                 path->nodes[level]->start,
5117                                                 path->nodes[level]->len,
5118                                                 &wc->refs[level],
5119                                                 &wc->flags[level]);
5120                         BUG_ON(ret);
5121                         BUG_ON(wc->refs[level] == 0);
5122
5123                         if (level == root_item->drop_level)
5124                                 break;
5125
5126                         btrfs_tree_unlock(path->nodes[level]);
5127                         WARN_ON(wc->refs[level] != 1);
5128                         level--;
5129                 }
5130         }
5131
5132         wc->level = level;
5133         wc->shared_level = -1;
5134         wc->stage = DROP_REFERENCE;
5135         wc->update_ref = update_ref;
5136         wc->keep_locks = 0;
5137
5138         while (1) {
5139                 ret = walk_down_tree(trans, root, path, wc);
5140                 if (ret < 0) {
5141                         err = ret;
5142                         break;
5143                 }
5144
5145                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5146                 if (ret < 0) {
5147                         err = ret;
5148                         break;
5149                 }
5150
5151                 if (ret > 0) {
5152                         BUG_ON(wc->stage != DROP_REFERENCE);
5153                         break;
5154                 }
5155
5156                 if (wc->stage == DROP_REFERENCE) {
5157                         level = wc->level;
5158                         btrfs_node_key(path->nodes[level],
5159                                        &root_item->drop_progress,
5160                                        path->slots[level]);
5161                         root_item->drop_level = level;
5162                 }
5163
5164                 BUG_ON(wc->level == 0);
5165                 if (trans->transaction->in_commit ||
5166                     trans->transaction->delayed_refs.flushing) {
5167                         ret = btrfs_update_root(trans, tree_root,
5168                                                 &root->root_key,
5169                                                 root_item);
5170                         BUG_ON(ret);
5171
5172                         btrfs_end_transaction(trans, tree_root);
5173                         trans = btrfs_start_transaction(tree_root, 1);
5174                 } else {
5175                         unsigned long update;
5176                         update = trans->delayed_ref_updates;
5177                         trans->delayed_ref_updates = 0;
5178                         if (update)
5179                                 btrfs_run_delayed_refs(trans, tree_root,
5180                                                        update);
5181                 }
5182         }
5183         btrfs_release_path(root, path);
5184         BUG_ON(err);
5185
5186         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5187         BUG_ON(ret);
5188
5189         free_extent_buffer(root->node);
5190         free_extent_buffer(root->commit_root);
5191         kfree(root);
5192 out:
5193         btrfs_end_transaction(trans, tree_root);
5194         kfree(wc);
5195         btrfs_free_path(path);
5196         return err;
5197 }
5198
5199 /*
5200  * drop subtree rooted at tree block 'node'.
5201  *
5202  * NOTE: this function will unlock and release tree block 'node'
5203  */
5204 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5205                         struct btrfs_root *root,
5206                         struct extent_buffer *node,
5207                         struct extent_buffer *parent)
5208 {
5209         struct btrfs_path *path;
5210         struct walk_control *wc;
5211         int level;
5212         int parent_level;
5213         int ret = 0;
5214         int wret;
5215
5216         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5217
5218         path = btrfs_alloc_path();
5219         BUG_ON(!path);
5220
5221         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5222         BUG_ON(!wc);
5223
5224         btrfs_assert_tree_locked(parent);
5225         parent_level = btrfs_header_level(parent);
5226         extent_buffer_get(parent);
5227         path->nodes[parent_level] = parent;
5228         path->slots[parent_level] = btrfs_header_nritems(parent);
5229
5230         btrfs_assert_tree_locked(node);
5231         level = btrfs_header_level(node);
5232         path->nodes[level] = node;
5233         path->slots[level] = 0;
5234         path->locks[level] = 1;
5235
5236         wc->refs[parent_level] = 1;
5237         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5238         wc->level = level;
5239         wc->shared_level = -1;
5240         wc->stage = DROP_REFERENCE;
5241         wc->update_ref = 0;
5242         wc->keep_locks = 1;
5243
5244         while (1) {
5245                 wret = walk_down_tree(trans, root, path, wc);
5246                 if (wret < 0) {
5247                         ret = wret;
5248                         break;
5249                 }
5250
5251                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5252                 if (wret < 0)
5253                         ret = wret;
5254                 if (wret != 0)
5255                         break;
5256         }
5257
5258         kfree(wc);
5259         btrfs_free_path(path);
5260         return ret;
5261 }
5262
5263 #if 0
5264 static unsigned long calc_ra(unsigned long start, unsigned long last,
5265                              unsigned long nr)
5266 {
5267         return min(last, start + nr - 1);
5268 }
5269
5270 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5271                                          u64 len)
5272 {
5273         u64 page_start;
5274         u64 page_end;
5275         unsigned long first_index;
5276         unsigned long last_index;
5277         unsigned long i;
5278         struct page *page;
5279         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5280         struct file_ra_state *ra;
5281         struct btrfs_ordered_extent *ordered;
5282         unsigned int total_read = 0;
5283         unsigned int total_dirty = 0;
5284         int ret = 0;
5285
5286         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5287
5288         mutex_lock(&inode->i_mutex);
5289         first_index = start >> PAGE_CACHE_SHIFT;
5290         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5291
5292         /* make sure the dirty trick played by the caller work */
5293         ret = invalidate_inode_pages2_range(inode->i_mapping,
5294                                             first_index, last_index);
5295         if (ret)
5296                 goto out_unlock;
5297
5298         file_ra_state_init(ra, inode->i_mapping);
5299
5300         for (i = first_index ; i <= last_index; i++) {
5301                 if (total_read % ra->ra_pages == 0) {
5302                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5303                                        calc_ra(i, last_index, ra->ra_pages));
5304                 }
5305                 total_read++;
5306 again:
5307                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5308                         BUG_ON(1);
5309                 page = grab_cache_page(inode->i_mapping, i);
5310                 if (!page) {
5311                         ret = -ENOMEM;
5312                         goto out_unlock;
5313                 }
5314                 if (!PageUptodate(page)) {
5315                         btrfs_readpage(NULL, page);
5316                         lock_page(page);
5317                         if (!PageUptodate(page)) {
5318                                 unlock_page(page);
5319                                 page_cache_release(page);
5320                                 ret = -EIO;
5321                                 goto out_unlock;
5322                         }
5323                 }
5324                 wait_on_page_writeback(page);
5325
5326                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5327                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5328                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5329
5330                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5331                 if (ordered) {
5332                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5333                         unlock_page(page);
5334                         page_cache_release(page);
5335                         btrfs_start_ordered_extent(inode, ordered, 1);
5336                         btrfs_put_ordered_extent(ordered);
5337                         goto again;
5338                 }
5339                 set_page_extent_mapped(page);
5340
5341                 if (i == first_index)
5342                         set_extent_bits(io_tree, page_start, page_end,
5343                                         EXTENT_BOUNDARY, GFP_NOFS);
5344                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5345
5346                 set_page_dirty(page);
5347                 total_dirty++;
5348
5349                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5350                 unlock_page(page);
5351                 page_cache_release(page);
5352         }
5353
5354 out_unlock:
5355         kfree(ra);
5356         mutex_unlock(&inode->i_mutex);
5357         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5358         return ret;
5359 }
5360
5361 static noinline int relocate_data_extent(struct inode *reloc_inode,
5362                                          struct btrfs_key *extent_key,
5363                                          u64 offset)
5364 {
5365         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5366         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5367         struct extent_map *em;
5368         u64 start = extent_key->objectid - offset;
5369         u64 end = start + extent_key->offset - 1;
5370
5371         em = alloc_extent_map(GFP_NOFS);
5372         BUG_ON(!em || IS_ERR(em));
5373
5374         em->start = start;
5375         em->len = extent_key->offset;
5376         em->block_len = extent_key->offset;
5377         em->block_start = extent_key->objectid;
5378         em->bdev = root->fs_info->fs_devices->latest_bdev;
5379         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5380
5381         /* setup extent map to cheat btrfs_readpage */
5382         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5383         while (1) {
5384                 int ret;
5385                 spin_lock(&em_tree->lock);
5386                 ret = add_extent_mapping(em_tree, em);
5387                 spin_unlock(&em_tree->lock);
5388                 if (ret != -EEXIST) {
5389                         free_extent_map(em);
5390                         break;
5391                 }
5392                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5393         }
5394         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5395
5396         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5397 }
5398
5399 struct btrfs_ref_path {
5400         u64 extent_start;
5401         u64 nodes[BTRFS_MAX_LEVEL];
5402         u64 root_objectid;
5403         u64 root_generation;
5404         u64 owner_objectid;
5405         u32 num_refs;
5406         int lowest_level;
5407         int current_level;
5408         int shared_level;
5409
5410         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5411         u64 new_nodes[BTRFS_MAX_LEVEL];
5412 };
5413
5414 struct disk_extent {
5415         u64 ram_bytes;
5416         u64 disk_bytenr;
5417         u64 disk_num_bytes;
5418         u64 offset;
5419         u64 num_bytes;
5420         u8 compression;
5421         u8 encryption;
5422         u16 other_encoding;
5423 };
5424
5425 static int is_cowonly_root(u64 root_objectid)
5426 {
5427         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5428             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5429             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5430             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5431             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5432             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5433                 return 1;
5434         return 0;
5435 }
5436
5437 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5438                                     struct btrfs_root *extent_root,
5439                                     struct btrfs_ref_path *ref_path,
5440                                     int first_time)
5441 {
5442         struct extent_buffer *leaf;
5443         struct btrfs_path *path;
5444         struct btrfs_extent_ref *ref;
5445         struct btrfs_key key;
5446         struct btrfs_key found_key;
5447         u64 bytenr;
5448         u32 nritems;
5449         int level;
5450         int ret = 1;
5451
5452         path = btrfs_alloc_path();
5453         if (!path)
5454                 return -ENOMEM;
5455
5456         if (first_time) {
5457                 ref_path->lowest_level = -1;
5458                 ref_path->current_level = -1;
5459                 ref_path->shared_level = -1;
5460                 goto walk_up;
5461         }
5462 walk_down:
5463         level = ref_path->current_level - 1;
5464         while (level >= -1) {
5465                 u64 parent;
5466                 if (level < ref_path->lowest_level)
5467                         break;
5468
5469                 if (level >= 0)
5470                         bytenr = ref_path->nodes[level];
5471                 else
5472                         bytenr = ref_path->extent_start;
5473                 BUG_ON(bytenr == 0);
5474
5475                 parent = ref_path->nodes[level + 1];
5476                 ref_path->nodes[level + 1] = 0;
5477                 ref_path->current_level = level;
5478                 BUG_ON(parent == 0);
5479
5480                 key.objectid = bytenr;
5481                 key.offset = parent + 1;
5482                 key.type = BTRFS_EXTENT_REF_KEY;
5483
5484                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5485                 if (ret < 0)
5486                         goto out;
5487                 BUG_ON(ret == 0);
5488
5489                 leaf = path->nodes[0];
5490                 nritems = btrfs_header_nritems(leaf);
5491                 if (path->slots[0] >= nritems) {
5492                         ret = btrfs_next_leaf(extent_root, path);
5493                         if (ret < 0)
5494                                 goto out;
5495                         if (ret > 0)
5496                                 goto next;
5497                         leaf = path->nodes[0];
5498                 }
5499
5500                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5501                 if (found_key.objectid == bytenr &&
5502                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5503                         if (level < ref_path->shared_level)
5504                                 ref_path->shared_level = level;
5505                         goto found;
5506                 }
5507 next:
5508                 level--;
5509                 btrfs_release_path(extent_root, path);
5510                 cond_resched();
5511         }
5512         /* reached lowest level */
5513         ret = 1;
5514         goto out;
5515 walk_up:
5516         level = ref_path->current_level;
5517         while (level < BTRFS_MAX_LEVEL - 1) {
5518                 u64 ref_objectid;
5519
5520                 if (level >= 0)
5521                         bytenr = ref_path->nodes[level];
5522                 else
5523                         bytenr = ref_path->extent_start;
5524
5525                 BUG_ON(bytenr == 0);
5526
5527                 key.objectid = bytenr;
5528                 key.offset = 0;
5529                 key.type = BTRFS_EXTENT_REF_KEY;
5530
5531                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5532                 if (ret < 0)
5533                         goto out;
5534
5535                 leaf = path->nodes[0];
5536                 nritems = btrfs_header_nritems(leaf);
5537                 if (path->slots[0] >= nritems) {
5538                         ret = btrfs_next_leaf(extent_root, path);
5539                         if (ret < 0)
5540                                 goto out;
5541                         if (ret > 0) {
5542                                 /* the extent was freed by someone */
5543                                 if (ref_path->lowest_level == level)
5544                                         goto out;
5545                                 btrfs_release_path(extent_root, path);
5546                                 goto walk_down;
5547                         }
5548                         leaf = path->nodes[0];
5549                 }
5550
5551                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5552                 if (found_key.objectid != bytenr ||
5553                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5554                         /* the extent was freed by someone */
5555                         if (ref_path->lowest_level == level) {
5556                                 ret = 1;
5557                                 goto out;
5558                         }
5559                         btrfs_release_path(extent_root, path);
5560                         goto walk_down;
5561                 }
5562 found:
5563                 ref = btrfs_item_ptr(leaf, path->slots[0],
5564                                 struct btrfs_extent_ref);
5565                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5566                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5567                         if (first_time) {
5568                                 level = (int)ref_objectid;
5569                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5570                                 ref_path->lowest_level = level;
5571                                 ref_path->current_level = level;
5572                                 ref_path->nodes[level] = bytenr;
5573                         } else {
5574                                 WARN_ON(ref_objectid != level);
5575                         }
5576                 } else {
5577                         WARN_ON(level != -1);
5578                 }
5579                 first_time = 0;
5580
5581                 if (ref_path->lowest_level == level) {
5582                         ref_path->owner_objectid = ref_objectid;
5583                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5584                 }
5585
5586                 /*
5587                  * the block is tree root or the block isn't in reference
5588                  * counted tree.
5589                  */
5590                 if (found_key.objectid == found_key.offset ||
5591                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5592                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5593                         ref_path->root_generation =
5594                                 btrfs_ref_generation(leaf, ref);
5595                         if (level < 0) {
5596                                 /* special reference from the tree log */
5597                                 ref_path->nodes[0] = found_key.offset;
5598                                 ref_path->current_level = 0;
5599                         }
5600                         ret = 0;
5601                         goto out;
5602                 }
5603
5604                 level++;
5605                 BUG_ON(ref_path->nodes[level] != 0);
5606                 ref_path->nodes[level] = found_key.offset;
5607                 ref_path->current_level = level;
5608
5609                 /*
5610                  * the reference was created in the running transaction,
5611                  * no need to continue walking up.
5612                  */
5613                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5614                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5615                         ref_path->root_generation =
5616                                 btrfs_ref_generation(leaf, ref);
5617                         ret = 0;
5618                         goto out;
5619                 }
5620
5621                 btrfs_release_path(extent_root, path);
5622                 cond_resched();
5623         }
5624         /* reached max tree level, but no tree root found. */
5625         BUG();
5626 out:
5627         btrfs_free_path(path);
5628         return ret;
5629 }
5630
5631 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5632                                 struct btrfs_root *extent_root,
5633                                 struct btrfs_ref_path *ref_path,
5634                                 u64 extent_start)
5635 {
5636         memset(ref_path, 0, sizeof(*ref_path));
5637         ref_path->extent_start = extent_start;
5638
5639         return __next_ref_path(trans, extent_root, ref_path, 1);
5640 }
5641
5642 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5643                                struct btrfs_root *extent_root,
5644                                struct btrfs_ref_path *ref_path)
5645 {
5646         return __next_ref_path(trans, extent_root, ref_path, 0);
5647 }
5648
5649 static noinline int get_new_locations(struct inode *reloc_inode,
5650                                       struct btrfs_key *extent_key,
5651                                       u64 offset, int no_fragment,
5652                                       struct disk_extent **extents,
5653                                       int *nr_extents)
5654 {
5655         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5656         struct btrfs_path *path;
5657         struct btrfs_file_extent_item *fi;
5658         struct extent_buffer *leaf;
5659         struct disk_extent *exts = *extents;
5660         struct btrfs_key found_key;
5661         u64 cur_pos;
5662         u64 last_byte;
5663         u32 nritems;
5664         int nr = 0;
5665         int max = *nr_extents;
5666         int ret;
5667
5668         WARN_ON(!no_fragment && *extents);
5669         if (!exts) {
5670                 max = 1;
5671                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5672                 if (!exts)
5673                         return -ENOMEM;
5674         }
5675
5676         path = btrfs_alloc_path();
5677         BUG_ON(!path);
5678
5679         cur_pos = extent_key->objectid - offset;
5680         last_byte = extent_key->objectid + extent_key->offset;
5681         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5682                                        cur_pos, 0);
5683         if (ret < 0)
5684                 goto out;
5685         if (ret > 0) {
5686                 ret = -ENOENT;
5687                 goto out;
5688         }
5689
5690         while (1) {
5691                 leaf = path->nodes[0];
5692                 nritems = btrfs_header_nritems(leaf);
5693                 if (path->slots[0] >= nritems) {
5694                         ret = btrfs_next_leaf(root, path);
5695                         if (ret < 0)
5696                                 goto out;
5697                         if (ret > 0)
5698                                 break;
5699                         leaf = path->nodes[0];
5700                 }
5701
5702                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5703                 if (found_key.offset != cur_pos ||
5704                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
5705                     found_key.objectid != reloc_inode->i_ino)
5706                         break;
5707
5708                 fi = btrfs_item_ptr(leaf, path->slots[0],
5709                                     struct btrfs_file_extent_item);
5710                 if (btrfs_file_extent_type(leaf, fi) !=
5711                     BTRFS_FILE_EXTENT_REG ||
5712                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5713                         break;
5714
5715                 if (nr == max) {
5716                         struct disk_extent *old = exts;
5717                         max *= 2;
5718                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5719                         memcpy(exts, old, sizeof(*exts) * nr);
5720                         if (old != *extents)
5721                                 kfree(old);
5722                 }
5723
5724                 exts[nr].disk_bytenr =
5725                         btrfs_file_extent_disk_bytenr(leaf, fi);
5726                 exts[nr].disk_num_bytes =
5727                         btrfs_file_extent_disk_num_bytes(leaf, fi);
5728                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5729                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5730                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5731                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5732                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5733                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5734                                                                            fi);
5735                 BUG_ON(exts[nr].offset > 0);
5736                 BUG_ON(exts[nr].compression || exts[nr].encryption);
5737                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5738
5739                 cur_pos += exts[nr].num_bytes;
5740                 nr++;
5741
5742                 if (cur_pos + offset >= last_byte)
5743                         break;
5744
5745                 if (no_fragment) {
5746                         ret = 1;
5747                         goto out;
5748                 }
5749                 path->slots[0]++;
5750         }
5751
5752         BUG_ON(cur_pos + offset > last_byte);
5753         if (cur_pos + offset < last_byte) {
5754                 ret = -ENOENT;
5755                 goto out;
5756         }
5757         ret = 0;
5758 out:
5759         btrfs_free_path(path);
5760         if (ret) {
5761                 if (exts != *extents)
5762                         kfree(exts);
5763         } else {
5764                 *extents = exts;
5765                 *nr_extents = nr;
5766         }
5767         return ret;
5768 }
5769
5770 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5771                                         struct btrfs_root *root,
5772                                         struct btrfs_path *path,
5773                                         struct btrfs_key *extent_key,
5774                                         struct btrfs_key *leaf_key,
5775                                         struct btrfs_ref_path *ref_path,
5776                                         struct disk_extent *new_extents,
5777                                         int nr_extents)
5778 {
5779         struct extent_buffer *leaf;
5780         struct btrfs_file_extent_item *fi;
5781         struct inode *inode = NULL;
5782         struct btrfs_key key;
5783         u64 lock_start = 0;
5784         u64 lock_end = 0;
5785         u64 num_bytes;
5786         u64 ext_offset;
5787         u64 search_end = (u64)-1;
5788         u32 nritems;
5789         int nr_scaned = 0;
5790         int extent_locked = 0;
5791         int extent_type;
5792         int ret;
5793
5794         memcpy(&key, leaf_key, sizeof(key));
5795         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5796                 if (key.objectid < ref_path->owner_objectid ||
5797                     (key.objectid == ref_path->owner_objectid &&
5798                      key.type < BTRFS_EXTENT_DATA_KEY)) {
5799                         key.objectid = ref_path->owner_objectid;
5800                         key.type = BTRFS_EXTENT_DATA_KEY;
5801                         key.offset = 0;
5802                 }
5803         }
5804
5805         while (1) {
5806                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5807                 if (ret < 0)
5808                         goto out;
5809
5810                 leaf = path->nodes[0];
5811                 nritems = btrfs_header_nritems(leaf);
5812 next:
5813                 if (extent_locked && ret > 0) {
5814                         /*
5815                          * the file extent item was modified by someone
5816                          * before the extent got locked.
5817                          */
5818                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5819                                       lock_end, GFP_NOFS);
5820                         extent_locked = 0;
5821                 }
5822
5823                 if (path->slots[0] >= nritems) {
5824                         if (++nr_scaned > 2)
5825                                 break;
5826
5827                         BUG_ON(extent_locked);
5828                         ret = btrfs_next_leaf(root, path);
5829                         if (ret < 0)
5830                                 goto out;
5831                         if (ret > 0)
5832                                 break;
5833                         leaf = path->nodes[0];
5834                         nritems = btrfs_header_nritems(leaf);
5835                 }
5836
5837                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5838
5839                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5840                         if ((key.objectid > ref_path->owner_objectid) ||
5841                             (key.objectid == ref_path->owner_objectid &&
5842                              key.type > BTRFS_EXTENT_DATA_KEY) ||
5843                             key.offset >= search_end)
5844                                 break;
5845                 }
5846
5847                 if (inode && key.objectid != inode->i_ino) {
5848                         BUG_ON(extent_locked);
5849                         btrfs_release_path(root, path);
5850                         mutex_unlock(&inode->i_mutex);
5851                         iput(inode);
5852                         inode = NULL;
5853                         continue;
5854                 }
5855
5856                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5857                         path->slots[0]++;
5858                         ret = 1;
5859                         goto next;
5860                 }
5861                 fi = btrfs_item_ptr(leaf, path->slots[0],
5862                                     struct btrfs_file_extent_item);
5863                 extent_type = btrfs_file_extent_type(leaf, fi);
5864                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5865                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5866                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5867                      extent_key->objectid)) {
5868                         path->slots[0]++;
5869                         ret = 1;
5870                         goto next;
5871                 }
5872
5873                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5874                 ext_offset = btrfs_file_extent_offset(leaf, fi);
5875
5876                 if (search_end == (u64)-1) {
5877                         search_end = key.offset - ext_offset +
5878                                 btrfs_file_extent_ram_bytes(leaf, fi);
5879                 }
5880
5881                 if (!extent_locked) {
5882                         lock_start = key.offset;
5883                         lock_end = lock_start + num_bytes - 1;
5884                 } else {
5885                         if (lock_start > key.offset ||
5886                             lock_end + 1 < key.offset + num_bytes) {
5887                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5888                                               lock_start, lock_end, GFP_NOFS);
5889                                 extent_locked = 0;
5890                         }
5891                 }
5892
5893                 if (!inode) {
5894                         btrfs_release_path(root, path);
5895
5896                         inode = btrfs_iget_locked(root->fs_info->sb,
5897                                                   key.objectid, root);
5898                         if (inode->i_state & I_NEW) {
5899                                 BTRFS_I(inode)->root = root;
5900                                 BTRFS_I(inode)->location.objectid =
5901                                         key.objectid;
5902                                 BTRFS_I(inode)->location.type =
5903                                         BTRFS_INODE_ITEM_KEY;
5904                                 BTRFS_I(inode)->location.offset = 0;
5905                                 btrfs_read_locked_inode(inode);
5906                                 unlock_new_inode(inode);
5907                         }
5908                         /*
5909                          * some code call btrfs_commit_transaction while
5910                          * holding the i_mutex, so we can't use mutex_lock
5911                          * here.
5912                          */
5913                         if (is_bad_inode(inode) ||
5914                             !mutex_trylock(&inode->i_mutex)) {
5915                                 iput(inode);
5916                                 inode = NULL;
5917                                 key.offset = (u64)-1;
5918                                 goto skip;
5919                         }
5920                 }
5921
5922                 if (!extent_locked) {
5923                         struct btrfs_ordered_extent *ordered;
5924
5925                         btrfs_release_path(root, path);
5926
5927                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5928                                     lock_end, GFP_NOFS);
5929                         ordered = btrfs_lookup_first_ordered_extent(inode,
5930                                                                     lock_end);
5931                         if (ordered &&
5932                             ordered->file_offset <= lock_end &&
5933                             ordered->file_offset + ordered->len > lock_start) {
5934                                 unlock_extent(&BTRFS_I(inode)->io_tree,
5935                                               lock_start, lock_end, GFP_NOFS);
5936                                 btrfs_start_ordered_extent(inode, ordered, 1);
5937                                 btrfs_put_ordered_extent(ordered);
5938                                 key.offset += num_bytes;
5939                                 goto skip;
5940                         }
5941                         if (ordered)
5942                                 btrfs_put_ordered_extent(ordered);
5943
5944                         extent_locked = 1;
5945                         continue;
5946                 }
5947
5948                 if (nr_extents == 1) {
5949                         /* update extent pointer in place */
5950                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
5951                                                 new_extents[0].disk_bytenr);
5952                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5953                                                 new_extents[0].disk_num_bytes);
5954                         btrfs_mark_buffer_dirty(leaf);
5955
5956                         btrfs_drop_extent_cache(inode, key.offset,
5957                                                 key.offset + num_bytes - 1, 0);
5958
5959                         ret = btrfs_inc_extent_ref(trans, root,
5960                                                 new_extents[0].disk_bytenr,
5961                                                 new_extents[0].disk_num_bytes,
5962                                                 leaf->start,
5963                                                 root->root_key.objectid,
5964                                                 trans->transid,
5965                                                 key.objectid);
5966                         BUG_ON(ret);
5967
5968                         ret = btrfs_free_extent(trans, root,
5969                                                 extent_key->objectid,
5970                                                 extent_key->offset,
5971                                                 leaf->start,
5972                                                 btrfs_header_owner(leaf),
5973                                                 btrfs_header_generation(leaf),
5974                                                 key.objectid, 0);
5975                         BUG_ON(ret);
5976
5977                         btrfs_release_path(root, path);
5978                         key.offset += num_bytes;
5979                 } else {
5980                         BUG_ON(1);
5981 #if 0
5982                         u64 alloc_hint;
5983                         u64 extent_len;
5984                         int i;
5985                         /*
5986                          * drop old extent pointer at first, then insert the
5987                          * new pointers one bye one
5988                          */
5989                         btrfs_release_path(root, path);
5990                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
5991                                                  key.offset + num_bytes,
5992                                                  key.offset, &alloc_hint);
5993                         BUG_ON(ret);
5994
5995                         for (i = 0; i < nr_extents; i++) {
5996                                 if (ext_offset >= new_extents[i].num_bytes) {
5997                                         ext_offset -= new_extents[i].num_bytes;
5998                                         continue;
5999                                 }
6000                                 extent_len = min(new_extents[i].num_bytes -
6001                                                  ext_offset, num_bytes);
6002
6003                                 ret = btrfs_insert_empty_item(trans, root,
6004                                                               path, &key,
6005                                                               sizeof(*fi));
6006                                 BUG_ON(ret);
6007
6008                                 leaf = path->nodes[0];
6009                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6010                                                 struct btrfs_file_extent_item);
6011                                 btrfs_set_file_extent_generation(leaf, fi,
6012                                                         trans->transid);
6013                                 btrfs_set_file_extent_type(leaf, fi,
6014                                                         BTRFS_FILE_EXTENT_REG);
6015                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6016                                                 new_extents[i].disk_bytenr);
6017                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6018                                                 new_extents[i].disk_num_bytes);
6019                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6020                                                 new_extents[i].ram_bytes);
6021
6022                                 btrfs_set_file_extent_compression(leaf, fi,
6023                                                 new_extents[i].compression);
6024                                 btrfs_set_file_extent_encryption(leaf, fi,
6025                                                 new_extents[i].encryption);
6026                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6027                                                 new_extents[i].other_encoding);
6028
6029                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6030                                                         extent_len);
6031                                 ext_offset += new_extents[i].offset;
6032                                 btrfs_set_file_extent_offset(leaf, fi,
6033                                                         ext_offset);
6034                                 btrfs_mark_buffer_dirty(leaf);
6035
6036                                 btrfs_drop_extent_cache(inode, key.offset,
6037                                                 key.offset + extent_len - 1, 0);
6038
6039                                 ret = btrfs_inc_extent_ref(trans, root,
6040                                                 new_extents[i].disk_bytenr,
6041                                                 new_extents[i].disk_num_bytes,
6042                                                 leaf->start,
6043                                                 root->root_key.objectid,
6044                                                 trans->transid, key.objectid);
6045                                 BUG_ON(ret);
6046                                 btrfs_release_path(root, path);
6047
6048                                 inode_add_bytes(inode, extent_len);
6049
6050                                 ext_offset = 0;
6051                                 num_bytes -= extent_len;
6052                                 key.offset += extent_len;
6053
6054                                 if (num_bytes == 0)
6055                                         break;
6056                         }
6057                         BUG_ON(i >= nr_extents);
6058 #endif
6059                 }
6060
6061                 if (extent_locked) {
6062                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6063                                       lock_end, GFP_NOFS);
6064                         extent_locked = 0;
6065                 }
6066 skip:
6067                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6068                     key.offset >= search_end)
6069                         break;
6070
6071                 cond_resched();
6072         }
6073         ret = 0;
6074 out:
6075         btrfs_release_path(root, path);
6076         if (inode) {
6077                 mutex_unlock(&inode->i_mutex);
6078                 if (extent_locked) {
6079                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6080                                       lock_end, GFP_NOFS);
6081                 }
6082                 iput(inode);
6083         }
6084         return ret;
6085 }
6086
6087 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6088                                struct btrfs_root *root,
6089                                struct extent_buffer *buf, u64 orig_start)
6090 {
6091         int level;
6092         int ret;
6093
6094         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6095         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6096
6097         level = btrfs_header_level(buf);
6098         if (level == 0) {
6099                 struct btrfs_leaf_ref *ref;
6100                 struct btrfs_leaf_ref *orig_ref;
6101
6102                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6103                 if (!orig_ref)
6104                         return -ENOENT;
6105
6106                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6107                 if (!ref) {
6108                         btrfs_free_leaf_ref(root, orig_ref);
6109                         return -ENOMEM;
6110                 }
6111
6112                 ref->nritems = orig_ref->nritems;
6113                 memcpy(ref->extents, orig_ref->extents,
6114                         sizeof(ref->extents[0]) * ref->nritems);
6115
6116                 btrfs_free_leaf_ref(root, orig_ref);
6117
6118                 ref->root_gen = trans->transid;
6119                 ref->bytenr = buf->start;
6120                 ref->owner = btrfs_header_owner(buf);
6121                 ref->generation = btrfs_header_generation(buf);
6122
6123                 ret = btrfs_add_leaf_ref(root, ref, 0);
6124                 WARN_ON(ret);
6125                 btrfs_free_leaf_ref(root, ref);
6126         }
6127         return 0;
6128 }
6129
6130 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6131                                         struct extent_buffer *leaf,
6132                                         struct btrfs_block_group_cache *group,
6133                                         struct btrfs_root *target_root)
6134 {
6135         struct btrfs_key key;
6136         struct inode *inode = NULL;
6137         struct btrfs_file_extent_item *fi;
6138         u64 num_bytes;
6139         u64 skip_objectid = 0;
6140         u32 nritems;
6141         u32 i;
6142
6143         nritems = btrfs_header_nritems(leaf);
6144         for (i = 0; i < nritems; i++) {
6145                 btrfs_item_key_to_cpu(leaf, &key, i);
6146                 if (key.objectid == skip_objectid ||
6147                     key.type != BTRFS_EXTENT_DATA_KEY)
6148                         continue;
6149                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6150                 if (btrfs_file_extent_type(leaf, fi) ==
6151                     BTRFS_FILE_EXTENT_INLINE)
6152                         continue;
6153                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6154                         continue;
6155                 if (!inode || inode->i_ino != key.objectid) {
6156                         iput(inode);
6157                         inode = btrfs_ilookup(target_root->fs_info->sb,
6158                                               key.objectid, target_root, 1);
6159                 }
6160                 if (!inode) {
6161                         skip_objectid = key.objectid;
6162                         continue;
6163                 }
6164                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6165
6166                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6167                             key.offset + num_bytes - 1, GFP_NOFS);
6168                 btrfs_drop_extent_cache(inode, key.offset,
6169                                         key.offset + num_bytes - 1, 1);
6170                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6171                               key.offset + num_bytes - 1, GFP_NOFS);
6172                 cond_resched();
6173         }
6174         iput(inode);
6175         return 0;
6176 }
6177
6178 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6179                                         struct btrfs_root *root,
6180                                         struct extent_buffer *leaf,
6181                                         struct btrfs_block_group_cache *group,
6182                                         struct inode *reloc_inode)
6183 {
6184         struct btrfs_key key;
6185         struct btrfs_key extent_key;
6186         struct btrfs_file_extent_item *fi;
6187         struct btrfs_leaf_ref *ref;
6188         struct disk_extent *new_extent;
6189         u64 bytenr;
6190         u64 num_bytes;
6191         u32 nritems;
6192         u32 i;
6193         int ext_index;
6194         int nr_extent;
6195         int ret;
6196
6197         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6198         BUG_ON(!new_extent);
6199
6200         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6201         BUG_ON(!ref);
6202
6203         ext_index = -1;
6204         nritems = btrfs_header_nritems(leaf);
6205         for (i = 0; i < nritems; i++) {
6206                 btrfs_item_key_to_cpu(leaf, &key, i);
6207                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6208                         continue;
6209                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6210                 if (btrfs_file_extent_type(leaf, fi) ==
6211                     BTRFS_FILE_EXTENT_INLINE)
6212                         continue;
6213                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6214                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6215                 if (bytenr == 0)
6216                         continue;
6217
6218                 ext_index++;
6219                 if (bytenr >= group->key.objectid + group->key.offset ||
6220                     bytenr + num_bytes <= group->key.objectid)
6221                         continue;
6222
6223                 extent_key.objectid = bytenr;
6224                 extent_key.offset = num_bytes;
6225                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6226                 nr_extent = 1;
6227                 ret = get_new_locations(reloc_inode, &extent_key,
6228                                         group->key.objectid, 1,
6229                                         &new_extent, &nr_extent);
6230                 if (ret > 0)
6231                         continue;
6232                 BUG_ON(ret < 0);
6233
6234                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6235                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6236                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6237                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6238
6239                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6240                                                 new_extent->disk_bytenr);
6241                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6242                                                 new_extent->disk_num_bytes);
6243                 btrfs_mark_buffer_dirty(leaf);
6244
6245                 ret = btrfs_inc_extent_ref(trans, root,
6246                                         new_extent->disk_bytenr,
6247                                         new_extent->disk_num_bytes,
6248                                         leaf->start,
6249                                         root->root_key.objectid,
6250                                         trans->transid, key.objectid);
6251                 BUG_ON(ret);
6252
6253                 ret = btrfs_free_extent(trans, root,
6254                                         bytenr, num_bytes, leaf->start,
6255                                         btrfs_header_owner(leaf),
6256                                         btrfs_header_generation(leaf),
6257                                         key.objectid, 0);
6258                 BUG_ON(ret);
6259                 cond_resched();
6260         }
6261         kfree(new_extent);
6262         BUG_ON(ext_index + 1 != ref->nritems);
6263         btrfs_free_leaf_ref(root, ref);
6264         return 0;
6265 }
6266
6267 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6268                           struct btrfs_root *root)
6269 {
6270         struct btrfs_root *reloc_root;
6271         int ret;
6272
6273         if (root->reloc_root) {
6274                 reloc_root = root->reloc_root;
6275                 root->reloc_root = NULL;
6276                 list_add(&reloc_root->dead_list,
6277                          &root->fs_info->dead_reloc_roots);
6278
6279                 btrfs_set_root_bytenr(&reloc_root->root_item,
6280                                       reloc_root->node->start);
6281                 btrfs_set_root_level(&root->root_item,
6282                                      btrfs_header_level(reloc_root->node));
6283                 memset(&reloc_root->root_item.drop_progress, 0,
6284                         sizeof(struct btrfs_disk_key));
6285                 reloc_root->root_item.drop_level = 0;
6286
6287                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6288                                         &reloc_root->root_key,
6289                                         &reloc_root->root_item);
6290                 BUG_ON(ret);
6291         }
6292         return 0;
6293 }
6294
6295 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6296 {
6297         struct btrfs_trans_handle *trans;
6298         struct btrfs_root *reloc_root;
6299         struct btrfs_root *prev_root = NULL;
6300         struct list_head dead_roots;
6301         int ret;
6302         unsigned long nr;
6303
6304         INIT_LIST_HEAD(&dead_roots);
6305         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6306
6307         while (!list_empty(&dead_roots)) {
6308                 reloc_root = list_entry(dead_roots.prev,
6309                                         struct btrfs_root, dead_list);
6310                 list_del_init(&reloc_root->dead_list);
6311
6312                 BUG_ON(reloc_root->commit_root != NULL);
6313                 while (1) {
6314                         trans = btrfs_join_transaction(root, 1);
6315                         BUG_ON(!trans);
6316
6317                         mutex_lock(&root->fs_info->drop_mutex);
6318                         ret = btrfs_drop_snapshot(trans, reloc_root);
6319                         if (ret != -EAGAIN)
6320                                 break;
6321                         mutex_unlock(&root->fs_info->drop_mutex);
6322
6323                         nr = trans->blocks_used;
6324                         ret = btrfs_end_transaction(trans, root);
6325                         BUG_ON(ret);
6326                         btrfs_btree_balance_dirty(root, nr);
6327                 }
6328
6329                 free_extent_buffer(reloc_root->node);
6330
6331                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6332                                      &reloc_root->root_key);
6333                 BUG_ON(ret);
6334                 mutex_unlock(&root->fs_info->drop_mutex);
6335
6336                 nr = trans->blocks_used;
6337                 ret = btrfs_end_transaction(trans, root);
6338                 BUG_ON(ret);
6339                 btrfs_btree_balance_dirty(root, nr);
6340
6341                 kfree(prev_root);
6342                 prev_root = reloc_root;
6343         }
6344         if (prev_root) {
6345                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6346                 kfree(prev_root);
6347         }
6348         return 0;
6349 }
6350
6351 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6352 {
6353         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6354         return 0;
6355 }
6356
6357 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6358 {
6359         struct btrfs_root *reloc_root;
6360         struct btrfs_trans_handle *trans;
6361         struct btrfs_key location;
6362         int found;
6363         int ret;
6364
6365         mutex_lock(&root->fs_info->tree_reloc_mutex);
6366         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6367         BUG_ON(ret);
6368         found = !list_empty(&root->fs_info->dead_reloc_roots);
6369         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6370
6371         if (found) {
6372                 trans = btrfs_start_transaction(root, 1);
6373                 BUG_ON(!trans);
6374                 ret = btrfs_commit_transaction(trans, root);
6375                 BUG_ON(ret);
6376         }
6377
6378         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6379         location.offset = (u64)-1;
6380         location.type = BTRFS_ROOT_ITEM_KEY;
6381
6382         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6383         BUG_ON(!reloc_root);
6384         btrfs_orphan_cleanup(reloc_root);
6385         return 0;
6386 }
6387
6388 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6389                                     struct btrfs_root *root)
6390 {
6391         struct btrfs_root *reloc_root;
6392         struct extent_buffer *eb;
6393         struct btrfs_root_item *root_item;
6394         struct btrfs_key root_key;
6395         int ret;
6396
6397         BUG_ON(!root->ref_cows);
6398         if (root->reloc_root)
6399                 return 0;
6400
6401         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6402         BUG_ON(!root_item);
6403
6404         ret = btrfs_copy_root(trans, root, root->commit_root,
6405                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6406         BUG_ON(ret);
6407
6408         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6409         root_key.offset = root->root_key.objectid;
6410         root_key.type = BTRFS_ROOT_ITEM_KEY;
6411
6412         memcpy(root_item, &root->root_item, sizeof(root_item));
6413         btrfs_set_root_refs(root_item, 0);
6414         btrfs_set_root_bytenr(root_item, eb->start);
6415         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6416         btrfs_set_root_generation(root_item, trans->transid);
6417
6418         btrfs_tree_unlock(eb);
6419         free_extent_buffer(eb);
6420
6421         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6422                                 &root_key, root_item);
6423         BUG_ON(ret);
6424         kfree(root_item);
6425
6426         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6427                                                  &root_key);
6428         BUG_ON(!reloc_root);
6429         reloc_root->last_trans = trans->transid;
6430         reloc_root->commit_root = NULL;
6431         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6432
6433         root->reloc_root = reloc_root;
6434         return 0;
6435 }
6436
6437 /*
6438  * Core function of space balance.
6439  *
6440  * The idea is using reloc trees to relocate tree blocks in reference
6441  * counted roots. There is one reloc tree for each subvol, and all
6442  * reloc trees share same root key objectid. Reloc trees are snapshots
6443  * of the latest committed roots of subvols (root->commit_root).
6444  *
6445  * To relocate a tree block referenced by a subvol, there are two steps.
6446  * COW the block through subvol's reloc tree, then update block pointer
6447  * in the subvol to point to the new block. Since all reloc trees share
6448  * same root key objectid, doing special handing for tree blocks owned
6449  * by them is easy. Once a tree block has been COWed in one reloc tree,
6450  * we can use the resulting new block directly when the same block is
6451  * required to COW again through other reloc trees. By this way, relocated
6452  * tree blocks are shared between reloc trees, so they are also shared
6453  * between subvols.
6454  */
6455 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6456                                       struct btrfs_root *root,
6457                                       struct btrfs_path *path,
6458                                       struct btrfs_key *first_key,
6459                                       struct btrfs_ref_path *ref_path,
6460                                       struct btrfs_block_group_cache *group,
6461                                       struct inode *reloc_inode)
6462 {
6463         struct btrfs_root *reloc_root;
6464         struct extent_buffer *eb = NULL;
6465         struct btrfs_key *keys;
6466         u64 *nodes;
6467         int level;
6468         int shared_level;
6469         int lowest_level = 0;
6470         int ret;
6471
6472         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6473                 lowest_level = ref_path->owner_objectid;
6474
6475         if (!root->ref_cows) {
6476                 path->lowest_level = lowest_level;
6477                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6478                 BUG_ON(ret < 0);
6479                 path->lowest_level = 0;
6480                 btrfs_release_path(root, path);
6481                 return 0;
6482         }
6483
6484         mutex_lock(&root->fs_info->tree_reloc_mutex);
6485         ret = init_reloc_tree(trans, root);
6486         BUG_ON(ret);
6487         reloc_root = root->reloc_root;
6488
6489         shared_level = ref_path->shared_level;
6490         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6491
6492         keys = ref_path->node_keys;
6493         nodes = ref_path->new_nodes;
6494         memset(&keys[shared_level + 1], 0,
6495                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6496         memset(&nodes[shared_level + 1], 0,
6497                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6498
6499         if (nodes[lowest_level] == 0) {
6500                 path->lowest_level = lowest_level;
6501                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6502                                         0, 1);
6503                 BUG_ON(ret);
6504                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6505                         eb = path->nodes[level];
6506                         if (!eb || eb == reloc_root->node)
6507                                 break;
6508                         nodes[level] = eb->start;
6509                         if (level == 0)
6510                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6511                         else
6512                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6513                 }
6514                 if (nodes[0] &&
6515                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6516                         eb = path->nodes[0];
6517                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6518                                                       group, reloc_inode);
6519                         BUG_ON(ret);
6520                 }
6521                 btrfs_release_path(reloc_root, path);
6522         } else {
6523                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6524                                        lowest_level);
6525                 BUG_ON(ret);
6526         }
6527
6528         /*
6529          * replace tree blocks in the fs tree with tree blocks in
6530          * the reloc tree.
6531          */
6532         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6533         BUG_ON(ret < 0);
6534
6535         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6536                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6537                                         0, 0);
6538                 BUG_ON(ret);
6539                 extent_buffer_get(path->nodes[0]);
6540                 eb = path->nodes[0];
6541                 btrfs_release_path(reloc_root, path);
6542                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6543                 BUG_ON(ret);
6544                 free_extent_buffer(eb);
6545         }
6546
6547         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6548         path->lowest_level = 0;
6549         return 0;
6550 }
6551
6552 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6553                                         struct btrfs_root *root,
6554                                         struct btrfs_path *path,
6555                                         struct btrfs_key *first_key,
6556                                         struct btrfs_ref_path *ref_path)
6557 {
6558         int ret;
6559
6560         ret = relocate_one_path(trans, root, path, first_key,
6561                                 ref_path, NULL, NULL);
6562         BUG_ON(ret);
6563
6564         return 0;
6565 }
6566
6567 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6568                                     struct btrfs_root *extent_root,
6569                                     struct btrfs_path *path,
6570                                     struct btrfs_key *extent_key)
6571 {
6572         int ret;
6573
6574         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6575         if (ret)
6576                 goto out;
6577         ret = btrfs_del_item(trans, extent_root, path);
6578 out:
6579         btrfs_release_path(extent_root, path);
6580         return ret;
6581 }
6582
6583 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6584                                                 struct btrfs_ref_path *ref_path)
6585 {
6586         struct btrfs_key root_key;
6587
6588         root_key.objectid = ref_path->root_objectid;
6589         root_key.type = BTRFS_ROOT_ITEM_KEY;
6590         if (is_cowonly_root(ref_path->root_objectid))
6591                 root_key.offset = 0;
6592         else
6593                 root_key.offset = (u64)-1;
6594
6595         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6596 }
6597
6598 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6599                                         struct btrfs_path *path,
6600                                         struct btrfs_key *extent_key,
6601                                         struct btrfs_block_group_cache *group,
6602                                         struct inode *reloc_inode, int pass)
6603 {
6604         struct btrfs_trans_handle *trans;
6605         struct btrfs_root *found_root;
6606         struct btrfs_ref_path *ref_path = NULL;
6607         struct disk_extent *new_extents = NULL;
6608         int nr_extents = 0;
6609         int loops;
6610         int ret;
6611         int level;
6612         struct btrfs_key first_key;
6613         u64 prev_block = 0;
6614
6615
6616         trans = btrfs_start_transaction(extent_root, 1);
6617         BUG_ON(!trans);
6618
6619         if (extent_key->objectid == 0) {
6620                 ret = del_extent_zero(trans, extent_root, path, extent_key);
6621                 goto out;
6622         }
6623
6624         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6625         if (!ref_path) {
6626                 ret = -ENOMEM;
6627                 goto out;
6628         }
6629
6630         for (loops = 0; ; loops++) {
6631                 if (loops == 0) {
6632                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6633                                                    extent_key->objectid);
6634                 } else {
6635                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6636                 }
6637                 if (ret < 0)
6638                         goto out;
6639                 if (ret > 0)
6640                         break;
6641
6642                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6643                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6644                         continue;
6645
6646                 found_root = read_ref_root(extent_root->fs_info, ref_path);
6647                 BUG_ON(!found_root);
6648                 /*
6649                  * for reference counted tree, only process reference paths
6650                  * rooted at the latest committed root.
6651                  */
6652                 if (found_root->ref_cows &&
6653                     ref_path->root_generation != found_root->root_key.offset)
6654                         continue;
6655
6656                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6657                         if (pass == 0) {
6658                                 /*
6659                                  * copy data extents to new locations
6660                                  */
6661                                 u64 group_start = group->key.objectid;
6662                                 ret = relocate_data_extent(reloc_inode,
6663                                                            extent_key,
6664                                                            group_start);
6665                                 if (ret < 0)
6666                                         goto out;
6667                                 break;
6668                         }
6669                         level = 0;
6670                 } else {
6671                         level = ref_path->owner_objectid;
6672                 }
6673
6674                 if (prev_block != ref_path->nodes[level]) {
6675                         struct extent_buffer *eb;
6676                         u64 block_start = ref_path->nodes[level];
6677                         u64 block_size = btrfs_level_size(found_root, level);
6678
6679                         eb = read_tree_block(found_root, block_start,
6680                                              block_size, 0);
6681                         btrfs_tree_lock(eb);
6682                         BUG_ON(level != btrfs_header_level(eb));
6683
6684                         if (level == 0)
6685                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
6686                         else
6687                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
6688
6689                         btrfs_tree_unlock(eb);
6690                         free_extent_buffer(eb);
6691                         prev_block = block_start;
6692                 }
6693
6694                 mutex_lock(&extent_root->fs_info->trans_mutex);
6695                 btrfs_record_root_in_trans(found_root);
6696                 mutex_unlock(&extent_root->fs_info->trans_mutex);
6697                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6698                         /*
6699                          * try to update data extent references while
6700                          * keeping metadata shared between snapshots.
6701                          */
6702                         if (pass == 1) {
6703                                 ret = relocate_one_path(trans, found_root,
6704                                                 path, &first_key, ref_path,
6705                                                 group, reloc_inode);
6706                                 if (ret < 0)
6707                                         goto out;
6708                                 continue;
6709                         }
6710                         /*
6711                          * use fallback method to process the remaining
6712                          * references.
6713                          */
6714                         if (!new_extents) {
6715                                 u64 group_start = group->key.objectid;
6716                                 new_extents = kmalloc(sizeof(*new_extents),
6717                                                       GFP_NOFS);
6718                                 nr_extents = 1;
6719                                 ret = get_new_locations(reloc_inode,
6720                                                         extent_key,
6721                                                         group_start, 1,
6722                                                         &new_extents,
6723                                                         &nr_extents);
6724                                 if (ret)
6725                                         goto out;
6726                         }
6727                         ret = replace_one_extent(trans, found_root,
6728                                                 path, extent_key,
6729                                                 &first_key, ref_path,
6730                                                 new_extents, nr_extents);
6731                 } else {
6732                         ret = relocate_tree_block(trans, found_root, path,
6733                                                   &first_key, ref_path);
6734                 }
6735                 if (ret < 0)
6736                         goto out;
6737         }
6738         ret = 0;
6739 out:
6740         btrfs_end_transaction(trans, extent_root);
6741         kfree(new_extents);
6742         kfree(ref_path);
6743         return ret;
6744 }
6745 #endif
6746
6747 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6748 {
6749         u64 num_devices;
6750         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6751                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6752
6753         num_devices = root->fs_info->fs_devices->rw_devices;
6754         if (num_devices == 1) {
6755                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6756                 stripped = flags & ~stripped;
6757
6758                 /* turn raid0 into single device chunks */
6759                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6760                         return stripped;
6761
6762                 /* turn mirroring into duplication */
6763                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6764                              BTRFS_BLOCK_GROUP_RAID10))
6765                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6766                 return flags;
6767         } else {
6768                 /* they already had raid on here, just return */
6769                 if (flags & stripped)
6770                         return flags;
6771
6772                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6773                 stripped = flags & ~stripped;
6774
6775                 /* switch duplicated blocks with raid1 */
6776                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6777                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6778
6779                 /* turn single device chunks into raid0 */
6780                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6781         }
6782         return flags;
6783 }
6784
6785 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6786                      struct btrfs_block_group_cache *shrink_block_group,
6787                      int force)
6788 {
6789         struct btrfs_trans_handle *trans;
6790         u64 new_alloc_flags;
6791         u64 calc;
6792
6793         spin_lock(&shrink_block_group->lock);
6794         if (btrfs_block_group_used(&shrink_block_group->item) +
6795             shrink_block_group->reserved > 0) {
6796                 spin_unlock(&shrink_block_group->lock);
6797
6798                 trans = btrfs_start_transaction(root, 1);
6799                 spin_lock(&shrink_block_group->lock);
6800
6801                 new_alloc_flags = update_block_group_flags(root,
6802                                                    shrink_block_group->flags);
6803                 if (new_alloc_flags != shrink_block_group->flags) {
6804                         calc =
6805                              btrfs_block_group_used(&shrink_block_group->item);
6806                 } else {
6807                         calc = shrink_block_group->key.offset;
6808                 }
6809                 spin_unlock(&shrink_block_group->lock);
6810
6811                 do_chunk_alloc(trans, root->fs_info->extent_root,
6812                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
6813
6814                 btrfs_end_transaction(trans, root);
6815         } else
6816                 spin_unlock(&shrink_block_group->lock);
6817         return 0;
6818 }
6819
6820
6821 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6822                                          struct btrfs_block_group_cache *group)
6823
6824 {
6825         __alloc_chunk_for_shrink(root, group, 1);
6826         set_block_group_readonly(group);
6827         return 0;
6828 }
6829
6830 #if 0
6831 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6832                                  struct btrfs_root *root,
6833                                  u64 objectid, u64 size)
6834 {
6835         struct btrfs_path *path;
6836         struct btrfs_inode_item *item;
6837         struct extent_buffer *leaf;
6838         int ret;
6839
6840         path = btrfs_alloc_path();
6841         if (!path)
6842                 return -ENOMEM;
6843
6844         path->leave_spinning = 1;
6845         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6846         if (ret)
6847                 goto out;
6848
6849         leaf = path->nodes[0];
6850         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6851         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6852         btrfs_set_inode_generation(leaf, item, 1);
6853         btrfs_set_inode_size(leaf, item, size);
6854         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6855         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6856         btrfs_mark_buffer_dirty(leaf);
6857         btrfs_release_path(root, path);
6858 out:
6859         btrfs_free_path(path);
6860         return ret;
6861 }
6862
6863 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6864                                         struct btrfs_block_group_cache *group)
6865 {
6866         struct inode *inode = NULL;
6867         struct btrfs_trans_handle *trans;
6868         struct btrfs_root *root;
6869         struct btrfs_key root_key;
6870         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6871         int err = 0;
6872
6873         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6874         root_key.type = BTRFS_ROOT_ITEM_KEY;
6875         root_key.offset = (u64)-1;
6876         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6877         if (IS_ERR(root))
6878                 return ERR_CAST(root);
6879
6880         trans = btrfs_start_transaction(root, 1);
6881         BUG_ON(!trans);
6882
6883         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6884         if (err)
6885                 goto out;
6886
6887         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6888         BUG_ON(err);
6889
6890         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6891                                        group->key.offset, 0, group->key.offset,
6892                                        0, 0, 0);
6893         BUG_ON(err);
6894
6895         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6896         if (inode->i_state & I_NEW) {
6897                 BTRFS_I(inode)->root = root;
6898                 BTRFS_I(inode)->location.objectid = objectid;
6899                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6900                 BTRFS_I(inode)->location.offset = 0;
6901                 btrfs_read_locked_inode(inode);
6902                 unlock_new_inode(inode);
6903                 BUG_ON(is_bad_inode(inode));
6904         } else {
6905                 BUG_ON(1);
6906         }
6907         BTRFS_I(inode)->index_cnt = group->key.objectid;
6908
6909         err = btrfs_orphan_add(trans, inode);
6910 out:
6911         btrfs_end_transaction(trans, root);
6912         if (err) {
6913                 if (inode)
6914                         iput(inode);
6915                 inode = ERR_PTR(err);
6916         }
6917         return inode;
6918 }
6919
6920 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6921 {
6922
6923         struct btrfs_ordered_sum *sums;
6924         struct btrfs_sector_sum *sector_sum;
6925         struct btrfs_ordered_extent *ordered;
6926         struct btrfs_root *root = BTRFS_I(inode)->root;
6927         struct list_head list;
6928         size_t offset;
6929         int ret;
6930         u64 disk_bytenr;
6931
6932         INIT_LIST_HEAD(&list);
6933
6934         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6935         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6936
6937         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6938         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6939                                        disk_bytenr + len - 1, &list);
6940
6941         while (!list_empty(&list)) {
6942                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6943                 list_del_init(&sums->list);
6944
6945                 sector_sum = sums->sums;
6946                 sums->bytenr = ordered->start;
6947
6948                 offset = 0;
6949                 while (offset < sums->len) {
6950                         sector_sum->bytenr += ordered->start - disk_bytenr;
6951                         sector_sum++;
6952                         offset += root->sectorsize;
6953                 }
6954
6955                 btrfs_add_ordered_sum(inode, ordered, sums);
6956         }
6957         btrfs_put_ordered_extent(ordered);
6958         return 0;
6959 }
6960
6961 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6962 {
6963         struct btrfs_trans_handle *trans;
6964         struct btrfs_path *path;
6965         struct btrfs_fs_info *info = root->fs_info;
6966         struct extent_buffer *leaf;
6967         struct inode *reloc_inode;
6968         struct btrfs_block_group_cache *block_group;
6969         struct btrfs_key key;
6970         u64 skipped;
6971         u64 cur_byte;
6972         u64 total_found;
6973         u32 nritems;
6974         int ret;
6975         int progress;
6976         int pass = 0;
6977
6978         root = root->fs_info->extent_root;
6979
6980         block_group = btrfs_lookup_block_group(info, group_start);
6981         BUG_ON(!block_group);
6982
6983         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6984                (unsigned long long)block_group->key.objectid,
6985                (unsigned long long)block_group->flags);
6986
6987         path = btrfs_alloc_path();
6988         BUG_ON(!path);
6989
6990         reloc_inode = create_reloc_inode(info, block_group);
6991         BUG_ON(IS_ERR(reloc_inode));
6992
6993         __alloc_chunk_for_shrink(root, block_group, 1);
6994         set_block_group_readonly(block_group);
6995
6996         btrfs_start_delalloc_inodes(info->tree_root);
6997         btrfs_wait_ordered_extents(info->tree_root, 0);
6998 again:
6999         skipped = 0;
7000         total_found = 0;
7001         progress = 0;
7002         key.objectid = block_group->key.objectid;
7003         key.offset = 0;
7004         key.type = 0;
7005         cur_byte = key.objectid;
7006
7007         trans = btrfs_start_transaction(info->tree_root, 1);
7008         btrfs_commit_transaction(trans, info->tree_root);
7009
7010         mutex_lock(&root->fs_info->cleaner_mutex);
7011         btrfs_clean_old_snapshots(info->tree_root);
7012         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
7013         mutex_unlock(&root->fs_info->cleaner_mutex);
7014
7015         trans = btrfs_start_transaction(info->tree_root, 1);
7016         btrfs_commit_transaction(trans, info->tree_root);
7017
7018         while (1) {
7019                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7020                 if (ret < 0)
7021                         goto out;
7022 next:
7023                 leaf = path->nodes[0];
7024                 nritems = btrfs_header_nritems(leaf);
7025                 if (path->slots[0] >= nritems) {
7026                         ret = btrfs_next_leaf(root, path);
7027                         if (ret < 0)
7028                                 goto out;
7029                         if (ret == 1) {
7030                                 ret = 0;
7031                                 break;
7032                         }
7033                         leaf = path->nodes[0];
7034                         nritems = btrfs_header_nritems(leaf);
7035                 }
7036
7037                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7038
7039                 if (key.objectid >= block_group->key.objectid +
7040                     block_group->key.offset)
7041                         break;
7042
7043                 if (progress && need_resched()) {
7044                         btrfs_release_path(root, path);
7045                         cond_resched();
7046                         progress = 0;
7047                         continue;
7048                 }
7049                 progress = 1;
7050
7051                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7052                     key.objectid + key.offset <= cur_byte) {
7053                         path->slots[0]++;
7054                         goto next;
7055                 }
7056
7057                 total_found++;
7058                 cur_byte = key.objectid + key.offset;
7059                 btrfs_release_path(root, path);
7060
7061                 __alloc_chunk_for_shrink(root, block_group, 0);
7062                 ret = relocate_one_extent(root, path, &key, block_group,
7063                                           reloc_inode, pass);
7064                 BUG_ON(ret < 0);
7065                 if (ret > 0)
7066                         skipped++;
7067
7068                 key.objectid = cur_byte;
7069                 key.type = 0;
7070                 key.offset = 0;
7071         }
7072
7073         btrfs_release_path(root, path);
7074
7075         if (pass == 0) {
7076                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7077                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7078         }
7079
7080         if (total_found > 0) {
7081                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7082                        (unsigned long long)total_found, pass);
7083                 pass++;
7084                 if (total_found == skipped && pass > 2) {
7085                         iput(reloc_inode);
7086                         reloc_inode = create_reloc_inode(info, block_group);
7087                         pass = 0;
7088                 }
7089                 goto again;
7090         }
7091
7092         /* delete reloc_inode */
7093         iput(reloc_inode);
7094
7095         /* unpin extents in this range */
7096         trans = btrfs_start_transaction(info->tree_root, 1);
7097         btrfs_commit_transaction(trans, info->tree_root);
7098
7099         spin_lock(&block_group->lock);
7100         WARN_ON(block_group->pinned > 0);
7101         WARN_ON(block_group->reserved > 0);
7102         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7103         spin_unlock(&block_group->lock);
7104         btrfs_put_block_group(block_group);
7105         ret = 0;
7106 out:
7107         btrfs_free_path(path);
7108         return ret;
7109 }
7110 #endif
7111
7112 static int find_first_block_group(struct btrfs_root *root,
7113                 struct btrfs_path *path, struct btrfs_key *key)
7114 {
7115         int ret = 0;
7116         struct btrfs_key found_key;
7117         struct extent_buffer *leaf;
7118         int slot;
7119
7120         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7121         if (ret < 0)
7122                 goto out;
7123
7124         while (1) {
7125                 slot = path->slots[0];
7126                 leaf = path->nodes[0];
7127                 if (slot >= btrfs_header_nritems(leaf)) {
7128                         ret = btrfs_next_leaf(root, path);
7129                         if (ret == 0)
7130                                 continue;
7131                         if (ret < 0)
7132                                 goto out;
7133                         break;
7134                 }
7135                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7136
7137                 if (found_key.objectid >= key->objectid &&
7138                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7139                         ret = 0;
7140                         goto out;
7141                 }
7142                 path->slots[0]++;
7143         }
7144         ret = -ENOENT;
7145 out:
7146         return ret;
7147 }
7148
7149 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7150 {
7151         struct btrfs_block_group_cache *block_group;
7152         struct btrfs_space_info *space_info;
7153         struct rb_node *n;
7154
7155         spin_lock(&info->block_group_cache_lock);
7156         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7157                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7158                                        cache_node);
7159                 rb_erase(&block_group->cache_node,
7160                          &info->block_group_cache_tree);
7161                 spin_unlock(&info->block_group_cache_lock);
7162
7163                 down_write(&block_group->space_info->groups_sem);
7164                 list_del(&block_group->list);
7165                 up_write(&block_group->space_info->groups_sem);
7166
7167                 if (block_group->cached == BTRFS_CACHE_STARTED)
7168                         wait_event(block_group->caching_q,
7169                                    block_group_cache_done(block_group));
7170
7171                 btrfs_remove_free_space_cache(block_group);
7172
7173                 WARN_ON(atomic_read(&block_group->count) != 1);
7174                 kfree(block_group);
7175
7176                 spin_lock(&info->block_group_cache_lock);
7177         }
7178         spin_unlock(&info->block_group_cache_lock);
7179
7180         /* now that all the block groups are freed, go through and
7181          * free all the space_info structs.  This is only called during
7182          * the final stages of unmount, and so we know nobody is
7183          * using them.  We call synchronize_rcu() once before we start,
7184          * just to be on the safe side.
7185          */
7186         synchronize_rcu();
7187
7188         while(!list_empty(&info->space_info)) {
7189                 space_info = list_entry(info->space_info.next,
7190                                         struct btrfs_space_info,
7191                                         list);
7192
7193                 list_del(&space_info->list);
7194                 kfree(space_info);
7195         }
7196         return 0;
7197 }
7198
7199 int btrfs_read_block_groups(struct btrfs_root *root)
7200 {
7201         struct btrfs_path *path;
7202         int ret;
7203         struct btrfs_block_group_cache *cache;
7204         struct btrfs_fs_info *info = root->fs_info;
7205         struct btrfs_space_info *space_info;
7206         struct btrfs_key key;
7207         struct btrfs_key found_key;
7208         struct extent_buffer *leaf;
7209
7210         root = info->extent_root;
7211         key.objectid = 0;
7212         key.offset = 0;
7213         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7214         path = btrfs_alloc_path();
7215         if (!path)
7216                 return -ENOMEM;
7217
7218         while (1) {
7219                 ret = find_first_block_group(root, path, &key);
7220                 if (ret > 0) {
7221                         ret = 0;
7222                         goto error;
7223                 }
7224                 if (ret != 0)
7225                         goto error;
7226
7227                 leaf = path->nodes[0];
7228                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7229                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7230                 if (!cache) {
7231                         ret = -ENOMEM;
7232                         break;
7233                 }
7234
7235                 atomic_set(&cache->count, 1);
7236                 spin_lock_init(&cache->lock);
7237                 spin_lock_init(&cache->tree_lock);
7238                 cache->fs_info = info;
7239                 init_waitqueue_head(&cache->caching_q);
7240                 INIT_LIST_HEAD(&cache->list);
7241                 INIT_LIST_HEAD(&cache->cluster_list);
7242
7243                 /*
7244                  * we only want to have 32k of ram per block group for keeping
7245                  * track of free space, and if we pass 1/2 of that we want to
7246                  * start converting things over to using bitmaps
7247                  */
7248                 cache->extents_thresh = ((1024 * 32) / 2) /
7249                         sizeof(struct btrfs_free_space);
7250
7251                 read_extent_buffer(leaf, &cache->item,
7252                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7253                                    sizeof(cache->item));
7254                 memcpy(&cache->key, &found_key, sizeof(found_key));
7255
7256                 key.objectid = found_key.objectid + found_key.offset;
7257                 btrfs_release_path(root, path);
7258                 cache->flags = btrfs_block_group_flags(&cache->item);
7259                 cache->sectorsize = root->sectorsize;
7260
7261                 remove_sb_from_cache(root, cache);
7262
7263                 /*
7264                  * check for two cases, either we are full, and therefore
7265                  * don't need to bother with the caching work since we won't
7266                  * find any space, or we are empty, and we can just add all
7267                  * the space in and be done with it.  This saves us _alot_ of
7268                  * time, particularly in the full case.
7269                  */
7270                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7271                         cache->cached = BTRFS_CACHE_FINISHED;
7272                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7273                         cache->cached = BTRFS_CACHE_FINISHED;
7274                         add_new_free_space(cache, root->fs_info,
7275                                            found_key.objectid,
7276                                            found_key.objectid +
7277                                            found_key.offset);
7278                 }
7279
7280                 ret = update_space_info(info, cache->flags, found_key.offset,
7281                                         btrfs_block_group_used(&cache->item),
7282                                         &space_info);
7283                 BUG_ON(ret);
7284                 cache->space_info = space_info;
7285                 down_write(&space_info->groups_sem);
7286                 list_add_tail(&cache->list, &space_info->block_groups);
7287                 up_write(&space_info->groups_sem);
7288
7289                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7290                 BUG_ON(ret);
7291
7292                 set_avail_alloc_bits(root->fs_info, cache->flags);
7293                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7294                         set_block_group_readonly(cache);
7295         }
7296         ret = 0;
7297 error:
7298         btrfs_free_path(path);
7299         return ret;
7300 }
7301
7302 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7303                            struct btrfs_root *root, u64 bytes_used,
7304                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7305                            u64 size)
7306 {
7307         int ret;
7308         struct btrfs_root *extent_root;
7309         struct btrfs_block_group_cache *cache;
7310
7311         extent_root = root->fs_info->extent_root;
7312
7313         root->fs_info->last_trans_log_full_commit = trans->transid;
7314
7315         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7316         if (!cache)
7317                 return -ENOMEM;
7318
7319         cache->key.objectid = chunk_offset;
7320         cache->key.offset = size;
7321         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7322         cache->sectorsize = root->sectorsize;
7323
7324         /*
7325          * we only want to have 32k of ram per block group for keeping track
7326          * of free space, and if we pass 1/2 of that we want to start
7327          * converting things over to using bitmaps
7328          */
7329         cache->extents_thresh = ((1024 * 32) / 2) /
7330                 sizeof(struct btrfs_free_space);
7331         atomic_set(&cache->count, 1);
7332         spin_lock_init(&cache->lock);
7333         spin_lock_init(&cache->tree_lock);
7334         init_waitqueue_head(&cache->caching_q);
7335         INIT_LIST_HEAD(&cache->list);
7336         INIT_LIST_HEAD(&cache->cluster_list);
7337
7338         btrfs_set_block_group_used(&cache->item, bytes_used);
7339         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7340         cache->flags = type;
7341         btrfs_set_block_group_flags(&cache->item, type);
7342
7343         cache->cached = BTRFS_CACHE_FINISHED;
7344         remove_sb_from_cache(root, cache);
7345
7346         add_new_free_space(cache, root->fs_info, chunk_offset,
7347                            chunk_offset + size);
7348
7349         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7350                                 &cache->space_info);
7351         BUG_ON(ret);
7352         down_write(&cache->space_info->groups_sem);
7353         list_add_tail(&cache->list, &cache->space_info->block_groups);
7354         up_write(&cache->space_info->groups_sem);
7355
7356         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7357         BUG_ON(ret);
7358
7359         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7360                                 sizeof(cache->item));
7361         BUG_ON(ret);
7362
7363         set_avail_alloc_bits(extent_root->fs_info, type);
7364
7365         return 0;
7366 }
7367
7368 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7369                              struct btrfs_root *root, u64 group_start)
7370 {
7371         struct btrfs_path *path;
7372         struct btrfs_block_group_cache *block_group;
7373         struct btrfs_free_cluster *cluster;
7374         struct btrfs_key key;
7375         int ret;
7376
7377         root = root->fs_info->extent_root;
7378
7379         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7380         BUG_ON(!block_group);
7381         BUG_ON(!block_group->ro);
7382
7383         memcpy(&key, &block_group->key, sizeof(key));
7384
7385         /* make sure this block group isn't part of an allocation cluster */
7386         cluster = &root->fs_info->data_alloc_cluster;
7387         spin_lock(&cluster->refill_lock);
7388         btrfs_return_cluster_to_free_space(block_group, cluster);
7389         spin_unlock(&cluster->refill_lock);
7390
7391         /*
7392          * make sure this block group isn't part of a metadata
7393          * allocation cluster
7394          */
7395         cluster = &root->fs_info->meta_alloc_cluster;
7396         spin_lock(&cluster->refill_lock);
7397         btrfs_return_cluster_to_free_space(block_group, cluster);
7398         spin_unlock(&cluster->refill_lock);
7399
7400         path = btrfs_alloc_path();
7401         BUG_ON(!path);
7402
7403         spin_lock(&root->fs_info->block_group_cache_lock);
7404         rb_erase(&block_group->cache_node,
7405                  &root->fs_info->block_group_cache_tree);
7406         spin_unlock(&root->fs_info->block_group_cache_lock);
7407
7408         down_write(&block_group->space_info->groups_sem);
7409         /*
7410          * we must use list_del_init so people can check to see if they
7411          * are still on the list after taking the semaphore
7412          */
7413         list_del_init(&block_group->list);
7414         up_write(&block_group->space_info->groups_sem);
7415
7416         if (block_group->cached == BTRFS_CACHE_STARTED)
7417                 wait_event(block_group->caching_q,
7418                            block_group_cache_done(block_group));
7419
7420         btrfs_remove_free_space_cache(block_group);
7421
7422         spin_lock(&block_group->space_info->lock);
7423         block_group->space_info->total_bytes -= block_group->key.offset;
7424         block_group->space_info->bytes_readonly -= block_group->key.offset;
7425         spin_unlock(&block_group->space_info->lock);
7426
7427         btrfs_clear_space_info_full(root->fs_info);
7428
7429         btrfs_put_block_group(block_group);
7430         btrfs_put_block_group(block_group);
7431
7432         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7433         if (ret > 0)
7434                 ret = -EIO;
7435         if (ret < 0)
7436                 goto out;
7437
7438         ret = btrfs_del_item(trans, root, path);
7439 out:
7440         btrfs_free_path(path);
7441         return ret;
7442 }