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