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 <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc,
  39                               int mark_free);
  40 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
  41                                    u64 num_bytes, int reserve);
  42 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  43                                 struct btrfs_root *root,
  44                                 u64 bytenr, u64 num_bytes, u64 parent,
  45                                 u64 root_objectid, u64 owner_objectid,
  46                                 u64 owner_offset, int refs_to_drop,
  47                                 struct btrfs_delayed_extent_op *extra_op);
  48 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  49                                     struct extent_buffer *leaf,
  50                                     struct btrfs_extent_item *ei);
  51 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  52                                       struct btrfs_root *root,
  53                                       u64 parent, u64 root_objectid,
  54                                       u64 flags, u64 owner, u64 offset,
  55                                       struct btrfs_key *ins, int ref_mod);
  56 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  57                                      struct btrfs_root *root,
  58                                      u64 parent, u64 root_objectid,
  59                                      u64 flags, struct btrfs_disk_key *key,
  60                                      int level, struct btrfs_key *ins);
  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 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  65                           struct btrfs_root *root,
  66                           struct btrfs_path *path,
  67                           u64 bytenr, u64 num_bytes,
  68                           int is_data, int reserved,
  69                           struct extent_buffer **must_clean);
  70 static int find_next_key(struct btrfs_path *path, int level,
  71                          struct btrfs_key *key);
  72 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  73                             int dump_block_groups);
  74 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
  75                                 struct btrfs_root *root,
  76                                 struct btrfs_space_info *sinfo, u64 num_bytes);
  77 static int shrink_delalloc(struct btrfs_trans_handle *trans,
  78                            struct btrfs_root *root,
  79                            struct btrfs_space_info *sinfo, u64 to_reclaim);
  80
  81 static noinline int
  82 block_group_cache_done(struct btrfs_block_group_cache *cache)
  83 {
  84         smp_mb();
  85         return cache->cached == BTRFS_CACHE_FINISHED;
  86 }
  87
  88 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  89 {
  90         return (cache->flags & bits) == bits;
  91 }
  92
  93 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  94 {
  95         atomic_inc(&cache->count);
  96 }
  97
  98 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  99 {
 100         if (atomic_dec_and_test(&cache->count))
 101                 kfree(cache);
 102 }
 103
 104 /*
 105  * this adds the block group to the fs_info rb tree for the block group
 106  * cache
 107  */
 108 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 109                                 struct btrfs_block_group_cache *block_group)
 110 {
 111         struct rb_node **p;
 112         struct rb_node *parent = NULL;
 113         struct btrfs_block_group_cache *cache;
 114
 115         spin_lock(&info->block_group_cache_lock);
 116         p = &info->block_group_cache_tree.rb_node;
 117
 118         while (*p) {
 119                 parent = *p;
 120                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 121                                  cache_node);
 122                 if (block_group->key.objectid < cache->key.objectid) {
 123                         p = &(*p)->rb_left;
 124                 } else if (block_group->key.objectid > cache->key.objectid) {
 125                         p = &(*p)->rb_right;
 126                 } else {
 127                         spin_unlock(&info->block_group_cache_lock);
 128                         return -EEXIST;
 129                 }
 130         }
 131
 132         rb_link_node(&block_group->cache_node, parent, p);
 133         rb_insert_color(&block_group->cache_node,
 134                         &info->block_group_cache_tree);
 135         spin_unlock(&info->block_group_cache_lock);
 136
 137         return 0;
 138 }
 139
 140 /*
 141  * This will return the block group at or after bytenr if contains is 0, else
 142  * it will return the block group that contains the bytenr
 143  */
 144 static struct btrfs_block_group_cache *
 145 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 146                               int contains)
 147 {
 148         struct btrfs_block_group_cache *cache, *ret = NULL;
 149         struct rb_node *n;
 150         u64 end, start;
 151
 152         spin_lock(&info->block_group_cache_lock);
 153         n = info->block_group_cache_tree.rb_node;
 154
 155         while (n) {
 156                 cache = rb_entry(n, struct btrfs_block_group_cache,
 157                                  cache_node);
 158                 end = cache->key.objectid + cache->key.offset - 1;
 159                 start = cache->key.objectid;
 160
 161                 if (bytenr < start) {
 162                         if (!contains && (!ret || start < ret->key.objectid))
 163                                 ret = cache;
 164                         n = n->rb_left;
 165                 } else if (bytenr > start) {
 166                         if (contains && bytenr <= end) {
 167                                 ret = cache;
 168                                 break;
 169                         }
 170                         n = n->rb_right;
 171                 } else {
 172                         ret = cache;
 173                         break;
 174                 }
 175         }
 176         if (ret)
 177                 btrfs_get_block_group(ret);
 178         spin_unlock(&info->block_group_cache_lock);
 179
 180         return ret;
 181 }
 182
 183 static int add_excluded_extent(struct btrfs_root *root,
 184                                u64 start, u64 num_bytes)
 185 {
 186         u64 end = start + num_bytes - 1;
 187         set_extent_bits(&root->fs_info->freed_extents[0],
 188                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 189         set_extent_bits(&root->fs_info->freed_extents[1],
 190                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 191         return 0;
 192 }
 193
 194 static void free_excluded_extents(struct btrfs_root *root,
 195                                   struct btrfs_block_group_cache *cache)
 196 {
 197         u64 start, end;
 198
 199         start = cache->key.objectid;
 200         end = start + cache->key.offset - 1;
 201
 202         clear_extent_bits(&root->fs_info->freed_extents[0],
 203                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 204         clear_extent_bits(&root->fs_info->freed_extents[1],
 205                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 206 }
 207
 208 static int exclude_super_stripes(struct btrfs_root *root,
 209                                  struct btrfs_block_group_cache *cache)
 210 {
 211         u64 bytenr;
 212         u64 *logical;
 213         int stripe_len;
 214         int i, nr, ret;
 215
 216         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 217                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 218                 cache->bytes_super += stripe_len;
 219                 ret = add_excluded_extent(root, cache->key.objectid,
 220                                           stripe_len);
 221                 BUG_ON(ret);
 222         }
 223
 224         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 225                 bytenr = btrfs_sb_offset(i);
 226                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 227                                        cache->key.objectid, bytenr,
 228                                        0, &logical, &nr, &stripe_len);
 229                 BUG_ON(ret);
 230
 231                 while (nr--) {
 232                         cache->bytes_super += stripe_len;
 233                         ret = add_excluded_extent(root, logical[nr],
 234                                                   stripe_len);
 235                         BUG_ON(ret);
 236                 }
 237
 238                 kfree(logical);
 239         }
 240         return 0;
 241 }
 242
 243 static struct btrfs_caching_control *
 244 get_caching_control(struct btrfs_block_group_cache *cache)
 245 {
 246         struct btrfs_caching_control *ctl;
 247
 248         spin_lock(&cache->lock);
 249         if (cache->cached != BTRFS_CACHE_STARTED) {
 250                 spin_unlock(&cache->lock);
 251                 return NULL;
 252         }
 253
 254         ctl = cache->caching_ctl;
 255         atomic_inc(&ctl->count);
 256         spin_unlock(&cache->lock);
 257         return ctl;
 258 }
 259
 260 static void put_caching_control(struct btrfs_caching_control *ctl)
 261 {
 262         if (atomic_dec_and_test(&ctl->count))
 263                 kfree(ctl);
 264 }
 265
 266 /*
 267  * this is only called by cache_block_group, since we could have freed extents
 268  * we need to check the pinned_extents for any extents that can't be used yet
 269  * since their free space will be released as soon as the transaction commits.
 270  */
 271 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 272                               struct btrfs_fs_info *info, u64 start, u64 end)
 273 {
 274         u64 extent_start, extent_end, size, total_added = 0;
 275         int ret;
 276
 277         while (start < end) {
 278                 ret = find_first_extent_bit(info->pinned_extents, start,
 279                                             &extent_start, &extent_end,
 280                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 281                 if (ret)
 282                         break;
 283
 284                 if (extent_start <= start) {
 285                         start = extent_end + 1;
 286                 } else if (extent_start > start && extent_start < end) {
 287                         size = extent_start - start;
 288                         total_added += size;
 289                         ret = btrfs_add_free_space(block_group, start,
 290                                                    size);
 291                         BUG_ON(ret);
 292                         start = extent_end + 1;
 293                 } else {
 294                         break;
 295                 }
 296         }
 297
 298         if (start < end) {
 299                 size = end - start;
 300                 total_added += size;
 301                 ret = btrfs_add_free_space(block_group, start, size);
 302                 BUG_ON(ret);
 303         }
 304
 305         return total_added;
 306 }
 307
 308 static int caching_kthread(void *data)
 309 {
 310         struct btrfs_block_group_cache *block_group = data;
 311         struct btrfs_fs_info *fs_info = block_group->fs_info;
 312         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 313         struct btrfs_root *extent_root = fs_info->extent_root;
 314         struct btrfs_path *path;
 315         struct extent_buffer *leaf;
 316         struct btrfs_key key;
 317         u64 total_found = 0;
 318         u64 last = 0;
 319         u32 nritems;
 320         int ret = 0;
 321
 322         path = btrfs_alloc_path();
 323         if (!path)
 324                 return -ENOMEM;
 325
 326         exclude_super_stripes(extent_root, block_group);
 327         spin_lock(&block_group->space_info->lock);
 328         block_group->space_info->bytes_super += block_group->bytes_super;
 329         spin_unlock(&block_group->space_info->lock);
 330
 331         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 332
 333         /*
 334          * We don't want to deadlock with somebody trying to allocate a new
 335          * extent for the extent root while also trying to search the extent
 336          * root to add free space.  So we skip locking and search the commit
 337          * root, since its read-only
 338          */
 339         path->skip_locking = 1;
 340         path->search_commit_root = 1;
 341         path->reada = 2;
 342
 343         key.objectid = last;
 344         key.offset = 0;
 345         key.type = BTRFS_EXTENT_ITEM_KEY;
 346 again:
 347         mutex_lock(&caching_ctl->mutex);
 348         /* need to make sure the commit_root doesn't disappear */
 349         down_read(&fs_info->extent_commit_sem);
 350
 351         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 352         if (ret < 0)
 353                 goto err;
 354
 355         leaf = path->nodes[0];
 356         nritems = btrfs_header_nritems(leaf);
 357
 358         while (1) {
 359                 smp_mb();
 360                 if (fs_info->closing > 1) {
 361                         last = (u64)-1;
 362                         break;
 363                 }
 364
 365                 if (path->slots[0] < nritems) {
 366                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 367                 } else {
 368                         ret = find_next_key(path, 0, &key);
 369                         if (ret)
 370                                 break;
 371
 372                         caching_ctl->progress = last;
 373                         btrfs_release_path(extent_root, path);
 374                         up_read(&fs_info->extent_commit_sem);
 375                         mutex_unlock(&caching_ctl->mutex);
 376                         if (btrfs_transaction_in_commit(fs_info))
 377                                 schedule_timeout(1);
 378                         else
 379                                 cond_resched();
 380                         goto again;
 381                 }
 382
 383                 if (key.objectid < block_group->key.objectid) {
 384                         path->slots[0]++;
 385                         continue;
 386                 }
 387
 388                 if (key.objectid >= block_group->key.objectid +
 389                     block_group->key.offset)
 390                         break;
 391
 392                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 393                         total_found += add_new_free_space(block_group,
 394                                                           fs_info, last,
 395                                                           key.objectid);
 396                         last = key.objectid + key.offset;
 397
 398                         if (total_found > (1024 * 1024 * 2)) {
 399                                 total_found = 0;
 400                                 wake_up(&caching_ctl->wait);
 401                         }
 402                 }
 403                 path->slots[0]++;
 404         }
 405         ret = 0;
 406
 407         total_found += add_new_free_space(block_group, fs_info, last,
 408                                           block_group->key.objectid +
 409                                           block_group->key.offset);
 410         caching_ctl->progress = (u64)-1;
 411
 412         spin_lock(&block_group->lock);
 413         block_group->caching_ctl = NULL;
 414         block_group->cached = BTRFS_CACHE_FINISHED;
 415         spin_unlock(&block_group->lock);
 416
 417 err:
 418         btrfs_free_path(path);
 419         up_read(&fs_info->extent_commit_sem);
 420
 421         free_excluded_extents(extent_root, block_group);
 422
 423         mutex_unlock(&caching_ctl->mutex);
 424         wake_up(&caching_ctl->wait);
 425
 426         put_caching_control(caching_ctl);
 427         atomic_dec(&block_group->space_info->caching_threads);
 428         btrfs_put_block_group(block_group);
 429
 430         return 0;
 431 }
 432
 433 static int cache_block_group(struct btrfs_block_group_cache *cache)
 434 {
 435         struct btrfs_fs_info *fs_info = cache->fs_info;
 436         struct btrfs_caching_control *caching_ctl;
 437         struct task_struct *tsk;
 438         int ret = 0;
 439
 440         smp_mb();
 441         if (cache->cached != BTRFS_CACHE_NO)
 442                 return 0;
 443
 444         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 445         BUG_ON(!caching_ctl);
 446
 447         INIT_LIST_HEAD(&caching_ctl->list);
 448         mutex_init(&caching_ctl->mutex);
 449         init_waitqueue_head(&caching_ctl->wait);
 450         caching_ctl->block_group = cache;
 451         caching_ctl->progress = cache->key.objectid;
 452         /* one for caching kthread, one for caching block group list */
 453         atomic_set(&caching_ctl->count, 2);
 454
 455         spin_lock(&cache->lock);
 456         if (cache->cached != BTRFS_CACHE_NO) {
 457                 spin_unlock(&cache->lock);
 458                 kfree(caching_ctl);
 459                 return 0;
 460         }
 461         cache->caching_ctl = caching_ctl;
 462         cache->cached = BTRFS_CACHE_STARTED;
 463         spin_unlock(&cache->lock);
 464
 465         down_write(&fs_info->extent_commit_sem);
 466         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 467         up_write(&fs_info->extent_commit_sem);
 468
 469         atomic_inc(&cache->space_info->caching_threads);
 470         btrfs_get_block_group(cache);
 471
 472         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 473                           cache->key.objectid);
 474         if (IS_ERR(tsk)) {
 475                 ret = PTR_ERR(tsk);
 476                 printk(KERN_ERR "error running thread %d\n", ret);
 477                 BUG();
 478         }
 479
 480         return ret;
 481 }
 482
 483 /*
 484  * return the block group that starts at or after bytenr
 485  */
 486 static struct btrfs_block_group_cache *
 487 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 488 {
 489         struct btrfs_block_group_cache *cache;
 490
 491         cache = block_group_cache_tree_search(info, bytenr, 0);
 492
 493         return cache;
 494 }
 495
 496 /*
 497  * return the block group that contains the given bytenr
 498  */
 499 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 500                                                  struct btrfs_fs_info *info,
 501                                                  u64 bytenr)
 502 {
 503         struct btrfs_block_group_cache *cache;
 504
 505         cache = block_group_cache_tree_search(info, bytenr, 1);
 506
 507         return cache;
 508 }
 509
 510 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 511                                                   u64 flags)
 512 {
 513         struct list_head *head = &info->space_info;
 514         struct btrfs_space_info *found;
 515
 516         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 517                  BTRFS_BLOCK_GROUP_METADATA;
 518
 519         rcu_read_lock();
 520         list_for_each_entry_rcu(found, head, list) {
 521                 if (found->flags == flags) {
 522                         rcu_read_unlock();
 523                         return found;
 524                 }
 525         }
 526         rcu_read_unlock();
 527         return NULL;
 528 }
 529
 530 /*
 531  * after adding space to the filesystem, we need to clear the full flags
 532  * on all the space infos.
 533  */
 534 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 535 {
 536         struct list_head *head = &info->space_info;
 537         struct btrfs_space_info *found;
 538
 539         rcu_read_lock();
 540         list_for_each_entry_rcu(found, head, list)
 541                 found->full = 0;
 542         rcu_read_unlock();
 543 }
 544
 545 static u64 div_factor(u64 num, int factor)
 546 {
 547         if (factor == 10)
 548                 return num;
 549         num *= factor;
 550         do_div(num, 10);
 551         return num;
 552 }
 553
 554 u64 btrfs_find_block_group(struct btrfs_root *root,
 555                            u64 search_start, u64 search_hint, int owner)
 556 {
 557         struct btrfs_block_group_cache *cache;
 558         u64 used;
 559         u64 last = max(search_hint, search_start);
 560         u64 group_start = 0;
 561         int full_search = 0;
 562         int factor = 9;
 563         int wrapped = 0;
 564 again:
 565         while (1) {
 566                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 567                 if (!cache)
 568                         break;
 569
 570                 spin_lock(&cache->lock);
 571                 last = cache->key.objectid + cache->key.offset;
 572                 used = btrfs_block_group_used(&cache->item);
 573
 574                 if ((full_search || !cache->ro) &&
 575                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 576                         if (used + cache->pinned + cache->reserved <
 577                             div_factor(cache->key.offset, factor)) {
 578                                 group_start = cache->key.objectid;
 579                                 spin_unlock(&cache->lock);
 580                                 btrfs_put_block_group(cache);
 581                                 goto found;
 582                         }
 583                 }
 584                 spin_unlock(&cache->lock);
 585                 btrfs_put_block_group(cache);
 586                 cond_resched();
 587         }
 588         if (!wrapped) {
 589                 last = search_start;
 590                 wrapped = 1;
 591                 goto again;
 592         }
 593         if (!full_search && factor < 10) {
 594                 last = search_start;
 595                 full_search = 1;
 596                 factor = 10;
 597                 goto again;
 598         }
 599 found:
 600         return group_start;
 601 }
 602
 603 /* simple helper to search for an existing extent at a given offset */
 604 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 605 {
 606         int ret;
 607         struct btrfs_key key;
 608         struct btrfs_path *path;
 609
 610         path = btrfs_alloc_path();
 611         BUG_ON(!path);
 612         key.objectid = start;
 613         key.offset = len;
 614         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 615         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 616                                 0, 0);
 617         btrfs_free_path(path);
 618         return ret;
 619 }
 620
 621 /*
 622  * Back reference rules.  Back refs have three main goals:
 623  *
 624  * 1) differentiate between all holders of references to an extent so that
 625  *    when a reference is dropped we can make sure it was a valid reference
 626  *    before freeing the extent.
 627  *
 628  * 2) Provide enough information to quickly find the holders of an extent
 629  *    if we notice a given block is corrupted or bad.
 630  *
 631  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 632  *    maintenance.  This is actually the same as #2, but with a slightly
 633  *    different use case.
 634  *
 635  * There are two kinds of back refs. The implicit back refs is optimized
 636  * for pointers in non-shared tree blocks. For a given pointer in a block,
 637  * back refs of this kind provide information about the block's owner tree
 638  * and the pointer's key. These information allow us to find the block by
 639  * b-tree searching. The full back refs is for pointers in tree blocks not
 640  * referenced by their owner trees. The location of tree block is recorded
 641  * in the back refs. Actually the full back refs is generic, and can be
 642  * used in all cases the implicit back refs is used. The major shortcoming
 643  * of the full back refs is its overhead. Every time a tree block gets
 644  * COWed, we have to update back refs entry for all pointers in it.
 645  *
 646  * For a newly allocated tree block, we use implicit back refs for
 647  * pointers in it. This means most tree related operations only involve
 648  * implicit back refs. For a tree block created in old transaction, the
 649  * only way to drop a reference to it is COW it. So we can detect the
 650  * event that tree block loses its owner tree's reference and do the
 651  * back refs conversion.
 652  *
 653  * When a tree block is COW'd through a tree, there are four cases:
 654  *
 655  * The reference count of the block is one and the tree is the block's
 656  * owner tree. Nothing to do in this case.
 657  *
 658  * The reference count of the block is one and the tree is not the
 659  * block's owner tree. In this case, full back refs is used for pointers
 660  * in the block. Remove these full back refs, add implicit back refs for
 661  * every pointers in the new block.
 662  *
 663  * The reference count of the block is greater than one and the tree is
 664  * the block's owner tree. In this case, implicit back refs is used for
 665  * pointers in the block. Add full back refs for every pointers in the
 666  * block, increase lower level extents' reference counts. The original
 667  * implicit back refs are entailed to the new block.
 668  *
 669  * The reference count of the block is greater than one and the tree is
 670  * not the block's owner tree. Add implicit back refs for every pointer in
 671  * the new block, increase lower level extents' reference count.
 672  *
 673  * Back Reference Key composing:
 674  *
 675  * The key objectid corresponds to the first byte in the extent,
 676  * The key type is used to differentiate between types of back refs.
 677  * There are different meanings of the key offset for different types
 678  * of back refs.
 679  *
 680  * File extents can be referenced by:
 681  *
 682  * - multiple snapshots, subvolumes, or different generations in one subvol
 683  * - different files inside a single subvolume
 684  * - different offsets inside a file (bookend extents in file.c)
 685  *
 686  * The extent ref structure for the implicit back refs has fields for:
 687  *
 688  * - Objectid of the subvolume root
 689  * - objectid of the file holding the reference
 690  * - original offset in the file
 691  * - how many bookend extents
 692  *
 693  * The key offset for the implicit back refs is hash of the first
 694  * three fields.
 695  *
 696  * The extent ref structure for the full back refs has field for:
 697  *
 698  * - number of pointers in the tree leaf
 699  *
 700  * The key offset for the implicit back refs is the first byte of
 701  * the tree leaf
 702  *
 703  * When a file extent is allocated, The implicit back refs is used.
 704  * the fields are filled in:
 705  *
 706  *     (root_key.objectid, inode objectid, offset in file, 1)
 707  *
 708  * When a file extent is removed file truncation, we find the
 709  * corresponding implicit back refs and check the following fields:
 710  *
 711  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 712  *
 713  * Btree extents can be referenced by:
 714  *
 715  * - Different subvolumes
 716  *
 717  * Both the implicit back refs and the full back refs for tree blocks
 718  * only consist of key. The key offset for the implicit back refs is
 719  * objectid of block's owner tree. The key offset for the full back refs
 720  * is the first byte of parent block.
 721  *
 722  * When implicit back refs is used, information about the lowest key and
 723  * level of the tree block are required. These information are stored in
 724  * tree block info structure.
 725  */
 726
 727 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 728 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 729                                   struct btrfs_root *root,
 730                                   struct btrfs_path *path,
 731                                   u64 owner, u32 extra_size)
 732 {
 733         struct btrfs_extent_item *item;
 734         struct btrfs_extent_item_v0 *ei0;
 735         struct btrfs_extent_ref_v0 *ref0;
 736         struct btrfs_tree_block_info *bi;
 737         struct extent_buffer *leaf;
 738         struct btrfs_key key;
 739         struct btrfs_key found_key;
 740         u32 new_size = sizeof(*item);
 741         u64 refs;
 742         int ret;
 743
 744         leaf = path->nodes[0];
 745         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 746
 747         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 748         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 749                              struct btrfs_extent_item_v0);
 750         refs = btrfs_extent_refs_v0(leaf, ei0);
 751
 752         if (owner == (u64)-1) {
 753                 while (1) {
 754                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 755                                 ret = btrfs_next_leaf(root, path);
 756                                 if (ret < 0)
 757                                         return ret;
 758                                 BUG_ON(ret > 0);
 759                                 leaf = path->nodes[0];
 760                         }
 761                         btrfs_item_key_to_cpu(leaf, &found_key,
 762                                               path->slots[0]);
 763                         BUG_ON(key.objectid != found_key.objectid);
 764                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 765                                 path->slots[0]++;
 766                                 continue;
 767                         }
 768                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 769                                               struct btrfs_extent_ref_v0);
 770                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 771                         break;
 772                 }
 773         }
 774         btrfs_release_path(root, path);
 775
 776         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 777                 new_size += sizeof(*bi);
 778
 779         new_size -= sizeof(*ei0);
 780         ret = btrfs_search_slot(trans, root, &key, path,
 781                                 new_size + extra_size, 1);
 782         if (ret < 0)
 783                 return ret;
 784         BUG_ON(ret);
 785
 786         ret = btrfs_extend_item(trans, root, path, new_size);
 787         BUG_ON(ret);
 788
 789         leaf = path->nodes[0];
 790         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 791         btrfs_set_extent_refs(leaf, item, refs);
 792         /* FIXME: get real generation */
 793         btrfs_set_extent_generation(leaf, item, 0);
 794         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 795                 btrfs_set_extent_flags(leaf, item,
 796                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 797                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 798                 bi = (struct btrfs_tree_block_info *)(item + 1);
 799                 /* FIXME: get first key of the block */
 800                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 801                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 802         } else {
 803                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 804         }
 805         btrfs_mark_buffer_dirty(leaf);
 806         return 0;
 807 }
 808 #endif
 809
 810 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 811 {
 812         u32 high_crc = ~(u32)0;
 813         u32 low_crc = ~(u32)0;
 814         __le64 lenum;
 815
 816         lenum = cpu_to_le64(root_objectid);
 817         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 818         lenum = cpu_to_le64(owner);
 819         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 820         lenum = cpu_to_le64(offset);
 821         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 822
 823         return ((u64)high_crc << 31) ^ (u64)low_crc;
 824 }
 825
 826 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 827                                      struct btrfs_extent_data_ref *ref)
 828 {
 829         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 830                                     btrfs_extent_data_ref_objectid(leaf, ref),
 831                                     btrfs_extent_data_ref_offset(leaf, ref));
 832 }
 833
 834 static int match_extent_data_ref(struct extent_buffer *leaf,
 835                                  struct btrfs_extent_data_ref *ref,
 836                                  u64 root_objectid, u64 owner, u64 offset)
 837 {
 838         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 839             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 840             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 841                 return 0;
 842         return 1;
 843 }
 844
 845 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 846                                            struct btrfs_root *root,
 847                                            struct btrfs_path *path,
 848                                            u64 bytenr, u64 parent,
 849                                            u64 root_objectid,
 850                                            u64 owner, u64 offset)
 851 {
 852         struct btrfs_key key;
 853         struct btrfs_extent_data_ref *ref;
 854         struct extent_buffer *leaf;
 855         u32 nritems;
 856         int ret;
 857         int recow;
 858         int err = -ENOENT;
 859
 860         key.objectid = bytenr;
 861         if (parent) {
 862                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 863                 key.offset = parent;
 864         } else {
 865                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 866                 key.offset = hash_extent_data_ref(root_objectid,
 867                                                   owner, offset);
 868         }
 869 again:
 870         recow = 0;
 871         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 872         if (ret < 0) {
 873                 err = ret;
 874                 goto fail;
 875         }
 876
 877         if (parent) {
 878                 if (!ret)
 879                         return 0;
 880 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 881                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 882                 btrfs_release_path(root, path);
 883                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 884                 if (ret < 0) {
 885                         err = ret;
 886                         goto fail;
 887                 }
 888                 if (!ret)
 889                         return 0;
 890 #endif
 891                 goto fail;
 892         }
 893
 894         leaf = path->nodes[0];
 895         nritems = btrfs_header_nritems(leaf);
 896         while (1) {
 897                 if (path->slots[0] >= nritems) {
 898                         ret = btrfs_next_leaf(root, path);
 899                         if (ret < 0)
 900                                 err = ret;
 901                         if (ret)
 902                                 goto fail;
 903
 904                         leaf = path->nodes[0];
 905                         nritems = btrfs_header_nritems(leaf);
 906                         recow = 1;
 907                 }
 908
 909                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 910                 if (key.objectid != bytenr ||
 911                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 912                         goto fail;
 913
 914                 ref = btrfs_item_ptr(leaf, path->slots[0],
 915                                      struct btrfs_extent_data_ref);
 916
 917                 if (match_extent_data_ref(leaf, ref, root_objectid,
 918                                           owner, offset)) {
 919                         if (recow) {
 920                                 btrfs_release_path(root, path);
 921                                 goto again;
 922                         }
 923                         err = 0;
 924                         break;
 925                 }
 926                 path->slots[0]++;
 927         }
 928 fail:
 929         return err;
 930 }
 931
 932 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 933                                            struct btrfs_root *root,
 934                                            struct btrfs_path *path,
 935                                            u64 bytenr, u64 parent,
 936                                            u64 root_objectid, u64 owner,
 937                                            u64 offset, int refs_to_add)
 938 {
 939         struct btrfs_key key;
 940         struct extent_buffer *leaf;
 941         u32 size;
 942         u32 num_refs;
 943         int ret;
 944
 945         key.objectid = bytenr;
 946         if (parent) {
 947                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 948                 key.offset = parent;
 949                 size = sizeof(struct btrfs_shared_data_ref);
 950         } else {
 951                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 952                 key.offset = hash_extent_data_ref(root_objectid,
 953                                                   owner, offset);
 954                 size = sizeof(struct btrfs_extent_data_ref);
 955         }
 956
 957         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 958         if (ret && ret != -EEXIST)
 959                 goto fail;
 960
 961         leaf = path->nodes[0];
 962         if (parent) {
 963                 struct btrfs_shared_data_ref *ref;
 964                 ref = btrfs_item_ptr(leaf, path->slots[0],
 965                                      struct btrfs_shared_data_ref);
 966                 if (ret == 0) {
 967                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 968                 } else {
 969                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 970                         num_refs += refs_to_add;
 971                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 972                 }
 973         } else {
 974                 struct btrfs_extent_data_ref *ref;
 975                 while (ret == -EEXIST) {
 976                         ref = btrfs_item_ptr(leaf, path->slots[0],
 977                                              struct btrfs_extent_data_ref);
 978                         if (match_extent_data_ref(leaf, ref, root_objectid,
 979                                                   owner, offset))
 980                                 break;
 981                         btrfs_release_path(root, path);
 982                         key.offset++;
 983                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 984                                                       size);
 985                         if (ret && ret != -EEXIST)
 986                                 goto fail;
 987
 988                         leaf = path->nodes[0];
 989                 }
 990                 ref = btrfs_item_ptr(leaf, path->slots[0],
 991                                      struct btrfs_extent_data_ref);
 992                 if (ret == 0) {
 993                         btrfs_set_extent_data_ref_root(leaf, ref,
 994                                                        root_objectid);
 995                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 996                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 997                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 998                 } else {
 999                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1000                         num_refs += refs_to_add;
1001                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1002                 }
1003         }
1004         btrfs_mark_buffer_dirty(leaf);
1005         ret = 0;
1006 fail:
1007         btrfs_release_path(root, path);
1008         return ret;
1009 }
1010
1011 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1012                                            struct btrfs_root *root,
1013                                            struct btrfs_path *path,
1014                                            int refs_to_drop)
1015 {
1016         struct btrfs_key key;
1017         struct btrfs_extent_data_ref *ref1 = NULL;
1018         struct btrfs_shared_data_ref *ref2 = NULL;
1019         struct extent_buffer *leaf;
1020         u32 num_refs = 0;
1021         int ret = 0;
1022
1023         leaf = path->nodes[0];
1024         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1025
1026         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1027                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1028                                       struct btrfs_extent_data_ref);
1029                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1030         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1031                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1032                                       struct btrfs_shared_data_ref);
1033                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1034 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1035         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1036                 struct btrfs_extent_ref_v0 *ref0;
1037                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1038                                       struct btrfs_extent_ref_v0);
1039                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1040 #endif
1041         } else {
1042                 BUG();
1043         }
1044
1045         BUG_ON(num_refs < refs_to_drop);
1046         num_refs -= refs_to_drop;
1047
1048         if (num_refs == 0) {
1049                 ret = btrfs_del_item(trans, root, path);
1050         } else {
1051                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1052                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1053                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1054                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1055 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1056                 else {
1057                         struct btrfs_extent_ref_v0 *ref0;
1058                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1059                                         struct btrfs_extent_ref_v0);
1060                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1061                 }
1062 #endif
1063                 btrfs_mark_buffer_dirty(leaf);
1064         }
1065         return ret;
1066 }
1067
1068 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1069                                           struct btrfs_path *path,
1070                                           struct btrfs_extent_inline_ref *iref)
1071 {
1072         struct btrfs_key key;
1073         struct extent_buffer *leaf;
1074         struct btrfs_extent_data_ref *ref1;
1075         struct btrfs_shared_data_ref *ref2;
1076         u32 num_refs = 0;
1077
1078         leaf = path->nodes[0];
1079         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1080         if (iref) {
1081                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1082                     BTRFS_EXTENT_DATA_REF_KEY) {
1083                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1084                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1085                 } else {
1086                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1087                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1088                 }
1089         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1090                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1091                                       struct btrfs_extent_data_ref);
1092                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1093         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1094                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1095                                       struct btrfs_shared_data_ref);
1096                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1097 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1098         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1099                 struct btrfs_extent_ref_v0 *ref0;
1100                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1101                                       struct btrfs_extent_ref_v0);
1102                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1103 #endif
1104         } else {
1105                 WARN_ON(1);
1106         }
1107         return num_refs;
1108 }
1109
1110 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1111                                           struct btrfs_root *root,
1112                                           struct btrfs_path *path,
1113                                           u64 bytenr, u64 parent,
1114                                           u64 root_objectid)
1115 {
1116         struct btrfs_key key;
1117         int ret;
1118
1119         key.objectid = bytenr;
1120         if (parent) {
1121                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1122                 key.offset = parent;
1123         } else {
1124                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1125                 key.offset = root_objectid;
1126         }
1127
1128         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1129         if (ret > 0)
1130                 ret = -ENOENT;
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132         if (ret == -ENOENT && parent) {
1133                 btrfs_release_path(root, path);
1134                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1135                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1136                 if (ret > 0)
1137                         ret = -ENOENT;
1138         }
1139 #endif
1140         return ret;
1141 }
1142
1143 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1144                                           struct btrfs_root *root,
1145                                           struct btrfs_path *path,
1146                                           u64 bytenr, u64 parent,
1147                                           u64 root_objectid)
1148 {
1149         struct btrfs_key key;
1150         int ret;
1151
1152         key.objectid = bytenr;
1153         if (parent) {
1154                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1155                 key.offset = parent;
1156         } else {
1157                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1158                 key.offset = root_objectid;
1159         }
1160
1161         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1162         btrfs_release_path(root, path);
1163         return ret;
1164 }
1165
1166 static inline int extent_ref_type(u64 parent, u64 owner)
1167 {
1168         int type;
1169         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1170                 if (parent > 0)
1171                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1172                 else
1173                         type = BTRFS_TREE_BLOCK_REF_KEY;
1174         } else {
1175                 if (parent > 0)
1176                         type = BTRFS_SHARED_DATA_REF_KEY;
1177                 else
1178                         type = BTRFS_EXTENT_DATA_REF_KEY;
1179         }
1180         return type;
1181 }
1182
1183 static int find_next_key(struct btrfs_path *path, int level,
1184                          struct btrfs_key *key)
1185
1186 {
1187         for (; level < BTRFS_MAX_LEVEL; level++) {
1188                 if (!path->nodes[level])
1189                         break;
1190                 if (path->slots[level] + 1 >=
1191                     btrfs_header_nritems(path->nodes[level]))
1192                         continue;
1193                 if (level == 0)
1194                         btrfs_item_key_to_cpu(path->nodes[level], key,
1195                                               path->slots[level] + 1);
1196                 else
1197                         btrfs_node_key_to_cpu(path->nodes[level], key,
1198                                               path->slots[level] + 1);
1199                 return 0;
1200         }
1201         return 1;
1202 }
1203
1204 /*
1205  * look for inline back ref. if back ref is found, *ref_ret is set
1206  * to the address of inline back ref, and 0 is returned.
1207  *
1208  * if back ref isn't found, *ref_ret is set to the address where it
1209  * should be inserted, and -ENOENT is returned.
1210  *
1211  * if insert is true and there are too many inline back refs, the path
1212  * points to the extent item, and -EAGAIN is returned.
1213  *
1214  * NOTE: inline back refs are ordered in the same way that back ref
1215  *       items in the tree are ordered.
1216  */
1217 static noinline_for_stack
1218 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1219                                  struct btrfs_root *root,
1220                                  struct btrfs_path *path,
1221                                  struct btrfs_extent_inline_ref **ref_ret,
1222                                  u64 bytenr, u64 num_bytes,
1223                                  u64 parent, u64 root_objectid,
1224                                  u64 owner, u64 offset, int insert)
1225 {
1226         struct btrfs_key key;
1227         struct extent_buffer *leaf;
1228         struct btrfs_extent_item *ei;
1229         struct btrfs_extent_inline_ref *iref;
1230         u64 flags;
1231         u64 item_size;
1232         unsigned long ptr;
1233         unsigned long end;
1234         int extra_size;
1235         int type;
1236         int want;
1237         int ret;
1238         int err = 0;
1239
1240         key.objectid = bytenr;
1241         key.type = BTRFS_EXTENT_ITEM_KEY;
1242         key.offset = num_bytes;
1243
1244         want = extent_ref_type(parent, owner);
1245         if (insert) {
1246                 extra_size = btrfs_extent_inline_ref_size(want);
1247                 path->keep_locks = 1;
1248         } else
1249                 extra_size = -1;
1250         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1251         if (ret < 0) {
1252                 err = ret;
1253                 goto out;
1254         }
1255         BUG_ON(ret);
1256
1257         leaf = path->nodes[0];
1258         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260         if (item_size < sizeof(*ei)) {
1261                 if (!insert) {
1262                         err = -ENOENT;
1263                         goto out;
1264                 }
1265                 ret = convert_extent_item_v0(trans, root, path, owner,
1266                                              extra_size);
1267                 if (ret < 0) {
1268                         err = ret;
1269                         goto out;
1270                 }
1271                 leaf = path->nodes[0];
1272                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1273         }
1274 #endif
1275         BUG_ON(item_size < sizeof(*ei));
1276
1277         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1278         flags = btrfs_extent_flags(leaf, ei);
1279
1280         ptr = (unsigned long)(ei + 1);
1281         end = (unsigned long)ei + item_size;
1282
1283         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1284                 ptr += sizeof(struct btrfs_tree_block_info);
1285                 BUG_ON(ptr > end);
1286         } else {
1287                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1288         }
1289
1290         err = -ENOENT;
1291         while (1) {
1292                 if (ptr >= end) {
1293                         WARN_ON(ptr > end);
1294                         break;
1295                 }
1296                 iref = (struct btrfs_extent_inline_ref *)ptr;
1297                 type = btrfs_extent_inline_ref_type(leaf, iref);
1298                 if (want < type)
1299                         break;
1300                 if (want > type) {
1301                         ptr += btrfs_extent_inline_ref_size(type);
1302                         continue;
1303                 }
1304
1305                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1306                         struct btrfs_extent_data_ref *dref;
1307                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1308                         if (match_extent_data_ref(leaf, dref, root_objectid,
1309                                                   owner, offset)) {
1310                                 err = 0;
1311                                 break;
1312                         }
1313                         if (hash_extent_data_ref_item(leaf, dref) <
1314                             hash_extent_data_ref(root_objectid, owner, offset))
1315                                 break;
1316                 } else {
1317                         u64 ref_offset;
1318                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1319                         if (parent > 0) {
1320                                 if (parent == ref_offset) {
1321                                         err = 0;
1322                                         break;
1323                                 }
1324                                 if (ref_offset < parent)
1325                                         break;
1326                         } else {
1327                                 if (root_objectid == ref_offset) {
1328                                         err = 0;
1329                                         break;
1330                                 }
1331                                 if (ref_offset < root_objectid)
1332                                         break;
1333                         }
1334                 }
1335                 ptr += btrfs_extent_inline_ref_size(type);
1336         }
1337         if (err == -ENOENT && insert) {
1338                 if (item_size + extra_size >=
1339                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1340                         err = -EAGAIN;
1341                         goto out;
1342                 }
1343                 /*
1344                  * To add new inline back ref, we have to make sure
1345                  * there is no corresponding back ref item.
1346                  * For simplicity, we just do not add new inline back
1347                  * ref if there is any kind of item for this block
1348                  */
1349                 if (find_next_key(path, 0, &key) == 0 &&
1350                     key.objectid == bytenr &&
1351                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1352                         err = -EAGAIN;
1353                         goto out;
1354                 }
1355         }
1356         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1357 out:
1358         if (insert) {
1359                 path->keep_locks = 0;
1360                 btrfs_unlock_up_safe(path, 1);
1361         }
1362         return err;
1363 }
1364
1365 /*
1366  * helper to add new inline back ref
1367  */
1368 static noinline_for_stack
1369 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1370                                 struct btrfs_root *root,
1371                                 struct btrfs_path *path,
1372                                 struct btrfs_extent_inline_ref *iref,
1373                                 u64 parent, u64 root_objectid,
1374                                 u64 owner, u64 offset, int refs_to_add,
1375                                 struct btrfs_delayed_extent_op *extent_op)
1376 {
1377         struct extent_buffer *leaf;
1378         struct btrfs_extent_item *ei;
1379         unsigned long ptr;
1380         unsigned long end;
1381         unsigned long item_offset;
1382         u64 refs;
1383         int size;
1384         int type;
1385         int ret;
1386
1387         leaf = path->nodes[0];
1388         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1389         item_offset = (unsigned long)iref - (unsigned long)ei;
1390
1391         type = extent_ref_type(parent, owner);
1392         size = btrfs_extent_inline_ref_size(type);
1393
1394         ret = btrfs_extend_item(trans, root, path, size);
1395         BUG_ON(ret);
1396
1397         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1398         refs = btrfs_extent_refs(leaf, ei);
1399         refs += refs_to_add;
1400         btrfs_set_extent_refs(leaf, ei, refs);
1401         if (extent_op)
1402                 __run_delayed_extent_op(extent_op, leaf, ei);
1403
1404         ptr = (unsigned long)ei + item_offset;
1405         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1406         if (ptr < end - size)
1407                 memmove_extent_buffer(leaf, ptr + size, ptr,
1408                                       end - size - ptr);
1409
1410         iref = (struct btrfs_extent_inline_ref *)ptr;
1411         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1412         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1413                 struct btrfs_extent_data_ref *dref;
1414                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1415                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1416                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1417                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1418                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1419         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1420                 struct btrfs_shared_data_ref *sref;
1421                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1422                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1423                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1424         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1425                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1426         } else {
1427                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1428         }
1429         btrfs_mark_buffer_dirty(leaf);
1430         return 0;
1431 }
1432
1433 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1434                                  struct btrfs_root *root,
1435                                  struct btrfs_path *path,
1436                                  struct btrfs_extent_inline_ref **ref_ret,
1437                                  u64 bytenr, u64 num_bytes, u64 parent,
1438                                  u64 root_objectid, u64 owner, u64 offset)
1439 {
1440         int ret;
1441
1442         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1443                                            bytenr, num_bytes, parent,
1444                                            root_objectid, owner, offset, 0);
1445         if (ret != -ENOENT)
1446                 return ret;
1447
1448         btrfs_release_path(root, path);
1449         *ref_ret = NULL;
1450
1451         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1452                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1453                                             root_objectid);
1454         } else {
1455                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1456                                              root_objectid, owner, offset);
1457         }
1458         return ret;
1459 }
1460
1461 /*
1462  * helper to update/remove inline back ref
1463  */
1464 static noinline_for_stack
1465 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1466                                  struct btrfs_root *root,
1467                                  struct btrfs_path *path,
1468                                  struct btrfs_extent_inline_ref *iref,
1469                                  int refs_to_mod,
1470                                  struct btrfs_delayed_extent_op *extent_op)
1471 {
1472         struct extent_buffer *leaf;
1473         struct btrfs_extent_item *ei;
1474         struct btrfs_extent_data_ref *dref = NULL;
1475         struct btrfs_shared_data_ref *sref = NULL;
1476         unsigned long ptr;
1477         unsigned long end;
1478         u32 item_size;
1479         int size;
1480         int type;
1481         int ret;
1482         u64 refs;
1483
1484         leaf = path->nodes[0];
1485         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1486         refs = btrfs_extent_refs(leaf, ei);
1487         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1488         refs += refs_to_mod;
1489         btrfs_set_extent_refs(leaf, ei, refs);
1490         if (extent_op)
1491                 __run_delayed_extent_op(extent_op, leaf, ei);
1492
1493         type = btrfs_extent_inline_ref_type(leaf, iref);
1494
1495         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1496                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1497                 refs = btrfs_extent_data_ref_count(leaf, dref);
1498         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1499                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1500                 refs = btrfs_shared_data_ref_count(leaf, sref);
1501         } else {
1502                 refs = 1;
1503                 BUG_ON(refs_to_mod != -1);
1504         }
1505
1506         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1507         refs += refs_to_mod;
1508
1509         if (refs > 0) {
1510                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1511                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1512                 else
1513                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1514         } else {
1515                 size =  btrfs_extent_inline_ref_size(type);
1516                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1517                 ptr = (unsigned long)iref;
1518                 end = (unsigned long)ei + item_size;
1519                 if (ptr + size < end)
1520                         memmove_extent_buffer(leaf, ptr, ptr + size,
1521                                               end - ptr - size);
1522                 item_size -= size;
1523                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1524                 BUG_ON(ret);
1525         }
1526         btrfs_mark_buffer_dirty(leaf);
1527         return 0;
1528 }
1529
1530 static noinline_for_stack
1531 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1532                                  struct btrfs_root *root,
1533                                  struct btrfs_path *path,
1534                                  u64 bytenr, u64 num_bytes, u64 parent,
1535                                  u64 root_objectid, u64 owner,
1536                                  u64 offset, int refs_to_add,
1537                                  struct btrfs_delayed_extent_op *extent_op)
1538 {
1539         struct btrfs_extent_inline_ref *iref;
1540         int ret;
1541
1542         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1543                                            bytenr, num_bytes, parent,
1544                                            root_objectid, owner, offset, 1);
1545         if (ret == 0) {
1546                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1547                 ret = update_inline_extent_backref(trans, root, path, iref,
1548                                                    refs_to_add, extent_op);
1549         } else if (ret == -ENOENT) {
1550                 ret = setup_inline_extent_backref(trans, root, path, iref,
1551                                                   parent, root_objectid,
1552                                                   owner, offset, refs_to_add,
1553                                                   extent_op);
1554         }
1555         return ret;
1556 }
1557
1558 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1559                                  struct btrfs_root *root,
1560                                  struct btrfs_path *path,
1561                                  u64 bytenr, u64 parent, u64 root_objectid,
1562                                  u64 owner, u64 offset, int refs_to_add)
1563 {
1564         int ret;
1565         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1566                 BUG_ON(refs_to_add != 1);
1567                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1568                                             parent, root_objectid);
1569         } else {
1570                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1571                                              parent, root_objectid,
1572                                              owner, offset, refs_to_add);
1573         }
1574         return ret;
1575 }
1576
1577 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1578                                  struct btrfs_root *root,
1579                                  struct btrfs_path *path,
1580                                  struct btrfs_extent_inline_ref *iref,
1581                                  int refs_to_drop, int is_data)
1582 {
1583         int ret;
1584
1585         BUG_ON(!is_data && refs_to_drop != 1);
1586         if (iref) {
1587                 ret = update_inline_extent_backref(trans, root, path, iref,
1588                                                    -refs_to_drop, NULL);
1589         } else if (is_data) {
1590                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1591         } else {
1592                 ret = btrfs_del_item(trans, root, path);
1593         }
1594         return ret;
1595 }
1596
1597 static void btrfs_issue_discard(struct block_device *bdev,
1598                                 u64 start, u64 len)
1599 {
1600         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1601                              DISCARD_FL_BARRIER);
1602 }
1603
1604 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1605                                 u64 num_bytes)
1606 {
1607         int ret;
1608         u64 map_length = num_bytes;
1609         struct btrfs_multi_bio *multi = NULL;
1610
1611         if (!btrfs_test_opt(root, DISCARD))
1612                 return 0;
1613
1614         /* Tell the block device(s) that the sectors can be discarded */
1615         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1616                               bytenr, &map_length, &multi, 0);
1617         if (!ret) {
1618                 struct btrfs_bio_stripe *stripe = multi->stripes;
1619                 int i;
1620
1621                 if (map_length > num_bytes)
1622                         map_length = num_bytes;
1623
1624                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1625                         btrfs_issue_discard(stripe->dev->bdev,
1626                                             stripe->physical,
1627                                             map_length);
1628                 }
1629                 kfree(multi);
1630         }
1631
1632         return ret;
1633 }
1634
1635 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1636                          struct btrfs_root *root,
1637                          u64 bytenr, u64 num_bytes, u64 parent,
1638                          u64 root_objectid, u64 owner, u64 offset)
1639 {
1640         int ret;
1641         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1642                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1643
1644         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1645                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1646                                         parent, root_objectid, (int)owner,
1647                                         BTRFS_ADD_DELAYED_REF, NULL);
1648         } else {
1649                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1650                                         parent, root_objectid, owner, offset,
1651                                         BTRFS_ADD_DELAYED_REF, NULL);
1652         }
1653         return ret;
1654 }
1655
1656 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1657                                   struct btrfs_root *root,
1658                                   u64 bytenr, u64 num_bytes,
1659                                   u64 parent, u64 root_objectid,
1660                                   u64 owner, u64 offset, int refs_to_add,
1661                                   struct btrfs_delayed_extent_op *extent_op)
1662 {
1663         struct btrfs_path *path;
1664         struct extent_buffer *leaf;
1665         struct btrfs_extent_item *item;
1666         u64 refs;
1667         int ret;
1668         int err = 0;
1669
1670         path = btrfs_alloc_path();
1671         if (!path)
1672                 return -ENOMEM;
1673
1674         path->reada = 1;
1675         path->leave_spinning = 1;
1676         /* this will setup the path even if it fails to insert the back ref */
1677         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1678                                            path, bytenr, num_bytes, parent,
1679                                            root_objectid, owner, offset,
1680                                            refs_to_add, extent_op);
1681         if (ret == 0)
1682                 goto out;
1683
1684         if (ret != -EAGAIN) {
1685                 err = ret;
1686                 goto out;
1687         }
1688
1689         leaf = path->nodes[0];
1690         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1691         refs = btrfs_extent_refs(leaf, item);
1692         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1693         if (extent_op)
1694                 __run_delayed_extent_op(extent_op, leaf, item);
1695
1696         btrfs_mark_buffer_dirty(leaf);
1697         btrfs_release_path(root->fs_info->extent_root, path);
1698
1699         path->reada = 1;
1700         path->leave_spinning = 1;
1701
1702         /* now insert the actual backref */
1703         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1704                                     path, bytenr, parent, root_objectid,
1705                                     owner, offset, refs_to_add);
1706         BUG_ON(ret);
1707 out:
1708         btrfs_free_path(path);
1709         return err;
1710 }
1711
1712 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1713                                 struct btrfs_root *root,
1714                                 struct btrfs_delayed_ref_node *node,
1715                                 struct btrfs_delayed_extent_op *extent_op,
1716                                 int insert_reserved)
1717 {
1718         int ret = 0;
1719         struct btrfs_delayed_data_ref *ref;
1720         struct btrfs_key ins;
1721         u64 parent = 0;
1722         u64 ref_root = 0;
1723         u64 flags = 0;
1724
1725         ins.objectid = node->bytenr;
1726         ins.offset = node->num_bytes;
1727         ins.type = BTRFS_EXTENT_ITEM_KEY;
1728
1729         ref = btrfs_delayed_node_to_data_ref(node);
1730         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1731                 parent = ref->parent;
1732         else
1733                 ref_root = ref->root;
1734
1735         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1736                 if (extent_op) {
1737                         BUG_ON(extent_op->update_key);
1738                         flags |= extent_op->flags_to_set;
1739                 }
1740                 ret = alloc_reserved_file_extent(trans, root,
1741                                                  parent, ref_root, flags,
1742                                                  ref->objectid, ref->offset,
1743                                                  &ins, node->ref_mod);
1744         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1745                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1746                                              node->num_bytes, parent,
1747                                              ref_root, ref->objectid,
1748                                              ref->offset, node->ref_mod,
1749                                              extent_op);
1750         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1751                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1752                                           node->num_bytes, parent,
1753                                           ref_root, ref->objectid,
1754                                           ref->offset, node->ref_mod,
1755                                           extent_op);
1756         } else {
1757                 BUG();
1758         }
1759         return ret;
1760 }
1761
1762 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1763                                     struct extent_buffer *leaf,
1764                                     struct btrfs_extent_item *ei)
1765 {
1766         u64 flags = btrfs_extent_flags(leaf, ei);
1767         if (extent_op->update_flags) {
1768                 flags |= extent_op->flags_to_set;
1769                 btrfs_set_extent_flags(leaf, ei, flags);
1770         }
1771
1772         if (extent_op->update_key) {
1773                 struct btrfs_tree_block_info *bi;
1774                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1775                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1776                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1777         }
1778 }
1779
1780 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1781                                  struct btrfs_root *root,
1782                                  struct btrfs_delayed_ref_node *node,
1783                                  struct btrfs_delayed_extent_op *extent_op)
1784 {
1785         struct btrfs_key key;
1786         struct btrfs_path *path;
1787         struct btrfs_extent_item *ei;
1788         struct extent_buffer *leaf;
1789         u32 item_size;
1790         int ret;
1791         int err = 0;
1792
1793         path = btrfs_alloc_path();
1794         if (!path)
1795                 return -ENOMEM;
1796
1797         key.objectid = node->bytenr;
1798         key.type = BTRFS_EXTENT_ITEM_KEY;
1799         key.offset = node->num_bytes;
1800
1801         path->reada = 1;
1802         path->leave_spinning = 1;
1803         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1804                                 path, 0, 1);
1805         if (ret < 0) {
1806                 err = ret;
1807                 goto out;
1808         }
1809         if (ret > 0) {
1810                 err = -EIO;
1811                 goto out;
1812         }
1813
1814         leaf = path->nodes[0];
1815         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1816 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1817         if (item_size < sizeof(*ei)) {
1818                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1819                                              path, (u64)-1, 0);
1820                 if (ret < 0) {
1821                         err = ret;
1822                         goto out;
1823                 }
1824                 leaf = path->nodes[0];
1825                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1826         }
1827 #endif
1828         BUG_ON(item_size < sizeof(*ei));
1829         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1830         __run_delayed_extent_op(extent_op, leaf, ei);
1831
1832         btrfs_mark_buffer_dirty(leaf);
1833 out:
1834         btrfs_free_path(path);
1835         return err;
1836 }
1837
1838 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1839                                 struct btrfs_root *root,
1840                                 struct btrfs_delayed_ref_node *node,
1841                                 struct btrfs_delayed_extent_op *extent_op,
1842                                 int insert_reserved)
1843 {
1844         int ret = 0;
1845         struct btrfs_delayed_tree_ref *ref;
1846         struct btrfs_key ins;
1847         u64 parent = 0;
1848         u64 ref_root = 0;
1849
1850         ins.objectid = node->bytenr;
1851         ins.offset = node->num_bytes;
1852         ins.type = BTRFS_EXTENT_ITEM_KEY;
1853
1854         ref = btrfs_delayed_node_to_tree_ref(node);
1855         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1856                 parent = ref->parent;
1857         else
1858                 ref_root = ref->root;
1859
1860         BUG_ON(node->ref_mod != 1);
1861         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1862                 BUG_ON(!extent_op || !extent_op->update_flags ||
1863                        !extent_op->update_key);
1864                 ret = alloc_reserved_tree_block(trans, root,
1865                                                 parent, ref_root,
1866                                                 extent_op->flags_to_set,
1867                                                 &extent_op->key,
1868                                                 ref->level, &ins);
1869         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1870                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1871                                              node->num_bytes, parent, ref_root,
1872                                              ref->level, 0, 1, extent_op);
1873         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1874                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1875                                           node->num_bytes, parent, ref_root,
1876                                           ref->level, 0, 1, extent_op);
1877         } else {
1878                 BUG();
1879         }
1880         return ret;
1881 }
1882
1883
1884 /* helper function to actually process a single delayed ref entry */
1885 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1886                                struct btrfs_root *root,
1887                                struct btrfs_delayed_ref_node *node,
1888                                struct btrfs_delayed_extent_op *extent_op,
1889                                int insert_reserved)
1890 {
1891         int ret;
1892         if (btrfs_delayed_ref_is_head(node)) {
1893                 struct btrfs_delayed_ref_head *head;
1894                 /*
1895                  * we've hit the end of the chain and we were supposed
1896                  * to insert this extent into the tree.  But, it got
1897                  * deleted before we ever needed to insert it, so all
1898                  * we have to do is clean up the accounting
1899                  */
1900                 BUG_ON(extent_op);
1901                 head = btrfs_delayed_node_to_head(node);
1902                 if (insert_reserved) {
1903                         int mark_free = 0;
1904                         struct extent_buffer *must_clean = NULL;
1905
1906                         ret = pin_down_bytes(trans, root, NULL,
1907                                              node->bytenr, node->num_bytes,
1908                                              head->is_data, 1, &must_clean);
1909                         if (ret > 0)
1910                                 mark_free = 1;
1911
1912                         if (must_clean) {
1913                                 clean_tree_block(NULL, root, must_clean);
1914                                 btrfs_tree_unlock(must_clean);
1915                                 free_extent_buffer(must_clean);
1916                         }
1917                         if (head->is_data) {
1918                                 ret = btrfs_del_csums(trans, root,
1919                                                       node->bytenr,
1920                                                       node->num_bytes);
1921                                 BUG_ON(ret);
1922                         }
1923                         if (mark_free) {
1924                                 ret = btrfs_free_reserved_extent(root,
1925                                                         node->bytenr,
1926                                                         node->num_bytes);
1927                                 BUG_ON(ret);
1928                         }
1929                 }
1930                 mutex_unlock(&head->mutex);
1931                 return 0;
1932         }
1933
1934         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1935             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1936                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1937                                            insert_reserved);
1938         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1939                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1940                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1941                                            insert_reserved);
1942         else
1943                 BUG();
1944         return ret;
1945 }
1946
1947 static noinline struct btrfs_delayed_ref_node *
1948 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1949 {
1950         struct rb_node *node;
1951         struct btrfs_delayed_ref_node *ref;
1952         int action = BTRFS_ADD_DELAYED_REF;
1953 again:
1954         /*
1955          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1956          * this prevents ref count from going down to zero when
1957          * there still are pending delayed ref.
1958          */
1959         node = rb_prev(&head->node.rb_node);
1960         while (1) {
1961                 if (!node)
1962                         break;
1963                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1964                                 rb_node);
1965                 if (ref->bytenr != head->node.bytenr)
1966                         break;
1967                 if (ref->action == action)
1968                         return ref;
1969                 node = rb_prev(node);
1970         }
1971         if (action == BTRFS_ADD_DELAYED_REF) {
1972                 action = BTRFS_DROP_DELAYED_REF;
1973                 goto again;
1974         }
1975         return NULL;
1976 }
1977
1978 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1979                                        struct btrfs_root *root,
1980                                        struct list_head *cluster)
1981 {
1982         struct btrfs_delayed_ref_root *delayed_refs;
1983         struct btrfs_delayed_ref_node *ref;
1984         struct btrfs_delayed_ref_head *locked_ref = NULL;
1985         struct btrfs_delayed_extent_op *extent_op;
1986         int ret;
1987         int count = 0;
1988         int must_insert_reserved = 0;
1989
1990         delayed_refs = &trans->transaction->delayed_refs;
1991         while (1) {
1992                 if (!locked_ref) {
1993                         /* pick a new head ref from the cluster list */
1994                         if (list_empty(cluster))
1995                                 break;
1996
1997                         locked_ref = list_entry(cluster->next,
1998                                      struct btrfs_delayed_ref_head, cluster);
1999
2000                         /* grab the lock that says we are going to process
2001                          * all the refs for this head */
2002                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2003
2004                         /*
2005                          * we may have dropped the spin lock to get the head
2006                          * mutex lock, and that might have given someone else
2007                          * time to free the head.  If that's true, it has been
2008                          * removed from our list and we can move on.
2009                          */
2010                         if (ret == -EAGAIN) {
2011                                 locked_ref = NULL;
2012                                 count++;
2013                                 continue;
2014                         }
2015                 }
2016
2017                 /*
2018                  * record the must insert reserved flag before we
2019                  * drop the spin lock.
2020                  */
2021                 must_insert_reserved = locked_ref->must_insert_reserved;
2022                 locked_ref->must_insert_reserved = 0;
2023
2024                 extent_op = locked_ref->extent_op;
2025                 locked_ref->extent_op = NULL;
2026
2027                 /*
2028                  * locked_ref is the head node, so we have to go one
2029                  * node back for any delayed ref updates
2030                  */
2031                 ref = select_delayed_ref(locked_ref);
2032                 if (!ref) {
2033                         /* All delayed refs have been processed, Go ahead
2034                          * and send the head node to run_one_delayed_ref,
2035                          * so that any accounting fixes can happen
2036                          */
2037                         ref = &locked_ref->node;
2038
2039                         if (extent_op && must_insert_reserved) {
2040                                 kfree(extent_op);
2041                                 extent_op = NULL;
2042                         }
2043
2044                         if (extent_op) {
2045                                 spin_unlock(&delayed_refs->lock);
2046
2047                                 ret = run_delayed_extent_op(trans, root,
2048                                                             ref, extent_op);
2049                                 BUG_ON(ret);
2050                                 kfree(extent_op);
2051
2052                                 cond_resched();
2053                                 spin_lock(&delayed_refs->lock);
2054                                 continue;
2055                         }
2056
2057                         list_del_init(&locked_ref->cluster);
2058                         locked_ref = NULL;
2059                 }
2060
2061                 ref->in_tree = 0;
2062                 rb_erase(&ref->rb_node, &delayed_refs->root);
2063                 delayed_refs->num_entries--;
2064
2065                 spin_unlock(&delayed_refs->lock);
2066
2067                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2068                                           must_insert_reserved);
2069                 BUG_ON(ret);
2070
2071                 btrfs_put_delayed_ref(ref);
2072                 kfree(extent_op);
2073                 count++;
2074
2075                 cond_resched();
2076                 spin_lock(&delayed_refs->lock);
2077         }
2078         return count;
2079 }
2080
2081 /*
2082  * this starts processing the delayed reference count updates and
2083  * extent insertions we have queued up so far.  count can be
2084  * 0, which means to process everything in the tree at the start
2085  * of the run (but not newly added entries), or it can be some target
2086  * number you'd like to process.
2087  */
2088 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2089                            struct btrfs_root *root, unsigned long count)
2090 {
2091         struct rb_node *node;
2092         struct btrfs_delayed_ref_root *delayed_refs;
2093         struct btrfs_delayed_ref_node *ref;
2094         struct list_head cluster;
2095         int ret;
2096         int run_all = count == (unsigned long)-1;
2097         int run_most = 0;
2098
2099         if (root == root->fs_info->extent_root)
2100                 root = root->fs_info->tree_root;
2101
2102         delayed_refs = &trans->transaction->delayed_refs;
2103         INIT_LIST_HEAD(&cluster);
2104 again:
2105         spin_lock(&delayed_refs->lock);
2106         if (count == 0) {
2107                 count = delayed_refs->num_entries * 2;
2108                 run_most = 1;
2109         }
2110         while (1) {
2111                 if (!(run_all || run_most) &&
2112                     delayed_refs->num_heads_ready < 64)
2113                         break;
2114
2115                 /*
2116                  * go find something we can process in the rbtree.  We start at
2117                  * the beginning of the tree, and then build a cluster
2118                  * of refs to process starting at the first one we are able to
2119                  * lock
2120                  */
2121                 ret = btrfs_find_ref_cluster(trans, &cluster,
2122                                              delayed_refs->run_delayed_start);
2123                 if (ret)
2124                         break;
2125
2126                 ret = run_clustered_refs(trans, root, &cluster);
2127                 BUG_ON(ret < 0);
2128
2129                 count -= min_t(unsigned long, ret, count);
2130
2131                 if (count == 0)
2132                         break;
2133         }
2134
2135         if (run_all) {
2136                 node = rb_first(&delayed_refs->root);
2137                 if (!node)
2138                         goto out;
2139                 count = (unsigned long)-1;
2140
2141                 while (node) {
2142                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2143                                        rb_node);
2144                         if (btrfs_delayed_ref_is_head(ref)) {
2145                                 struct btrfs_delayed_ref_head *head;
2146
2147                                 head = btrfs_delayed_node_to_head(ref);
2148                                 atomic_inc(&ref->refs);
2149
2150                                 spin_unlock(&delayed_refs->lock);
2151                                 mutex_lock(&head->mutex);
2152                                 mutex_unlock(&head->mutex);
2153
2154                                 btrfs_put_delayed_ref(ref);
2155                                 cond_resched();
2156                                 goto again;
2157                         }
2158                         node = rb_next(node);
2159                 }
2160                 spin_unlock(&delayed_refs->lock);
2161                 schedule_timeout(1);
2162                 goto again;
2163         }
2164 out:
2165         spin_unlock(&delayed_refs->lock);
2166         return 0;
2167 }
2168
2169 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2170                                 struct btrfs_root *root,
2171                                 u64 bytenr, u64 num_bytes, u64 flags,
2172                                 int is_data)
2173 {
2174         struct btrfs_delayed_extent_op *extent_op;
2175         int ret;
2176
2177         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2178         if (!extent_op)
2179                 return -ENOMEM;
2180
2181         extent_op->flags_to_set = flags;
2182         extent_op->update_flags = 1;
2183         extent_op->update_key = 0;
2184         extent_op->is_data = is_data ? 1 : 0;
2185
2186         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2187         if (ret)
2188                 kfree(extent_op);
2189         return ret;
2190 }
2191
2192 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2193                                       struct btrfs_root *root,
2194                                       struct btrfs_path *path,
2195                                       u64 objectid, u64 offset, u64 bytenr)
2196 {
2197         struct btrfs_delayed_ref_head *head;
2198         struct btrfs_delayed_ref_node *ref;
2199         struct btrfs_delayed_data_ref *data_ref;
2200         struct btrfs_delayed_ref_root *delayed_refs;
2201         struct rb_node *node;
2202         int ret = 0;
2203
2204         ret = -ENOENT;
2205         delayed_refs = &trans->transaction->delayed_refs;
2206         spin_lock(&delayed_refs->lock);
2207         head = btrfs_find_delayed_ref_head(trans, bytenr);
2208         if (!head)
2209                 goto out;
2210
2211         if (!mutex_trylock(&head->mutex)) {
2212                 atomic_inc(&head->node.refs);
2213                 spin_unlock(&delayed_refs->lock);
2214
2215                 btrfs_release_path(root->fs_info->extent_root, path);
2216
2217                 mutex_lock(&head->mutex);
2218                 mutex_unlock(&head->mutex);
2219                 btrfs_put_delayed_ref(&head->node);
2220                 return -EAGAIN;
2221         }
2222
2223         node = rb_prev(&head->node.rb_node);
2224         if (!node)
2225                 goto out_unlock;
2226
2227         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2228
2229         if (ref->bytenr != bytenr)
2230                 goto out_unlock;
2231
2232         ret = 1;
2233         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2234                 goto out_unlock;
2235
2236         data_ref = btrfs_delayed_node_to_data_ref(ref);
2237
2238         node = rb_prev(node);
2239         if (node) {
2240                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2241                 if (ref->bytenr == bytenr)
2242                         goto out_unlock;
2243         }
2244
2245         if (data_ref->root != root->root_key.objectid ||
2246             data_ref->objectid != objectid || data_ref->offset != offset)
2247                 goto out_unlock;
2248
2249         ret = 0;
2250 out_unlock:
2251         mutex_unlock(&head->mutex);
2252 out:
2253         spin_unlock(&delayed_refs->lock);
2254         return ret;
2255 }
2256
2257 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2258                                         struct btrfs_root *root,
2259                                         struct btrfs_path *path,
2260                                         u64 objectid, u64 offset, u64 bytenr)
2261 {
2262         struct btrfs_root *extent_root = root->fs_info->extent_root;
2263         struct extent_buffer *leaf;
2264         struct btrfs_extent_data_ref *ref;
2265         struct btrfs_extent_inline_ref *iref;
2266         struct btrfs_extent_item *ei;
2267         struct btrfs_key key;
2268         u32 item_size;
2269         int ret;
2270
2271         key.objectid = bytenr;
2272         key.offset = (u64)-1;
2273         key.type = BTRFS_EXTENT_ITEM_KEY;
2274
2275         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2276         if (ret < 0)
2277                 goto out;
2278         BUG_ON(ret == 0);
2279
2280         ret = -ENOENT;
2281         if (path->slots[0] == 0)
2282                 goto out;
2283
2284         path->slots[0]--;
2285         leaf = path->nodes[0];
2286         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287
2288         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2289                 goto out;
2290
2291         ret = 1;
2292         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2294         if (item_size < sizeof(*ei)) {
2295                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2296                 goto out;
2297         }
2298 #endif
2299         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2300
2301         if (item_size != sizeof(*ei) +
2302             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2303                 goto out;
2304
2305         if (btrfs_extent_generation(leaf, ei) <=
2306             btrfs_root_last_snapshot(&root->root_item))
2307                 goto out;
2308
2309         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2310         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2311             BTRFS_EXTENT_DATA_REF_KEY)
2312                 goto out;
2313
2314         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2315         if (btrfs_extent_refs(leaf, ei) !=
2316             btrfs_extent_data_ref_count(leaf, ref) ||
2317             btrfs_extent_data_ref_root(leaf, ref) !=
2318             root->root_key.objectid ||
2319             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2320             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2321                 goto out;
2322
2323         ret = 0;
2324 out:
2325         return ret;
2326 }
2327
2328 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2329                           struct btrfs_root *root,
2330                           u64 objectid, u64 offset, u64 bytenr)
2331 {
2332         struct btrfs_path *path;
2333         int ret;
2334         int ret2;
2335
2336         path = btrfs_alloc_path();
2337         if (!path)
2338                 return -ENOENT;
2339
2340         do {
2341                 ret = check_committed_ref(trans, root, path, objectid,
2342                                           offset, bytenr);
2343                 if (ret && ret != -ENOENT)
2344                         goto out;
2345
2346                 ret2 = check_delayed_ref(trans, root, path, objectid,
2347                                          offset, bytenr);
2348         } while (ret2 == -EAGAIN);
2349
2350         if (ret2 && ret2 != -ENOENT) {
2351                 ret = ret2;
2352                 goto out;
2353         }
2354
2355         if (ret != -ENOENT || ret2 != -ENOENT)
2356                 ret = 0;
2357 out:
2358         btrfs_free_path(path);
2359         return ret;
2360 }
2361
2362 #if 0
2363 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2364                     struct extent_buffer *buf, u32 nr_extents)
2365 {
2366         struct btrfs_key key;
2367         struct btrfs_file_extent_item *fi;
2368         u64 root_gen;
2369         u32 nritems;
2370         int i;
2371         int level;
2372         int ret = 0;
2373         int shared = 0;
2374
2375         if (!root->ref_cows)
2376                 return 0;
2377
2378         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2379                 shared = 0;
2380                 root_gen = root->root_key.offset;
2381         } else {
2382                 shared = 1;
2383                 root_gen = trans->transid - 1;
2384         }
2385
2386         level = btrfs_header_level(buf);
2387         nritems = btrfs_header_nritems(buf);
2388
2389         if (level == 0) {
2390                 struct btrfs_leaf_ref *ref;
2391                 struct btrfs_extent_info *info;
2392
2393                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2394                 if (!ref) {
2395                         ret = -ENOMEM;
2396                         goto out;
2397                 }
2398
2399                 ref->root_gen = root_gen;
2400                 ref->bytenr = buf->start;
2401                 ref->owner = btrfs_header_owner(buf);
2402                 ref->generation = btrfs_header_generation(buf);
2403                 ref->nritems = nr_extents;
2404                 info = ref->extents;
2405
2406                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2407                         u64 disk_bytenr;
2408                         btrfs_item_key_to_cpu(buf, &key, i);
2409                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2410                                 continue;
2411                         fi = btrfs_item_ptr(buf, i,
2412                                             struct btrfs_file_extent_item);
2413                         if (btrfs_file_extent_type(buf, fi) ==
2414                             BTRFS_FILE_EXTENT_INLINE)
2415                                 continue;
2416                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2417                         if (disk_bytenr == 0)
2418                                 continue;
2419
2420                         info->bytenr = disk_bytenr;
2421                         info->num_bytes =
2422                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2423                         info->objectid = key.objectid;
2424                         info->offset = key.offset;
2425                         info++;
2426                 }
2427
2428                 ret = btrfs_add_leaf_ref(root, ref, shared);
2429                 if (ret == -EEXIST && shared) {
2430                         struct btrfs_leaf_ref *old;
2431                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2432                         BUG_ON(!old);
2433                         btrfs_remove_leaf_ref(root, old);
2434                         btrfs_free_leaf_ref(root, old);
2435                         ret = btrfs_add_leaf_ref(root, ref, shared);
2436                 }
2437                 WARN_ON(ret);
2438                 btrfs_free_leaf_ref(root, ref);
2439         }
2440 out:
2441         return ret;
2442 }
2443
2444 /* when a block goes through cow, we update the reference counts of
2445  * everything that block points to.  The internal pointers of the block
2446  * can be in just about any order, and it is likely to have clusters of
2447  * things that are close together and clusters of things that are not.
2448  *
2449  * To help reduce the seeks that come with updating all of these reference
2450  * counts, sort them by byte number before actual updates are done.
2451  *
2452  * struct refsort is used to match byte number to slot in the btree block.
2453  * we sort based on the byte number and then use the slot to actually
2454  * find the item.
2455  *
2456  * struct refsort is smaller than strcut btrfs_item and smaller than
2457  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2458  * for a btree block, there's no way for a kmalloc of refsorts for a
2459  * single node to be bigger than a page.
2460  */
2461 struct refsort {
2462         u64 bytenr;
2463         u32 slot;
2464 };
2465
2466 /*
2467  * for passing into sort()
2468  */
2469 static int refsort_cmp(const void *a_void, const void *b_void)
2470 {
2471         const struct refsort *a = a_void;
2472         const struct refsort *b = b_void;
2473
2474         if (a->bytenr < b->bytenr)
2475                 return -1;
2476         if (a->bytenr > b->bytenr)
2477                 return 1;
2478         return 0;
2479 }
2480 #endif
2481
2482 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2483                            struct btrfs_root *root,
2484                            struct extent_buffer *buf,
2485                            int full_backref, int inc)
2486 {
2487         u64 bytenr;
2488         u64 num_bytes;
2489         u64 parent;
2490         u64 ref_root;
2491         u32 nritems;
2492         struct btrfs_key key;
2493         struct btrfs_file_extent_item *fi;
2494         int i;
2495         int level;
2496         int ret = 0;
2497         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2498                             u64, u64, u64, u64, u64, u64);
2499
2500         ref_root = btrfs_header_owner(buf);
2501         nritems = btrfs_header_nritems(buf);
2502         level = btrfs_header_level(buf);
2503
2504         if (!root->ref_cows && level == 0)
2505                 return 0;
2506
2507         if (inc)
2508                 process_func = btrfs_inc_extent_ref;
2509         else
2510                 process_func = btrfs_free_extent;
2511
2512         if (full_backref)
2513                 parent = buf->start;
2514         else
2515                 parent = 0;
2516
2517         for (i = 0; i < nritems; i++) {
2518                 if (level == 0) {
2519                         btrfs_item_key_to_cpu(buf, &key, i);
2520                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2521                                 continue;
2522                         fi = btrfs_item_ptr(buf, i,
2523                                             struct btrfs_file_extent_item);
2524                         if (btrfs_file_extent_type(buf, fi) ==
2525                             BTRFS_FILE_EXTENT_INLINE)
2526                                 continue;
2527                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2528                         if (bytenr == 0)
2529                                 continue;
2530
2531                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2532                         key.offset -= btrfs_file_extent_offset(buf, fi);
2533                         ret = process_func(trans, root, bytenr, num_bytes,
2534                                            parent, ref_root, key.objectid,
2535                                            key.offset);
2536                         if (ret)
2537                                 goto fail;
2538                 } else {
2539                         bytenr = btrfs_node_blockptr(buf, i);
2540                         num_bytes = btrfs_level_size(root, level - 1);
2541                         ret = process_func(trans, root, bytenr, num_bytes,
2542                                            parent, ref_root, level - 1, 0);
2543                         if (ret)
2544                                 goto fail;
2545                 }
2546         }
2547         return 0;
2548 fail:
2549         BUG();
2550         return ret;
2551 }
2552
2553 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2554                   struct extent_buffer *buf, int full_backref)
2555 {
2556         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2557 }
2558
2559 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2560                   struct extent_buffer *buf, int full_backref)
2561 {
2562         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2563 }
2564
2565 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2566                                  struct btrfs_root *root,
2567                                  struct btrfs_path *path,
2568                                  struct btrfs_block_group_cache *cache)
2569 {
2570         int ret;
2571         struct btrfs_root *extent_root = root->fs_info->extent_root;
2572         unsigned long bi;
2573         struct extent_buffer *leaf;
2574
2575         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2576         if (ret < 0)
2577                 goto fail;
2578         BUG_ON(ret);
2579
2580         leaf = path->nodes[0];
2581         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2582         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2583         btrfs_mark_buffer_dirty(leaf);
2584         btrfs_release_path(extent_root, path);
2585 fail:
2586         if (ret)
2587                 return ret;
2588         return 0;
2589
2590 }
2591
2592 static struct btrfs_block_group_cache *
2593 next_block_group(struct btrfs_root *root,
2594                  struct btrfs_block_group_cache *cache)
2595 {
2596         struct rb_node *node;
2597         spin_lock(&root->fs_info->block_group_cache_lock);
2598         node = rb_next(&cache->cache_node);
2599         btrfs_put_block_group(cache);
2600         if (node) {
2601                 cache = rb_entry(node, struct btrfs_block_group_cache,
2602                                  cache_node);
2603                 btrfs_get_block_group(cache);
2604         } else
2605                 cache = NULL;
2606         spin_unlock(&root->fs_info->block_group_cache_lock);
2607         return cache;
2608 }
2609
2610 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2611                                    struct btrfs_root *root)
2612 {
2613         struct btrfs_block_group_cache *cache;
2614         int err = 0;
2615         struct btrfs_path *path;
2616         u64 last = 0;
2617
2618         path = btrfs_alloc_path();
2619         if (!path)
2620                 return -ENOMEM;
2621
2622         while (1) {
2623                 if (last == 0) {
2624                         err = btrfs_run_delayed_refs(trans, root,
2625                                                      (unsigned long)-1);
2626                         BUG_ON(err);
2627                 }
2628
2629                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2630                 while (cache) {
2631                         if (cache->dirty)
2632                                 break;
2633                         cache = next_block_group(root, cache);
2634                 }
2635                 if (!cache) {
2636                         if (last == 0)
2637                                 break;
2638                         last = 0;
2639                         continue;
2640                 }
2641
2642                 cache->dirty = 0;
2643                 last = cache->key.objectid + cache->key.offset;
2644
2645                 err = write_one_cache_group(trans, root, path, cache);
2646                 BUG_ON(err);
2647                 btrfs_put_block_group(cache);
2648         }
2649
2650         btrfs_free_path(path);
2651         return 0;
2652 }
2653
2654 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2655 {
2656         struct btrfs_block_group_cache *block_group;
2657         int readonly = 0;
2658
2659         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2660         if (!block_group || block_group->ro)
2661                 readonly = 1;
2662         if (block_group)
2663                 btrfs_put_block_group(block_group);
2664         return readonly;
2665 }
2666
2667 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2668                              u64 total_bytes, u64 bytes_used,
2669                              struct btrfs_space_info **space_info)
2670 {
2671         struct btrfs_space_info *found;
2672         int i;
2673         int factor;
2674
2675         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2676                      BTRFS_BLOCK_GROUP_RAID10))
2677                 factor = 2;
2678         else
2679                 factor = 1;
2680
2681         found = __find_space_info(info, flags);
2682         if (found) {
2683                 spin_lock(&found->lock);
2684                 found->total_bytes += total_bytes;
2685                 found->bytes_used += bytes_used;
2686                 found->disk_used += bytes_used * factor;
2687                 found->full = 0;
2688                 spin_unlock(&found->lock);
2689                 *space_info = found;
2690                 return 0;
2691         }
2692         found = kzalloc(sizeof(*found), GFP_NOFS);
2693         if (!found)
2694                 return -ENOMEM;
2695
2696         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2697                 INIT_LIST_HEAD(&found->block_groups[i]);
2698         init_rwsem(&found->groups_sem);
2699         spin_lock_init(&found->lock);
2700         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2701                                 BTRFS_BLOCK_GROUP_SYSTEM |
2702                                 BTRFS_BLOCK_GROUP_METADATA);
2703         found->total_bytes = total_bytes;
2704         found->bytes_used = bytes_used;
2705         found->disk_used = bytes_used * factor;
2706         found->bytes_pinned = 0;
2707         found->bytes_reserved = 0;
2708         found->bytes_readonly = 0;
2709         found->bytes_delalloc = 0;
2710         found->full = 0;
2711         found->force_alloc = 0;
2712         *space_info = found;
2713         list_add_rcu(&found->list, &info->space_info);
2714         atomic_set(&found->caching_threads, 0);
2715         return 0;
2716 }
2717
2718 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2719 {
2720         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2721                                    BTRFS_BLOCK_GROUP_RAID1 |
2722                                    BTRFS_BLOCK_GROUP_RAID10 |
2723                                    BTRFS_BLOCK_GROUP_DUP);
2724         if (extra_flags) {
2725                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2726                         fs_info->avail_data_alloc_bits |= extra_flags;
2727                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2728                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2729                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2730                         fs_info->avail_system_alloc_bits |= extra_flags;
2731         }
2732 }
2733
2734 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2735 {
2736         spin_lock(&cache->space_info->lock);
2737         spin_lock(&cache->lock);
2738         if (!cache->ro) {
2739                 cache->space_info->bytes_readonly += cache->key.offset -
2740                                         btrfs_block_group_used(&cache->item);
2741                 cache->ro = 1;
2742         }
2743         spin_unlock(&cache->lock);
2744         spin_unlock(&cache->space_info->lock);
2745 }
2746
2747 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2748 {
2749         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2750
2751         if (num_devices == 1)
2752                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2753         if (num_devices < 4)
2754                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2755
2756         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2757             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2758                       BTRFS_BLOCK_GROUP_RAID10))) {
2759                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2760         }
2761
2762         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2763             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2764                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2765         }
2766
2767         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2768             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2769              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2770              (flags & BTRFS_BLOCK_GROUP_DUP)))
2771                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2772         return flags;
2773 }
2774
2775 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2776 {
2777         if (flags & BTRFS_BLOCK_GROUP_DATA)
2778                 flags |= root->fs_info->avail_data_alloc_bits &
2779                          root->fs_info->data_alloc_profile;
2780         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2781                 flags |= root->fs_info->avail_system_alloc_bits &
2782                          root->fs_info->system_alloc_profile;
2783         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2784                 flags |= root->fs_info->avail_metadata_alloc_bits &
2785                          root->fs_info->metadata_alloc_profile;
2786         return btrfs_reduce_alloc_profile(root, flags);
2787 }
2788
2789 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2790 {
2791         u64 flags;
2792
2793         if (data)
2794                 flags = BTRFS_BLOCK_GROUP_DATA;
2795         else if (root == root->fs_info->chunk_root)
2796                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2797         else
2798                 flags = BTRFS_BLOCK_GROUP_METADATA;
2799
2800         return get_alloc_profile(root, flags);
2801 }
2802
2803 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2804 {
2805         u64 alloc_target;
2806
2807         alloc_target = btrfs_get_alloc_profile(root, 1);
2808         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2809                                                        alloc_target);
2810 }
2811
2812 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2813 {
2814         u64 num_bytes;
2815         int level;
2816
2817         level = BTRFS_MAX_LEVEL - 2;
2818         /*
2819          * NOTE: these calculations are absolutely the worst possible case.
2820          * This assumes that _every_ item we insert will require a new leaf, and
2821          * that the tree has grown to its maximum level size.
2822          */
2823
2824         /*
2825          * for every item we insert we could insert both an extent item and a
2826          * extent ref item.  Then for ever item we insert, we will need to cow
2827          * both the original leaf, plus the leaf to the left and right of it.
2828          *
2829          * Unless we are talking about the extent root, then we just want the
2830          * number of items * 2, since we just need the extent item plus its ref.
2831          */
2832         if (root == root->fs_info->extent_root)
2833                 num_bytes = num_items * 2;
2834         else
2835                 num_bytes = (num_items + (2 * num_items)) * 3;
2836
2837         /*
2838          * num_bytes is total number of leaves we could need times the leaf
2839          * size, and then for every leaf we could end up cow'ing 2 nodes per
2840          * level, down to the leaf level.
2841          */
2842         num_bytes = (num_bytes * root->leafsize) +
2843                 (num_bytes * (level * 2)) * root->nodesize;
2844
2845         return num_bytes;
2846 }
2847
2848 /*
2849  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2850  * we have extents, this function does nothing.
2851  */
2852 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2853                                           struct inode *inode, int num_items)
2854 {
2855         struct btrfs_fs_info *info = root->fs_info;
2856         struct btrfs_space_info *meta_sinfo;
2857         u64 num_bytes;
2858         u64 alloc_target;
2859         bool bug = false;
2860
2861         /* get the space info for where the metadata will live */
2862         alloc_target = btrfs_get_alloc_profile(root, 0);
2863         meta_sinfo = __find_space_info(info, alloc_target);
2864
2865         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2866                                            num_items);
2867
2868         spin_lock(&meta_sinfo->lock);
2869         spin_lock(&BTRFS_I(inode)->accounting_lock);
2870         if (BTRFS_I(inode)->reserved_extents <=
2871             BTRFS_I(inode)->outstanding_extents) {
2872                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2873                 spin_unlock(&meta_sinfo->lock);
2874                 return 0;
2875         }
2876         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2877
2878         BTRFS_I(inode)->reserved_extents -= num_items;
2879         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2880
2881         if (meta_sinfo->bytes_delalloc < num_bytes) {
2882                 bug = true;
2883                 meta_sinfo->bytes_delalloc = 0;
2884         } else {
2885                 meta_sinfo->bytes_delalloc -= num_bytes;
2886         }
2887         spin_unlock(&meta_sinfo->lock);
2888
2889         BUG_ON(bug);
2890
2891         return 0;
2892 }
2893
2894 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2895 {
2896         u64 thresh;
2897
2898         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2899                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2900                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2901                 meta_sinfo->bytes_may_use;
2902
2903         thresh = meta_sinfo->total_bytes - thresh;
2904         thresh *= 80;
2905         do_div(thresh, 100);
2906         if (thresh <= meta_sinfo->bytes_delalloc)
2907                 meta_sinfo->force_delalloc = 1;
2908         else
2909                 meta_sinfo->force_delalloc = 0;
2910 }
2911
2912 /*
2913  * Reserve metadata space for delalloc.
2914  */
2915 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
2916                                         struct inode *inode, int num_items)
2917 {
2918         struct btrfs_fs_info *info = root->fs_info;
2919         struct btrfs_space_info *meta_sinfo;
2920         u64 num_bytes;
2921         u64 used;
2922         u64 alloc_target;
2923         int flushed = 0;
2924         int force_delalloc;
2925
2926         /* get the space info for where the metadata will live */
2927         alloc_target = btrfs_get_alloc_profile(root, 0);
2928         meta_sinfo = __find_space_info(info, alloc_target);
2929
2930         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2931                                            num_items);
2932 again:
2933         spin_lock(&meta_sinfo->lock);
2934
2935         force_delalloc = meta_sinfo->force_delalloc;
2936
2937         if (unlikely(!meta_sinfo->bytes_root))
2938                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
2939
2940         if (!flushed)
2941                 meta_sinfo->bytes_delalloc += num_bytes;
2942
2943         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2944                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2945                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2946                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
2947
2948         if (used > meta_sinfo->total_bytes) {
2949                 flushed++;
2950
2951                 if (flushed == 1) {
2952                         if (maybe_allocate_chunk(NULL, root, meta_sinfo,
2953                                                  num_bytes))
2954                                 goto again;
2955                         flushed++;
2956                 } else {
2957                         spin_unlock(&meta_sinfo->lock);
2958                 }
2959
2960                 if (flushed == 2) {
2961                         filemap_flush(inode->i_mapping);
2962                         goto again;
2963                 } else if (flushed == 3) {
2964                         shrink_delalloc(NULL, root, meta_sinfo, num_bytes);
2965                         goto again;
2966                 }
2967                 spin_lock(&meta_sinfo->lock);
2968                 meta_sinfo->bytes_delalloc -= num_bytes;
2969                 spin_unlock(&meta_sinfo->lock);
2970                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
2971                        BTRFS_I(inode)->outstanding_extents,
2972                        BTRFS_I(inode)->reserved_extents);
2973                 dump_space_info(meta_sinfo, 0, 0);
2974                 return -ENOSPC;
2975         }
2976
2977         BTRFS_I(inode)->reserved_extents += num_items;
2978         check_force_delalloc(meta_sinfo);
2979         spin_unlock(&meta_sinfo->lock);
2980
2981         if (!flushed && force_delalloc)
2982                 filemap_flush(inode->i_mapping);
2983
2984         return 0;
2985 }
2986
2987 /*
2988  * unreserve num_items number of items worth of metadata space.  This needs to
2989  * be paired with btrfs_reserve_metadata_space.
2990  *
2991  * NOTE: if you have the option, run this _AFTER_ you do a
2992  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
2993  * oprations which will result in more used metadata, so we want to make sure we
2994  * can do that without issue.
2995  */
2996 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
2997 {
2998         struct btrfs_fs_info *info = root->fs_info;
2999         struct btrfs_space_info *meta_sinfo;
3000         u64 num_bytes;
3001         u64 alloc_target;
3002         bool bug = false;
3003
3004         /* get the space info for where the metadata will live */
3005         alloc_target = btrfs_get_alloc_profile(root, 0);
3006         meta_sinfo = __find_space_info(info, alloc_target);
3007
3008         num_bytes = calculate_bytes_needed(root, num_items);
3009
3010         spin_lock(&meta_sinfo->lock);
3011         if (meta_sinfo->bytes_may_use < num_bytes) {
3012                 bug = true;
3013                 meta_sinfo->bytes_may_use = 0;
3014         } else {
3015                 meta_sinfo->bytes_may_use -= num_bytes;
3016         }
3017         spin_unlock(&meta_sinfo->lock);
3018
3019         BUG_ON(bug);
3020
3021         return 0;
3022 }
3023
3024 /*
3025  * Reserve some metadata space for use.  We'll calculate the worste case number
3026  * of bytes that would be needed to modify num_items number of items.  If we
3027  * have space, fantastic, if not, you get -ENOSPC.  Please call
3028  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3029  * items you reserved, since whatever metadata you needed should have already
3030  * been allocated.
3031  *
3032  * This will commit the transaction to make more space if we don't have enough
3033  * metadata space.  THe only time we don't do this is if we're reserving space
3034  * inside of a transaction, then we will just return -ENOSPC and it is the
3035  * callers responsibility to handle it properly.
3036  */
3037 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3038 {
3039         struct btrfs_fs_info *info = root->fs_info;
3040         struct btrfs_space_info *meta_sinfo;
3041         u64 num_bytes;
3042         u64 used;
3043         u64 alloc_target;
3044         int retries = 0;
3045
3046         /* get the space info for where the metadata will live */
3047         alloc_target = btrfs_get_alloc_profile(root, 0);
3048         meta_sinfo = __find_space_info(info, alloc_target);
3049
3050         num_bytes = calculate_bytes_needed(root, num_items);
3051 again:
3052         spin_lock(&meta_sinfo->lock);
3053
3054         if (unlikely(!meta_sinfo->bytes_root))
3055                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3056
3057         if (!retries)
3058                 meta_sinfo->bytes_may_use += num_bytes;
3059
3060         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3061                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3062                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3063                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3064
3065         if (used > meta_sinfo->total_bytes) {
3066                 retries++;
3067                 if (retries == 1) {
3068                         if (maybe_allocate_chunk(NULL, root, meta_sinfo,
3069                                                  num_bytes))
3070                                 goto again;
3071                         retries++;
3072                 } else {
3073                         spin_unlock(&meta_sinfo->lock);
3074                 }
3075
3076                 if (retries == 2) {
3077                         shrink_delalloc(NULL, root, meta_sinfo, num_bytes);
3078                         goto again;
3079                 }
3080                 spin_lock(&meta_sinfo->lock);
3081                 meta_sinfo->bytes_may_use -= num_bytes;
3082                 spin_unlock(&meta_sinfo->lock);
3083
3084                 dump_space_info(meta_sinfo, 0, 0);
3085                 return -ENOSPC;
3086         }
3087
3088         check_force_delalloc(meta_sinfo);
3089         spin_unlock(&meta_sinfo->lock);
3090
3091         return 0;
3092 }
3093
3094 /*
3095  * This will check the space that the inode allocates from to make sure we have
3096  * enough space for bytes.
3097  */
3098 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3099                                 u64 bytes)
3100 {
3101         struct btrfs_space_info *data_sinfo;
3102         u64 used;
3103         int ret = 0, committed = 0;
3104
3105         /* make sure bytes are sectorsize aligned */
3106         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3107
3108         data_sinfo = BTRFS_I(inode)->space_info;
3109         if (!data_sinfo)
3110                 goto alloc;
3111
3112 again:
3113         /* make sure we have enough space to handle the data first */
3114         spin_lock(&data_sinfo->lock);
3115         used = data_sinfo->bytes_used + data_sinfo->bytes_delalloc +
3116                 data_sinfo->bytes_reserved + data_sinfo->bytes_pinned +
3117                 data_sinfo->bytes_readonly + data_sinfo->bytes_may_use +
3118                 data_sinfo->bytes_super;
3119
3120         if (used + bytes > data_sinfo->total_bytes) {
3121                 struct btrfs_trans_handle *trans;
3122
3123                 /*
3124                  * if we don't have enough free bytes in this space then we need
3125                  * to alloc a new chunk.
3126                  */
3127                 if (!data_sinfo->full) {
3128                         u64 alloc_target;
3129
3130                         data_sinfo->force_alloc = 1;
3131                         spin_unlock(&data_sinfo->lock);
3132 alloc:
3133                         alloc_target = btrfs_get_alloc_profile(root, 1);
3134                         trans = btrfs_start_transaction(root, 1);
3135                         if (!trans)
3136                                 return -ENOMEM;
3137
3138                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3139                                              bytes + 2 * 1024 * 1024,
3140                                              alloc_target, 0);
3141                         btrfs_end_transaction(trans, root);
3142                         if (ret)
3143                                 return ret;
3144
3145                         if (!data_sinfo) {
3146                                 btrfs_set_inode_space_info(root, inode);
3147                                 data_sinfo = BTRFS_I(inode)->space_info;
3148                         }
3149                         goto again;
3150                 }
3151                 spin_unlock(&data_sinfo->lock);
3152
3153                 /* commit the current transaction and try again */
3154                 if (!committed && !root->fs_info->open_ioctl_trans) {
3155                         committed = 1;
3156                         trans = btrfs_join_transaction(root, 1);
3157                         if (!trans)
3158                                 return -ENOMEM;
3159                         ret = btrfs_commit_transaction(trans, root);
3160                         if (ret)
3161                                 return ret;
3162                         goto again;
3163                 }
3164
3165                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3166                        ", %llu bytes_used, %llu bytes_reserved, "
3167                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3168                        "%llu total\n", (unsigned long long)bytes,
3169                        (unsigned long long)data_sinfo->bytes_delalloc,
3170                        (unsigned long long)data_sinfo->bytes_used,
3171                        (unsigned long long)data_sinfo->bytes_reserved,
3172                        (unsigned long long)data_sinfo->bytes_pinned,
3173                        (unsigned long long)data_sinfo->bytes_readonly,
3174                        (unsigned long long)data_sinfo->bytes_may_use,
3175                        (unsigned long long)data_sinfo->total_bytes);
3176                 return -ENOSPC;
3177         }
3178         data_sinfo->bytes_may_use += bytes;
3179         BTRFS_I(inode)->reserved_bytes += bytes;
3180         spin_unlock(&data_sinfo->lock);
3181
3182         return 0;
3183 }
3184
3185 /*
3186  * if there was an error for whatever reason after calling
3187  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3188  */
3189 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3190                                     struct inode *inode, u64 bytes)
3191 {
3192         struct btrfs_space_info *data_sinfo;
3193
3194         /* make sure bytes are sectorsize aligned */
3195         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3196
3197         data_sinfo = BTRFS_I(inode)->space_info;
3198         spin_lock(&data_sinfo->lock);
3199         data_sinfo->bytes_may_use -= bytes;
3200         BTRFS_I(inode)->reserved_bytes -= bytes;
3201         spin_unlock(&data_sinfo->lock);
3202 }
3203
3204 /* called when we are adding a delalloc extent to the inode's io_tree */
3205 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3206                                   u64 bytes)
3207 {
3208         struct btrfs_space_info *data_sinfo;
3209
3210         /* get the space info for where this inode will be storing its data */
3211         data_sinfo = BTRFS_I(inode)->space_info;
3212
3213         /* make sure we have enough space to handle the data first */
3214         spin_lock(&data_sinfo->lock);
3215         data_sinfo->bytes_delalloc += bytes;
3216
3217         /*
3218          * we are adding a delalloc extent without calling
3219          * btrfs_check_data_free_space first.  This happens on a weird
3220          * writepage condition, but shouldn't hurt our accounting
3221          */
3222         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3223                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3224                 BTRFS_I(inode)->reserved_bytes = 0;
3225         } else {
3226                 data_sinfo->bytes_may_use -= bytes;
3227                 BTRFS_I(inode)->reserved_bytes -= bytes;
3228         }
3229
3230         spin_unlock(&data_sinfo->lock);
3231 }
3232
3233 /* called when we are clearing an delalloc extent from the inode's io_tree */
3234 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3235                               u64 bytes)
3236 {
3237         struct btrfs_space_info *info;
3238
3239         info = BTRFS_I(inode)->space_info;
3240
3241         spin_lock(&info->lock);
3242         info->bytes_delalloc -= bytes;
3243         spin_unlock(&info->lock);
3244 }
3245
3246 static void force_metadata_allocation(struct btrfs_fs_info *info)
3247 {
3248         struct list_head *head = &info->space_info;
3249         struct btrfs_space_info *found;
3250
3251         rcu_read_lock();
3252         list_for_each_entry_rcu(found, head, list) {
3253                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3254                         found->force_alloc = 1;
3255         }
3256         rcu_read_unlock();
3257 }
3258
3259 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3260                               u64 alloc_bytes)
3261 {
3262         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3263
3264         if (sinfo->bytes_used + sinfo->bytes_reserved +
3265             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3266                 return 0;
3267
3268         if (sinfo->bytes_used + sinfo->bytes_reserved +
3269             alloc_bytes < div_factor(num_bytes, 8))
3270                 return 0;
3271
3272         return 1;
3273 }
3274
3275 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3276                           struct btrfs_root *extent_root, u64 alloc_bytes,
3277                           u64 flags, int force)
3278 {
3279         struct btrfs_space_info *space_info;
3280         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3281         int ret = 0;
3282
3283         mutex_lock(&fs_info->chunk_mutex);
3284
3285         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3286
3287         space_info = __find_space_info(extent_root->fs_info, flags);
3288         if (!space_info) {
3289                 ret = update_space_info(extent_root->fs_info, flags,
3290                                         0, 0, &space_info);
3291                 BUG_ON(ret);
3292         }
3293         BUG_ON(!space_info);
3294
3295         spin_lock(&space_info->lock);
3296         if (space_info->force_alloc)
3297                 force = 1;
3298         if (space_info->full) {
3299                 spin_unlock(&space_info->lock);
3300                 goto out;
3301         }
3302
3303         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3304                 spin_unlock(&space_info->lock);
3305                 goto out;
3306         }
3307         spin_unlock(&space_info->lock);
3308
3309         /*
3310          * if we're doing a data chunk, go ahead and make sure that
3311          * we keep a reasonable number of metadata chunks allocated in the
3312          * FS as well.
3313          */
3314         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3315                 fs_info->data_chunk_allocations++;
3316                 if (!(fs_info->data_chunk_allocations %
3317                       fs_info->metadata_ratio))
3318                         force_metadata_allocation(fs_info);
3319         }
3320
3321         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3322         spin_lock(&space_info->lock);
3323         if (ret)
3324                 space_info->full = 1;
3325         else
3326                 ret = 1;
3327         space_info->force_alloc = 0;
3328         spin_unlock(&space_info->lock);
3329 out:
3330         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3331         return ret;
3332 }
3333
3334 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3335                                 struct btrfs_root *root,
3336                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3337 {
3338         int ret;
3339         int end_trans = 0;
3340
3341         if (sinfo->full)
3342                 return 0;
3343
3344         spin_lock(&sinfo->lock);
3345         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3346         spin_unlock(&sinfo->lock);
3347         if (!ret)
3348                 return 0;
3349
3350         if (!trans) {
3351                 trans = btrfs_join_transaction(root, 1);
3352                 BUG_ON(IS_ERR(trans));
3353                 end_trans = 1;
3354         }
3355
3356         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3357                              num_bytes + 2 * 1024 * 1024,
3358                              get_alloc_profile(root, sinfo->flags), 0);
3359
3360         if (end_trans)
3361                 btrfs_end_transaction(trans, root);
3362
3363         return ret == 1 ? 1 : 0;
3364 }
3365
3366 /*
3367  * shrink metadata reservation for delalloc
3368  */
3369 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3370                            struct btrfs_root *root,
3371                            struct btrfs_space_info *sinfo, u64 to_reclaim)
3372 {
3373         u64 reserved;
3374         u64 max_reclaim;
3375         u64 reclaimed = 0;
3376         int pause = 1;
3377         int ret;
3378
3379         spin_lock(&sinfo->lock);
3380         reserved = sinfo->bytes_delalloc;
3381         spin_unlock(&sinfo->lock);
3382
3383         if (reserved == 0)
3384                 return 0;
3385
3386         max_reclaim = min(reserved, to_reclaim);
3387
3388         while (1) {
3389                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3390                 if (!ret) {
3391                         __set_current_state(TASK_INTERRUPTIBLE);
3392                         schedule_timeout(pause);
3393                         pause <<= 1;
3394                         if (pause > HZ / 10)
3395                                 pause = HZ / 10;
3396                 } else {
3397                         pause = 1;
3398                 }
3399
3400                 spin_lock(&sinfo->lock);
3401                 if (reserved > sinfo->bytes_delalloc)
3402                         reclaimed = reserved - sinfo->bytes_delalloc;
3403                 reserved = sinfo->bytes_delalloc;
3404                 spin_unlock(&sinfo->lock);
3405
3406                 if (reserved == 0 || reclaimed >= max_reclaim)
3407                         break;
3408
3409                 if (trans && trans->transaction->blocked)
3410                         return -EAGAIN;
3411         }
3412         return reclaimed >= to_reclaim;
3413 }
3414
3415 static int update_block_group(struct btrfs_trans_handle *trans,
3416                               struct btrfs_root *root,
3417                               u64 bytenr, u64 num_bytes, int alloc,
3418                               int mark_free)
3419 {
3420         struct btrfs_block_group_cache *cache;
3421         struct btrfs_fs_info *info = root->fs_info;
3422         int factor;
3423         u64 total = num_bytes;
3424         u64 old_val;
3425         u64 byte_in_group;
3426
3427         /* block accounting for super block */
3428         spin_lock(&info->delalloc_lock);
3429         old_val = btrfs_super_bytes_used(&info->super_copy);
3430         if (alloc)
3431                 old_val += num_bytes;
3432         else
3433                 old_val -= num_bytes;
3434         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3435         spin_unlock(&info->delalloc_lock);
3436
3437         while (total) {
3438                 cache = btrfs_lookup_block_group(info, bytenr);
3439                 if (!cache)
3440                         return -1;
3441                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3442                                     BTRFS_BLOCK_GROUP_RAID1 |
3443                                     BTRFS_BLOCK_GROUP_RAID10))
3444                         factor = 2;
3445                 else
3446                         factor = 1;
3447                 byte_in_group = bytenr - cache->key.objectid;
3448                 WARN_ON(byte_in_group > cache->key.offset);
3449
3450                 spin_lock(&cache->space_info->lock);
3451                 spin_lock(&cache->lock);
3452                 cache->dirty = 1;
3453                 old_val = btrfs_block_group_used(&cache->item);
3454                 num_bytes = min(total, cache->key.offset - byte_in_group);
3455                 if (alloc) {
3456                         old_val += num_bytes;
3457                         btrfs_set_block_group_used(&cache->item, old_val);
3458                         cache->reserved -= num_bytes;
3459                         cache->space_info->bytes_reserved -= num_bytes;
3460                         cache->space_info->bytes_used += num_bytes;
3461                         cache->space_info->disk_used += num_bytes * factor;
3462                         if (cache->ro)
3463                                 cache->space_info->bytes_readonly -= num_bytes;
3464                         spin_unlock(&cache->lock);
3465                         spin_unlock(&cache->space_info->lock);
3466                 } else {
3467                         old_val -= num_bytes;
3468                         btrfs_set_block_group_used(&cache->item, old_val);
3469                         cache->space_info->bytes_used -= num_bytes;
3470                         cache->space_info->disk_used -= num_bytes * factor;
3471                         if (cache->ro)
3472                                 cache->space_info->bytes_readonly += num_bytes;
3473                         spin_unlock(&cache->lock);
3474                         spin_unlock(&cache->space_info->lock);
3475                         if (mark_free) {
3476                                 int ret;
3477
3478                                 ret = btrfs_discard_extent(root, bytenr,
3479                                                            num_bytes);
3480                                 WARN_ON(ret);
3481
3482                                 ret = btrfs_add_free_space(cache, bytenr,
3483                                                            num_bytes);
3484                                 WARN_ON(ret);
3485                         }
3486                 }
3487                 btrfs_put_block_group(cache);
3488                 total -= num_bytes;
3489                 bytenr += num_bytes;
3490         }
3491         return 0;
3492 }
3493
3494 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3495 {
3496         struct btrfs_block_group_cache *cache;
3497         u64 bytenr;
3498
3499         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3500         if (!cache)
3501                 return 0;
3502
3503         bytenr = cache->key.objectid;
3504         btrfs_put_block_group(cache);
3505
3506         return bytenr;
3507 }
3508
3509 /*
3510  * this function must be called within transaction
3511  */
3512 int btrfs_pin_extent(struct btrfs_root *root,
3513                      u64 bytenr, u64 num_bytes, int reserved)
3514 {
3515         struct btrfs_fs_info *fs_info = root->fs_info;
3516         struct btrfs_block_group_cache *cache;
3517
3518         cache = btrfs_lookup_block_group(fs_info, bytenr);
3519         BUG_ON(!cache);
3520
3521         spin_lock(&cache->space_info->lock);
3522         spin_lock(&cache->lock);
3523         cache->pinned += num_bytes;
3524         cache->space_info->bytes_pinned += num_bytes;
3525         if (reserved) {
3526                 cache->reserved -= num_bytes;
3527                 cache->space_info->bytes_reserved -= num_bytes;
3528         }
3529         spin_unlock(&cache->lock);
3530         spin_unlock(&cache->space_info->lock);
3531
3532         btrfs_put_block_group(cache);
3533
3534         set_extent_dirty(fs_info->pinned_extents,
3535                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3536         return 0;
3537 }
3538
3539 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3540                                    u64 num_bytes, int reserve)
3541 {
3542         spin_lock(&cache->space_info->lock);
3543         spin_lock(&cache->lock);
3544         if (reserve) {
3545                 cache->reserved += num_bytes;
3546                 cache->space_info->bytes_reserved += num_bytes;
3547         } else {
3548                 cache->reserved -= num_bytes;
3549                 cache->space_info->bytes_reserved -= num_bytes;
3550         }
3551         spin_unlock(&cache->lock);
3552         spin_unlock(&cache->space_info->lock);
3553         return 0;
3554 }
3555
3556 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3557                                 struct btrfs_root *root)
3558 {
3559         struct btrfs_fs_info *fs_info = root->fs_info;
3560         struct btrfs_caching_control *next;
3561         struct btrfs_caching_control *caching_ctl;
3562         struct btrfs_block_group_cache *cache;
3563
3564         down_write(&fs_info->extent_commit_sem);
3565
3566         list_for_each_entry_safe(caching_ctl, next,
3567                                  &fs_info->caching_block_groups, list) {
3568                 cache = caching_ctl->block_group;
3569                 if (block_group_cache_done(cache)) {
3570                         cache->last_byte_to_unpin = (u64)-1;
3571                         list_del_init(&caching_ctl->list);
3572                         put_caching_control(caching_ctl);
3573                 } else {
3574                         cache->last_byte_to_unpin = caching_ctl->progress;
3575                 }
3576         }
3577
3578         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3579                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3580         else
3581                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3582
3583         up_write(&fs_info->extent_commit_sem);
3584         return 0;
3585 }
3586
3587 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3588 {
3589         struct btrfs_fs_info *fs_info = root->fs_info;
3590         struct btrfs_block_group_cache *cache = NULL;
3591         u64 len;
3592
3593         while (start <= end) {
3594                 if (!cache ||
3595                     start >= cache->key.objectid + cache->key.offset) {
3596                         if (cache)
3597                                 btrfs_put_block_group(cache);
3598                         cache = btrfs_lookup_block_group(fs_info, start);
3599                         BUG_ON(!cache);
3600                 }
3601
3602                 len = cache->key.objectid + cache->key.offset - start;
3603                 len = min(len, end + 1 - start);
3604
3605                 if (start < cache->last_byte_to_unpin) {
3606                         len = min(len, cache->last_byte_to_unpin - start);
3607                         btrfs_add_free_space(cache, start, len);
3608                 }
3609
3610                 spin_lock(&cache->space_info->lock);
3611                 spin_lock(&cache->lock);
3612                 cache->pinned -= len;
3613                 cache->space_info->bytes_pinned -= len;
3614                 spin_unlock(&cache->lock);
3615                 spin_unlock(&cache->space_info->lock);
3616
3617                 start += len;
3618         }
3619
3620         if (cache)
3621                 btrfs_put_block_group(cache);
3622         return 0;
3623 }
3624
3625 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3626                                struct btrfs_root *root)
3627 {
3628         struct btrfs_fs_info *fs_info = root->fs_info;
3629         struct extent_io_tree *unpin;
3630         u64 start;
3631         u64 end;
3632         int ret;
3633
3634         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3635                 unpin = &fs_info->freed_extents[1];
3636         else
3637                 unpin = &fs_info->freed_extents[0];
3638
3639         while (1) {
3640                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3641                                             EXTENT_DIRTY);
3642                 if (ret)
3643                         break;
3644
3645                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3646
3647                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3648                 unpin_extent_range(root, start, end);
3649                 cond_resched();
3650         }
3651
3652         return ret;
3653 }
3654
3655 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3656                           struct btrfs_root *root,
3657                           struct btrfs_path *path,
3658                           u64 bytenr, u64 num_bytes,
3659                           int is_data, int reserved,
3660                           struct extent_buffer **must_clean)
3661 {
3662         int err = 0;
3663         struct extent_buffer *buf;
3664
3665         if (is_data)
3666                 goto pinit;
3667
3668         /*
3669          * discard is sloooow, and so triggering discards on
3670          * individual btree blocks isn't a good plan.  Just
3671          * pin everything in discard mode.
3672          */
3673         if (btrfs_test_opt(root, DISCARD))
3674                 goto pinit;
3675
3676         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3677         if (!buf)
3678                 goto pinit;
3679
3680         /* we can reuse a block if it hasn't been written
3681          * and it is from this transaction.  We can't
3682          * reuse anything from the tree log root because
3683          * it has tiny sub-transactions.
3684          */
3685         if (btrfs_buffer_uptodate(buf, 0) &&
3686             btrfs_try_tree_lock(buf)) {
3687                 u64 header_owner = btrfs_header_owner(buf);
3688                 u64 header_transid = btrfs_header_generation(buf);
3689                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3690                     header_transid == trans->transid &&
3691                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3692                         *must_clean = buf;
3693                         return 1;
3694                 }
3695                 btrfs_tree_unlock(buf);
3696         }
3697         free_extent_buffer(buf);
3698 pinit:
3699         if (path)
3700                 btrfs_set_path_blocking(path);
3701         /* unlocks the pinned mutex */
3702         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3703
3704         BUG_ON(err < 0);
3705         return 0;
3706 }
3707
3708 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3709                                 struct btrfs_root *root,
3710                                 u64 bytenr, u64 num_bytes, u64 parent,
3711                                 u64 root_objectid, u64 owner_objectid,
3712                                 u64 owner_offset, int refs_to_drop,
3713                                 struct btrfs_delayed_extent_op *extent_op)
3714 {
3715         struct btrfs_key key;
3716         struct btrfs_path *path;
3717         struct btrfs_fs_info *info = root->fs_info;
3718         struct btrfs_root *extent_root = info->extent_root;
3719         struct extent_buffer *leaf;
3720         struct btrfs_extent_item *ei;
3721         struct btrfs_extent_inline_ref *iref;
3722         int ret;
3723         int is_data;
3724         int extent_slot = 0;
3725         int found_extent = 0;
3726         int num_to_del = 1;
3727         u32 item_size;
3728         u64 refs;
3729
3730         path = btrfs_alloc_path();
3731         if (!path)
3732                 return -ENOMEM;
3733
3734         path->reada = 1;
3735         path->leave_spinning = 1;
3736
3737         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3738         BUG_ON(!is_data && refs_to_drop != 1);
3739
3740         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3741                                     bytenr, num_bytes, parent,
3742                                     root_objectid, owner_objectid,
3743                                     owner_offset);
3744         if (ret == 0) {
3745                 extent_slot = path->slots[0];
3746                 while (extent_slot >= 0) {
3747                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3748                                               extent_slot);
3749                         if (key.objectid != bytenr)
3750                                 break;
3751                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3752                             key.offset == num_bytes) {
3753                                 found_extent = 1;
3754                                 break;
3755                         }
3756                         if (path->slots[0] - extent_slot > 5)
3757                                 break;
3758                         extent_slot--;
3759                 }
3760 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3761                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3762                 if (found_extent && item_size < sizeof(*ei))
3763                         found_extent = 0;
3764 #endif
3765                 if (!found_extent) {
3766                         BUG_ON(iref);
3767                         ret = remove_extent_backref(trans, extent_root, path,
3768                                                     NULL, refs_to_drop,
3769                                                     is_data);
3770                         BUG_ON(ret);
3771                         btrfs_release_path(extent_root, path);
3772                         path->leave_spinning = 1;
3773
3774                         key.objectid = bytenr;
3775                         key.type = BTRFS_EXTENT_ITEM_KEY;
3776                         key.offset = num_bytes;
3777
3778                         ret = btrfs_search_slot(trans, extent_root,
3779                                                 &key, path, -1, 1);
3780                         if (ret) {
3781                                 printk(KERN_ERR "umm, got %d back from search"
3782                                        ", was looking for %llu\n", ret,
3783                                        (unsigned long long)bytenr);
3784                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3785                         }
3786                         BUG_ON(ret);
3787                         extent_slot = path->slots[0];
3788                 }
3789         } else {
3790                 btrfs_print_leaf(extent_root, path->nodes[0]);
3791                 WARN_ON(1);
3792                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3793                        "parent %llu root %llu  owner %llu offset %llu\n",
3794                        (unsigned long long)bytenr,
3795                        (unsigned long long)parent,
3796                        (unsigned long long)root_objectid,
3797                        (unsigned long long)owner_objectid,
3798                        (unsigned long long)owner_offset);
3799         }
3800
3801         leaf = path->nodes[0];
3802         item_size = btrfs_item_size_nr(leaf, extent_slot);
3803 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3804         if (item_size < sizeof(*ei)) {
3805                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3806                 ret = convert_extent_item_v0(trans, extent_root, path,
3807                                              owner_objectid, 0);
3808                 BUG_ON(ret < 0);
3809
3810                 btrfs_release_path(extent_root, path);
3811                 path->leave_spinning = 1;
3812
3813                 key.objectid = bytenr;
3814                 key.type = BTRFS_EXTENT_ITEM_KEY;
3815                 key.offset = num_bytes;
3816
3817                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3818                                         -1, 1);
3819                 if (ret) {
3820                         printk(KERN_ERR "umm, got %d back from search"
3821                                ", was looking for %llu\n", ret,
3822                                (unsigned long long)bytenr);
3823                         btrfs_print_leaf(extent_root, path->nodes[0]);
3824                 }
3825                 BUG_ON(ret);
3826                 extent_slot = path->slots[0];
3827                 leaf = path->nodes[0];
3828                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3829         }
3830 #endif
3831         BUG_ON(item_size < sizeof(*ei));
3832         ei = btrfs_item_ptr(leaf, extent_slot,
3833                             struct btrfs_extent_item);
3834         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3835                 struct btrfs_tree_block_info *bi;
3836                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3837                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3838                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3839         }
3840
3841         refs = btrfs_extent_refs(leaf, ei);
3842         BUG_ON(refs < refs_to_drop);
3843         refs -= refs_to_drop;
3844
3845         if (refs > 0) {
3846                 if (extent_op)
3847                         __run_delayed_extent_op(extent_op, leaf, ei);
3848                 /*
3849                  * In the case of inline back ref, reference count will
3850                  * be updated by remove_extent_backref
3851                  */
3852                 if (iref) {
3853                         BUG_ON(!found_extent);
3854                 } else {
3855                         btrfs_set_extent_refs(leaf, ei, refs);
3856                         btrfs_mark_buffer_dirty(leaf);
3857                 }
3858                 if (found_extent) {
3859                         ret = remove_extent_backref(trans, extent_root, path,
3860                                                     iref, refs_to_drop,
3861                                                     is_data);
3862                         BUG_ON(ret);
3863                 }
3864         } else {
3865                 int mark_free = 0;
3866                 struct extent_buffer *must_clean = NULL;
3867
3868                 if (found_extent) {
3869                         BUG_ON(is_data && refs_to_drop !=
3870                                extent_data_ref_count(root, path, iref));
3871                         if (iref) {
3872                                 BUG_ON(path->slots[0] != extent_slot);
3873                         } else {
3874                                 BUG_ON(path->slots[0] != extent_slot + 1);
3875                                 path->slots[0] = extent_slot;
3876                                 num_to_del = 2;
3877                         }
3878                 }
3879
3880                 ret = pin_down_bytes(trans, root, path, bytenr,
3881                                      num_bytes, is_data, 0, &must_clean);
3882                 if (ret > 0)
3883                         mark_free = 1;
3884                 BUG_ON(ret < 0);
3885                 /*
3886                  * it is going to be very rare for someone to be waiting
3887                  * on the block we're freeing.  del_items might need to
3888                  * schedule, so rather than get fancy, just force it
3889                  * to blocking here
3890                  */
3891                 if (must_clean)
3892                         btrfs_set_lock_blocking(must_clean);
3893
3894                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3895                                       num_to_del);
3896                 BUG_ON(ret);
3897                 btrfs_release_path(extent_root, path);
3898
3899                 if (must_clean) {
3900                         clean_tree_block(NULL, root, must_clean);
3901                         btrfs_tree_unlock(must_clean);
3902                         free_extent_buffer(must_clean);
3903                 }
3904
3905                 if (is_data) {
3906                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3907                         BUG_ON(ret);
3908                 } else {
3909                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3910                              bytenr >> PAGE_CACHE_SHIFT,
3911                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3912                 }
3913
3914                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3915                                          mark_free);
3916                 BUG_ON(ret);
3917         }
3918         btrfs_free_path(path);
3919         return ret;
3920 }
3921
3922 /*
3923  * when we free an extent, it is possible (and likely) that we free the last
3924  * delayed ref for that extent as well.  This searches the delayed ref tree for
3925  * a given extent, and if there are no other delayed refs to be processed, it
3926  * removes it from the tree.
3927  */
3928 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3929                                       struct btrfs_root *root, u64 bytenr)
3930 {
3931         struct btrfs_delayed_ref_head *head;
3932         struct btrfs_delayed_ref_root *delayed_refs;
3933         struct btrfs_delayed_ref_node *ref;
3934         struct rb_node *node;
3935         int ret;
3936
3937         delayed_refs = &trans->transaction->delayed_refs;
3938         spin_lock(&delayed_refs->lock);
3939         head = btrfs_find_delayed_ref_head(trans, bytenr);
3940         if (!head)
3941                 goto out;
3942
3943         node = rb_prev(&head->node.rb_node);
3944         if (!node)
3945                 goto out;
3946
3947         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3948
3949         /* there are still entries for this ref, we can't drop it */
3950         if (ref->bytenr == bytenr)
3951                 goto out;
3952
3953         if (head->extent_op) {
3954                 if (!head->must_insert_reserved)
3955                         goto out;
3956                 kfree(head->extent_op);
3957                 head->extent_op = NULL;
3958         }
3959
3960         /*
3961          * waiting for the lock here would deadlock.  If someone else has it
3962          * locked they are already in the process of dropping it anyway
3963          */
3964         if (!mutex_trylock(&head->mutex))
3965                 goto out;
3966
3967         /*
3968          * at this point we have a head with no other entries.  Go
3969          * ahead and process it.
3970          */
3971         head->node.in_tree = 0;
3972         rb_erase(&head->node.rb_node, &delayed_refs->root);
3973
3974         delayed_refs->num_entries--;
3975
3976         /*
3977          * we don't take a ref on the node because we're removing it from the
3978          * tree, so we just steal the ref the tree was holding.
3979          */
3980         delayed_refs->num_heads--;
3981         if (list_empty(&head->cluster))
3982                 delayed_refs->num_heads_ready--;
3983
3984         list_del_init(&head->cluster);
3985         spin_unlock(&delayed_refs->lock);
3986
3987         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3988                                   &head->node, head->extent_op,
3989                                   head->must_insert_reserved);
3990         BUG_ON(ret);
3991         btrfs_put_delayed_ref(&head->node);
3992         return 0;
3993 out:
3994         spin_unlock(&delayed_refs->lock);
3995         return 0;
3996 }
3997
3998 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3999                       struct btrfs_root *root,
4000                       u64 bytenr, u64 num_bytes, u64 parent,
4001                       u64 root_objectid, u64 owner, u64 offset)
4002 {
4003         int ret;
4004
4005         /*
4006          * tree log blocks never actually go into the extent allocation
4007          * tree, just update pinning info and exit early.
4008          */
4009         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4010                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4011                 /* unlocks the pinned mutex */
4012                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4013                 ret = 0;
4014         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4015                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4016                                         parent, root_objectid, (int)owner,
4017                                         BTRFS_DROP_DELAYED_REF, NULL);
4018                 BUG_ON(ret);
4019                 ret = check_ref_cleanup(trans, root, bytenr);
4020                 BUG_ON(ret);
4021         } else {
4022                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4023                                         parent, root_objectid, owner,
4024                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4025                 BUG_ON(ret);
4026         }
4027         return ret;
4028 }
4029
4030 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4031                           struct btrfs_root *root,
4032                           u64 bytenr, u32 blocksize,
4033                           u64 parent, u64 root_objectid, int level)
4034 {
4035         u64 used;
4036         spin_lock(&root->node_lock);
4037         used = btrfs_root_used(&root->root_item) - blocksize;
4038         btrfs_set_root_used(&root->root_item, used);
4039         spin_unlock(&root->node_lock);
4040
4041         return btrfs_free_extent(trans, root, bytenr, blocksize,
4042                                  parent, root_objectid, level, 0);
4043 }
4044
4045 static u64 stripe_align(struct btrfs_root *root, u64 val)
4046 {
4047         u64 mask = ((u64)root->stripesize - 1);
4048         u64 ret = (val + mask) & ~mask;
4049         return ret;
4050 }
4051
4052 /*
4053  * when we wait for progress in the block group caching, its because
4054  * our allocation attempt failed at least once.  So, we must sleep
4055  * and let some progress happen before we try again.
4056  *
4057  * This function will sleep at least once waiting for new free space to
4058  * show up, and then it will check the block group free space numbers
4059  * for our min num_bytes.  Another option is to have it go ahead
4060  * and look in the rbtree for a free extent of a given size, but this
4061  * is a good start.
4062  */
4063 static noinline int
4064 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4065                                 u64 num_bytes)
4066 {
4067         struct btrfs_caching_control *caching_ctl;
4068         DEFINE_WAIT(wait);
4069
4070         caching_ctl = get_caching_control(cache);
4071         if (!caching_ctl)
4072                 return 0;
4073
4074         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4075                    (cache->free_space >= num_bytes));
4076
4077         put_caching_control(caching_ctl);
4078         return 0;
4079 }
4080
4081 static noinline int
4082 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4083 {
4084         struct btrfs_caching_control *caching_ctl;
4085         DEFINE_WAIT(wait);
4086
4087         caching_ctl = get_caching_control(cache);
4088         if (!caching_ctl)
4089                 return 0;
4090
4091         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4092
4093         put_caching_control(caching_ctl);
4094         return 0;
4095 }
4096
4097 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4098 {
4099         int index;
4100         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4101                 index = 0;
4102         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4103                 index = 1;
4104         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4105                 index = 2;
4106         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4107                 index = 3;
4108         else
4109                 index = 4;
4110         return index;
4111 }
4112
4113 enum btrfs_loop_type {
4114         LOOP_FIND_IDEAL = 0,
4115         LOOP_CACHING_NOWAIT = 1,
4116         LOOP_CACHING_WAIT = 2,
4117         LOOP_ALLOC_CHUNK = 3,
4118         LOOP_NO_EMPTY_SIZE = 4,
4119 };
4120
4121 /*
4122  * walks the btree of allocated extents and find a hole of a given size.
4123  * The key ins is changed to record the hole:
4124  * ins->objectid == block start
4125  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4126  * ins->offset == number of blocks
4127  * Any available blocks before search_start are skipped.
4128  */
4129 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4130                                      struct btrfs_root *orig_root,
4131                                      u64 num_bytes, u64 empty_size,
4132                                      u64 search_start, u64 search_end,
4133                                      u64 hint_byte, struct btrfs_key *ins,
4134                                      u64 exclude_start, u64 exclude_nr,
4135                                      int data)
4136 {
4137         int ret = 0;
4138         struct btrfs_root *root = orig_root->fs_info->extent_root;
4139         struct btrfs_free_cluster *last_ptr = NULL;
4140         struct btrfs_block_group_cache *block_group = NULL;
4141         int empty_cluster = 2 * 1024 * 1024;
4142         int allowed_chunk_alloc = 0;
4143         int done_chunk_alloc = 0;
4144         struct btrfs_space_info *space_info;
4145         int last_ptr_loop = 0;
4146         int index = 0;
4147         int loop = 0;
4148         bool found_uncached_bg = false;
4149         bool failed_cluster_refill = false;
4150         bool failed_alloc = false;
4151         u64 ideal_cache_percent = 0;
4152         u64 ideal_cache_offset = 0;
4153
4154         WARN_ON(num_bytes < root->sectorsize);
4155         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4156         ins->objectid = 0;
4157         ins->offset = 0;
4158
4159         space_info = __find_space_info(root->fs_info, data);
4160         if (!space_info) {
4161                 printk(KERN_ERR "No space info for %d\n", data);
4162                 return -ENOSPC;
4163         }
4164
4165         if (orig_root->ref_cows || empty_size)
4166                 allowed_chunk_alloc = 1;
4167
4168         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4169                 last_ptr = &root->fs_info->meta_alloc_cluster;
4170                 if (!btrfs_test_opt(root, SSD))
4171                         empty_cluster = 64 * 1024;
4172         }
4173
4174         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4175                 last_ptr = &root->fs_info->data_alloc_cluster;
4176         }
4177
4178         if (last_ptr) {
4179                 spin_lock(&last_ptr->lock);
4180                 if (last_ptr->block_group)
4181                         hint_byte = last_ptr->window_start;
4182                 spin_unlock(&last_ptr->lock);
4183         }
4184
4185         search_start = max(search_start, first_logical_byte(root, 0));
4186         search_start = max(search_start, hint_byte);
4187
4188         if (!last_ptr)
4189                 empty_cluster = 0;
4190
4191         if (search_start == hint_byte) {
4192 ideal_cache:
4193                 block_group = btrfs_lookup_block_group(root->fs_info,
4194                                                        search_start);
4195                 /*
4196                  * we don't want to use the block group if it doesn't match our
4197                  * allocation bits, or if its not cached.
4198                  *
4199                  * However if we are re-searching with an ideal block group
4200                  * picked out then we don't care that the block group is cached.
4201                  */
4202                 if (block_group && block_group_bits(block_group, data) &&
4203                     (block_group->cached != BTRFS_CACHE_NO ||
4204                      search_start == ideal_cache_offset)) {
4205                         down_read(&space_info->groups_sem);
4206                         if (list_empty(&block_group->list) ||
4207                             block_group->ro) {
4208                                 /*
4209                                  * someone is removing this block group,
4210                                  * we can't jump into the have_block_group
4211                                  * target because our list pointers are not
4212                                  * valid
4213                                  */
4214                                 btrfs_put_block_group(block_group);
4215                                 up_read(&space_info->groups_sem);
4216                         } else {
4217                                 index = get_block_group_index(block_group);
4218                                 goto have_block_group;
4219                         }
4220                 } else if (block_group) {
4221                         btrfs_put_block_group(block_group);
4222                 }
4223         }
4224 search:
4225         down_read(&space_info->groups_sem);
4226         list_for_each_entry(block_group, &space_info->block_groups[index],
4227                             list) {
4228                 u64 offset;
4229                 int cached;
4230
4231                 btrfs_get_block_group(block_group);
4232                 search_start = block_group->key.objectid;
4233
4234 have_block_group:
4235                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4236                         u64 free_percent;
4237
4238                         free_percent = btrfs_block_group_used(&block_group->item);
4239                         free_percent *= 100;
4240                         free_percent = div64_u64(free_percent,
4241                                                  block_group->key.offset);
4242                         free_percent = 100 - free_percent;
4243                         if (free_percent > ideal_cache_percent &&
4244                             likely(!block_group->ro)) {
4245                                 ideal_cache_offset = block_group->key.objectid;
4246                                 ideal_cache_percent = free_percent;
4247                         }
4248
4249                         /*
4250                          * We only want to start kthread caching if we are at
4251                          * the point where we will wait for caching to make
4252                          * progress, or if our ideal search is over and we've
4253                          * found somebody to start caching.
4254                          */
4255                         if (loop > LOOP_CACHING_NOWAIT ||
4256                             (loop > LOOP_FIND_IDEAL &&
4257                              atomic_read(&space_info->caching_threads) < 2)) {
4258                                 ret = cache_block_group(block_group);
4259                                 BUG_ON(ret);
4260                         }
4261                         found_uncached_bg = true;
4262
4263                         /*
4264                          * If loop is set for cached only, try the next block
4265                          * group.
4266                          */
4267                         if (loop == LOOP_FIND_IDEAL)
4268                                 goto loop;
4269                 }
4270
4271                 cached = block_group_cache_done(block_group);
4272                 if (unlikely(!cached))
4273                         found_uncached_bg = true;
4274
4275                 if (unlikely(block_group->ro))
4276                         goto loop;
4277
4278                 /*
4279                  * Ok we want to try and use the cluster allocator, so lets look
4280                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4281                  * have tried the cluster allocator plenty of times at this
4282                  * point and not have found anything, so we are likely way too
4283                  * fragmented for the clustering stuff to find anything, so lets
4284                  * just skip it and let the allocator find whatever block it can
4285                  * find
4286                  */
4287                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4288                         /*
4289                          * the refill lock keeps out other
4290                          * people trying to start a new cluster
4291                          */
4292                         spin_lock(&last_ptr->refill_lock);
4293                         if (last_ptr->block_group &&
4294                             (last_ptr->block_group->ro ||
4295                             !block_group_bits(last_ptr->block_group, data))) {
4296                                 offset = 0;
4297                                 goto refill_cluster;
4298                         }
4299
4300                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4301                                                  num_bytes, search_start);
4302                         if (offset) {
4303                                 /* we have a block, we're done */
4304                                 spin_unlock(&last_ptr->refill_lock);
4305                                 goto checks;
4306                         }
4307
4308                         spin_lock(&last_ptr->lock);
4309                         /*
4310                          * whoops, this cluster doesn't actually point to
4311                          * this block group.  Get a ref on the block
4312                          * group is does point to and try again
4313                          */
4314                         if (!last_ptr_loop && last_ptr->block_group &&
4315                             last_ptr->block_group != block_group) {
4316
4317                                 btrfs_put_block_group(block_group);
4318                                 block_group = last_ptr->block_group;
4319                                 btrfs_get_block_group(block_group);
4320                                 spin_unlock(&last_ptr->lock);
4321                                 spin_unlock(&last_ptr->refill_lock);
4322
4323                                 last_ptr_loop = 1;
4324                                 search_start = block_group->key.objectid;
4325                                 /*
4326                                  * we know this block group is properly
4327                                  * in the list because
4328                                  * btrfs_remove_block_group, drops the
4329                                  * cluster before it removes the block
4330                                  * group from the list
4331                                  */
4332                                 goto have_block_group;
4333                         }
4334                         spin_unlock(&last_ptr->lock);
4335 refill_cluster:
4336                         /*
4337                          * this cluster didn't work out, free it and
4338                          * start over
4339                          */
4340                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4341
4342                         last_ptr_loop = 0;
4343
4344                         /* allocate a cluster in this block group */
4345                         ret = btrfs_find_space_cluster(trans, root,
4346                                                block_group, last_ptr,
4347                                                offset, num_bytes,
4348                                                empty_cluster + empty_size);
4349                         if (ret == 0) {
4350                                 /*
4351                                  * now pull our allocation out of this
4352                                  * cluster
4353                                  */
4354                                 offset = btrfs_alloc_from_cluster(block_group,
4355                                                   last_ptr, num_bytes,
4356                                                   search_start);
4357                                 if (offset) {
4358                                         /* we found one, proceed */
4359                                         spin_unlock(&last_ptr->refill_lock);
4360                                         goto checks;
4361                                 }
4362                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4363                                    && !failed_cluster_refill) {
4364                                 spin_unlock(&last_ptr->refill_lock);
4365
4366                                 failed_cluster_refill = true;
4367                                 wait_block_group_cache_progress(block_group,
4368                                        num_bytes + empty_cluster + empty_size);
4369                                 goto have_block_group;
4370                         }
4371
4372                         /*
4373                          * at this point we either didn't find a cluster
4374                          * or we weren't able to allocate a block from our
4375                          * cluster.  Free the cluster we've been trying
4376                          * to use, and go to the next block group
4377                          */
4378                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4379                         spin_unlock(&last_ptr->refill_lock);
4380                         goto loop;
4381                 }
4382
4383                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4384                                                     num_bytes, empty_size);
4385                 /*
4386                  * If we didn't find a chunk, and we haven't failed on this
4387                  * block group before, and this block group is in the middle of
4388                  * caching and we are ok with waiting, then go ahead and wait
4389                  * for progress to be made, and set failed_alloc to true.
4390                  *
4391                  * If failed_alloc is true then we've already waited on this
4392                  * block group once and should move on to the next block group.
4393                  */
4394                 if (!offset && !failed_alloc && !cached &&
4395                     loop > LOOP_CACHING_NOWAIT) {
4396                         wait_block_group_cache_progress(block_group,
4397                                                 num_bytes + empty_size);
4398                         failed_alloc = true;
4399                         goto have_block_group;
4400                 } else if (!offset) {
4401                         goto loop;
4402                 }
4403 checks:
4404                 search_start = stripe_align(root, offset);
4405                 /* move on to the next group */
4406                 if (search_start + num_bytes >= search_end) {
4407                         btrfs_add_free_space(block_group, offset, num_bytes);
4408                         goto loop;
4409                 }
4410
4411                 /* move on to the next group */
4412                 if (search_start + num_bytes >
4413                     block_group->key.objectid + block_group->key.offset) {
4414                         btrfs_add_free_space(block_group, offset, num_bytes);
4415                         goto loop;
4416                 }
4417
4418                 if (exclude_nr > 0 &&
4419                     (search_start + num_bytes > exclude_start &&
4420                      search_start < exclude_start + exclude_nr)) {
4421                         search_start = exclude_start + exclude_nr;
4422
4423                         btrfs_add_free_space(block_group, offset, num_bytes);
4424                         /*
4425                          * if search_start is still in this block group
4426                          * then we just re-search this block group
4427                          */
4428                         if (search_start >= block_group->key.objectid &&
4429                             search_start < (block_group->key.objectid +
4430                                             block_group->key.offset))
4431                                 goto have_block_group;
4432                         goto loop;
4433                 }
4434
4435                 ins->objectid = search_start;
4436                 ins->offset = num_bytes;
4437
4438                 if (offset < search_start)
4439                         btrfs_add_free_space(block_group, offset,
4440                                              search_start - offset);
4441                 BUG_ON(offset > search_start);
4442
4443                 update_reserved_extents(block_group, num_bytes, 1);
4444
4445                 /* we are all good, lets return */
4446                 break;
4447 loop:
4448                 failed_cluster_refill = false;
4449                 failed_alloc = false;
4450                 BUG_ON(index != get_block_group_index(block_group));
4451                 btrfs_put_block_group(block_group);
4452         }
4453         up_read(&space_info->groups_sem);
4454
4455         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4456                 goto search;
4457
4458         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4459          *                      for them to make caching progress.  Also
4460          *                      determine the best possible bg to cache
4461          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4462          *                      caching kthreads as we move along
4463          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4464          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4465          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4466          *                      again
4467          */
4468         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4469             (found_uncached_bg || empty_size || empty_cluster ||
4470              allowed_chunk_alloc)) {
4471                 index = 0;
4472                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4473                         found_uncached_bg = false;
4474                         loop++;
4475                         if (!ideal_cache_percent &&
4476                             atomic_read(&space_info->caching_threads))
4477                                 goto search;
4478
4479                         /*
4480                          * 1 of the following 2 things have happened so far
4481                          *
4482                          * 1) We found an ideal block group for caching that
4483                          * is mostly full and will cache quickly, so we might
4484                          * as well wait for it.
4485                          *
4486                          * 2) We searched for cached only and we didn't find
4487                          * anything, and we didn't start any caching kthreads
4488                          * either, so chances are we will loop through and
4489                          * start a couple caching kthreads, and then come back
4490                          * around and just wait for them.  This will be slower
4491                          * because we will have 2 caching kthreads reading at
4492                          * the same time when we could have just started one
4493                          * and waited for it to get far enough to give us an
4494                          * allocation, so go ahead and go to the wait caching
4495                          * loop.
4496                          */
4497                         loop = LOOP_CACHING_WAIT;
4498                         search_start = ideal_cache_offset;
4499                         ideal_cache_percent = 0;
4500                         goto ideal_cache;
4501                 } else if (loop == LOOP_FIND_IDEAL) {
4502                         /*
4503                          * Didn't find a uncached bg, wait on anything we find
4504                          * next.
4505                          */
4506                         loop = LOOP_CACHING_WAIT;
4507                         goto search;
4508                 }
4509
4510                 if (loop < LOOP_CACHING_WAIT) {
4511                         loop++;
4512                         goto search;
4513                 }
4514
4515                 if (loop == LOOP_ALLOC_CHUNK) {
4516                         empty_size = 0;
4517                         empty_cluster = 0;
4518                 }
4519
4520                 if (allowed_chunk_alloc) {
4521                         ret = do_chunk_alloc(trans, root, num_bytes +
4522                                              2 * 1024 * 1024, data, 1);
4523                         allowed_chunk_alloc = 0;
4524                         done_chunk_alloc = 1;
4525                 } else if (!done_chunk_alloc) {
4526                         space_info->force_alloc = 1;
4527                 }
4528
4529                 if (loop < LOOP_NO_EMPTY_SIZE) {
4530                         loop++;
4531                         goto search;
4532                 }
4533                 ret = -ENOSPC;
4534         } else if (!ins->objectid) {
4535                 ret = -ENOSPC;
4536         }
4537
4538         /* we found what we needed */
4539         if (ins->objectid) {
4540                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4541                         trans->block_group = block_group->key.objectid;
4542
4543                 btrfs_put_block_group(block_group);
4544                 ret = 0;
4545         }
4546
4547         return ret;
4548 }
4549
4550 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4551                             int dump_block_groups)
4552 {
4553         struct btrfs_block_group_cache *cache;
4554         int index = 0;
4555
4556         spin_lock(&info->lock);
4557         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4558                (unsigned long long)(info->total_bytes - info->bytes_used -
4559                                     info->bytes_pinned - info->bytes_reserved -
4560                                     info->bytes_super),
4561                (info->full) ? "" : "not ");
4562         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4563                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4564                "\n",
4565                (unsigned long long)info->total_bytes,
4566                (unsigned long long)info->bytes_pinned,
4567                (unsigned long long)info->bytes_delalloc,
4568                (unsigned long long)info->bytes_may_use,
4569                (unsigned long long)info->bytes_used,
4570                (unsigned long long)info->bytes_root,
4571                (unsigned long long)info->bytes_super,
4572                (unsigned long long)info->bytes_reserved);
4573         spin_unlock(&info->lock);
4574
4575         if (!dump_block_groups)
4576                 return;
4577
4578         down_read(&info->groups_sem);
4579 again:
4580         list_for_each_entry(cache, &info->block_groups[index], list) {
4581                 spin_lock(&cache->lock);
4582                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4583                        "%llu pinned %llu reserved\n",
4584                        (unsigned long long)cache->key.objectid,
4585                        (unsigned long long)cache->key.offset,
4586                        (unsigned long long)btrfs_block_group_used(&cache->item),
4587                        (unsigned long long)cache->pinned,
4588                        (unsigned long long)cache->reserved);
4589                 btrfs_dump_free_space(cache, bytes);
4590                 spin_unlock(&cache->lock);
4591         }
4592         if (++index < BTRFS_NR_RAID_TYPES)
4593                 goto again;
4594         up_read(&info->groups_sem);
4595 }
4596
4597 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4598                          struct btrfs_root *root,
4599                          u64 num_bytes, u64 min_alloc_size,
4600                          u64 empty_size, u64 hint_byte,
4601                          u64 search_end, struct btrfs_key *ins,
4602                          u64 data)
4603 {
4604         int ret;
4605         u64 search_start = 0;
4606
4607         data = btrfs_get_alloc_profile(root, data);
4608 again:
4609         /*
4610          * the only place that sets empty_size is btrfs_realloc_node, which
4611          * is not called recursively on allocations
4612          */
4613         if (empty_size || root->ref_cows)
4614                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4615                                      num_bytes + 2 * 1024 * 1024, data, 0);
4616
4617         WARN_ON(num_bytes < root->sectorsize);
4618         ret = find_free_extent(trans, root, num_bytes, empty_size,
4619                                search_start, search_end, hint_byte, ins,
4620                                trans->alloc_exclude_start,
4621                                trans->alloc_exclude_nr, data);
4622
4623         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4624                 num_bytes = num_bytes >> 1;
4625                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4626                 num_bytes = max(num_bytes, min_alloc_size);
4627                 do_chunk_alloc(trans, root->fs_info->extent_root,
4628                                num_bytes, data, 1);
4629                 goto again;
4630         }
4631         if (ret == -ENOSPC) {
4632                 struct btrfs_space_info *sinfo;
4633
4634                 sinfo = __find_space_info(root->fs_info, data);
4635                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4636                        "wanted %llu\n", (unsigned long long)data,
4637                        (unsigned long long)num_bytes);
4638                 dump_space_info(sinfo, num_bytes, 1);
4639         }
4640
4641         return ret;
4642 }
4643
4644 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4645 {
4646         struct btrfs_block_group_cache *cache;
4647         int ret = 0;
4648
4649         cache = btrfs_lookup_block_group(root->fs_info, start);
4650         if (!cache) {
4651                 printk(KERN_ERR "Unable to find block group for %llu\n",
4652                        (unsigned long long)start);
4653                 return -ENOSPC;
4654         }
4655
4656         ret = btrfs_discard_extent(root, start, len);
4657
4658         btrfs_add_free_space(cache, start, len);
4659         update_reserved_extents(cache, len, 0);
4660         btrfs_put_block_group(cache);
4661
4662         return ret;
4663 }
4664
4665 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4666                                       struct btrfs_root *root,
4667                                       u64 parent, u64 root_objectid,
4668                                       u64 flags, u64 owner, u64 offset,
4669                                       struct btrfs_key *ins, int ref_mod)
4670 {
4671         int ret;
4672         struct btrfs_fs_info *fs_info = root->fs_info;
4673         struct btrfs_extent_item *extent_item;
4674         struct btrfs_extent_inline_ref *iref;
4675         struct btrfs_path *path;
4676         struct extent_buffer *leaf;
4677         int type;
4678         u32 size;
4679
4680         if (parent > 0)
4681                 type = BTRFS_SHARED_DATA_REF_KEY;
4682         else
4683                 type = BTRFS_EXTENT_DATA_REF_KEY;
4684
4685         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4686
4687         path = btrfs_alloc_path();
4688         BUG_ON(!path);
4689
4690         path->leave_spinning = 1;
4691         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4692                                       ins, size);
4693         BUG_ON(ret);
4694
4695         leaf = path->nodes[0];
4696         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4697                                      struct btrfs_extent_item);
4698         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4699         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4700         btrfs_set_extent_flags(leaf, extent_item,
4701                                flags | BTRFS_EXTENT_FLAG_DATA);
4702
4703         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4704         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4705         if (parent > 0) {
4706                 struct btrfs_shared_data_ref *ref;
4707                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4708                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4709                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4710         } else {
4711                 struct btrfs_extent_data_ref *ref;
4712                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4713                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4714                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4715                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4716                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4717         }
4718
4719         btrfs_mark_buffer_dirty(path->nodes[0]);
4720         btrfs_free_path(path);
4721
4722         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4723                                  1, 0);
4724         if (ret) {
4725                 printk(KERN_ERR "btrfs update block group failed for %llu "
4726                        "%llu\n", (unsigned long long)ins->objectid,
4727                        (unsigned long long)ins->offset);
4728                 BUG();
4729         }
4730         return ret;
4731 }
4732
4733 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4734                                      struct btrfs_root *root,
4735                                      u64 parent, u64 root_objectid,
4736                                      u64 flags, struct btrfs_disk_key *key,
4737                                      int level, struct btrfs_key *ins)
4738 {
4739         int ret;
4740         struct btrfs_fs_info *fs_info = root->fs_info;
4741         struct btrfs_extent_item *extent_item;
4742         struct btrfs_tree_block_info *block_info;
4743         struct btrfs_extent_inline_ref *iref;
4744         struct btrfs_path *path;
4745         struct extent_buffer *leaf;
4746         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4747
4748         path = btrfs_alloc_path();
4749         BUG_ON(!path);
4750
4751         path->leave_spinning = 1;
4752         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4753                                       ins, size);
4754         BUG_ON(ret);
4755
4756         leaf = path->nodes[0];
4757         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4758                                      struct btrfs_extent_item);
4759         btrfs_set_extent_refs(leaf, extent_item, 1);
4760         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4761         btrfs_set_extent_flags(leaf, extent_item,
4762                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4763         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4764
4765         btrfs_set_tree_block_key(leaf, block_info, key);
4766         btrfs_set_tree_block_level(leaf, block_info, level);
4767
4768         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4769         if (parent > 0) {
4770                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4771                 btrfs_set_extent_inline_ref_type(leaf, iref,
4772                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4773                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4774         } else {
4775                 btrfs_set_extent_inline_ref_type(leaf, iref,
4776                                                  BTRFS_TREE_BLOCK_REF_KEY);
4777                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4778         }
4779
4780         btrfs_mark_buffer_dirty(leaf);
4781         btrfs_free_path(path);
4782
4783         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4784                                  1, 0);
4785         if (ret) {
4786                 printk(KERN_ERR "btrfs update block group failed for %llu "
4787                        "%llu\n", (unsigned long long)ins->objectid,
4788                        (unsigned long long)ins->offset);
4789                 BUG();
4790         }
4791         return ret;
4792 }
4793
4794 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4795                                      struct btrfs_root *root,
4796                                      u64 root_objectid, u64 owner,
4797                                      u64 offset, struct btrfs_key *ins)
4798 {
4799         int ret;
4800
4801         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4802
4803         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4804                                          0, root_objectid, owner, offset,
4805                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4806         return ret;
4807 }
4808
4809 /*
4810  * this is used by the tree logging recovery code.  It records that
4811  * an extent has been allocated and makes sure to clear the free
4812  * space cache bits as well
4813  */
4814 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4815                                    struct btrfs_root *root,
4816                                    u64 root_objectid, u64 owner, u64 offset,
4817                                    struct btrfs_key *ins)
4818 {
4819         int ret;
4820         struct btrfs_block_group_cache *block_group;
4821         struct btrfs_caching_control *caching_ctl;
4822         u64 start = ins->objectid;
4823         u64 num_bytes = ins->offset;
4824
4825         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4826         cache_block_group(block_group);
4827         caching_ctl = get_caching_control(block_group);
4828
4829         if (!caching_ctl) {
4830                 BUG_ON(!block_group_cache_done(block_group));
4831                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4832                 BUG_ON(ret);
4833         } else {
4834                 mutex_lock(&caching_ctl->mutex);
4835
4836                 if (start >= caching_ctl->progress) {
4837                         ret = add_excluded_extent(root, start, num_bytes);
4838                         BUG_ON(ret);
4839                 } else if (start + num_bytes <= caching_ctl->progress) {
4840                         ret = btrfs_remove_free_space(block_group,
4841                                                       start, num_bytes);
4842                         BUG_ON(ret);
4843                 } else {
4844                         num_bytes = caching_ctl->progress - start;
4845                         ret = btrfs_remove_free_space(block_group,
4846                                                       start, num_bytes);
4847                         BUG_ON(ret);
4848
4849                         start = caching_ctl->progress;
4850                         num_bytes = ins->objectid + ins->offset -
4851                                     caching_ctl->progress;
4852                         ret = add_excluded_extent(root, start, num_bytes);
4853                         BUG_ON(ret);
4854                 }
4855
4856                 mutex_unlock(&caching_ctl->mutex);
4857                 put_caching_control(caching_ctl);
4858         }
4859
4860         update_reserved_extents(block_group, ins->offset, 1);
4861         btrfs_put_block_group(block_group);
4862         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4863                                          0, owner, offset, ins, 1);
4864         return ret;
4865 }
4866
4867 /*
4868  * finds a free extent and does all the dirty work required for allocation
4869  * returns the key for the extent through ins, and a tree buffer for
4870  * the first block of the extent through buf.
4871  *
4872  * returns 0 if everything worked, non-zero otherwise.
4873  */
4874 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4875                             struct btrfs_root *root,
4876                             u64 num_bytes, u64 parent, u64 root_objectid,
4877                             struct btrfs_disk_key *key, int level,
4878                             u64 empty_size, u64 hint_byte, u64 search_end,
4879                             struct btrfs_key *ins)
4880 {
4881         int ret;
4882         u64 flags = 0;
4883
4884         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4885                                    empty_size, hint_byte, search_end,
4886                                    ins, 0);
4887         if (ret)
4888                 return ret;
4889
4890         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4891                 if (parent == 0)
4892                         parent = ins->objectid;
4893                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4894         } else
4895                 BUG_ON(parent > 0);
4896
4897         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4898                 struct btrfs_delayed_extent_op *extent_op;
4899                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4900                 BUG_ON(!extent_op);
4901                 if (key)
4902                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4903                 else
4904                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4905                 extent_op->flags_to_set = flags;
4906                 extent_op->update_key = 1;
4907                 extent_op->update_flags = 1;
4908                 extent_op->is_data = 0;
4909
4910                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4911                                         ins->offset, parent, root_objectid,
4912                                         level, BTRFS_ADD_DELAYED_EXTENT,
4913                                         extent_op);
4914                 BUG_ON(ret);
4915         }
4916
4917         if (root_objectid == root->root_key.objectid) {
4918                 u64 used;
4919                 spin_lock(&root->node_lock);
4920                 used = btrfs_root_used(&root->root_item) + num_bytes;
4921                 btrfs_set_root_used(&root->root_item, used);
4922                 spin_unlock(&root->node_lock);
4923         }
4924         return ret;
4925 }
4926
4927 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4928                                             struct btrfs_root *root,
4929                                             u64 bytenr, u32 blocksize,
4930                                             int level)
4931 {
4932         struct extent_buffer *buf;
4933
4934         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4935         if (!buf)
4936                 return ERR_PTR(-ENOMEM);
4937         btrfs_set_header_generation(buf, trans->transid);
4938         btrfs_set_buffer_lockdep_class(buf, level);
4939         btrfs_tree_lock(buf);
4940         clean_tree_block(trans, root, buf);
4941
4942         btrfs_set_lock_blocking(buf);
4943         btrfs_set_buffer_uptodate(buf);
4944
4945         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4946                 /*
4947                  * we allow two log transactions at a time, use different
4948                  * EXENT bit to differentiate dirty pages.
4949                  */
4950                 if (root->log_transid % 2 == 0)
4951                         set_extent_dirty(&root->dirty_log_pages, buf->start,
4952                                         buf->start + buf->len - 1, GFP_NOFS);
4953                 else
4954                         set_extent_new(&root->dirty_log_pages, buf->start,
4955                                         buf->start + buf->len - 1, GFP_NOFS);
4956         } else {
4957                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4958                          buf->start + buf->len - 1, GFP_NOFS);
4959         }
4960         trans->blocks_used++;
4961         /* this returns a buffer locked for blocking */
4962         return buf;
4963 }
4964
4965 /*
4966  * helper function to allocate a block for a given tree
4967  * returns the tree buffer or NULL.
4968  */
4969 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4970                                         struct btrfs_root *root, u32 blocksize,
4971                                         u64 parent, u64 root_objectid,
4972                                         struct btrfs_disk_key *key, int level,
4973                                         u64 hint, u64 empty_size)
4974 {
4975         struct btrfs_key ins;
4976         int ret;
4977         struct extent_buffer *buf;
4978
4979         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4980                                key, level, empty_size, hint, (u64)-1, &ins);
4981         if (ret) {
4982                 BUG_ON(ret > 0);
4983                 return ERR_PTR(ret);
4984         }
4985
4986         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4987                                     blocksize, level);
4988         return buf;
4989 }
4990
4991 struct walk_control {
4992         u64 refs[BTRFS_MAX_LEVEL];
4993         u64 flags[BTRFS_MAX_LEVEL];
4994         struct btrfs_key update_progress;
4995         int stage;
4996         int level;
4997         int shared_level;
4998         int update_ref;
4999         int keep_locks;
5000         int reada_slot;
5001         int reada_count;
5002 };
5003
5004 #define DROP_REFERENCE  1
5005 #define UPDATE_BACKREF  2
5006
5007 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5008                                      struct btrfs_root *root,
5009                                      struct walk_control *wc,
5010                                      struct btrfs_path *path)
5011 {
5012         u64 bytenr;
5013         u64 generation;
5014         u64 refs;
5015         u64 flags;
5016         u64 last = 0;
5017         u32 nritems;
5018         u32 blocksize;
5019         struct btrfs_key key;
5020         struct extent_buffer *eb;
5021         int ret;
5022         int slot;
5023         int nread = 0;
5024
5025         if (path->slots[wc->level] < wc->reada_slot) {
5026                 wc->reada_count = wc->reada_count * 2 / 3;
5027                 wc->reada_count = max(wc->reada_count, 2);
5028         } else {
5029                 wc->reada_count = wc->reada_count * 3 / 2;
5030                 wc->reada_count = min_t(int, wc->reada_count,
5031                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5032         }
5033
5034         eb = path->nodes[wc->level];
5035         nritems = btrfs_header_nritems(eb);
5036         blocksize = btrfs_level_size(root, wc->level - 1);
5037
5038         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5039                 if (nread >= wc->reada_count)
5040                         break;
5041
5042                 cond_resched();
5043                 bytenr = btrfs_node_blockptr(eb, slot);
5044                 generation = btrfs_node_ptr_generation(eb, slot);
5045
5046                 if (slot == path->slots[wc->level])
5047                         goto reada;
5048
5049                 if (wc->stage == UPDATE_BACKREF &&
5050                     generation <= root->root_key.offset)
5051                         continue;
5052
5053                 /* We don't lock the tree block, it's OK to be racy here */
5054                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5055                                                &refs, &flags);
5056                 BUG_ON(ret);
5057                 BUG_ON(refs == 0);
5058
5059                 if (wc->stage == DROP_REFERENCE) {
5060                         if (refs == 1)
5061                                 goto reada;
5062
5063                         if (wc->level == 1 &&
5064                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5065                                 continue;
5066                         if (!wc->update_ref ||
5067                             generation <= root->root_key.offset)
5068                                 continue;
5069                         btrfs_node_key_to_cpu(eb, &key, slot);
5070                         ret = btrfs_comp_cpu_keys(&key,
5071                                                   &wc->update_progress);
5072                         if (ret < 0)
5073                                 continue;
5074                 } else {
5075                         if (wc->level == 1 &&
5076                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5077                                 continue;
5078                 }
5079 reada:
5080                 ret = readahead_tree_block(root, bytenr, blocksize,
5081                                            generation);
5082                 if (ret)
5083                         break;
5084                 last = bytenr + blocksize;
5085                 nread++;
5086         }
5087         wc->reada_slot = slot;
5088 }
5089
5090 /*
5091  * hepler to process tree block while walking down the tree.
5092  *
5093  * when wc->stage == UPDATE_BACKREF, this function updates
5094  * back refs for pointers in the block.
5095  *
5096  * NOTE: return value 1 means we should stop walking down.
5097  */
5098 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5099                                    struct btrfs_root *root,
5100                                    struct btrfs_path *path,
5101                                    struct walk_control *wc, int lookup_info)
5102 {
5103         int level = wc->level;
5104         struct extent_buffer *eb = path->nodes[level];
5105         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5106         int ret;
5107
5108         if (wc->stage == UPDATE_BACKREF &&
5109             btrfs_header_owner(eb) != root->root_key.objectid)
5110                 return 1;
5111
5112         /*
5113          * when reference count of tree block is 1, it won't increase
5114          * again. once full backref flag is set, we never clear it.
5115          */
5116         if (lookup_info &&
5117             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5118              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5119                 BUG_ON(!path->locks[level]);
5120                 ret = btrfs_lookup_extent_info(trans, root,
5121                                                eb->start, eb->len,
5122                                                &wc->refs[level],
5123                                                &wc->flags[level]);
5124                 BUG_ON(ret);
5125                 BUG_ON(wc->refs[level] == 0);
5126         }
5127
5128         if (wc->stage == DROP_REFERENCE) {
5129                 if (wc->refs[level] > 1)
5130                         return 1;
5131
5132                 if (path->locks[level] && !wc->keep_locks) {
5133                         btrfs_tree_unlock(eb);
5134                         path->locks[level] = 0;
5135                 }
5136                 return 0;
5137         }
5138
5139         /* wc->stage == UPDATE_BACKREF */
5140         if (!(wc->flags[level] & flag)) {
5141                 BUG_ON(!path->locks[level]);
5142                 ret = btrfs_inc_ref(trans, root, eb, 1);
5143                 BUG_ON(ret);
5144                 ret = btrfs_dec_ref(trans, root, eb, 0);
5145                 BUG_ON(ret);
5146                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5147                                                   eb->len, flag, 0);
5148                 BUG_ON(ret);
5149                 wc->flags[level] |= flag;
5150         }
5151
5152         /*
5153          * the block is shared by multiple trees, so it's not good to
5154          * keep the tree lock
5155          */
5156         if (path->locks[level] && level > 0) {
5157                 btrfs_tree_unlock(eb);
5158                 path->locks[level] = 0;
5159         }
5160         return 0;
5161 }
5162
5163 /*
5164  * hepler to process tree block pointer.
5165  *
5166  * when wc->stage == DROP_REFERENCE, this function checks
5167  * reference count of the block pointed to. if the block
5168  * is shared and we need update back refs for the subtree
5169  * rooted at the block, this function changes wc->stage to
5170  * UPDATE_BACKREF. if the block is shared and there is no
5171  * need to update back, this function drops the reference
5172  * to the block.
5173  *
5174  * NOTE: return value 1 means we should stop walking down.
5175  */
5176 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5177                                  struct btrfs_root *root,
5178                                  struct btrfs_path *path,
5179                                  struct walk_control *wc, int *lookup_info)
5180 {
5181         u64 bytenr;
5182         u64 generation;
5183         u64 parent;
5184         u32 blocksize;
5185         struct btrfs_key key;
5186         struct extent_buffer *next;
5187         int level = wc->level;
5188         int reada = 0;
5189         int ret = 0;
5190
5191         generation = btrfs_node_ptr_generation(path->nodes[level],
5192                                                path->slots[level]);
5193         /*
5194          * if the lower level block was created before the snapshot
5195          * was created, we know there is no need to update back refs
5196          * for the subtree
5197          */
5198         if (wc->stage == UPDATE_BACKREF &&
5199             generation <= root->root_key.offset) {
5200                 *lookup_info = 1;
5201                 return 1;
5202         }
5203
5204         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5205         blocksize = btrfs_level_size(root, level - 1);
5206
5207         next = btrfs_find_tree_block(root, bytenr, blocksize);
5208         if (!next) {
5209                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5210                 if (!next)
5211                         return -ENOMEM;
5212                 reada = 1;
5213         }
5214         btrfs_tree_lock(next);
5215         btrfs_set_lock_blocking(next);
5216
5217         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5218                                        &wc->refs[level - 1],
5219                                        &wc->flags[level - 1]);
5220         BUG_ON(ret);
5221         BUG_ON(wc->refs[level - 1] == 0);
5222         *lookup_info = 0;
5223
5224         if (wc->stage == DROP_REFERENCE) {
5225                 if (wc->refs[level - 1] > 1) {
5226                         if (level == 1 &&
5227                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5228                                 goto skip;
5229
5230                         if (!wc->update_ref ||
5231                             generation <= root->root_key.offset)
5232                                 goto skip;
5233
5234                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5235                                               path->slots[level]);
5236                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5237                         if (ret < 0)
5238                                 goto skip;
5239
5240                         wc->stage = UPDATE_BACKREF;
5241                         wc->shared_level = level - 1;
5242                 }
5243         } else {
5244                 if (level == 1 &&
5245                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5246                         goto skip;
5247         }
5248
5249         if (!btrfs_buffer_uptodate(next, generation)) {
5250                 btrfs_tree_unlock(next);
5251                 free_extent_buffer(next);
5252                 next = NULL;
5253                 *lookup_info = 1;
5254         }
5255
5256         if (!next) {
5257                 if (reada && level == 1)
5258                         reada_walk_down(trans, root, wc, path);
5259                 next = read_tree_block(root, bytenr, blocksize, generation);
5260                 btrfs_tree_lock(next);
5261                 btrfs_set_lock_blocking(next);
5262         }
5263
5264         level--;
5265         BUG_ON(level != btrfs_header_level(next));
5266         path->nodes[level] = next;
5267         path->slots[level] = 0;
5268         path->locks[level] = 1;
5269         wc->level = level;
5270         if (wc->level == 1)
5271                 wc->reada_slot = 0;
5272         return 0;
5273 skip:
5274         wc->refs[level - 1] = 0;
5275         wc->flags[level - 1] = 0;
5276         if (wc->stage == DROP_REFERENCE) {
5277                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5278                         parent = path->nodes[level]->start;
5279                 } else {
5280                         BUG_ON(root->root_key.objectid !=
5281                                btrfs_header_owner(path->nodes[level]));
5282                         parent = 0;
5283                 }
5284
5285                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5286                                         root->root_key.objectid, level - 1, 0);
5287                 BUG_ON(ret);
5288         }
5289         btrfs_tree_unlock(next);
5290         free_extent_buffer(next);
5291         *lookup_info = 1;
5292         return 1;
5293 }
5294
5295 /*
5296  * hepler to process tree block while walking up the tree.
5297  *
5298  * when wc->stage == DROP_REFERENCE, this function drops
5299  * reference count on the block.
5300  *
5301  * when wc->stage == UPDATE_BACKREF, this function changes
5302  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5303  * to UPDATE_BACKREF previously while processing the block.
5304  *
5305  * NOTE: return value 1 means we should stop walking up.
5306  */
5307 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5308                                  struct btrfs_root *root,
5309                                  struct btrfs_path *path,
5310                                  struct walk_control *wc)
5311 {
5312         int ret = 0;
5313         int level = wc->level;
5314         struct extent_buffer *eb = path->nodes[level];
5315         u64 parent = 0;
5316
5317         if (wc->stage == UPDATE_BACKREF) {
5318                 BUG_ON(wc->shared_level < level);
5319                 if (level < wc->shared_level)
5320                         goto out;
5321
5322                 ret = find_next_key(path, level + 1, &wc->update_progress);
5323                 if (ret > 0)
5324                         wc->update_ref = 0;
5325
5326                 wc->stage = DROP_REFERENCE;
5327                 wc->shared_level = -1;
5328                 path->slots[level] = 0;
5329
5330                 /*
5331                  * check reference count again if the block isn't locked.
5332                  * we should start walking down the tree again if reference
5333                  * count is one.
5334                  */
5335                 if (!path->locks[level]) {
5336                         BUG_ON(level == 0);
5337                         btrfs_tree_lock(eb);
5338                         btrfs_set_lock_blocking(eb);
5339                         path->locks[level] = 1;
5340
5341                         ret = btrfs_lookup_extent_info(trans, root,
5342                                                        eb->start, eb->len,
5343                                                        &wc->refs[level],
5344                                                        &wc->flags[level]);
5345                         BUG_ON(ret);
5346                         BUG_ON(wc->refs[level] == 0);
5347                         if (wc->refs[level] == 1) {
5348                                 btrfs_tree_unlock(eb);
5349                                 path->locks[level] = 0;
5350                                 return 1;
5351                         }
5352                 }
5353         }
5354
5355         /* wc->stage == DROP_REFERENCE */
5356         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5357
5358         if (wc->refs[level] == 1) {
5359                 if (level == 0) {
5360                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5361                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5362                         else
5363                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5364                         BUG_ON(ret);
5365                 }
5366                 /* make block locked assertion in clean_tree_block happy */
5367                 if (!path->locks[level] &&
5368                     btrfs_header_generation(eb) == trans->transid) {
5369                         btrfs_tree_lock(eb);
5370                         btrfs_set_lock_blocking(eb);
5371                         path->locks[level] = 1;
5372                 }
5373                 clean_tree_block(trans, root, eb);
5374         }
5375
5376         if (eb == root->node) {
5377                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5378                         parent = eb->start;
5379                 else
5380                         BUG_ON(root->root_key.objectid !=
5381                                btrfs_header_owner(eb));
5382         } else {
5383                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5384                         parent = path->nodes[level + 1]->start;
5385                 else
5386                         BUG_ON(root->root_key.objectid !=
5387                                btrfs_header_owner(path->nodes[level + 1]));
5388         }
5389
5390         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5391                                 root->root_key.objectid, level, 0);
5392         BUG_ON(ret);
5393 out:
5394         wc->refs[level] = 0;
5395         wc->flags[level] = 0;
5396         return ret;
5397 }
5398
5399 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5400                                    struct btrfs_root *root,
5401                                    struct btrfs_path *path,
5402                                    struct walk_control *wc)
5403 {
5404         int level = wc->level;
5405         int lookup_info = 1;
5406         int ret;
5407
5408         while (level >= 0) {
5409                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5410                 if (ret > 0)
5411                         break;
5412
5413                 if (level == 0)
5414                         break;
5415
5416                 if (path->slots[level] >=
5417                     btrfs_header_nritems(path->nodes[level]))
5418                         break;
5419
5420                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5421                 if (ret > 0) {
5422                         path->slots[level]++;
5423                         continue;
5424                 } else if (ret < 0)
5425                         return ret;
5426                 level = wc->level;
5427         }
5428         return 0;
5429 }
5430
5431 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5432                                  struct btrfs_root *root,
5433                                  struct btrfs_path *path,
5434                                  struct walk_control *wc, int max_level)
5435 {
5436         int level = wc->level;
5437         int ret;
5438
5439         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5440         while (level < max_level && path->nodes[level]) {
5441                 wc->level = level;
5442                 if (path->slots[level] + 1 <
5443                     btrfs_header_nritems(path->nodes[level])) {
5444                         path->slots[level]++;
5445                         return 0;
5446                 } else {
5447                         ret = walk_up_proc(trans, root, path, wc);
5448                         if (ret > 0)
5449                                 return 0;
5450
5451                         if (path->locks[level]) {
5452                                 btrfs_tree_unlock(path->nodes[level]);
5453                                 path->locks[level] = 0;
5454                         }
5455                         free_extent_buffer(path->nodes[level]);
5456                         path->nodes[level] = NULL;
5457                         level++;
5458                 }
5459         }
5460         return 1;
5461 }
5462
5463 /*
5464  * drop a subvolume tree.
5465  *
5466  * this function traverses the tree freeing any blocks that only
5467  * referenced by the tree.
5468  *
5469  * when a shared tree block is found. this function decreases its
5470  * reference count by one. if update_ref is true, this function
5471  * also make sure backrefs for the shared block and all lower level
5472  * blocks are properly updated.
5473  */
5474 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5475 {
5476         struct btrfs_path *path;
5477         struct btrfs_trans_handle *trans;
5478         struct btrfs_root *tree_root = root->fs_info->tree_root;
5479         struct btrfs_root_item *root_item = &root->root_item;
5480         struct walk_control *wc;
5481         struct btrfs_key key;
5482         int err = 0;
5483         int ret;
5484         int level;
5485
5486         path = btrfs_alloc_path();
5487         BUG_ON(!path);
5488
5489         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5490         BUG_ON(!wc);
5491
5492         trans = btrfs_start_transaction(tree_root, 1);
5493
5494         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5495                 level = btrfs_header_level(root->node);
5496                 path->nodes[level] = btrfs_lock_root_node(root);
5497                 btrfs_set_lock_blocking(path->nodes[level]);
5498                 path->slots[level] = 0;
5499                 path->locks[level] = 1;
5500                 memset(&wc->update_progress, 0,
5501                        sizeof(wc->update_progress));
5502         } else {
5503                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5504                 memcpy(&wc->update_progress, &key,
5505                        sizeof(wc->update_progress));
5506
5507                 level = root_item->drop_level;
5508                 BUG_ON(level == 0);
5509                 path->lowest_level = level;
5510                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5511                 path->lowest_level = 0;
5512                 if (ret < 0) {
5513                         err = ret;
5514                         goto out;
5515                 }
5516                 WARN_ON(ret > 0);
5517
5518                 /*
5519                  * unlock our path, this is safe because only this
5520                  * function is allowed to delete this snapshot
5521                  */
5522                 btrfs_unlock_up_safe(path, 0);
5523
5524                 level = btrfs_header_level(root->node);
5525                 while (1) {
5526                         btrfs_tree_lock(path->nodes[level]);
5527                         btrfs_set_lock_blocking(path->nodes[level]);
5528
5529                         ret = btrfs_lookup_extent_info(trans, root,
5530                                                 path->nodes[level]->start,
5531                                                 path->nodes[level]->len,
5532                                                 &wc->refs[level],
5533                                                 &wc->flags[level]);
5534                         BUG_ON(ret);
5535                         BUG_ON(wc->refs[level] == 0);
5536
5537                         if (level == root_item->drop_level)
5538                                 break;
5539
5540                         btrfs_tree_unlock(path->nodes[level]);
5541                         WARN_ON(wc->refs[level] != 1);
5542                         level--;
5543                 }
5544         }
5545
5546         wc->level = level;
5547         wc->shared_level = -1;
5548         wc->stage = DROP_REFERENCE;
5549         wc->update_ref = update_ref;
5550         wc->keep_locks = 0;
5551         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5552
5553         while (1) {
5554                 ret = walk_down_tree(trans, root, path, wc);
5555                 if (ret < 0) {
5556                         err = ret;
5557                         break;
5558                 }
5559
5560                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5561                 if (ret < 0) {
5562                         err = ret;
5563                         break;
5564                 }
5565
5566                 if (ret > 0) {
5567                         BUG_ON(wc->stage != DROP_REFERENCE);
5568                         break;
5569                 }
5570
5571                 if (wc->stage == DROP_REFERENCE) {
5572                         level = wc->level;
5573                         btrfs_node_key(path->nodes[level],
5574                                        &root_item->drop_progress,
5575                                        path->slots[level]);
5576                         root_item->drop_level = level;
5577                 }
5578
5579                 BUG_ON(wc->level == 0);
5580                 if (trans->transaction->in_commit ||
5581                     trans->transaction->delayed_refs.flushing) {
5582                         ret = btrfs_update_root(trans, tree_root,
5583                                                 &root->root_key,
5584                                                 root_item);
5585                         BUG_ON(ret);
5586
5587                         btrfs_end_transaction(trans, tree_root);
5588                         trans = btrfs_start_transaction(tree_root, 1);
5589                 } else {
5590                         unsigned long update;
5591                         update = trans->delayed_ref_updates;
5592                         trans->delayed_ref_updates = 0;
5593                         if (update)
5594                                 btrfs_run_delayed_refs(trans, tree_root,
5595                                                        update);
5596                 }
5597         }
5598         btrfs_release_path(root, path);
5599         BUG_ON(err);
5600
5601         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5602         BUG_ON(ret);
5603
5604         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5605                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5606                                            NULL, NULL);
5607                 BUG_ON(ret < 0);
5608                 if (ret > 0) {
5609                         ret = btrfs_del_orphan_item(trans, tree_root,
5610                                                     root->root_key.objectid);
5611                         BUG_ON(ret);
5612                 }
5613         }
5614
5615         if (root->in_radix) {
5616                 btrfs_free_fs_root(tree_root->fs_info, root);
5617         } else {
5618                 free_extent_buffer(root->node);
5619                 free_extent_buffer(root->commit_root);
5620                 kfree(root);
5621         }
5622 out:
5623         btrfs_end_transaction(trans, tree_root);
5624         kfree(wc);
5625         btrfs_free_path(path);
5626         return err;
5627 }
5628
5629 /*
5630  * drop subtree rooted at tree block 'node'.
5631  *
5632  * NOTE: this function will unlock and release tree block 'node'
5633  */
5634 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5635                         struct btrfs_root *root,
5636                         struct extent_buffer *node,
5637                         struct extent_buffer *parent)
5638 {
5639         struct btrfs_path *path;
5640         struct walk_control *wc;
5641         int level;
5642         int parent_level;
5643         int ret = 0;
5644         int wret;
5645
5646         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5647
5648         path = btrfs_alloc_path();
5649         BUG_ON(!path);
5650
5651         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5652         BUG_ON(!wc);
5653
5654         btrfs_assert_tree_locked(parent);
5655         parent_level = btrfs_header_level(parent);
5656         extent_buffer_get(parent);
5657         path->nodes[parent_level] = parent;
5658         path->slots[parent_level] = btrfs_header_nritems(parent);
5659
5660         btrfs_assert_tree_locked(node);
5661         level = btrfs_header_level(node);
5662         path->nodes[level] = node;
5663         path->slots[level] = 0;
5664         path->locks[level] = 1;
5665
5666         wc->refs[parent_level] = 1;
5667         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5668         wc->level = level;
5669         wc->shared_level = -1;
5670         wc->stage = DROP_REFERENCE;
5671         wc->update_ref = 0;
5672         wc->keep_locks = 1;
5673         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5674
5675         while (1) {
5676                 wret = walk_down_tree(trans, root, path, wc);
5677                 if (wret < 0) {
5678                         ret = wret;
5679                         break;
5680                 }
5681
5682                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5683                 if (wret < 0)
5684                         ret = wret;
5685                 if (wret != 0)
5686                         break;
5687         }
5688
5689         kfree(wc);
5690         btrfs_free_path(path);
5691         return ret;
5692 }
5693
5694 #if 0
5695 static unsigned long calc_ra(unsigned long start, unsigned long last,
5696                              unsigned long nr)
5697 {
5698         return min(last, start + nr - 1);
5699 }
5700
5701 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5702                                          u64 len)
5703 {
5704         u64 page_start;
5705         u64 page_end;
5706         unsigned long first_index;
5707         unsigned long last_index;
5708         unsigned long i;
5709         struct page *page;
5710         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5711         struct file_ra_state *ra;
5712         struct btrfs_ordered_extent *ordered;
5713         unsigned int total_read = 0;
5714         unsigned int total_dirty = 0;
5715         int ret = 0;
5716
5717         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5718
5719         mutex_lock(&inode->i_mutex);
5720         first_index = start >> PAGE_CACHE_SHIFT;
5721         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5722
5723         /* make sure the dirty trick played by the caller work */
5724         ret = invalidate_inode_pages2_range(inode->i_mapping,
5725                                             first_index, last_index);
5726         if (ret)
5727                 goto out_unlock;
5728
5729         file_ra_state_init(ra, inode->i_mapping);
5730
5731         for (i = first_index ; i <= last_index; i++) {
5732                 if (total_read % ra->ra_pages == 0) {
5733                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5734                                        calc_ra(i, last_index, ra->ra_pages));
5735                 }
5736                 total_read++;
5737 again:
5738                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5739                         BUG_ON(1);
5740                 page = grab_cache_page(inode->i_mapping, i);
5741                 if (!page) {
5742                         ret = -ENOMEM;
5743                         goto out_unlock;
5744                 }
5745                 if (!PageUptodate(page)) {
5746                         btrfs_readpage(NULL, page);
5747                         lock_page(page);
5748                         if (!PageUptodate(page)) {
5749                                 unlock_page(page);
5750                                 page_cache_release(page);
5751                                 ret = -EIO;
5752                                 goto out_unlock;
5753                         }
5754                 }
5755                 wait_on_page_writeback(page);
5756
5757                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5758                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5759                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5760
5761                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5762                 if (ordered) {
5763                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5764                         unlock_page(page);
5765                         page_cache_release(page);
5766                         btrfs_start_ordered_extent(inode, ordered, 1);
5767                         btrfs_put_ordered_extent(ordered);
5768                         goto again;
5769                 }
5770                 set_page_extent_mapped(page);
5771
5772                 if (i == first_index)
5773                         set_extent_bits(io_tree, page_start, page_end,
5774                                         EXTENT_BOUNDARY, GFP_NOFS);
5775                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5776
5777                 set_page_dirty(page);
5778                 total_dirty++;
5779
5780                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5781                 unlock_page(page);
5782                 page_cache_release(page);
5783         }
5784
5785 out_unlock:
5786         kfree(ra);
5787         mutex_unlock(&inode->i_mutex);
5788         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5789         return ret;
5790 }
5791
5792 static noinline int relocate_data_extent(struct inode *reloc_inode,
5793                                          struct btrfs_key *extent_key,
5794                                          u64 offset)
5795 {
5796         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5797         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5798         struct extent_map *em;
5799         u64 start = extent_key->objectid - offset;
5800         u64 end = start + extent_key->offset - 1;
5801
5802         em = alloc_extent_map(GFP_NOFS);
5803         BUG_ON(!em || IS_ERR(em));
5804
5805         em->start = start;
5806         em->len = extent_key->offset;
5807         em->block_len = extent_key->offset;
5808         em->block_start = extent_key->objectid;
5809         em->bdev = root->fs_info->fs_devices->latest_bdev;
5810         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5811
5812         /* setup extent map to cheat btrfs_readpage */
5813         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5814         while (1) {
5815                 int ret;
5816                 write_lock(&em_tree->lock);
5817                 ret = add_extent_mapping(em_tree, em);
5818                 write_unlock(&em_tree->lock);
5819                 if (ret != -EEXIST) {
5820                         free_extent_map(em);
5821                         break;
5822                 }
5823                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5824         }
5825         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5826
5827         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5828 }
5829
5830 struct btrfs_ref_path {
5831         u64 extent_start;
5832         u64 nodes[BTRFS_MAX_LEVEL];
5833         u64 root_objectid;
5834         u64 root_generation;
5835         u64 owner_objectid;
5836         u32 num_refs;
5837         int lowest_level;
5838         int current_level;
5839         int shared_level;
5840
5841         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5842         u64 new_nodes[BTRFS_MAX_LEVEL];
5843 };
5844
5845 struct disk_extent {
5846         u64 ram_bytes;
5847         u64 disk_bytenr;
5848         u64 disk_num_bytes;
5849         u64 offset;
5850         u64 num_bytes;
5851         u8 compression;
5852         u8 encryption;
5853         u16 other_encoding;
5854 };
5855
5856 static int is_cowonly_root(u64 root_objectid)
5857 {
5858         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5859             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5860             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5861             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5862             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5863             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5864                 return 1;
5865         return 0;
5866 }
5867
5868 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5869                                     struct btrfs_root *extent_root,
5870                                     struct btrfs_ref_path *ref_path,
5871                                     int first_time)
5872 {
5873         struct extent_buffer *leaf;
5874         struct btrfs_path *path;
5875         struct btrfs_extent_ref *ref;
5876         struct btrfs_key key;
5877         struct btrfs_key found_key;
5878         u64 bytenr;
5879         u32 nritems;
5880         int level;
5881         int ret = 1;
5882
5883         path = btrfs_alloc_path();
5884         if (!path)
5885                 return -ENOMEM;
5886
5887         if (first_time) {
5888                 ref_path->lowest_level = -1;
5889                 ref_path->current_level = -1;
5890                 ref_path->shared_level = -1;
5891                 goto walk_up;
5892         }
5893 walk_down:
5894         level = ref_path->current_level - 1;
5895         while (level >= -1) {
5896                 u64 parent;
5897                 if (level < ref_path->lowest_level)
5898                         break;
5899
5900                 if (level >= 0)
5901                         bytenr = ref_path->nodes[level];
5902                 else
5903                         bytenr = ref_path->extent_start;
5904                 BUG_ON(bytenr == 0);
5905
5906                 parent = ref_path->nodes[level + 1];
5907                 ref_path->nodes[level + 1] = 0;
5908                 ref_path->current_level = level;
5909                 BUG_ON(parent == 0);
5910
5911                 key.objectid = bytenr;
5912                 key.offset = parent + 1;
5913                 key.type = BTRFS_EXTENT_REF_KEY;
5914
5915                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5916                 if (ret < 0)
5917                         goto out;
5918                 BUG_ON(ret == 0);
5919
5920                 leaf = path->nodes[0];
5921                 nritems = btrfs_header_nritems(leaf);
5922                 if (path->slots[0] >= nritems) {
5923                         ret = btrfs_next_leaf(extent_root, path);
5924                         if (ret < 0)
5925                                 goto out;
5926                         if (ret > 0)
5927                                 goto next;
5928                         leaf = path->nodes[0];
5929                 }
5930
5931                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5932                 if (found_key.objectid == bytenr &&
5933                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5934                         if (level < ref_path->shared_level)
5935                                 ref_path->shared_level = level;
5936                         goto found;
5937                 }
5938 next:
5939                 level--;
5940                 btrfs_release_path(extent_root, path);
5941                 cond_resched();
5942         }
5943         /* reached lowest level */
5944         ret = 1;
5945         goto out;
5946 walk_up:
5947         level = ref_path->current_level;
5948         while (level < BTRFS_MAX_LEVEL - 1) {
5949                 u64 ref_objectid;
5950
5951                 if (level >= 0)
5952                         bytenr = ref_path->nodes[level];
5953                 else
5954                         bytenr = ref_path->extent_start;
5955
5956                 BUG_ON(bytenr == 0);
5957
5958                 key.objectid = bytenr;
5959                 key.offset = 0;
5960                 key.type = BTRFS_EXTENT_REF_KEY;
5961
5962                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5963                 if (ret < 0)
5964                         goto out;
5965
5966                 leaf = path->nodes[0];
5967                 nritems = btrfs_header_nritems(leaf);
5968                 if (path->slots[0] >= nritems) {
5969                         ret = btrfs_next_leaf(extent_root, path);
5970                         if (ret < 0)
5971                                 goto out;
5972                         if (ret > 0) {
5973                                 /* the extent was freed by someone */
5974                                 if (ref_path->lowest_level == level)
5975                                         goto out;
5976                                 btrfs_release_path(extent_root, path);
5977                                 goto walk_down;
5978                         }
5979                         leaf = path->nodes[0];
5980                 }
5981
5982                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5983                 if (found_key.objectid != bytenr ||
5984                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5985                         /* the extent was freed by someone */
5986                         if (ref_path->lowest_level == level) {
5987                                 ret = 1;
5988                                 goto out;
5989                         }
5990                         btrfs_release_path(extent_root, path);
5991                         goto walk_down;
5992                 }
5993 found:
5994                 ref = btrfs_item_ptr(leaf, path->slots[0],
5995                                 struct btrfs_extent_ref);
5996                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5997                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5998                         if (first_time) {
5999                                 level = (int)ref_objectid;
6000                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6001                                 ref_path->lowest_level = level;
6002                                 ref_path->current_level = level;
6003                                 ref_path->nodes[level] = bytenr;
6004                         } else {
6005                                 WARN_ON(ref_objectid != level);
6006                         }
6007                 } else {
6008                         WARN_ON(level != -1);
6009                 }
6010                 first_time = 0;
6011
6012                 if (ref_path->lowest_level == level) {
6013                         ref_path->owner_objectid = ref_objectid;
6014                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6015                 }
6016
6017                 /*
6018                  * the block is tree root or the block isn't in reference
6019                  * counted tree.
6020                  */
6021                 if (found_key.objectid == found_key.offset ||
6022                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6023                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6024                         ref_path->root_generation =
6025                                 btrfs_ref_generation(leaf, ref);
6026                         if (level < 0) {
6027                                 /* special reference from the tree log */
6028                                 ref_path->nodes[0] = found_key.offset;
6029                                 ref_path->current_level = 0;
6030                         }
6031                         ret = 0;
6032                         goto out;
6033                 }
6034
6035                 level++;
6036                 BUG_ON(ref_path->nodes[level] != 0);
6037                 ref_path->nodes[level] = found_key.offset;
6038                 ref_path->current_level = level;
6039
6040                 /*
6041                  * the reference was created in the running transaction,
6042                  * no need to continue walking up.
6043                  */
6044                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6045                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6046                         ref_path->root_generation =
6047                                 btrfs_ref_generation(leaf, ref);
6048                         ret = 0;
6049                         goto out;
6050                 }
6051
6052                 btrfs_release_path(extent_root, path);
6053                 cond_resched();
6054         }
6055         /* reached max tree level, but no tree root found. */
6056         BUG();
6057 out:
6058         btrfs_free_path(path);
6059         return ret;
6060 }
6061
6062 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6063                                 struct btrfs_root *extent_root,
6064                                 struct btrfs_ref_path *ref_path,
6065                                 u64 extent_start)
6066 {
6067         memset(ref_path, 0, sizeof(*ref_path));
6068         ref_path->extent_start = extent_start;
6069
6070         return __next_ref_path(trans, extent_root, ref_path, 1);
6071 }
6072
6073 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6074                                struct btrfs_root *extent_root,
6075                                struct btrfs_ref_path *ref_path)
6076 {
6077         return __next_ref_path(trans, extent_root, ref_path, 0);
6078 }
6079
6080 static noinline int get_new_locations(struct inode *reloc_inode,
6081                                       struct btrfs_key *extent_key,
6082                                       u64 offset, int no_fragment,
6083                                       struct disk_extent **extents,
6084                                       int *nr_extents)
6085 {
6086         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6087         struct btrfs_path *path;
6088         struct btrfs_file_extent_item *fi;
6089         struct extent_buffer *leaf;
6090         struct disk_extent *exts = *extents;
6091         struct btrfs_key found_key;
6092         u64 cur_pos;
6093         u64 last_byte;
6094         u32 nritems;
6095         int nr = 0;
6096         int max = *nr_extents;
6097         int ret;
6098
6099         WARN_ON(!no_fragment && *extents);
6100         if (!exts) {
6101                 max = 1;
6102                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6103                 if (!exts)
6104                         return -ENOMEM;
6105         }
6106
6107         path = btrfs_alloc_path();
6108         BUG_ON(!path);
6109
6110         cur_pos = extent_key->objectid - offset;
6111         last_byte = extent_key->objectid + extent_key->offset;
6112         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6113                                        cur_pos, 0);
6114         if (ret < 0)
6115                 goto out;
6116         if (ret > 0) {
6117                 ret = -ENOENT;
6118                 goto out;
6119         }
6120
6121         while (1) {
6122                 leaf = path->nodes[0];
6123                 nritems = btrfs_header_nritems(leaf);
6124                 if (path->slots[0] >= nritems) {
6125                         ret = btrfs_next_leaf(root, path);
6126                         if (ret < 0)
6127                                 goto out;
6128                         if (ret > 0)
6129                                 break;
6130                         leaf = path->nodes[0];
6131                 }
6132
6133                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6134                 if (found_key.offset != cur_pos ||
6135                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6136                     found_key.objectid != reloc_inode->i_ino)
6137                         break;
6138
6139                 fi = btrfs_item_ptr(leaf, path->slots[0],
6140                                     struct btrfs_file_extent_item);
6141                 if (btrfs_file_extent_type(leaf, fi) !=
6142                     BTRFS_FILE_EXTENT_REG ||
6143                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6144                         break;
6145
6146                 if (nr == max) {
6147                         struct disk_extent *old = exts;
6148                         max *= 2;
6149                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6150                         memcpy(exts, old, sizeof(*exts) * nr);
6151                         if (old != *extents)
6152                                 kfree(old);
6153                 }
6154
6155                 exts[nr].disk_bytenr =
6156                         btrfs_file_extent_disk_bytenr(leaf, fi);
6157                 exts[nr].disk_num_bytes =
6158                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6159                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6160                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6161                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6162                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6163                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6164                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6165                                                                            fi);
6166                 BUG_ON(exts[nr].offset > 0);
6167                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6168                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6169
6170                 cur_pos += exts[nr].num_bytes;
6171                 nr++;
6172
6173                 if (cur_pos + offset >= last_byte)
6174                         break;
6175
6176                 if (no_fragment) {
6177                         ret = 1;
6178                         goto out;
6179                 }
6180                 path->slots[0]++;
6181         }
6182
6183         BUG_ON(cur_pos + offset > last_byte);
6184         if (cur_pos + offset < last_byte) {
6185                 ret = -ENOENT;
6186                 goto out;
6187         }
6188         ret = 0;
6189 out:
6190         btrfs_free_path(path);
6191         if (ret) {
6192                 if (exts != *extents)
6193                         kfree(exts);
6194         } else {
6195                 *extents = exts;
6196                 *nr_extents = nr;
6197         }
6198         return ret;
6199 }
6200
6201 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6202                                         struct btrfs_root *root,
6203                                         struct btrfs_path *path,
6204                                         struct btrfs_key *extent_key,
6205                                         struct btrfs_key *leaf_key,
6206                                         struct btrfs_ref_path *ref_path,
6207                                         struct disk_extent *new_extents,
6208                                         int nr_extents)
6209 {
6210         struct extent_buffer *leaf;
6211         struct btrfs_file_extent_item *fi;
6212         struct inode *inode = NULL;
6213         struct btrfs_key key;
6214         u64 lock_start = 0;
6215         u64 lock_end = 0;
6216         u64 num_bytes;
6217         u64 ext_offset;
6218         u64 search_end = (u64)-1;
6219         u32 nritems;
6220         int nr_scaned = 0;
6221         int extent_locked = 0;
6222         int extent_type;
6223         int ret;
6224
6225         memcpy(&key, leaf_key, sizeof(key));
6226         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6227                 if (key.objectid < ref_path->owner_objectid ||
6228                     (key.objectid == ref_path->owner_objectid &&
6229                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6230                         key.objectid = ref_path->owner_objectid;
6231                         key.type = BTRFS_EXTENT_DATA_KEY;
6232                         key.offset = 0;
6233                 }
6234         }
6235
6236         while (1) {
6237                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6238                 if (ret < 0)
6239                         goto out;
6240
6241                 leaf = path->nodes[0];
6242                 nritems = btrfs_header_nritems(leaf);
6243 next:
6244                 if (extent_locked && ret > 0) {
6245                         /*
6246                          * the file extent item was modified by someone
6247                          * before the extent got locked.
6248                          */
6249                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6250                                       lock_end, GFP_NOFS);
6251                         extent_locked = 0;
6252                 }
6253
6254                 if (path->slots[0] >= nritems) {
6255                         if (++nr_scaned > 2)
6256                                 break;
6257
6258                         BUG_ON(extent_locked);
6259                         ret = btrfs_next_leaf(root, path);
6260                         if (ret < 0)
6261                                 goto out;
6262                         if (ret > 0)
6263                                 break;
6264                         leaf = path->nodes[0];
6265                         nritems = btrfs_header_nritems(leaf);
6266                 }
6267
6268                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6269
6270                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6271                         if ((key.objectid > ref_path->owner_objectid) ||
6272                             (key.objectid == ref_path->owner_objectid &&
6273                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6274                             key.offset >= search_end)
6275                                 break;
6276                 }
6277
6278                 if (inode && key.objectid != inode->i_ino) {
6279                         BUG_ON(extent_locked);
6280                         btrfs_release_path(root, path);
6281                         mutex_unlock(&inode->i_mutex);
6282                         iput(inode);
6283                         inode = NULL;
6284                         continue;
6285                 }
6286
6287                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6288                         path->slots[0]++;
6289                         ret = 1;
6290                         goto next;
6291                 }
6292                 fi = btrfs_item_ptr(leaf, path->slots[0],
6293                                     struct btrfs_file_extent_item);
6294                 extent_type = btrfs_file_extent_type(leaf, fi);
6295                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6296                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6297                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6298                      extent_key->objectid)) {
6299                         path->slots[0]++;
6300                         ret = 1;
6301                         goto next;
6302                 }
6303
6304                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6305                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6306
6307                 if (search_end == (u64)-1) {
6308                         search_end = key.offset - ext_offset +
6309                                 btrfs_file_extent_ram_bytes(leaf, fi);
6310                 }
6311
6312                 if (!extent_locked) {
6313                         lock_start = key.offset;
6314                         lock_end = lock_start + num_bytes - 1;
6315                 } else {
6316                         if (lock_start > key.offset ||
6317                             lock_end + 1 < key.offset + num_bytes) {
6318                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6319                                               lock_start, lock_end, GFP_NOFS);
6320                                 extent_locked = 0;
6321                         }
6322                 }
6323
6324                 if (!inode) {
6325                         btrfs_release_path(root, path);
6326
6327                         inode = btrfs_iget_locked(root->fs_info->sb,
6328                                                   key.objectid, root);
6329                         if (inode->i_state & I_NEW) {
6330                                 BTRFS_I(inode)->root = root;
6331                                 BTRFS_I(inode)->location.objectid =
6332                                         key.objectid;
6333                                 BTRFS_I(inode)->location.type =
6334                                         BTRFS_INODE_ITEM_KEY;
6335                                 BTRFS_I(inode)->location.offset = 0;
6336                                 btrfs_read_locked_inode(inode);
6337                                 unlock_new_inode(inode);
6338                         }
6339                         /*
6340                          * some code call btrfs_commit_transaction while
6341                          * holding the i_mutex, so we can't use mutex_lock
6342                          * here.
6343                          */
6344                         if (is_bad_inode(inode) ||
6345                             !mutex_trylock(&inode->i_mutex)) {
6346                                 iput(inode);
6347                                 inode = NULL;
6348                                 key.offset = (u64)-1;
6349                                 goto skip;
6350                         }
6351                 }
6352
6353                 if (!extent_locked) {
6354                         struct btrfs_ordered_extent *ordered;
6355
6356                         btrfs_release_path(root, path);
6357
6358                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6359                                     lock_end, GFP_NOFS);
6360                         ordered = btrfs_lookup_first_ordered_extent(inode,
6361                                                                     lock_end);
6362                         if (ordered &&
6363                             ordered->file_offset <= lock_end &&
6364                             ordered->file_offset + ordered->len > lock_start) {
6365                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6366                                               lock_start, lock_end, GFP_NOFS);
6367                                 btrfs_start_ordered_extent(inode, ordered, 1);
6368                                 btrfs_put_ordered_extent(ordered);
6369                                 key.offset += num_bytes;
6370                                 goto skip;
6371                         }
6372                         if (ordered)
6373                                 btrfs_put_ordered_extent(ordered);
6374
6375                         extent_locked = 1;
6376                         continue;
6377                 }
6378
6379                 if (nr_extents == 1) {
6380                         /* update extent pointer in place */
6381                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6382                                                 new_extents[0].disk_bytenr);
6383                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6384                                                 new_extents[0].disk_num_bytes);
6385                         btrfs_mark_buffer_dirty(leaf);
6386
6387                         btrfs_drop_extent_cache(inode, key.offset,
6388                                                 key.offset + num_bytes - 1, 0);
6389
6390                         ret = btrfs_inc_extent_ref(trans, root,
6391                                                 new_extents[0].disk_bytenr,
6392                                                 new_extents[0].disk_num_bytes,
6393                                                 leaf->start,
6394                                                 root->root_key.objectid,
6395                                                 trans->transid,
6396                                                 key.objectid);
6397                         BUG_ON(ret);
6398
6399                         ret = btrfs_free_extent(trans, root,
6400                                                 extent_key->objectid,
6401                                                 extent_key->offset,
6402                                                 leaf->start,
6403                                                 btrfs_header_owner(leaf),
6404                                                 btrfs_header_generation(leaf),
6405                                                 key.objectid, 0);
6406                         BUG_ON(ret);
6407
6408                         btrfs_release_path(root, path);
6409                         key.offset += num_bytes;
6410                 } else {
6411                         BUG_ON(1);
6412 #if 0
6413                         u64 alloc_hint;
6414                         u64 extent_len;
6415                         int i;
6416                         /*
6417                          * drop old extent pointer at first, then insert the
6418                          * new pointers one bye one
6419                          */
6420                         btrfs_release_path(root, path);
6421                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6422                                                  key.offset + num_bytes,
6423                                                  key.offset, &alloc_hint);
6424                         BUG_ON(ret);
6425
6426                         for (i = 0; i < nr_extents; i++) {
6427                                 if (ext_offset >= new_extents[i].num_bytes) {
6428                                         ext_offset -= new_extents[i].num_bytes;
6429                                         continue;
6430                                 }
6431                                 extent_len = min(new_extents[i].num_bytes -
6432                                                  ext_offset, num_bytes);
6433
6434                                 ret = btrfs_insert_empty_item(trans, root,
6435                                                               path, &key,
6436                                                               sizeof(*fi));
6437                                 BUG_ON(ret);
6438
6439                                 leaf = path->nodes[0];
6440                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6441                                                 struct btrfs_file_extent_item);
6442                                 btrfs_set_file_extent_generation(leaf, fi,
6443                                                         trans->transid);
6444                                 btrfs_set_file_extent_type(leaf, fi,
6445                                                         BTRFS_FILE_EXTENT_REG);
6446                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6447                                                 new_extents[i].disk_bytenr);
6448                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6449                                                 new_extents[i].disk_num_bytes);
6450                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6451                                                 new_extents[i].ram_bytes);
6452
6453                                 btrfs_set_file_extent_compression(leaf, fi,
6454                                                 new_extents[i].compression);
6455                                 btrfs_set_file_extent_encryption(leaf, fi,
6456                                                 new_extents[i].encryption);
6457                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6458                                                 new_extents[i].other_encoding);
6459
6460                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6461                                                         extent_len);
6462                                 ext_offset += new_extents[i].offset;
6463                                 btrfs_set_file_extent_offset(leaf, fi,
6464                                                         ext_offset);
6465                                 btrfs_mark_buffer_dirty(leaf);
6466
6467                                 btrfs_drop_extent_cache(inode, key.offset,
6468                                                 key.offset + extent_len - 1, 0);
6469
6470                                 ret = btrfs_inc_extent_ref(trans, root,
6471                                                 new_extents[i].disk_bytenr,
6472                                                 new_extents[i].disk_num_bytes,
6473                                                 leaf->start,
6474                                                 root->root_key.objectid,
6475                                                 trans->transid, key.objectid);
6476                                 BUG_ON(ret);
6477                                 btrfs_release_path(root, path);
6478
6479                                 inode_add_bytes(inode, extent_len);
6480
6481                                 ext_offset = 0;
6482                                 num_bytes -= extent_len;
6483                                 key.offset += extent_len;
6484
6485                                 if (num_bytes == 0)
6486                                         break;
6487                         }
6488                         BUG_ON(i >= nr_extents);
6489 #endif
6490                 }
6491
6492                 if (extent_locked) {
6493                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6494                                       lock_end, GFP_NOFS);
6495                         extent_locked = 0;
6496                 }
6497 skip:
6498                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6499                     key.offset >= search_end)
6500                         break;
6501
6502                 cond_resched();
6503         }
6504         ret = 0;
6505 out:
6506         btrfs_release_path(root, path);
6507         if (inode) {
6508                 mutex_unlock(&inode->i_mutex);
6509                 if (extent_locked) {
6510                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6511                                       lock_end, GFP_NOFS);
6512                 }
6513                 iput(inode);
6514         }
6515         return ret;
6516 }
6517
6518 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6519                                struct btrfs_root *root,
6520                                struct extent_buffer *buf, u64 orig_start)
6521 {
6522         int level;
6523         int ret;
6524
6525         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6526         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6527
6528         level = btrfs_header_level(buf);
6529         if (level == 0) {
6530                 struct btrfs_leaf_ref *ref;
6531                 struct btrfs_leaf_ref *orig_ref;
6532
6533                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6534                 if (!orig_ref)
6535                         return -ENOENT;
6536
6537                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6538                 if (!ref) {
6539                         btrfs_free_leaf_ref(root, orig_ref);
6540                         return -ENOMEM;
6541                 }
6542
6543                 ref->nritems = orig_ref->nritems;
6544                 memcpy(ref->extents, orig_ref->extents,
6545                         sizeof(ref->extents[0]) * ref->nritems);
6546
6547                 btrfs_free_leaf_ref(root, orig_ref);
6548
6549                 ref->root_gen = trans->transid;
6550                 ref->bytenr = buf->start;
6551                 ref->owner = btrfs_header_owner(buf);
6552                 ref->generation = btrfs_header_generation(buf);
6553
6554                 ret = btrfs_add_leaf_ref(root, ref, 0);
6555                 WARN_ON(ret);
6556                 btrfs_free_leaf_ref(root, ref);
6557         }
6558         return 0;
6559 }
6560
6561 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6562                                         struct extent_buffer *leaf,
6563                                         struct btrfs_block_group_cache *group,
6564                                         struct btrfs_root *target_root)
6565 {
6566         struct btrfs_key key;
6567         struct inode *inode = NULL;
6568         struct btrfs_file_extent_item *fi;
6569         struct extent_state *cached_state = NULL;
6570         u64 num_bytes;
6571         u64 skip_objectid = 0;
6572         u32 nritems;
6573         u32 i;
6574
6575         nritems = btrfs_header_nritems(leaf);
6576         for (i = 0; i < nritems; i++) {
6577                 btrfs_item_key_to_cpu(leaf, &key, i);
6578                 if (key.objectid == skip_objectid ||
6579                     key.type != BTRFS_EXTENT_DATA_KEY)
6580                         continue;
6581                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6582                 if (btrfs_file_extent_type(leaf, fi) ==
6583                     BTRFS_FILE_EXTENT_INLINE)
6584                         continue;
6585                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6586                         continue;
6587                 if (!inode || inode->i_ino != key.objectid) {
6588                         iput(inode);
6589                         inode = btrfs_ilookup(target_root->fs_info->sb,
6590                                               key.objectid, target_root, 1);
6591                 }
6592                 if (!inode) {
6593                         skip_objectid = key.objectid;
6594                         continue;
6595                 }
6596                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6597
6598                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
6599                                  key.offset + num_bytes - 1, 0, &cached_state,
6600                                  GFP_NOFS);
6601                 btrfs_drop_extent_cache(inode, key.offset,
6602                                         key.offset + num_bytes - 1, 1);
6603                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
6604                                      key.offset + num_bytes - 1, &cached_state,
6605                                      GFP_NOFS);
6606                 cond_resched();
6607         }
6608         iput(inode);
6609         return 0;
6610 }
6611
6612 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6613                                         struct btrfs_root *root,
6614                                         struct extent_buffer *leaf,
6615                                         struct btrfs_block_group_cache *group,
6616                                         struct inode *reloc_inode)
6617 {
6618         struct btrfs_key key;
6619         struct btrfs_key extent_key;
6620         struct btrfs_file_extent_item *fi;
6621         struct btrfs_leaf_ref *ref;
6622         struct disk_extent *new_extent;
6623         u64 bytenr;
6624         u64 num_bytes;
6625         u32 nritems;
6626         u32 i;
6627         int ext_index;
6628         int nr_extent;
6629         int ret;
6630
6631         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6632         BUG_ON(!new_extent);
6633
6634         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6635         BUG_ON(!ref);
6636
6637         ext_index = -1;
6638         nritems = btrfs_header_nritems(leaf);
6639         for (i = 0; i < nritems; i++) {
6640                 btrfs_item_key_to_cpu(leaf, &key, i);
6641                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6642                         continue;
6643                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6644                 if (btrfs_file_extent_type(leaf, fi) ==
6645                     BTRFS_FILE_EXTENT_INLINE)
6646                         continue;
6647                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6648                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6649                 if (bytenr == 0)
6650                         continue;
6651
6652                 ext_index++;
6653                 if (bytenr >= group->key.objectid + group->key.offset ||
6654                     bytenr + num_bytes <= group->key.objectid)
6655                         continue;
6656
6657                 extent_key.objectid = bytenr;
6658                 extent_key.offset = num_bytes;
6659                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6660                 nr_extent = 1;
6661                 ret = get_new_locations(reloc_inode, &extent_key,
6662                                         group->key.objectid, 1,
6663                                         &new_extent, &nr_extent);
6664                 if (ret > 0)
6665                         continue;
6666                 BUG_ON(ret < 0);
6667
6668                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6669                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6670                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6671                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6672
6673                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6674                                                 new_extent->disk_bytenr);
6675                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6676                                                 new_extent->disk_num_bytes);
6677                 btrfs_mark_buffer_dirty(leaf);
6678
6679                 ret = btrfs_inc_extent_ref(trans, root,
6680                                         new_extent->disk_bytenr,
6681                                         new_extent->disk_num_bytes,
6682                                         leaf->start,
6683                                         root->root_key.objectid,
6684                                         trans->transid, key.objectid);
6685                 BUG_ON(ret);
6686
6687                 ret = btrfs_free_extent(trans, root,
6688                                         bytenr, num_bytes, leaf->start,
6689                                         btrfs_header_owner(leaf),
6690                                         btrfs_header_generation(leaf),
6691                                         key.objectid, 0);
6692                 BUG_ON(ret);
6693                 cond_resched();
6694         }
6695         kfree(new_extent);
6696         BUG_ON(ext_index + 1 != ref->nritems);
6697         btrfs_free_leaf_ref(root, ref);
6698         return 0;
6699 }
6700
6701 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6702                           struct btrfs_root *root)
6703 {
6704         struct btrfs_root *reloc_root;
6705         int ret;
6706
6707         if (root->reloc_root) {
6708                 reloc_root = root->reloc_root;
6709                 root->reloc_root = NULL;
6710                 list_add(&reloc_root->dead_list,
6711                          &root->fs_info->dead_reloc_roots);
6712
6713                 btrfs_set_root_bytenr(&reloc_root->root_item,
6714                                       reloc_root->node->start);
6715                 btrfs_set_root_level(&root->root_item,
6716                                      btrfs_header_level(reloc_root->node));
6717                 memset(&reloc_root->root_item.drop_progress, 0,
6718                         sizeof(struct btrfs_disk_key));
6719                 reloc_root->root_item.drop_level = 0;
6720
6721                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6722                                         &reloc_root->root_key,
6723                                         &reloc_root->root_item);
6724                 BUG_ON(ret);
6725         }
6726         return 0;
6727 }
6728
6729 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6730 {
6731         struct btrfs_trans_handle *trans;
6732         struct btrfs_root *reloc_root;
6733         struct btrfs_root *prev_root = NULL;
6734         struct list_head dead_roots;
6735         int ret;
6736         unsigned long nr;
6737
6738         INIT_LIST_HEAD(&dead_roots);
6739         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6740
6741         while (!list_empty(&dead_roots)) {
6742                 reloc_root = list_entry(dead_roots.prev,
6743                                         struct btrfs_root, dead_list);
6744                 list_del_init(&reloc_root->dead_list);
6745
6746                 BUG_ON(reloc_root->commit_root != NULL);
6747                 while (1) {
6748                         trans = btrfs_join_transaction(root, 1);
6749                         BUG_ON(!trans);
6750
6751                         mutex_lock(&root->fs_info->drop_mutex);
6752                         ret = btrfs_drop_snapshot(trans, reloc_root);
6753                         if (ret != -EAGAIN)
6754                                 break;
6755                         mutex_unlock(&root->fs_info->drop_mutex);
6756
6757                         nr = trans->blocks_used;
6758                         ret = btrfs_end_transaction(trans, root);
6759                         BUG_ON(ret);
6760                         btrfs_btree_balance_dirty(root, nr);
6761                 }
6762
6763                 free_extent_buffer(reloc_root->node);
6764
6765                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6766                                      &reloc_root->root_key);
6767                 BUG_ON(ret);
6768                 mutex_unlock(&root->fs_info->drop_mutex);
6769
6770                 nr = trans->blocks_used;
6771                 ret = btrfs_end_transaction(trans, root);
6772                 BUG_ON(ret);
6773                 btrfs_btree_balance_dirty(root, nr);
6774
6775                 kfree(prev_root);
6776                 prev_root = reloc_root;
6777         }
6778         if (prev_root) {
6779                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6780                 kfree(prev_root);
6781         }
6782         return 0;
6783 }
6784
6785 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6786 {
6787         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6788         return 0;
6789 }
6790
6791 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6792 {
6793         struct btrfs_root *reloc_root;
6794         struct btrfs_trans_handle *trans;
6795         struct btrfs_key location;
6796         int found;
6797         int ret;
6798
6799         mutex_lock(&root->fs_info->tree_reloc_mutex);
6800         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6801         BUG_ON(ret);
6802         found = !list_empty(&root->fs_info->dead_reloc_roots);
6803         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6804
6805         if (found) {
6806                 trans = btrfs_start_transaction(root, 1);
6807                 BUG_ON(!trans);
6808                 ret = btrfs_commit_transaction(trans, root);
6809                 BUG_ON(ret);
6810         }
6811
6812         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6813         location.offset = (u64)-1;
6814         location.type = BTRFS_ROOT_ITEM_KEY;
6815
6816         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6817         BUG_ON(!reloc_root);
6818         btrfs_orphan_cleanup(reloc_root);
6819         return 0;
6820 }
6821
6822 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6823                                     struct btrfs_root *root)
6824 {
6825         struct btrfs_root *reloc_root;
6826         struct extent_buffer *eb;
6827         struct btrfs_root_item *root_item;
6828         struct btrfs_key root_key;
6829         int ret;
6830
6831         BUG_ON(!root->ref_cows);
6832         if (root->reloc_root)
6833                 return 0;
6834
6835         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6836         BUG_ON(!root_item);
6837
6838         ret = btrfs_copy_root(trans, root, root->commit_root,
6839                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6840         BUG_ON(ret);
6841
6842         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6843         root_key.offset = root->root_key.objectid;
6844         root_key.type = BTRFS_ROOT_ITEM_KEY;
6845
6846         memcpy(root_item, &root->root_item, sizeof(root_item));
6847         btrfs_set_root_refs(root_item, 0);
6848         btrfs_set_root_bytenr(root_item, eb->start);
6849         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6850         btrfs_set_root_generation(root_item, trans->transid);
6851
6852         btrfs_tree_unlock(eb);
6853         free_extent_buffer(eb);
6854
6855         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6856                                 &root_key, root_item);
6857         BUG_ON(ret);
6858         kfree(root_item);
6859
6860         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6861                                                  &root_key);
6862         BUG_ON(!reloc_root);
6863         reloc_root->last_trans = trans->transid;
6864         reloc_root->commit_root = NULL;
6865         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6866
6867         root->reloc_root = reloc_root;
6868         return 0;
6869 }
6870
6871 /*
6872  * Core function of space balance.
6873  *
6874  * The idea is using reloc trees to relocate tree blocks in reference
6875  * counted roots. There is one reloc tree for each subvol, and all
6876  * reloc trees share same root key objectid. Reloc trees are snapshots
6877  * of the latest committed roots of subvols (root->commit_root).
6878  *
6879  * To relocate a tree block referenced by a subvol, there are two steps.
6880  * COW the block through subvol's reloc tree, then update block pointer
6881  * in the subvol to point to the new block. Since all reloc trees share
6882  * same root key objectid, doing special handing for tree blocks owned
6883  * by them is easy. Once a tree block has been COWed in one reloc tree,
6884  * we can use the resulting new block directly when the same block is
6885  * required to COW again through other reloc trees. By this way, relocated
6886  * tree blocks are shared between reloc trees, so they are also shared
6887  * between subvols.
6888  */
6889 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6890                                       struct btrfs_root *root,
6891                                       struct btrfs_path *path,
6892                                       struct btrfs_key *first_key,
6893                                       struct btrfs_ref_path *ref_path,
6894                                       struct btrfs_block_group_cache *group,
6895                                       struct inode *reloc_inode)
6896 {
6897         struct btrfs_root *reloc_root;
6898         struct extent_buffer *eb = NULL;
6899         struct btrfs_key *keys;
6900         u64 *nodes;
6901         int level;
6902         int shared_level;
6903         int lowest_level = 0;
6904         int ret;
6905
6906         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6907                 lowest_level = ref_path->owner_objectid;
6908
6909         if (!root->ref_cows) {
6910                 path->lowest_level = lowest_level;
6911                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6912                 BUG_ON(ret < 0);
6913                 path->lowest_level = 0;
6914                 btrfs_release_path(root, path);
6915                 return 0;
6916         }
6917
6918         mutex_lock(&root->fs_info->tree_reloc_mutex);
6919         ret = init_reloc_tree(trans, root);
6920         BUG_ON(ret);
6921         reloc_root = root->reloc_root;
6922
6923         shared_level = ref_path->shared_level;
6924         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6925
6926         keys = ref_path->node_keys;
6927         nodes = ref_path->new_nodes;
6928         memset(&keys[shared_level + 1], 0,
6929                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6930         memset(&nodes[shared_level + 1], 0,
6931                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6932
6933         if (nodes[lowest_level] == 0) {
6934                 path->lowest_level = lowest_level;
6935                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6936                                         0, 1);
6937                 BUG_ON(ret);
6938                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6939                         eb = path->nodes[level];
6940                         if (!eb || eb == reloc_root->node)
6941                                 break;
6942                         nodes[level] = eb->start;
6943                         if (level == 0)
6944                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6945                         else
6946                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6947                 }
6948                 if (nodes[0] &&
6949                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6950                         eb = path->nodes[0];
6951                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6952                                                       group, reloc_inode);
6953                         BUG_ON(ret);
6954                 }
6955                 btrfs_release_path(reloc_root, path);
6956         } else {
6957                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6958                                        lowest_level);
6959                 BUG_ON(ret);
6960         }
6961
6962         /*
6963          * replace tree blocks in the fs tree with tree blocks in
6964          * the reloc tree.
6965          */
6966         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6967         BUG_ON(ret < 0);
6968
6969         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6970                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6971                                         0, 0);
6972                 BUG_ON(ret);
6973                 extent_buffer_get(path->nodes[0]);
6974                 eb = path->nodes[0];
6975                 btrfs_release_path(reloc_root, path);
6976                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6977                 BUG_ON(ret);
6978                 free_extent_buffer(eb);
6979         }
6980
6981         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6982         path->lowest_level = 0;
6983         return 0;
6984 }
6985
6986 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6987                                         struct btrfs_root *root,
6988                                         struct btrfs_path *path,
6989                                         struct btrfs_key *first_key,
6990                                         struct btrfs_ref_path *ref_path)
6991 {
6992         int ret;
6993
6994         ret = relocate_one_path(trans, root, path, first_key,
6995                                 ref_path, NULL, NULL);
6996         BUG_ON(ret);
6997
6998         return 0;
6999 }
7000
7001 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7002                                     struct btrfs_root *extent_root,
7003                                     struct btrfs_path *path,
7004                                     struct btrfs_key *extent_key)
7005 {
7006         int ret;
7007
7008         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7009         if (ret)
7010                 goto out;
7011         ret = btrfs_del_item(trans, extent_root, path);
7012 out:
7013         btrfs_release_path(extent_root, path);
7014         return ret;
7015 }
7016
7017 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7018                                                 struct btrfs_ref_path *ref_path)
7019 {
7020         struct btrfs_key root_key;
7021
7022         root_key.objectid = ref_path->root_objectid;
7023         root_key.type = BTRFS_ROOT_ITEM_KEY;
7024         if (is_cowonly_root(ref_path->root_objectid))
7025                 root_key.offset = 0;
7026         else
7027                 root_key.offset = (u64)-1;
7028
7029         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7030 }
7031
7032 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7033                                         struct btrfs_path *path,
7034                                         struct btrfs_key *extent_key,
7035                                         struct btrfs_block_group_cache *group,
7036                                         struct inode *reloc_inode, int pass)
7037 {
7038         struct btrfs_trans_handle *trans;
7039         struct btrfs_root *found_root;
7040         struct btrfs_ref_path *ref_path = NULL;
7041         struct disk_extent *new_extents = NULL;
7042         int nr_extents = 0;
7043         int loops;
7044         int ret;
7045         int level;
7046         struct btrfs_key first_key;
7047         u64 prev_block = 0;
7048
7049
7050         trans = btrfs_start_transaction(extent_root, 1);
7051         BUG_ON(!trans);
7052
7053         if (extent_key->objectid == 0) {
7054                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7055                 goto out;
7056         }
7057
7058         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7059         if (!ref_path) {
7060                 ret = -ENOMEM;
7061                 goto out;
7062         }
7063
7064         for (loops = 0; ; loops++) {
7065                 if (loops == 0) {
7066                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7067                                                    extent_key->objectid);
7068                 } else {
7069                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7070                 }
7071                 if (ret < 0)
7072                         goto out;
7073                 if (ret > 0)
7074                         break;
7075
7076                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7077                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7078                         continue;
7079
7080                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7081                 BUG_ON(!found_root);
7082                 /*
7083                  * for reference counted tree, only process reference paths
7084                  * rooted at the latest committed root.
7085                  */
7086                 if (found_root->ref_cows &&
7087                     ref_path->root_generation != found_root->root_key.offset)
7088                         continue;
7089
7090                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7091                         if (pass == 0) {
7092                                 /*
7093                                  * copy data extents to new locations
7094                                  */
7095                                 u64 group_start = group->key.objectid;
7096                                 ret = relocate_data_extent(reloc_inode,
7097                                                            extent_key,
7098                                                            group_start);
7099                                 if (ret < 0)
7100                                         goto out;
7101                                 break;
7102                         }
7103                         level = 0;
7104                 } else {
7105                         level = ref_path->owner_objectid;
7106                 }
7107
7108                 if (prev_block != ref_path->nodes[level]) {
7109                         struct extent_buffer *eb;
7110                         u64 block_start = ref_path->nodes[level];
7111                         u64 block_size = btrfs_level_size(found_root, level);
7112
7113                         eb = read_tree_block(found_root, block_start,
7114                                              block_size, 0);
7115                         btrfs_tree_lock(eb);
7116                         BUG_ON(level != btrfs_header_level(eb));
7117
7118                         if (level == 0)
7119                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7120                         else
7121                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7122
7123                         btrfs_tree_unlock(eb);
7124                         free_extent_buffer(eb);
7125                         prev_block = block_start;
7126                 }
7127
7128                 mutex_lock(&extent_root->fs_info->trans_mutex);
7129                 btrfs_record_root_in_trans(found_root);
7130                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7131                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7132                         /*
7133                          * try to update data extent references while
7134                          * keeping metadata shared between snapshots.
7135                          */
7136                         if (pass == 1) {
7137                                 ret = relocate_one_path(trans, found_root,
7138                                                 path, &first_key, ref_path,
7139                                                 group, reloc_inode);
7140                                 if (ret < 0)
7141                                         goto out;
7142                                 continue;
7143                         }
7144                         /*
7145                          * use fallback method to process the remaining
7146                          * references.
7147                          */
7148                         if (!new_extents) {
7149                                 u64 group_start = group->key.objectid;
7150                                 new_extents = kmalloc(sizeof(*new_extents),
7151                                                       GFP_NOFS);
7152                                 nr_extents = 1;
7153                                 ret = get_new_locations(reloc_inode,
7154                                                         extent_key,
7155                                                         group_start, 1,
7156                                                         &new_extents,
7157                                                         &nr_extents);
7158                                 if (ret)
7159                                         goto out;
7160                         }
7161                         ret = replace_one_extent(trans, found_root,
7162                                                 path, extent_key,
7163                                                 &first_key, ref_path,
7164                                                 new_extents, nr_extents);
7165                 } else {
7166                         ret = relocate_tree_block(trans, found_root, path,
7167                                                   &first_key, ref_path);
7168                 }
7169                 if (ret < 0)
7170                         goto out;
7171         }
7172         ret = 0;
7173 out:
7174         btrfs_end_transaction(trans, extent_root);
7175         kfree(new_extents);
7176         kfree(ref_path);
7177         return ret;
7178 }
7179 #endif
7180
7181 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7182 {
7183         u64 num_devices;
7184         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7185                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7186
7187         num_devices = root->fs_info->fs_devices->rw_devices;
7188         if (num_devices == 1) {
7189                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7190                 stripped = flags & ~stripped;
7191
7192                 /* turn raid0 into single device chunks */
7193                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7194                         return stripped;
7195
7196                 /* turn mirroring into duplication */
7197                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7198                              BTRFS_BLOCK_GROUP_RAID10))
7199                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7200                 return flags;
7201         } else {
7202                 /* they already had raid on here, just return */
7203                 if (flags & stripped)
7204                         return flags;
7205
7206                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7207                 stripped = flags & ~stripped;
7208
7209                 /* switch duplicated blocks with raid1 */
7210                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7211                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7212
7213                 /* turn single device chunks into raid0 */
7214                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7215         }
7216         return flags;
7217 }
7218
7219 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7220                      struct btrfs_block_group_cache *shrink_block_group,
7221                      int force)
7222 {
7223         struct btrfs_trans_handle *trans;
7224         u64 new_alloc_flags;
7225         u64 calc;
7226
7227         spin_lock(&shrink_block_group->lock);
7228         if (btrfs_block_group_used(&shrink_block_group->item) +
7229             shrink_block_group->reserved > 0) {
7230                 spin_unlock(&shrink_block_group->lock);
7231
7232                 trans = btrfs_start_transaction(root, 1);
7233                 spin_lock(&shrink_block_group->lock);
7234
7235                 new_alloc_flags = update_block_group_flags(root,
7236                                                    shrink_block_group->flags);
7237                 if (new_alloc_flags != shrink_block_group->flags) {
7238                         calc =
7239                              btrfs_block_group_used(&shrink_block_group->item);
7240                 } else {
7241                         calc = shrink_block_group->key.offset;
7242                 }
7243                 spin_unlock(&shrink_block_group->lock);
7244
7245                 do_chunk_alloc(trans, root->fs_info->extent_root,
7246                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7247
7248                 btrfs_end_transaction(trans, root);
7249         } else
7250                 spin_unlock(&shrink_block_group->lock);
7251         return 0;
7252 }
7253
7254
7255 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7256                                          struct btrfs_block_group_cache *group)
7257
7258 {
7259         __alloc_chunk_for_shrink(root, group, 1);
7260         set_block_group_readonly(group);
7261         return 0;
7262 }
7263
7264 /*
7265  * checks to see if its even possible to relocate this block group.
7266  *
7267  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7268  * ok to go ahead and try.
7269  */
7270 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7271 {
7272         struct btrfs_block_group_cache *block_group;
7273         struct btrfs_space_info *space_info;
7274         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7275         struct btrfs_device *device;
7276         int full = 0;
7277         int ret = 0;
7278
7279         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7280
7281         /* odd, couldn't find the block group, leave it alone */
7282         if (!block_group)
7283                 return -1;
7284
7285         /* no bytes used, we're good */
7286         if (!btrfs_block_group_used(&block_group->item))
7287                 goto out;
7288
7289         space_info = block_group->space_info;
7290         spin_lock(&space_info->lock);
7291
7292         full = space_info->full;
7293
7294         /*
7295          * if this is the last block group we have in this space, we can't
7296          * relocate it unless we're able to allocate a new chunk below.
7297          *
7298          * Otherwise, we need to make sure we have room in the space to handle
7299          * all of the extents from this block group.  If we can, we're good
7300          */
7301         if ((space_info->total_bytes != block_group->key.offset) &&
7302            (space_info->bytes_used + space_info->bytes_reserved +
7303             space_info->bytes_pinned + space_info->bytes_readonly +
7304             btrfs_block_group_used(&block_group->item) <
7305             space_info->total_bytes)) {
7306                 spin_unlock(&space_info->lock);
7307                 goto out;
7308         }
7309         spin_unlock(&space_info->lock);
7310
7311         /*
7312          * ok we don't have enough space, but maybe we have free space on our
7313          * devices to allocate new chunks for relocation, so loop through our
7314          * alloc devices and guess if we have enough space.  However, if we
7315          * were marked as full, then we know there aren't enough chunks, and we
7316          * can just return.
7317          */
7318         ret = -1;
7319         if (full)
7320                 goto out;
7321
7322         mutex_lock(&root->fs_info->chunk_mutex);
7323         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7324                 u64 min_free = btrfs_block_group_used(&block_group->item);
7325                 u64 dev_offset, max_avail;
7326
7327                 /*
7328                  * check to make sure we can actually find a chunk with enough
7329                  * space to fit our block group in.
7330                  */
7331                 if (device->total_bytes > device->bytes_used + min_free) {
7332                         ret = find_free_dev_extent(NULL, device, min_free,
7333                                                    &dev_offset, &max_avail);
7334                         if (!ret)
7335                                 break;
7336                         ret = -1;
7337                 }
7338         }
7339         mutex_unlock(&root->fs_info->chunk_mutex);
7340 out:
7341         btrfs_put_block_group(block_group);
7342         return ret;
7343 }
7344
7345 static int find_first_block_group(struct btrfs_root *root,
7346                 struct btrfs_path *path, struct btrfs_key *key)
7347 {
7348         int ret = 0;
7349         struct btrfs_key found_key;
7350         struct extent_buffer *leaf;
7351         int slot;
7352
7353         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7354         if (ret < 0)
7355                 goto out;
7356
7357         while (1) {
7358                 slot = path->slots[0];
7359                 leaf = path->nodes[0];
7360                 if (slot >= btrfs_header_nritems(leaf)) {
7361                         ret = btrfs_next_leaf(root, path);
7362                         if (ret == 0)
7363                                 continue;
7364                         if (ret < 0)
7365                                 goto out;
7366                         break;
7367                 }
7368                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7369
7370                 if (found_key.objectid >= key->objectid &&
7371                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7372                         ret = 0;
7373                         goto out;
7374                 }
7375                 path->slots[0]++;
7376         }
7377 out:
7378         return ret;
7379 }
7380
7381 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7382 {
7383         struct btrfs_block_group_cache *block_group;
7384         struct btrfs_space_info *space_info;
7385         struct btrfs_caching_control *caching_ctl;
7386         struct rb_node *n;
7387
7388         down_write(&info->extent_commit_sem);
7389         while (!list_empty(&info->caching_block_groups)) {
7390                 caching_ctl = list_entry(info->caching_block_groups.next,
7391                                          struct btrfs_caching_control, list);
7392                 list_del(&caching_ctl->list);
7393                 put_caching_control(caching_ctl);
7394         }
7395         up_write(&info->extent_commit_sem);
7396
7397         spin_lock(&info->block_group_cache_lock);
7398         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7399                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7400                                        cache_node);
7401                 rb_erase(&block_group->cache_node,
7402                          &info->block_group_cache_tree);
7403                 spin_unlock(&info->block_group_cache_lock);
7404
7405                 down_write(&block_group->space_info->groups_sem);
7406                 list_del(&block_group->list);
7407                 up_write(&block_group->space_info->groups_sem);
7408
7409                 if (block_group->cached == BTRFS_CACHE_STARTED)
7410                         wait_block_group_cache_done(block_group);
7411
7412                 btrfs_remove_free_space_cache(block_group);
7413                 btrfs_put_block_group(block_group);
7414
7415                 spin_lock(&info->block_group_cache_lock);
7416         }
7417         spin_unlock(&info->block_group_cache_lock);
7418
7419         /* now that all the block groups are freed, go through and
7420          * free all the space_info structs.  This is only called during
7421          * the final stages of unmount, and so we know nobody is
7422          * using them.  We call synchronize_rcu() once before we start,
7423          * just to be on the safe side.
7424          */
7425         synchronize_rcu();
7426
7427         while(!list_empty(&info->space_info)) {
7428                 space_info = list_entry(info->space_info.next,
7429                                         struct btrfs_space_info,
7430                                         list);
7431
7432                 list_del(&space_info->list);
7433                 kfree(space_info);
7434         }
7435         return 0;
7436 }
7437
7438 static void __link_block_group(struct btrfs_space_info *space_info,
7439                                struct btrfs_block_group_cache *cache)
7440 {
7441         int index = get_block_group_index(cache);
7442
7443         down_write(&space_info->groups_sem);
7444         list_add_tail(&cache->list, &space_info->block_groups[index]);
7445         up_write(&space_info->groups_sem);
7446 }
7447
7448 int btrfs_read_block_groups(struct btrfs_root *root)
7449 {
7450         struct btrfs_path *path;
7451         int ret;
7452         struct btrfs_block_group_cache *cache;
7453         struct btrfs_fs_info *info = root->fs_info;
7454         struct btrfs_space_info *space_info;
7455         struct btrfs_key key;
7456         struct btrfs_key found_key;
7457         struct extent_buffer *leaf;
7458
7459         root = info->extent_root;
7460         key.objectid = 0;
7461         key.offset = 0;
7462         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7463         path = btrfs_alloc_path();
7464         if (!path)
7465                 return -ENOMEM;
7466
7467         while (1) {
7468                 ret = find_first_block_group(root, path, &key);
7469                 if (ret > 0)
7470                         break;
7471                 if (ret != 0)
7472                         goto error;
7473
7474                 leaf = path->nodes[0];
7475                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7476                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7477                 if (!cache) {
7478                         ret = -ENOMEM;
7479                         break;
7480                 }
7481
7482                 atomic_set(&cache->count, 1);
7483                 spin_lock_init(&cache->lock);
7484                 spin_lock_init(&cache->tree_lock);
7485                 cache->fs_info = info;
7486                 INIT_LIST_HEAD(&cache->list);
7487                 INIT_LIST_HEAD(&cache->cluster_list);
7488
7489                 /*
7490                  * we only want to have 32k of ram per block group for keeping
7491                  * track of free space, and if we pass 1/2 of that we want to
7492                  * start converting things over to using bitmaps
7493                  */
7494                 cache->extents_thresh = ((1024 * 32) / 2) /
7495                         sizeof(struct btrfs_free_space);
7496
7497                 read_extent_buffer(leaf, &cache->item,
7498                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7499                                    sizeof(cache->item));
7500                 memcpy(&cache->key, &found_key, sizeof(found_key));
7501
7502                 key.objectid = found_key.objectid + found_key.offset;
7503                 btrfs_release_path(root, path);
7504                 cache->flags = btrfs_block_group_flags(&cache->item);
7505                 cache->sectorsize = root->sectorsize;
7506
7507                 /*
7508                  * check for two cases, either we are full, and therefore
7509                  * don't need to bother with the caching work since we won't
7510                  * find any space, or we are empty, and we can just add all
7511                  * the space in and be done with it.  This saves us _alot_ of
7512                  * time, particularly in the full case.
7513                  */
7514                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7515                         exclude_super_stripes(root, cache);
7516                         cache->last_byte_to_unpin = (u64)-1;
7517                         cache->cached = BTRFS_CACHE_FINISHED;
7518                         free_excluded_extents(root, cache);
7519                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7520                         exclude_super_stripes(root, cache);
7521                         cache->last_byte_to_unpin = (u64)-1;
7522                         cache->cached = BTRFS_CACHE_FINISHED;
7523                         add_new_free_space(cache, root->fs_info,
7524                                            found_key.objectid,
7525                                            found_key.objectid +
7526                                            found_key.offset);
7527                         free_excluded_extents(root, cache);
7528                 }
7529
7530                 ret = update_space_info(info, cache->flags, found_key.offset,
7531                                         btrfs_block_group_used(&cache->item),
7532                                         &space_info);
7533                 BUG_ON(ret);
7534                 cache->space_info = space_info;
7535                 spin_lock(&cache->space_info->lock);
7536                 cache->space_info->bytes_super += cache->bytes_super;
7537                 spin_unlock(&cache->space_info->lock);
7538
7539                 __link_block_group(space_info, cache);
7540
7541                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7542                 BUG_ON(ret);
7543
7544                 set_avail_alloc_bits(root->fs_info, cache->flags);
7545                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7546                         set_block_group_readonly(cache);
7547         }
7548
7549         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7550                 if (!(get_alloc_profile(root, space_info->flags) &
7551                       (BTRFS_BLOCK_GROUP_RAID10 |
7552                        BTRFS_BLOCK_GROUP_RAID1 |
7553                        BTRFS_BLOCK_GROUP_DUP)))
7554                         continue;
7555                 /*
7556                  * avoid allocating from un-mirrored block group if there are
7557                  * mirrored block groups.
7558                  */
7559                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7560                         set_block_group_readonly(cache);
7561                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7562                         set_block_group_readonly(cache);
7563         }
7564         ret = 0;
7565 error:
7566         btrfs_free_path(path);
7567         return ret;
7568 }
7569
7570 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7571                            struct btrfs_root *root, u64 bytes_used,
7572                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7573                            u64 size)
7574 {
7575         int ret;
7576         struct btrfs_root *extent_root;
7577         struct btrfs_block_group_cache *cache;
7578
7579         extent_root = root->fs_info->extent_root;
7580
7581         root->fs_info->last_trans_log_full_commit = trans->transid;
7582
7583         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7584         if (!cache)
7585                 return -ENOMEM;
7586
7587         cache->key.objectid = chunk_offset;
7588         cache->key.offset = size;
7589         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7590         cache->sectorsize = root->sectorsize;
7591
7592         /*
7593          * we only want to have 32k of ram per block group for keeping track
7594          * of free space, and if we pass 1/2 of that we want to start
7595          * converting things over to using bitmaps
7596          */
7597         cache->extents_thresh = ((1024 * 32) / 2) /
7598                 sizeof(struct btrfs_free_space);
7599         atomic_set(&cache->count, 1);
7600         spin_lock_init(&cache->lock);
7601         spin_lock_init(&cache->tree_lock);
7602         INIT_LIST_HEAD(&cache->list);
7603         INIT_LIST_HEAD(&cache->cluster_list);
7604
7605         btrfs_set_block_group_used(&cache->item, bytes_used);
7606         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7607         cache->flags = type;
7608         btrfs_set_block_group_flags(&cache->item, type);
7609
7610         cache->last_byte_to_unpin = (u64)-1;
7611         cache->cached = BTRFS_CACHE_FINISHED;
7612         exclude_super_stripes(root, cache);
7613
7614         add_new_free_space(cache, root->fs_info, chunk_offset,
7615                            chunk_offset + size);
7616
7617         free_excluded_extents(root, cache);
7618
7619         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7620                                 &cache->space_info);
7621         BUG_ON(ret);
7622
7623         spin_lock(&cache->space_info->lock);
7624         cache->space_info->bytes_super += cache->bytes_super;
7625         spin_unlock(&cache->space_info->lock);
7626
7627         __link_block_group(cache->space_info, cache);
7628
7629         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7630         BUG_ON(ret);
7631
7632         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7633                                 sizeof(cache->item));
7634         BUG_ON(ret);
7635
7636         set_avail_alloc_bits(extent_root->fs_info, type);
7637
7638         return 0;
7639 }
7640
7641 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7642                              struct btrfs_root *root, u64 group_start)
7643 {
7644         struct btrfs_path *path;
7645         struct btrfs_block_group_cache *block_group;
7646         struct btrfs_free_cluster *cluster;
7647         struct btrfs_key key;
7648         int ret;
7649
7650         root = root->fs_info->extent_root;
7651
7652         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7653         BUG_ON(!block_group);
7654         BUG_ON(!block_group->ro);
7655
7656         memcpy(&key, &block_group->key, sizeof(key));
7657
7658         /* make sure this block group isn't part of an allocation cluster */
7659         cluster = &root->fs_info->data_alloc_cluster;
7660         spin_lock(&cluster->refill_lock);
7661         btrfs_return_cluster_to_free_space(block_group, cluster);
7662         spin_unlock(&cluster->refill_lock);
7663
7664         /*
7665          * make sure this block group isn't part of a metadata
7666          * allocation cluster
7667          */
7668         cluster = &root->fs_info->meta_alloc_cluster;
7669         spin_lock(&cluster->refill_lock);
7670         btrfs_return_cluster_to_free_space(block_group, cluster);
7671         spin_unlock(&cluster->refill_lock);
7672
7673         path = btrfs_alloc_path();
7674         BUG_ON(!path);
7675
7676         spin_lock(&root->fs_info->block_group_cache_lock);
7677         rb_erase(&block_group->cache_node,
7678                  &root->fs_info->block_group_cache_tree);
7679         spin_unlock(&root->fs_info->block_group_cache_lock);
7680
7681         down_write(&block_group->space_info->groups_sem);
7682         /*
7683          * we must use list_del_init so people can check to see if they
7684          * are still on the list after taking the semaphore
7685          */
7686         list_del_init(&block_group->list);
7687         up_write(&block_group->space_info->groups_sem);
7688
7689         if (block_group->cached == BTRFS_CACHE_STARTED)
7690                 wait_block_group_cache_done(block_group);
7691
7692         btrfs_remove_free_space_cache(block_group);
7693
7694         spin_lock(&block_group->space_info->lock);
7695         block_group->space_info->total_bytes -= block_group->key.offset;
7696         block_group->space_info->bytes_readonly -= block_group->key.offset;
7697         spin_unlock(&block_group->space_info->lock);
7698
7699         btrfs_clear_space_info_full(root->fs_info);
7700
7701         btrfs_put_block_group(block_group);
7702         btrfs_put_block_group(block_group);
7703
7704         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7705         if (ret > 0)
7706                 ret = -EIO;
7707         if (ret < 0)
7708                 goto out;
7709
7710         ret = btrfs_del_item(trans, root, path);
7711 out:
7712         btrfs_free_path(path);
7713         return ret;
7714 }