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
  78 static noinline int
  79 block_group_cache_done(struct btrfs_block_group_cache *cache)
  80 {
  81         smp_mb();
  82         return cache->cached == BTRFS_CACHE_FINISHED;
  83 }
  84
  85 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  86 {
  87         return (cache->flags & bits) == bits;
  88 }
  89
  90 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  91 {
  92         atomic_inc(&cache->count);
  93 }
  94
  95 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  96 {
  97         if (atomic_dec_and_test(&cache->count))
  98                 kfree(cache);
  99 }
 100
 101 /*
 102  * this adds the block group to the fs_info rb tree for the block group
 103  * cache
 104  */
 105 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 106                                 struct btrfs_block_group_cache *block_group)
 107 {
 108         struct rb_node **p;
 109         struct rb_node *parent = NULL;
 110         struct btrfs_block_group_cache *cache;
 111
 112         spin_lock(&info->block_group_cache_lock);
 113         p = &info->block_group_cache_tree.rb_node;
 114
 115         while (*p) {
 116                 parent = *p;
 117                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 118                                  cache_node);
 119                 if (block_group->key.objectid < cache->key.objectid) {
 120                         p = &(*p)->rb_left;
 121                 } else if (block_group->key.objectid > cache->key.objectid) {
 122                         p = &(*p)->rb_right;
 123                 } else {
 124                         spin_unlock(&info->block_group_cache_lock);
 125                         return -EEXIST;
 126                 }
 127         }
 128
 129         rb_link_node(&block_group->cache_node, parent, p);
 130         rb_insert_color(&block_group->cache_node,
 131                         &info->block_group_cache_tree);
 132         spin_unlock(&info->block_group_cache_lock);
 133
 134         return 0;
 135 }
 136
 137 /*
 138  * This will return the block group at or after bytenr if contains is 0, else
 139  * it will return the block group that contains the bytenr
 140  */
 141 static struct btrfs_block_group_cache *
 142 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 143                               int contains)
 144 {
 145         struct btrfs_block_group_cache *cache, *ret = NULL;
 146         struct rb_node *n;
 147         u64 end, start;
 148
 149         spin_lock(&info->block_group_cache_lock);
 150         n = info->block_group_cache_tree.rb_node;
 151
 152         while (n) {
 153                 cache = rb_entry(n, struct btrfs_block_group_cache,
 154                                  cache_node);
 155                 end = cache->key.objectid + cache->key.offset - 1;
 156                 start = cache->key.objectid;
 157
 158                 if (bytenr < start) {
 159                         if (!contains && (!ret || start < ret->key.objectid))
 160                                 ret = cache;
 161                         n = n->rb_left;
 162                 } else if (bytenr > start) {
 163                         if (contains && bytenr <= end) {
 164                                 ret = cache;
 165                                 break;
 166                         }
 167                         n = n->rb_right;
 168                 } else {
 169                         ret = cache;
 170                         break;
 171                 }
 172         }
 173         if (ret)
 174                 btrfs_get_block_group(ret);
 175         spin_unlock(&info->block_group_cache_lock);
 176
 177         return ret;
 178 }
 179
 180 static int add_excluded_extent(struct btrfs_root *root,
 181                                u64 start, u64 num_bytes)
 182 {
 183         u64 end = start + num_bytes - 1;
 184         set_extent_bits(&root->fs_info->freed_extents[0],
 185                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 186         set_extent_bits(&root->fs_info->freed_extents[1],
 187                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 188         return 0;
 189 }
 190
 191 static void free_excluded_extents(struct btrfs_root *root,
 192                                   struct btrfs_block_group_cache *cache)
 193 {
 194         u64 start, end;
 195
 196         start = cache->key.objectid;
 197         end = start + cache->key.offset - 1;
 198
 199         clear_extent_bits(&root->fs_info->freed_extents[0],
 200                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 201         clear_extent_bits(&root->fs_info->freed_extents[1],
 202                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 203 }
 204
 205 static int exclude_super_stripes(struct btrfs_root *root,
 206                                  struct btrfs_block_group_cache *cache)
 207 {
 208         u64 bytenr;
 209         u64 *logical;
 210         int stripe_len;
 211         int i, nr, ret;
 212
 213         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 214                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 215                 cache->bytes_super += stripe_len;
 216                 ret = add_excluded_extent(root, cache->key.objectid,
 217                                           stripe_len);
 218                 BUG_ON(ret);
 219         }
 220
 221         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 222                 bytenr = btrfs_sb_offset(i);
 223                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 224                                        cache->key.objectid, bytenr,
 225                                        0, &logical, &nr, &stripe_len);
 226                 BUG_ON(ret);
 227
 228                 while (nr--) {
 229                         cache->bytes_super += stripe_len;
 230                         ret = add_excluded_extent(root, logical[nr],
 231                                                   stripe_len);
 232                         BUG_ON(ret);
 233                 }
 234
 235                 kfree(logical);
 236         }
 237         return 0;
 238 }
 239
 240 static struct btrfs_caching_control *
 241 get_caching_control(struct btrfs_block_group_cache *cache)
 242 {
 243         struct btrfs_caching_control *ctl;
 244
 245         spin_lock(&cache->lock);
 246         if (cache->cached != BTRFS_CACHE_STARTED) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         exclude_super_stripes(extent_root, block_group);
 324         spin_lock(&block_group->space_info->lock);
 325         block_group->space_info->bytes_super += block_group->bytes_super;
 326         spin_unlock(&block_group->space_info->lock);
 327
 328         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 329
 330         /*
 331          * We don't want to deadlock with somebody trying to allocate a new
 332          * extent for the extent root while also trying to search the extent
 333          * root to add free space.  So we skip locking and search the commit
 334          * root, since its read-only
 335          */
 336         path->skip_locking = 1;
 337         path->search_commit_root = 1;
 338         path->reada = 2;
 339
 340         key.objectid = last;
 341         key.offset = 0;
 342         key.type = BTRFS_EXTENT_ITEM_KEY;
 343 again:
 344         mutex_lock(&caching_ctl->mutex);
 345         /* need to make sure the commit_root doesn't disappear */
 346         down_read(&fs_info->extent_commit_sem);
 347
 348         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 349         if (ret < 0)
 350                 goto err;
 351
 352         leaf = path->nodes[0];
 353         nritems = btrfs_header_nritems(leaf);
 354
 355         while (1) {
 356                 smp_mb();
 357                 if (fs_info->closing > 1) {
 358                         last = (u64)-1;
 359                         break;
 360                 }
 361
 362                 if (path->slots[0] < nritems) {
 363                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 364                 } else {
 365                         ret = find_next_key(path, 0, &key);
 366                         if (ret)
 367                                 break;
 368
 369                         caching_ctl->progress = last;
 370                         btrfs_release_path(extent_root, path);
 371                         up_read(&fs_info->extent_commit_sem);
 372                         mutex_unlock(&caching_ctl->mutex);
 373                         if (btrfs_transaction_in_commit(fs_info))
 374                                 schedule_timeout(1);
 375                         else
 376                                 cond_resched();
 377                         goto again;
 378                 }
 379
 380                 if (key.objectid < block_group->key.objectid) {
 381                         path->slots[0]++;
 382                         continue;
 383                 }
 384
 385                 if (key.objectid >= block_group->key.objectid +
 386                     block_group->key.offset)
 387                         break;
 388
 389                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 390                         total_found += add_new_free_space(block_group,
 391                                                           fs_info, last,
 392                                                           key.objectid);
 393                         last = key.objectid + key.offset;
 394
 395                         if (total_found > (1024 * 1024 * 2)) {
 396                                 total_found = 0;
 397                                 wake_up(&caching_ctl->wait);
 398                         }
 399                 }
 400                 path->slots[0]++;
 401         }
 402         ret = 0;
 403
 404         total_found += add_new_free_space(block_group, fs_info, last,
 405                                           block_group->key.objectid +
 406                                           block_group->key.offset);
 407         caching_ctl->progress = (u64)-1;
 408
 409         spin_lock(&block_group->lock);
 410         block_group->caching_ctl = NULL;
 411         block_group->cached = BTRFS_CACHE_FINISHED;
 412         spin_unlock(&block_group->lock);
 413
 414 err:
 415         btrfs_free_path(path);
 416         up_read(&fs_info->extent_commit_sem);
 417
 418         free_excluded_extents(extent_root, block_group);
 419
 420         mutex_unlock(&caching_ctl->mutex);
 421         wake_up(&caching_ctl->wait);
 422
 423         put_caching_control(caching_ctl);
 424         atomic_dec(&block_group->space_info->caching_threads);
 425         btrfs_put_block_group(block_group);
 426
 427         return 0;
 428 }
 429
 430 static int cache_block_group(struct btrfs_block_group_cache *cache)
 431 {
 432         struct btrfs_fs_info *fs_info = cache->fs_info;
 433         struct btrfs_caching_control *caching_ctl;
 434         struct task_struct *tsk;
 435         int ret = 0;
 436
 437         smp_mb();
 438         if (cache->cached != BTRFS_CACHE_NO)
 439                 return 0;
 440
 441         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 442         BUG_ON(!caching_ctl);
 443
 444         INIT_LIST_HEAD(&caching_ctl->list);
 445         mutex_init(&caching_ctl->mutex);
 446         init_waitqueue_head(&caching_ctl->wait);
 447         caching_ctl->block_group = cache;
 448         caching_ctl->progress = cache->key.objectid;
 449         /* one for caching kthread, one for caching block group list */
 450         atomic_set(&caching_ctl->count, 2);
 451
 452         spin_lock(&cache->lock);
 453         if (cache->cached != BTRFS_CACHE_NO) {
 454                 spin_unlock(&cache->lock);
 455                 kfree(caching_ctl);
 456                 return 0;
 457         }
 458         cache->caching_ctl = caching_ctl;
 459         cache->cached = BTRFS_CACHE_STARTED;
 460         spin_unlock(&cache->lock);
 461
 462         down_write(&fs_info->extent_commit_sem);
 463         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 464         up_write(&fs_info->extent_commit_sem);
 465
 466         atomic_inc(&cache->space_info->caching_threads);
 467         btrfs_get_block_group(cache);
 468
 469         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 470                           cache->key.objectid);
 471         if (IS_ERR(tsk)) {
 472                 ret = PTR_ERR(tsk);
 473                 printk(KERN_ERR "error running thread %d\n", ret);
 474                 BUG();
 475         }
 476
 477         return ret;
 478 }
 479
 480 /*
 481  * return the block group that starts at or after bytenr
 482  */
 483 static struct btrfs_block_group_cache *
 484 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 485 {
 486         struct btrfs_block_group_cache *cache;
 487
 488         cache = block_group_cache_tree_search(info, bytenr, 0);
 489
 490         return cache;
 491 }
 492
 493 /*
 494  * return the block group that contains the given bytenr
 495  */
 496 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 497                                                  struct btrfs_fs_info *info,
 498                                                  u64 bytenr)
 499 {
 500         struct btrfs_block_group_cache *cache;
 501
 502         cache = block_group_cache_tree_search(info, bytenr, 1);
 503
 504         return cache;
 505 }
 506
 507 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 508                                                   u64 flags)
 509 {
 510         struct list_head *head = &info->space_info;
 511         struct btrfs_space_info *found;
 512
 513         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 514                  BTRFS_BLOCK_GROUP_METADATA;
 515
 516         rcu_read_lock();
 517         list_for_each_entry_rcu(found, head, list) {
 518                 if (found->flags == flags) {
 519                         rcu_read_unlock();
 520                         return found;
 521                 }
 522         }
 523         rcu_read_unlock();
 524         return NULL;
 525 }
 526
 527 /*
 528  * after adding space to the filesystem, we need to clear the full flags
 529  * on all the space infos.
 530  */
 531 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 532 {
 533         struct list_head *head = &info->space_info;
 534         struct btrfs_space_info *found;
 535
 536         rcu_read_lock();
 537         list_for_each_entry_rcu(found, head, list)
 538                 found->full = 0;
 539         rcu_read_unlock();
 540 }
 541
 542 static u64 div_factor(u64 num, int factor)
 543 {
 544         if (factor == 10)
 545                 return num;
 546         num *= factor;
 547         do_div(num, 10);
 548         return num;
 549 }
 550
 551 u64 btrfs_find_block_group(struct btrfs_root *root,
 552                            u64 search_start, u64 search_hint, int owner)
 553 {
 554         struct btrfs_block_group_cache *cache;
 555         u64 used;
 556         u64 last = max(search_hint, search_start);
 557         u64 group_start = 0;
 558         int full_search = 0;
 559         int factor = 9;
 560         int wrapped = 0;
 561 again:
 562         while (1) {
 563                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 564                 if (!cache)
 565                         break;
 566
 567                 spin_lock(&cache->lock);
 568                 last = cache->key.objectid + cache->key.offset;
 569                 used = btrfs_block_group_used(&cache->item);
 570
 571                 if ((full_search || !cache->ro) &&
 572                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 573                         if (used + cache->pinned + cache->reserved <
 574                             div_factor(cache->key.offset, factor)) {
 575                                 group_start = cache->key.objectid;
 576                                 spin_unlock(&cache->lock);
 577                                 btrfs_put_block_group(cache);
 578                                 goto found;
 579                         }
 580                 }
 581                 spin_unlock(&cache->lock);
 582                 btrfs_put_block_group(cache);
 583                 cond_resched();
 584         }
 585         if (!wrapped) {
 586                 last = search_start;
 587                 wrapped = 1;
 588                 goto again;
 589         }
 590         if (!full_search && factor < 10) {
 591                 last = search_start;
 592                 full_search = 1;
 593                 factor = 10;
 594                 goto again;
 595         }
 596 found:
 597         return group_start;
 598 }
 599
 600 /* simple helper to search for an existing extent at a given offset */
 601 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 602 {
 603         int ret;
 604         struct btrfs_key key;
 605         struct btrfs_path *path;
 606
 607         path = btrfs_alloc_path();
 608         BUG_ON(!path);
 609         key.objectid = start;
 610         key.offset = len;
 611         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 612         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 613                                 0, 0);
 614         btrfs_free_path(path);
 615         return ret;
 616 }
 617
 618 /*
 619  * Back reference rules.  Back refs have three main goals:
 620  *
 621  * 1) differentiate between all holders of references to an extent so that
 622  *    when a reference is dropped we can make sure it was a valid reference
 623  *    before freeing the extent.
 624  *
 625  * 2) Provide enough information to quickly find the holders of an extent
 626  *    if we notice a given block is corrupted or bad.
 627  *
 628  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 629  *    maintenance.  This is actually the same as #2, but with a slightly
 630  *    different use case.
 631  *
 632  * There are two kinds of back refs. The implicit back refs is optimized
 633  * for pointers in non-shared tree blocks. For a given pointer in a block,
 634  * back refs of this kind provide information about the block's owner tree
 635  * and the pointer's key. These information allow us to find the block by
 636  * b-tree searching. The full back refs is for pointers in tree blocks not
 637  * referenced by their owner trees. The location of tree block is recorded
 638  * in the back refs. Actually the full back refs is generic, and can be
 639  * used in all cases the implicit back refs is used. The major shortcoming
 640  * of the full back refs is its overhead. Every time a tree block gets
 641  * COWed, we have to update back refs entry for all pointers in it.
 642  *
 643  * For a newly allocated tree block, we use implicit back refs for
 644  * pointers in it. This means most tree related operations only involve
 645  * implicit back refs. For a tree block created in old transaction, the
 646  * only way to drop a reference to it is COW it. So we can detect the
 647  * event that tree block loses its owner tree's reference and do the
 648  * back refs conversion.
 649  *
 650  * When a tree block is COW'd through a tree, there are four cases:
 651  *
 652  * The reference count of the block is one and the tree is the block's
 653  * owner tree. Nothing to do in this case.
 654  *
 655  * The reference count of the block is one and the tree is not the
 656  * block's owner tree. In this case, full back refs is used for pointers
 657  * in the block. Remove these full back refs, add implicit back refs for
 658  * every pointers in the new block.
 659  *
 660  * The reference count of the block is greater than one and the tree is
 661  * the block's owner tree. In this case, implicit back refs is used for
 662  * pointers in the block. Add full back refs for every pointers in the
 663  * block, increase lower level extents' reference counts. The original
 664  * implicit back refs are entailed to the new block.
 665  *
 666  * The reference count of the block is greater than one and the tree is
 667  * not the block's owner tree. Add implicit back refs for every pointer in
 668  * the new block, increase lower level extents' reference count.
 669  *
 670  * Back Reference Key composing:
 671  *
 672  * The key objectid corresponds to the first byte in the extent,
 673  * The key type is used to differentiate between types of back refs.
 674  * There are different meanings of the key offset for different types
 675  * of back refs.
 676  *
 677  * File extents can be referenced by:
 678  *
 679  * - multiple snapshots, subvolumes, or different generations in one subvol
 680  * - different files inside a single subvolume
 681  * - different offsets inside a file (bookend extents in file.c)
 682  *
 683  * The extent ref structure for the implicit back refs has fields for:
 684  *
 685  * - Objectid of the subvolume root
 686  * - objectid of the file holding the reference
 687  * - original offset in the file
 688  * - how many bookend extents
 689  *
 690  * The key offset for the implicit back refs is hash of the first
 691  * three fields.
 692  *
 693  * The extent ref structure for the full back refs has field for:
 694  *
 695  * - number of pointers in the tree leaf
 696  *
 697  * The key offset for the implicit back refs is the first byte of
 698  * the tree leaf
 699  *
 700  * When a file extent is allocated, The implicit back refs is used.
 701  * the fields are filled in:
 702  *
 703  *     (root_key.objectid, inode objectid, offset in file, 1)
 704  *
 705  * When a file extent is removed file truncation, we find the
 706  * corresponding implicit back refs and check the following fields:
 707  *
 708  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 709  *
 710  * Btree extents can be referenced by:
 711  *
 712  * - Different subvolumes
 713  *
 714  * Both the implicit back refs and the full back refs for tree blocks
 715  * only consist of key. The key offset for the implicit back refs is
 716  * objectid of block's owner tree. The key offset for the full back refs
 717  * is the first byte of parent block.
 718  *
 719  * When implicit back refs is used, information about the lowest key and
 720  * level of the tree block are required. These information are stored in
 721  * tree block info structure.
 722  */
 723
 724 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 725 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 726                                   struct btrfs_root *root,
 727                                   struct btrfs_path *path,
 728                                   u64 owner, u32 extra_size)
 729 {
 730         struct btrfs_extent_item *item;
 731         struct btrfs_extent_item_v0 *ei0;
 732         struct btrfs_extent_ref_v0 *ref0;
 733         struct btrfs_tree_block_info *bi;
 734         struct extent_buffer *leaf;
 735         struct btrfs_key key;
 736         struct btrfs_key found_key;
 737         u32 new_size = sizeof(*item);
 738         u64 refs;
 739         int ret;
 740
 741         leaf = path->nodes[0];
 742         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 743
 744         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 745         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 746                              struct btrfs_extent_item_v0);
 747         refs = btrfs_extent_refs_v0(leaf, ei0);
 748
 749         if (owner == (u64)-1) {
 750                 while (1) {
 751                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 752                                 ret = btrfs_next_leaf(root, path);
 753                                 if (ret < 0)
 754                                         return ret;
 755                                 BUG_ON(ret > 0);
 756                                 leaf = path->nodes[0];
 757                         }
 758                         btrfs_item_key_to_cpu(leaf, &found_key,
 759                                               path->slots[0]);
 760                         BUG_ON(key.objectid != found_key.objectid);
 761                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 762                                 path->slots[0]++;
 763                                 continue;
 764                         }
 765                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 766                                               struct btrfs_extent_ref_v0);
 767                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 768                         break;
 769                 }
 770         }
 771         btrfs_release_path(root, path);
 772
 773         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 774                 new_size += sizeof(*bi);
 775
 776         new_size -= sizeof(*ei0);
 777         ret = btrfs_search_slot(trans, root, &key, path,
 778                                 new_size + extra_size, 1);
 779         if (ret < 0)
 780                 return ret;
 781         BUG_ON(ret);
 782
 783         ret = btrfs_extend_item(trans, root, path, new_size);
 784         BUG_ON(ret);
 785
 786         leaf = path->nodes[0];
 787         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 788         btrfs_set_extent_refs(leaf, item, refs);
 789         /* FIXME: get real generation */
 790         btrfs_set_extent_generation(leaf, item, 0);
 791         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 792                 btrfs_set_extent_flags(leaf, item,
 793                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 794                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 795                 bi = (struct btrfs_tree_block_info *)(item + 1);
 796                 /* FIXME: get first key of the block */
 797                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 798                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 799         } else {
 800                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 801         }
 802         btrfs_mark_buffer_dirty(leaf);
 803         return 0;
 804 }
 805 #endif
 806
 807 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 808 {
 809         u32 high_crc = ~(u32)0;
 810         u32 low_crc = ~(u32)0;
 811         __le64 lenum;
 812
 813         lenum = cpu_to_le64(root_objectid);
 814         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 815         lenum = cpu_to_le64(owner);
 816         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 817         lenum = cpu_to_le64(offset);
 818         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 819
 820         return ((u64)high_crc << 31) ^ (u64)low_crc;
 821 }
 822
 823 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 824                                      struct btrfs_extent_data_ref *ref)
 825 {
 826         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 827                                     btrfs_extent_data_ref_objectid(leaf, ref),
 828                                     btrfs_extent_data_ref_offset(leaf, ref));
 829 }
 830
 831 static int match_extent_data_ref(struct extent_buffer *leaf,
 832                                  struct btrfs_extent_data_ref *ref,
 833                                  u64 root_objectid, u64 owner, u64 offset)
 834 {
 835         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 836             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 837             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 838                 return 0;
 839         return 1;
 840 }
 841
 842 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 843                                            struct btrfs_root *root,
 844                                            struct btrfs_path *path,
 845                                            u64 bytenr, u64 parent,
 846                                            u64 root_objectid,
 847                                            u64 owner, u64 offset)
 848 {
 849         struct btrfs_key key;
 850         struct btrfs_extent_data_ref *ref;
 851         struct extent_buffer *leaf;
 852         u32 nritems;
 853         int ret;
 854         int recow;
 855         int err = -ENOENT;
 856
 857         key.objectid = bytenr;
 858         if (parent) {
 859                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 860                 key.offset = parent;
 861         } else {
 862                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 863                 key.offset = hash_extent_data_ref(root_objectid,
 864                                                   owner, offset);
 865         }
 866 again:
 867         recow = 0;
 868         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 869         if (ret < 0) {
 870                 err = ret;
 871                 goto fail;
 872         }
 873
 874         if (parent) {
 875                 if (!ret)
 876                         return 0;
 877 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 878                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 879                 btrfs_release_path(root, path);
 880                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 881                 if (ret < 0) {
 882                         err = ret;
 883                         goto fail;
 884                 }
 885                 if (!ret)
 886                         return 0;
 887 #endif
 888                 goto fail;
 889         }
 890
 891         leaf = path->nodes[0];
 892         nritems = btrfs_header_nritems(leaf);
 893         while (1) {
 894                 if (path->slots[0] >= nritems) {
 895                         ret = btrfs_next_leaf(root, path);
 896                         if (ret < 0)
 897                                 err = ret;
 898                         if (ret)
 899                                 goto fail;
 900
 901                         leaf = path->nodes[0];
 902                         nritems = btrfs_header_nritems(leaf);
 903                         recow = 1;
 904                 }
 905
 906                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 907                 if (key.objectid != bytenr ||
 908                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 909                         goto fail;
 910
 911                 ref = btrfs_item_ptr(leaf, path->slots[0],
 912                                      struct btrfs_extent_data_ref);
 913
 914                 if (match_extent_data_ref(leaf, ref, root_objectid,
 915                                           owner, offset)) {
 916                         if (recow) {
 917                                 btrfs_release_path(root, path);
 918                                 goto again;
 919                         }
 920                         err = 0;
 921                         break;
 922                 }
 923                 path->slots[0]++;
 924         }
 925 fail:
 926         return err;
 927 }
 928
 929 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 930                                            struct btrfs_root *root,
 931                                            struct btrfs_path *path,
 932                                            u64 bytenr, u64 parent,
 933                                            u64 root_objectid, u64 owner,
 934                                            u64 offset, int refs_to_add)
 935 {
 936         struct btrfs_key key;
 937         struct extent_buffer *leaf;
 938         u32 size;
 939         u32 num_refs;
 940         int ret;
 941
 942         key.objectid = bytenr;
 943         if (parent) {
 944                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 945                 key.offset = parent;
 946                 size = sizeof(struct btrfs_shared_data_ref);
 947         } else {
 948                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 949                 key.offset = hash_extent_data_ref(root_objectid,
 950                                                   owner, offset);
 951                 size = sizeof(struct btrfs_extent_data_ref);
 952         }
 953
 954         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 955         if (ret && ret != -EEXIST)
 956                 goto fail;
 957
 958         leaf = path->nodes[0];
 959         if (parent) {
 960                 struct btrfs_shared_data_ref *ref;
 961                 ref = btrfs_item_ptr(leaf, path->slots[0],
 962                                      struct btrfs_shared_data_ref);
 963                 if (ret == 0) {
 964                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 965                 } else {
 966                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 967                         num_refs += refs_to_add;
 968                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 969                 }
 970         } else {
 971                 struct btrfs_extent_data_ref *ref;
 972                 while (ret == -EEXIST) {
 973                         ref = btrfs_item_ptr(leaf, path->slots[0],
 974                                              struct btrfs_extent_data_ref);
 975                         if (match_extent_data_ref(leaf, ref, root_objectid,
 976                                                   owner, offset))
 977                                 break;
 978                         btrfs_release_path(root, path);
 979                         key.offset++;
 980                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 981                                                       size);
 982                         if (ret && ret != -EEXIST)
 983                                 goto fail;
 984
 985                         leaf = path->nodes[0];
 986                 }
 987                 ref = btrfs_item_ptr(leaf, path->slots[0],
 988                                      struct btrfs_extent_data_ref);
 989                 if (ret == 0) {
 990                         btrfs_set_extent_data_ref_root(leaf, ref,
 991                                                        root_objectid);
 992                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 993                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 994                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 995                 } else {
 996                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 997                         num_refs += refs_to_add;
 998                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 999                 }
1000         }
1001         btrfs_mark_buffer_dirty(leaf);
1002         ret = 0;
1003 fail:
1004         btrfs_release_path(root, path);
1005         return ret;
1006 }
1007
1008 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1009                                            struct btrfs_root *root,
1010                                            struct btrfs_path *path,
1011                                            int refs_to_drop)
1012 {
1013         struct btrfs_key key;
1014         struct btrfs_extent_data_ref *ref1 = NULL;
1015         struct btrfs_shared_data_ref *ref2 = NULL;
1016         struct extent_buffer *leaf;
1017         u32 num_refs = 0;
1018         int ret = 0;
1019
1020         leaf = path->nodes[0];
1021         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1022
1023         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1024                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1025                                       struct btrfs_extent_data_ref);
1026                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1027         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1028                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1029                                       struct btrfs_shared_data_ref);
1030                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1031 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1032         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1033                 struct btrfs_extent_ref_v0 *ref0;
1034                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1035                                       struct btrfs_extent_ref_v0);
1036                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1037 #endif
1038         } else {
1039                 BUG();
1040         }
1041
1042         BUG_ON(num_refs < refs_to_drop);
1043         num_refs -= refs_to_drop;
1044
1045         if (num_refs == 0) {
1046                 ret = btrfs_del_item(trans, root, path);
1047         } else {
1048                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1049                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1050                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1051                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1052 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1053                 else {
1054                         struct btrfs_extent_ref_v0 *ref0;
1055                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1056                                         struct btrfs_extent_ref_v0);
1057                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1058                 }
1059 #endif
1060                 btrfs_mark_buffer_dirty(leaf);
1061         }
1062         return ret;
1063 }
1064
1065 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1066                                           struct btrfs_path *path,
1067                                           struct btrfs_extent_inline_ref *iref)
1068 {
1069         struct btrfs_key key;
1070         struct extent_buffer *leaf;
1071         struct btrfs_extent_data_ref *ref1;
1072         struct btrfs_shared_data_ref *ref2;
1073         u32 num_refs = 0;
1074
1075         leaf = path->nodes[0];
1076         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1077         if (iref) {
1078                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1079                     BTRFS_EXTENT_DATA_REF_KEY) {
1080                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1081                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1082                 } else {
1083                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1084                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1085                 }
1086         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1087                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1088                                       struct btrfs_extent_data_ref);
1089                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1090         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1091                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1092                                       struct btrfs_shared_data_ref);
1093                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1094 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1095         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1096                 struct btrfs_extent_ref_v0 *ref0;
1097                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1098                                       struct btrfs_extent_ref_v0);
1099                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1100 #endif
1101         } else {
1102                 WARN_ON(1);
1103         }
1104         return num_refs;
1105 }
1106
1107 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1108                                           struct btrfs_root *root,
1109                                           struct btrfs_path *path,
1110                                           u64 bytenr, u64 parent,
1111                                           u64 root_objectid)
1112 {
1113         struct btrfs_key key;
1114         int ret;
1115
1116         key.objectid = bytenr;
1117         if (parent) {
1118                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1119                 key.offset = parent;
1120         } else {
1121                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1122                 key.offset = root_objectid;
1123         }
1124
1125         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1126         if (ret > 0)
1127                 ret = -ENOENT;
1128 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1129         if (ret == -ENOENT && parent) {
1130                 btrfs_release_path(root, path);
1131                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1133                 if (ret > 0)
1134                         ret = -ENOENT;
1135         }
1136 #endif
1137         return ret;
1138 }
1139
1140 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1141                                           struct btrfs_root *root,
1142                                           struct btrfs_path *path,
1143                                           u64 bytenr, u64 parent,
1144                                           u64 root_objectid)
1145 {
1146         struct btrfs_key key;
1147         int ret;
1148
1149         key.objectid = bytenr;
1150         if (parent) {
1151                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1152                 key.offset = parent;
1153         } else {
1154                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1155                 key.offset = root_objectid;
1156         }
1157
1158         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1159         btrfs_release_path(root, path);
1160         return ret;
1161 }
1162
1163 static inline int extent_ref_type(u64 parent, u64 owner)
1164 {
1165         int type;
1166         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1167                 if (parent > 0)
1168                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1169                 else
1170                         type = BTRFS_TREE_BLOCK_REF_KEY;
1171         } else {
1172                 if (parent > 0)
1173                         type = BTRFS_SHARED_DATA_REF_KEY;
1174                 else
1175                         type = BTRFS_EXTENT_DATA_REF_KEY;
1176         }
1177         return type;
1178 }
1179
1180 static int find_next_key(struct btrfs_path *path, int level,
1181                          struct btrfs_key *key)
1182
1183 {
1184         for (; level < BTRFS_MAX_LEVEL; level++) {
1185                 if (!path->nodes[level])
1186                         break;
1187                 if (path->slots[level] + 1 >=
1188                     btrfs_header_nritems(path->nodes[level]))
1189                         continue;
1190                 if (level == 0)
1191                         btrfs_item_key_to_cpu(path->nodes[level], key,
1192                                               path->slots[level] + 1);
1193                 else
1194                         btrfs_node_key_to_cpu(path->nodes[level], key,
1195                                               path->slots[level] + 1);
1196                 return 0;
1197         }
1198         return 1;
1199 }
1200
1201 /*
1202  * look for inline back ref. if back ref is found, *ref_ret is set
1203  * to the address of inline back ref, and 0 is returned.
1204  *
1205  * if back ref isn't found, *ref_ret is set to the address where it
1206  * should be inserted, and -ENOENT is returned.
1207  *
1208  * if insert is true and there are too many inline back refs, the path
1209  * points to the extent item, and -EAGAIN is returned.
1210  *
1211  * NOTE: inline back refs are ordered in the same way that back ref
1212  *       items in the tree are ordered.
1213  */
1214 static noinline_for_stack
1215 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1216                                  struct btrfs_root *root,
1217                                  struct btrfs_path *path,
1218                                  struct btrfs_extent_inline_ref **ref_ret,
1219                                  u64 bytenr, u64 num_bytes,
1220                                  u64 parent, u64 root_objectid,
1221                                  u64 owner, u64 offset, int insert)
1222 {
1223         struct btrfs_key key;
1224         struct extent_buffer *leaf;
1225         struct btrfs_extent_item *ei;
1226         struct btrfs_extent_inline_ref *iref;
1227         u64 flags;
1228         u64 item_size;
1229         unsigned long ptr;
1230         unsigned long end;
1231         int extra_size;
1232         int type;
1233         int want;
1234         int ret;
1235         int err = 0;
1236
1237         key.objectid = bytenr;
1238         key.type = BTRFS_EXTENT_ITEM_KEY;
1239         key.offset = num_bytes;
1240
1241         want = extent_ref_type(parent, owner);
1242         if (insert) {
1243                 extra_size = btrfs_extent_inline_ref_size(want);
1244                 path->keep_locks = 1;
1245         } else
1246                 extra_size = -1;
1247         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1248         if (ret < 0) {
1249                 err = ret;
1250                 goto out;
1251         }
1252         BUG_ON(ret);
1253
1254         leaf = path->nodes[0];
1255         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1256 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1257         if (item_size < sizeof(*ei)) {
1258                 if (!insert) {
1259                         err = -ENOENT;
1260                         goto out;
1261                 }
1262                 ret = convert_extent_item_v0(trans, root, path, owner,
1263                                              extra_size);
1264                 if (ret < 0) {
1265                         err = ret;
1266                         goto out;
1267                 }
1268                 leaf = path->nodes[0];
1269                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1270         }
1271 #endif
1272         BUG_ON(item_size < sizeof(*ei));
1273
1274         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1275         flags = btrfs_extent_flags(leaf, ei);
1276
1277         ptr = (unsigned long)(ei + 1);
1278         end = (unsigned long)ei + item_size;
1279
1280         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1281                 ptr += sizeof(struct btrfs_tree_block_info);
1282                 BUG_ON(ptr > end);
1283         } else {
1284                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1285         }
1286
1287         err = -ENOENT;
1288         while (1) {
1289                 if (ptr >= end) {
1290                         WARN_ON(ptr > end);
1291                         break;
1292                 }
1293                 iref = (struct btrfs_extent_inline_ref *)ptr;
1294                 type = btrfs_extent_inline_ref_type(leaf, iref);
1295                 if (want < type)
1296                         break;
1297                 if (want > type) {
1298                         ptr += btrfs_extent_inline_ref_size(type);
1299                         continue;
1300                 }
1301
1302                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1303                         struct btrfs_extent_data_ref *dref;
1304                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1305                         if (match_extent_data_ref(leaf, dref, root_objectid,
1306                                                   owner, offset)) {
1307                                 err = 0;
1308                                 break;
1309                         }
1310                         if (hash_extent_data_ref_item(leaf, dref) <
1311                             hash_extent_data_ref(root_objectid, owner, offset))
1312                                 break;
1313                 } else {
1314                         u64 ref_offset;
1315                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1316                         if (parent > 0) {
1317                                 if (parent == ref_offset) {
1318                                         err = 0;
1319                                         break;
1320                                 }
1321                                 if (ref_offset < parent)
1322                                         break;
1323                         } else {
1324                                 if (root_objectid == ref_offset) {
1325                                         err = 0;
1326                                         break;
1327                                 }
1328                                 if (ref_offset < root_objectid)
1329                                         break;
1330                         }
1331                 }
1332                 ptr += btrfs_extent_inline_ref_size(type);
1333         }
1334         if (err == -ENOENT && insert) {
1335                 if (item_size + extra_size >=
1336                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1337                         err = -EAGAIN;
1338                         goto out;
1339                 }
1340                 /*
1341                  * To add new inline back ref, we have to make sure
1342                  * there is no corresponding back ref item.
1343                  * For simplicity, we just do not add new inline back
1344                  * ref if there is any kind of item for this block
1345                  */
1346                 if (find_next_key(path, 0, &key) == 0 &&
1347                     key.objectid == bytenr &&
1348                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1349                         err = -EAGAIN;
1350                         goto out;
1351                 }
1352         }
1353         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1354 out:
1355         if (insert) {
1356                 path->keep_locks = 0;
1357                 btrfs_unlock_up_safe(path, 1);
1358         }
1359         return err;
1360 }
1361
1362 /*
1363  * helper to add new inline back ref
1364  */
1365 static noinline_for_stack
1366 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1367                                 struct btrfs_root *root,
1368                                 struct btrfs_path *path,
1369                                 struct btrfs_extent_inline_ref *iref,
1370                                 u64 parent, u64 root_objectid,
1371                                 u64 owner, u64 offset, int refs_to_add,
1372                                 struct btrfs_delayed_extent_op *extent_op)
1373 {
1374         struct extent_buffer *leaf;
1375         struct btrfs_extent_item *ei;
1376         unsigned long ptr;
1377         unsigned long end;
1378         unsigned long item_offset;
1379         u64 refs;
1380         int size;
1381         int type;
1382         int ret;
1383
1384         leaf = path->nodes[0];
1385         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1386         item_offset = (unsigned long)iref - (unsigned long)ei;
1387
1388         type = extent_ref_type(parent, owner);
1389         size = btrfs_extent_inline_ref_size(type);
1390
1391         ret = btrfs_extend_item(trans, root, path, size);
1392         BUG_ON(ret);
1393
1394         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1395         refs = btrfs_extent_refs(leaf, ei);
1396         refs += refs_to_add;
1397         btrfs_set_extent_refs(leaf, ei, refs);
1398         if (extent_op)
1399                 __run_delayed_extent_op(extent_op, leaf, ei);
1400
1401         ptr = (unsigned long)ei + item_offset;
1402         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1403         if (ptr < end - size)
1404                 memmove_extent_buffer(leaf, ptr + size, ptr,
1405                                       end - size - ptr);
1406
1407         iref = (struct btrfs_extent_inline_ref *)ptr;
1408         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1409         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1410                 struct btrfs_extent_data_ref *dref;
1411                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1412                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1413                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1414                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1415                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1416         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1417                 struct btrfs_shared_data_ref *sref;
1418                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1419                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1420                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1421         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1422                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1423         } else {
1424                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1425         }
1426         btrfs_mark_buffer_dirty(leaf);
1427         return 0;
1428 }
1429
1430 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1431                                  struct btrfs_root *root,
1432                                  struct btrfs_path *path,
1433                                  struct btrfs_extent_inline_ref **ref_ret,
1434                                  u64 bytenr, u64 num_bytes, u64 parent,
1435                                  u64 root_objectid, u64 owner, u64 offset)
1436 {
1437         int ret;
1438
1439         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1440                                            bytenr, num_bytes, parent,
1441                                            root_objectid, owner, offset, 0);
1442         if (ret != -ENOENT)
1443                 return ret;
1444
1445         btrfs_release_path(root, path);
1446         *ref_ret = NULL;
1447
1448         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1449                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1450                                             root_objectid);
1451         } else {
1452                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1453                                              root_objectid, owner, offset);
1454         }
1455         return ret;
1456 }
1457
1458 /*
1459  * helper to update/remove inline back ref
1460  */
1461 static noinline_for_stack
1462 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1463                                  struct btrfs_root *root,
1464                                  struct btrfs_path *path,
1465                                  struct btrfs_extent_inline_ref *iref,
1466                                  int refs_to_mod,
1467                                  struct btrfs_delayed_extent_op *extent_op)
1468 {
1469         struct extent_buffer *leaf;
1470         struct btrfs_extent_item *ei;
1471         struct btrfs_extent_data_ref *dref = NULL;
1472         struct btrfs_shared_data_ref *sref = NULL;
1473         unsigned long ptr;
1474         unsigned long end;
1475         u32 item_size;
1476         int size;
1477         int type;
1478         int ret;
1479         u64 refs;
1480
1481         leaf = path->nodes[0];
1482         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1483         refs = btrfs_extent_refs(leaf, ei);
1484         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1485         refs += refs_to_mod;
1486         btrfs_set_extent_refs(leaf, ei, refs);
1487         if (extent_op)
1488                 __run_delayed_extent_op(extent_op, leaf, ei);
1489
1490         type = btrfs_extent_inline_ref_type(leaf, iref);
1491
1492         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1493                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1494                 refs = btrfs_extent_data_ref_count(leaf, dref);
1495         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1496                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1497                 refs = btrfs_shared_data_ref_count(leaf, sref);
1498         } else {
1499                 refs = 1;
1500                 BUG_ON(refs_to_mod != -1);
1501         }
1502
1503         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1504         refs += refs_to_mod;
1505
1506         if (refs > 0) {
1507                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1508                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1509                 else
1510                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1511         } else {
1512                 size =  btrfs_extent_inline_ref_size(type);
1513                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1514                 ptr = (unsigned long)iref;
1515                 end = (unsigned long)ei + item_size;
1516                 if (ptr + size < end)
1517                         memmove_extent_buffer(leaf, ptr, ptr + size,
1518                                               end - ptr - size);
1519                 item_size -= size;
1520                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1521                 BUG_ON(ret);
1522         }
1523         btrfs_mark_buffer_dirty(leaf);
1524         return 0;
1525 }
1526
1527 static noinline_for_stack
1528 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1529                                  struct btrfs_root *root,
1530                                  struct btrfs_path *path,
1531                                  u64 bytenr, u64 num_bytes, u64 parent,
1532                                  u64 root_objectid, u64 owner,
1533                                  u64 offset, int refs_to_add,
1534                                  struct btrfs_delayed_extent_op *extent_op)
1535 {
1536         struct btrfs_extent_inline_ref *iref;
1537         int ret;
1538
1539         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1540                                            bytenr, num_bytes, parent,
1541                                            root_objectid, owner, offset, 1);
1542         if (ret == 0) {
1543                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1544                 ret = update_inline_extent_backref(trans, root, path, iref,
1545                                                    refs_to_add, extent_op);
1546         } else if (ret == -ENOENT) {
1547                 ret = setup_inline_extent_backref(trans, root, path, iref,
1548                                                   parent, root_objectid,
1549                                                   owner, offset, refs_to_add,
1550                                                   extent_op);
1551         }
1552         return ret;
1553 }
1554
1555 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1556                                  struct btrfs_root *root,
1557                                  struct btrfs_path *path,
1558                                  u64 bytenr, u64 parent, u64 root_objectid,
1559                                  u64 owner, u64 offset, int refs_to_add)
1560 {
1561         int ret;
1562         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1563                 BUG_ON(refs_to_add != 1);
1564                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1565                                             parent, root_objectid);
1566         } else {
1567                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1568                                              parent, root_objectid,
1569                                              owner, offset, refs_to_add);
1570         }
1571         return ret;
1572 }
1573
1574 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1575                                  struct btrfs_root *root,
1576                                  struct btrfs_path *path,
1577                                  struct btrfs_extent_inline_ref *iref,
1578                                  int refs_to_drop, int is_data)
1579 {
1580         int ret;
1581
1582         BUG_ON(!is_data && refs_to_drop != 1);
1583         if (iref) {
1584                 ret = update_inline_extent_backref(trans, root, path, iref,
1585                                                    -refs_to_drop, NULL);
1586         } else if (is_data) {
1587                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1588         } else {
1589                 ret = btrfs_del_item(trans, root, path);
1590         }
1591         return ret;
1592 }
1593
1594 static void btrfs_issue_discard(struct block_device *bdev,
1595                                 u64 start, u64 len)
1596 {
1597         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1598                              DISCARD_FL_BARRIER);
1599 }
1600
1601 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1602                                 u64 num_bytes)
1603 {
1604         int ret;
1605         u64 map_length = num_bytes;
1606         struct btrfs_multi_bio *multi = NULL;
1607
1608         if (!btrfs_test_opt(root, DISCARD))
1609                 return 0;
1610
1611         /* Tell the block device(s) that the sectors can be discarded */
1612         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1613                               bytenr, &map_length, &multi, 0);
1614         if (!ret) {
1615                 struct btrfs_bio_stripe *stripe = multi->stripes;
1616                 int i;
1617
1618                 if (map_length > num_bytes)
1619                         map_length = num_bytes;
1620
1621                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1622                         btrfs_issue_discard(stripe->dev->bdev,
1623                                             stripe->physical,
1624                                             map_length);
1625                 }
1626                 kfree(multi);
1627         }
1628
1629         return ret;
1630 }
1631
1632 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1633                          struct btrfs_root *root,
1634                          u64 bytenr, u64 num_bytes, u64 parent,
1635                          u64 root_objectid, u64 owner, u64 offset)
1636 {
1637         int ret;
1638         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1639                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1640
1641         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1642                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1643                                         parent, root_objectid, (int)owner,
1644                                         BTRFS_ADD_DELAYED_REF, NULL);
1645         } else {
1646                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1647                                         parent, root_objectid, owner, offset,
1648                                         BTRFS_ADD_DELAYED_REF, NULL);
1649         }
1650         return ret;
1651 }
1652
1653 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1654                                   struct btrfs_root *root,
1655                                   u64 bytenr, u64 num_bytes,
1656                                   u64 parent, u64 root_objectid,
1657                                   u64 owner, u64 offset, int refs_to_add,
1658                                   struct btrfs_delayed_extent_op *extent_op)
1659 {
1660         struct btrfs_path *path;
1661         struct extent_buffer *leaf;
1662         struct btrfs_extent_item *item;
1663         u64 refs;
1664         int ret;
1665         int err = 0;
1666
1667         path = btrfs_alloc_path();
1668         if (!path)
1669                 return -ENOMEM;
1670
1671         path->reada = 1;
1672         path->leave_spinning = 1;
1673         /* this will setup the path even if it fails to insert the back ref */
1674         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1675                                            path, bytenr, num_bytes, parent,
1676                                            root_objectid, owner, offset,
1677                                            refs_to_add, extent_op);
1678         if (ret == 0)
1679                 goto out;
1680
1681         if (ret != -EAGAIN) {
1682                 err = ret;
1683                 goto out;
1684         }
1685
1686         leaf = path->nodes[0];
1687         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688         refs = btrfs_extent_refs(leaf, item);
1689         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1690         if (extent_op)
1691                 __run_delayed_extent_op(extent_op, leaf, item);
1692
1693         btrfs_mark_buffer_dirty(leaf);
1694         btrfs_release_path(root->fs_info->extent_root, path);
1695
1696         path->reada = 1;
1697         path->leave_spinning = 1;
1698
1699         /* now insert the actual backref */
1700         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1701                                     path, bytenr, parent, root_objectid,
1702                                     owner, offset, refs_to_add);
1703         BUG_ON(ret);
1704 out:
1705         btrfs_free_path(path);
1706         return err;
1707 }
1708
1709 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1710                                 struct btrfs_root *root,
1711                                 struct btrfs_delayed_ref_node *node,
1712                                 struct btrfs_delayed_extent_op *extent_op,
1713                                 int insert_reserved)
1714 {
1715         int ret = 0;
1716         struct btrfs_delayed_data_ref *ref;
1717         struct btrfs_key ins;
1718         u64 parent = 0;
1719         u64 ref_root = 0;
1720         u64 flags = 0;
1721
1722         ins.objectid = node->bytenr;
1723         ins.offset = node->num_bytes;
1724         ins.type = BTRFS_EXTENT_ITEM_KEY;
1725
1726         ref = btrfs_delayed_node_to_data_ref(node);
1727         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1728                 parent = ref->parent;
1729         else
1730                 ref_root = ref->root;
1731
1732         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1733                 if (extent_op) {
1734                         BUG_ON(extent_op->update_key);
1735                         flags |= extent_op->flags_to_set;
1736                 }
1737                 ret = alloc_reserved_file_extent(trans, root,
1738                                                  parent, ref_root, flags,
1739                                                  ref->objectid, ref->offset,
1740                                                  &ins, node->ref_mod);
1741         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1742                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1743                                              node->num_bytes, parent,
1744                                              ref_root, ref->objectid,
1745                                              ref->offset, node->ref_mod,
1746                                              extent_op);
1747         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1748                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1749                                           node->num_bytes, parent,
1750                                           ref_root, ref->objectid,
1751                                           ref->offset, node->ref_mod,
1752                                           extent_op);
1753         } else {
1754                 BUG();
1755         }
1756         return ret;
1757 }
1758
1759 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1760                                     struct extent_buffer *leaf,
1761                                     struct btrfs_extent_item *ei)
1762 {
1763         u64 flags = btrfs_extent_flags(leaf, ei);
1764         if (extent_op->update_flags) {
1765                 flags |= extent_op->flags_to_set;
1766                 btrfs_set_extent_flags(leaf, ei, flags);
1767         }
1768
1769         if (extent_op->update_key) {
1770                 struct btrfs_tree_block_info *bi;
1771                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1772                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1773                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1774         }
1775 }
1776
1777 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1778                                  struct btrfs_root *root,
1779                                  struct btrfs_delayed_ref_node *node,
1780                                  struct btrfs_delayed_extent_op *extent_op)
1781 {
1782         struct btrfs_key key;
1783         struct btrfs_path *path;
1784         struct btrfs_extent_item *ei;
1785         struct extent_buffer *leaf;
1786         u32 item_size;
1787         int ret;
1788         int err = 0;
1789
1790         path = btrfs_alloc_path();
1791         if (!path)
1792                 return -ENOMEM;
1793
1794         key.objectid = node->bytenr;
1795         key.type = BTRFS_EXTENT_ITEM_KEY;
1796         key.offset = node->num_bytes;
1797
1798         path->reada = 1;
1799         path->leave_spinning = 1;
1800         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1801                                 path, 0, 1);
1802         if (ret < 0) {
1803                 err = ret;
1804                 goto out;
1805         }
1806         if (ret > 0) {
1807                 err = -EIO;
1808                 goto out;
1809         }
1810
1811         leaf = path->nodes[0];
1812         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1813 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1814         if (item_size < sizeof(*ei)) {
1815                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1816                                              path, (u64)-1, 0);
1817                 if (ret < 0) {
1818                         err = ret;
1819                         goto out;
1820                 }
1821                 leaf = path->nodes[0];
1822                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1823         }
1824 #endif
1825         BUG_ON(item_size < sizeof(*ei));
1826         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1827         __run_delayed_extent_op(extent_op, leaf, ei);
1828
1829         btrfs_mark_buffer_dirty(leaf);
1830 out:
1831         btrfs_free_path(path);
1832         return err;
1833 }
1834
1835 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1836                                 struct btrfs_root *root,
1837                                 struct btrfs_delayed_ref_node *node,
1838                                 struct btrfs_delayed_extent_op *extent_op,
1839                                 int insert_reserved)
1840 {
1841         int ret = 0;
1842         struct btrfs_delayed_tree_ref *ref;
1843         struct btrfs_key ins;
1844         u64 parent = 0;
1845         u64 ref_root = 0;
1846
1847         ins.objectid = node->bytenr;
1848         ins.offset = node->num_bytes;
1849         ins.type = BTRFS_EXTENT_ITEM_KEY;
1850
1851         ref = btrfs_delayed_node_to_tree_ref(node);
1852         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1853                 parent = ref->parent;
1854         else
1855                 ref_root = ref->root;
1856
1857         BUG_ON(node->ref_mod != 1);
1858         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1859                 BUG_ON(!extent_op || !extent_op->update_flags ||
1860                        !extent_op->update_key);
1861                 ret = alloc_reserved_tree_block(trans, root,
1862                                                 parent, ref_root,
1863                                                 extent_op->flags_to_set,
1864                                                 &extent_op->key,
1865                                                 ref->level, &ins);
1866         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1867                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1868                                              node->num_bytes, parent, ref_root,
1869                                              ref->level, 0, 1, extent_op);
1870         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1871                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1872                                           node->num_bytes, parent, ref_root,
1873                                           ref->level, 0, 1, extent_op);
1874         } else {
1875                 BUG();
1876         }
1877         return ret;
1878 }
1879
1880
1881 /* helper function to actually process a single delayed ref entry */
1882 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1883                                struct btrfs_root *root,
1884                                struct btrfs_delayed_ref_node *node,
1885                                struct btrfs_delayed_extent_op *extent_op,
1886                                int insert_reserved)
1887 {
1888         int ret;
1889         if (btrfs_delayed_ref_is_head(node)) {
1890                 struct btrfs_delayed_ref_head *head;
1891                 /*
1892                  * we've hit the end of the chain and we were supposed
1893                  * to insert this extent into the tree.  But, it got
1894                  * deleted before we ever needed to insert it, so all
1895                  * we have to do is clean up the accounting
1896                  */
1897                 BUG_ON(extent_op);
1898                 head = btrfs_delayed_node_to_head(node);
1899                 if (insert_reserved) {
1900                         int mark_free = 0;
1901                         struct extent_buffer *must_clean = NULL;
1902
1903                         ret = pin_down_bytes(trans, root, NULL,
1904                                              node->bytenr, node->num_bytes,
1905                                              head->is_data, 1, &must_clean);
1906                         if (ret > 0)
1907                                 mark_free = 1;
1908
1909                         if (must_clean) {
1910                                 clean_tree_block(NULL, root, must_clean);
1911                                 btrfs_tree_unlock(must_clean);
1912                                 free_extent_buffer(must_clean);
1913                         }
1914                         if (head->is_data) {
1915                                 ret = btrfs_del_csums(trans, root,
1916                                                       node->bytenr,
1917                                                       node->num_bytes);
1918                                 BUG_ON(ret);
1919                         }
1920                         if (mark_free) {
1921                                 ret = btrfs_free_reserved_extent(root,
1922                                                         node->bytenr,
1923                                                         node->num_bytes);
1924                                 BUG_ON(ret);
1925                         }
1926                 }
1927                 mutex_unlock(&head->mutex);
1928                 return 0;
1929         }
1930
1931         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1932             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1933                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1934                                            insert_reserved);
1935         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1936                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1937                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1938                                            insert_reserved);
1939         else
1940                 BUG();
1941         return ret;
1942 }
1943
1944 static noinline struct btrfs_delayed_ref_node *
1945 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1946 {
1947         struct rb_node *node;
1948         struct btrfs_delayed_ref_node *ref;
1949         int action = BTRFS_ADD_DELAYED_REF;
1950 again:
1951         /*
1952          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1953          * this prevents ref count from going down to zero when
1954          * there still are pending delayed ref.
1955          */
1956         node = rb_prev(&head->node.rb_node);
1957         while (1) {
1958                 if (!node)
1959                         break;
1960                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1961                                 rb_node);
1962                 if (ref->bytenr != head->node.bytenr)
1963                         break;
1964                 if (ref->action == action)
1965                         return ref;
1966                 node = rb_prev(node);
1967         }
1968         if (action == BTRFS_ADD_DELAYED_REF) {
1969                 action = BTRFS_DROP_DELAYED_REF;
1970                 goto again;
1971         }
1972         return NULL;
1973 }
1974
1975 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1976                                        struct btrfs_root *root,
1977                                        struct list_head *cluster)
1978 {
1979         struct btrfs_delayed_ref_root *delayed_refs;
1980         struct btrfs_delayed_ref_node *ref;
1981         struct btrfs_delayed_ref_head *locked_ref = NULL;
1982         struct btrfs_delayed_extent_op *extent_op;
1983         int ret;
1984         int count = 0;
1985         int must_insert_reserved = 0;
1986
1987         delayed_refs = &trans->transaction->delayed_refs;
1988         while (1) {
1989                 if (!locked_ref) {
1990                         /* pick a new head ref from the cluster list */
1991                         if (list_empty(cluster))
1992                                 break;
1993
1994                         locked_ref = list_entry(cluster->next,
1995                                      struct btrfs_delayed_ref_head, cluster);
1996
1997                         /* grab the lock that says we are going to process
1998                          * all the refs for this head */
1999                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2000
2001                         /*
2002                          * we may have dropped the spin lock to get the head
2003                          * mutex lock, and that might have given someone else
2004                          * time to free the head.  If that's true, it has been
2005                          * removed from our list and we can move on.
2006                          */
2007                         if (ret == -EAGAIN) {
2008                                 locked_ref = NULL;
2009                                 count++;
2010                                 continue;
2011                         }
2012                 }
2013
2014                 /*
2015                  * record the must insert reserved flag before we
2016                  * drop the spin lock.
2017                  */
2018                 must_insert_reserved = locked_ref->must_insert_reserved;
2019                 locked_ref->must_insert_reserved = 0;
2020
2021                 extent_op = locked_ref->extent_op;
2022                 locked_ref->extent_op = NULL;
2023
2024                 /*
2025                  * locked_ref is the head node, so we have to go one
2026                  * node back for any delayed ref updates
2027                  */
2028                 ref = select_delayed_ref(locked_ref);
2029                 if (!ref) {
2030                         /* All delayed refs have been processed, Go ahead
2031                          * and send the head node to run_one_delayed_ref,
2032                          * so that any accounting fixes can happen
2033                          */
2034                         ref = &locked_ref->node;
2035
2036                         if (extent_op && must_insert_reserved) {
2037                                 kfree(extent_op);
2038                                 extent_op = NULL;
2039                         }
2040
2041                         if (extent_op) {
2042                                 spin_unlock(&delayed_refs->lock);
2043
2044                                 ret = run_delayed_extent_op(trans, root,
2045                                                             ref, extent_op);
2046                                 BUG_ON(ret);
2047                                 kfree(extent_op);
2048
2049                                 cond_resched();
2050                                 spin_lock(&delayed_refs->lock);
2051                                 continue;
2052                         }
2053
2054                         list_del_init(&locked_ref->cluster);
2055                         locked_ref = NULL;
2056                 }
2057
2058                 ref->in_tree = 0;
2059                 rb_erase(&ref->rb_node, &delayed_refs->root);
2060                 delayed_refs->num_entries--;
2061
2062                 spin_unlock(&delayed_refs->lock);
2063
2064                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2065                                           must_insert_reserved);
2066                 BUG_ON(ret);
2067
2068                 btrfs_put_delayed_ref(ref);
2069                 kfree(extent_op);
2070                 count++;
2071
2072                 cond_resched();
2073                 spin_lock(&delayed_refs->lock);
2074         }
2075         return count;
2076 }
2077
2078 /*
2079  * this starts processing the delayed reference count updates and
2080  * extent insertions we have queued up so far.  count can be
2081  * 0, which means to process everything in the tree at the start
2082  * of the run (but not newly added entries), or it can be some target
2083  * number you'd like to process.
2084  */
2085 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2086                            struct btrfs_root *root, unsigned long count)
2087 {
2088         struct rb_node *node;
2089         struct btrfs_delayed_ref_root *delayed_refs;
2090         struct btrfs_delayed_ref_node *ref;
2091         struct list_head cluster;
2092         int ret;
2093         int run_all = count == (unsigned long)-1;
2094         int run_most = 0;
2095
2096         if (root == root->fs_info->extent_root)
2097                 root = root->fs_info->tree_root;
2098
2099         delayed_refs = &trans->transaction->delayed_refs;
2100         INIT_LIST_HEAD(&cluster);
2101 again:
2102         spin_lock(&delayed_refs->lock);
2103         if (count == 0) {
2104                 count = delayed_refs->num_entries * 2;
2105                 run_most = 1;
2106         }
2107         while (1) {
2108                 if (!(run_all || run_most) &&
2109                     delayed_refs->num_heads_ready < 64)
2110                         break;
2111
2112                 /*
2113                  * go find something we can process in the rbtree.  We start at
2114                  * the beginning of the tree, and then build a cluster
2115                  * of refs to process starting at the first one we are able to
2116                  * lock
2117                  */
2118                 ret = btrfs_find_ref_cluster(trans, &cluster,
2119                                              delayed_refs->run_delayed_start);
2120                 if (ret)
2121                         break;
2122
2123                 ret = run_clustered_refs(trans, root, &cluster);
2124                 BUG_ON(ret < 0);
2125
2126                 count -= min_t(unsigned long, ret, count);
2127
2128                 if (count == 0)
2129                         break;
2130         }
2131
2132         if (run_all) {
2133                 node = rb_first(&delayed_refs->root);
2134                 if (!node)
2135                         goto out;
2136                 count = (unsigned long)-1;
2137
2138                 while (node) {
2139                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2140                                        rb_node);
2141                         if (btrfs_delayed_ref_is_head(ref)) {
2142                                 struct btrfs_delayed_ref_head *head;
2143
2144                                 head = btrfs_delayed_node_to_head(ref);
2145                                 atomic_inc(&ref->refs);
2146
2147                                 spin_unlock(&delayed_refs->lock);
2148                                 mutex_lock(&head->mutex);
2149                                 mutex_unlock(&head->mutex);
2150
2151                                 btrfs_put_delayed_ref(ref);
2152                                 cond_resched();
2153                                 goto again;
2154                         }
2155                         node = rb_next(node);
2156                 }
2157                 spin_unlock(&delayed_refs->lock);
2158                 schedule_timeout(1);
2159                 goto again;
2160         }
2161 out:
2162         spin_unlock(&delayed_refs->lock);
2163         return 0;
2164 }
2165
2166 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2167                                 struct btrfs_root *root,
2168                                 u64 bytenr, u64 num_bytes, u64 flags,
2169                                 int is_data)
2170 {
2171         struct btrfs_delayed_extent_op *extent_op;
2172         int ret;
2173
2174         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2175         if (!extent_op)
2176                 return -ENOMEM;
2177
2178         extent_op->flags_to_set = flags;
2179         extent_op->update_flags = 1;
2180         extent_op->update_key = 0;
2181         extent_op->is_data = is_data ? 1 : 0;
2182
2183         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2184         if (ret)
2185                 kfree(extent_op);
2186         return ret;
2187 }
2188
2189 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2190                                       struct btrfs_root *root,
2191                                       struct btrfs_path *path,
2192                                       u64 objectid, u64 offset, u64 bytenr)
2193 {
2194         struct btrfs_delayed_ref_head *head;
2195         struct btrfs_delayed_ref_node *ref;
2196         struct btrfs_delayed_data_ref *data_ref;
2197         struct btrfs_delayed_ref_root *delayed_refs;
2198         struct rb_node *node;
2199         int ret = 0;
2200
2201         ret = -ENOENT;
2202         delayed_refs = &trans->transaction->delayed_refs;
2203         spin_lock(&delayed_refs->lock);
2204         head = btrfs_find_delayed_ref_head(trans, bytenr);
2205         if (!head)
2206                 goto out;
2207
2208         if (!mutex_trylock(&head->mutex)) {
2209                 atomic_inc(&head->node.refs);
2210                 spin_unlock(&delayed_refs->lock);
2211
2212                 btrfs_release_path(root->fs_info->extent_root, path);
2213
2214                 mutex_lock(&head->mutex);
2215                 mutex_unlock(&head->mutex);
2216                 btrfs_put_delayed_ref(&head->node);
2217                 return -EAGAIN;
2218         }
2219
2220         node = rb_prev(&head->node.rb_node);
2221         if (!node)
2222                 goto out_unlock;
2223
2224         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2225
2226         if (ref->bytenr != bytenr)
2227                 goto out_unlock;
2228
2229         ret = 1;
2230         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2231                 goto out_unlock;
2232
2233         data_ref = btrfs_delayed_node_to_data_ref(ref);
2234
2235         node = rb_prev(node);
2236         if (node) {
2237                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2238                 if (ref->bytenr == bytenr)
2239                         goto out_unlock;
2240         }
2241
2242         if (data_ref->root != root->root_key.objectid ||
2243             data_ref->objectid != objectid || data_ref->offset != offset)
2244                 goto out_unlock;
2245
2246         ret = 0;
2247 out_unlock:
2248         mutex_unlock(&head->mutex);
2249 out:
2250         spin_unlock(&delayed_refs->lock);
2251         return ret;
2252 }
2253
2254 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2255                                         struct btrfs_root *root,
2256                                         struct btrfs_path *path,
2257                                         u64 objectid, u64 offset, u64 bytenr)
2258 {
2259         struct btrfs_root *extent_root = root->fs_info->extent_root;
2260         struct extent_buffer *leaf;
2261         struct btrfs_extent_data_ref *ref;
2262         struct btrfs_extent_inline_ref *iref;
2263         struct btrfs_extent_item *ei;
2264         struct btrfs_key key;
2265         u32 item_size;
2266         int ret;
2267
2268         key.objectid = bytenr;
2269         key.offset = (u64)-1;
2270         key.type = BTRFS_EXTENT_ITEM_KEY;
2271
2272         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2273         if (ret < 0)
2274                 goto out;
2275         BUG_ON(ret == 0);
2276
2277         ret = -ENOENT;
2278         if (path->slots[0] == 0)
2279                 goto out;
2280
2281         path->slots[0]--;
2282         leaf = path->nodes[0];
2283         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2284
2285         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2286                 goto out;
2287
2288         ret = 1;
2289         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2290 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2291         if (item_size < sizeof(*ei)) {
2292                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2293                 goto out;
2294         }
2295 #endif
2296         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2297
2298         if (item_size != sizeof(*ei) +
2299             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2300                 goto out;
2301
2302         if (btrfs_extent_generation(leaf, ei) <=
2303             btrfs_root_last_snapshot(&root->root_item))
2304                 goto out;
2305
2306         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2307         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2308             BTRFS_EXTENT_DATA_REF_KEY)
2309                 goto out;
2310
2311         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2312         if (btrfs_extent_refs(leaf, ei) !=
2313             btrfs_extent_data_ref_count(leaf, ref) ||
2314             btrfs_extent_data_ref_root(leaf, ref) !=
2315             root->root_key.objectid ||
2316             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2317             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2318                 goto out;
2319
2320         ret = 0;
2321 out:
2322         return ret;
2323 }
2324
2325 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2326                           struct btrfs_root *root,
2327                           u64 objectid, u64 offset, u64 bytenr)
2328 {
2329         struct btrfs_path *path;
2330         int ret;
2331         int ret2;
2332
2333         path = btrfs_alloc_path();
2334         if (!path)
2335                 return -ENOENT;
2336
2337         do {
2338                 ret = check_committed_ref(trans, root, path, objectid,
2339                                           offset, bytenr);
2340                 if (ret && ret != -ENOENT)
2341                         goto out;
2342
2343                 ret2 = check_delayed_ref(trans, root, path, objectid,
2344                                          offset, bytenr);
2345         } while (ret2 == -EAGAIN);
2346
2347         if (ret2 && ret2 != -ENOENT) {
2348                 ret = ret2;
2349                 goto out;
2350         }
2351
2352         if (ret != -ENOENT || ret2 != -ENOENT)
2353                 ret = 0;
2354 out:
2355         btrfs_free_path(path);
2356         return ret;
2357 }
2358
2359 #if 0
2360 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2361                     struct extent_buffer *buf, u32 nr_extents)
2362 {
2363         struct btrfs_key key;
2364         struct btrfs_file_extent_item *fi;
2365         u64 root_gen;
2366         u32 nritems;
2367         int i;
2368         int level;
2369         int ret = 0;
2370         int shared = 0;
2371
2372         if (!root->ref_cows)
2373                 return 0;
2374
2375         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2376                 shared = 0;
2377                 root_gen = root->root_key.offset;
2378         } else {
2379                 shared = 1;
2380                 root_gen = trans->transid - 1;
2381         }
2382
2383         level = btrfs_header_level(buf);
2384         nritems = btrfs_header_nritems(buf);
2385
2386         if (level == 0) {
2387                 struct btrfs_leaf_ref *ref;
2388                 struct btrfs_extent_info *info;
2389
2390                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2391                 if (!ref) {
2392                         ret = -ENOMEM;
2393                         goto out;
2394                 }
2395
2396                 ref->root_gen = root_gen;
2397                 ref->bytenr = buf->start;
2398                 ref->owner = btrfs_header_owner(buf);
2399                 ref->generation = btrfs_header_generation(buf);
2400                 ref->nritems = nr_extents;
2401                 info = ref->extents;
2402
2403                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2404                         u64 disk_bytenr;
2405                         btrfs_item_key_to_cpu(buf, &key, i);
2406                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2407                                 continue;
2408                         fi = btrfs_item_ptr(buf, i,
2409                                             struct btrfs_file_extent_item);
2410                         if (btrfs_file_extent_type(buf, fi) ==
2411                             BTRFS_FILE_EXTENT_INLINE)
2412                                 continue;
2413                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2414                         if (disk_bytenr == 0)
2415                                 continue;
2416
2417                         info->bytenr = disk_bytenr;
2418                         info->num_bytes =
2419                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2420                         info->objectid = key.objectid;
2421                         info->offset = key.offset;
2422                         info++;
2423                 }
2424
2425                 ret = btrfs_add_leaf_ref(root, ref, shared);
2426                 if (ret == -EEXIST && shared) {
2427                         struct btrfs_leaf_ref *old;
2428                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2429                         BUG_ON(!old);
2430                         btrfs_remove_leaf_ref(root, old);
2431                         btrfs_free_leaf_ref(root, old);
2432                         ret = btrfs_add_leaf_ref(root, ref, shared);
2433                 }
2434                 WARN_ON(ret);
2435                 btrfs_free_leaf_ref(root, ref);
2436         }
2437 out:
2438         return ret;
2439 }
2440
2441 /* when a block goes through cow, we update the reference counts of
2442  * everything that block points to.  The internal pointers of the block
2443  * can be in just about any order, and it is likely to have clusters of
2444  * things that are close together and clusters of things that are not.
2445  *
2446  * To help reduce the seeks that come with updating all of these reference
2447  * counts, sort them by byte number before actual updates are done.
2448  *
2449  * struct refsort is used to match byte number to slot in the btree block.
2450  * we sort based on the byte number and then use the slot to actually
2451  * find the item.
2452  *
2453  * struct refsort is smaller than strcut btrfs_item and smaller than
2454  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2455  * for a btree block, there's no way for a kmalloc of refsorts for a
2456  * single node to be bigger than a page.
2457  */
2458 struct refsort {
2459         u64 bytenr;
2460         u32 slot;
2461 };
2462
2463 /*
2464  * for passing into sort()
2465  */
2466 static int refsort_cmp(const void *a_void, const void *b_void)
2467 {
2468         const struct refsort *a = a_void;
2469         const struct refsort *b = b_void;
2470
2471         if (a->bytenr < b->bytenr)
2472                 return -1;
2473         if (a->bytenr > b->bytenr)
2474                 return 1;
2475         return 0;
2476 }
2477 #endif
2478
2479 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2480                            struct btrfs_root *root,
2481                            struct extent_buffer *buf,
2482                            int full_backref, int inc)
2483 {
2484         u64 bytenr;
2485         u64 num_bytes;
2486         u64 parent;
2487         u64 ref_root;
2488         u32 nritems;
2489         struct btrfs_key key;
2490         struct btrfs_file_extent_item *fi;
2491         int i;
2492         int level;
2493         int ret = 0;
2494         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2495                             u64, u64, u64, u64, u64, u64);
2496
2497         ref_root = btrfs_header_owner(buf);
2498         nritems = btrfs_header_nritems(buf);
2499         level = btrfs_header_level(buf);
2500
2501         if (!root->ref_cows && level == 0)
2502                 return 0;
2503
2504         if (inc)
2505                 process_func = btrfs_inc_extent_ref;
2506         else
2507                 process_func = btrfs_free_extent;
2508
2509         if (full_backref)
2510                 parent = buf->start;
2511         else
2512                 parent = 0;
2513
2514         for (i = 0; i < nritems; i++) {
2515                 if (level == 0) {
2516                         btrfs_item_key_to_cpu(buf, &key, i);
2517                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2518                                 continue;
2519                         fi = btrfs_item_ptr(buf, i,
2520                                             struct btrfs_file_extent_item);
2521                         if (btrfs_file_extent_type(buf, fi) ==
2522                             BTRFS_FILE_EXTENT_INLINE)
2523                                 continue;
2524                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2525                         if (bytenr == 0)
2526                                 continue;
2527
2528                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2529                         key.offset -= btrfs_file_extent_offset(buf, fi);
2530                         ret = process_func(trans, root, bytenr, num_bytes,
2531                                            parent, ref_root, key.objectid,
2532                                            key.offset);
2533                         if (ret)
2534                                 goto fail;
2535                 } else {
2536                         bytenr = btrfs_node_blockptr(buf, i);
2537                         num_bytes = btrfs_level_size(root, level - 1);
2538                         ret = process_func(trans, root, bytenr, num_bytes,
2539                                            parent, ref_root, level - 1, 0);
2540                         if (ret)
2541                                 goto fail;
2542                 }
2543         }
2544         return 0;
2545 fail:
2546         BUG();
2547         return ret;
2548 }
2549
2550 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2551                   struct extent_buffer *buf, int full_backref)
2552 {
2553         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2554 }
2555
2556 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2557                   struct extent_buffer *buf, int full_backref)
2558 {
2559         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2560 }
2561
2562 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2563                                  struct btrfs_root *root,
2564                                  struct btrfs_path *path,
2565                                  struct btrfs_block_group_cache *cache)
2566 {
2567         int ret;
2568         struct btrfs_root *extent_root = root->fs_info->extent_root;
2569         unsigned long bi;
2570         struct extent_buffer *leaf;
2571
2572         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2573         if (ret < 0)
2574                 goto fail;
2575         BUG_ON(ret);
2576
2577         leaf = path->nodes[0];
2578         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2579         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2580         btrfs_mark_buffer_dirty(leaf);
2581         btrfs_release_path(extent_root, path);
2582 fail:
2583         if (ret)
2584                 return ret;
2585         return 0;
2586
2587 }
2588
2589 static struct btrfs_block_group_cache *
2590 next_block_group(struct btrfs_root *root,
2591                  struct btrfs_block_group_cache *cache)
2592 {
2593         struct rb_node *node;
2594         spin_lock(&root->fs_info->block_group_cache_lock);
2595         node = rb_next(&cache->cache_node);
2596         btrfs_put_block_group(cache);
2597         if (node) {
2598                 cache = rb_entry(node, struct btrfs_block_group_cache,
2599                                  cache_node);
2600                 btrfs_get_block_group(cache);
2601         } else
2602                 cache = NULL;
2603         spin_unlock(&root->fs_info->block_group_cache_lock);
2604         return cache;
2605 }
2606
2607 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2608                                    struct btrfs_root *root)
2609 {
2610         struct btrfs_block_group_cache *cache;
2611         int err = 0;
2612         struct btrfs_path *path;
2613         u64 last = 0;
2614
2615         path = btrfs_alloc_path();
2616         if (!path)
2617                 return -ENOMEM;
2618
2619         while (1) {
2620                 if (last == 0) {
2621                         err = btrfs_run_delayed_refs(trans, root,
2622                                                      (unsigned long)-1);
2623                         BUG_ON(err);
2624                 }
2625
2626                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2627                 while (cache) {
2628                         if (cache->dirty)
2629                                 break;
2630                         cache = next_block_group(root, cache);
2631                 }
2632                 if (!cache) {
2633                         if (last == 0)
2634                                 break;
2635                         last = 0;
2636                         continue;
2637                 }
2638
2639                 cache->dirty = 0;
2640                 last = cache->key.objectid + cache->key.offset;
2641
2642                 err = write_one_cache_group(trans, root, path, cache);
2643                 BUG_ON(err);
2644                 btrfs_put_block_group(cache);
2645         }
2646
2647         btrfs_free_path(path);
2648         return 0;
2649 }
2650
2651 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2652 {
2653         struct btrfs_block_group_cache *block_group;
2654         int readonly = 0;
2655
2656         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2657         if (!block_group || block_group->ro)
2658                 readonly = 1;
2659         if (block_group)
2660                 btrfs_put_block_group(block_group);
2661         return readonly;
2662 }
2663
2664 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2665                              u64 total_bytes, u64 bytes_used,
2666                              struct btrfs_space_info **space_info)
2667 {
2668         struct btrfs_space_info *found;
2669         int i;
2670         int factor;
2671
2672         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2673                      BTRFS_BLOCK_GROUP_RAID10))
2674                 factor = 2;
2675         else
2676                 factor = 1;
2677
2678         found = __find_space_info(info, flags);
2679         if (found) {
2680                 spin_lock(&found->lock);
2681                 found->total_bytes += total_bytes;
2682                 found->bytes_used += bytes_used;
2683                 found->disk_used += bytes_used * factor;
2684                 found->full = 0;
2685                 spin_unlock(&found->lock);
2686                 *space_info = found;
2687                 return 0;
2688         }
2689         found = kzalloc(sizeof(*found), GFP_NOFS);
2690         if (!found)
2691                 return -ENOMEM;
2692
2693         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2694                 INIT_LIST_HEAD(&found->block_groups[i]);
2695         init_rwsem(&found->groups_sem);
2696         init_waitqueue_head(&found->flush_wait);
2697         spin_lock_init(&found->lock);
2698         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2699                                 BTRFS_BLOCK_GROUP_SYSTEM |
2700                                 BTRFS_BLOCK_GROUP_METADATA);
2701         found->total_bytes = total_bytes;
2702         found->bytes_used = bytes_used;
2703         found->disk_used = bytes_used * factor;
2704         found->bytes_pinned = 0;
2705         found->bytes_reserved = 0;
2706         found->bytes_readonly = 0;
2707         found->bytes_delalloc = 0;
2708         found->full = 0;
2709         found->force_alloc = 0;
2710         *space_info = found;
2711         list_add_rcu(&found->list, &info->space_info);
2712         atomic_set(&found->caching_threads, 0);
2713         return 0;
2714 }
2715
2716 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2717 {
2718         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2719                                    BTRFS_BLOCK_GROUP_RAID1 |
2720                                    BTRFS_BLOCK_GROUP_RAID10 |
2721                                    BTRFS_BLOCK_GROUP_DUP);
2722         if (extra_flags) {
2723                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2724                         fs_info->avail_data_alloc_bits |= extra_flags;
2725                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2726                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2727                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2728                         fs_info->avail_system_alloc_bits |= extra_flags;
2729         }
2730 }
2731
2732 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2733 {
2734         spin_lock(&cache->space_info->lock);
2735         spin_lock(&cache->lock);
2736         if (!cache->ro) {
2737                 cache->space_info->bytes_readonly += cache->key.offset -
2738                                         btrfs_block_group_used(&cache->item);
2739                 cache->ro = 1;
2740         }
2741         spin_unlock(&cache->lock);
2742         spin_unlock(&cache->space_info->lock);
2743 }
2744
2745 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2746 {
2747         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2748
2749         if (num_devices == 1)
2750                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2751         if (num_devices < 4)
2752                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2753
2754         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2755             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2756                       BTRFS_BLOCK_GROUP_RAID10))) {
2757                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2758         }
2759
2760         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2761             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2762                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2763         }
2764
2765         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2766             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2767              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2768              (flags & BTRFS_BLOCK_GROUP_DUP)))
2769                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2770         return flags;
2771 }
2772
2773 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2774 {
2775         if (flags & BTRFS_BLOCK_GROUP_DATA)
2776                 flags |= root->fs_info->avail_data_alloc_bits &
2777                          root->fs_info->data_alloc_profile;
2778         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2779                 flags |= root->fs_info->avail_system_alloc_bits &
2780                          root->fs_info->system_alloc_profile;
2781         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2782                 flags |= root->fs_info->avail_metadata_alloc_bits &
2783                          root->fs_info->metadata_alloc_profile;
2784         return btrfs_reduce_alloc_profile(root, flags);
2785 }
2786
2787 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2788 {
2789         u64 flags;
2790
2791         if (data)
2792                 flags = BTRFS_BLOCK_GROUP_DATA;
2793         else if (root == root->fs_info->chunk_root)
2794                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2795         else
2796                 flags = BTRFS_BLOCK_GROUP_METADATA;
2797
2798         return get_alloc_profile(root, flags);
2799 }
2800
2801 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2802 {
2803         u64 alloc_target;
2804
2805         alloc_target = btrfs_get_alloc_profile(root, 1);
2806         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2807                                                        alloc_target);
2808 }
2809
2810 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2811 {
2812         u64 num_bytes;
2813         int level;
2814
2815         level = BTRFS_MAX_LEVEL - 2;
2816         /*
2817          * NOTE: these calculations are absolutely the worst possible case.
2818          * This assumes that _every_ item we insert will require a new leaf, and
2819          * that the tree has grown to its maximum level size.
2820          */
2821
2822         /*
2823          * for every item we insert we could insert both an extent item and a
2824          * extent ref item.  Then for ever item we insert, we will need to cow
2825          * both the original leaf, plus the leaf to the left and right of it.
2826          *
2827          * Unless we are talking about the extent root, then we just want the
2828          * number of items * 2, since we just need the extent item plus its ref.
2829          */
2830         if (root == root->fs_info->extent_root)
2831                 num_bytes = num_items * 2;
2832         else
2833                 num_bytes = (num_items + (2 * num_items)) * 3;
2834
2835         /*
2836          * num_bytes is total number of leaves we could need times the leaf
2837          * size, and then for every leaf we could end up cow'ing 2 nodes per
2838          * level, down to the leaf level.
2839          */
2840         num_bytes = (num_bytes * root->leafsize) +
2841                 (num_bytes * (level * 2)) * root->nodesize;
2842
2843         return num_bytes;
2844 }
2845
2846 /*
2847  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2848  * we have extents, this function does nothing.
2849  */
2850 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2851                                           struct inode *inode, int num_items)
2852 {
2853         struct btrfs_fs_info *info = root->fs_info;
2854         struct btrfs_space_info *meta_sinfo;
2855         u64 num_bytes;
2856         u64 alloc_target;
2857         bool bug = false;
2858
2859         /* get the space info for where the metadata will live */
2860         alloc_target = btrfs_get_alloc_profile(root, 0);
2861         meta_sinfo = __find_space_info(info, alloc_target);
2862
2863         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2864                                            num_items);
2865
2866         spin_lock(&meta_sinfo->lock);
2867         spin_lock(&BTRFS_I(inode)->accounting_lock);
2868         if (BTRFS_I(inode)->reserved_extents <=
2869             BTRFS_I(inode)->outstanding_extents) {
2870                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2871                 spin_unlock(&meta_sinfo->lock);
2872                 return 0;
2873         }
2874         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2875
2876         BTRFS_I(inode)->reserved_extents -= num_items;
2877         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2878
2879         if (meta_sinfo->bytes_delalloc < num_bytes) {
2880                 bug = true;
2881                 meta_sinfo->bytes_delalloc = 0;
2882         } else {
2883                 meta_sinfo->bytes_delalloc -= num_bytes;
2884         }
2885         spin_unlock(&meta_sinfo->lock);
2886
2887         BUG_ON(bug);
2888
2889         return 0;
2890 }
2891
2892 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2893 {
2894         u64 thresh;
2895
2896         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2897                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2898                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2899                 meta_sinfo->bytes_may_use;
2900
2901         thresh = meta_sinfo->total_bytes - thresh;
2902         thresh *= 80;
2903         do_div(thresh, 100);
2904         if (thresh <= meta_sinfo->bytes_delalloc)
2905                 meta_sinfo->force_delalloc = 1;
2906         else
2907                 meta_sinfo->force_delalloc = 0;
2908 }
2909
2910 struct async_flush {
2911         struct btrfs_root *root;
2912         struct btrfs_space_info *info;
2913         struct btrfs_work work;
2914 };
2915
2916 static noinline void flush_delalloc_async(struct btrfs_work *work)
2917 {
2918         struct async_flush *async;
2919         struct btrfs_root *root;
2920         struct btrfs_space_info *info;
2921
2922         async = container_of(work, struct async_flush, work);
2923         root = async->root;
2924         info = async->info;
2925
2926         btrfs_start_delalloc_inodes(root, 0);
2927         wake_up(&info->flush_wait);
2928         btrfs_wait_ordered_extents(root, 0, 0);
2929
2930         spin_lock(&info->lock);
2931         info->flushing = 0;
2932         spin_unlock(&info->lock);
2933         wake_up(&info->flush_wait);
2934
2935         kfree(async);
2936 }
2937
2938 static void wait_on_flush(struct btrfs_space_info *info)
2939 {
2940         DEFINE_WAIT(wait);
2941         u64 used;
2942
2943         while (1) {
2944                 prepare_to_wait(&info->flush_wait, &wait,
2945                                 TASK_UNINTERRUPTIBLE);
2946                 spin_lock(&info->lock);
2947                 if (!info->flushing) {
2948                         spin_unlock(&info->lock);
2949                         break;
2950                 }
2951
2952                 used = info->bytes_used + info->bytes_reserved +
2953                         info->bytes_pinned + info->bytes_readonly +
2954                         info->bytes_super + info->bytes_root +
2955                         info->bytes_may_use + info->bytes_delalloc;
2956                 if (used < info->total_bytes) {
2957                         spin_unlock(&info->lock);
2958                         break;
2959                 }
2960                 spin_unlock(&info->lock);
2961                 schedule();
2962         }
2963         finish_wait(&info->flush_wait, &wait);
2964 }
2965
2966 static void flush_delalloc(struct btrfs_root *root,
2967                                  struct btrfs_space_info *info)
2968 {
2969         struct async_flush *async;
2970         bool wait = false;
2971
2972         spin_lock(&info->lock);
2973
2974         if (!info->flushing)
2975                 info->flushing = 1;
2976         else
2977                 wait = true;
2978
2979         spin_unlock(&info->lock);
2980
2981         if (wait) {
2982                 wait_on_flush(info);
2983                 return;
2984         }
2985
2986         async = kzalloc(sizeof(*async), GFP_NOFS);
2987         if (!async)
2988                 goto flush;
2989
2990         async->root = root;
2991         async->info = info;
2992         async->work.func = flush_delalloc_async;
2993
2994         btrfs_queue_worker(&root->fs_info->enospc_workers,
2995                            &async->work);
2996         wait_on_flush(info);
2997         return;
2998
2999 flush:
3000         btrfs_start_delalloc_inodes(root, 0);
3001         btrfs_wait_ordered_extents(root, 0, 0);
3002
3003         spin_lock(&info->lock);
3004         info->flushing = 0;
3005         spin_unlock(&info->lock);
3006         wake_up(&info->flush_wait);
3007 }
3008
3009 /*
3010  * Reserve metadata space for delalloc.
3011  */
3012 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3013                                         struct inode *inode, int num_items)
3014 {
3015         struct btrfs_fs_info *info = root->fs_info;
3016         struct btrfs_space_info *meta_sinfo;
3017         u64 num_bytes;
3018         u64 used;
3019         u64 alloc_target;
3020         int flushed = 0;
3021         int force_delalloc;
3022
3023         /* get the space info for where the metadata will live */
3024         alloc_target = btrfs_get_alloc_profile(root, 0);
3025         meta_sinfo = __find_space_info(info, alloc_target);
3026
3027         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3028                                            num_items);
3029 again:
3030         spin_lock(&meta_sinfo->lock);
3031
3032         force_delalloc = meta_sinfo->force_delalloc;
3033
3034         if (unlikely(!meta_sinfo->bytes_root))
3035                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3036
3037         if (!flushed)
3038                 meta_sinfo->bytes_delalloc += num_bytes;
3039
3040         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3041                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3042                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3043                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3044
3045         if (used > meta_sinfo->total_bytes) {
3046                 flushed++;
3047
3048                 if (flushed == 1) {
3049                         if (maybe_allocate_chunk(NULL, root, meta_sinfo,
3050                                                  num_bytes))
3051                                 goto again;
3052                         flushed++;
3053                 } else {
3054                         spin_unlock(&meta_sinfo->lock);
3055                 }
3056
3057                 if (flushed == 2) {
3058                         filemap_flush(inode->i_mapping);
3059                         goto again;
3060                 } else if (flushed == 3) {
3061                         flush_delalloc(root, meta_sinfo);
3062                         goto again;
3063                 }
3064                 spin_lock(&meta_sinfo->lock);
3065                 meta_sinfo->bytes_delalloc -= num_bytes;
3066                 spin_unlock(&meta_sinfo->lock);
3067                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3068                        BTRFS_I(inode)->outstanding_extents,
3069                        BTRFS_I(inode)->reserved_extents);
3070                 dump_space_info(meta_sinfo, 0, 0);
3071                 return -ENOSPC;
3072         }
3073
3074         BTRFS_I(inode)->reserved_extents += num_items;
3075         check_force_delalloc(meta_sinfo);
3076         spin_unlock(&meta_sinfo->lock);
3077
3078         if (!flushed && force_delalloc)
3079                 filemap_flush(inode->i_mapping);
3080
3081         return 0;
3082 }
3083
3084 /*
3085  * unreserve num_items number of items worth of metadata space.  This needs to
3086  * be paired with btrfs_reserve_metadata_space.
3087  *
3088  * NOTE: if you have the option, run this _AFTER_ you do a
3089  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3090  * oprations which will result in more used metadata, so we want to make sure we
3091  * can do that without issue.
3092  */
3093 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3094 {
3095         struct btrfs_fs_info *info = root->fs_info;
3096         struct btrfs_space_info *meta_sinfo;
3097         u64 num_bytes;
3098         u64 alloc_target;
3099         bool bug = false;
3100
3101         /* get the space info for where the metadata will live */
3102         alloc_target = btrfs_get_alloc_profile(root, 0);
3103         meta_sinfo = __find_space_info(info, alloc_target);
3104
3105         num_bytes = calculate_bytes_needed(root, num_items);
3106
3107         spin_lock(&meta_sinfo->lock);
3108         if (meta_sinfo->bytes_may_use < num_bytes) {
3109                 bug = true;
3110                 meta_sinfo->bytes_may_use = 0;
3111         } else {
3112                 meta_sinfo->bytes_may_use -= num_bytes;
3113         }
3114         spin_unlock(&meta_sinfo->lock);
3115
3116         BUG_ON(bug);
3117
3118         return 0;
3119 }
3120
3121 /*
3122  * Reserve some metadata space for use.  We'll calculate the worste case number
3123  * of bytes that would be needed to modify num_items number of items.  If we
3124  * have space, fantastic, if not, you get -ENOSPC.  Please call
3125  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3126  * items you reserved, since whatever metadata you needed should have already
3127  * been allocated.
3128  *
3129  * This will commit the transaction to make more space if we don't have enough
3130  * metadata space.  THe only time we don't do this is if we're reserving space
3131  * inside of a transaction, then we will just return -ENOSPC and it is the
3132  * callers responsibility to handle it properly.
3133  */
3134 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3135 {
3136         struct btrfs_fs_info *info = root->fs_info;
3137         struct btrfs_space_info *meta_sinfo;
3138         u64 num_bytes;
3139         u64 used;
3140         u64 alloc_target;
3141         int retries = 0;
3142
3143         /* get the space info for where the metadata will live */
3144         alloc_target = btrfs_get_alloc_profile(root, 0);
3145         meta_sinfo = __find_space_info(info, alloc_target);
3146
3147         num_bytes = calculate_bytes_needed(root, num_items);
3148 again:
3149         spin_lock(&meta_sinfo->lock);
3150
3151         if (unlikely(!meta_sinfo->bytes_root))
3152                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3153
3154         if (!retries)
3155                 meta_sinfo->bytes_may_use += num_bytes;
3156
3157         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3158                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3159                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3160                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3161
3162         if (used > meta_sinfo->total_bytes) {
3163                 retries++;
3164                 if (retries == 1) {
3165                         if (maybe_allocate_chunk(NULL, root, meta_sinfo,
3166                                                  num_bytes))
3167                                 goto again;
3168                         retries++;
3169                 } else {
3170                         spin_unlock(&meta_sinfo->lock);
3171                 }
3172
3173                 if (retries == 2) {
3174                         flush_delalloc(root, meta_sinfo);
3175                         goto again;
3176                 }
3177                 spin_lock(&meta_sinfo->lock);
3178                 meta_sinfo->bytes_may_use -= num_bytes;
3179                 spin_unlock(&meta_sinfo->lock);
3180
3181                 dump_space_info(meta_sinfo, 0, 0);
3182                 return -ENOSPC;
3183         }
3184
3185         check_force_delalloc(meta_sinfo);
3186         spin_unlock(&meta_sinfo->lock);
3187
3188         return 0;
3189 }
3190
3191 /*
3192  * This will check the space that the inode allocates from to make sure we have
3193  * enough space for bytes.
3194  */
3195 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3196                                 u64 bytes)
3197 {
3198         struct btrfs_space_info *data_sinfo;
3199         u64 used;
3200         int ret = 0, committed = 0, flushed = 0;
3201
3202         /* make sure bytes are sectorsize aligned */
3203         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3204
3205         data_sinfo = BTRFS_I(inode)->space_info;
3206         if (!data_sinfo)
3207                 goto alloc;
3208
3209 again:
3210         /* make sure we have enough space to handle the data first */
3211         spin_lock(&data_sinfo->lock);
3212         used = data_sinfo->bytes_used + data_sinfo->bytes_delalloc +
3213                 data_sinfo->bytes_reserved + data_sinfo->bytes_pinned +
3214                 data_sinfo->bytes_readonly + data_sinfo->bytes_may_use +
3215                 data_sinfo->bytes_super;
3216
3217         if (used + bytes > data_sinfo->total_bytes) {
3218                 struct btrfs_trans_handle *trans;
3219
3220                 if (!flushed) {
3221                         spin_unlock(&data_sinfo->lock);
3222                         flush_delalloc(root, data_sinfo);
3223                         flushed = 1;
3224                         goto again;
3225                 }
3226
3227                 /*
3228                  * if we don't have enough free bytes in this space then we need
3229                  * to alloc a new chunk.
3230                  */
3231                 if (!data_sinfo->full) {
3232                         u64 alloc_target;
3233
3234                         data_sinfo->force_alloc = 1;
3235                         spin_unlock(&data_sinfo->lock);
3236 alloc:
3237                         alloc_target = btrfs_get_alloc_profile(root, 1);
3238                         trans = btrfs_start_transaction(root, 1);
3239                         if (!trans)
3240                                 return -ENOMEM;
3241
3242                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3243                                              bytes + 2 * 1024 * 1024,
3244                                              alloc_target, 0);
3245                         btrfs_end_transaction(trans, root);
3246                         if (ret)
3247                                 return ret;
3248
3249                         if (!data_sinfo) {
3250                                 btrfs_set_inode_space_info(root, inode);
3251                                 data_sinfo = BTRFS_I(inode)->space_info;
3252                         }
3253                         goto again;
3254                 }
3255                 spin_unlock(&data_sinfo->lock);
3256
3257                 /* commit the current transaction and try again */
3258                 if (!committed && !root->fs_info->open_ioctl_trans) {
3259                         committed = 1;
3260                         trans = btrfs_join_transaction(root, 1);
3261                         if (!trans)
3262                                 return -ENOMEM;
3263                         ret = btrfs_commit_transaction(trans, root);
3264                         if (ret)
3265                                 return ret;
3266                         goto again;
3267                 }
3268
3269                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3270                        ", %llu bytes_used, %llu bytes_reserved, "
3271                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3272                        "%llu total\n", (unsigned long long)bytes,
3273                        (unsigned long long)data_sinfo->bytes_delalloc,
3274                        (unsigned long long)data_sinfo->bytes_used,
3275                        (unsigned long long)data_sinfo->bytes_reserved,
3276                        (unsigned long long)data_sinfo->bytes_pinned,
3277                        (unsigned long long)data_sinfo->bytes_readonly,
3278                        (unsigned long long)data_sinfo->bytes_may_use,
3279                        (unsigned long long)data_sinfo->total_bytes);
3280                 return -ENOSPC;
3281         }
3282         data_sinfo->bytes_may_use += bytes;
3283         BTRFS_I(inode)->reserved_bytes += bytes;
3284         spin_unlock(&data_sinfo->lock);
3285
3286         return 0;
3287 }
3288
3289 /*
3290  * if there was an error for whatever reason after calling
3291  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3292  */
3293 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3294                                     struct inode *inode, u64 bytes)
3295 {
3296         struct btrfs_space_info *data_sinfo;
3297
3298         /* make sure bytes are sectorsize aligned */
3299         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3300
3301         data_sinfo = BTRFS_I(inode)->space_info;
3302         spin_lock(&data_sinfo->lock);
3303         data_sinfo->bytes_may_use -= bytes;
3304         BTRFS_I(inode)->reserved_bytes -= bytes;
3305         spin_unlock(&data_sinfo->lock);
3306 }
3307
3308 /* called when we are adding a delalloc extent to the inode's io_tree */
3309 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3310                                   u64 bytes)
3311 {
3312         struct btrfs_space_info *data_sinfo;
3313
3314         /* get the space info for where this inode will be storing its data */
3315         data_sinfo = BTRFS_I(inode)->space_info;
3316
3317         /* make sure we have enough space to handle the data first */
3318         spin_lock(&data_sinfo->lock);
3319         data_sinfo->bytes_delalloc += bytes;
3320
3321         /*
3322          * we are adding a delalloc extent without calling
3323          * btrfs_check_data_free_space first.  This happens on a weird
3324          * writepage condition, but shouldn't hurt our accounting
3325          */
3326         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3327                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3328                 BTRFS_I(inode)->reserved_bytes = 0;
3329         } else {
3330                 data_sinfo->bytes_may_use -= bytes;
3331                 BTRFS_I(inode)->reserved_bytes -= bytes;
3332         }
3333
3334         spin_unlock(&data_sinfo->lock);
3335 }
3336
3337 /* called when we are clearing an delalloc extent from the inode's io_tree */
3338 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3339                               u64 bytes)
3340 {
3341         struct btrfs_space_info *info;
3342
3343         info = BTRFS_I(inode)->space_info;
3344
3345         spin_lock(&info->lock);
3346         info->bytes_delalloc -= bytes;
3347         spin_unlock(&info->lock);
3348 }
3349
3350 static void force_metadata_allocation(struct btrfs_fs_info *info)
3351 {
3352         struct list_head *head = &info->space_info;
3353         struct btrfs_space_info *found;
3354
3355         rcu_read_lock();
3356         list_for_each_entry_rcu(found, head, list) {
3357                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3358                         found->force_alloc = 1;
3359         }
3360         rcu_read_unlock();
3361 }
3362
3363 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3364                               u64 alloc_bytes)
3365 {
3366         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3367
3368         if (sinfo->bytes_used + sinfo->bytes_reserved +
3369             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3370                 return 0;
3371
3372         if (sinfo->bytes_used + sinfo->bytes_reserved +
3373             alloc_bytes < div_factor(num_bytes, 8))
3374                 return 0;
3375
3376         return 1;
3377 }
3378
3379 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3380                           struct btrfs_root *extent_root, u64 alloc_bytes,
3381                           u64 flags, int force)
3382 {
3383         struct btrfs_space_info *space_info;
3384         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3385         int ret = 0;
3386
3387         mutex_lock(&fs_info->chunk_mutex);
3388
3389         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3390
3391         space_info = __find_space_info(extent_root->fs_info, flags);
3392         if (!space_info) {
3393                 ret = update_space_info(extent_root->fs_info, flags,
3394                                         0, 0, &space_info);
3395                 BUG_ON(ret);
3396         }
3397         BUG_ON(!space_info);
3398
3399         spin_lock(&space_info->lock);
3400         if (space_info->force_alloc)
3401                 force = 1;
3402         if (space_info->full) {
3403                 spin_unlock(&space_info->lock);
3404                 goto out;
3405         }
3406
3407         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3408                 spin_unlock(&space_info->lock);
3409                 goto out;
3410         }
3411         spin_unlock(&space_info->lock);
3412
3413         /*
3414          * if we're doing a data chunk, go ahead and make sure that
3415          * we keep a reasonable number of metadata chunks allocated in the
3416          * FS as well.
3417          */
3418         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3419                 fs_info->data_chunk_allocations++;
3420                 if (!(fs_info->data_chunk_allocations %
3421                       fs_info->metadata_ratio))
3422                         force_metadata_allocation(fs_info);
3423         }
3424
3425         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3426         spin_lock(&space_info->lock);
3427         if (ret)
3428                 space_info->full = 1;
3429         else
3430                 ret = 1;
3431         space_info->force_alloc = 0;
3432         spin_unlock(&space_info->lock);
3433 out:
3434         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3435         return ret;
3436 }
3437
3438 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3439                                 struct btrfs_root *root,
3440                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3441 {
3442         int ret;
3443         int end_trans = 0;
3444
3445         if (sinfo->full)
3446                 return 0;
3447
3448         spin_lock(&sinfo->lock);
3449         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3450         spin_unlock(&sinfo->lock);
3451         if (!ret)
3452                 return 0;
3453
3454         if (!trans) {
3455                 trans = btrfs_join_transaction(root, 1);
3456                 BUG_ON(IS_ERR(trans));
3457                 end_trans = 1;
3458         }
3459
3460         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3461                              num_bytes + 2 * 1024 * 1024,
3462                              get_alloc_profile(root, sinfo->flags), 0);
3463
3464         if (end_trans)
3465                 btrfs_end_transaction(trans, root);
3466
3467         return ret == 1 ? 1 : 0;
3468 }
3469
3470 static int update_block_group(struct btrfs_trans_handle *trans,
3471                               struct btrfs_root *root,
3472                               u64 bytenr, u64 num_bytes, int alloc,
3473                               int mark_free)
3474 {
3475         struct btrfs_block_group_cache *cache;
3476         struct btrfs_fs_info *info = root->fs_info;
3477         int factor;
3478         u64 total = num_bytes;
3479         u64 old_val;
3480         u64 byte_in_group;
3481
3482         /* block accounting for super block */
3483         spin_lock(&info->delalloc_lock);
3484         old_val = btrfs_super_bytes_used(&info->super_copy);
3485         if (alloc)
3486                 old_val += num_bytes;
3487         else
3488                 old_val -= num_bytes;
3489         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3490         spin_unlock(&info->delalloc_lock);
3491
3492         while (total) {
3493                 cache = btrfs_lookup_block_group(info, bytenr);
3494                 if (!cache)
3495                         return -1;
3496                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3497                                     BTRFS_BLOCK_GROUP_RAID1 |
3498                                     BTRFS_BLOCK_GROUP_RAID10))
3499                         factor = 2;
3500                 else
3501                         factor = 1;
3502                 byte_in_group = bytenr - cache->key.objectid;
3503                 WARN_ON(byte_in_group > cache->key.offset);
3504
3505                 spin_lock(&cache->space_info->lock);
3506                 spin_lock(&cache->lock);
3507                 cache->dirty = 1;
3508                 old_val = btrfs_block_group_used(&cache->item);
3509                 num_bytes = min(total, cache->key.offset - byte_in_group);
3510                 if (alloc) {
3511                         old_val += num_bytes;
3512                         btrfs_set_block_group_used(&cache->item, old_val);
3513                         cache->reserved -= num_bytes;
3514                         cache->space_info->bytes_reserved -= num_bytes;
3515                         cache->space_info->bytes_used += num_bytes;
3516                         cache->space_info->disk_used += num_bytes * factor;
3517                         if (cache->ro)
3518                                 cache->space_info->bytes_readonly -= num_bytes;
3519                         spin_unlock(&cache->lock);
3520                         spin_unlock(&cache->space_info->lock);
3521                 } else {
3522                         old_val -= num_bytes;
3523                         btrfs_set_block_group_used(&cache->item, old_val);
3524                         cache->space_info->bytes_used -= num_bytes;
3525                         cache->space_info->disk_used -= num_bytes * factor;
3526                         if (cache->ro)
3527                                 cache->space_info->bytes_readonly += num_bytes;
3528                         spin_unlock(&cache->lock);
3529                         spin_unlock(&cache->space_info->lock);
3530                         if (mark_free) {
3531                                 int ret;
3532
3533                                 ret = btrfs_discard_extent(root, bytenr,
3534                                                            num_bytes);
3535                                 WARN_ON(ret);
3536
3537                                 ret = btrfs_add_free_space(cache, bytenr,
3538                                                            num_bytes);
3539                                 WARN_ON(ret);
3540                         }
3541                 }
3542                 btrfs_put_block_group(cache);
3543                 total -= num_bytes;
3544                 bytenr += num_bytes;
3545         }
3546         return 0;
3547 }
3548
3549 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3550 {
3551         struct btrfs_block_group_cache *cache;
3552         u64 bytenr;
3553
3554         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3555         if (!cache)
3556                 return 0;
3557
3558         bytenr = cache->key.objectid;
3559         btrfs_put_block_group(cache);
3560
3561         return bytenr;
3562 }
3563
3564 /*
3565  * this function must be called within transaction
3566  */
3567 int btrfs_pin_extent(struct btrfs_root *root,
3568                      u64 bytenr, u64 num_bytes, int reserved)
3569 {
3570         struct btrfs_fs_info *fs_info = root->fs_info;
3571         struct btrfs_block_group_cache *cache;
3572
3573         cache = btrfs_lookup_block_group(fs_info, bytenr);
3574         BUG_ON(!cache);
3575
3576         spin_lock(&cache->space_info->lock);
3577         spin_lock(&cache->lock);
3578         cache->pinned += num_bytes;
3579         cache->space_info->bytes_pinned += num_bytes;
3580         if (reserved) {
3581                 cache->reserved -= num_bytes;
3582                 cache->space_info->bytes_reserved -= num_bytes;
3583         }
3584         spin_unlock(&cache->lock);
3585         spin_unlock(&cache->space_info->lock);
3586
3587         btrfs_put_block_group(cache);
3588
3589         set_extent_dirty(fs_info->pinned_extents,
3590                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3591         return 0;
3592 }
3593
3594 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3595                                    u64 num_bytes, int reserve)
3596 {
3597         spin_lock(&cache->space_info->lock);
3598         spin_lock(&cache->lock);
3599         if (reserve) {
3600                 cache->reserved += num_bytes;
3601                 cache->space_info->bytes_reserved += num_bytes;
3602         } else {
3603                 cache->reserved -= num_bytes;
3604                 cache->space_info->bytes_reserved -= num_bytes;
3605         }
3606         spin_unlock(&cache->lock);
3607         spin_unlock(&cache->space_info->lock);
3608         return 0;
3609 }
3610
3611 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3612                                 struct btrfs_root *root)
3613 {
3614         struct btrfs_fs_info *fs_info = root->fs_info;
3615         struct btrfs_caching_control *next;
3616         struct btrfs_caching_control *caching_ctl;
3617         struct btrfs_block_group_cache *cache;
3618
3619         down_write(&fs_info->extent_commit_sem);
3620
3621         list_for_each_entry_safe(caching_ctl, next,
3622                                  &fs_info->caching_block_groups, list) {
3623                 cache = caching_ctl->block_group;
3624                 if (block_group_cache_done(cache)) {
3625                         cache->last_byte_to_unpin = (u64)-1;
3626                         list_del_init(&caching_ctl->list);
3627                         put_caching_control(caching_ctl);
3628                 } else {
3629                         cache->last_byte_to_unpin = caching_ctl->progress;
3630                 }
3631         }
3632
3633         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3634                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3635         else
3636                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3637
3638         up_write(&fs_info->extent_commit_sem);
3639         return 0;
3640 }
3641
3642 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3643 {
3644         struct btrfs_fs_info *fs_info = root->fs_info;
3645         struct btrfs_block_group_cache *cache = NULL;
3646         u64 len;
3647
3648         while (start <= end) {
3649                 if (!cache ||
3650                     start >= cache->key.objectid + cache->key.offset) {
3651                         if (cache)
3652                                 btrfs_put_block_group(cache);
3653                         cache = btrfs_lookup_block_group(fs_info, start);
3654                         BUG_ON(!cache);
3655                 }
3656
3657                 len = cache->key.objectid + cache->key.offset - start;
3658                 len = min(len, end + 1 - start);
3659
3660                 if (start < cache->last_byte_to_unpin) {
3661                         len = min(len, cache->last_byte_to_unpin - start);
3662                         btrfs_add_free_space(cache, start, len);
3663                 }
3664
3665                 spin_lock(&cache->space_info->lock);
3666                 spin_lock(&cache->lock);
3667                 cache->pinned -= len;
3668                 cache->space_info->bytes_pinned -= len;
3669                 spin_unlock(&cache->lock);
3670                 spin_unlock(&cache->space_info->lock);
3671
3672                 start += len;
3673         }
3674
3675         if (cache)
3676                 btrfs_put_block_group(cache);
3677         return 0;
3678 }
3679
3680 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3681                                struct btrfs_root *root)
3682 {
3683         struct btrfs_fs_info *fs_info = root->fs_info;
3684         struct extent_io_tree *unpin;
3685         u64 start;
3686         u64 end;
3687         int ret;
3688
3689         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3690                 unpin = &fs_info->freed_extents[1];
3691         else
3692                 unpin = &fs_info->freed_extents[0];
3693
3694         while (1) {
3695                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3696                                             EXTENT_DIRTY);
3697                 if (ret)
3698                         break;
3699
3700                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3701
3702                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3703                 unpin_extent_range(root, start, end);
3704                 cond_resched();
3705         }
3706
3707         return ret;
3708 }
3709
3710 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3711                           struct btrfs_root *root,
3712                           struct btrfs_path *path,
3713                           u64 bytenr, u64 num_bytes,
3714                           int is_data, int reserved,
3715                           struct extent_buffer **must_clean)
3716 {
3717         int err = 0;
3718         struct extent_buffer *buf;
3719
3720         if (is_data)
3721                 goto pinit;
3722
3723         /*
3724          * discard is sloooow, and so triggering discards on
3725          * individual btree blocks isn't a good plan.  Just
3726          * pin everything in discard mode.
3727          */
3728         if (btrfs_test_opt(root, DISCARD))
3729                 goto pinit;
3730
3731         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3732         if (!buf)
3733                 goto pinit;
3734
3735         /* we can reuse a block if it hasn't been written
3736          * and it is from this transaction.  We can't
3737          * reuse anything from the tree log root because
3738          * it has tiny sub-transactions.
3739          */
3740         if (btrfs_buffer_uptodate(buf, 0) &&
3741             btrfs_try_tree_lock(buf)) {
3742                 u64 header_owner = btrfs_header_owner(buf);
3743                 u64 header_transid = btrfs_header_generation(buf);
3744                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3745                     header_transid == trans->transid &&
3746                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3747                         *must_clean = buf;
3748                         return 1;
3749                 }
3750                 btrfs_tree_unlock(buf);
3751         }
3752         free_extent_buffer(buf);
3753 pinit:
3754         if (path)
3755                 btrfs_set_path_blocking(path);
3756         /* unlocks the pinned mutex */
3757         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3758
3759         BUG_ON(err < 0);
3760         return 0;
3761 }
3762
3763 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3764                                 struct btrfs_root *root,
3765                                 u64 bytenr, u64 num_bytes, u64 parent,
3766                                 u64 root_objectid, u64 owner_objectid,
3767                                 u64 owner_offset, int refs_to_drop,
3768                                 struct btrfs_delayed_extent_op *extent_op)
3769 {
3770         struct btrfs_key key;
3771         struct btrfs_path *path;
3772         struct btrfs_fs_info *info = root->fs_info;
3773         struct btrfs_root *extent_root = info->extent_root;
3774         struct extent_buffer *leaf;
3775         struct btrfs_extent_item *ei;
3776         struct btrfs_extent_inline_ref *iref;
3777         int ret;
3778         int is_data;
3779         int extent_slot = 0;
3780         int found_extent = 0;
3781         int num_to_del = 1;
3782         u32 item_size;
3783         u64 refs;
3784
3785         path = btrfs_alloc_path();
3786         if (!path)
3787                 return -ENOMEM;
3788
3789         path->reada = 1;
3790         path->leave_spinning = 1;
3791
3792         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3793         BUG_ON(!is_data && refs_to_drop != 1);
3794
3795         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3796                                     bytenr, num_bytes, parent,
3797                                     root_objectid, owner_objectid,
3798                                     owner_offset);
3799         if (ret == 0) {
3800                 extent_slot = path->slots[0];
3801                 while (extent_slot >= 0) {
3802                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3803                                               extent_slot);
3804                         if (key.objectid != bytenr)
3805                                 break;
3806                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3807                             key.offset == num_bytes) {
3808                                 found_extent = 1;
3809                                 break;
3810                         }
3811                         if (path->slots[0] - extent_slot > 5)
3812                                 break;
3813                         extent_slot--;
3814                 }
3815 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3816                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3817                 if (found_extent && item_size < sizeof(*ei))
3818                         found_extent = 0;
3819 #endif
3820                 if (!found_extent) {
3821                         BUG_ON(iref);
3822                         ret = remove_extent_backref(trans, extent_root, path,
3823                                                     NULL, refs_to_drop,
3824                                                     is_data);
3825                         BUG_ON(ret);
3826                         btrfs_release_path(extent_root, path);
3827                         path->leave_spinning = 1;
3828
3829                         key.objectid = bytenr;
3830                         key.type = BTRFS_EXTENT_ITEM_KEY;
3831                         key.offset = num_bytes;
3832
3833                         ret = btrfs_search_slot(trans, extent_root,
3834                                                 &key, path, -1, 1);
3835                         if (ret) {
3836                                 printk(KERN_ERR "umm, got %d back from search"
3837                                        ", was looking for %llu\n", ret,
3838                                        (unsigned long long)bytenr);
3839                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3840                         }
3841                         BUG_ON(ret);
3842                         extent_slot = path->slots[0];
3843                 }
3844         } else {
3845                 btrfs_print_leaf(extent_root, path->nodes[0]);
3846                 WARN_ON(1);
3847                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3848                        "parent %llu root %llu  owner %llu offset %llu\n",
3849                        (unsigned long long)bytenr,
3850                        (unsigned long long)parent,
3851                        (unsigned long long)root_objectid,
3852                        (unsigned long long)owner_objectid,
3853                        (unsigned long long)owner_offset);
3854         }
3855
3856         leaf = path->nodes[0];
3857         item_size = btrfs_item_size_nr(leaf, extent_slot);
3858 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3859         if (item_size < sizeof(*ei)) {
3860                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3861                 ret = convert_extent_item_v0(trans, extent_root, path,
3862                                              owner_objectid, 0);
3863                 BUG_ON(ret < 0);
3864
3865                 btrfs_release_path(extent_root, path);
3866                 path->leave_spinning = 1;
3867
3868                 key.objectid = bytenr;
3869                 key.type = BTRFS_EXTENT_ITEM_KEY;
3870                 key.offset = num_bytes;
3871
3872                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3873                                         -1, 1);
3874                 if (ret) {
3875                         printk(KERN_ERR "umm, got %d back from search"
3876                                ", was looking for %llu\n", ret,
3877                                (unsigned long long)bytenr);
3878                         btrfs_print_leaf(extent_root, path->nodes[0]);
3879                 }
3880                 BUG_ON(ret);
3881                 extent_slot = path->slots[0];
3882                 leaf = path->nodes[0];
3883                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3884         }
3885 #endif
3886         BUG_ON(item_size < sizeof(*ei));
3887         ei = btrfs_item_ptr(leaf, extent_slot,
3888                             struct btrfs_extent_item);
3889         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3890                 struct btrfs_tree_block_info *bi;
3891                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3892                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3893                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3894         }
3895
3896         refs = btrfs_extent_refs(leaf, ei);
3897         BUG_ON(refs < refs_to_drop);
3898         refs -= refs_to_drop;
3899
3900         if (refs > 0) {
3901                 if (extent_op)
3902                         __run_delayed_extent_op(extent_op, leaf, ei);
3903                 /*
3904                  * In the case of inline back ref, reference count will
3905                  * be updated by remove_extent_backref
3906                  */
3907                 if (iref) {
3908                         BUG_ON(!found_extent);
3909                 } else {
3910                         btrfs_set_extent_refs(leaf, ei, refs);
3911                         btrfs_mark_buffer_dirty(leaf);
3912                 }
3913                 if (found_extent) {
3914                         ret = remove_extent_backref(trans, extent_root, path,
3915                                                     iref, refs_to_drop,
3916                                                     is_data);
3917                         BUG_ON(ret);
3918                 }
3919         } else {
3920                 int mark_free = 0;
3921                 struct extent_buffer *must_clean = NULL;
3922
3923                 if (found_extent) {
3924                         BUG_ON(is_data && refs_to_drop !=
3925                                extent_data_ref_count(root, path, iref));
3926                         if (iref) {
3927                                 BUG_ON(path->slots[0] != extent_slot);
3928                         } else {
3929                                 BUG_ON(path->slots[0] != extent_slot + 1);
3930                                 path->slots[0] = extent_slot;
3931                                 num_to_del = 2;
3932                         }
3933                 }
3934
3935                 ret = pin_down_bytes(trans, root, path, bytenr,
3936                                      num_bytes, is_data, 0, &must_clean);
3937                 if (ret > 0)
3938                         mark_free = 1;
3939                 BUG_ON(ret < 0);
3940                 /*
3941                  * it is going to be very rare for someone to be waiting
3942                  * on the block we're freeing.  del_items might need to
3943                  * schedule, so rather than get fancy, just force it
3944                  * to blocking here
3945                  */
3946                 if (must_clean)
3947                         btrfs_set_lock_blocking(must_clean);
3948
3949                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3950                                       num_to_del);
3951                 BUG_ON(ret);
3952                 btrfs_release_path(extent_root, path);
3953
3954                 if (must_clean) {
3955                         clean_tree_block(NULL, root, must_clean);
3956                         btrfs_tree_unlock(must_clean);
3957                         free_extent_buffer(must_clean);
3958                 }
3959
3960                 if (is_data) {
3961                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3962                         BUG_ON(ret);
3963                 } else {
3964                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3965                              bytenr >> PAGE_CACHE_SHIFT,
3966                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3967                 }
3968
3969                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3970                                          mark_free);
3971                 BUG_ON(ret);
3972         }
3973         btrfs_free_path(path);
3974         return ret;
3975 }
3976
3977 /*
3978  * when we free an extent, it is possible (and likely) that we free the last
3979  * delayed ref for that extent as well.  This searches the delayed ref tree for
3980  * a given extent, and if there are no other delayed refs to be processed, it
3981  * removes it from the tree.
3982  */
3983 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3984                                       struct btrfs_root *root, u64 bytenr)
3985 {
3986         struct btrfs_delayed_ref_head *head;
3987         struct btrfs_delayed_ref_root *delayed_refs;
3988         struct btrfs_delayed_ref_node *ref;
3989         struct rb_node *node;
3990         int ret;
3991
3992         delayed_refs = &trans->transaction->delayed_refs;
3993         spin_lock(&delayed_refs->lock);
3994         head = btrfs_find_delayed_ref_head(trans, bytenr);
3995         if (!head)
3996                 goto out;
3997
3998         node = rb_prev(&head->node.rb_node);
3999         if (!node)
4000                 goto out;
4001
4002         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4003
4004         /* there are still entries for this ref, we can't drop it */
4005         if (ref->bytenr == bytenr)
4006                 goto out;
4007
4008         if (head->extent_op) {
4009                 if (!head->must_insert_reserved)
4010                         goto out;
4011                 kfree(head->extent_op);
4012                 head->extent_op = NULL;
4013         }
4014
4015         /*
4016          * waiting for the lock here would deadlock.  If someone else has it
4017          * locked they are already in the process of dropping it anyway
4018          */
4019         if (!mutex_trylock(&head->mutex))
4020                 goto out;
4021
4022         /*
4023          * at this point we have a head with no other entries.  Go
4024          * ahead and process it.
4025          */
4026         head->node.in_tree = 0;
4027         rb_erase(&head->node.rb_node, &delayed_refs->root);
4028
4029         delayed_refs->num_entries--;
4030
4031         /*
4032          * we don't take a ref on the node because we're removing it from the
4033          * tree, so we just steal the ref the tree was holding.
4034          */
4035         delayed_refs->num_heads--;
4036         if (list_empty(&head->cluster))
4037                 delayed_refs->num_heads_ready--;
4038
4039         list_del_init(&head->cluster);
4040         spin_unlock(&delayed_refs->lock);
4041
4042         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4043                                   &head->node, head->extent_op,
4044                                   head->must_insert_reserved);
4045         BUG_ON(ret);
4046         btrfs_put_delayed_ref(&head->node);
4047         return 0;
4048 out:
4049         spin_unlock(&delayed_refs->lock);
4050         return 0;
4051 }
4052
4053 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4054                       struct btrfs_root *root,
4055                       u64 bytenr, u64 num_bytes, u64 parent,
4056                       u64 root_objectid, u64 owner, u64 offset)
4057 {
4058         int ret;
4059
4060         /*
4061          * tree log blocks never actually go into the extent allocation
4062          * tree, just update pinning info and exit early.
4063          */
4064         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4065                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4066                 /* unlocks the pinned mutex */
4067                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4068                 ret = 0;
4069         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4070                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4071                                         parent, root_objectid, (int)owner,
4072                                         BTRFS_DROP_DELAYED_REF, NULL);
4073                 BUG_ON(ret);
4074                 ret = check_ref_cleanup(trans, root, bytenr);
4075                 BUG_ON(ret);
4076         } else {
4077                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4078                                         parent, root_objectid, owner,
4079                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4080                 BUG_ON(ret);
4081         }
4082         return ret;
4083 }
4084
4085 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4086                           struct btrfs_root *root,
4087                           u64 bytenr, u32 blocksize,
4088                           u64 parent, u64 root_objectid, int level)
4089 {
4090         u64 used;
4091         spin_lock(&root->node_lock);
4092         used = btrfs_root_used(&root->root_item) - blocksize;
4093         btrfs_set_root_used(&root->root_item, used);
4094         spin_unlock(&root->node_lock);
4095
4096         return btrfs_free_extent(trans, root, bytenr, blocksize,
4097                                  parent, root_objectid, level, 0);
4098 }
4099
4100 static u64 stripe_align(struct btrfs_root *root, u64 val)
4101 {
4102         u64 mask = ((u64)root->stripesize - 1);
4103         u64 ret = (val + mask) & ~mask;
4104         return ret;
4105 }
4106
4107 /*
4108  * when we wait for progress in the block group caching, its because
4109  * our allocation attempt failed at least once.  So, we must sleep
4110  * and let some progress happen before we try again.
4111  *
4112  * This function will sleep at least once waiting for new free space to
4113  * show up, and then it will check the block group free space numbers
4114  * for our min num_bytes.  Another option is to have it go ahead
4115  * and look in the rbtree for a free extent of a given size, but this
4116  * is a good start.
4117  */
4118 static noinline int
4119 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4120                                 u64 num_bytes)
4121 {
4122         struct btrfs_caching_control *caching_ctl;
4123         DEFINE_WAIT(wait);
4124
4125         caching_ctl = get_caching_control(cache);
4126         if (!caching_ctl)
4127                 return 0;
4128
4129         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4130                    (cache->free_space >= num_bytes));
4131
4132         put_caching_control(caching_ctl);
4133         return 0;
4134 }
4135
4136 static noinline int
4137 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4138 {
4139         struct btrfs_caching_control *caching_ctl;
4140         DEFINE_WAIT(wait);
4141
4142         caching_ctl = get_caching_control(cache);
4143         if (!caching_ctl)
4144                 return 0;
4145
4146         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4147
4148         put_caching_control(caching_ctl);
4149         return 0;
4150 }
4151
4152 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4153 {
4154         int index;
4155         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4156                 index = 0;
4157         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4158                 index = 1;
4159         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4160                 index = 2;
4161         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4162                 index = 3;
4163         else
4164                 index = 4;
4165         return index;
4166 }
4167
4168 enum btrfs_loop_type {
4169         LOOP_FIND_IDEAL = 0,
4170         LOOP_CACHING_NOWAIT = 1,
4171         LOOP_CACHING_WAIT = 2,
4172         LOOP_ALLOC_CHUNK = 3,
4173         LOOP_NO_EMPTY_SIZE = 4,
4174 };
4175
4176 /*
4177  * walks the btree of allocated extents and find a hole of a given size.
4178  * The key ins is changed to record the hole:
4179  * ins->objectid == block start
4180  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4181  * ins->offset == number of blocks
4182  * Any available blocks before search_start are skipped.
4183  */
4184 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4185                                      struct btrfs_root *orig_root,
4186                                      u64 num_bytes, u64 empty_size,
4187                                      u64 search_start, u64 search_end,
4188                                      u64 hint_byte, struct btrfs_key *ins,
4189                                      u64 exclude_start, u64 exclude_nr,
4190                                      int data)
4191 {
4192         int ret = 0;
4193         struct btrfs_root *root = orig_root->fs_info->extent_root;
4194         struct btrfs_free_cluster *last_ptr = NULL;
4195         struct btrfs_block_group_cache *block_group = NULL;
4196         int empty_cluster = 2 * 1024 * 1024;
4197         int allowed_chunk_alloc = 0;
4198         int done_chunk_alloc = 0;
4199         struct btrfs_space_info *space_info;
4200         int last_ptr_loop = 0;
4201         int index = 0;
4202         int loop = 0;
4203         bool found_uncached_bg = false;
4204         bool failed_cluster_refill = false;
4205         bool failed_alloc = false;
4206         u64 ideal_cache_percent = 0;
4207         u64 ideal_cache_offset = 0;
4208
4209         WARN_ON(num_bytes < root->sectorsize);
4210         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4211         ins->objectid = 0;
4212         ins->offset = 0;
4213
4214         space_info = __find_space_info(root->fs_info, data);
4215         if (!space_info) {
4216                 printk(KERN_ERR "No space info for %d\n", data);
4217                 return -ENOSPC;
4218         }
4219
4220         if (orig_root->ref_cows || empty_size)
4221                 allowed_chunk_alloc = 1;
4222
4223         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4224                 last_ptr = &root->fs_info->meta_alloc_cluster;
4225                 if (!btrfs_test_opt(root, SSD))
4226                         empty_cluster = 64 * 1024;
4227         }
4228
4229         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4230                 last_ptr = &root->fs_info->data_alloc_cluster;
4231         }
4232
4233         if (last_ptr) {
4234                 spin_lock(&last_ptr->lock);
4235                 if (last_ptr->block_group)
4236                         hint_byte = last_ptr->window_start;
4237                 spin_unlock(&last_ptr->lock);
4238         }
4239
4240         search_start = max(search_start, first_logical_byte(root, 0));
4241         search_start = max(search_start, hint_byte);
4242
4243         if (!last_ptr)
4244                 empty_cluster = 0;
4245
4246         if (search_start == hint_byte) {
4247 ideal_cache:
4248                 block_group = btrfs_lookup_block_group(root->fs_info,
4249                                                        search_start);
4250                 /*
4251                  * we don't want to use the block group if it doesn't match our
4252                  * allocation bits, or if its not cached.
4253                  *
4254                  * However if we are re-searching with an ideal block group
4255                  * picked out then we don't care that the block group is cached.
4256                  */
4257                 if (block_group && block_group_bits(block_group, data) &&
4258                     (block_group->cached != BTRFS_CACHE_NO ||
4259                      search_start == ideal_cache_offset)) {
4260                         down_read(&space_info->groups_sem);
4261                         if (list_empty(&block_group->list) ||
4262                             block_group->ro) {
4263                                 /*
4264                                  * someone is removing this block group,
4265                                  * we can't jump into the have_block_group
4266                                  * target because our list pointers are not
4267                                  * valid
4268                                  */
4269                                 btrfs_put_block_group(block_group);
4270                                 up_read(&space_info->groups_sem);
4271                         } else {
4272                                 index = get_block_group_index(block_group);
4273                                 goto have_block_group;
4274                         }
4275                 } else if (block_group) {
4276                         btrfs_put_block_group(block_group);
4277                 }
4278         }
4279 search:
4280         down_read(&space_info->groups_sem);
4281         list_for_each_entry(block_group, &space_info->block_groups[index],
4282                             list) {
4283                 u64 offset;
4284                 int cached;
4285
4286                 btrfs_get_block_group(block_group);
4287                 search_start = block_group->key.objectid;
4288
4289 have_block_group:
4290                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4291                         u64 free_percent;
4292
4293                         free_percent = btrfs_block_group_used(&block_group->item);
4294                         free_percent *= 100;
4295                         free_percent = div64_u64(free_percent,
4296                                                  block_group->key.offset);
4297                         free_percent = 100 - free_percent;
4298                         if (free_percent > ideal_cache_percent &&
4299                             likely(!block_group->ro)) {
4300                                 ideal_cache_offset = block_group->key.objectid;
4301                                 ideal_cache_percent = free_percent;
4302                         }
4303
4304                         /*
4305                          * We only want to start kthread caching if we are at
4306                          * the point where we will wait for caching to make
4307                          * progress, or if our ideal search is over and we've
4308                          * found somebody to start caching.
4309                          */
4310                         if (loop > LOOP_CACHING_NOWAIT ||
4311                             (loop > LOOP_FIND_IDEAL &&
4312                              atomic_read(&space_info->caching_threads) < 2)) {
4313                                 ret = cache_block_group(block_group);
4314                                 BUG_ON(ret);
4315                         }
4316                         found_uncached_bg = true;
4317
4318                         /*
4319                          * If loop is set for cached only, try the next block
4320                          * group.
4321                          */
4322                         if (loop == LOOP_FIND_IDEAL)
4323                                 goto loop;
4324                 }
4325
4326                 cached = block_group_cache_done(block_group);
4327                 if (unlikely(!cached))
4328                         found_uncached_bg = true;
4329
4330                 if (unlikely(block_group->ro))
4331                         goto loop;
4332
4333                 /*
4334                  * Ok we want to try and use the cluster allocator, so lets look
4335                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4336                  * have tried the cluster allocator plenty of times at this
4337                  * point and not have found anything, so we are likely way too
4338                  * fragmented for the clustering stuff to find anything, so lets
4339                  * just skip it and let the allocator find whatever block it can
4340                  * find
4341                  */
4342                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4343                         /*
4344                          * the refill lock keeps out other
4345                          * people trying to start a new cluster
4346                          */
4347                         spin_lock(&last_ptr->refill_lock);
4348                         if (last_ptr->block_group &&
4349                             (last_ptr->block_group->ro ||
4350                             !block_group_bits(last_ptr->block_group, data))) {
4351                                 offset = 0;
4352                                 goto refill_cluster;
4353                         }
4354
4355                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4356                                                  num_bytes, search_start);
4357                         if (offset) {
4358                                 /* we have a block, we're done */
4359                                 spin_unlock(&last_ptr->refill_lock);
4360                                 goto checks;
4361                         }
4362
4363                         spin_lock(&last_ptr->lock);
4364                         /*
4365                          * whoops, this cluster doesn't actually point to
4366                          * this block group.  Get a ref on the block
4367                          * group is does point to and try again
4368                          */
4369                         if (!last_ptr_loop && last_ptr->block_group &&
4370                             last_ptr->block_group != block_group) {
4371
4372                                 btrfs_put_block_group(block_group);
4373                                 block_group = last_ptr->block_group;
4374                                 btrfs_get_block_group(block_group);
4375                                 spin_unlock(&last_ptr->lock);
4376                                 spin_unlock(&last_ptr->refill_lock);
4377
4378                                 last_ptr_loop = 1;
4379                                 search_start = block_group->key.objectid;
4380                                 /*
4381                                  * we know this block group is properly
4382                                  * in the list because
4383                                  * btrfs_remove_block_group, drops the
4384                                  * cluster before it removes the block
4385                                  * group from the list
4386                                  */
4387                                 goto have_block_group;
4388                         }
4389                         spin_unlock(&last_ptr->lock);
4390 refill_cluster:
4391                         /*
4392                          * this cluster didn't work out, free it and
4393                          * start over
4394                          */
4395                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4396
4397                         last_ptr_loop = 0;
4398
4399                         /* allocate a cluster in this block group */
4400                         ret = btrfs_find_space_cluster(trans, root,
4401                                                block_group, last_ptr,
4402                                                offset, num_bytes,
4403                                                empty_cluster + empty_size);
4404                         if (ret == 0) {
4405                                 /*
4406                                  * now pull our allocation out of this
4407                                  * cluster
4408                                  */
4409                                 offset = btrfs_alloc_from_cluster(block_group,
4410                                                   last_ptr, num_bytes,
4411                                                   search_start);
4412                                 if (offset) {
4413                                         /* we found one, proceed */
4414                                         spin_unlock(&last_ptr->refill_lock);
4415                                         goto checks;
4416                                 }
4417                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4418                                    && !failed_cluster_refill) {
4419                                 spin_unlock(&last_ptr->refill_lock);
4420
4421                                 failed_cluster_refill = true;
4422                                 wait_block_group_cache_progress(block_group,
4423                                        num_bytes + empty_cluster + empty_size);
4424                                 goto have_block_group;
4425                         }
4426
4427                         /*
4428                          * at this point we either didn't find a cluster
4429                          * or we weren't able to allocate a block from our
4430                          * cluster.  Free the cluster we've been trying
4431                          * to use, and go to the next block group
4432                          */
4433                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4434                         spin_unlock(&last_ptr->refill_lock);
4435                         goto loop;
4436                 }
4437
4438                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4439                                                     num_bytes, empty_size);
4440                 /*
4441                  * If we didn't find a chunk, and we haven't failed on this
4442                  * block group before, and this block group is in the middle of
4443                  * caching and we are ok with waiting, then go ahead and wait
4444                  * for progress to be made, and set failed_alloc to true.
4445                  *
4446                  * If failed_alloc is true then we've already waited on this
4447                  * block group once and should move on to the next block group.
4448                  */
4449                 if (!offset && !failed_alloc && !cached &&
4450                     loop > LOOP_CACHING_NOWAIT) {
4451                         wait_block_group_cache_progress(block_group,
4452                                                 num_bytes + empty_size);
4453                         failed_alloc = true;
4454                         goto have_block_group;
4455                 } else if (!offset) {
4456                         goto loop;
4457                 }
4458 checks:
4459                 search_start = stripe_align(root, offset);
4460                 /* move on to the next group */
4461                 if (search_start + num_bytes >= search_end) {
4462                         btrfs_add_free_space(block_group, offset, num_bytes);
4463                         goto loop;
4464                 }
4465
4466                 /* move on to the next group */
4467                 if (search_start + num_bytes >
4468                     block_group->key.objectid + block_group->key.offset) {
4469                         btrfs_add_free_space(block_group, offset, num_bytes);
4470                         goto loop;
4471                 }
4472
4473                 if (exclude_nr > 0 &&
4474                     (search_start + num_bytes > exclude_start &&
4475                      search_start < exclude_start + exclude_nr)) {
4476                         search_start = exclude_start + exclude_nr;
4477
4478                         btrfs_add_free_space(block_group, offset, num_bytes);
4479                         /*
4480                          * if search_start is still in this block group
4481                          * then we just re-search this block group
4482                          */
4483                         if (search_start >= block_group->key.objectid &&
4484                             search_start < (block_group->key.objectid +
4485                                             block_group->key.offset))
4486                                 goto have_block_group;
4487                         goto loop;
4488                 }
4489
4490                 ins->objectid = search_start;
4491                 ins->offset = num_bytes;
4492
4493                 if (offset < search_start)
4494                         btrfs_add_free_space(block_group, offset,
4495                                              search_start - offset);
4496                 BUG_ON(offset > search_start);
4497
4498                 update_reserved_extents(block_group, num_bytes, 1);
4499
4500                 /* we are all good, lets return */
4501                 break;
4502 loop:
4503                 failed_cluster_refill = false;
4504                 failed_alloc = false;
4505                 BUG_ON(index != get_block_group_index(block_group));
4506                 btrfs_put_block_group(block_group);
4507         }
4508         up_read(&space_info->groups_sem);
4509
4510         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4511                 goto search;
4512
4513         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4514          *                      for them to make caching progress.  Also
4515          *                      determine the best possible bg to cache
4516          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4517          *                      caching kthreads as we move along
4518          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4519          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4520          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4521          *                      again
4522          */
4523         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4524             (found_uncached_bg || empty_size || empty_cluster ||
4525              allowed_chunk_alloc)) {
4526                 index = 0;
4527                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4528                         found_uncached_bg = false;
4529                         loop++;
4530                         if (!ideal_cache_percent &&
4531                             atomic_read(&space_info->caching_threads))
4532                                 goto search;
4533
4534                         /*
4535                          * 1 of the following 2 things have happened so far
4536                          *
4537                          * 1) We found an ideal block group for caching that
4538                          * is mostly full and will cache quickly, so we might
4539                          * as well wait for it.
4540                          *
4541                          * 2) We searched for cached only and we didn't find
4542                          * anything, and we didn't start any caching kthreads
4543                          * either, so chances are we will loop through and
4544                          * start a couple caching kthreads, and then come back
4545                          * around and just wait for them.  This will be slower
4546                          * because we will have 2 caching kthreads reading at
4547                          * the same time when we could have just started one
4548                          * and waited for it to get far enough to give us an
4549                          * allocation, so go ahead and go to the wait caching
4550                          * loop.
4551                          */
4552                         loop = LOOP_CACHING_WAIT;
4553                         search_start = ideal_cache_offset;
4554                         ideal_cache_percent = 0;
4555                         goto ideal_cache;
4556                 } else if (loop == LOOP_FIND_IDEAL) {
4557                         /*
4558                          * Didn't find a uncached bg, wait on anything we find
4559                          * next.
4560                          */
4561                         loop = LOOP_CACHING_WAIT;
4562                         goto search;
4563                 }
4564
4565                 if (loop < LOOP_CACHING_WAIT) {
4566                         loop++;
4567                         goto search;
4568                 }
4569
4570                 if (loop == LOOP_ALLOC_CHUNK) {
4571                         empty_size = 0;
4572                         empty_cluster = 0;
4573                 }
4574
4575                 if (allowed_chunk_alloc) {
4576                         ret = do_chunk_alloc(trans, root, num_bytes +
4577                                              2 * 1024 * 1024, data, 1);
4578                         allowed_chunk_alloc = 0;
4579                         done_chunk_alloc = 1;
4580                 } else if (!done_chunk_alloc) {
4581                         space_info->force_alloc = 1;
4582                 }
4583
4584                 if (loop < LOOP_NO_EMPTY_SIZE) {
4585                         loop++;
4586                         goto search;
4587                 }
4588                 ret = -ENOSPC;
4589         } else if (!ins->objectid) {
4590                 ret = -ENOSPC;
4591         }
4592
4593         /* we found what we needed */
4594         if (ins->objectid) {
4595                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4596                         trans->block_group = block_group->key.objectid;
4597
4598                 btrfs_put_block_group(block_group);
4599                 ret = 0;
4600         }
4601
4602         return ret;
4603 }
4604
4605 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4606                             int dump_block_groups)
4607 {
4608         struct btrfs_block_group_cache *cache;
4609         int index = 0;
4610
4611         spin_lock(&info->lock);
4612         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4613                (unsigned long long)(info->total_bytes - info->bytes_used -
4614                                     info->bytes_pinned - info->bytes_reserved -
4615                                     info->bytes_super),
4616                (info->full) ? "" : "not ");
4617         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4618                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4619                "\n",
4620                (unsigned long long)info->total_bytes,
4621                (unsigned long long)info->bytes_pinned,
4622                (unsigned long long)info->bytes_delalloc,
4623                (unsigned long long)info->bytes_may_use,
4624                (unsigned long long)info->bytes_used,
4625                (unsigned long long)info->bytes_root,
4626                (unsigned long long)info->bytes_super,
4627                (unsigned long long)info->bytes_reserved);
4628         spin_unlock(&info->lock);
4629
4630         if (!dump_block_groups)
4631                 return;
4632
4633         down_read(&info->groups_sem);
4634 again:
4635         list_for_each_entry(cache, &info->block_groups[index], list) {
4636                 spin_lock(&cache->lock);
4637                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4638                        "%llu pinned %llu reserved\n",
4639                        (unsigned long long)cache->key.objectid,
4640                        (unsigned long long)cache->key.offset,
4641                        (unsigned long long)btrfs_block_group_used(&cache->item),
4642                        (unsigned long long)cache->pinned,
4643                        (unsigned long long)cache->reserved);
4644                 btrfs_dump_free_space(cache, bytes);
4645                 spin_unlock(&cache->lock);
4646         }
4647         if (++index < BTRFS_NR_RAID_TYPES)
4648                 goto again;
4649         up_read(&info->groups_sem);
4650 }
4651
4652 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4653                          struct btrfs_root *root,
4654                          u64 num_bytes, u64 min_alloc_size,
4655                          u64 empty_size, u64 hint_byte,
4656                          u64 search_end, struct btrfs_key *ins,
4657                          u64 data)
4658 {
4659         int ret;
4660         u64 search_start = 0;
4661
4662         data = btrfs_get_alloc_profile(root, data);
4663 again:
4664         /*
4665          * the only place that sets empty_size is btrfs_realloc_node, which
4666          * is not called recursively on allocations
4667          */
4668         if (empty_size || root->ref_cows)
4669                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4670                                      num_bytes + 2 * 1024 * 1024, data, 0);
4671
4672         WARN_ON(num_bytes < root->sectorsize);
4673         ret = find_free_extent(trans, root, num_bytes, empty_size,
4674                                search_start, search_end, hint_byte, ins,
4675                                trans->alloc_exclude_start,
4676                                trans->alloc_exclude_nr, data);
4677
4678         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4679                 num_bytes = num_bytes >> 1;
4680                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4681                 num_bytes = max(num_bytes, min_alloc_size);
4682                 do_chunk_alloc(trans, root->fs_info->extent_root,
4683                                num_bytes, data, 1);
4684                 goto again;
4685         }
4686         if (ret == -ENOSPC) {
4687                 struct btrfs_space_info *sinfo;
4688
4689                 sinfo = __find_space_info(root->fs_info, data);
4690                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4691                        "wanted %llu\n", (unsigned long long)data,
4692                        (unsigned long long)num_bytes);
4693                 dump_space_info(sinfo, num_bytes, 1);
4694         }
4695
4696         return ret;
4697 }
4698
4699 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4700 {
4701         struct btrfs_block_group_cache *cache;
4702         int ret = 0;
4703
4704         cache = btrfs_lookup_block_group(root->fs_info, start);
4705         if (!cache) {
4706                 printk(KERN_ERR "Unable to find block group for %llu\n",
4707                        (unsigned long long)start);
4708                 return -ENOSPC;
4709         }
4710
4711         ret = btrfs_discard_extent(root, start, len);
4712
4713         btrfs_add_free_space(cache, start, len);
4714         update_reserved_extents(cache, len, 0);
4715         btrfs_put_block_group(cache);
4716
4717         return ret;
4718 }
4719
4720 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4721                                       struct btrfs_root *root,
4722                                       u64 parent, u64 root_objectid,
4723                                       u64 flags, u64 owner, u64 offset,
4724                                       struct btrfs_key *ins, int ref_mod)
4725 {
4726         int ret;
4727         struct btrfs_fs_info *fs_info = root->fs_info;
4728         struct btrfs_extent_item *extent_item;
4729         struct btrfs_extent_inline_ref *iref;
4730         struct btrfs_path *path;
4731         struct extent_buffer *leaf;
4732         int type;
4733         u32 size;
4734
4735         if (parent > 0)
4736                 type = BTRFS_SHARED_DATA_REF_KEY;
4737         else
4738                 type = BTRFS_EXTENT_DATA_REF_KEY;
4739
4740         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4741
4742         path = btrfs_alloc_path();
4743         BUG_ON(!path);
4744
4745         path->leave_spinning = 1;
4746         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4747                                       ins, size);
4748         BUG_ON(ret);
4749
4750         leaf = path->nodes[0];
4751         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4752                                      struct btrfs_extent_item);
4753         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4754         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4755         btrfs_set_extent_flags(leaf, extent_item,
4756                                flags | BTRFS_EXTENT_FLAG_DATA);
4757
4758         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4759         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4760         if (parent > 0) {
4761                 struct btrfs_shared_data_ref *ref;
4762                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4763                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4764                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4765         } else {
4766                 struct btrfs_extent_data_ref *ref;
4767                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4768                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4769                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4770                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4771                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4772         }
4773
4774         btrfs_mark_buffer_dirty(path->nodes[0]);
4775         btrfs_free_path(path);
4776
4777         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4778                                  1, 0);
4779         if (ret) {
4780                 printk(KERN_ERR "btrfs update block group failed for %llu "
4781                        "%llu\n", (unsigned long long)ins->objectid,
4782                        (unsigned long long)ins->offset);
4783                 BUG();
4784         }
4785         return ret;
4786 }
4787
4788 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4789                                      struct btrfs_root *root,
4790                                      u64 parent, u64 root_objectid,
4791                                      u64 flags, struct btrfs_disk_key *key,
4792                                      int level, struct btrfs_key *ins)
4793 {
4794         int ret;
4795         struct btrfs_fs_info *fs_info = root->fs_info;
4796         struct btrfs_extent_item *extent_item;
4797         struct btrfs_tree_block_info *block_info;
4798         struct btrfs_extent_inline_ref *iref;
4799         struct btrfs_path *path;
4800         struct extent_buffer *leaf;
4801         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4802
4803         path = btrfs_alloc_path();
4804         BUG_ON(!path);
4805
4806         path->leave_spinning = 1;
4807         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4808                                       ins, size);
4809         BUG_ON(ret);
4810
4811         leaf = path->nodes[0];
4812         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4813                                      struct btrfs_extent_item);
4814         btrfs_set_extent_refs(leaf, extent_item, 1);
4815         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4816         btrfs_set_extent_flags(leaf, extent_item,
4817                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4818         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4819
4820         btrfs_set_tree_block_key(leaf, block_info, key);
4821         btrfs_set_tree_block_level(leaf, block_info, level);
4822
4823         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4824         if (parent > 0) {
4825                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4826                 btrfs_set_extent_inline_ref_type(leaf, iref,
4827                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4828                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4829         } else {
4830                 btrfs_set_extent_inline_ref_type(leaf, iref,
4831                                                  BTRFS_TREE_BLOCK_REF_KEY);
4832                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4833         }
4834
4835         btrfs_mark_buffer_dirty(leaf);
4836         btrfs_free_path(path);
4837
4838         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4839                                  1, 0);
4840         if (ret) {
4841                 printk(KERN_ERR "btrfs update block group failed for %llu "
4842                        "%llu\n", (unsigned long long)ins->objectid,
4843                        (unsigned long long)ins->offset);
4844                 BUG();
4845         }
4846         return ret;
4847 }
4848
4849 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4850                                      struct btrfs_root *root,
4851                                      u64 root_objectid, u64 owner,
4852                                      u64 offset, struct btrfs_key *ins)
4853 {
4854         int ret;
4855
4856         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4857
4858         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4859                                          0, root_objectid, owner, offset,
4860                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4861         return ret;
4862 }
4863
4864 /*
4865  * this is used by the tree logging recovery code.  It records that
4866  * an extent has been allocated and makes sure to clear the free
4867  * space cache bits as well
4868  */
4869 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4870                                    struct btrfs_root *root,
4871                                    u64 root_objectid, u64 owner, u64 offset,
4872                                    struct btrfs_key *ins)
4873 {
4874         int ret;
4875         struct btrfs_block_group_cache *block_group;
4876         struct btrfs_caching_control *caching_ctl;
4877         u64 start = ins->objectid;
4878         u64 num_bytes = ins->offset;
4879
4880         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4881         cache_block_group(block_group);
4882         caching_ctl = get_caching_control(block_group);
4883
4884         if (!caching_ctl) {
4885                 BUG_ON(!block_group_cache_done(block_group));
4886                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4887                 BUG_ON(ret);
4888         } else {
4889                 mutex_lock(&caching_ctl->mutex);
4890
4891                 if (start >= caching_ctl->progress) {
4892                         ret = add_excluded_extent(root, start, num_bytes);
4893                         BUG_ON(ret);
4894                 } else if (start + num_bytes <= caching_ctl->progress) {
4895                         ret = btrfs_remove_free_space(block_group,
4896                                                       start, num_bytes);
4897                         BUG_ON(ret);
4898                 } else {
4899                         num_bytes = caching_ctl->progress - start;
4900                         ret = btrfs_remove_free_space(block_group,
4901                                                       start, num_bytes);
4902                         BUG_ON(ret);
4903
4904                         start = caching_ctl->progress;
4905                         num_bytes = ins->objectid + ins->offset -
4906                                     caching_ctl->progress;
4907                         ret = add_excluded_extent(root, start, num_bytes);
4908                         BUG_ON(ret);
4909                 }
4910
4911                 mutex_unlock(&caching_ctl->mutex);
4912                 put_caching_control(caching_ctl);
4913         }
4914
4915         update_reserved_extents(block_group, ins->offset, 1);
4916         btrfs_put_block_group(block_group);
4917         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4918                                          0, owner, offset, ins, 1);
4919         return ret;
4920 }
4921
4922 /*
4923  * finds a free extent and does all the dirty work required for allocation
4924  * returns the key for the extent through ins, and a tree buffer for
4925  * the first block of the extent through buf.
4926  *
4927  * returns 0 if everything worked, non-zero otherwise.
4928  */
4929 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4930                             struct btrfs_root *root,
4931                             u64 num_bytes, u64 parent, u64 root_objectid,
4932                             struct btrfs_disk_key *key, int level,
4933                             u64 empty_size, u64 hint_byte, u64 search_end,
4934                             struct btrfs_key *ins)
4935 {
4936         int ret;
4937         u64 flags = 0;
4938
4939         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4940                                    empty_size, hint_byte, search_end,
4941                                    ins, 0);
4942         if (ret)
4943                 return ret;
4944
4945         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4946                 if (parent == 0)
4947                         parent = ins->objectid;
4948                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4949         } else
4950                 BUG_ON(parent > 0);
4951
4952         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4953                 struct btrfs_delayed_extent_op *extent_op;
4954                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4955                 BUG_ON(!extent_op);
4956                 if (key)
4957                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4958                 else
4959                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4960                 extent_op->flags_to_set = flags;
4961                 extent_op->update_key = 1;
4962                 extent_op->update_flags = 1;
4963                 extent_op->is_data = 0;
4964
4965                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4966                                         ins->offset, parent, root_objectid,
4967                                         level, BTRFS_ADD_DELAYED_EXTENT,
4968                                         extent_op);
4969                 BUG_ON(ret);
4970         }
4971
4972         if (root_objectid == root->root_key.objectid) {
4973                 u64 used;
4974                 spin_lock(&root->node_lock);
4975                 used = btrfs_root_used(&root->root_item) + num_bytes;
4976                 btrfs_set_root_used(&root->root_item, used);
4977                 spin_unlock(&root->node_lock);
4978         }
4979         return ret;
4980 }
4981
4982 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4983                                             struct btrfs_root *root,
4984                                             u64 bytenr, u32 blocksize,
4985                                             int level)
4986 {
4987         struct extent_buffer *buf;
4988
4989         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4990         if (!buf)
4991                 return ERR_PTR(-ENOMEM);
4992         btrfs_set_header_generation(buf, trans->transid);
4993         btrfs_set_buffer_lockdep_class(buf, level);
4994         btrfs_tree_lock(buf);
4995         clean_tree_block(trans, root, buf);
4996
4997         btrfs_set_lock_blocking(buf);
4998         btrfs_set_buffer_uptodate(buf);
4999
5000         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5001                 /*
5002                  * we allow two log transactions at a time, use different
5003                  * EXENT bit to differentiate dirty pages.
5004                  */
5005                 if (root->log_transid % 2 == 0)
5006                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5007                                         buf->start + buf->len - 1, GFP_NOFS);
5008                 else
5009                         set_extent_new(&root->dirty_log_pages, buf->start,
5010                                         buf->start + buf->len - 1, GFP_NOFS);
5011         } else {
5012                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5013                          buf->start + buf->len - 1, GFP_NOFS);
5014         }
5015         trans->blocks_used++;
5016         /* this returns a buffer locked for blocking */
5017         return buf;
5018 }
5019
5020 /*
5021  * helper function to allocate a block for a given tree
5022  * returns the tree buffer or NULL.
5023  */
5024 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5025                                         struct btrfs_root *root, u32 blocksize,
5026                                         u64 parent, u64 root_objectid,
5027                                         struct btrfs_disk_key *key, int level,
5028                                         u64 hint, u64 empty_size)
5029 {
5030         struct btrfs_key ins;
5031         int ret;
5032         struct extent_buffer *buf;
5033
5034         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
5035                                key, level, empty_size, hint, (u64)-1, &ins);
5036         if (ret) {
5037                 BUG_ON(ret > 0);
5038                 return ERR_PTR(ret);
5039         }
5040
5041         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5042                                     blocksize, level);
5043         return buf;
5044 }
5045
5046 struct walk_control {
5047         u64 refs[BTRFS_MAX_LEVEL];
5048         u64 flags[BTRFS_MAX_LEVEL];
5049         struct btrfs_key update_progress;
5050         int stage;
5051         int level;
5052         int shared_level;
5053         int update_ref;
5054         int keep_locks;
5055         int reada_slot;
5056         int reada_count;
5057 };
5058
5059 #define DROP_REFERENCE  1
5060 #define UPDATE_BACKREF  2
5061
5062 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5063                                      struct btrfs_root *root,
5064                                      struct walk_control *wc,
5065                                      struct btrfs_path *path)
5066 {
5067         u64 bytenr;
5068         u64 generation;
5069         u64 refs;
5070         u64 flags;
5071         u64 last = 0;
5072         u32 nritems;
5073         u32 blocksize;
5074         struct btrfs_key key;
5075         struct extent_buffer *eb;
5076         int ret;
5077         int slot;
5078         int nread = 0;
5079
5080         if (path->slots[wc->level] < wc->reada_slot) {
5081                 wc->reada_count = wc->reada_count * 2 / 3;
5082                 wc->reada_count = max(wc->reada_count, 2);
5083         } else {
5084                 wc->reada_count = wc->reada_count * 3 / 2;
5085                 wc->reada_count = min_t(int, wc->reada_count,
5086                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5087         }
5088
5089         eb = path->nodes[wc->level];
5090         nritems = btrfs_header_nritems(eb);
5091         blocksize = btrfs_level_size(root, wc->level - 1);
5092
5093         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5094                 if (nread >= wc->reada_count)
5095                         break;
5096
5097                 cond_resched();
5098                 bytenr = btrfs_node_blockptr(eb, slot);
5099                 generation = btrfs_node_ptr_generation(eb, slot);
5100
5101                 if (slot == path->slots[wc->level])
5102                         goto reada;
5103
5104                 if (wc->stage == UPDATE_BACKREF &&
5105                     generation <= root->root_key.offset)
5106                         continue;
5107
5108                 /* We don't lock the tree block, it's OK to be racy here */
5109                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5110                                                &refs, &flags);
5111                 BUG_ON(ret);
5112                 BUG_ON(refs == 0);
5113
5114                 if (wc->stage == DROP_REFERENCE) {
5115                         if (refs == 1)
5116                                 goto reada;
5117
5118                         if (wc->level == 1 &&
5119                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5120                                 continue;
5121                         if (!wc->update_ref ||
5122                             generation <= root->root_key.offset)
5123                                 continue;
5124                         btrfs_node_key_to_cpu(eb, &key, slot);
5125                         ret = btrfs_comp_cpu_keys(&key,
5126                                                   &wc->update_progress);
5127                         if (ret < 0)
5128                                 continue;
5129                 } else {
5130                         if (wc->level == 1 &&
5131                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5132                                 continue;
5133                 }
5134 reada:
5135                 ret = readahead_tree_block(root, bytenr, blocksize,
5136                                            generation);
5137                 if (ret)
5138                         break;
5139                 last = bytenr + blocksize;
5140                 nread++;
5141         }
5142         wc->reada_slot = slot;
5143 }
5144
5145 /*
5146  * hepler to process tree block while walking down the tree.
5147  *
5148  * when wc->stage == UPDATE_BACKREF, this function updates
5149  * back refs for pointers in the block.
5150  *
5151  * NOTE: return value 1 means we should stop walking down.
5152  */
5153 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5154                                    struct btrfs_root *root,
5155                                    struct btrfs_path *path,
5156                                    struct walk_control *wc, int lookup_info)
5157 {
5158         int level = wc->level;
5159         struct extent_buffer *eb = path->nodes[level];
5160         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5161         int ret;
5162
5163         if (wc->stage == UPDATE_BACKREF &&
5164             btrfs_header_owner(eb) != root->root_key.objectid)
5165                 return 1;
5166
5167         /*
5168          * when reference count of tree block is 1, it won't increase
5169          * again. once full backref flag is set, we never clear it.
5170          */
5171         if (lookup_info &&
5172             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5173              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5174                 BUG_ON(!path->locks[level]);
5175                 ret = btrfs_lookup_extent_info(trans, root,
5176                                                eb->start, eb->len,
5177                                                &wc->refs[level],
5178                                                &wc->flags[level]);
5179                 BUG_ON(ret);
5180                 BUG_ON(wc->refs[level] == 0);
5181         }
5182
5183         if (wc->stage == DROP_REFERENCE) {
5184                 if (wc->refs[level] > 1)
5185                         return 1;
5186
5187                 if (path->locks[level] && !wc->keep_locks) {
5188                         btrfs_tree_unlock(eb);
5189                         path->locks[level] = 0;
5190                 }
5191                 return 0;
5192         }
5193
5194         /* wc->stage == UPDATE_BACKREF */
5195         if (!(wc->flags[level] & flag)) {
5196                 BUG_ON(!path->locks[level]);
5197                 ret = btrfs_inc_ref(trans, root, eb, 1);
5198                 BUG_ON(ret);
5199                 ret = btrfs_dec_ref(trans, root, eb, 0);
5200                 BUG_ON(ret);
5201                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5202                                                   eb->len, flag, 0);
5203                 BUG_ON(ret);
5204                 wc->flags[level] |= flag;
5205         }
5206
5207         /*
5208          * the block is shared by multiple trees, so it's not good to
5209          * keep the tree lock
5210          */
5211         if (path->locks[level] && level > 0) {
5212                 btrfs_tree_unlock(eb);
5213                 path->locks[level] = 0;
5214         }
5215         return 0;
5216 }
5217
5218 /*
5219  * hepler to process tree block pointer.
5220  *
5221  * when wc->stage == DROP_REFERENCE, this function checks
5222  * reference count of the block pointed to. if the block
5223  * is shared and we need update back refs for the subtree
5224  * rooted at the block, this function changes wc->stage to
5225  * UPDATE_BACKREF. if the block is shared and there is no
5226  * need to update back, this function drops the reference
5227  * to the block.
5228  *
5229  * NOTE: return value 1 means we should stop walking down.
5230  */
5231 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5232                                  struct btrfs_root *root,
5233                                  struct btrfs_path *path,
5234                                  struct walk_control *wc, int *lookup_info)
5235 {
5236         u64 bytenr;
5237         u64 generation;
5238         u64 parent;
5239         u32 blocksize;
5240         struct btrfs_key key;
5241         struct extent_buffer *next;
5242         int level = wc->level;
5243         int reada = 0;
5244         int ret = 0;
5245
5246         generation = btrfs_node_ptr_generation(path->nodes[level],
5247                                                path->slots[level]);
5248         /*
5249          * if the lower level block was created before the snapshot
5250          * was created, we know there is no need to update back refs
5251          * for the subtree
5252          */
5253         if (wc->stage == UPDATE_BACKREF &&
5254             generation <= root->root_key.offset) {
5255                 *lookup_info = 1;
5256                 return 1;
5257         }
5258
5259         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5260         blocksize = btrfs_level_size(root, level - 1);
5261
5262         next = btrfs_find_tree_block(root, bytenr, blocksize);
5263         if (!next) {
5264                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5265                 if (!next)
5266                         return -ENOMEM;
5267                 reada = 1;
5268         }
5269         btrfs_tree_lock(next);
5270         btrfs_set_lock_blocking(next);
5271
5272         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5273                                        &wc->refs[level - 1],
5274                                        &wc->flags[level - 1]);
5275         BUG_ON(ret);
5276         BUG_ON(wc->refs[level - 1] == 0);
5277         *lookup_info = 0;
5278
5279         if (wc->stage == DROP_REFERENCE) {
5280                 if (wc->refs[level - 1] > 1) {
5281                         if (level == 1 &&
5282                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5283                                 goto skip;
5284
5285                         if (!wc->update_ref ||
5286                             generation <= root->root_key.offset)
5287                                 goto skip;
5288
5289                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5290                                               path->slots[level]);
5291                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5292                         if (ret < 0)
5293                                 goto skip;
5294
5295                         wc->stage = UPDATE_BACKREF;
5296                         wc->shared_level = level - 1;
5297                 }
5298         } else {
5299                 if (level == 1 &&
5300                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5301                         goto skip;
5302         }
5303
5304         if (!btrfs_buffer_uptodate(next, generation)) {
5305                 btrfs_tree_unlock(next);
5306                 free_extent_buffer(next);
5307                 next = NULL;
5308                 *lookup_info = 1;
5309         }
5310
5311         if (!next) {
5312                 if (reada && level == 1)
5313                         reada_walk_down(trans, root, wc, path);
5314                 next = read_tree_block(root, bytenr, blocksize, generation);
5315                 btrfs_tree_lock(next);
5316                 btrfs_set_lock_blocking(next);
5317         }
5318
5319         level--;
5320         BUG_ON(level != btrfs_header_level(next));
5321         path->nodes[level] = next;
5322         path->slots[level] = 0;
5323         path->locks[level] = 1;
5324         wc->level = level;
5325         if (wc->level == 1)
5326                 wc->reada_slot = 0;
5327         return 0;
5328 skip:
5329         wc->refs[level - 1] = 0;
5330         wc->flags[level - 1] = 0;
5331         if (wc->stage == DROP_REFERENCE) {
5332                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5333                         parent = path->nodes[level]->start;
5334                 } else {
5335                         BUG_ON(root->root_key.objectid !=
5336                                btrfs_header_owner(path->nodes[level]));
5337                         parent = 0;
5338                 }
5339
5340                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5341                                         root->root_key.objectid, level - 1, 0);
5342                 BUG_ON(ret);
5343         }
5344         btrfs_tree_unlock(next);
5345         free_extent_buffer(next);
5346         *lookup_info = 1;
5347         return 1;
5348 }
5349
5350 /*
5351  * hepler to process tree block while walking up the tree.
5352  *
5353  * when wc->stage == DROP_REFERENCE, this function drops
5354  * reference count on the block.
5355  *
5356  * when wc->stage == UPDATE_BACKREF, this function changes
5357  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5358  * to UPDATE_BACKREF previously while processing the block.
5359  *
5360  * NOTE: return value 1 means we should stop walking up.
5361  */
5362 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5363                                  struct btrfs_root *root,
5364                                  struct btrfs_path *path,
5365                                  struct walk_control *wc)
5366 {
5367         int ret = 0;
5368         int level = wc->level;
5369         struct extent_buffer *eb = path->nodes[level];
5370         u64 parent = 0;
5371
5372         if (wc->stage == UPDATE_BACKREF) {
5373                 BUG_ON(wc->shared_level < level);
5374                 if (level < wc->shared_level)
5375                         goto out;
5376
5377                 ret = find_next_key(path, level + 1, &wc->update_progress);
5378                 if (ret > 0)
5379                         wc->update_ref = 0;
5380
5381                 wc->stage = DROP_REFERENCE;
5382                 wc->shared_level = -1;
5383                 path->slots[level] = 0;
5384
5385                 /*
5386                  * check reference count again if the block isn't locked.
5387                  * we should start walking down the tree again if reference
5388                  * count is one.
5389                  */
5390                 if (!path->locks[level]) {
5391                         BUG_ON(level == 0);
5392                         btrfs_tree_lock(eb);
5393                         btrfs_set_lock_blocking(eb);
5394                         path->locks[level] = 1;
5395
5396                         ret = btrfs_lookup_extent_info(trans, root,
5397                                                        eb->start, eb->len,
5398                                                        &wc->refs[level],
5399                                                        &wc->flags[level]);
5400                         BUG_ON(ret);
5401                         BUG_ON(wc->refs[level] == 0);
5402                         if (wc->refs[level] == 1) {
5403                                 btrfs_tree_unlock(eb);
5404                                 path->locks[level] = 0;
5405                                 return 1;
5406                         }
5407                 }
5408         }
5409
5410         /* wc->stage == DROP_REFERENCE */
5411         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5412
5413         if (wc->refs[level] == 1) {
5414                 if (level == 0) {
5415                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5416                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5417                         else
5418                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5419                         BUG_ON(ret);
5420                 }
5421                 /* make block locked assertion in clean_tree_block happy */
5422                 if (!path->locks[level] &&
5423                     btrfs_header_generation(eb) == trans->transid) {
5424                         btrfs_tree_lock(eb);
5425                         btrfs_set_lock_blocking(eb);
5426                         path->locks[level] = 1;
5427                 }
5428                 clean_tree_block(trans, root, eb);
5429         }
5430
5431         if (eb == root->node) {
5432                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5433                         parent = eb->start;
5434                 else
5435                         BUG_ON(root->root_key.objectid !=
5436                                btrfs_header_owner(eb));
5437         } else {
5438                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5439                         parent = path->nodes[level + 1]->start;
5440                 else
5441                         BUG_ON(root->root_key.objectid !=
5442                                btrfs_header_owner(path->nodes[level + 1]));
5443         }
5444
5445         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5446                                 root->root_key.objectid, level, 0);
5447         BUG_ON(ret);
5448 out:
5449         wc->refs[level] = 0;
5450         wc->flags[level] = 0;
5451         return ret;
5452 }
5453
5454 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5455                                    struct btrfs_root *root,
5456                                    struct btrfs_path *path,
5457                                    struct walk_control *wc)
5458 {
5459         int level = wc->level;
5460         int lookup_info = 1;
5461         int ret;
5462
5463         while (level >= 0) {
5464                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5465                 if (ret > 0)
5466                         break;
5467
5468                 if (level == 0)
5469                         break;
5470
5471                 if (path->slots[level] >=
5472                     btrfs_header_nritems(path->nodes[level]))
5473                         break;
5474
5475                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5476                 if (ret > 0) {
5477                         path->slots[level]++;
5478                         continue;
5479                 } else if (ret < 0)
5480                         return ret;
5481                 level = wc->level;
5482         }
5483         return 0;
5484 }
5485
5486 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5487                                  struct btrfs_root *root,
5488                                  struct btrfs_path *path,
5489                                  struct walk_control *wc, int max_level)
5490 {
5491         int level = wc->level;
5492         int ret;
5493
5494         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5495         while (level < max_level && path->nodes[level]) {
5496                 wc->level = level;
5497                 if (path->slots[level] + 1 <
5498                     btrfs_header_nritems(path->nodes[level])) {
5499                         path->slots[level]++;
5500                         return 0;
5501                 } else {
5502                         ret = walk_up_proc(trans, root, path, wc);
5503                         if (ret > 0)
5504                                 return 0;
5505
5506                         if (path->locks[level]) {
5507                                 btrfs_tree_unlock(path->nodes[level]);
5508                                 path->locks[level] = 0;
5509                         }
5510                         free_extent_buffer(path->nodes[level]);
5511                         path->nodes[level] = NULL;
5512                         level++;
5513                 }
5514         }
5515         return 1;
5516 }
5517
5518 /*
5519  * drop a subvolume tree.
5520  *
5521  * this function traverses the tree freeing any blocks that only
5522  * referenced by the tree.
5523  *
5524  * when a shared tree block is found. this function decreases its
5525  * reference count by one. if update_ref is true, this function
5526  * also make sure backrefs for the shared block and all lower level
5527  * blocks are properly updated.
5528  */
5529 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5530 {
5531         struct btrfs_path *path;
5532         struct btrfs_trans_handle *trans;
5533         struct btrfs_root *tree_root = root->fs_info->tree_root;
5534         struct btrfs_root_item *root_item = &root->root_item;
5535         struct walk_control *wc;
5536         struct btrfs_key key;
5537         int err = 0;
5538         int ret;
5539         int level;
5540
5541         path = btrfs_alloc_path();
5542         BUG_ON(!path);
5543
5544         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5545         BUG_ON(!wc);
5546
5547         trans = btrfs_start_transaction(tree_root, 1);
5548
5549         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5550                 level = btrfs_header_level(root->node);
5551                 path->nodes[level] = btrfs_lock_root_node(root);
5552                 btrfs_set_lock_blocking(path->nodes[level]);
5553                 path->slots[level] = 0;
5554                 path->locks[level] = 1;
5555                 memset(&wc->update_progress, 0,
5556                        sizeof(wc->update_progress));
5557         } else {
5558                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5559                 memcpy(&wc->update_progress, &key,
5560                        sizeof(wc->update_progress));
5561
5562                 level = root_item->drop_level;
5563                 BUG_ON(level == 0);
5564                 path->lowest_level = level;
5565                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5566                 path->lowest_level = 0;
5567                 if (ret < 0) {
5568                         err = ret;
5569                         goto out;
5570                 }
5571                 WARN_ON(ret > 0);
5572
5573                 /*
5574                  * unlock our path, this is safe because only this
5575                  * function is allowed to delete this snapshot
5576                  */
5577                 btrfs_unlock_up_safe(path, 0);
5578
5579                 level = btrfs_header_level(root->node);
5580                 while (1) {
5581                         btrfs_tree_lock(path->nodes[level]);
5582                         btrfs_set_lock_blocking(path->nodes[level]);
5583
5584                         ret = btrfs_lookup_extent_info(trans, root,
5585                                                 path->nodes[level]->start,
5586                                                 path->nodes[level]->len,
5587                                                 &wc->refs[level],
5588                                                 &wc->flags[level]);
5589                         BUG_ON(ret);
5590                         BUG_ON(wc->refs[level] == 0);
5591
5592                         if (level == root_item->drop_level)
5593                                 break;
5594
5595                         btrfs_tree_unlock(path->nodes[level]);
5596                         WARN_ON(wc->refs[level] != 1);
5597                         level--;
5598                 }
5599         }
5600
5601         wc->level = level;
5602         wc->shared_level = -1;
5603         wc->stage = DROP_REFERENCE;
5604         wc->update_ref = update_ref;
5605         wc->keep_locks = 0;
5606         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5607
5608         while (1) {
5609                 ret = walk_down_tree(trans, root, path, wc);
5610                 if (ret < 0) {
5611                         err = ret;
5612                         break;
5613                 }
5614
5615                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5616                 if (ret < 0) {
5617                         err = ret;
5618                         break;
5619                 }
5620
5621                 if (ret > 0) {
5622                         BUG_ON(wc->stage != DROP_REFERENCE);
5623                         break;
5624                 }
5625
5626                 if (wc->stage == DROP_REFERENCE) {
5627                         level = wc->level;
5628                         btrfs_node_key(path->nodes[level],
5629                                        &root_item->drop_progress,
5630                                        path->slots[level]);
5631                         root_item->drop_level = level;
5632                 }
5633
5634                 BUG_ON(wc->level == 0);
5635                 if (trans->transaction->in_commit ||
5636                     trans->transaction->delayed_refs.flushing) {
5637                         ret = btrfs_update_root(trans, tree_root,
5638                                                 &root->root_key,
5639                                                 root_item);
5640                         BUG_ON(ret);
5641
5642                         btrfs_end_transaction(trans, tree_root);
5643                         trans = btrfs_start_transaction(tree_root, 1);
5644                 } else {
5645                         unsigned long update;
5646                         update = trans->delayed_ref_updates;
5647                         trans->delayed_ref_updates = 0;
5648                         if (update)
5649                                 btrfs_run_delayed_refs(trans, tree_root,
5650                                                        update);
5651                 }
5652         }
5653         btrfs_release_path(root, path);
5654         BUG_ON(err);
5655
5656         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5657         BUG_ON(ret);
5658
5659         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5660                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5661                                            NULL, NULL);
5662                 BUG_ON(ret < 0);
5663                 if (ret > 0) {
5664                         ret = btrfs_del_orphan_item(trans, tree_root,
5665                                                     root->root_key.objectid);
5666                         BUG_ON(ret);
5667                 }
5668         }
5669
5670         if (root->in_radix) {
5671                 btrfs_free_fs_root(tree_root->fs_info, root);
5672         } else {
5673                 free_extent_buffer(root->node);
5674                 free_extent_buffer(root->commit_root);
5675                 kfree(root);
5676         }
5677 out:
5678         btrfs_end_transaction(trans, tree_root);
5679         kfree(wc);
5680         btrfs_free_path(path);
5681         return err;
5682 }
5683
5684 /*
5685  * drop subtree rooted at tree block 'node'.
5686  *
5687  * NOTE: this function will unlock and release tree block 'node'
5688  */
5689 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5690                         struct btrfs_root *root,
5691                         struct extent_buffer *node,
5692                         struct extent_buffer *parent)
5693 {
5694         struct btrfs_path *path;
5695         struct walk_control *wc;
5696         int level;
5697         int parent_level;
5698         int ret = 0;
5699         int wret;
5700
5701         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5702
5703         path = btrfs_alloc_path();
5704         BUG_ON(!path);
5705
5706         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5707         BUG_ON(!wc);
5708
5709         btrfs_assert_tree_locked(parent);
5710         parent_level = btrfs_header_level(parent);
5711         extent_buffer_get(parent);
5712         path->nodes[parent_level] = parent;
5713         path->slots[parent_level] = btrfs_header_nritems(parent);
5714
5715         btrfs_assert_tree_locked(node);
5716         level = btrfs_header_level(node);
5717         path->nodes[level] = node;
5718         path->slots[level] = 0;
5719         path->locks[level] = 1;
5720
5721         wc->refs[parent_level] = 1;
5722         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5723         wc->level = level;
5724         wc->shared_level = -1;
5725         wc->stage = DROP_REFERENCE;
5726         wc->update_ref = 0;
5727         wc->keep_locks = 1;
5728         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5729
5730         while (1) {
5731                 wret = walk_down_tree(trans, root, path, wc);
5732                 if (wret < 0) {
5733                         ret = wret;
5734                         break;
5735                 }
5736
5737                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5738                 if (wret < 0)
5739                         ret = wret;
5740                 if (wret != 0)
5741                         break;
5742         }
5743
5744         kfree(wc);
5745         btrfs_free_path(path);
5746         return ret;
5747 }
5748
5749 #if 0
5750 static unsigned long calc_ra(unsigned long start, unsigned long last,
5751                              unsigned long nr)
5752 {
5753         return min(last, start + nr - 1);
5754 }
5755
5756 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5757                                          u64 len)
5758 {
5759         u64 page_start;
5760         u64 page_end;
5761         unsigned long first_index;
5762         unsigned long last_index;
5763         unsigned long i;
5764         struct page *page;
5765         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5766         struct file_ra_state *ra;
5767         struct btrfs_ordered_extent *ordered;
5768         unsigned int total_read = 0;
5769         unsigned int total_dirty = 0;
5770         int ret = 0;
5771
5772         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5773
5774         mutex_lock(&inode->i_mutex);
5775         first_index = start >> PAGE_CACHE_SHIFT;
5776         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5777
5778         /* make sure the dirty trick played by the caller work */
5779         ret = invalidate_inode_pages2_range(inode->i_mapping,
5780                                             first_index, last_index);
5781         if (ret)
5782                 goto out_unlock;
5783
5784         file_ra_state_init(ra, inode->i_mapping);
5785
5786         for (i = first_index ; i <= last_index; i++) {
5787                 if (total_read % ra->ra_pages == 0) {
5788                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5789                                        calc_ra(i, last_index, ra->ra_pages));
5790                 }
5791                 total_read++;
5792 again:
5793                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5794                         BUG_ON(1);
5795                 page = grab_cache_page(inode->i_mapping, i);
5796                 if (!page) {
5797                         ret = -ENOMEM;
5798                         goto out_unlock;
5799                 }
5800                 if (!PageUptodate(page)) {
5801                         btrfs_readpage(NULL, page);
5802                         lock_page(page);
5803                         if (!PageUptodate(page)) {
5804                                 unlock_page(page);
5805                                 page_cache_release(page);
5806                                 ret = -EIO;
5807                                 goto out_unlock;
5808                         }
5809                 }
5810                 wait_on_page_writeback(page);
5811
5812                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5813                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5814                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5815
5816                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5817                 if (ordered) {
5818                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5819                         unlock_page(page);
5820                         page_cache_release(page);
5821                         btrfs_start_ordered_extent(inode, ordered, 1);
5822                         btrfs_put_ordered_extent(ordered);
5823                         goto again;
5824                 }
5825                 set_page_extent_mapped(page);
5826
5827                 if (i == first_index)
5828                         set_extent_bits(io_tree, page_start, page_end,
5829                                         EXTENT_BOUNDARY, GFP_NOFS);
5830                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5831
5832                 set_page_dirty(page);
5833                 total_dirty++;
5834
5835                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5836                 unlock_page(page);
5837                 page_cache_release(page);
5838         }
5839
5840 out_unlock:
5841         kfree(ra);
5842         mutex_unlock(&inode->i_mutex);
5843         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5844         return ret;
5845 }
5846
5847 static noinline int relocate_data_extent(struct inode *reloc_inode,
5848                                          struct btrfs_key *extent_key,
5849                                          u64 offset)
5850 {
5851         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5852         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5853         struct extent_map *em;
5854         u64 start = extent_key->objectid - offset;
5855         u64 end = start + extent_key->offset - 1;
5856
5857         em = alloc_extent_map(GFP_NOFS);
5858         BUG_ON(!em || IS_ERR(em));
5859
5860         em->start = start;
5861         em->len = extent_key->offset;
5862         em->block_len = extent_key->offset;
5863         em->block_start = extent_key->objectid;
5864         em->bdev = root->fs_info->fs_devices->latest_bdev;
5865         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5866
5867         /* setup extent map to cheat btrfs_readpage */
5868         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5869         while (1) {
5870                 int ret;
5871                 write_lock(&em_tree->lock);
5872                 ret = add_extent_mapping(em_tree, em);
5873                 write_unlock(&em_tree->lock);
5874                 if (ret != -EEXIST) {
5875                         free_extent_map(em);
5876                         break;
5877                 }
5878                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5879         }
5880         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5881
5882         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5883 }
5884
5885 struct btrfs_ref_path {
5886         u64 extent_start;
5887         u64 nodes[BTRFS_MAX_LEVEL];
5888         u64 root_objectid;
5889         u64 root_generation;
5890         u64 owner_objectid;
5891         u32 num_refs;
5892         int lowest_level;
5893         int current_level;
5894         int shared_level;
5895
5896         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5897         u64 new_nodes[BTRFS_MAX_LEVEL];
5898 };
5899
5900 struct disk_extent {
5901         u64 ram_bytes;
5902         u64 disk_bytenr;
5903         u64 disk_num_bytes;
5904         u64 offset;
5905         u64 num_bytes;
5906         u8 compression;
5907         u8 encryption;
5908         u16 other_encoding;
5909 };
5910
5911 static int is_cowonly_root(u64 root_objectid)
5912 {
5913         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5914             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5915             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5916             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5917             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5918             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5919                 return 1;
5920         return 0;
5921 }
5922
5923 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5924                                     struct btrfs_root *extent_root,
5925                                     struct btrfs_ref_path *ref_path,
5926                                     int first_time)
5927 {
5928         struct extent_buffer *leaf;
5929         struct btrfs_path *path;
5930         struct btrfs_extent_ref *ref;
5931         struct btrfs_key key;
5932         struct btrfs_key found_key;
5933         u64 bytenr;
5934         u32 nritems;
5935         int level;
5936         int ret = 1;
5937
5938         path = btrfs_alloc_path();
5939         if (!path)
5940                 return -ENOMEM;
5941
5942         if (first_time) {
5943                 ref_path->lowest_level = -1;
5944                 ref_path->current_level = -1;
5945                 ref_path->shared_level = -1;
5946                 goto walk_up;
5947         }
5948 walk_down:
5949         level = ref_path->current_level - 1;
5950         while (level >= -1) {
5951                 u64 parent;
5952                 if (level < ref_path->lowest_level)
5953                         break;
5954
5955                 if (level >= 0)
5956                         bytenr = ref_path->nodes[level];
5957                 else
5958                         bytenr = ref_path->extent_start;
5959                 BUG_ON(bytenr == 0);
5960
5961                 parent = ref_path->nodes[level + 1];
5962                 ref_path->nodes[level + 1] = 0;
5963                 ref_path->current_level = level;
5964                 BUG_ON(parent == 0);
5965
5966                 key.objectid = bytenr;
5967                 key.offset = parent + 1;
5968                 key.type = BTRFS_EXTENT_REF_KEY;
5969
5970                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5971                 if (ret < 0)
5972                         goto out;
5973                 BUG_ON(ret == 0);
5974
5975                 leaf = path->nodes[0];
5976                 nritems = btrfs_header_nritems(leaf);
5977                 if (path->slots[0] >= nritems) {
5978                         ret = btrfs_next_leaf(extent_root, path);
5979                         if (ret < 0)
5980                                 goto out;
5981                         if (ret > 0)
5982                                 goto next;
5983                         leaf = path->nodes[0];
5984                 }
5985
5986                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5987                 if (found_key.objectid == bytenr &&
5988                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5989                         if (level < ref_path->shared_level)
5990                                 ref_path->shared_level = level;
5991                         goto found;
5992                 }
5993 next:
5994                 level--;
5995                 btrfs_release_path(extent_root, path);
5996                 cond_resched();
5997         }
5998         /* reached lowest level */
5999         ret = 1;
6000         goto out;
6001 walk_up:
6002         level = ref_path->current_level;
6003         while (level < BTRFS_MAX_LEVEL - 1) {
6004                 u64 ref_objectid;
6005
6006                 if (level >= 0)
6007                         bytenr = ref_path->nodes[level];
6008                 else
6009                         bytenr = ref_path->extent_start;
6010
6011                 BUG_ON(bytenr == 0);
6012
6013                 key.objectid = bytenr;
6014                 key.offset = 0;
6015                 key.type = BTRFS_EXTENT_REF_KEY;
6016
6017                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6018                 if (ret < 0)
6019                         goto out;
6020
6021                 leaf = path->nodes[0];
6022                 nritems = btrfs_header_nritems(leaf);
6023                 if (path->slots[0] >= nritems) {
6024                         ret = btrfs_next_leaf(extent_root, path);
6025                         if (ret < 0)
6026                                 goto out;
6027                         if (ret > 0) {
6028                                 /* the extent was freed by someone */
6029                                 if (ref_path->lowest_level == level)
6030                                         goto out;
6031                                 btrfs_release_path(extent_root, path);
6032                                 goto walk_down;
6033                         }
6034                         leaf = path->nodes[0];
6035                 }
6036
6037                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6038                 if (found_key.objectid != bytenr ||
6039                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6040                         /* the extent was freed by someone */
6041                         if (ref_path->lowest_level == level) {
6042                                 ret = 1;
6043                                 goto out;
6044                         }
6045                         btrfs_release_path(extent_root, path);
6046                         goto walk_down;
6047                 }
6048 found:
6049                 ref = btrfs_item_ptr(leaf, path->slots[0],
6050                                 struct btrfs_extent_ref);
6051                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6052                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6053                         if (first_time) {
6054                                 level = (int)ref_objectid;
6055                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6056                                 ref_path->lowest_level = level;
6057                                 ref_path->current_level = level;
6058                                 ref_path->nodes[level] = bytenr;
6059                         } else {
6060                                 WARN_ON(ref_objectid != level);
6061                         }
6062                 } else {
6063                         WARN_ON(level != -1);
6064                 }
6065                 first_time = 0;
6066
6067                 if (ref_path->lowest_level == level) {
6068                         ref_path->owner_objectid = ref_objectid;
6069                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6070                 }
6071
6072                 /*
6073                  * the block is tree root or the block isn't in reference
6074                  * counted tree.
6075                  */
6076                 if (found_key.objectid == found_key.offset ||
6077                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6078                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6079                         ref_path->root_generation =
6080                                 btrfs_ref_generation(leaf, ref);
6081                         if (level < 0) {
6082                                 /* special reference from the tree log */
6083                                 ref_path->nodes[0] = found_key.offset;
6084                                 ref_path->current_level = 0;
6085                         }
6086                         ret = 0;
6087                         goto out;
6088                 }
6089
6090                 level++;
6091                 BUG_ON(ref_path->nodes[level] != 0);
6092                 ref_path->nodes[level] = found_key.offset;
6093                 ref_path->current_level = level;
6094
6095                 /*
6096                  * the reference was created in the running transaction,
6097                  * no need to continue walking up.
6098                  */
6099                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6100                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6101                         ref_path->root_generation =
6102                                 btrfs_ref_generation(leaf, ref);
6103                         ret = 0;
6104                         goto out;
6105                 }
6106
6107                 btrfs_release_path(extent_root, path);
6108                 cond_resched();
6109         }
6110         /* reached max tree level, but no tree root found. */
6111         BUG();
6112 out:
6113         btrfs_free_path(path);
6114         return ret;
6115 }
6116
6117 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6118                                 struct btrfs_root *extent_root,
6119                                 struct btrfs_ref_path *ref_path,
6120                                 u64 extent_start)
6121 {
6122         memset(ref_path, 0, sizeof(*ref_path));
6123         ref_path->extent_start = extent_start;
6124
6125         return __next_ref_path(trans, extent_root, ref_path, 1);
6126 }
6127
6128 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6129                                struct btrfs_root *extent_root,
6130                                struct btrfs_ref_path *ref_path)
6131 {
6132         return __next_ref_path(trans, extent_root, ref_path, 0);
6133 }
6134
6135 static noinline int get_new_locations(struct inode *reloc_inode,
6136                                       struct btrfs_key *extent_key,
6137                                       u64 offset, int no_fragment,
6138                                       struct disk_extent **extents,
6139                                       int *nr_extents)
6140 {
6141         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6142         struct btrfs_path *path;
6143         struct btrfs_file_extent_item *fi;
6144         struct extent_buffer *leaf;
6145         struct disk_extent *exts = *extents;
6146         struct btrfs_key found_key;
6147         u64 cur_pos;
6148         u64 last_byte;
6149         u32 nritems;
6150         int nr = 0;
6151         int max = *nr_extents;
6152         int ret;
6153
6154         WARN_ON(!no_fragment && *extents);
6155         if (!exts) {
6156                 max = 1;
6157                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6158                 if (!exts)
6159                         return -ENOMEM;
6160         }
6161
6162         path = btrfs_alloc_path();
6163         BUG_ON(!path);
6164
6165         cur_pos = extent_key->objectid - offset;
6166         last_byte = extent_key->objectid + extent_key->offset;
6167         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6168                                        cur_pos, 0);
6169         if (ret < 0)
6170                 goto out;
6171         if (ret > 0) {
6172                 ret = -ENOENT;
6173                 goto out;
6174         }
6175
6176         while (1) {
6177                 leaf = path->nodes[0];
6178                 nritems = btrfs_header_nritems(leaf);
6179                 if (path->slots[0] >= nritems) {
6180                         ret = btrfs_next_leaf(root, path);
6181                         if (ret < 0)
6182                                 goto out;
6183                         if (ret > 0)
6184                                 break;
6185                         leaf = path->nodes[0];
6186                 }
6187
6188                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6189                 if (found_key.offset != cur_pos ||
6190                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6191                     found_key.objectid != reloc_inode->i_ino)
6192                         break;
6193
6194                 fi = btrfs_item_ptr(leaf, path->slots[0],
6195                                     struct btrfs_file_extent_item);
6196                 if (btrfs_file_extent_type(leaf, fi) !=
6197                     BTRFS_FILE_EXTENT_REG ||
6198                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6199                         break;
6200
6201                 if (nr == max) {
6202                         struct disk_extent *old = exts;
6203                         max *= 2;
6204                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6205                         memcpy(exts, old, sizeof(*exts) * nr);
6206                         if (old != *extents)
6207                                 kfree(old);
6208                 }
6209
6210                 exts[nr].disk_bytenr =
6211                         btrfs_file_extent_disk_bytenr(leaf, fi);
6212                 exts[nr].disk_num_bytes =
6213                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6214                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6215                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6216                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6217                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6218                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6219                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6220                                                                            fi);
6221                 BUG_ON(exts[nr].offset > 0);
6222                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6223                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6224
6225                 cur_pos += exts[nr].num_bytes;
6226                 nr++;
6227
6228                 if (cur_pos + offset >= last_byte)
6229                         break;
6230
6231                 if (no_fragment) {
6232                         ret = 1;
6233                         goto out;
6234                 }
6235                 path->slots[0]++;
6236         }
6237
6238         BUG_ON(cur_pos + offset > last_byte);
6239         if (cur_pos + offset < last_byte) {
6240                 ret = -ENOENT;
6241                 goto out;
6242         }
6243         ret = 0;
6244 out:
6245         btrfs_free_path(path);
6246         if (ret) {
6247                 if (exts != *extents)
6248                         kfree(exts);
6249         } else {
6250                 *extents = exts;
6251                 *nr_extents = nr;
6252         }
6253         return ret;
6254 }
6255
6256 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6257                                         struct btrfs_root *root,
6258                                         struct btrfs_path *path,
6259                                         struct btrfs_key *extent_key,
6260                                         struct btrfs_key *leaf_key,
6261                                         struct btrfs_ref_path *ref_path,
6262                                         struct disk_extent *new_extents,
6263                                         int nr_extents)
6264 {
6265         struct extent_buffer *leaf;
6266         struct btrfs_file_extent_item *fi;
6267         struct inode *inode = NULL;
6268         struct btrfs_key key;
6269         u64 lock_start = 0;
6270         u64 lock_end = 0;
6271         u64 num_bytes;
6272         u64 ext_offset;
6273         u64 search_end = (u64)-1;
6274         u32 nritems;
6275         int nr_scaned = 0;
6276         int extent_locked = 0;
6277         int extent_type;
6278         int ret;
6279
6280         memcpy(&key, leaf_key, sizeof(key));
6281         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6282                 if (key.objectid < ref_path->owner_objectid ||
6283                     (key.objectid == ref_path->owner_objectid &&
6284                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6285                         key.objectid = ref_path->owner_objectid;
6286                         key.type = BTRFS_EXTENT_DATA_KEY;
6287                         key.offset = 0;
6288                 }
6289         }
6290
6291         while (1) {
6292                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6293                 if (ret < 0)
6294                         goto out;
6295
6296                 leaf = path->nodes[0];
6297                 nritems = btrfs_header_nritems(leaf);
6298 next:
6299                 if (extent_locked && ret > 0) {
6300                         /*
6301                          * the file extent item was modified by someone
6302                          * before the extent got locked.
6303                          */
6304                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6305                                       lock_end, GFP_NOFS);
6306                         extent_locked = 0;
6307                 }
6308
6309                 if (path->slots[0] >= nritems) {
6310                         if (++nr_scaned > 2)
6311                                 break;
6312
6313                         BUG_ON(extent_locked);
6314                         ret = btrfs_next_leaf(root, path);
6315                         if (ret < 0)
6316                                 goto out;
6317                         if (ret > 0)
6318                                 break;
6319                         leaf = path->nodes[0];
6320                         nritems = btrfs_header_nritems(leaf);
6321                 }
6322
6323                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6324
6325                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6326                         if ((key.objectid > ref_path->owner_objectid) ||
6327                             (key.objectid == ref_path->owner_objectid &&
6328                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6329                             key.offset >= search_end)
6330                                 break;
6331                 }
6332
6333                 if (inode && key.objectid != inode->i_ino) {
6334                         BUG_ON(extent_locked);
6335                         btrfs_release_path(root, path);
6336                         mutex_unlock(&inode->i_mutex);
6337                         iput(inode);
6338                         inode = NULL;
6339                         continue;
6340                 }
6341
6342                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6343                         path->slots[0]++;
6344                         ret = 1;
6345                         goto next;
6346                 }
6347                 fi = btrfs_item_ptr(leaf, path->slots[0],
6348                                     struct btrfs_file_extent_item);
6349                 extent_type = btrfs_file_extent_type(leaf, fi);
6350                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6351                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6352                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6353                      extent_key->objectid)) {
6354                         path->slots[0]++;
6355                         ret = 1;
6356                         goto next;
6357                 }
6358
6359                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6360                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6361
6362                 if (search_end == (u64)-1) {
6363                         search_end = key.offset - ext_offset +
6364                                 btrfs_file_extent_ram_bytes(leaf, fi);
6365                 }
6366
6367                 if (!extent_locked) {
6368                         lock_start = key.offset;
6369                         lock_end = lock_start + num_bytes - 1;
6370                 } else {
6371                         if (lock_start > key.offset ||
6372                             lock_end + 1 < key.offset + num_bytes) {
6373                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6374                                               lock_start, lock_end, GFP_NOFS);
6375                                 extent_locked = 0;
6376                         }
6377                 }
6378
6379                 if (!inode) {
6380                         btrfs_release_path(root, path);
6381
6382                         inode = btrfs_iget_locked(root->fs_info->sb,
6383                                                   key.objectid, root);
6384                         if (inode->i_state & I_NEW) {
6385                                 BTRFS_I(inode)->root = root;
6386                                 BTRFS_I(inode)->location.objectid =
6387                                         key.objectid;
6388                                 BTRFS_I(inode)->location.type =
6389                                         BTRFS_INODE_ITEM_KEY;
6390                                 BTRFS_I(inode)->location.offset = 0;
6391                                 btrfs_read_locked_inode(inode);
6392                                 unlock_new_inode(inode);
6393                         }
6394                         /*
6395                          * some code call btrfs_commit_transaction while
6396                          * holding the i_mutex, so we can't use mutex_lock
6397                          * here.
6398                          */
6399                         if (is_bad_inode(inode) ||
6400                             !mutex_trylock(&inode->i_mutex)) {
6401                                 iput(inode);
6402                                 inode = NULL;
6403                                 key.offset = (u64)-1;
6404                                 goto skip;
6405                         }
6406                 }
6407
6408                 if (!extent_locked) {
6409                         struct btrfs_ordered_extent *ordered;
6410
6411                         btrfs_release_path(root, path);
6412
6413                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6414                                     lock_end, GFP_NOFS);
6415                         ordered = btrfs_lookup_first_ordered_extent(inode,
6416                                                                     lock_end);
6417                         if (ordered &&
6418                             ordered->file_offset <= lock_end &&
6419                             ordered->file_offset + ordered->len > lock_start) {
6420                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6421                                               lock_start, lock_end, GFP_NOFS);
6422                                 btrfs_start_ordered_extent(inode, ordered, 1);
6423                                 btrfs_put_ordered_extent(ordered);
6424                                 key.offset += num_bytes;
6425                                 goto skip;
6426                         }
6427                         if (ordered)
6428                                 btrfs_put_ordered_extent(ordered);
6429
6430                         extent_locked = 1;
6431                         continue;
6432                 }
6433
6434                 if (nr_extents == 1) {
6435                         /* update extent pointer in place */
6436                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6437                                                 new_extents[0].disk_bytenr);
6438                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6439                                                 new_extents[0].disk_num_bytes);
6440                         btrfs_mark_buffer_dirty(leaf);
6441
6442                         btrfs_drop_extent_cache(inode, key.offset,
6443                                                 key.offset + num_bytes - 1, 0);
6444
6445                         ret = btrfs_inc_extent_ref(trans, root,
6446                                                 new_extents[0].disk_bytenr,
6447                                                 new_extents[0].disk_num_bytes,
6448                                                 leaf->start,
6449                                                 root->root_key.objectid,
6450                                                 trans->transid,
6451                                                 key.objectid);
6452                         BUG_ON(ret);
6453
6454                         ret = btrfs_free_extent(trans, root,
6455                                                 extent_key->objectid,
6456                                                 extent_key->offset,
6457                                                 leaf->start,
6458                                                 btrfs_header_owner(leaf),
6459                                                 btrfs_header_generation(leaf),
6460                                                 key.objectid, 0);
6461                         BUG_ON(ret);
6462
6463                         btrfs_release_path(root, path);
6464                         key.offset += num_bytes;
6465                 } else {
6466                         BUG_ON(1);
6467 #if 0
6468                         u64 alloc_hint;
6469                         u64 extent_len;
6470                         int i;
6471                         /*
6472                          * drop old extent pointer at first, then insert the
6473                          * new pointers one bye one
6474                          */
6475                         btrfs_release_path(root, path);
6476                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6477                                                  key.offset + num_bytes,
6478                                                  key.offset, &alloc_hint);
6479                         BUG_ON(ret);
6480
6481                         for (i = 0; i < nr_extents; i++) {
6482                                 if (ext_offset >= new_extents[i].num_bytes) {
6483                                         ext_offset -= new_extents[i].num_bytes;
6484                                         continue;
6485                                 }
6486                                 extent_len = min(new_extents[i].num_bytes -
6487                                                  ext_offset, num_bytes);
6488
6489                                 ret = btrfs_insert_empty_item(trans, root,
6490                                                               path, &key,
6491                                                               sizeof(*fi));
6492                                 BUG_ON(ret);
6493
6494                                 leaf = path->nodes[0];
6495                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6496                                                 struct btrfs_file_extent_item);
6497                                 btrfs_set_file_extent_generation(leaf, fi,
6498                                                         trans->transid);
6499                                 btrfs_set_file_extent_type(leaf, fi,
6500                                                         BTRFS_FILE_EXTENT_REG);
6501                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6502                                                 new_extents[i].disk_bytenr);
6503                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6504                                                 new_extents[i].disk_num_bytes);
6505                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6506                                                 new_extents[i].ram_bytes);
6507
6508                                 btrfs_set_file_extent_compression(leaf, fi,
6509                                                 new_extents[i].compression);
6510                                 btrfs_set_file_extent_encryption(leaf, fi,
6511                                                 new_extents[i].encryption);
6512                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6513                                                 new_extents[i].other_encoding);
6514
6515                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6516                                                         extent_len);
6517                                 ext_offset += new_extents[i].offset;
6518                                 btrfs_set_file_extent_offset(leaf, fi,
6519                                                         ext_offset);
6520                                 btrfs_mark_buffer_dirty(leaf);
6521
6522                                 btrfs_drop_extent_cache(inode, key.offset,
6523                                                 key.offset + extent_len - 1, 0);
6524
6525                                 ret = btrfs_inc_extent_ref(trans, root,
6526                                                 new_extents[i].disk_bytenr,
6527                                                 new_extents[i].disk_num_bytes,
6528                                                 leaf->start,
6529                                                 root->root_key.objectid,
6530                                                 trans->transid, key.objectid);
6531                                 BUG_ON(ret);
6532                                 btrfs_release_path(root, path);
6533
6534                                 inode_add_bytes(inode, extent_len);
6535
6536                                 ext_offset = 0;
6537                                 num_bytes -= extent_len;
6538                                 key.offset += extent_len;
6539
6540                                 if (num_bytes == 0)
6541                                         break;
6542                         }
6543                         BUG_ON(i >= nr_extents);
6544 #endif
6545                 }
6546
6547                 if (extent_locked) {
6548                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6549                                       lock_end, GFP_NOFS);
6550                         extent_locked = 0;
6551                 }
6552 skip:
6553                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6554                     key.offset >= search_end)
6555                         break;
6556
6557                 cond_resched();
6558         }
6559         ret = 0;
6560 out:
6561         btrfs_release_path(root, path);
6562         if (inode) {
6563                 mutex_unlock(&inode->i_mutex);
6564                 if (extent_locked) {
6565                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6566                                       lock_end, GFP_NOFS);
6567                 }
6568                 iput(inode);
6569         }
6570         return ret;
6571 }
6572
6573 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6574                                struct btrfs_root *root,
6575                                struct extent_buffer *buf, u64 orig_start)
6576 {
6577         int level;
6578         int ret;
6579
6580         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6581         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6582
6583         level = btrfs_header_level(buf);
6584         if (level == 0) {
6585                 struct btrfs_leaf_ref *ref;
6586                 struct btrfs_leaf_ref *orig_ref;
6587
6588                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6589                 if (!orig_ref)
6590                         return -ENOENT;
6591
6592                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6593                 if (!ref) {
6594                         btrfs_free_leaf_ref(root, orig_ref);
6595                         return -ENOMEM;
6596                 }
6597
6598                 ref->nritems = orig_ref->nritems;
6599                 memcpy(ref->extents, orig_ref->extents,
6600                         sizeof(ref->extents[0]) * ref->nritems);
6601
6602                 btrfs_free_leaf_ref(root, orig_ref);
6603
6604                 ref->root_gen = trans->transid;
6605                 ref->bytenr = buf->start;
6606                 ref->owner = btrfs_header_owner(buf);
6607                 ref->generation = btrfs_header_generation(buf);
6608
6609                 ret = btrfs_add_leaf_ref(root, ref, 0);
6610                 WARN_ON(ret);
6611                 btrfs_free_leaf_ref(root, ref);
6612         }
6613         return 0;
6614 }
6615
6616 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6617                                         struct extent_buffer *leaf,
6618                                         struct btrfs_block_group_cache *group,
6619                                         struct btrfs_root *target_root)
6620 {
6621         struct btrfs_key key;
6622         struct inode *inode = NULL;
6623         struct btrfs_file_extent_item *fi;
6624         struct extent_state *cached_state = NULL;
6625         u64 num_bytes;
6626         u64 skip_objectid = 0;
6627         u32 nritems;
6628         u32 i;
6629
6630         nritems = btrfs_header_nritems(leaf);
6631         for (i = 0; i < nritems; i++) {
6632                 btrfs_item_key_to_cpu(leaf, &key, i);
6633                 if (key.objectid == skip_objectid ||
6634                     key.type != BTRFS_EXTENT_DATA_KEY)
6635                         continue;
6636                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6637                 if (btrfs_file_extent_type(leaf, fi) ==
6638                     BTRFS_FILE_EXTENT_INLINE)
6639                         continue;
6640                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6641                         continue;
6642                 if (!inode || inode->i_ino != key.objectid) {
6643                         iput(inode);
6644                         inode = btrfs_ilookup(target_root->fs_info->sb,
6645                                               key.objectid, target_root, 1);
6646                 }
6647                 if (!inode) {
6648                         skip_objectid = key.objectid;
6649                         continue;
6650                 }
6651                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6652
6653                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
6654                                  key.offset + num_bytes - 1, 0, &cached_state,
6655                                  GFP_NOFS);
6656                 btrfs_drop_extent_cache(inode, key.offset,
6657                                         key.offset + num_bytes - 1, 1);
6658                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
6659                                      key.offset + num_bytes - 1, &cached_state,
6660                                      GFP_NOFS);
6661                 cond_resched();
6662         }
6663         iput(inode);
6664         return 0;
6665 }
6666
6667 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6668                                         struct btrfs_root *root,
6669                                         struct extent_buffer *leaf,
6670                                         struct btrfs_block_group_cache *group,
6671                                         struct inode *reloc_inode)
6672 {
6673         struct btrfs_key key;
6674         struct btrfs_key extent_key;
6675         struct btrfs_file_extent_item *fi;
6676         struct btrfs_leaf_ref *ref;
6677         struct disk_extent *new_extent;
6678         u64 bytenr;
6679         u64 num_bytes;
6680         u32 nritems;
6681         u32 i;
6682         int ext_index;
6683         int nr_extent;
6684         int ret;
6685
6686         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6687         BUG_ON(!new_extent);
6688
6689         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6690         BUG_ON(!ref);
6691
6692         ext_index = -1;
6693         nritems = btrfs_header_nritems(leaf);
6694         for (i = 0; i < nritems; i++) {
6695                 btrfs_item_key_to_cpu(leaf, &key, i);
6696                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6697                         continue;
6698                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6699                 if (btrfs_file_extent_type(leaf, fi) ==
6700                     BTRFS_FILE_EXTENT_INLINE)
6701                         continue;
6702                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6703                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6704                 if (bytenr == 0)
6705                         continue;
6706
6707                 ext_index++;
6708                 if (bytenr >= group->key.objectid + group->key.offset ||
6709                     bytenr + num_bytes <= group->key.objectid)
6710                         continue;
6711
6712                 extent_key.objectid = bytenr;
6713                 extent_key.offset = num_bytes;
6714                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6715                 nr_extent = 1;
6716                 ret = get_new_locations(reloc_inode, &extent_key,
6717                                         group->key.objectid, 1,
6718                                         &new_extent, &nr_extent);
6719                 if (ret > 0)
6720                         continue;
6721                 BUG_ON(ret < 0);
6722
6723                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6724                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6725                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6726                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6727
6728                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6729                                                 new_extent->disk_bytenr);
6730                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6731                                                 new_extent->disk_num_bytes);
6732                 btrfs_mark_buffer_dirty(leaf);
6733
6734                 ret = btrfs_inc_extent_ref(trans, root,
6735                                         new_extent->disk_bytenr,
6736                                         new_extent->disk_num_bytes,
6737                                         leaf->start,
6738                                         root->root_key.objectid,
6739                                         trans->transid, key.objectid);
6740                 BUG_ON(ret);
6741
6742                 ret = btrfs_free_extent(trans, root,
6743                                         bytenr, num_bytes, leaf->start,
6744                                         btrfs_header_owner(leaf),
6745                                         btrfs_header_generation(leaf),
6746                                         key.objectid, 0);
6747                 BUG_ON(ret);
6748                 cond_resched();
6749         }
6750         kfree(new_extent);
6751         BUG_ON(ext_index + 1 != ref->nritems);
6752         btrfs_free_leaf_ref(root, ref);
6753         return 0;
6754 }
6755
6756 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6757                           struct btrfs_root *root)
6758 {
6759         struct btrfs_root *reloc_root;
6760         int ret;
6761
6762         if (root->reloc_root) {
6763                 reloc_root = root->reloc_root;
6764                 root->reloc_root = NULL;
6765                 list_add(&reloc_root->dead_list,
6766                          &root->fs_info->dead_reloc_roots);
6767
6768                 btrfs_set_root_bytenr(&reloc_root->root_item,
6769                                       reloc_root->node->start);
6770                 btrfs_set_root_level(&root->root_item,
6771                                      btrfs_header_level(reloc_root->node));
6772                 memset(&reloc_root->root_item.drop_progress, 0,
6773                         sizeof(struct btrfs_disk_key));
6774                 reloc_root->root_item.drop_level = 0;
6775
6776                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6777                                         &reloc_root->root_key,
6778                                         &reloc_root->root_item);
6779                 BUG_ON(ret);
6780         }
6781         return 0;
6782 }
6783
6784 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6785 {
6786         struct btrfs_trans_handle *trans;
6787         struct btrfs_root *reloc_root;
6788         struct btrfs_root *prev_root = NULL;
6789         struct list_head dead_roots;
6790         int ret;
6791         unsigned long nr;
6792
6793         INIT_LIST_HEAD(&dead_roots);
6794         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6795
6796         while (!list_empty(&dead_roots)) {
6797                 reloc_root = list_entry(dead_roots.prev,
6798                                         struct btrfs_root, dead_list);
6799                 list_del_init(&reloc_root->dead_list);
6800
6801                 BUG_ON(reloc_root->commit_root != NULL);
6802                 while (1) {
6803                         trans = btrfs_join_transaction(root, 1);
6804                         BUG_ON(!trans);
6805
6806                         mutex_lock(&root->fs_info->drop_mutex);
6807                         ret = btrfs_drop_snapshot(trans, reloc_root);
6808                         if (ret != -EAGAIN)
6809                                 break;
6810                         mutex_unlock(&root->fs_info->drop_mutex);
6811
6812                         nr = trans->blocks_used;
6813                         ret = btrfs_end_transaction(trans, root);
6814                         BUG_ON(ret);
6815                         btrfs_btree_balance_dirty(root, nr);
6816                 }
6817
6818                 free_extent_buffer(reloc_root->node);
6819
6820                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6821                                      &reloc_root->root_key);
6822                 BUG_ON(ret);
6823                 mutex_unlock(&root->fs_info->drop_mutex);
6824
6825                 nr = trans->blocks_used;
6826                 ret = btrfs_end_transaction(trans, root);
6827                 BUG_ON(ret);
6828                 btrfs_btree_balance_dirty(root, nr);
6829
6830                 kfree(prev_root);
6831                 prev_root = reloc_root;
6832         }
6833         if (prev_root) {
6834                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6835                 kfree(prev_root);
6836         }
6837         return 0;
6838 }
6839
6840 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6841 {
6842         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6843         return 0;
6844 }
6845
6846 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6847 {
6848         struct btrfs_root *reloc_root;
6849         struct btrfs_trans_handle *trans;
6850         struct btrfs_key location;
6851         int found;
6852         int ret;
6853
6854         mutex_lock(&root->fs_info->tree_reloc_mutex);
6855         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6856         BUG_ON(ret);
6857         found = !list_empty(&root->fs_info->dead_reloc_roots);
6858         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6859
6860         if (found) {
6861                 trans = btrfs_start_transaction(root, 1);
6862                 BUG_ON(!trans);
6863                 ret = btrfs_commit_transaction(trans, root);
6864                 BUG_ON(ret);
6865         }
6866
6867         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6868         location.offset = (u64)-1;
6869         location.type = BTRFS_ROOT_ITEM_KEY;
6870
6871         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6872         BUG_ON(!reloc_root);
6873         btrfs_orphan_cleanup(reloc_root);
6874         return 0;
6875 }
6876
6877 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6878                                     struct btrfs_root *root)
6879 {
6880         struct btrfs_root *reloc_root;
6881         struct extent_buffer *eb;
6882         struct btrfs_root_item *root_item;
6883         struct btrfs_key root_key;
6884         int ret;
6885
6886         BUG_ON(!root->ref_cows);
6887         if (root->reloc_root)
6888                 return 0;
6889
6890         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6891         BUG_ON(!root_item);
6892
6893         ret = btrfs_copy_root(trans, root, root->commit_root,
6894                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6895         BUG_ON(ret);
6896
6897         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6898         root_key.offset = root->root_key.objectid;
6899         root_key.type = BTRFS_ROOT_ITEM_KEY;
6900
6901         memcpy(root_item, &root->root_item, sizeof(root_item));
6902         btrfs_set_root_refs(root_item, 0);
6903         btrfs_set_root_bytenr(root_item, eb->start);
6904         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6905         btrfs_set_root_generation(root_item, trans->transid);
6906
6907         btrfs_tree_unlock(eb);
6908         free_extent_buffer(eb);
6909
6910         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6911                                 &root_key, root_item);
6912         BUG_ON(ret);
6913         kfree(root_item);
6914
6915         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6916                                                  &root_key);
6917         BUG_ON(!reloc_root);
6918         reloc_root->last_trans = trans->transid;
6919         reloc_root->commit_root = NULL;
6920         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6921
6922         root->reloc_root = reloc_root;
6923         return 0;
6924 }
6925
6926 /*
6927  * Core function of space balance.
6928  *
6929  * The idea is using reloc trees to relocate tree blocks in reference
6930  * counted roots. There is one reloc tree for each subvol, and all
6931  * reloc trees share same root key objectid. Reloc trees are snapshots
6932  * of the latest committed roots of subvols (root->commit_root).
6933  *
6934  * To relocate a tree block referenced by a subvol, there are two steps.
6935  * COW the block through subvol's reloc tree, then update block pointer
6936  * in the subvol to point to the new block. Since all reloc trees share
6937  * same root key objectid, doing special handing for tree blocks owned
6938  * by them is easy. Once a tree block has been COWed in one reloc tree,
6939  * we can use the resulting new block directly when the same block is
6940  * required to COW again through other reloc trees. By this way, relocated
6941  * tree blocks are shared between reloc trees, so they are also shared
6942  * between subvols.
6943  */
6944 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6945                                       struct btrfs_root *root,
6946                                       struct btrfs_path *path,
6947                                       struct btrfs_key *first_key,
6948                                       struct btrfs_ref_path *ref_path,
6949                                       struct btrfs_block_group_cache *group,
6950                                       struct inode *reloc_inode)
6951 {
6952         struct btrfs_root *reloc_root;
6953         struct extent_buffer *eb = NULL;
6954         struct btrfs_key *keys;
6955         u64 *nodes;
6956         int level;
6957         int shared_level;
6958         int lowest_level = 0;
6959         int ret;
6960
6961         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6962                 lowest_level = ref_path->owner_objectid;
6963
6964         if (!root->ref_cows) {
6965                 path->lowest_level = lowest_level;
6966                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6967                 BUG_ON(ret < 0);
6968                 path->lowest_level = 0;
6969                 btrfs_release_path(root, path);
6970                 return 0;
6971         }
6972
6973         mutex_lock(&root->fs_info->tree_reloc_mutex);
6974         ret = init_reloc_tree(trans, root);
6975         BUG_ON(ret);
6976         reloc_root = root->reloc_root;
6977
6978         shared_level = ref_path->shared_level;
6979         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6980
6981         keys = ref_path->node_keys;
6982         nodes = ref_path->new_nodes;
6983         memset(&keys[shared_level + 1], 0,
6984                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6985         memset(&nodes[shared_level + 1], 0,
6986                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6987
6988         if (nodes[lowest_level] == 0) {
6989                 path->lowest_level = lowest_level;
6990                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6991                                         0, 1);
6992                 BUG_ON(ret);
6993                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6994                         eb = path->nodes[level];
6995                         if (!eb || eb == reloc_root->node)
6996                                 break;
6997                         nodes[level] = eb->start;
6998                         if (level == 0)
6999                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7000                         else
7001                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7002                 }
7003                 if (nodes[0] &&
7004                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7005                         eb = path->nodes[0];
7006                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7007                                                       group, reloc_inode);
7008                         BUG_ON(ret);
7009                 }
7010                 btrfs_release_path(reloc_root, path);
7011         } else {
7012                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7013                                        lowest_level);
7014                 BUG_ON(ret);
7015         }
7016
7017         /*
7018          * replace tree blocks in the fs tree with tree blocks in
7019          * the reloc tree.
7020          */
7021         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7022         BUG_ON(ret < 0);
7023
7024         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7025                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7026                                         0, 0);
7027                 BUG_ON(ret);
7028                 extent_buffer_get(path->nodes[0]);
7029                 eb = path->nodes[0];
7030                 btrfs_release_path(reloc_root, path);
7031                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7032                 BUG_ON(ret);
7033                 free_extent_buffer(eb);
7034         }
7035
7036         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7037         path->lowest_level = 0;
7038         return 0;
7039 }
7040
7041 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7042                                         struct btrfs_root *root,
7043                                         struct btrfs_path *path,
7044                                         struct btrfs_key *first_key,
7045                                         struct btrfs_ref_path *ref_path)
7046 {
7047         int ret;
7048
7049         ret = relocate_one_path(trans, root, path, first_key,
7050                                 ref_path, NULL, NULL);
7051         BUG_ON(ret);
7052
7053         return 0;
7054 }
7055
7056 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7057                                     struct btrfs_root *extent_root,
7058                                     struct btrfs_path *path,
7059                                     struct btrfs_key *extent_key)
7060 {
7061         int ret;
7062
7063         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7064         if (ret)
7065                 goto out;
7066         ret = btrfs_del_item(trans, extent_root, path);
7067 out:
7068         btrfs_release_path(extent_root, path);
7069         return ret;
7070 }
7071
7072 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7073                                                 struct btrfs_ref_path *ref_path)
7074 {
7075         struct btrfs_key root_key;
7076
7077         root_key.objectid = ref_path->root_objectid;
7078         root_key.type = BTRFS_ROOT_ITEM_KEY;
7079         if (is_cowonly_root(ref_path->root_objectid))
7080                 root_key.offset = 0;
7081         else
7082                 root_key.offset = (u64)-1;
7083
7084         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7085 }
7086
7087 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7088                                         struct btrfs_path *path,
7089                                         struct btrfs_key *extent_key,
7090                                         struct btrfs_block_group_cache *group,
7091                                         struct inode *reloc_inode, int pass)
7092 {
7093         struct btrfs_trans_handle *trans;
7094         struct btrfs_root *found_root;
7095         struct btrfs_ref_path *ref_path = NULL;
7096         struct disk_extent *new_extents = NULL;
7097         int nr_extents = 0;
7098         int loops;
7099         int ret;
7100         int level;
7101         struct btrfs_key first_key;
7102         u64 prev_block = 0;
7103
7104
7105         trans = btrfs_start_transaction(extent_root, 1);
7106         BUG_ON(!trans);
7107
7108         if (extent_key->objectid == 0) {
7109                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7110                 goto out;
7111         }
7112
7113         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7114         if (!ref_path) {
7115                 ret = -ENOMEM;
7116                 goto out;
7117         }
7118
7119         for (loops = 0; ; loops++) {
7120                 if (loops == 0) {
7121                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7122                                                    extent_key->objectid);
7123                 } else {
7124                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7125                 }
7126                 if (ret < 0)
7127                         goto out;
7128                 if (ret > 0)
7129                         break;
7130
7131                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7132                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7133                         continue;
7134
7135                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7136                 BUG_ON(!found_root);
7137                 /*
7138                  * for reference counted tree, only process reference paths
7139                  * rooted at the latest committed root.
7140                  */
7141                 if (found_root->ref_cows &&
7142                     ref_path->root_generation != found_root->root_key.offset)
7143                         continue;
7144
7145                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7146                         if (pass == 0) {
7147                                 /*
7148                                  * copy data extents to new locations
7149                                  */
7150                                 u64 group_start = group->key.objectid;
7151                                 ret = relocate_data_extent(reloc_inode,
7152                                                            extent_key,
7153                                                            group_start);
7154                                 if (ret < 0)
7155                                         goto out;
7156                                 break;
7157                         }
7158                         level = 0;
7159                 } else {
7160                         level = ref_path->owner_objectid;
7161                 }
7162
7163                 if (prev_block != ref_path->nodes[level]) {
7164                         struct extent_buffer *eb;
7165                         u64 block_start = ref_path->nodes[level];
7166                         u64 block_size = btrfs_level_size(found_root, level);
7167
7168                         eb = read_tree_block(found_root, block_start,
7169                                              block_size, 0);
7170                         btrfs_tree_lock(eb);
7171                         BUG_ON(level != btrfs_header_level(eb));
7172
7173                         if (level == 0)
7174                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7175                         else
7176                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7177
7178                         btrfs_tree_unlock(eb);
7179                         free_extent_buffer(eb);
7180                         prev_block = block_start;
7181                 }
7182
7183                 mutex_lock(&extent_root->fs_info->trans_mutex);
7184                 btrfs_record_root_in_trans(found_root);
7185                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7186                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7187                         /*
7188                          * try to update data extent references while
7189                          * keeping metadata shared between snapshots.
7190                          */
7191                         if (pass == 1) {
7192                                 ret = relocate_one_path(trans, found_root,
7193                                                 path, &first_key, ref_path,
7194                                                 group, reloc_inode);
7195                                 if (ret < 0)
7196                                         goto out;
7197                                 continue;
7198                         }
7199                         /*
7200                          * use fallback method to process the remaining
7201                          * references.
7202                          */
7203                         if (!new_extents) {
7204                                 u64 group_start = group->key.objectid;
7205                                 new_extents = kmalloc(sizeof(*new_extents),
7206                                                       GFP_NOFS);
7207                                 nr_extents = 1;
7208                                 ret = get_new_locations(reloc_inode,
7209                                                         extent_key,
7210                                                         group_start, 1,
7211                                                         &new_extents,
7212                                                         &nr_extents);
7213                                 if (ret)
7214                                         goto out;
7215                         }
7216                         ret = replace_one_extent(trans, found_root,
7217                                                 path, extent_key,
7218                                                 &first_key, ref_path,
7219                                                 new_extents, nr_extents);
7220                 } else {
7221                         ret = relocate_tree_block(trans, found_root, path,
7222                                                   &first_key, ref_path);
7223                 }
7224                 if (ret < 0)
7225                         goto out;
7226         }
7227         ret = 0;
7228 out:
7229         btrfs_end_transaction(trans, extent_root);
7230         kfree(new_extents);
7231         kfree(ref_path);
7232         return ret;
7233 }
7234 #endif
7235
7236 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7237 {
7238         u64 num_devices;
7239         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7240                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7241
7242         num_devices = root->fs_info->fs_devices->rw_devices;
7243         if (num_devices == 1) {
7244                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7245                 stripped = flags & ~stripped;
7246
7247                 /* turn raid0 into single device chunks */
7248                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7249                         return stripped;
7250
7251                 /* turn mirroring into duplication */
7252                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7253                              BTRFS_BLOCK_GROUP_RAID10))
7254                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7255                 return flags;
7256         } else {
7257                 /* they already had raid on here, just return */
7258                 if (flags & stripped)
7259                         return flags;
7260
7261                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7262                 stripped = flags & ~stripped;
7263
7264                 /* switch duplicated blocks with raid1 */
7265                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7266                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7267
7268                 /* turn single device chunks into raid0 */
7269                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7270         }
7271         return flags;
7272 }
7273
7274 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7275                      struct btrfs_block_group_cache *shrink_block_group,
7276                      int force)
7277 {
7278         struct btrfs_trans_handle *trans;
7279         u64 new_alloc_flags;
7280         u64 calc;
7281
7282         spin_lock(&shrink_block_group->lock);
7283         if (btrfs_block_group_used(&shrink_block_group->item) +
7284             shrink_block_group->reserved > 0) {
7285                 spin_unlock(&shrink_block_group->lock);
7286
7287                 trans = btrfs_start_transaction(root, 1);
7288                 spin_lock(&shrink_block_group->lock);
7289
7290                 new_alloc_flags = update_block_group_flags(root,
7291                                                    shrink_block_group->flags);
7292                 if (new_alloc_flags != shrink_block_group->flags) {
7293                         calc =
7294                              btrfs_block_group_used(&shrink_block_group->item);
7295                 } else {
7296                         calc = shrink_block_group->key.offset;
7297                 }
7298                 spin_unlock(&shrink_block_group->lock);
7299
7300                 do_chunk_alloc(trans, root->fs_info->extent_root,
7301                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7302
7303                 btrfs_end_transaction(trans, root);
7304         } else
7305                 spin_unlock(&shrink_block_group->lock);
7306         return 0;
7307 }
7308
7309
7310 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7311                                          struct btrfs_block_group_cache *group)
7312
7313 {
7314         __alloc_chunk_for_shrink(root, group, 1);
7315         set_block_group_readonly(group);
7316         return 0;
7317 }
7318
7319 /*
7320  * checks to see if its even possible to relocate this block group.
7321  *
7322  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7323  * ok to go ahead and try.
7324  */
7325 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7326 {
7327         struct btrfs_block_group_cache *block_group;
7328         struct btrfs_space_info *space_info;
7329         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7330         struct btrfs_device *device;
7331         int full = 0;
7332         int ret = 0;
7333
7334         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7335
7336         /* odd, couldn't find the block group, leave it alone */
7337         if (!block_group)
7338                 return -1;
7339
7340         /* no bytes used, we're good */
7341         if (!btrfs_block_group_used(&block_group->item))
7342                 goto out;
7343
7344         space_info = block_group->space_info;
7345         spin_lock(&space_info->lock);
7346
7347         full = space_info->full;
7348
7349         /*
7350          * if this is the last block group we have in this space, we can't
7351          * relocate it unless we're able to allocate a new chunk below.
7352          *
7353          * Otherwise, we need to make sure we have room in the space to handle
7354          * all of the extents from this block group.  If we can, we're good
7355          */
7356         if ((space_info->total_bytes != block_group->key.offset) &&
7357            (space_info->bytes_used + space_info->bytes_reserved +
7358             space_info->bytes_pinned + space_info->bytes_readonly +
7359             btrfs_block_group_used(&block_group->item) <
7360             space_info->total_bytes)) {
7361                 spin_unlock(&space_info->lock);
7362                 goto out;
7363         }
7364         spin_unlock(&space_info->lock);
7365
7366         /*
7367          * ok we don't have enough space, but maybe we have free space on our
7368          * devices to allocate new chunks for relocation, so loop through our
7369          * alloc devices and guess if we have enough space.  However, if we
7370          * were marked as full, then we know there aren't enough chunks, and we
7371          * can just return.
7372          */
7373         ret = -1;
7374         if (full)
7375                 goto out;
7376
7377         mutex_lock(&root->fs_info->chunk_mutex);
7378         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7379                 u64 min_free = btrfs_block_group_used(&block_group->item);
7380                 u64 dev_offset, max_avail;
7381
7382                 /*
7383                  * check to make sure we can actually find a chunk with enough
7384                  * space to fit our block group in.
7385                  */
7386                 if (device->total_bytes > device->bytes_used + min_free) {
7387                         ret = find_free_dev_extent(NULL, device, min_free,
7388                                                    &dev_offset, &max_avail);
7389                         if (!ret)
7390                                 break;
7391                         ret = -1;
7392                 }
7393         }
7394         mutex_unlock(&root->fs_info->chunk_mutex);
7395 out:
7396         btrfs_put_block_group(block_group);
7397         return ret;
7398 }
7399
7400 static int find_first_block_group(struct btrfs_root *root,
7401                 struct btrfs_path *path, struct btrfs_key *key)
7402 {
7403         int ret = 0;
7404         struct btrfs_key found_key;
7405         struct extent_buffer *leaf;
7406         int slot;
7407
7408         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7409         if (ret < 0)
7410                 goto out;
7411
7412         while (1) {
7413                 slot = path->slots[0];
7414                 leaf = path->nodes[0];
7415                 if (slot >= btrfs_header_nritems(leaf)) {
7416                         ret = btrfs_next_leaf(root, path);
7417                         if (ret == 0)
7418                                 continue;
7419                         if (ret < 0)
7420                                 goto out;
7421                         break;
7422                 }
7423                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7424
7425                 if (found_key.objectid >= key->objectid &&
7426                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7427                         ret = 0;
7428                         goto out;
7429                 }
7430                 path->slots[0]++;
7431         }
7432 out:
7433         return ret;
7434 }
7435
7436 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7437 {
7438         struct btrfs_block_group_cache *block_group;
7439         struct btrfs_space_info *space_info;
7440         struct btrfs_caching_control *caching_ctl;
7441         struct rb_node *n;
7442
7443         down_write(&info->extent_commit_sem);
7444         while (!list_empty(&info->caching_block_groups)) {
7445                 caching_ctl = list_entry(info->caching_block_groups.next,
7446                                          struct btrfs_caching_control, list);
7447                 list_del(&caching_ctl->list);
7448                 put_caching_control(caching_ctl);
7449         }
7450         up_write(&info->extent_commit_sem);
7451
7452         spin_lock(&info->block_group_cache_lock);
7453         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7454                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7455                                        cache_node);
7456                 rb_erase(&block_group->cache_node,
7457                          &info->block_group_cache_tree);
7458                 spin_unlock(&info->block_group_cache_lock);
7459
7460                 down_write(&block_group->space_info->groups_sem);
7461                 list_del(&block_group->list);
7462                 up_write(&block_group->space_info->groups_sem);
7463
7464                 if (block_group->cached == BTRFS_CACHE_STARTED)
7465                         wait_block_group_cache_done(block_group);
7466
7467                 btrfs_remove_free_space_cache(block_group);
7468                 btrfs_put_block_group(block_group);
7469
7470                 spin_lock(&info->block_group_cache_lock);
7471         }
7472         spin_unlock(&info->block_group_cache_lock);
7473
7474         /* now that all the block groups are freed, go through and
7475          * free all the space_info structs.  This is only called during
7476          * the final stages of unmount, and so we know nobody is
7477          * using them.  We call synchronize_rcu() once before we start,
7478          * just to be on the safe side.
7479          */
7480         synchronize_rcu();
7481
7482         while(!list_empty(&info->space_info)) {
7483                 space_info = list_entry(info->space_info.next,
7484                                         struct btrfs_space_info,
7485                                         list);
7486
7487                 list_del(&space_info->list);
7488                 kfree(space_info);
7489         }
7490         return 0;
7491 }
7492
7493 static void __link_block_group(struct btrfs_space_info *space_info,
7494                                struct btrfs_block_group_cache *cache)
7495 {
7496         int index = get_block_group_index(cache);
7497
7498         down_write(&space_info->groups_sem);
7499         list_add_tail(&cache->list, &space_info->block_groups[index]);
7500         up_write(&space_info->groups_sem);
7501 }
7502
7503 int btrfs_read_block_groups(struct btrfs_root *root)
7504 {
7505         struct btrfs_path *path;
7506         int ret;
7507         struct btrfs_block_group_cache *cache;
7508         struct btrfs_fs_info *info = root->fs_info;
7509         struct btrfs_space_info *space_info;
7510         struct btrfs_key key;
7511         struct btrfs_key found_key;
7512         struct extent_buffer *leaf;
7513
7514         root = info->extent_root;
7515         key.objectid = 0;
7516         key.offset = 0;
7517         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7518         path = btrfs_alloc_path();
7519         if (!path)
7520                 return -ENOMEM;
7521
7522         while (1) {
7523                 ret = find_first_block_group(root, path, &key);
7524                 if (ret > 0)
7525                         break;
7526                 if (ret != 0)
7527                         goto error;
7528
7529                 leaf = path->nodes[0];
7530                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7531                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7532                 if (!cache) {
7533                         ret = -ENOMEM;
7534                         break;
7535                 }
7536
7537                 atomic_set(&cache->count, 1);
7538                 spin_lock_init(&cache->lock);
7539                 spin_lock_init(&cache->tree_lock);
7540                 cache->fs_info = info;
7541                 INIT_LIST_HEAD(&cache->list);
7542                 INIT_LIST_HEAD(&cache->cluster_list);
7543
7544                 /*
7545                  * we only want to have 32k of ram per block group for keeping
7546                  * track of free space, and if we pass 1/2 of that we want to
7547                  * start converting things over to using bitmaps
7548                  */
7549                 cache->extents_thresh = ((1024 * 32) / 2) /
7550                         sizeof(struct btrfs_free_space);
7551
7552                 read_extent_buffer(leaf, &cache->item,
7553                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7554                                    sizeof(cache->item));
7555                 memcpy(&cache->key, &found_key, sizeof(found_key));
7556
7557                 key.objectid = found_key.objectid + found_key.offset;
7558                 btrfs_release_path(root, path);
7559                 cache->flags = btrfs_block_group_flags(&cache->item);
7560                 cache->sectorsize = root->sectorsize;
7561
7562                 /*
7563                  * check for two cases, either we are full, and therefore
7564                  * don't need to bother with the caching work since we won't
7565                  * find any space, or we are empty, and we can just add all
7566                  * the space in and be done with it.  This saves us _alot_ of
7567                  * time, particularly in the full case.
7568                  */
7569                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7570                         exclude_super_stripes(root, cache);
7571                         cache->last_byte_to_unpin = (u64)-1;
7572                         cache->cached = BTRFS_CACHE_FINISHED;
7573                         free_excluded_extents(root, cache);
7574                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7575                         exclude_super_stripes(root, cache);
7576                         cache->last_byte_to_unpin = (u64)-1;
7577                         cache->cached = BTRFS_CACHE_FINISHED;
7578                         add_new_free_space(cache, root->fs_info,
7579                                            found_key.objectid,
7580                                            found_key.objectid +
7581                                            found_key.offset);
7582                         free_excluded_extents(root, cache);
7583                 }
7584
7585                 ret = update_space_info(info, cache->flags, found_key.offset,
7586                                         btrfs_block_group_used(&cache->item),
7587                                         &space_info);
7588                 BUG_ON(ret);
7589                 cache->space_info = space_info;
7590                 spin_lock(&cache->space_info->lock);
7591                 cache->space_info->bytes_super += cache->bytes_super;
7592                 spin_unlock(&cache->space_info->lock);
7593
7594                 __link_block_group(space_info, cache);
7595
7596                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7597                 BUG_ON(ret);
7598
7599                 set_avail_alloc_bits(root->fs_info, cache->flags);
7600                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7601                         set_block_group_readonly(cache);
7602         }
7603
7604         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7605                 if (!(get_alloc_profile(root, space_info->flags) &
7606                       (BTRFS_BLOCK_GROUP_RAID10 |
7607                        BTRFS_BLOCK_GROUP_RAID1 |
7608                        BTRFS_BLOCK_GROUP_DUP)))
7609                         continue;
7610                 /*
7611                  * avoid allocating from un-mirrored block group if there are
7612                  * mirrored block groups.
7613                  */
7614                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7615                         set_block_group_readonly(cache);
7616                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7617                         set_block_group_readonly(cache);
7618         }
7619         ret = 0;
7620 error:
7621         btrfs_free_path(path);
7622         return ret;
7623 }
7624
7625 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7626                            struct btrfs_root *root, u64 bytes_used,
7627                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7628                            u64 size)
7629 {
7630         int ret;
7631         struct btrfs_root *extent_root;
7632         struct btrfs_block_group_cache *cache;
7633
7634         extent_root = root->fs_info->extent_root;
7635
7636         root->fs_info->last_trans_log_full_commit = trans->transid;
7637
7638         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7639         if (!cache)
7640                 return -ENOMEM;
7641
7642         cache->key.objectid = chunk_offset;
7643         cache->key.offset = size;
7644         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7645         cache->sectorsize = root->sectorsize;
7646
7647         /*
7648          * we only want to have 32k of ram per block group for keeping track
7649          * of free space, and if we pass 1/2 of that we want to start
7650          * converting things over to using bitmaps
7651          */
7652         cache->extents_thresh = ((1024 * 32) / 2) /
7653                 sizeof(struct btrfs_free_space);
7654         atomic_set(&cache->count, 1);
7655         spin_lock_init(&cache->lock);
7656         spin_lock_init(&cache->tree_lock);
7657         INIT_LIST_HEAD(&cache->list);
7658         INIT_LIST_HEAD(&cache->cluster_list);
7659
7660         btrfs_set_block_group_used(&cache->item, bytes_used);
7661         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7662         cache->flags = type;
7663         btrfs_set_block_group_flags(&cache->item, type);
7664
7665         cache->last_byte_to_unpin = (u64)-1;
7666         cache->cached = BTRFS_CACHE_FINISHED;
7667         exclude_super_stripes(root, cache);
7668
7669         add_new_free_space(cache, root->fs_info, chunk_offset,
7670                            chunk_offset + size);
7671
7672         free_excluded_extents(root, cache);
7673
7674         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7675                                 &cache->space_info);
7676         BUG_ON(ret);
7677
7678         spin_lock(&cache->space_info->lock);
7679         cache->space_info->bytes_super += cache->bytes_super;
7680         spin_unlock(&cache->space_info->lock);
7681
7682         __link_block_group(cache->space_info, cache);
7683
7684         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7685         BUG_ON(ret);
7686
7687         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7688                                 sizeof(cache->item));
7689         BUG_ON(ret);
7690
7691         set_avail_alloc_bits(extent_root->fs_info, type);
7692
7693         return 0;
7694 }
7695
7696 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7697                              struct btrfs_root *root, u64 group_start)
7698 {
7699         struct btrfs_path *path;
7700         struct btrfs_block_group_cache *block_group;
7701         struct btrfs_free_cluster *cluster;
7702         struct btrfs_key key;
7703         int ret;
7704
7705         root = root->fs_info->extent_root;
7706
7707         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7708         BUG_ON(!block_group);
7709         BUG_ON(!block_group->ro);
7710
7711         memcpy(&key, &block_group->key, sizeof(key));
7712
7713         /* make sure this block group isn't part of an allocation cluster */
7714         cluster = &root->fs_info->data_alloc_cluster;
7715         spin_lock(&cluster->refill_lock);
7716         btrfs_return_cluster_to_free_space(block_group, cluster);
7717         spin_unlock(&cluster->refill_lock);
7718
7719         /*
7720          * make sure this block group isn't part of a metadata
7721          * allocation cluster
7722          */
7723         cluster = &root->fs_info->meta_alloc_cluster;
7724         spin_lock(&cluster->refill_lock);
7725         btrfs_return_cluster_to_free_space(block_group, cluster);
7726         spin_unlock(&cluster->refill_lock);
7727
7728         path = btrfs_alloc_path();
7729         BUG_ON(!path);
7730
7731         spin_lock(&root->fs_info->block_group_cache_lock);
7732         rb_erase(&block_group->cache_node,
7733                  &root->fs_info->block_group_cache_tree);
7734         spin_unlock(&root->fs_info->block_group_cache_lock);
7735
7736         down_write(&block_group->space_info->groups_sem);
7737         /*
7738          * we must use list_del_init so people can check to see if they
7739          * are still on the list after taking the semaphore
7740          */
7741         list_del_init(&block_group->list);
7742         up_write(&block_group->space_info->groups_sem);
7743
7744         if (block_group->cached == BTRFS_CACHE_STARTED)
7745                 wait_block_group_cache_done(block_group);
7746
7747         btrfs_remove_free_space_cache(block_group);
7748
7749         spin_lock(&block_group->space_info->lock);
7750         block_group->space_info->total_bytes -= block_group->key.offset;
7751         block_group->space_info->bytes_readonly -= block_group->key.offset;
7752         spin_unlock(&block_group->space_info->lock);
7753
7754         btrfs_clear_space_info_full(root->fs_info);
7755
7756         btrfs_put_block_group(block_group);
7757         btrfs_put_block_group(block_group);
7758
7759         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7760         if (ret > 0)
7761                 ret = -EIO;
7762         if (ret < 0)
7763                 goto out;
7764
7765         ret = btrfs_del_item(trans, root, path);
7766 out:
7767         btrfs_free_path(path);
7768         return ret;
7769 }